20260818 fe148 ours RP镜像统一panda-rpki命名：VERSION单一版本来源v1.3.0，镜像tag语义化，release出包强制HEAD已打同版git tag(#148)

VERSION

@@ -0,0 +1 @@
+v1.3.0

deploy/docker-installer/.env.example

@@ -39,14 +39,36 @@ RETAIN_RUNS=100
 # TAL/TA input mode:
 # file-with-ta: use packaged fixture TAL + TA only.
 # file-live-ta: use packaged fixture TAL; snapshot waits for live TA refresh, delta refreshes TA in background.
+# custom-file-with-ta: use the custom TAL/TA below (requires exactly RIRS=custom);
+#   for the local test repo compose component or a third-party test TA.
 # url: pass TAL URL to child process.
 TAL_INPUT_MODE=file-live-ta
 LIVE_TA_REFRESH_BEFORE_SNAPSHOT=1
 LIVE_TA_REFRESH_CONNECT_TIMEOUT_SECS=15
 LIVE_TA_REFRESH_MAX_TIME_SECS=120
 
+# Custom TAL/TA mode (only active with RIRS=custom + TAL_INPUT_MODE=custom-file-with-ta).
+# CUSTOM_FIXTURE_HOST_DIR is the host directory bind-mounted to /opt/ours-rp/custom-fixtures.
+# CUSTOM_TAL_URI is recorded in CIR as the TAL origin URI.
+# HTTP_ROOT_CERT_PATHS lists HTTPS TLS root certificates (space-separated, container
+# paths) for self-signed RRDP/TAL servers; it is independent of the RPKI TA.
+#RIRS=custom
+#TAL_INPUT_MODE=custom-file-with-ta
+#CUSTOM_FIXTURE_HOST_DIR=../custom-fixtures
+#CUSTOM_TAL_PATH=/opt/ours-rp/custom-fixtures/tal/custom.tal
+#CUSTOM_TA_PATH=/opt/ours-rp/custom-fixtures/ta/custom-ta.cer
+#CUSTOM_TAL_URI=https://rpki-local-rrdp:8443/tal/custom.tal
+#HTTP_ROOT_CERT_PATHS=/opt/ours-rp/custom-fixtures/certs/rrdp-ca.pem
+# Optional shared docker network between the soak container and the stack local
+# test repository component. When set, scripts/start.sh attaches the soak
+# container to this external network before starting it.
+#CUSTOM_REPO_NETWORK=rpki-local-repo-net
+
 # Sync and runtime behavior.
 RSYNC_SCOPE=module-root
+# HTTP User-Agent for all outgoing HTTP requests (RRDP / TAL / TA / dead-repo probes).
+# Unset, empty, or header-illegal values fall back to panda-rpki/0.2.
+#RPKI_HTTP_USER_AGENT=panda-rpki/0.2
 DISABLE_COMPETING_RPS=0
 RUN_ROOT=/var/lib/ours-rp
 DB_DIR=/var/lib/ours-rp/state/db

deploy/docker-installer/compose/docker-compose.yml

@@ -11,8 +11,10 @@ services:
       RUN_ROOT: /var/lib/ours-rp
       BIN_DIR: /opt/ours-rp/bin
       FIXTURE_DIR: /opt/ours-rp/fixtures
+      CUSTOM_FIXTURE_DIR: /opt/ours-rp/custom-fixtures
     volumes:
       - ../.env:/opt/ours-rp/.env:ro
+      - ${CUSTOM_FIXTURE_HOST_DIR:-../custom-fixtures}:/opt/ours-rp/custom-fixtures:ro
       - ${HOST_DATA_DIR:-/var/lib/ours-rp-package-installer}/state:/var/lib/ours-rp/state
       - ${HOST_DATA_DIR:-/var/lib/ours-rp-package-installer}/runs:/var/lib/ours-rp/runs
       - ${HOST_DATA_DIR:-/var/lib/ours-rp-package-installer}/logs:/var/lib/ours-rp/logs

deploy/docker-installer/custom-fixtures/README.md

@@ -0,0 +1,19 @@
+# custom-fixtures 占位目录
+
+默认空目录，仅作为 `compose/docker-compose.yml` 中
+`${CUSTOM_FIXTURE_HOST_DIR:-../custom-fixtures}` 挂载的缺省目标，保证
+RIR 默认模式下 compose 不因缺目录而失败。
+
+启用 custom TAL 模式（`RIRS=custom` + `TAL_INPUT_MODE=custom-file-with-ta`）时，
+把 `CUSTOM_FIXTURE_HOST_DIR` 指向包含以下结构的主机目录：
+
+```text
+<fixture-root>/
+├── tal/custom.tal          # 自定义 TAL
+├── ta/custom-ta.cer        # 自定义 TA DER
+└── certs/rrdp-ca.pem       # 自签 HTTPS 服务的 TLS root cert（如需要）
+```
+
+本仓库的 `rpki_local_repo_installer` 组件会在其 `$HOST_DATA_DIR/fixtures`
+下生成完全兼容的目录结构；stack installer 的 `STACK_TA_MODE=local` 会自动
+完成该指向。

deploy/docker-installer/docs/README.en.md

@@ -54,6 +54,12 @@ Defaults:
 - `ALLOW_CROSS_ARCH=0`
 - `RTR_REPORT_DIR=__HOST_DATA_DIR__/empty-rtr-report`, which mounts an empty fallback directory by default; after deploying a separate RTR service, register its same-host report directory through `scripts/register_rtr_monitor.sh`
 
+To integrate a local or third-party test TA, use the custom TAL mode
+(`RIRS=custom` + `TAL_INPUT_MODE=custom-file-with-ta` together with
+`CUSTOM_TAL_PATH` / `CUSTOM_TA_PATH` / `CUSTOM_TAL_URI` /
+`HTTP_ROOT_CERT_PATHS` and the `CUSTOM_FIXTURE_HOST_DIR` mount); see the
+commented section in `.env.example` for details.
+
 ## First Start Semantics
 
 If there is no successful run under `HOST_DATA_DIR/runs`, `start.sh` starts the core `ours-rp-soak` service first and waits for the first snapshot to succeed before starting metrics, Prometheus and Grafana.

deploy/docker-installer/docs/README.zh-CN.md

@@ -54,6 +54,11 @@ vim .env
 - `ALLOW_CROSS_ARCH=0`
 - `RTR_REPORT_DIR=__HOST_DATA_DIR__/empty-rtr-report`，默认挂载空目录；独立 RTR 服务部署完成后，使用 `scripts/register_rtr_monitor.sh` 注册同机 report 目录
 
+如需对接本地或第三方测试 TA，可使用 custom TAL 模式（`RIRS=custom` +
+`TAL_INPUT_MODE=custom-file-with-ta`，配合 `CUSTOM_TAL_PATH` / `CUSTOM_TA_PATH` /
+`CUSTOM_TAL_URI` / `HTTP_ROOT_CERT_PATHS` 与 `CUSTOM_FIXTURE_HOST_DIR` 挂载），
+配置项说明见 `.env.example` 中的注释段。
+
 ## 首次启动语义
 
 如果 `HOST_DATA_DIR/runs` 下没有成功 run，`start.sh` 会先启动核心 `ours-rp-soak`，等待第一轮 snapshot 成功后再启动 metrics、Prometheus 和 Grafana。

deploy/docker-installer/scripts/start.sh

@@ -47,7 +47,21 @@ if has_success_run; then
 fi
 
 log "starting core soak service"
-compose_cmd --profile core up -d ours-rp-soak
+if [[ -n "${CUSTOM_REPO_NETWORK:-}" ]]; then
+  # Custom TAL mode against the stack local test repository: create the soak
+  # container without starting it, attach the shared external network first,
+  # then start it. This avoids the first snapshot racing the network connect.
+  compose_cmd --profile core up -d --no-start ours-rp-soak 2>/dev/null \
+    || compose_cmd --profile core create ours-rp-soak
+  soak_container="${COMPOSE_PROJECT_NAME:-ours-rp-package-installer}-soak"
+  if ! docker inspect -f '{{json .NetworkSettings.Networks}}' "$soak_container" | grep -q "\"$CUSTOM_REPO_NETWORK\""; then
+    log "attaching soak container to custom repo network: $CUSTOM_REPO_NETWORK"
+    docker network connect "$CUSTOM_REPO_NETWORK" "$soak_container"
+  fi
+  compose_cmd --profile core start ours-rp-soak
+else
+  compose_cmd --profile core up -d ours-rp-soak
+fi
 
 if [[ "$had_success" == "0" && "$WAIT_FIRST_RUN" == "1" ]]; then
   log "no previous successful run found; waiting for first run timeout=${timeout_secs}s"

deploy/rpki-explorer/.env.example

@@ -0,0 +1,22 @@
+# rpki-explorer standalone stack configuration.
+# Copy to .env and adjust paths for the target host.
+
+# Soak outputs on the host (mounted read-only).
+SOAK_RUNS_DIR=/root/rpki096_x86_rtr_installer/data/runs
+SOAK_DB_DIR=/root/rpki096_x86_rtr_installer/data/state/db
+
+# Host directory where the query-db index is persisted (reused across restarts).
+QUERY_DB_DIR=/root/rpki128_explorer/query-db
+
+# Host loopback port the explorer is published on (SSH port-forward access).
+LISTEN_PORT=9517
+
+# Keep only the latest N indexed runs; first boot backfills the latest N runs.
+RETAIN_RUNS=10
+
+# Watch poll interval for newly completed runs.
+WATCH_INTERVAL_SECS=10
+
+# Image coordinates (built locally, transferred via docker save/load).
+IMAGE_NAME=rpki-explorer
+IMAGE_TAG=local

deploy/rpki-explorer/Dockerfile

@@ -0,0 +1,68 @@
+# rpki-explorer all-in-one image:
+#   - rpki_query_service (watch mode) + rpki_query_indexer
+#   - nginx serving the explorer SPA and proxying /api/v1/ to the query service
+#
+# Build context must be the repository root (rpki_2/rpki):
+#   docker build -f deploy/rpki-explorer/Dockerfile -t rpki-explorer:<tag> .
+#
+# Note: no `# syntax=` directive on purpose — the default images below are all
+# expected to be present locally, and pulling the dockerfile frontend from
+# docker.io may not be possible on offline build hosts.
+
+ARG RUST_BUILDER_IMAGE=rust:1-bookworm
+ARG NODE_BUILDER_IMAGE=node:slim
+ARG RUNTIME_IMAGE=nginx:1.27-bookworm
+
+FROM ${RUST_BUILDER_IMAGE} AS rust-builder
+
+WORKDIR /src
+
+RUN apt-get update \
+    && apt-get install -y --no-install-recommends \
+      build-essential \
+      ca-certificates \
+      clang \
+      cmake \
+      git \
+      libclang-dev \
+      make \
+      perl \
+      pkg-config \
+      python3 \
+    && rm -rf /var/lib/apt/lists/*
+
+COPY . .
+
+RUN --mount=type=cache,target=/usr/local/cargo/registry \
+    --mount=type=cache,target=/usr/local/cargo/git \
+    --mount=type=cache,target=/src/target \
+    cargo build --release --bin rpki_query_service --bin rpki_query_indexer \
+    && cp target/release/rpki_query_service target/release/rpki_query_indexer /usr/local/bin/
+
+FROM ${NODE_BUILDER_IMAGE} AS ui-builder
+
+WORKDIR /ui
+
+COPY ui/rpki-explorer/package.json ui/rpki-explorer/package-lock.json ./
+RUN npm ci --no-audit --no-fund
+
+COPY ui/rpki-explorer/ ./
+RUN npm run build
+
+FROM ${RUNTIME_IMAGE}
+
+ARG GIT_REV=unknown
+LABEL org.opencontainers.image.title="rpki-explorer" \
+      org.opencontainers.image.revision="${GIT_REV}"
+
+COPY --from=rust-builder /usr/local/bin/rpki_query_service /usr/local/bin/rpki_query_service
+COPY --from=rust-builder /usr/local/bin/rpki_query_indexer /usr/local/bin/rpki_query_indexer
+COPY --from=ui-builder /ui/dist /usr/share/nginx/html
+COPY deploy/rpki-explorer/nginx.conf /etc/nginx/conf.d/default.conf
+COPY deploy/rpki-explorer/entrypoint.sh /entrypoint.sh
+
+RUN chmod +x /entrypoint.sh && mkdir -p /data
+
+EXPOSE 8080
+
+ENTRYPOINT ["/entrypoint.sh"]

deploy/rpki-explorer/README.md

@@ -0,0 +1,76 @@
+# rpki-explorer standalone stack
+
+One container running the RPKI Explorer web UI plus the query backend
+(`rpki_query_service` in watch mode, spawning `rpki_query_indexer`), deployed
+next to a running ours-RP soak. It mounts the soak outputs **read-only** and
+keeps only the latest `RETAIN_RUNS` indexed runs.
+
+```text
+browser ──ssh -L──► 127.0.0.1:9517 ──► container
+                                        ├─ nginx :8080  (dist + /api/v1/ proxy)
+                                        └─ rpki_query_service :9557
+                                             ├─ /data/query-db        (host bind mount, rw, reused on restart)
+                                             ├─ /soak/db/repo-bytes.db (ro)
+                                             └─ /soak/runs             (ro, watched)
+```
+
+## Prerequisites
+
+- Docker with the compose plugin (`docker compose version`) on the target host.
+- A running soak whose `runs/` directories contain `report.json` (and
+  `run-summary.json` once finished) — only completed runs get indexed.
+
+## Build (local machine, same arch as the target)
+
+```bash
+deploy/rpki-explorer/build_image.sh
+# tarball + provenance land in target/rpki-explorer-image/
+```
+
+The build runs entirely inside Docker (rust release binaries + vite build),
+so no local toolchain version requirements apply.
+
+## Transfer and deploy (remote host)
+
+```bash
+cat target/rpki-explorer-image/rpki-explorer-<tag>.tar.gz | ssh root@<host> 'gunzip | docker load'
+
+ssh root@<host>
+mkdir -p /root/rpki128_explorer
+cd /root/rpki128_explorer
+# place docker-compose.yml and .env here (see .env.example)
+docker compose up -d
+docker logs -f rpki-explorer
+```
+
+On first boot the entrypoint computes
+`--watch-min-run-seq = latest_completed - RETAIN_RUNS + 1` so exactly the
+latest `RETAIN_RUNS` completed runs are backfilled. On later boots the
+existing query-db drives continuation — the index built earlier is reused, no
+re-backfill happens.
+
+## Access
+
+```bash
+ssh -L 9517:127.0.0.1:9517 root@<host>
+# open http://127.0.0.1:9517
+```
+
+The stack binds loopback only; there is no authentication or TLS.
+
+## Upgrade
+
+Rebuild the image with a new tag, transfer and `docker load`, update
+`IMAGE_TAG` in `.env`, then `docker compose up -d`. The query-db on the host
+is preserved.
+
+## Rollback / removal
+
+```bash
+docker compose down
+# keep /root/rpki128_explorer/query-db to reuse the index later, or delete it
+# to force a fresh backfill of the latest RETAIN_RUNS runs.
+```
+
+The soak stack is never touched; removing the explorer stack has no effect on
+it.

deploy/rpki-explorer/build_image.sh

@@ -0,0 +1,37 @@
+#!/usr/bin/env bash
+# Build the rpki-explorer all-in-one image locally and export it as a tarball
+# ready to transfer to a remote host (docker load on the other side).
+set -euo pipefail
+
+REPO_ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)"
+OUT_DIR="${OUT_DIR:-$REPO_ROOT/target/rpki-explorer-image}"
+IMAGE_NAME="${IMAGE_NAME:-rpki-explorer}"
+GIT_REV="$(git -C "$REPO_ROOT" rev-parse --short HEAD)"
+IMAGE_TAG="${IMAGE_TAG:-$GIT_REV}"
+
+dirty_count="$(git -C "$REPO_ROOT" status --porcelain | wc -l)"
+
+echo "[build] repo=$REPO_ROOT rev=$GIT_REV dirty_files=$dirty_count tag=$IMAGE_TAG"
+
+docker build \
+  --file "$REPO_ROOT/deploy/rpki-explorer/Dockerfile" \
+  --build-arg "GIT_REV=$GIT_REV" \
+  --tag "$IMAGE_NAME:$IMAGE_TAG" \
+  "$REPO_ROOT"
+
+mkdir -p "$OUT_DIR"
+TARBALL="$OUT_DIR/rpki-explorer-$IMAGE_TAG.tar.gz"
+docker save "$IMAGE_NAME:$IMAGE_TAG" | gzip > "$TARBALL"
+
+IMAGE_ID="$(docker image inspect --format '{{.Id}}' "$IMAGE_NAME:$IMAGE_TAG")"
+cat > "$OUT_DIR/rpki-explorer-$IMAGE_TAG.provenance.txt" <<EOF
+git_rev=$GIT_REV
+dirty_files=$dirty_count
+build_time_utc=$(date -u +%Y-%m-%dT%H:%M:%SZ)
+image=$IMAGE_NAME:$IMAGE_TAG
+image_id=$IMAGE_ID
+EOF
+
+echo "[build] image tarball: $TARBALL"
+echo "[build] provenance:    $OUT_DIR/rpki-explorer-$IMAGE_TAG.provenance.txt"
+echo "[build] transfer:      cat '$TARBALL' | ssh root@<host> 'gunzip | docker load'"

deploy/rpki-explorer/docker-compose.yml

@@ -0,0 +1,34 @@
+name: rpki-explorer
+
+# Standalone explorer stack: one container with nginx + rpki_query_service
+# (+ rpki_query_indexer spawned by the service watcher). It mounts the soak
+# run root and state db read-only and persists its query-db on the host so
+# container/host restarts reuse the already-built index.
+
+services:
+  rpki-explorer:
+    image: ${IMAGE_NAME:-rpki-explorer}:${IMAGE_TAG:-local}
+    build:
+      context: ../..
+      dockerfile: deploy/rpki-explorer/Dockerfile
+      args:
+        GIT_REV: ${GIT_REV:-unknown}
+    container_name: rpki-explorer
+    restart: unless-stopped
+    # RocksDB keeps table files open across 13 column families; the default
+    # 1024 nofile limit is too tight under concurrent reads.
+    ulimits:
+      nofile:
+        soft: 65536
+        hard: 65536
+    ports:
+      # Loopback only: access via SSH port forwarding, e.g.
+      #   ssh -L 9517:127.0.0.1:9517 root@<host>
+      - "127.0.0.1:${LISTEN_PORT:-9517}:8080"
+    environment:
+      RETAIN_RUNS: "${RETAIN_RUNS:-10}"
+      WATCH_INTERVAL_SECS: "${WATCH_INTERVAL_SECS:-10}"
+    volumes:
+      - "${SOAK_RUNS_DIR:?set SOAK_RUNS_DIR in .env}:/soak/runs:ro"
+      - "${SOAK_DB_DIR:?set SOAK_DB_DIR in .env}:/soak/db:ro"
+      - "${QUERY_DB_DIR:-./query-db}:/data"

deploy/rpki-explorer/entrypoint.sh

@@ -0,0 +1,92 @@
+#!/usr/bin/env bash
+# Container entrypoint: run rpki_query_service (watch mode) and nginx in one
+# container. nginx serves the explorer SPA and proxies /api/v1/ to the query
+# service. If either process exits, the container exits and the compose
+# restart policy brings it back.
+set -euo pipefail
+
+RUNS_DIR="${RUNS_DIR:-/soak/runs}"
+REPO_BYTES_DB="${REPO_BYTES_DB:-/soak/db/repo-bytes.db}"
+QUERY_DB="${QUERY_DB:-/data/query-db}"
+RETAIN_RUNS="${RETAIN_RUNS:-10}"
+WATCH_INTERVAL_SECS="${WATCH_INTERVAL_SECS:-10}"
+QUERY_LISTEN="${QUERY_LISTEN:-127.0.0.1:9557}"
+
+log() { echo "[entrypoint] $*"; }
+
+latest_completed_seq() {
+  local latest=0 dir base n
+  shopt -s nullglob
+  for dir in "$RUNS_DIR"/run_*/; do
+    [ -f "${dir}report.json" ] || continue
+    base="${dir%/}"
+    n="${base##*/run_}"
+    case "$n" in
+      *[!0-9]* | "") continue ;;
+    esac
+    if [ "$n" -gt "$latest" ]; then
+      latest="$n"
+    fi
+  done
+  echo "$latest"
+}
+
+mkdir -p "$(dirname "$QUERY_DB")"
+
+ARGS=(
+  --query-db "$QUERY_DB"
+  --listen "$QUERY_LISTEN"
+  --watch-run-root "$RUNS_DIR"
+  --watch-interval-secs "$WATCH_INTERVAL_SECS"
+  --retain-indexed-runs "$RETAIN_RUNS"
+)
+
+if [ -e "$REPO_BYTES_DB" ]; then
+  ARGS+=(--repo-bytes-db "$REPO_BYTES_DB")
+else
+  log "WARNING: repo-bytes db not found at $REPO_BYTES_DB; parsed/raw/export features will be unavailable"
+fi
+
+# Decide the watcher start point:
+# - explicit WATCH_MIN_RUN_SEQ always wins;
+# - on first boot (query-db does not exist yet) backfill the latest
+#   RETAIN_RUNS completed runs;
+# - on later boots the existing query-db drives continuation
+#   (max(min_run_seq, latest_ready+1)), so no flag is needed.
+if [ -n "${WATCH_MIN_RUN_SEQ:-}" ]; then
+  ARGS+=(--watch-min-run-seq "$WATCH_MIN_RUN_SEQ")
+  log "using explicit WATCH_MIN_RUN_SEQ=$WATCH_MIN_RUN_SEQ"
+elif [ ! -e "$QUERY_DB" ]; then
+  latest="$(latest_completed_seq)"
+  if [ "$latest" -gt "$RETAIN_RUNS" ]; then
+    min_seq=$((latest - RETAIN_RUNS + 1))
+  else
+    min_seq=1
+  fi
+  log "first boot: latest completed run seq=$latest -> --watch-min-run-seq $min_seq (backfill $RETAIN_RUNS runs)"
+  ARGS+=(--watch-min-run-seq "$min_seq")
+else
+  log "existing query-db at $QUERY_DB; resuming from indexed state"
+fi
+
+log "starting rpki_query_service ${ARGS[*]}"
+rpki_query_service "${ARGS[@]}" &
+QS_PID=$!
+
+log "starting nginx"
+nginx -g 'daemon off;' &
+NGINX_PID=$!
+
+shutdown() {
+  log "shutting down (qs=$QS_PID nginx=$NGINX_PID)"
+  kill -TERM "$QS_PID" 2>/dev/null || true
+  kill -TERM "$NGINX_PID" 2>/dev/null || true
+}
+trap shutdown TERM INT
+
+wait -n "$QS_PID" "$NGINX_PID"
+EXIT_CODE=$?
+log "a process exited (code=$EXIT_CODE); stopping the remaining one"
+shutdown
+wait "$QS_PID" "$NGINX_PID" 2>/dev/null || true
+exit "$EXIT_CODE"

deploy/rpki-explorer/nginx.conf

@@ -0,0 +1,48 @@
+server {
+    listen 8080;
+    server_name _;
+
+    access_log /dev/stdout;
+    error_log /dev/stderr;
+
+    root /usr/share/nginx/html;
+    index index.html;
+
+    # Keep API/proxy redirects relative so port-forwarded access (ssh -L) is
+    # not bounced to an unreachable absolute URL.
+    absolute_redirect off;
+    port_in_redirect off;
+
+    # Hand-typed API URLs that fail to parse (e.g. raw % in a path segment)
+    # should land on the SPA instead of nginx's default 400 page.
+    error_page 400 /index.html;
+
+    # `/api/v1` without trailing slash must reach the service info endpoint
+    # too, not the SPA fallback.
+    location = /api/v1 {
+        proxy_pass http://127.0.0.1:9557;
+        proxy_http_version 1.1;
+        proxy_set_header Host $host;
+        proxy_set_header X-Real-IP $remote_addr;
+        proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
+        proxy_read_timeout 300s;
+        proxy_send_timeout 300s;
+    }
+
+    # query service has no CORS headers and no authentication: the API must be
+    # served same-origin with the UI.
+    location /api/v1/ {
+        proxy_pass http://127.0.0.1:9557;
+        proxy_http_version 1.1;
+        proxy_set_header Host $host;
+        proxy_set_header X-Real-IP $remote_addr;
+        proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
+        proxy_read_timeout 300s;
+        proxy_send_timeout 300s;
+    }
+
+    # SPA fallback: routing is client-side.
+    location / {
+        try_files $uri $uri/ /index.html;
+    }
+}

docker/ours-rp-metrics.Dockerfile

@@ -139,7 +139,7 @@ ARG SOURCE_DIRTY=unknown
 ARG BUILD_TIMESTAMP_UTC=unknown
 ARG RPKI_CLIENT_VERSION=9.8
 
-LABEL org.opencontainers.image.title="ours-rp-metrics" \
+LABEL org.opencontainers.image.title="panda-rpki-metrics" \
       org.opencontainers.image.description="Ours RP artifact metrics image for multi-arch Docker Compose deployment" \
       org.opencontainers.image.revision="${SOURCE_COMMIT}" \
       org.opencontainers.image.created="${BUILD_TIMESTAMP_UTC}" \

docker/ours-rp-runtime.Dockerfile

@@ -124,7 +124,7 @@ ARG SOURCE_COMMIT=unknown
 ARG SOURCE_DIRTY=unknown
 ARG BUILD_TIMESTAMP_UTC=unknown
 
-LABEL org.opencontainers.image.title="ours-rp-runtime" \
+LABEL org.opencontainers.image.title="panda-rpki-rp" \
       org.opencontainers.image.description="Ours RP runtime image for multi-arch Docker Compose deployment" \
       org.opencontainers.image.revision="${SOURCE_COMMIT}" \
       org.opencontainers.image.created="${BUILD_TIMESTAMP_UTC}" \

scripts/docker/build_docker_installer_package.sh

@@ -5,6 +5,7 @@ SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
 REPO_ROOT="$(cd "$SCRIPT_DIR/../.." && pwd)"
 
 TARGET_ARCH="${TARGET_ARCH:-}"
+VERSION="${VERSION:-}"
 IMAGE_TAG=""
 IMAGE_TAR=""
 METRICS_IMAGE=""
@@ -25,6 +26,10 @@ Usage:
 
 Options:
   --arch <arch>       Target architecture: amd64|arm64
+  --version <vX.Y.Z>  Component version for image tags and the package name.
+                      Default: contents of the VERSION file at the repo root.
+                      Release builds (without --allow-dirty) require a git tag
+                      matching this version pointing at HEAD.
   --prometheus-image <tag>
                       Prometheus image tag to record and package.
   --prometheus-image-tar <path>
@@ -42,7 +47,7 @@ Options:
   -h, --help          Show help.
 
 Every package rebuilds runtime and metrics images from the current HEAD. Their
-tags are ours-rp-runtime-<arch>:<git8> and ours-rp-metrics-<arch>:<git8>.
+tags are panda-rpki-rp-<arch>:<version> and panda-rpki-metrics-<arch>:<version>.
 External runtime/metrics image tags and tar archives are intentionally rejected.
 USAGE
 }
@@ -72,17 +77,17 @@ safe_tag_name() {
 runtime_image_tag() {
   local arch="$1"
   local revision="$2"
-  printf 'ours-rp-runtime-%s:%s\n' "$arch" "$revision"
+  printf 'panda-rpki-rp-%s:%s\n' "$arch" "$revision"
 }
 
 metrics_image_tag() {
   local arch="$1"
   local revision="$2"
-  printf 'ours-rp-metrics-%s:%s\n' "$arch" "$revision"
+  printf 'panda-rpki-metrics-%s:%s\n' "$arch" "$revision"
 }
 
 default_package_prefix() {
-  printf 'ours-rp-installer-%s\n' "$1"
+  printf 'panda-rpki-rp-installer-%s\n' "$1"
 }
 
 default_host_data_dir() {
@@ -180,6 +185,10 @@ while [[ $# -gt 0 ]]; do
       }
       shift 2
       ;;
+    --version)
+      VERSION="$2"
+      shift 2
+      ;;
     --image|--image-tar|--metrics-image|--metrics-image-tar)
       echo "$1 is no longer supported: installer packages always rebuild runtime and metrics images from current HEAD" >&2
       exit 2
@@ -254,8 +263,36 @@ will be explicitly marked dirty.
 EOF
   exit 2
 fi
+
+# 版本来源：--version 覆盖，默认读仓库根 VERSION 文件。release 构建要求
+# HEAD 已打同版 git tag（自研组件打包前打 tag，与容器镜像 tag 保持一致）。
+if [[ -z "$VERSION" ]]; then
+  [[ -f "$REPO_ROOT/VERSION" ]] || {
+    echo "missing VERSION file at $REPO_ROOT/VERSION; pass --version <vX.Y.Z> or create it" >&2
+    exit 2
+  }
+  VERSION="$(tr -d '[:space:]' < "$REPO_ROOT/VERSION")"
+fi
+[[ "$VERSION" =~ ^v[0-9]+\.[0-9]+\.[0-9]+$ ]] || {
+  echo "invalid version '$VERSION' (expected vX.Y.Z)" >&2
+  exit 2
+}
+GIT_TAG_AT_HEAD=""
+if [[ "$ALLOW_DIRTY" != "1" ]]; then
+  GIT_TAG_AT_HEAD="$(git -C "$REPO_ROOT" tag --points-at HEAD | grep -Fx "$VERSION" || true)"
+  [[ -n "$GIT_TAG_AT_HEAD" ]] || {
+    cat >&2 <<EOF
+refusing to build a release installer: HEAD is not tagged with '$VERSION'.
+Create the tag first so the git tag matches the container image tag:
+  git -C $REPO_ROOT tag -a $VERSION -m "release $VERSION"
+Use --allow-dirty only for development verification (skips this check).
+EOF
+    exit 2
+  }
+fi
+
 TIMESTAMP="$(date -u +%Y%m%dT%H%M%SZ)"
-REVISION_TOKEN="$SOURCE_COMMIT_SHORT"
+REVISION_TOKEN="$VERSION"
 if [[ "$SOURCE_DIRTY" == "true" ]]; then
   REVISION_TOKEN="${REVISION_TOKEN}-dirty"
 fi
@@ -367,6 +404,8 @@ package_schema_version=2
 package_name=$package_name
 created_at_utc=$TIMESTAMP
 build_timestamp_utc=$TIMESTAMP
+version=$VERSION
+git_tag=${GIT_TAG_AT_HEAD:-}
 source_commit=$SOURCE_COMMIT_FULL
 source_commit_short=$SOURCE_COMMIT_SHORT
 source_dirty=$SOURCE_DIRTY
@@ -419,6 +458,8 @@ EOF
 
 cat > "$stage/PACKAGE-SUMMARY.txt" <<EOF
 package_name: $package_name
+version: $VERSION
+git_tag: ${GIT_TAG_AT_HEAD:-}
 source_commit: $SOURCE_COMMIT_FULL
 source_commit_short: $SOURCE_COMMIT_SHORT
 source_dirty: $SOURCE_DIRTY

scripts/docker/build_docker_metrics_image.sh

@@ -69,7 +69,7 @@ if [[ -z "$IMAGE_TAG" ]]; then
   if [[ -n "$(git -C "$REPO_ROOT" status --short 2>/dev/null)" ]]; then
     SOURCE_COMMIT_SHORT="${SOURCE_COMMIT_SHORT}-dirty"
   fi
-  IMAGE_TAG="ours-rp-metrics-${TARGET_ARCH}:${SOURCE_COMMIT_SHORT}"
+  IMAGE_TAG="panda-rpki-metrics-${TARGET_ARCH}:${SOURCE_COMMIT_SHORT}"
 fi
 
 if [[ "$dockerfile_explicit" != "1" ]]; then

scripts/docker/build_docker_runtime_image.sh

@@ -22,7 +22,7 @@ Usage:
 
 Options:
   --arch <arch>       Target architecture: amd64|arm64
-  --image <tag>       Docker image tag (default: ours-rp-runtime-<arch>:<git8>)
+  --image <tag>       Docker image tag (default: panda-rpki-rp-<arch>:<git8>)
   --out-dir <path>    Directory for docker save tar.gz (default: target/<arch>-docker)
   --dockerfile <path> Dockerfile path
   --builder <name>    buildx builder name
@@ -69,7 +69,7 @@ platform_for_arch() {
 default_image_tag() {
   local arch="$1"
   local commit_short="$2"
-  printf 'ours-rp-runtime-%s:%s\n' "$arch" "$commit_short"
+  printf 'panda-rpki-rp-%s:%s\n' "$arch" "$commit_short"
 }
 
 require_command() {

scripts/soak/portable-soak.env.example

@@ -66,6 +66,14 @@ PERIODIC_SNAPSHOT_MAX_DELTAS=100
 # 每隔多少轮执行一次 db_stats --exact。设置为空或 0 表示关闭 exact 统计。
 DB_STATS_EXACT_EVERY=3
 
+# 死 repo 传输层黑名单（#141，默认关）。设置状态文件路径即开启：
+# 连续 N 轮传输层抓取失败的 repo 会被拉黑并在后续轮次跳过其传输尝试；
+# daemon 会按间隔对黑名单条目做健康探针，恢复后自动摘除。
+# 留空表示完全关闭（行为与旧版本一致）。
+#DEAD_REPO_BLACKLIST_PATH=/var/lib/ours-rp/state/dead-repo-blacklist.json
+#DEAD_REPO_BLACKLIST_FAIL_THRESHOLD=3
+#DEAD_REPO_HEALTH_CHECK_INTERVAL_SECS=600
+
 # 是否开启 ours RP progress log。1 表示开启。
 RPKI_PROGRESS_LOG=1
 
@@ -94,6 +102,10 @@ ENABLE_CHILD_CERTIFICATE_VALIDATION_CACHE=0
 # 实验性 transport 预热：RPKI_EXTRA_ARGS="--enable-transport-request-prefetch --enable-roa-validation-cache"
 RPKI_EXTRA_ARGS=""
 
+# 所有出站 HTTP 请求（RRDP / TAL / TA / 死 repo 探针）的 User-Agent。
+# 不设置、为空或含非法 header 字符时回退默认 panda-rpki/0.2。
+#RPKI_HTTP_USER_AGENT=panda-rpki/0.2
+
 # 是否为每轮输出 timing profile 到 runs/run_xxxx/analyze/timing.json。
 # 性能 profile 或打点验证时设置为 1；普通 soak 建议保持 0，避免额外开销。
 RPKI_ANALYZE=0

scripts/soak/run_cache_ablation_experiment.sh

@@ -38,7 +38,7 @@ Environment:
   EXPERIMENT_RUN_ROOT          shared run root/state root; default PACKAGE_ROOT
   EXPERIMENT_DIR               experiment metadata output directory
   CASE_RUNS                    delta runs per case; default 10
-  EXPERIMENT_CASE_SET          default or cache-only; default runs the original 4-case matrix
+  EXPERIMENT_CASE_SET          default, cache-only, crypto-sig or sig-pp-compare; default runs the original 4-case matrix
   RUN_START_INTERVAL_SECS      fixed start cadence for all runs; default 600
   FIRST_RUN_DELAY_SECS         delay before the first scheduled run; default 0
   SNAPSHOT_EXTRA_ARGS          extra rpki args for snapshot warmup
@@ -95,7 +95,10 @@ case_count() {
   case "$EXPERIMENT_CASE_SET" in
     default) printf '%s' 4 ;;
     cache-only) printf '%s' 3 ;;
-    *) die "EXPERIMENT_CASE_SET must be default or cache-only: $EXPERIMENT_CASE_SET" ;;
+    crypto-sig) printf '%s' 4 ;;
+    sig-pp-compare) printf '%s' 2 ;;
+    baseline-combo) printf '%s' 2 ;;
+    *) die "EXPERIMENT_CASE_SET must be default, cache-only, crypto-sig or sig-pp-compare: $EXPERIMENT_CASE_SET" ;;
   esac
 }
 
@@ -108,6 +111,14 @@ case_id_for_index() {
     cache-only:1) printf '%s' "pp-only" ;;
     cache-only:2) printf '%s' "object-only" ;;
     cache-only:3) printf '%s' "pp-object-only" ;;
+    crypto-sig:1) printf '%s' "baseline" ;;
+    crypto-sig:2) printf '%s' "sig-only" ;;
+    crypto-sig:3) printf '%s' "pp-object" ;;
+    crypto-sig:4) printf '%s' "pp-object-sig" ;;
+    sig-pp-compare:1) printf '%s' "sig-only" ;;
+    sig-pp-compare:2) printf '%s' "pp-only" ;;
+    baseline-combo:1) printf '%s' "baseline" ;;
+    baseline-combo:2) printf '%s' "pp-object-sig" ;;
     *) die "unknown case index: $1 for set $EXPERIMENT_CASE_SET" ;;
   esac
 }
@@ -121,6 +132,14 @@ case_name_for_index() {
     cache-only:1) printf '%s' "pp-cache-only" ;;
     cache-only:2) printf '%s' "object-cache-only" ;;
     cache-only:3) printf '%s' "pp-cache-object-cache-only" ;;
+    crypto-sig:1) printf '%s' "all-cache-off" ;;
+    crypto-sig:2) printf '%s' "crypto-sig-cache-only" ;;
+    crypto-sig:3) printf '%s' "pp-cache-object-cache" ;;
+    crypto-sig:4) printf '%s' "pp-cache-object-crypto-sig-cache" ;;
+    sig-pp-compare:1) printf '%s' "crypto-sig-cache-only" ;;
+    sig-pp-compare:2) printf '%s' "pp-cache-only" ;;
+    baseline-combo:1) printf '%s' "all-cache-off" ;;
+    baseline-combo:2) printf '%s' "pp-cache-object-crypto-sig-cache" ;;
     *) die "unknown case index: $1 for set $EXPERIMENT_CASE_SET" ;;
   esac
 }
@@ -134,14 +153,22 @@ case_extra_args_for_index() {
     cache-only:1) printf '%s' "--enable-publication-point-validation-cache" ;;
     cache-only:2) printf '%s' "--enable-roa-validation-cache" ;;
     cache-only:3) printf '%s' "--enable-publication-point-validation-cache --enable-roa-validation-cache" ;;
+    crypto-sig:1) printf '%s' "" ;;
+    crypto-sig:2) printf '%s' "--enable-crypto-signature-cache" ;;
+    crypto-sig:3) printf '%s' "--enable-publication-point-validation-cache --enable-roa-validation-cache" ;;
+    crypto-sig:4) printf '%s' "--enable-publication-point-validation-cache --enable-roa-validation-cache --enable-crypto-signature-cache" ;;
+    sig-pp-compare:1) printf '%s' "--enable-crypto-signature-cache" ;;
+    sig-pp-compare:2) printf '%s' "--enable-publication-point-validation-cache" ;;
+    baseline-combo:1) printf '%s' "" ;;
+    baseline-combo:2) printf '%s' "--enable-publication-point-validation-cache --enable-roa-validation-cache --enable-crypto-signature-cache" ;;
     *) die "unknown case index: $1 for set $EXPERIMENT_CASE_SET" ;;
   esac
 }
 
 case_child_cert_cache_for_index() {
   case "$EXPERIMENT_CASE_SET:$1" in
-    default:4|cache-only:2|cache-only:3) printf '%s' "1" ;;
+    default:4|cache-only:2|cache-only:3|crypto-sig:3|crypto-sig:4|baseline-combo:2) printf '%s' "1" ;;
-    default:1|default:2|default:3|cache-only:1) printf '%s' "0" ;;
+    default:1|default:2|default:3|cache-only:1|crypto-sig:1|crypto-sig:2|sig-pp-compare:1|sig-pp-compare:2|baseline-combo:1) printf '%s' "0" ;;
     *) die "unknown case index: $1 for set $EXPERIMENT_CASE_SET" ;;
   esac
 }
@@ -163,6 +190,23 @@ elif case_set == "cache-only":
         {"caseId": "object-only", "caseName": "object-cache-only", "extraArgs": "--enable-roa-validation-cache", "enableChildCertificateValidationCache": True},
         {"caseId": "pp-object-only", "caseName": "pp-cache-object-cache-only", "extraArgs": "--enable-publication-point-validation-cache --enable-roa-validation-cache", "enableChildCertificateValidationCache": True},
     ]
+elif case_set == "crypto-sig":
+    cases = [
+        {"caseId": "baseline", "caseName": "all-cache-off", "extraArgs": "", "enableChildCertificateValidationCache": False},
+        {"caseId": "sig-only", "caseName": "crypto-sig-cache-only", "extraArgs": "--enable-crypto-signature-cache", "enableChildCertificateValidationCache": False},
+        {"caseId": "pp-object", "caseName": "pp-cache-object-cache", "extraArgs": "--enable-publication-point-validation-cache --enable-roa-validation-cache", "enableChildCertificateValidationCache": True},
+        {"caseId": "pp-object-sig", "caseName": "pp-cache-object-crypto-sig-cache", "extraArgs": "--enable-publication-point-validation-cache --enable-roa-validation-cache --enable-crypto-signature-cache", "enableChildCertificateValidationCache": True},
+    ]
+elif case_set == "sig-pp-compare":
+    cases = [
+        {"caseId": "sig-only", "caseName": "crypto-sig-cache-only", "extraArgs": "--enable-crypto-signature-cache", "enableChildCertificateValidationCache": False},
+        {"caseId": "pp-only", "caseName": "pp-cache-only", "extraArgs": "--enable-publication-point-validation-cache", "enableChildCertificateValidationCache": False},
+    ]
+elif case_set == "baseline-combo":
+    cases = [
+        {"caseId": "baseline", "caseName": "all-cache-off", "extraArgs": "", "enableChildCertificateValidationCache": False},
+        {"caseId": "pp-object-sig", "caseName": "pp-cache-object-crypto-sig-cache", "extraArgs": "--enable-publication-point-validation-cache --enable-roa-validation-cache --enable-crypto-signature-cache", "enableChildCertificateValidationCache": True},
+    ]
 else:
     raise SystemExit(f"unknown case set: {case_set}")
 print(json.dumps(cases, ensure_ascii=False, indent=8))
@@ -345,6 +389,7 @@ else:
         "publication_point_cache_index_load",
         "publication_point_cache_index_refresh",
         "roa_validation_cache",
+        "crypto_signature_cache_observe",
     ]:
         if key in stage:
             record[key] = stage.get(key)
@@ -360,6 +405,11 @@ else:
         "child_certificate_cache_miss_not_found",
         "fresh_publication_points",
         "fresh_manifest_files_total",
+        "crypto_signature_cache_observe_calls",
+        "crypto_signature_cache_observe_would_hit",
+        "crypto_signature_cache_observe_new_keys",
+        "crypto_signature_cache_verify_executed",
+        "crypto_signature_cache_verify_skipped",
     ]
     record["cacheCounts"] = {k: analysis_counts.get(k) for k in interesting if k in analysis_counts}
     record["repoSyncStats"] = summary.get("repoSyncStats")

scripts/soak/run_soak.sh

@@ -36,6 +36,10 @@ DISABLE_COMPETING_RPS="${DISABLE_COMPETING_RPS:-1}"
 ENABLE_CHILD_CERTIFICATE_VALIDATION_CACHE="${ENABLE_CHILD_CERTIFICATE_VALIDATION_CACHE:-0}"
 RPKI_EXTRA_ARGS="${RPKI_EXTRA_ARGS:-}"
 RPKI_ANALYZE="${RPKI_ANALYZE:-0}"
+CUSTOM_TAL_PATH="${CUSTOM_TAL_PATH:-$PACKAGE_ROOT/custom-fixtures/tal/custom.tal}"
+CUSTOM_TA_PATH="${CUSTOM_TA_PATH:-$PACKAGE_ROOT/custom-fixtures/ta/custom-ta.cer}"
+CUSTOM_TAL_URI="${CUSTOM_TAL_URI:-}"
+HTTP_ROOT_CERT_PATHS="${HTTP_ROOT_CERT_PATHS:-}"
 
 BIN_DIR="${BIN_DIR:-$PACKAGE_ROOT/bin}"
 FIXTURE_DIR="${FIXTURE_DIR:-$PACKAGE_ROOT/fixtures}"
@@ -119,10 +123,10 @@ validate_rsync_scope() {
 
 validate_tal_input_mode() {
   case "$TAL_INPUT_MODE" in
-    file-with-ta|file-live-ta|url)
+    file-with-ta|file-live-ta|custom-file-with-ta|url)
       ;;
     *)
-      die "TAL_INPUT_MODE must be file-with-ta, file-live-ta or url: $TAL_INPUT_MODE"
+      die "TAL_INPUT_MODE must be file-with-ta, file-live-ta, custom-file-with-ta or url: $TAL_INPUT_MODE"
       ;;
   esac
 }
@@ -156,14 +160,46 @@ parse_rirs() {
       afrinic|apnic|arin|lacnic|ripe)
         RIR_LIST+=("$normalized")
         ;;
+      custom)
+        [[ "$TAL_INPUT_MODE" == "custom-file-with-ta" ]] \
+          || die "RIRS=custom requires TAL_INPUT_MODE=custom-file-with-ta"
+        RIR_LIST+=("$normalized")
+        ;;
       *)
-        die "invalid RIRS entry: $raw_token; allowed: afrinic,apnic,arin,lacnic,ripe"
+        die "invalid RIRS entry: $raw_token; allowed: afrinic,apnic,arin,lacnic,ripe,custom (custom mode only)"
         ;;
     esac
   done
   [[ "${#RIR_LIST[@]}" -gt 0 ]] || die "RIRS must contain at least one RIR"
 }
 
+validate_custom_inputs() {
+  local root_cert_path
+  local -a root_cert_paths=()
+
+  if [[ "$TAL_INPUT_MODE" == "custom-file-with-ta" ]]; then
+    [[ "${#RIR_LIST[@]}" -eq 1 && "${RIR_LIST[0]}" == "custom" ]] \
+      || die "custom-file-with-ta requires exactly RIRS=custom"
+    [[ -s "$CUSTOM_TAL_PATH" ]] \
+      || die "missing custom TAL: $CUSTOM_TAL_PATH"
+    [[ -s "$CUSTOM_TA_PATH" ]] \
+      || die "missing custom TA DER: $CUSTOM_TA_PATH"
+    [[ -n "$CUSTOM_TAL_URI" ]] \
+      || die "CUSTOM_TAL_URI is required in custom-file-with-ta mode"
+  elif [[ "$RIRS" == *custom* ]]; then
+    die "RIRS=custom requires TAL_INPUT_MODE=custom-file-with-ta"
+  fi
+
+  if [[ -n "$HTTP_ROOT_CERT_PATHS" ]]; then
+    # shellcheck disable=SC2206
+    root_cert_paths=( $HTTP_ROOT_CERT_PATHS )
+    for root_cert_path in "${root_cert_paths[@]}"; do
+      [[ -s "$root_cert_path" ]] \
+        || die "missing HTTP root certificate: $root_cert_path"
+    done
+  fi
+}
+
 tal_file_for_rir() {
   case "$1" in
     afrinic) printf '%s' "$FIXTURE_DIR/tal/afrinic.tal" ;;
@@ -171,6 +207,7 @@ tal_file_for_rir() {
     arin) printf '%s' "$FIXTURE_DIR/tal/arin.tal" ;;
     lacnic) printf '%s' "$FIXTURE_DIR/tal/lacnic.tal" ;;
     ripe) printf '%s' "$FIXTURE_DIR/tal/ripe-ncc.tal" ;;
+    custom) printf '%s' "$CUSTOM_TAL_PATH" ;;
     *) die "unknown RIR: $1" ;;
   esac
 }
@@ -182,6 +219,7 @@ ta_file_for_rir() {
     arin) printf '%s' "$FIXTURE_DIR/ta/arin-ta.cer" ;;
     lacnic) printf '%s' "$FIXTURE_DIR/ta/lacnic-ta.cer" ;;
     ripe) printf '%s' "$FIXTURE_DIR/ta/ripe-ncc-ta.cer" ;;
+    custom) printf '%s' "$CUSTOM_TA_PATH" ;;
     *) die "unknown RIR: $1" ;;
   esac
 }
@@ -193,6 +231,7 @@ tal_url_for_rir() {
     arin) printf '%s' "https://www.arin.net/resources/manage/rpki/arin.tal" ;;
     lacnic) printf '%s' "https://www.lacnic.net/innovaportal/file/4983/1/lacnic.tal" ;;
     ripe) printf '%s' "https://tal.rpki.ripe.net/ripe-ncc.tal" ;;
+    custom) printf '%s' "$CUSTOM_TAL_URI" ;;
     *) die "unknown RIR: $1" ;;
   esac
 }
@@ -1058,6 +1097,17 @@ build_child_args() {
     CHILD_ARGS+=(--rsync-mirror-root "$TMP_DIR/rsync-mirror-{run_id}")
   fi
 
+  local root_cert_path
+  local -a root_cert_paths=()
+  if [[ -n "$HTTP_ROOT_CERT_PATHS" ]]; then
+    # HTTPS trust roots for RRDP/TAL fetches (independent of the RPKI TA).
+    # shellcheck disable=SC2206
+    root_cert_paths=( $HTTP_ROOT_CERT_PATHS )
+    for root_cert_path in "${root_cert_paths[@]}"; do
+      CHILD_ARGS+=(--http-root-cert "$root_cert_path")
+    done
+  fi
+
   CHILD_ARGS+=(
     --parallel-phase2-ready-batch-size 256
     --parallel-phase2-ready-batch-wall-time-budget-ms 100
@@ -1331,6 +1381,17 @@ run_one_round() {
       daemon_args+=(--db-stats-exact-every "$DB_STATS_EXACT_EVERY")
     fi
   fi
+  # Dead-repo transport blacklist (#141): opt-in via env; the daemon forwards
+  # the enable flag and threshold to the child unless already present there.
+  if [[ -n "${DEAD_REPO_BLACKLIST_PATH:-}" ]]; then
+    daemon_args+=(--dead-repo-blacklist "$DEAD_REPO_BLACKLIST_PATH")
+    if [[ -n "${DEAD_REPO_BLACKLIST_FAIL_THRESHOLD:-}" ]]; then
+      daemon_args+=(--dead-repo-blacklist-fail-threshold "$DEAD_REPO_BLACKLIST_FAIL_THRESHOLD")
+    fi
+    if [[ -n "${DEAD_REPO_HEALTH_CHECK_INTERVAL_SECS:-}" ]]; then
+      daemon_args+=(--dead-repo-health-check-interval-secs "$DEAD_REPO_HEALTH_CHECK_INTERVAL_SECS")
+    fi
+  fi
 
   set +e
   env \
@@ -1406,6 +1467,7 @@ main() {
     validate_positive_int "DB_STATS_EXACT_EVERY" "$DB_STATS_EXACT_EVERY"
   fi
   parse_rirs
+  validate_custom_inputs
   [[ -x "$RPKI_BIN" ]] || die "missing executable: $RPKI_BIN"
   [[ -x "$RPKI_DAEMON_BIN" ]] || die "missing executable: $RPKI_DAEMON_BIN"
 

scripts/verification/README.md

@@ -29,7 +29,7 @@
 - `vrps.csv`、`vaps.csv`
 - `stage-timing.json`、`validation-events.jsonl`
 - `validation-contract.json`、`verification-meta.json`
-- `compare/ccr-state-compare.json`、`compare/ccr-state-compare.md`
+- `compare/ccr-state-digest.json`、`compare/ccr-state-digest.md`
 - `stdout.log`、`stderr.log`、`process-time.txt`、`wrapper-timing.json`
 
 通过条件是 compare JSON 中 `stateDigestMatch=true`。`producedAt` 不参与比较。

scripts/verification/observe_remote231_verification_only.sh

@@ -0,0 +1,563 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+# Observe a Docker-managed soak from the host without joining its compose project.
+# The verifier binary must be copied byte-for-byte from the running runtime image;
+# validation-contract.json rejects a different binary SHA-256.
+
+usage() {
+  cat <<'USAGE'
+Usage:
+  observe_remote231_verification_only.sh \
+    --runs-root <normal_runs_dir> \
+    --state-root <normal_state_dir> \
+    --rpki-bin <host_side_rpki> \
+    --work-root <observer_root> \
+    --webhook-url-file <0600_file> \
+    [--target-runs 20] [--delay-secs 30] [--poll-secs 5] \
+    [--verification-timeout-secs 480] [--normal-interval-secs 600] \
+    [--normal-run-safety-secs 60] [--verification-object-workers 4] \
+    [--verification-child-path <path>]
+
+This is a host-side, finite observer. On its first start it records the latest
+already-terminal normal run and observes only later terminal runs. For each one
+it waits DELAY seconds, then uses a non-blocking state lock. It never waits for
+or blocks the Docker soak. Successful normal runs are verified with a copied
+host-side rpki binary; failed, stale, or lock-busy runs are recorded as skipped.
+
+The verification timeout is additionally bounded by the source run completion
+time plus NORMAL_INTERVAL_SECS minus NORMAL_RUN_SAFETY_SECS. This preserves a
+normal-run safety window even after an observer restart or delayed scan.
+
+Only compact JSON results are retained under WORK_ROOT/results/. The large
+verification output and any retained scratch DB are always deleted afterwards.
+
+When --verification-child-path is supplied, it replaces PATH only for the
+verification binary. Passing a directory without rsync (for example
+/nonexistent) is an adversarial acceptance check: an accidental SystemRsync
+fallback must fail instead of accessing the network.
+USAGE
+}
+
+RUNS_ROOT=""
+STATE_ROOT=""
+RPKI_BIN=""
+WORK_ROOT=""
+WEBHOOK_URL_FILE=""
+TARGET_RUNS=20
+DELAY_SECS=30
+POLL_SECS=5
+VERIFICATION_TIMEOUT_SECS=480
+NORMAL_INTERVAL_SECS=600
+NORMAL_RUN_SAFETY_SECS=60
+VERIFICATION_OBJECT_WORKERS=4
+VERIFICATION_CHILD_PATH=""
+
+while [[ $# -gt 0 ]]; do
+  case "$1" in
+    --runs-root) RUNS_ROOT="${2:?missing value}"; shift 2 ;;
+    --state-root) STATE_ROOT="${2:?missing value}"; shift 2 ;;
+    --rpki-bin) RPKI_BIN="${2:?missing value}"; shift 2 ;;
+    --work-root) WORK_ROOT="${2:?missing value}"; shift 2 ;;
+    --webhook-url-file) WEBHOOK_URL_FILE="${2:?missing value}"; shift 2 ;;
+    --target-runs) TARGET_RUNS="${2:?missing value}"; shift 2 ;;
+    --delay-secs) DELAY_SECS="${2:?missing value}"; shift 2 ;;
+    --poll-secs) POLL_SECS="${2:?missing value}"; shift 2 ;;
+    --verification-timeout-secs) VERIFICATION_TIMEOUT_SECS="${2:?missing value}"; shift 2 ;;
+    --normal-interval-secs) NORMAL_INTERVAL_SECS="${2:?missing value}"; shift 2 ;;
+    --normal-run-safety-secs) NORMAL_RUN_SAFETY_SECS="${2:?missing value}"; shift 2 ;;
+    --verification-object-workers) VERIFICATION_OBJECT_WORKERS="${2:?missing value}"; shift 2 ;;
+    --verification-child-path) VERIFICATION_CHILD_PATH="${2:?missing value}"; shift 2 ;;
+    -h|--help) usage; exit 0 ;;
+    *) echo "unknown argument: $1" >&2; usage >&2; exit 2 ;;
+  esac
+done
+
+for value_name in RUNS_ROOT STATE_ROOT RPKI_BIN WORK_ROOT WEBHOOK_URL_FILE; do
+  [[ -n "${!value_name}" ]] || { echo "$value_name is required" >&2; exit 2; }
+done
+for integer_name in TARGET_RUNS DELAY_SECS POLL_SECS VERIFICATION_TIMEOUT_SECS NORMAL_INTERVAL_SECS VERIFICATION_OBJECT_WORKERS; do
+  [[ "${!integer_name}" =~ ^[0-9]+$ ]] && [[ "${!integer_name}" != 0 ]] || {
+    echo "$integer_name must be a positive integer" >&2
+    exit 2
+  }
+done
+[[ "$NORMAL_RUN_SAFETY_SECS" =~ ^[0-9]+$ ]] || {
+  echo "NORMAL_RUN_SAFETY_SECS must be a non-negative integer" >&2
+  exit 2
+}
+(( NORMAL_RUN_SAFETY_SECS < NORMAL_INTERVAL_SECS )) || {
+  echo "NORMAL_RUN_SAFETY_SECS must be smaller than NORMAL_INTERVAL_SECS" >&2
+  exit 2
+}
+[[ -d "$RUNS_ROOT" ]] || { echo "runs root is not a directory: $RUNS_ROOT" >&2; exit 2; }
+[[ -d "$STATE_ROOT" ]] || { echo "state root is not a directory: $STATE_ROOT" >&2; exit 2; }
+[[ -x "$RPKI_BIN" ]] || { echo "rpki binary is not executable: $RPKI_BIN" >&2; exit 2; }
+[[ -f "$WEBHOOK_URL_FILE" ]] || { echo "webhook URL file is absent: $WEBHOOK_URL_FILE" >&2; exit 2; }
+for required_command in curl flock ionice jq nice python3 sha256sum timeout; do
+  command -v "$required_command" >/dev/null || {
+    echo "missing required command: $required_command" >&2
+    exit 2
+  }
+done
+
+mkdir -p "$WORK_ROOT/results" "$WORK_ROOT/scratch"
+STATE_FILE="$WORK_ROOT/observer-state.json"
+OBSERVER_LOCK="$WORK_ROOT/observer.lock"
+RESULTS_DIR="$WORK_ROOT/results"
+SCRATCH_ROOT="$WORK_ROOT/scratch"
+WEBHOOK_URL="$(tr -d '\r\n' < "$WEBHOOK_URL_FILE")"
+[[ "$WEBHOOK_URL" == https://open.feishu.cn/open-apis/bot/v2/hook/* ]] || {
+  echo "webhook URL does not have the expected Feishu incoming-webhook form" >&2
+  exit 2
+}
+
+now_utc() { date -u +%Y-%m-%dT%H:%M:%SZ; }
+now_ms() { python3 -c 'import time; print(time.time_ns() // 1_000_000)'; }
+
+remaining_verification_window_secs() {
+  local completed_at="$1"
+  python3 - "$completed_at" "$NORMAL_INTERVAL_SECS" "$NORMAL_RUN_SAFETY_SECS" <<'PY'
+import datetime
+import sys
+import time
+
+completed = datetime.datetime.fromisoformat(sys.argv[1].replace("Z", "+00:00"))
+interval = int(sys.argv[2])
+safety = int(sys.argv[3])
+print(int(completed.timestamp()) + interval - safety - int(time.time()))
+PY
+}
+
+atomic_replace() {
+  local destination="$1"
+  local temporary="${destination}.tmp.$$"
+  cat > "$temporary"
+  mv "$temporary" "$destination"
+}
+
+terminal_run_ids() {
+  local run_dir meta
+  for run_dir in "$RUNS_ROOT"/run_[0-9]*; do
+    [[ -d "$run_dir" ]] || continue
+    meta="$run_dir/run-meta.json"
+    [[ -f "$meta" ]] || continue
+    jq -e '(.completed_at_rfc3339_utc? // .completedAt? // null) != null' "$meta" >/dev/null 2>&1 || continue
+    basename "$run_dir"
+  done | sort -t_ -k2,2n
+}
+
+run_number() {
+  local run_id="$1"
+  printf '%d\n' "$((10#${run_id#run_}))"
+}
+
+latest_terminal_run_id() {
+  terminal_run_ids | tail -n 1
+}
+
+result_path() {
+  printf '%s/%s.json\n' "$RESULTS_DIR" "$1"
+}
+
+read_json_or_empty() {
+  local path="$1"
+  if [[ -f "$path" ]] && jq -e . "$path" >/dev/null 2>&1; then
+    jq -c . "$path"
+  else
+    printf '{}'
+  fi
+}
+
+update_cumulative() {
+  local observed normal_success passes mismatches errors skipped pending latest
+  local -a result_files
+  shopt -s nullglob
+  result_files=("$RESULTS_DIR"/run_*.json)
+  shopt -u nullglob
+  observed="${#result_files[@]}"
+  if (( observed == 0 )); then
+    normal_success=0
+    passes=0
+    mismatches=0
+    errors=0
+    skipped=0
+    pending=0
+  else
+    normal_success="$(jq -r 'select(.normalRun.status == "success") | 1' "${result_files[@]}" | wc -l)"
+    passes="$(jq -r 'select(.verification.status == "pass") | 1' "${result_files[@]}" | wc -l)"
+    mismatches="$(jq -r 'select(.verification.status == "mismatch") | 1' "${result_files[@]}" | wc -l)"
+    errors="$(jq -r 'select(.verification.status == "error") | 1' "${result_files[@]}" | wc -l)"
+    skipped="$(jq -r 'select(.verification.status | startswith("skipped_")) | 1' "${result_files[@]}" | wc -l)"
+    pending="$(jq -r 'select(.notification.status != "sent") | 1' "${result_files[@]}" | wc -l)"
+  fi
+  latest="$(find "$RESULTS_DIR" -maxdepth 1 -type f -name 'run_*.json' -printf '%f\n' | sed 's/.json$//' | sort -t_ -k2,2n | tail -n 1)"
+  jq -n \
+    --arg updated_at "$(now_utc)" \
+    --arg initial_run_id "$(jq -r '.initialLatestTerminalRunId' "$STATE_FILE")" \
+    --arg latest_run_id "$latest" \
+    --argjson target_runs "$TARGET_RUNS" \
+    --argjson observed_runs "$observed" \
+    --argjson normal_success_runs "$normal_success" \
+    --argjson verification_passes "$passes" \
+    --argjson verification_mismatches "$mismatches" \
+    --argjson verification_errors "$errors" \
+    --argjson verification_skipped "$skipped" \
+    --argjson notification_pending "$pending" \
+    '{schemaVersion: 1, updatedAt: $updated_at, initialLatestTerminalRunId: $initial_run_id,
+      latestProcessedRunId: $latest_run_id, targetRuns: $target_runs, observedRuns: $observed_runs,
+      normalSuccessRuns: $normal_success_runs, verificationPasses: $verification_passes,
+      verificationMismatches: $verification_mismatches, verificationErrors: $verification_errors,
+      verificationSkipped: $verification_skipped, notificationPending: $notification_pending}' \
+    | atomic_replace "$RESULTS_DIR/cumulative.json"
+}
+
+notify_result() {
+  local path="$1" run_id status observed target passes mismatches errors skipped analysis_available top_phase message payload response
+  run_id="$(jq -r '.runId' "$path")"
+  status="$(jq -r '.verification.status' "$path")"
+  observed="$(jq -r '.observedRuns' "$RESULTS_DIR/cumulative.json")"
+  target="$(jq -r '.targetRuns' "$RESULTS_DIR/cumulative.json")"
+  passes="$(jq -r '.verificationPasses' "$RESULTS_DIR/cumulative.json")"
+  mismatches="$(jq -r '.verificationMismatches' "$RESULTS_DIR/cumulative.json")"
+  errors="$(jq -r '.verificationErrors' "$RESULTS_DIR/cumulative.json")"
+  skipped="$(jq -r '.verificationSkipped' "$RESULTS_DIR/cumulative.json")"
+  analysis_available="$(jq -r '.verification.analysis.available // false' "$path")"
+  top_phase="$(jq -r '(.verification.analysis.phaseTotals // [])[0] | if . == null then "n/a" else "\(.name)=\(.totalMs)ms" end' "$path")"
+  message="From codex: #137 远端231 verification-only 旁路复核\n${run_id}: ${status}\n"
+  message+="CCR state digest match=$(jq -r 'if .verification.comparison | has("stateDigestMatch") then (.verification.comparison.stateDigestMatch | tostring) else "n/a" end' "$path"); fallback replay match=$(jq -r 'if .verification.comparison.fallbackReplaySelection | has("matches") then (.verification.comparison.fallbackReplaySelection.matches | tostring) else "n/a" end' "$path"); verificationProcessMs=$(jq -r '.verification.timing.verificationProcessMs // .verification.wallMs // "n/a"' "$path"); preValidationPreparationMs=$(jq -r '.verification.timing.preValidationPreparationMs // "n/a"' "$path"); lockWaitMs=$(jq -r '.verification.lockWaitMs // "n/a"' "$path")\n"
+  message+="analysisAvailable=${analysis_available}; topPhase=${top_phase}\n"
+  message+="累计 ${observed}/${target}: pass=${passes}, mismatch=${mismatches}, error=${errors}, skipped=${skipped}\n"
+  message+="结果: ${path}"
+  payload="$(jq -n --arg text "$message" '{msg_type: "text", content: {text: $text}}')"
+  if response="$(curl --fail-with-body --silent --show-error --connect-timeout 10 --max-time 30 \
+      -H 'Content-Type: application/json' --data "$payload" "$WEBHOOK_URL" 2>&1)"; then
+    if jq -e '((.code // .StatusCode // 0) == 0)' <<< "$response" >/dev/null 2>&1; then
+      jq --arg sent_at "$(now_utc)" --arg response "$response" \
+        '.notification = {status: "sent", sentAt: $sent_at, response: $response}' "$path" \
+        | atomic_replace "$path"
+      return 0
+    fi
+    response="Feishu rejected payload: $response"
+  fi
+  jq --arg attempted_at "$(now_utc)" --arg error "$response" \
+    '.notification = {status: "pending", lastAttemptAt: $attempted_at, lastError: $error}' "$path" \
+    | atomic_replace "$path"
+  return 1
+}
+
+deliver_pending_notifications() {
+  local path
+  for path in "$RESULTS_DIR"/run_*.json; do
+    [[ -f "$path" ]] || continue
+    [[ "$(jq -r '.notification.status // "pending"' "$path")" == "sent" ]] && continue
+    update_cumulative
+    notify_result "$path" || true
+  done
+  update_cumulative
+}
+
+write_result() {
+  local run_id="$1" normal_json="$2" verification_json="$3" result
+  result="$(result_path "$run_id")"
+  jq -n \
+    --arg recorded_at "$(now_utc)" \
+    --arg run_id "$run_id" \
+    --argjson normal_run "$normal_json" \
+    --argjson verification "$verification_json" \
+    '{schemaVersion: 1, recordedAt: $recorded_at, runId: $run_id, normalRun: $normal_run,
+      verification: $verification, notification: {status: "pending"}}' \
+    | atomic_replace "$result"
+  update_cumulative
+  notify_result "$result" || true
+  update_cumulative
+}
+
+normal_run_summary() {
+  local run_dir="$1"
+  if [[ ! -f "$run_dir/run-meta.json" ]]; then
+    printf '{"status":"invalid-meta"}'
+    return
+  fi
+  jq -cn --slurpfile meta "$run_dir/run-meta.json" --slurpfile summary "$run_dir/run-summary.json" \
+    '{status: ($meta[0].status // "unknown"), syncMode: ($meta[0].sync_mode // $meta[0].syncMode // "unknown"),
+      runIndex: ($meta[0].run_index // $meta[0].runIndex // null),
+      completedAt: ($meta[0].completed_at_rfc3339_utc // $meta[0].completedAt // null),
+      wallMs: ($summary[0].wall_ms // $summary[0].wallMs // null),
+      exitCode: ($summary[0].exit_code // $summary[0].exitCode // $meta[0].daemon_exit_code // null)}' 2>/dev/null \
+    || printf '{"status":"invalid-meta"}'
+}
+
+process_terminal_run() {
+  local run_id="$1" run_dir="$RUNS_ROOT/$run_id" normal_json status temporary_dir state_lock_fd lock_started_ms lock_acquired_ms
+  local binding_run_id binary_sha source_sha run_started_ms run_finished_ms exit_code comparison_json comparison_state_digest_match fallback_replay_comparison_json fallback_replay_selection_match verification_meta_json stage_timing_json analysis_json analysis_summary_json error_tail max_rss_kb verification_status verification_json timeout_secs completed_at
+  local -a verification_child_exec
+  local observer_started_ms delay_finished_ms preparation_started_ms binding_started_ms binding_finished_ms binary_check_started_ms binary_check_finished_ms window_started_ms window_finished_ms scratch_cleanup_started_ms scratch_cleanup_finished_ms post_validation_started_ms post_validation_collection_finished_ms result_cleanup_started_ms result_cleanup_finished_ms
+  normal_json="$(normal_run_summary "$run_dir")"
+  status="$(jq -r '.status' <<< "$normal_json")"
+  observer_started_ms="$(now_ms)"
+  sleep "$DELAY_SECS"
+  delay_finished_ms="$(now_ms)"
+  if [[ "$status" != "success" ]]; then
+    write_result "$run_id" "$normal_json" '{"status":"skipped_normal_run_not_success"}'
+    return
+  fi
+
+  preparation_started_ms="$delay_finished_ms"
+  mkdir -p "$STATE_ROOT/locks"
+  exec {state_lock_fd}>"$STATE_ROOT/locks/state-execution.lock"
+  lock_started_ms="$(now_ms)"
+  if ! flock -n "$state_lock_fd"; then
+    exec {state_lock_fd}>&-
+    write_result "$run_id" "$normal_json" '{"status":"skipped_state_lock_busy"}'
+    return
+  fi
+  lock_acquired_ms="$(now_ms)"
+  binding_started_ms="$(now_ms)"
+  binding_run_id="$(jq -r '.currentRunId // empty' "$STATE_ROOT/meta/current-state-binding.json" 2>/dev/null || true)"
+  binding_finished_ms="$(now_ms)"
+  if [[ "$binding_run_id" != "$run_id" ]]; then
+    flock -u "$state_lock_fd"
+    exec {state_lock_fd}>&-
+    verification_json="$(jq -n --arg bound_run "$binding_run_id" --argjson lock_wait_ms "$((lock_acquired_ms - lock_started_ms))" \
+      '{status: "skipped_source_no_longer_current", boundRunId: $bound_run, lockWaitMs: $lock_wait_ms}')"
+    write_result "$run_id" "$normal_json" "$verification_json"
+    return
+  fi
+  binary_check_started_ms="$(now_ms)"
+  source_sha="$(jq -r '.binarySha256 // empty' "$run_dir/validation-contract.json" 2>/dev/null || true)"
+  binary_sha="$(sha256sum "$RPKI_BIN" | awk '{print $1}')"
+  binary_check_finished_ms="$(now_ms)"
+  if [[ -z "$source_sha" || "$source_sha" != "$binary_sha" ]]; then
+    flock -u "$state_lock_fd"
+    exec {state_lock_fd}>&-
+    verification_json="$(jq -n --arg expected "$source_sha" --arg actual "$binary_sha" --argjson lock_wait_ms "$((lock_acquired_ms - lock_started_ms))" \
+      '{status: "error_binary_sha256_mismatch", expectedBinarySha256: $expected, actualBinarySha256: $actual, lockWaitMs: $lock_wait_ms}')"
+    write_result "$run_id" "$normal_json" "$verification_json"
+    return
+  fi
+
+  window_started_ms="$(now_ms)"
+  completed_at="$(jq -r '.completedAt // empty' <<< "$normal_json")"
+  timeout_secs="$(remaining_verification_window_secs "$completed_at")"
+  window_finished_ms="$(now_ms)"
+  if (( timeout_secs <= 0 )); then
+    flock -u "$state_lock_fd"
+    exec {state_lock_fd}>&-
+    verification_json="$(jq -n --arg completed_at "$completed_at" --argjson lock_wait_ms "$((lock_acquired_ms - lock_started_ms))" \
+      '{status: "skipped_insufficient_interval", sourceCompletedAt: $completed_at, lockWaitMs: $lock_wait_ms}')"
+    write_result "$run_id" "$normal_json" "$verification_json"
+    return
+  fi
+  if (( timeout_secs > VERIFICATION_TIMEOUT_SECS )); then
+    timeout_secs="$VERIFICATION_TIMEOUT_SECS"
+  fi
+
+  temporary_dir="$SCRATCH_ROOT/$run_id"
+  scratch_cleanup_started_ms="$(now_ms)"
+  rm -rf "$temporary_dir"
+  scratch_cleanup_finished_ms="$(now_ms)"
+  run_started_ms="$(now_ms)"
+  verification_child_exec=()
+  if [[ -n "$VERIFICATION_CHILD_PATH" ]]; then
+    verification_child_exec=(/usr/bin/env "PATH=$VERIFICATION_CHILD_PATH")
+  fi
+  set +e
+  RPKI_STATE_EXECUTION_LOCK_HELD=1 /usr/bin/time -v -o "$temporary_dir.process-time" -- \
+    /usr/bin/timeout --foreground --signal=TERM --kill-after=30s "$timeout_secs" \
+      /usr/bin/ionice -c2 -n0 "${verification_child_exec[@]}" "$RPKI_BIN" \
+        --verification-only --verification-source-run "$run_dir" \
+        --verification-source-state "$STATE_ROOT" --verification-out "$temporary_dir" \
+        --analyze --analysis-out "$temporary_dir/analyze" \
+        --parallel-phase2-object-workers "$VERIFICATION_OBJECT_WORKERS" \
+        >"$temporary_dir.stdout" 2>"$temporary_dir.stderr"
+  exit_code=$?
+  set -e
+  run_finished_ms="$(now_ms)"
+  post_validation_started_ms="$run_finished_ms"
+  flock -u "$state_lock_fd"
+  exec {state_lock_fd}>&-
+
+  comparison_json="$(read_json_or_empty "$temporary_dir/compare/ccr-state-digest.json")"
+  fallback_replay_comparison_json="$(read_json_or_empty "$temporary_dir/compare/fallback-replay-selection.json")"
+  verification_meta_json="$(read_json_or_empty "$temporary_dir/verification-meta.json")"
+  stage_timing_json="$(read_json_or_empty "$temporary_dir/stage-timing.json")"
+  analysis_json="$(read_json_or_empty "$temporary_dir/analyze/timing.json")"
+  analysis_summary_json="$(jq -cn --argjson stage "$stage_timing_json" --argjson analysis "$analysis_json" '
+    {enabled: true,
+     available: (($analysis | type) == "object" and ($analysis | length) > 0),
+     stageTiming: {
+       totalMs: ($stage.total_ms // null),
+       validationMs: ($stage.validation_ms // null),
+       repoSyncMsTotal: ($stage.repo_sync_ms_total // null),
+       publicationPointRepoSyncMsTotal: ($stage.publication_point_repo_sync_ms_total // null),
+       rrdpDownloadMsTotal: ($stage.rrdp_download_ms_total // null),
+       rsyncDownloadMsTotal: ($stage.rsync_download_ms_total // null),
+       reportBuildMs: ($stage.report_build_ms // null),
+       reportWriteMs: ($stage.report_write_ms // null),
+       ccrBuildMs: ($stage.ccr_build_ms // null),
+       ccrWriteMs: ($stage.ccr_write_ms // null),
+       cirTotalMs: ($stage.cir_total_ms // null)
+     },
+     phaseTotals: (($analysis.phases // {})
+       | to_entries
+       | map({name: .key, count: (.value.count // 0), totalMs: (((.value.total_nanos // 0) / 1000000) | floor)})
+       | sort_by(-.totalMs)
+       | .[0:20]),
+     topPublicationPoints: (($analysis.top_publication_points // [])[0:10]),
+     topPublicationPointSteps: (($analysis.top_publication_point_steps // [])[0:10]),
+     topRrdpRepos: (($analysis.top_rrdp_repos // [])[0:10]),
+     topRrdpRepoSteps: (($analysis.top_rrdp_repo_steps // [])[0:10])}')"
+  error_tail="$(tail -n 20 "$temporary_dir.stderr" 2>/dev/null || true)"
+  max_rss_kb="$(awk -F: '/Maximum resident set size/{gsub(/^[[:space:]]+/, "", $2); print $2}' "$temporary_dir.process-time" 2>/dev/null || true)"
+  comparison_state_digest_match="$(jq -r 'if has("stateDigestMatch") then (.stateDigestMatch | tostring) else "missing" end' <<< "$comparison_json")"
+  fallback_replay_selection_match="$(jq -r 'if has("matches") then (.matches | tostring) else "missing" end' <<< "$fallback_replay_comparison_json")"
+  if [[ "$comparison_state_digest_match" == "true" && "$fallback_replay_selection_match" == "true" && "$exit_code" -eq 0 ]]; then
+    verification_status="pass"
+  elif [[ "$comparison_state_digest_match" == "false" || "$fallback_replay_selection_match" == "false" ]]; then
+    verification_status="mismatch"
+  else
+    verification_status="error"
+  fi
+  post_validation_collection_finished_ms="$(now_ms)"
+  result_cleanup_started_ms="$(now_ms)"
+  rm -rf "$temporary_dir" "$temporary_dir.stdout" "$temporary_dir.stderr" "$temporary_dir.process-time"
+  result_cleanup_finished_ms="$(now_ms)"
+  verification_json="$(jq -n \
+    --arg status "$verification_status" \
+    --arg verification_child_path "$VERIFICATION_CHILD_PATH" \
+    --argjson exit_code "$exit_code" \
+    --argjson wall_ms "$((run_finished_ms - run_started_ms))" \
+    --argjson lock_wait_ms "$((lock_acquired_ms - lock_started_ms))" \
+    --argjson timeout_limit_secs "$timeout_secs" \
+    --argjson object_workers "$VERIFICATION_OBJECT_WORKERS" \
+    --arg max_rss_kb "${max_rss_kb:-}" \
+    --arg error_tail "$error_tail" \
+    --argjson comparison "$comparison_json" \
+    --argjson fallback_replay_comparison "$fallback_replay_comparison_json" \
+    --argjson meta "$verification_meta_json" \
+    --argjson analysis "$analysis_summary_json" \
+    --argjson post_run_delay_ms "$((delay_finished_ms - observer_started_ms))" \
+    --argjson observer_to_verification_start_ms "$((run_started_ms - observer_started_ms))" \
+    --argjson pre_validation_preparation_ms "$((run_started_ms - preparation_started_ms))" \
+    --argjson binding_check_ms "$((binding_finished_ms - binding_started_ms))" \
+    --argjson binary_contract_check_ms "$((binary_check_finished_ms - binary_check_started_ms))" \
+    --argjson interval_window_calculation_ms "$((window_finished_ms - window_started_ms))" \
+    --argjson scratch_preparation_cleanup_ms "$((scratch_cleanup_finished_ms - scratch_cleanup_started_ms))" \
+    --argjson verification_process_ms "$((run_finished_ms - run_started_ms))" \
+    --argjson post_validation_collection_ms "$((post_validation_collection_finished_ms - post_validation_started_ms))" \
+    --argjson result_cleanup_ms "$((result_cleanup_finished_ms - result_cleanup_started_ms))" \
+    '{status: $status, exitCode: $exit_code, wallMs: $wall_ms, lockWaitMs: $lock_wait_ms,
+      timeoutLimitSecs: $timeout_limit_secs, objectWorkers: $object_workers,
+      verificationChildPath: (if $verification_child_path == "" then null else $verification_child_path end),
+      timing: {postRunDelayMs: $post_run_delay_ms,
+               observerToVerificationStartMs: $observer_to_verification_start_ms,
+               preValidationPreparationMs: $pre_validation_preparation_ms,
+               preparationBreakdownMs: {stateLockWaitMs: $lock_wait_ms,
+                                        bindingCheckMs: $binding_check_ms,
+                                        binaryContractCheckMs: $binary_contract_check_ms,
+                                        intervalWindowCalculationMs: $interval_window_calculation_ms,
+                                        scratchPreparationCleanupMs: $scratch_preparation_cleanup_ms},
+               verificationProcessMs: $verification_process_ms,
+               postValidationCollectionMs: $post_validation_collection_ms,
+               resultCleanupMs: $result_cleanup_ms},
+      maxRssKb: (if $max_rss_kb == "" then null else ($max_rss_kb | tonumber) end),
+      comparison: {stateDigestMatch: ($comparison | if has("stateDigestMatch") then .stateDigestMatch else null end),
+                   mismatchedStates: ($comparison.mismatchedStates // []),
+                   states: (($comparison.states // []) | map({name, sourcePresent, verificationPresent,
+                                                               sourceHashHex, verificationHashHex, matches})),
+                   fallbackReplaySelection: {
+                     matches: ($fallback_replay_comparison | if has("matches") then .matches else null end),
+                     expectedCount: (($fallback_replay_comparison.expected // []) | length),
+                     actualCount: (($fallback_replay_comparison.actual // []) | length),
+                     missing: ($fallback_replay_comparison.missing // []),
+                     unexpected: ($fallback_replay_comparison.unexpected // [])
+                   }},
+      verificationMeta: {status: ($meta.status // null), wallMs: ($meta.wallMs // null),
+                         repositoryObjectsChecked: ($meta.repositoryObjectsChecked // null),
+                         repositoryBytesChecked: ($meta.repositoryBytesChecked // null)},
+      analysis: $analysis,
+      errorTail: $error_tail}')"
+  write_result "$run_id" "$normal_json" "$verification_json"
+}
+
+exec {observer_lock_fd}>"$OBSERVER_LOCK"
+if ! flock -n "$observer_lock_fd"; then
+  echo "another observer already owns $OBSERVER_LOCK" >&2
+  exit 3
+fi
+
+find "$SCRATCH_ROOT" -mindepth 1 -maxdepth 1 -exec rm -rf {} +
+
+if [[ ! -f "$STATE_FILE" ]]; then
+  initial_run_id="$(latest_terminal_run_id)"
+  [[ -n "$initial_run_id" ]] || { echo "no terminal normal run found under $RUNS_ROOT" >&2; exit 2; }
+  jq -n \
+    --arg started_at "$(now_utc)" \
+    --arg initial_run_id "$initial_run_id" \
+    --arg runs_root "$RUNS_ROOT" \
+    --arg state_root "$STATE_ROOT" \
+    --arg rpki_sha256 "$(sha256sum "$RPKI_BIN" | awk '{print $1}')" \
+    --argjson target_runs "$TARGET_RUNS" \
+    --argjson delay_secs "$DELAY_SECS" \
+    --argjson verification_timeout_secs "$VERIFICATION_TIMEOUT_SECS" \
+    --argjson normal_interval_secs "$NORMAL_INTERVAL_SECS" \
+    --argjson normal_run_safety_secs "$NORMAL_RUN_SAFETY_SECS" \
+    --argjson verification_object_workers "$VERIFICATION_OBJECT_WORKERS" \
+    --arg verification_child_path "$VERIFICATION_CHILD_PATH" \
+    '{schemaVersion: 1, startedAt: $started_at, initialLatestTerminalRunId: $initial_run_id,
+      runsRoot: $runs_root, stateRoot: $state_root, verifierBinarySha256: $rpki_sha256,
+      targetRuns: $target_runs, delaySecs: $delay_secs, verificationTimeoutSecs: $verification_timeout_secs,
+      normalIntervalSecs: $normal_interval_secs, normalRunSafetySecs: $normal_run_safety_secs,
+      verificationObjectWorkers: $verification_object_workers,
+      verificationChildPath: (if $verification_child_path == "" then null else $verification_child_path end),
+      verificationAnalyze: true}' \
+    | atomic_replace "$STATE_FILE"
+  update_cumulative
+fi
+
+# Preserve the original baseline run while recording the live configuration after
+# an observer restart. This is particularly important for the normal-run safety
+# window, whose length depends on the soak interval.
+jq --arg updated_at "$(now_utc)" \
+  --argjson target_runs "$TARGET_RUNS" \
+  --argjson delay_secs "$DELAY_SECS" \
+  --argjson verification_timeout_secs "$VERIFICATION_TIMEOUT_SECS" \
+  --argjson normal_interval_secs "$NORMAL_INTERVAL_SECS" \
+  --argjson normal_run_safety_secs "$NORMAL_RUN_SAFETY_SECS" \
+  --argjson verification_object_workers "$VERIFICATION_OBJECT_WORKERS" \
+  --arg verification_child_path "$VERIFICATION_CHILD_PATH" \
+  '.updatedAt = $updated_at
+   | .targetRuns = $target_runs
+   | .delaySecs = $delay_secs
+   | .verificationTimeoutSecs = $verification_timeout_secs
+   | .normalIntervalSecs = $normal_interval_secs
+   | .normalRunSafetySecs = $normal_run_safety_secs
+   | .verificationObjectWorkers = $verification_object_workers
+   | .verificationChildPath = (if $verification_child_path == "" then null else $verification_child_path end)
+   | .verificationAnalyze = true' "$STATE_FILE" \
+  | atomic_replace "$STATE_FILE"
+
+initial_run_number="$(run_number "$(jq -r '.initialLatestTerminalRunId' "$STATE_FILE")")"
+while true; do
+  deliver_pending_notifications
+  observed_count="$(find "$RESULTS_DIR" -maxdepth 1 -type f -name 'run_*.json' -printf '%f\n' | wc -l)"
+  pending_count="$(jq -r '.notificationPending' "$RESULTS_DIR/cumulative.json")"
+  if (( observed_count >= TARGET_RUNS && pending_count == 0 )); then
+    break
+  fi
+  if (( observed_count < TARGET_RUNS )); then
+    while IFS= read -r run_id; do
+      [[ -n "$run_id" ]] || continue
+      (( $(run_number "$run_id") > initial_run_number )) || continue
+      [[ -f "$(result_path "$run_id")" ]] && continue
+      process_terminal_run "$run_id"
+      observed_count="$(find "$RESULTS_DIR" -maxdepth 1 -type f -name 'run_*.json' -printf '%f\n' | wc -l)"
+      (( observed_count < TARGET_RUNS )) || break
+    done < <(terminal_run_ids)
+  fi
+  sleep "$POLL_SECS"
+done
+
+update_cumulative
+echo "completed $(jq -r '.observedRuns' "$RESULTS_DIR/cumulative.json") observed runs; results: $RESULTS_DIR/cumulative.json"

src/bin/rpki_query_service.rs

@@ -11,7 +11,7 @@ use rpki::query::report_stream::{ObjectFilter, ObjectScope};
 use rpki::query::vrp::VrpLookup;
 use rpki::query_db::{
     ChainEdgeRecord, ExportJobRecord, ObjectInstanceRecord, ObjectUriIndexRecord, QueryDb,
-    QueryDbError, ValidationExplainRecord,
+    QueryDbError, RepoSortOrder, ValidationExplainRecord,
 };
 use serde::Serialize;
 use serde_json::{Value, json};
@@ -150,9 +150,20 @@ fn real_main() -> Result<(), String> {
     let db = Arc::new(open_query_db_for_service(&args).map_err(|e| e.to_string())?);
     let export_jobs = Arc::new(Mutex::new(BTreeMap::new()));
     let repo_bytes = match args.repo_bytes_db.as_ref() {
-        Some(path) => Some(Arc::new(
+        Some(path) => {
-            ExternalRepoBytesDb::open_read_only(path).map_err(|e| e.to_string())?,
+            let opened = if args.watch_run_root.is_some() {
-        )),
+                // Watch mode tracks a live soak that keeps writing new object
+                // bytes. A plain read-only open is frozen at open time (new
+                // manifests/CRLs/objects become unparsable until restart), so
+                // use a secondary instance and catch up in the watcher loop.
+                let secondary_path = args.query_db.with_extension("repo-bytes.secondary");
+                ExternalRepoBytesDb::open_as_secondary(path, &secondary_path)
+                    .map_err(|e| e.to_string())?
+            } else {
+                ExternalRepoBytesDb::open_read_only(path).map_err(|e| e.to_string())?
+            };
+            Some(Arc::new(opened))
+        }
         None => None,
     };
     if let Some(run_root) = args.watch_run_root.clone() {
@@ -163,6 +174,7 @@ fn real_main() -> Result<(), String> {
         )?;
         spawn_indexer_watcher(
             Arc::clone(&db),
+            repo_bytes.clone(),
             args.query_db.clone(),
             resolve_indexer_bin(args.indexer_bin.clone())?,
             run_root,
@@ -229,6 +241,7 @@ fn resolve_effective_watch_min_run_seq(
 
 fn spawn_indexer_watcher(
     db: Arc<QueryDb>,
+    repo_bytes: Option<Arc<ExternalRepoBytesDb>>,
     query_db_path: PathBuf,
     indexer_bin: PathBuf,
     run_root: PathBuf,
@@ -250,6 +263,9 @@ fn spawn_indexer_watcher(
                     ) {
                         Ok(message) => {
                             let _ = db.try_catch_up_with_primary();
+                            if let Some(repo_bytes) = repo_bytes.as_ref() {
+                                let _ = repo_bytes.try_catch_up_with_primary();
+                            }
                             eprintln!("query watcher indexed {}: {message}", run_dir.display());
                         }
                         Err(err) => eprintln!(
@@ -260,6 +276,9 @@ fn spawn_indexer_watcher(
                 }
                 Ok(None) => {
                     let _ = db.try_catch_up_with_primary();
+                    if let Some(repo_bytes) = repo_bytes.as_ref() {
+                        let _ = repo_bytes.try_catch_up_with_primary();
+                    }
                 }
                 Err(err) => eprintln!("query watcher index failed: {err}"),
             }
@@ -523,7 +542,18 @@ impl ApiError {
 
 impl From<QueryDbError> for ApiError {
     fn from(value: QueryDbError) -> Self {
-        Self::new(500, value.to_string())
+        match value {
+            // The run is still indexed but its report.json was removed by
+            // retention cleanup: this is a client-visible "gone", not a
+            // server fault.
+            QueryDbError::RunDataGone(run_id) => Self::new(
+                404,
+                format!(
+                    "report data for run {run_id} has expired and was cleaned up by retention"
+                ),
+            ),
+            other => Self::new(500, other.to_string()),
+        }
     }
 }
 
@@ -536,11 +566,10 @@ fn route_request(
     let (path, query) = split_target(target);
     let query = parse_query(query);
     let path = path.trim_end_matches('/');
-    let segments = path
+    let decoded_segments = decode_path_segments(path);
-        .trim_start_matches("/api/v1")
+    let segments = decoded_segments
-        .trim_start_matches('/')
+        .iter()
-        .split('/')
+        .map(String::as_str)
-        .filter(|s| !s.is_empty())
         .collect::<Vec<_>>();
     if segments.is_empty() {
         return Ok(json!({"data":{"service":"rpki_query_service","version":1}}));
@@ -617,10 +646,11 @@ fn route_request(
         }
         ["runs", raw_run_id, "repos"] => {
             let run_id = resolve_run(db, raw_run_id)?;
-            page_response(
+            let page = match repo_sort_from_query(&query)? {
-                db.list_repos(&run_id, limit(&query), cursor(&query))?,
+                Some(order) => db.list_repos_sorted(&run_id, limit(&query), cursor(&query), order)?,
-                Some(run_id),
+                None => db.list_repos(&run_id, limit(&query), cursor(&query))?,
-            )
+            };
+            page_response(page, Some(run_id))
         }
         ["runs", raw_run_id, "repos", repo_id] => {
             let run_id = resolve_run(db, raw_run_id)?;
@@ -907,10 +937,11 @@ fn route_request(
         }
         ["runs", raw_run_id, "stats", "repos"] => {
             let run_id = resolve_run(db, raw_run_id)?;
-            page_response(
+            let page = match repo_sort_from_query(&query)? {
-                db.list_repos(&run_id, limit(&query), cursor(&query))?,
+                Some(order) => db.list_repos_sorted(&run_id, limit(&query), cursor(&query), order)?,
-                Some(run_id),
+                None => db.list_repos(&run_id, limit(&query), cursor(&query))?,
-            )
+            };
+            page_response(page, Some(run_id))
         }
         ["runs", raw_run_id, "stats", "publication-points"] => {
             let run_id = resolve_run(db, raw_run_id)?;
@@ -983,11 +1014,10 @@ fn route_post_request(
 ) -> Result<Value, ApiError> {
     let (path, _) = split_target(target);
     let path = path.trim_end_matches('/');
-    let segments = path
+    let decoded_segments = decode_path_segments(path);
-        .trim_start_matches("/api/v1")
+    let segments = decoded_segments
-        .trim_start_matches('/')
+        .iter()
-        .split('/')
+        .map(String::as_str)
-        .filter(|s| !s.is_empty())
         .collect::<Vec<_>>();
     match segments.as_slice() {
         ["runs", raw_run_id, "exports"] => {
@@ -1140,11 +1170,10 @@ fn route_raw_request(
 ) -> Option<Result<Vec<u8>, ApiError>> {
     let (path, _) = split_target(target);
     let path = path.trim_end_matches('/');
-    let segments = path
+    let decoded_segments = decode_path_segments(path);
-        .trim_start_matches("/api/v1")
+    let segments = decoded_segments
-        .trim_start_matches('/')
+        .iter()
-        .split('/')
+        .map(String::as_str)
-        .filter(|s| !s.is_empty())
         .collect::<Vec<_>>();
     match segments.as_slice() {
         ["runs", raw_run_id, "objects", object_instance_id, "raw"] => Some(raw_object_response(
@@ -1205,7 +1234,20 @@ fn export_download_response(
         .as_ref()
         .ok_or_else(|| ApiError::new(404, "export output path not found"))?;
     let bytes = fs::read(path).map_err(|err| ApiError::new(404, err.to_string()))?;
-    Ok(binary_response(200, "application/x-tar", bytes))
+    let filename = format!(
+        "rpki-export-{}-{}.tar",
+        sanitize_filename_token(&run_id),
+        sanitize_filename_token(job_id)
+    );
+    Ok(binary_response_with_headers(
+        200,
+        "application/x-tar",
+        &[(
+            "Content-Disposition".to_string(),
+            format!("attachment; filename=\"{filename}\""),
+        )],
+        bytes,
+    ))
 }
 
 fn export_job_store_key(run_id: &str, job_id: &str) -> String {
@@ -2079,6 +2121,18 @@ fn hex_value(byte: u8) -> Option<u8> {
     }
 }
 
+/// Split an API path into percent-decoded segments. Real-world ids are plain
+/// hex, so decoding is a no-op for them; it only matters for hand-built URLs
+/// (e.g. an id containing a literal `%2F`).
+fn decode_path_segments(path: &str) -> Vec<String> {
+    path.trim_start_matches("/api/v1")
+        .trim_start_matches('/')
+        .split('/')
+        .filter(|s| !s.is_empty())
+        .map(|segment| percent_decode(segment).unwrap_or_else(|| segment.to_string()))
+        .collect()
+}
+
 fn limit(query: &BTreeMap<String, String>) -> usize {
     query
         .get("limit")
@@ -2103,6 +2157,18 @@ fn exports_limit(query: &BTreeMap<String, String>) -> usize {
         .clamp(1, 200)
 }
 
+fn repo_sort_from_query(query: &BTreeMap<String, String>) -> Result<Option<RepoSortOrder>, ApiError> {
+    match query.get("sort").map(|value| value.as_str()) {
+        None => Ok(None),
+        Some("objects") => match query.get("order").map(|value| value.as_str()) {
+            None | Some("desc") => Ok(Some(RepoSortOrder::ObjectsDesc)),
+            Some("asc") => Ok(Some(RepoSortOrder::ObjectsAsc)),
+            Some(other) => Err(ApiError::new(400, format!("invalid repos order: {other}"))),
+        },
+        Some(other) => Err(ApiError::new(400, format!("invalid repos sort: {other}"))),
+    }
+}
+
 fn object_filter_from_query(query: &BTreeMap<String, String>) -> Result<ObjectFilter, ApiError> {
     let mut filter = ObjectFilter::default();
     if let Some(value) = query.get("type") {
@@ -2229,13 +2295,26 @@ fn json_response(status: u16, value: &Value) -> Vec<u8> {
 }
 
 fn binary_response(status: u16, content_type: &str, body: Vec<u8>) -> Vec<u8> {
+    binary_response_with_headers(status, content_type, &[], body)
+}
+
+fn binary_response_with_headers(
+    status: u16,
+    content_type: &str,
+    extra_headers: &[(String, String)],
+    body: Vec<u8>,
+) -> Vec<u8> {
     let reason = match status {
         200 => "OK",
         404 => "Not Found",
         _ => "Internal Server Error",
     };
+    let mut head = format!("HTTP/1.1 {status} {reason}\r\nContent-Type: {content_type}\r\n");
+    for (name, value) in extra_headers {
+        head.push_str(&format!("{name}: {value}\r\n"));
+    }
     let mut response = format!(
-        "HTTP/1.1 {status} {reason}\r\nContent-Type: {content_type}\r\nContent-Length: {}\r\nConnection: close\r\n\r\n",
+        "{head}Content-Length: {}\r\nConnection: close\r\n\r\n",
         body.len()
     )
     .into_bytes();
@@ -2243,6 +2322,15 @@ fn binary_response(status: u16, content_type: &str, body: Vec<u8>) -> Vec<u8> {
     response
 }
 
+/// Header-safe token for use inside a `Content-Disposition` filename: strips
+/// anything that could smuggle CRLF or quotes into the response head.
+fn sanitize_filename_token(value: &str) -> String {
+    value
+        .chars()
+        .filter(|ch| ch.is_ascii_alphanumeric() || matches!(ch, '-' | '_' | '.'))
+        .collect()
+}
+
 fn is_sha256_hex(value: &str) -> bool {
     value.len() == 64 && value.bytes().all(|byte| byte.is_ascii_hexdigit())
 }
@@ -2607,6 +2695,94 @@ mod tests {
         );
     }
 
+    fn index_sample_run(temp: &tempfile::TempDir) -> QueryDb {
+        let run_dir = temp.path().join("runs/run_0001");
+        fs::create_dir_all(&run_dir).expect("run dir");
+        let object_bytes =
+            fs::read("tests/fixtures/repository/rpki.cernet.net/repo/cernet/0/AS4538.roa")
+                .expect("fixture roa");
+        let object_sha = hex::encode(Sha256::digest(&object_bytes));
+        write_sample_run(&run_dir, &object_sha);
+        let query_db_path = temp.path().join("query-db");
+        index_artifacts(&ArtifactIndexerConfig {
+            query_db_path: query_db_path.clone(),
+            run_root: Some(temp.path().to_path_buf()),
+            run_dir: None,
+            repo_bytes_db_path: None,
+            projection_entry_limit: 5,
+            min_run_seq: None,
+            retain_indexed_runs: None,
+        })
+        .expect("index artifacts");
+        QueryDb::open(&query_db_path).expect("query db")
+    }
+
+    #[test]
+    fn unknown_run_id_returns_404_instead_of_empty_payload() {
+        let temp = tempfile::tempdir().expect("tempdir");
+        let db = index_sample_run(&temp);
+
+        for target in [
+            "/api/v1/runs/run_9999",
+            "/api/v1/runs/run_9999/summary",
+            "/api/v1/runs/run_9999/objects",
+            "/api/v1/runs/run_9999/issues",
+            "/api/v1/runs/run_9999/stats/overview",
+            "/api/v1/runs/run_9999/stats/validation-events?name=manifest",
+        ] {
+            let err = match test_route_request(&db, None, target) {
+                Ok(_) => panic!("expected 404 for {target}"),
+                Err(err) => err,
+            };
+            assert_eq!(err.status, 404, "{target}");
+        }
+        // Existing runs and "latest" keep working.
+        assert!(
+            test_route_request(&db, None, "/api/v1/runs/run_0001/stats/overview").is_ok()
+        );
+        assert!(test_route_request(&db, None, "/api/v1/runs/latest/objects").is_ok());
+    }
+
+    #[test]
+    fn expired_run_report_returns_404_run_data_gone() {
+        let temp = tempfile::tempdir().expect("tempdir");
+        let db = index_sample_run(&temp);
+        assert!(test_route_request(&db, None, "/api/v1/runs/run_0001/objects").is_ok());
+
+        // Retention cleanup removes the run directory while the index record
+        // survives: object routes must report a distinguishable 404.
+        fs::remove_dir_all(temp.path().join("runs/run_0001")).expect("remove run dir");
+        let err = test_route_request(&db, None, "/api/v1/runs/run_0001/objects")
+            .expect_err("run data gone");
+        assert_eq!(err.status, 404);
+        assert!(
+            err.message.contains("expired"),
+            "unexpected message: {}",
+            err.message
+        );
+        let err = test_route_request(&db, None, "/api/v1/runs/run_0001/issues")
+            .expect_err("run data gone");
+        assert_eq!(err.status, 404);
+        // Stats served from the index keep working.
+        assert!(
+            test_route_request(&db, None, "/api/v1/runs/run_0001/stats/overview").is_ok()
+        );
+    }
+
+    #[test]
+    fn decode_path_segments_percent_decodes_ids() {
+        let segments = decode_path_segments("/api/v1/runs/run_0001/objects/abc%2Fdef");
+        assert_eq!(segments, vec!["runs", "run_0001", "objects", "abc/def"]);
+        assert!(decode_path_segments("/api/v1").is_empty());
+        assert!(decode_path_segments("/api/v1/").is_empty());
+    }
+
+    #[test]
+    fn sanitize_filename_token_strips_crlf_and_quotes() {
+        assert_eq!(sanitize_filename_token("run_0001"), "run_0001");
+        assert_eq!(sanitize_filename_token("a\r\nb\"c d/e"), "abcde");
+    }
+
     #[test]
     fn raw_route_downloads_object_bytes_from_repo_bytes_db() {
         let temp = tempfile::tempdir().expect("tempdir");
@@ -2794,6 +2970,31 @@ mod tests {
         assert_eq!(job.status, "complete");
         assert_eq!(job.object_count, 1);
         assert!(job.bytes_written > 0);
+
+        let download = test_route_raw_request(
+            &db,
+            None,
+            &format!("/api/v1/runs/latest/exports/{job_id}/download"),
+        )
+        .expect("download route")
+        .expect("download response");
+        let head_end = download
+            .windows(4)
+            .position(|window| window == b"\r\n\r\n")
+            .expect("header separator");
+        let head = String::from_utf8_lossy(&download[..head_end]);
+        assert!(head.starts_with("HTTP/1.1 200 OK"), "head: {head}");
+        assert!(
+            head.contains("Content-Type: application/x-tar\r\n"),
+            "head: {head}"
+        );
+        let expected_disposition = format!(
+            "Content-Disposition: attachment; filename=\"rpki-export-run_0001-{job_id}.tar\""
+        );
+        assert!(
+            head.contains(&expected_disposition),
+            "head: {head}"
+        );
     }
 
     #[test]

src/blob_store.rs

@@ -78,6 +78,7 @@ pub struct ExternalRepoBytesDb {
     path: PathBuf,
     db: Arc<DB>,
     read_only: bool,
+    secondary: bool,
 }
 
 impl ExternalRawStoreDb {
@@ -89,6 +90,7 @@ impl ExternalRawStoreDb {
         let mut opts = Options::default();
         opts.create_if_missing(true);
         opts.set_compression_type(rocksdb::DBCompressionType::Lz4);
+        opts.set_max_open_files(512);
         let db = DB::open(&opts, &path).map_err(|e| StorageError::RocksDb(e.to_string()))?;
         Ok(Self {
             path,
@@ -186,11 +188,13 @@ impl ExternalRepoBytesDb {
         let mut opts = Options::default();
         opts.create_if_missing(true);
         opts.set_compression_type(rocksdb::DBCompressionType::Lz4);
+        opts.set_max_open_files(512);
         let db = DB::open(&opts, &path).map_err(|e| StorageError::RocksDb(e.to_string()))?;
         Ok(Self {
             path,
             db: Arc::new(db),
             read_only: false,
+            secondary: false,
         })
     }
 
@@ -198,15 +202,60 @@ impl ExternalRepoBytesDb {
         let path = path.into();
         let mut opts = Options::default();
         opts.set_compression_type(rocksdb::DBCompressionType::Lz4);
+        opts.set_max_open_files(512);
         let db = DB::open_for_read_only(&opts, &path, false)
             .map_err(|e| StorageError::RocksDb(e.to_string()))?;
         Ok(Self {
             path,
             db: Arc::new(db),
             read_only: true,
+            secondary: false,
         })
     }
 
+    /// Open the repo-bytes DB as a RocksDB secondary instance.
+    ///
+    /// Unlike `open_read_only` (a frozen point-in-time view), a secondary
+    /// instance can follow the live primary via `try_catch_up_with_primary`,
+    /// which is required when the soak keeps writing new object bytes while
+    /// the query service is running.
+    pub fn open_as_secondary(
+        path: impl Into<PathBuf>,
+        secondary_path: impl Into<PathBuf>,
+    ) -> StorageResult<Self> {
+        let path = path.into();
+        let secondary_path = secondary_path.into();
+        if let Some(parent) = secondary_path.parent() {
+            std::fs::create_dir_all(parent).map_err(|e| StorageError::RocksDb(e.to_string()))?;
+        }
+        let mut opts = Options::default();
+        opts.set_compression_type(rocksdb::DBCompressionType::Lz4);
+        opts.set_max_open_files(512);
+        let db = DB::open_as_secondary(&opts, &path, &secondary_path)
+            .map_err(|e| StorageError::RocksDb(e.to_string()))?;
+        Ok(Self {
+            path,
+            db: Arc::new(db),
+            read_only: true,
+            secondary: true,
+        })
+    }
+
+    /// Pull the secondary view up to the primary's current state. No-op for
+    /// non-secondary handles.
+    pub fn try_catch_up_with_primary(&self) -> StorageResult<()> {
+        if self.secondary {
+            self.db
+                .try_catch_up_with_primary()
+                .map_err(|e| StorageError::RocksDb(e.to_string()))?;
+        }
+        Ok(())
+    }
+
+    pub(crate) fn is_secondary(&self) -> bool {
+        self.secondary
+    }
+
     pub fn put_blob_bytes_batch(&self, blobs: &[(String, Vec<u8>)]) -> StorageResult<()> {
         if blobs.is_empty() {
             return Ok(());
@@ -740,4 +789,48 @@ mod tests {
         );
         assert!(repo_bytes.get_blob_bytes("not-a-valid-hash").is_err());
     }
+
+    #[test]
+    fn external_repo_bytes_db_secondary_catches_up_with_live_primary() {
+        let td = tempfile::tempdir().expect("tempdir");
+        let primary_path = td.path().join("repo-bytes.db");
+        let secondary_path = td.path().join("repo-bytes.secondary");
+
+        let primary = ExternalRepoBytesDb::open(&primary_path).expect("open primary");
+        let bytes_a = b"repo-bytes-a".to_vec();
+        let hash_a = sha256_hex(&bytes_a);
+        primary
+            .put_blob_bytes_batch(&[(hash_a.clone(), bytes_a.clone())])
+            .expect("put a");
+
+        let secondary =
+            ExternalRepoBytesDb::open_as_secondary(&primary_path, &secondary_path)
+                .expect("open secondary");
+        assert!(secondary.is_secondary());
+
+        // A secondary open does not necessarily see pre-existing data until it
+        // catches up with the primary's manifest.
+        secondary
+            .try_catch_up_with_primary()
+            .expect("initial catch up");
+        assert_eq!(
+            secondary.get_blob_bytes(&hash_a).expect("get a via secondary"),
+            Some(bytes_a)
+        );
+
+        // Bytes written by the primary *after* the secondary open become
+        // visible after another catch-up (this is the live-soak scenario).
+        let bytes_b = b"repo-bytes-b".to_vec();
+        let hash_b = sha256_hex(&bytes_b);
+        primary
+            .put_blob_bytes_batch(&[(hash_b.clone(), bytes_b.clone())])
+            .expect("put b after secondary open");
+        secondary
+            .try_catch_up_with_primary()
+            .expect("second catch up");
+        assert_eq!(
+            secondary.get_blob_bytes(&hash_b).expect("get b via secondary"),
+            Some(bytes_b)
+        );
+    }
 }

src/ccr/accumulator.rs

@@ -406,6 +406,7 @@ mod tests {
                 sha256: manifest_hash,
                 object_type: Some("mft".to_string()),
                 validation_status: VcirArtifactValidationStatus::Accepted,
+                reject_reason: None,
             }],
             summary: VcirSummary {
                 local_vrp_count: 0,

src/ccr/build.rs

@@ -729,6 +729,7 @@ mod tests {
                 sha256: hash,
                 object_type: Some("mft".to_string()),
                 validation_status: crate::storage::VcirArtifactValidationStatus::Accepted,
+                reject_reason: None,
             }],
             summary: crate::storage::VcirSummary {
                 local_vrp_count: 0,

src/ccr/export.rs

@@ -267,6 +267,7 @@ mod tests {
                 sha256: hash,
                 object_type: Some("mft".to_string()),
                 validation_status: VcirArtifactValidationStatus::Accepted,
+                reject_reason: None,
             }],
             summary: VcirSummary {
                 local_vrp_count: 0,

src/ccr/verify.rs

@@ -645,6 +645,7 @@ mod tests {
                 sha256: "ff".repeat(32),
                 object_type: Some("mft".to_string()),
                 validation_status: VcirArtifactValidationStatus::Accepted,
+                reject_reason: None,
             }],
             summary: VcirSummary {
                 local_vrp_count: 0,

src/cir/accumulator.rs

@@ -140,6 +140,26 @@ impl CirInputAccumulator {
         rejects.entry(rsync_uri.to_string()).or_insert(reason);
     }
 
+    /// Merge another accumulator into this one. Object digests conflict-check
+    /// exactly like repeated `insert_object` calls; rejects keep the first
+    /// reason seen, matching sequential submission order when `other` holds
+    /// later publication points than `self`.
+    pub fn merge(&mut self, other: CirInputAccumulator) -> Result<(), CirExportError> {
+        for (rsync_uri, digest) in other.fresh_objects {
+            self.insert_object_digest(CirInputSection::Fresh, &rsync_uri, digest)?;
+        }
+        for (rsync_uri, digest) in other.cached_objects {
+            self.insert_object_digest(CirInputSection::Cached, &rsync_uri, digest)?;
+        }
+        for (rsync_uri, reason) in other.fresh_rejects {
+            self.insert_reject(CirInputSection::Fresh, &rsync_uri, reason);
+        }
+        for (rsync_uri, reason) in other.cached_rejects {
+            self.insert_reject(CirInputSection::Cached, &rsync_uri, reason);
+        }
+        Ok(())
+    }
+
     pub fn finalize(self) -> CirInputSnapshot {
         CirInputSnapshot {
             fresh_validated_objects: finalize_objects(self.fresh_objects),
@@ -280,4 +300,95 @@ mod tests {
         assert_eq!(merged.cached_validated_objects.len(), 1);
         assert_eq!(merged.fresh_rejected_objects.len(), 1);
     }
+
+    #[test]
+    fn accumulator_merge_matches_sequential_submission() {
+        let mut sequential = CirInputAccumulator::default();
+        sequential
+            .submit_audit_entries(
+                CirInputSection::Fresh,
+                &[
+                    entry("rsync://example.net/a.roa", 0x11, AuditObjectResult::Ok),
+                    entry("rsync://example.net/b.roa", 0x22, AuditObjectResult::Error),
+                ],
+            )
+            .unwrap();
+        sequential
+            .submit_audit_entries(
+                CirInputSection::Cached,
+                &[entry(
+                    "rsync://example.net/c.roa",
+                    0x33,
+                    AuditObjectResult::Ok,
+                )],
+            )
+            .unwrap();
+
+        let mut first = CirInputAccumulator::default();
+        first
+            .submit_audit_entries(
+                CirInputSection::Fresh,
+                &[entry("rsync://example.net/a.roa", 0x11, AuditObjectResult::Ok)],
+            )
+            .unwrap();
+        let mut second = CirInputAccumulator::default();
+        second
+            .submit_audit_entries(
+                CirInputSection::Fresh,
+                &[entry(
+                    "rsync://example.net/b.roa",
+                    0x22,
+                    AuditObjectResult::Error,
+                )],
+            )
+            .unwrap();
+        second
+            .submit_audit_entries(
+                CirInputSection::Cached,
+                &[entry(
+                    "rsync://example.net/c.roa",
+                    0x33,
+                    AuditObjectResult::Ok,
+                )],
+            )
+            .unwrap();
+        first.merge(second).unwrap();
+        assert_eq!(sequential.finalize(), first.finalize());
+    }
+
+    #[test]
+    fn accumulator_merge_detects_conflicting_hashes_across_shards() {
+        let mut first = CirInputAccumulator::default();
+        first
+            .insert_object(CirInputSection::Fresh, "rsync://example.net/a.roa", &"11".repeat(32))
+            .unwrap();
+        let mut second = CirInputAccumulator::default();
+        second
+            .insert_object(CirInputSection::Fresh, "rsync://example.net/a.roa", &"22".repeat(32))
+            .unwrap();
+        let err = first.merge(second).unwrap_err();
+        assert!(matches!(err, CirExportError::ConflictingObjectHash { .. }));
+    }
+
+    #[test]
+    fn accumulator_merge_keeps_first_reject_reason() {
+        let mut first = CirInputAccumulator::default();
+        first.insert_reject(
+            CirInputSection::Fresh,
+            "rsync://example.net/a.roa",
+            Some("first".to_string()),
+        );
+        let mut second = CirInputAccumulator::default();
+        second.insert_reject(
+            CirInputSection::Fresh,
+            "rsync://example.net/a.roa",
+            Some("second".to_string()),
+        );
+        first.merge(second).unwrap();
+        let snapshot = first.finalize();
+        assert_eq!(
+            snapshot.fresh_rejected_objects[0].reason,
+            Some("first".to_string())
+        );
+    }
 }

src/cli.rs

@@ -55,6 +55,7 @@ struct RunStageTiming {
     enable_child_certificate_validation_cache: bool,
     publication_point_cache_observe_only: bool,
     enable_publication_point_validation_cache: bool,
+    crypto_signature_cache_observe: Option<crate::crypto_sig_cache::CryptoSigCacheSummary>,
     enable_transport_request_prefetch: bool,
     report_build_ms: u64,
     report_write_ms: Option<u64>,
@@ -145,6 +146,8 @@ pub struct CliArgs {
     pub enable_child_certificate_validation_cache: bool,
     pub publication_point_cache_observe_only: bool,
     pub enable_publication_point_validation_cache: bool,
+    pub crypto_signature_cache_observe_only: bool,
+    pub enable_crypto_signature_cache: bool,
     pub enable_transport_request_prefetch: bool,
     pub ccr_out_path: Option<PathBuf>,
     pub vrps_csv_out_path: Option<PathBuf>,
@@ -222,6 +225,12 @@ Options:
                              Evaluate publication-point cache eligibility without changing results
   --enable-publication-point-validation-cache
                              Experimental: reuse complete publication-point validation projections
+  --crypto-signature-cache-observe-only
+                             Measure crypto signature cache key hit rates and verify durations
+                             without changing validation behavior (default: off)
+  --enable-crypto-signature-cache
+                             Experimental: skip cryptographic signature verification on cache
+                             hits (positive conclusions only; default: off)
   --enable-transport-request-prefetch
                              Experimental: prefetch previous run transport repo requests before tree traversal
   --ccr-out <path>           Write CCR DER ContentInfo to this path (optional)
@@ -267,6 +276,16 @@ Options:
                              Legacy Phase 2 scheduler finalize time budget; dedicated finalize worker ignores it (default: 100)
   --parallel-phase2-finalize-queue-capacity <n>
                              Phase 2 dedicated finalize worker queue capacity (default: 32768)
+  --control-plane-stage-workers <n>
+                             Experimental: Phase 2 ready publication point stage worker count;
+                             0 disables the stage pool and keeps inline staging (default: 0)
+  --dead-repo-blacklist <path>
+                             Enable the dead-repo transport blacklist persisted at this JSON
+                             path (default: disabled). Blacklisted rrdp repos skip straight to
+                             rsync; dual-blacklisted repos terminate instantly.
+  --dead-repo-blacklist-fail-threshold <n>
+                             Consecutive runs with transport-class fetch failure before a
+                             (repo, transport) entry is blacklisted (default: 3)
 
   --rsync-local-dir <path>   Use LocalDirRsyncFetcher rooted at this directory (offline tests)
   --disable-rrdp            Disable RRDP and synchronize only via rsync
@@ -301,6 +320,8 @@ pub fn parse_args(argv: &[String]) -> Result<CliArgs, String> {
     let mut ta_paths: Vec<PathBuf> = Vec::new();
     let mut parallel_phase1_cfg = ParallelPhase1Config::default();
     let mut parallel_phase2_cfg = ParallelPhase2Config::default();
+    let mut dead_repo_blacklist_path: Option<PathBuf> = None;
+    let mut dead_repo_blacklist_fail_threshold: Option<u32> = None;
 
     let mut db_path: Option<PathBuf> = None;
     let mut raw_store_db: Option<PathBuf> = None;
@@ -315,6 +336,8 @@ pub fn parse_args(argv: &[String]) -> Result<CliArgs, String> {
     let mut enable_roa_validation_cache: bool = false;
     let mut enable_child_certificate_validation_cache: bool = false;
     let mut publication_point_cache_observe_only: bool = false;
+    let mut crypto_signature_cache_observe_only: bool = false;
+    let mut enable_crypto_signature_cache: bool = false;
     let mut enable_publication_point_validation_cache: bool = false;
     let mut enable_transport_request_prefetch: bool = false;
     let mut ccr_out_path: Option<PathBuf> = None;
@@ -508,6 +531,32 @@ pub fn parse_args(argv: &[String]) -> Result<CliArgs, String> {
                         format!("invalid --parallel-phase2-finalize-queue-capacity: {v}")
                     })?;
             }
+            "--control-plane-stage-workers" => {
+                i += 1;
+                let v = argv
+                    .get(i)
+                    .ok_or("--control-plane-stage-workers requires a value")?;
+                parallel_phase2_cfg.stage_workers = v
+                    .parse::<usize>()
+                    .map_err(|_| format!("invalid --control-plane-stage-workers: {v}"))?;
+            }
+            "--dead-repo-blacklist" => {
+                i += 1;
+                let v = argv
+                    .get(i)
+                    .ok_or("--dead-repo-blacklist requires a value")?;
+                dead_repo_blacklist_path = Some(PathBuf::from(v));
+            }
+            "--dead-repo-blacklist-fail-threshold" => {
+                i += 1;
+                let v = argv
+                    .get(i)
+                    .ok_or("--dead-repo-blacklist-fail-threshold requires a value")?;
+                dead_repo_blacklist_fail_threshold = Some(
+                    v.parse::<u32>()
+                        .map_err(|_| format!("invalid --dead-repo-blacklist-fail-threshold: {v}"))?,
+                );
+            }
             "--db" => {
                 i += 1;
                 let v = argv.get(i).ok_or("--db requires a value")?;
@@ -579,6 +628,12 @@ pub fn parse_args(argv: &[String]) -> Result<CliArgs, String> {
             "--enable-publication-point-validation-cache" => {
                 enable_publication_point_validation_cache = true;
             }
+            "--crypto-signature-cache-observe-only" => {
+                crypto_signature_cache_observe_only = true;
+            }
+            "--enable-crypto-signature-cache" => {
+                enable_crypto_signature_cache = true;
+            }
             "--enable-transport-request-prefetch" => {
                 enable_transport_request_prefetch = true;
             }
@@ -808,6 +863,8 @@ pub fn parse_args(argv: &[String]) -> Result<CliArgs, String> {
             || enable_child_certificate_validation_cache
             || publication_point_cache_observe_only
             || enable_publication_point_validation_cache
+            || crypto_signature_cache_observe_only
+            || enable_crypto_signature_cache
             || enable_transport_request_prefetch
             || rsync_local_dir.is_some()
             || rsync_command.is_some()
@@ -1091,6 +1148,25 @@ pub fn parse_args(argv: &[String]) -> Result<CliArgs, String> {
         }
     }
 
+    if dead_repo_blacklist_fail_threshold.is_some() && dead_repo_blacklist_path.is_none() {
+        return Err(
+            "--dead-repo-blacklist-fail-threshold requires --dead-repo-blacklist".to_string()
+        );
+    }
+    if let Some(path) = dead_repo_blacklist_path {
+        let fail_threshold = dead_repo_blacklist_fail_threshold.unwrap_or(3);
+        if fail_threshold == 0 {
+            return Err("--dead-repo-blacklist-fail-threshold must be >= 1".to_string());
+        }
+        parallel_phase1_cfg.dead_repo_blacklist =
+            Some(crate::parallel::dead_repo_blacklist::DeadRepoBlacklistConfig {
+                path,
+                fail_threshold,
+                capacity:
+                    crate::parallel::dead_repo_blacklist::DEAD_REPO_BLACKLIST_DEFAULT_CAPACITY,
+            });
+    }
+
     Ok(CliArgs {
         verification_only,
         verification_source_run_dir,
@@ -1121,6 +1197,8 @@ pub fn parse_args(argv: &[String]) -> Result<CliArgs, String> {
         enable_child_certificate_validation_cache,
         publication_point_cache_observe_only,
         enable_publication_point_validation_cache,
+        crypto_signature_cache_observe_only,
+        enable_crypto_signature_cache,
         enable_transport_request_prefetch,
         ccr_out_path,
         vrps_csv_out_path,
@@ -2152,6 +2230,36 @@ where
     }
 }
 
+fn policy_for_verification_contract(
+    contract: &crate::verification_only::ValidationContract,
+) -> crate::policy::Policy {
+    // Preserve every contracted policy field, including sync_preference.  The
+    // verification fetchers themselves are network-disabled; changing this
+    // field would change the VCIR reuse-identity fingerprint and falsely make
+    // a contract-selected fallback ineligible.
+    let mut policy = contract.policy.clone();
+    policy.ca_failed_fetch_policy = if contract.fallback_replay_selections.is_empty() {
+        crate::policy::CaFailedFetchPolicy::StopAllOutput
+    } else {
+        crate::policy::CaFailedFetchPolicy::ReuseCurrentInstanceVcir
+    };
+    policy.verification_forced_vcir_reuse_manifest_uris = contract
+        .fallback_replay_selections
+        .iter()
+        .map(|selection| selection.manifest_rsync_uri.clone())
+        .collect();
+    policy
+}
+
+fn transport_prefetch_for_verification_contract(
+    contract: &crate::verification_only::ValidationContract,
+) -> bool {
+    // This only restores the source run's transport scheduling and request
+    // deduplication topology. It does not enable validation-result reuse or
+    // permit network access.
+    contract.cache.transport_prefetch
+}
+
 pub fn run(argv: &[String]) -> Result<(), String> {
     let mut args = parse_args(argv)?;
     let mut prepared_verification = if args.verification_only {
@@ -2176,7 +2284,11 @@ pub fn run(argv: &[String]) -> Result<(), String> {
         args.enable_child_certificate_validation_cache = false;
         args.publication_point_cache_observe_only = false;
         args.enable_publication_point_validation_cache = false;
-        args.enable_transport_request_prefetch = false;
+        args.crypto_signature_cache_observe_only = false;
+        args.enable_crypto_signature_cache = false;
+        args.enable_transport_request_prefetch =
+            transport_prefetch_for_verification_contract(&prepared.contract);
+        args.rsync_scope_policy = prepared.contract.rsync_scope_policy;
         args.ccr_out_path = Some(prepared.artifacts.result_ccr.clone());
         args.vrps_csv_out_path = Some(prepared.artifacts.vrps_csv.clone());
         args.vaps_csv_out_path = Some(prepared.artifacts.vaps_csv.clone());
@@ -2196,7 +2308,7 @@ pub fn run(argv: &[String]) -> Result<(), String> {
     };
 
     let mut policy = if let Some(prepared) = prepared_verification.as_ref() {
-        prepared.contract.policy.clone()
+        policy_for_verification_contract(&prepared.contract)
     } else {
         read_policy(args.policy_path.as_deref())?
     };
@@ -2210,9 +2322,6 @@ pub fn run(argv: &[String]) -> Result<(), String> {
         if args.disable_rrdp {
             policy.sync_preference = crate::policy::SyncPreference::RsyncOnly;
         }
-    } else {
-        policy.sync_preference = crate::policy::SyncPreference::RsyncOnly;
-        policy.ca_failed_fetch_policy = crate::policy::CaFailedFetchPolicy::StopAllOutput;
     }
     let validation_time = args
         .validation_time
@@ -2220,21 +2329,6 @@ pub fn run(argv: &[String]) -> Result<(), String> {
     let validation_time =
         time::OffsetDateTime::from_unix_timestamp(validation_time.unix_timestamp())
             .map_err(|error| format!("normalize validation time failed: {error}"))?;
-    if let Some(path) = args.validation_contract_out_path.as_deref() {
-        let contract = crate::verification_only::ValidationContract::for_current_binary(
-            validation_time,
-            policy.clone(),
-            args.max_ca_depth,
-            args.max_instances,
-            crate::verification_only::ValidationCacheContract {
-                publication_point: args.enable_publication_point_validation_cache,
-                roa: args.enable_roa_validation_cache,
-                child_certificate: args.enable_child_certificate_validation_cache,
-                transport_prefetch: args.enable_transport_request_prefetch,
-            },
-        )?;
-        crate::verification_only::write_validation_contract(path, &contract)?;
-    }
     let http_root_certificates_pem = args
         .http_root_cert_paths
         .iter()
@@ -2351,11 +2445,24 @@ pub fn run(argv: &[String]) -> Result<(), String> {
         store.as_ref(),
     );
     let validation_started = std::time::Instant::now();
+    let crypto_sig_cache =
+        if args.crypto_signature_cache_observe_only || args.enable_crypto_signature_cache {
+            let cache_file = crate::crypto_sig_cache::default_cache_file_path(&args.db_path);
+            let cache = Arc::new(crate::crypto_sig_cache::CryptoSigCache::load_or_rebuild(
+                cache_file,
+                args.enable_crypto_signature_cache,
+            ));
+            crate::crypto_sig_cache::install_global(Arc::clone(&cache));
+            Some(cache)
+        } else {
+            None
+        };
     let collect_current_repo_objects = false;
     let out = if args.verification_only {
         let http = crate::verification_only::NetworkDisabledHttpFetcher;
         let rsync = crate::fetch::current_repository::CurrentRepositoryViewRsyncFetcher::new(
             Arc::clone(&store),
+            args.rsync_scope_policy,
         );
         run_online_validation_with_fetchers(
             Arc::clone(&store),
@@ -2616,48 +2723,129 @@ pub fn run(argv: &[String]) -> Result<(), String> {
         ReportJsonFormat::Pretty
     };
     let ccr_produced_at = time::OffsetDateTime::now_utc();
-    let (report_result, ccr_result) = if args.skip_report_build {
+    let compare_view_trust_anchor = args
-        (
+        .compare_view_trust_anchor
-            Ok(ReportTaskOutput::skipped()),
+        .as_deref()
-            run_ccr_task(
+        .unwrap_or("unknown");
-                store.as_ref(),
+    let cir_tal_uris = if args.cir_enabled {
+        Some(effective_cir_tal_uris_for_discoveries(
+            &args,
             &shared,
-                args.ccr_out_path.as_deref(),
+            resolve_cir_export_tal_uris(&args)?,
-                ccr_produced_at,
+        )?)
-            ),
+    } else {
+        None
+    };
+    let cir_out_path = if args.cir_enabled {
+        Some(
+            args.cir_out_path
+                .as_deref()
+                .expect("validated by parse_args for cir"),
         )
     } else {
+        None
+    };
+    // Take the CIR input snapshot before the output stage so the CIR export can
+    // run inside the same scoped-thread group as report/ccr/compare_view; no
+    // other output task reads `shared.cir_input`.
+    let cir_input_owned = args
+        .cir_enabled
+        .then(|| std::mem::take(&mut shared.cir_input));
+    let (report_result, ccr_result, compare_view_result, cir_result) =
         std::thread::scope(|scope| {
-            let report_handle = scope.spawn(|| {
+            // Reborrow `shared` as a plain reference so the scoped output tasks
+            // (incl. the `move` CIR task) capture the reference instead of
+            // moving fields out of the owned value.
+            let shared = &shared;
+            let report_handle = if args.skip_report_build {
+                None
+            } else {
+                Some(scope.spawn(|| {
                     run_report_task(
                         &policy,
                         validation_time,
-                    &shared,
+                        shared,
                         args.report_json_path.as_deref(),
                         report_json_format,
                     )
-            });
+                }))
+            };
             let ccr_handle = scope.spawn(|| {
                 run_ccr_task(
                     store.as_ref(),
-                    &shared,
+                    shared,
                     args.ccr_out_path.as_deref(),
                     ccr_produced_at,
                 )
             });
-            let report_result = report_handle
+            let compare_view_handle = scope.spawn(|| {
+                run_compare_view_task(
+                    shared,
+                    args.vrps_csv_out_path.as_deref(),
+                    args.vaps_csv_out_path.as_deref(),
+                    compare_view_trust_anchor,
+                )
+            });
+            let cir_handle = match (cir_tal_uris.as_ref(), cir_out_path, cir_input_owned) {
+                (Some(cir_tal_uris), Some(cir_out_path), Some(cir_input)) => {
+                    Some(scope.spawn(move || {
+                        if cir_tal_uris.len() != shared.discoveries.len() {
+                            return Err(format!(
+                                "CIR export TAL URI count ({}) does not match discovery count ({})",
+                                cir_tal_uris.len(),
+                                shared.discoveries.len()
+                            ));
+                        }
+                        let tal_bindings = shared
+                            .discoveries
+                            .iter()
+                            .zip(cir_tal_uris.iter())
+                            .map(|(discovery, tal_uri)| CirTrustAnchorBinding {
+                                trust_anchor: &discovery.trust_anchor,
+                                tal_uri: tal_uri.as_str(),
+                            })
+                            .collect::<Vec<_>>();
+                        export_cir_from_input_snapshot_multi(
+                            &tal_bindings,
+                            validation_time,
+                            cir_input,
+                            cir_out_path,
+                        )
+                        .map_err(|e| e.to_string())
+                    }))
+                }
+                _ => None,
+            };
+            let report_result = match report_handle {
+                Some(handle) => handle
                     .join()
                     .map_err(|_| "report task panicked".to_string())
-                .and_then(|result| result);
+                    .and_then(|result| result),
+                None => Ok(ReportTaskOutput::skipped()),
+            };
             let ccr_result = ccr_handle
                 .join()
                 .map_err(|_| "ccr task panicked".to_string())
                 .and_then(|result| result);
-            (report_result, ccr_result)
+            let compare_view_result = compare_view_handle
-        })
+                .join()
-    };
+                .map_err(|_| "compare view task panicked".to_string())
+                .and_then(|result| result);
+            let cir_result = cir_handle.map(|handle| {
+                handle
+                    .join()
+                    .map_err(|_| "cir task panicked".to_string())
+                    .and_then(|result| result)
+            });
+            (report_result, ccr_result, compare_view_result, cir_result)
+        });
     let report_output = report_result?;
     let ccr_output = ccr_result?;
+    let compare_view_output = compare_view_result?;
+    let cir_summary = match cir_result {
+        Some(result) => Some(result?),
+        None => None,
+    };
     record_memory_checkpoint(
         &mut memory_checkpoints,
         "after_report_and_ccr",
@@ -2678,16 +2866,6 @@ pub fn run(argv: &[String]) -> Result<(), String> {
     let ccr_build_ms = ccr_output.ccr_build_ms;
     let ccr_build_breakdown = ccr_output.ccr_build_breakdown;
     let ccr_write_ms = ccr_output.ccr_write_ms;
-    let compare_view_trust_anchor = args
-        .compare_view_trust_anchor
-        .as_deref()
-        .unwrap_or("unknown");
-    let compare_view_output = run_compare_view_task(
-        &shared,
-        args.vrps_csv_out_path.as_deref(),
-        args.vaps_csv_out_path.as_deref(),
-        compare_view_trust_anchor,
-    )?;
     let compare_view_build_ms = compare_view_output.build_ms;
     let compare_view_write_ms = compare_view_output.write_ms;
     record_memory_checkpoint(
@@ -2700,45 +2878,13 @@ pub fn run(argv: &[String]) -> Result<(), String> {
     let mut cir_build_cir_ms = None;
     let mut cir_write_cir_ms = None;
     let mut cir_total_ms = None;
-    if args.cir_enabled {
+    if let Some(summary) = cir_summary {
-        let cir_tal_uris = effective_cir_tal_uris_for_discoveries(
-            &args,
-            &shared,
-            resolve_cir_export_tal_uris(&args)?,
-        )?;
-        if cir_tal_uris.len() != shared.discoveries.len() {
-            return Err(format!(
-                "CIR export TAL URI count ({}) does not match discovery count ({})",
-                cir_tal_uris.len(),
-                shared.discoveries.len()
-            ));
-        }
-        let cir_out_path = args
-            .cir_out_path
-            .as_deref()
-            .expect("validated by parse_args for cir");
-        let tal_bindings = shared
-            .discoveries
-            .iter()
-            .zip(cir_tal_uris.iter())
-            .map(|(discovery, tal_uri)| CirTrustAnchorBinding {
-                trust_anchor: &discovery.trust_anchor,
-                tal_uri: tal_uri.as_str(),
-            })
-            .collect::<Vec<_>>();
-        let summary = export_cir_from_input_snapshot_multi(
-            &tal_bindings,
-            validation_time,
-            std::mem::take(&mut shared.cir_input),
-            cir_out_path,
-        )
-        .map_err(|e| e.to_string())?;
         cir_build_cir_ms = Some(summary.timing.build_cir_ms);
         cir_write_cir_ms = Some(summary.timing.write_cir_ms);
         cir_total_ms = Some(summary.timing.total_ms);
         eprintln!(
             "wrote CIR: {} (objects={}, trust_anchors={}, build_cir_ms={}, write_cir_ms={}, total_ms={})",
-            cir_out_path.display(),
+            cir_out_path.expect("cir path present when cir enabled").display(),
             summary.object_count,
             summary.trust_anchor_count,
             summary.timing.build_cir_ms,
@@ -2778,6 +2924,37 @@ pub fn run(argv: &[String]) -> Result<(), String> {
         None
     };
     let publication_point_cache_index_load = store.publication_point_cache_mmap_index_load_stats();
+    let crypto_signature_cache_observe = crypto_sig_cache.as_ref().map(|cache| {
+        if let Err(e) = cache.persist() {
+            crate::progress_log::emit(
+                "crypto_signature_cache_persist",
+                serde_json::json!({
+                    "state": "failed",
+                    "error": e,
+                }),
+            );
+        }
+        crate::crypto_sig_cache::clear_global();
+        cache.summary()
+    });
+    if let (Some((_, t)), Some(summary)) =
+        (timing.as_ref(), crypto_signature_cache_observe.as_ref())
+    {
+        let (mut calls, mut would_hit, mut new_keys, mut executed, mut skipped) =
+            (0u64, 0u64, 0u64, 0u64, 0u64);
+        for stats in summary.per_point.values() {
+            calls += stats.calls;
+            would_hit += stats.would_hit;
+            new_keys += stats.new_keys;
+            executed += stats.verify_executed;
+            skipped += stats.verify_skipped;
+        }
+        t.record_count("crypto_signature_cache_observe_calls", calls);
+        t.record_count("crypto_signature_cache_observe_would_hit", would_hit);
+        t.record_count("crypto_signature_cache_observe_new_keys", new_keys);
+        t.record_count("crypto_signature_cache_verify_executed", executed);
+        t.record_count("crypto_signature_cache_verify_skipped", skipped);
+    }
     let timing_report_snapshot = timing
         .as_ref()
         .map(|(_, handle)| handle.report_snapshot(50));
@@ -2787,6 +2964,7 @@ pub fn run(argv: &[String]) -> Result<(), String> {
         enable_child_certificate_validation_cache: args.enable_child_certificate_validation_cache,
         publication_point_cache_observe_only: args.publication_point_cache_observe_only,
         enable_publication_point_validation_cache: args.enable_publication_point_validation_cache,
+        crypto_signature_cache_observe,
         enable_transport_request_prefetch: args.enable_transport_request_prefetch,
         report_build_ms,
         report_write_ms,
@@ -2851,6 +3029,36 @@ pub fn run(argv: &[String]) -> Result<(), String> {
         .or(args.vrps_csv_out_path.as_deref());
     write_stage_timing(stage_timing_anchor_path, &stage_timing)?;
 
+    // A normal run's contract is intentionally finalized only after its
+    // validation outputs have all been produced successfully.  This prevents a
+    // failed/incomplete run from leaving a contract that appears eligible for
+    // verification-only, and captures the fallback set actually exercised by
+    // the completed traversal rather than a prediction made before it starts.
+    if !args.verification_only {
+        if let Some(path) = args.validation_contract_out_path.as_deref() {
+            let mut contract = crate::verification_only::ValidationContract::for_current_binary(
+                validation_time,
+                policy.clone(),
+                args.max_ca_depth,
+                args.max_instances,
+                crate::verification_only::ValidationCacheContract {
+                    publication_point: args.enable_publication_point_validation_cache,
+                    roa: args.enable_roa_validation_cache,
+                    child_certificate: args.enable_child_certificate_validation_cache,
+                    transport_prefetch: args.enable_transport_request_prefetch,
+                    crypto_signature: args.enable_crypto_signature_cache,
+                },
+                args.rsync_scope_policy,
+            )?;
+            contract.fallback_replay_selections =
+                crate::verification_only::collect_fallback_replay_selections(
+                    store.as_ref(),
+                    shared.publication_points.as_ref(),
+                )?;
+            crate::verification_only::write_validation_contract(path, &contract)?;
+        }
+    }
+
     if let Some((out_dir, t)) = timing.as_ref() {
         t.record_count("vrps", shared.vrps.len() as u64);
         t.record_count("aspas", shared.aspas.len() as u64);
@@ -2913,7 +3121,26 @@ pub fn run(argv: &[String]) -> Result<(), String> {
             Some(&prepared.artifacts.compare_dir.join("ccr-state-digest.md")),
             &comparison,
         )?;
-        let scratch_retained = !comparison.state_digest_match || args.keep_verification_scratch;
+        let actual_fallback_replay_selections =
+            crate::verification_only::collect_fallback_replay_selections(
+                store.as_ref(),
+                shared.publication_points.as_ref(),
+            )?;
+        let fallback_replay_selection_comparison =
+            crate::verification_only::compare_fallback_replay_selections(
+                &prepared.contract.fallback_replay_selections,
+                &actual_fallback_replay_selections,
+            );
+        crate::verification_only::write_fallback_replay_selection_comparison(
+            &prepared
+                .artifacts
+                .compare_dir
+                .join("fallback-replay-selection.json"),
+            &fallback_replay_selection_comparison,
+        )?;
+        let verification_match =
+            comparison.state_digest_match && fallback_replay_selection_comparison.matches;
+        let scratch_retained = !verification_match || args.keep_verification_scratch;
         let source_run_id = prepared
             .artifacts
             .source_run_dir
@@ -2925,7 +3152,7 @@ pub fn run(argv: &[String]) -> Result<(), String> {
             &prepared.artifacts.verification_meta,
             &crate::verification_only::VerificationRunMeta {
                 schema_version: 1,
-                status: if comparison.state_digest_match {
+                status: if verification_match {
                     "success".to_string()
                 } else {
                     "mismatch".to_string()
@@ -2936,15 +3163,28 @@ pub fn run(argv: &[String]) -> Result<(), String> {
                 repository_objects_checked: prepared.blob_summary.current_objects,
                 repository_bytes_checked: prepared.blob_summary.bytes_verified,
                 state_digest_match: comparison.state_digest_match,
+                fallback_replay_selection_match: fallback_replay_selection_comparison.matches,
+                fallback_replay_materialization: prepared.fallback_replay_materialization.clone(),
                 scratch_retained,
             },
         )?;
+        if !verification_match {
+            let mut mismatch_parts = Vec::new();
             if !comparison.state_digest_match {
-            return Err(format!(
+                mismatch_parts.push(format!(
-                "verification-only CCR state digest mismatch: {}",
+                    "CCR state digest mismatch: {}",
                     comparison.mismatched_states.join(",")
                 ));
             }
+            if !fallback_replay_selection_comparison.matches {
+                mismatch_parts.push(format!(
+                    "fallback replay selection mismatch: missing={}, unexpected={}",
+                    fallback_replay_selection_comparison.missing.len(),
+                    fallback_replay_selection_comparison.unexpected.len()
+                ));
+            }
+            return Err(format!("verification-only {}", mismatch_parts.join("; ")));
+        }
         if !args.keep_verification_scratch {
             drop(shared);
             drop(store);

src/cli/tests.rs

@@ -108,6 +108,73 @@ fn parse_accepts_normal_validation_contract_output() {
     );
 }
 
+#[test]
+fn verification_policy_preserves_contract_identity_and_selects_only_bound_fallbacks() {
+    let manifest_rsync_uri = "rsync://example.test/repo/current.mft".to_string();
+    let contract = crate::verification_only::ValidationContract {
+        schema_version: crate::verification_only::VALIDATION_CONTRACT_SCHEMA_VERSION,
+        validation_time: "2026-07-16T00:00:00Z".to_string(),
+        binary_sha256: "11".repeat(32),
+        policy: Policy {
+            sync_preference: crate::policy::SyncPreference::RrdpThenRsync,
+            ..Policy::default()
+        },
+        max_ca_depth: 32,
+        max_instances: None,
+        cache: crate::verification_only::ValidationCacheContract {
+            publication_point: true,
+            roa: false,
+            child_certificate: false,
+            transport_prefetch: false,
+            crypto_signature: false,
+        },
+        rsync_scope_policy: crate::fetch::rsync_system::RsyncScopePolicy::ModuleRoot,
+        fallback_replay_selections: vec![crate::verification_only::FallbackReplaySelection {
+            manifest_rsync_uri: manifest_rsync_uri.clone(),
+            vcir_sha256: "22".repeat(32),
+            reuse_identity_sha256: "33".repeat(32),
+        }],
+    };
+
+    let policy = policy_for_verification_contract(&contract);
+    assert_eq!(
+        policy.sync_preference,
+        crate::policy::SyncPreference::RrdpThenRsync,
+        "sync preference is part of the stored VCIR reuse-identity fingerprint"
+    );
+    assert_eq!(
+        policy.ca_failed_fetch_policy,
+        crate::policy::CaFailedFetchPolicy::ReuseCurrentInstanceVcir
+    );
+    assert!(
+        policy
+            .verification_forced_vcir_reuse_manifest_uris
+            .contains(&manifest_rsync_uri)
+    );
+    assert!(
+        !serde_json::to_string(&policy)
+            .expect("serialize policy")
+            .contains("verification_forced_vcir_reuse_manifest_uris"),
+        "the runtime-only override must not perturb the contracted policy fingerprint"
+    );
+
+    assert!(
+        !transport_prefetch_for_verification_contract(&contract),
+        "verification-only must preserve a disabled source transport prefetch setting"
+    );
+    let mut contract_with_prefetch = contract;
+    contract_with_prefetch.cache.transport_prefetch = true;
+    assert_eq!(
+        contract_with_prefetch.rsync_scope_policy,
+        crate::fetch::rsync_system::RsyncScopePolicy::ModuleRoot,
+        "transport scope is part of the source contract"
+    );
+    assert!(
+        transport_prefetch_for_verification_contract(&contract_with_prefetch),
+        "verification-only must preserve a source transport prefetch setting"
+    );
+}
+
 #[test]
 fn parse_rejects_both_tal_url_and_tal_path() {
     let argv = vec![
@@ -289,6 +356,56 @@ fn parse_accepts_publication_point_cache_flags() {
     assert!(args.enable_publication_point_validation_cache);
 }
 
+#[test]
+fn parse_disables_crypto_signature_cache_observe_only_by_default() {
+    let argv = vec![
+        "rpki".to_string(),
+        "--db".to_string(),
+        "db".to_string(),
+        "--tal-path".to_string(),
+        "x.tal".to_string(),
+        "--ta-path".to_string(),
+        "x.cer".to_string(),
+    ];
+    let args = parse_args(&argv).expect("parse args");
+    assert!(!args.crypto_signature_cache_observe_only);
+    assert!(!args.enable_crypto_signature_cache);
+}
+
+#[test]
+fn parse_accepts_crypto_signature_cache_observe_only() {
+    let argv = vec![
+        "rpki".to_string(),
+        "--db".to_string(),
+        "db".to_string(),
+        "--tal-path".to_string(),
+        "x.tal".to_string(),
+        "--ta-path".to_string(),
+        "x.cer".to_string(),
+        "--crypto-signature-cache-observe-only".to_string(),
+    ];
+    let args = parse_args(&argv).expect("parse args");
+    assert!(args.crypto_signature_cache_observe_only);
+    assert!(!args.enable_crypto_signature_cache);
+}
+
+#[test]
+fn parse_accepts_enable_crypto_signature_cache() {
+    let argv = vec![
+        "rpki".to_string(),
+        "--db".to_string(),
+        "db".to_string(),
+        "--tal-path".to_string(),
+        "x.tal".to_string(),
+        "--ta-path".to_string(),
+        "x.cer".to_string(),
+        "--enable-crypto-signature-cache".to_string(),
+    ];
+    let args = parse_args(&argv).expect("parse args");
+    assert!(args.enable_crypto_signature_cache);
+    assert!(!args.crypto_signature_cache_observe_only);
+}
+
 #[test]
 fn parse_accepts_enable_transport_request_prefetch() {
     let argv = vec![
@@ -1807,6 +1924,7 @@ fn run_report_task_and_stage_timing_work() {
         enable_child_certificate_validation_cache: false,
         publication_point_cache_observe_only: false,
         enable_publication_point_validation_cache: false,
+        crypto_signature_cache_observe: None,
         enable_transport_request_prefetch: false,
         report_build_ms: report_output.report_build_ms,
         report_write_ms: report_output.report_write_ms,
@@ -1913,6 +2031,7 @@ fn stage_timing_serializes_memory_telemetry() {
         enable_child_certificate_validation_cache: true,
         publication_point_cache_observe_only: false,
         enable_publication_point_validation_cache: false,
+        crypto_signature_cache_observe: None,
         enable_transport_request_prefetch: true,
         report_build_ms: 2,
         report_write_ms: None,

src/crypto_sig_cache.rs

@@ -0,0 +1,876 @@
+//! Crypto signature verification cache (feature #126).
+//!
+//! Caches the context-free cryptographic fact "this signature over these exact bytes verifies
+//! under this public key" so that later runs can skip the RSA verification entirely. Only
+//! positive conclusions are stored (no negative caching). The cache never replaces any other
+//! validation semantics: revocation (CRL), time-window, and policy checks all run unchanged on
+//! the hit path — the hook wraps exactly the crypto verify call and nothing else.
+//!
+//! Modes (CLI):
+//!
+//! - `--crypto-signature-cache-observe-only` (M1): compute keys, maintain the persistent set,
+//!   record hit/miss/timing statistics, always run the real verification.
+//! - `--enable-crypto-signature-cache` (M2): on a cache hit, skip the real verification and
+//!   return success; on a miss, run the real verification and store the positive conclusion.
+//!   Statistics are recorded the same way; `verify_skipped`/`verify_executed` report what
+//!   actually happened. Both flags may be combined (reuse + statistics).
+//!
+//! The five choke points (see the development plan for the full caller inventory):
+//!
+//! - `CmsSignedObject`: `data_model::signed_object` CMS verify (ROA / ASPA / manifest on the
+//!   main validation path; GBR objects are not signature-verified by the tree runner).
+//! - `EeCertFast`: `validation::cert_path::verify_ee_cert_signature_fast` (ring, TBS + signature
+//!   bytes from the signed-object embedded EE certificate; ROA / ASPA / manifest EE paths).
+//! - `EeCertX509`: `validation::cert_path::verify_ee_cert_signature` (x509-parser; router
+//!   certificates and the `validate_ee_cert_path` family).
+//! - `ChildCaCert`: `validation::ca_path::verify_child_signature` (x509-parser; subordinate CA
+//!   certificates).
+//! - `Crl`: `data_model::crl::verify_signature_with_issuer_spki` (x509-parser; issuer CRLs).
+//!
+//! Trust-anchor self-signature verification is deliberately excluded (a handful of calls per
+//! run).
+//!
+//! Cache key (M1 decision: verify-input bytes, not full object bytes — the key material is in
+//! hand inside each choke point, while the repo-bytes content hash is only available at the
+//! upper processing layer):
+//!
+//! ```text
+//! key = SHA-256( "rpki-crypto-sig-cache-key-v1" | point-tag | "sha256WithRSAEncryption"
+//!                | SHA-256(verify input bytes) | SHA-256(signature bytes)
+//!                | SHA-256(verifier SPKI DER) )
+//! ```
+//!
+//! The signature bytes are part of the key so that a different signature over the same input
+//! under the same key can never inherit a positive verification fact.
+//!
+//! Storage:
+//!
+//! - Single small file next to the work DB (`<work-db-name>.crypto-sig-cache`), never inside
+//!   the work DB, not touched by periodic snapshot reset (#091/#092): the reset paths only
+//!   remove/replace the work DB directory and the PP/child-cert cache artifacts, not siblings
+//!   of their own choosing.
+//! - Binary format: 8-byte magic `RPKICSC2` | u64 record-count | records of
+//!   key[32] | verified_at_unix_secs(i64) | crypto_impl_version(u32) | reserved(u32).
+//!   A missing, corrupt, or magic-mismatched file (including the M1 `RPKICSC1` seen-set
+//!   format) is silently rebuilt empty. Writes are atomic (write-then-rename).
+//! - The cache is persisted once at the end of the validation stage (same point as the M1
+//!   seen-set). If a run fails mid-validation, entries added during that run are lost;
+//!   previously persisted entries are unaffected.
+//! - Size cap: approximate上限 `capacity` entries (default 2,000,000; measured M1 APNIC
+//!   ~148k keys/run, all5 estimated ~750k). When an insert would exceed the cap, the whole
+//!   cache is cleared and rebuilt from scratch (full-rebuild eviction — the simplest policy;
+//!   concurrent inserts may briefly overshoot the cap by a few entries). Each eviction is
+//!   counted in `evictions_total`.
+//! - Concurrency: entries live in 64 shards keyed by the first key byte, so parallel
+//!   validation workers rarely contend; statistics use relaxed atomics.
+
+use std::collections::{BTreeMap, HashMap};
+use std::path::{Path, PathBuf};
+use std::sync::atomic::{AtomicU64, AtomicUsize, Ordering};
+use std::sync::{Arc, Mutex, RwLock};
+
+use serde::Serialize;
+use sha2::Digest;
+
+const CACHE_FILE_MAGIC: &[u8; 8] = b"RPKICSC2";
+const KEY_DOMAIN: &[u8] = b"rpki-crypto-sig-cache-key-v1";
+const ALG_ID: &[u8] = b"sha256WithRSAEncryption";
+const SHARD_COUNT: usize = 64;
+const RECORD_BYTES: usize = 32 + 8 + 4 + 4;
+
+/// The schema version reported in summaries. The on-disk magic already carries the format
+/// version; this constant is the semantic version of the key/value definitions.
+pub const CACHE_SCHEMA_VERSION: u32 = 2;
+
+/// Default maximum number of cached positive conclusions (see module docs).
+pub const DEFAULT_CAPACITY: usize = 2_000_000;
+
+/// Crypto implementation identifier stored in every value. Bump when the verification
+/// backend changes so old entries can be invalidated wholesale.
+/// `1` = ring 0.17 `RSA_PKCS1_2048_8192_SHA256` (direct ring calls and x509-parser's
+/// `verify` feature, which is backed by ring in this crate).
+pub const CRYPTO_IMPL_VERSION: u32 = 1;
+
+#[derive(Clone, Copy, Debug, PartialEq, Eq, Hash)]
+pub enum CryptoSigVerifyPoint {
+    CmsSignedObject,
+    EeCertFast,
+    EeCertX509,
+    ChildCaCert,
+    Crl,
+}
+
+impl CryptoSigVerifyPoint {
+    pub const ALL: [Self; 5] = [
+        Self::CmsSignedObject,
+        Self::EeCertFast,
+        Self::EeCertX509,
+        Self::ChildCaCert,
+        Self::Crl,
+    ];
+
+    fn tag(self) -> u8 {
+        match self {
+            Self::CmsSignedObject => 1,
+            Self::EeCertFast => 2,
+            Self::EeCertX509 => 3,
+            Self::ChildCaCert => 4,
+            Self::Crl => 5,
+        }
+    }
+
+    pub fn name(self) -> &'static str {
+        match self {
+            Self::CmsSignedObject => "cms_signed_object",
+            Self::EeCertFast => "ee_cert_fast",
+            Self::EeCertX509 => "ee_cert_x509",
+            Self::ChildCaCert => "child_ca_cert",
+            Self::Crl => "crl",
+        }
+    }
+
+    fn from_index(index: usize) -> Self {
+        Self::ALL[index]
+    }
+}
+
+/// Compute the cache key for one verification fact.
+///
+/// `input_bytes` is the exact byte string fed to the verifier (CMS signedAttrs DER for
+/// signature, TBSCertificate DER, TBSCertList DER), `signature_bytes` the signature value,
+/// and `verifier_spki_der` the DER of the verifying SubjectPublicKeyInfo.
+pub fn compute_cache_key(
+    point: CryptoSigVerifyPoint,
+    input_bytes: &[u8],
+    signature_bytes: &[u8],
+    verifier_spki_der: &[u8],
+) -> [u8; 32] {
+    let mut h = sha2::Sha256::new();
+    h.update(KEY_DOMAIN);
+    h.update([point.tag()]);
+    h.update(ALG_ID);
+    for part in [input_bytes, signature_bytes, verifier_spki_der] {
+        let part_hash = sha2::Sha256::digest(part);
+        h.update(part_hash);
+    }
+    h.finalize().into()
+}
+
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+pub struct CacheValue {
+    pub verified_at_unix_secs: i64,
+    pub crypto_impl_version: u32,
+}
+
+#[derive(Clone, Debug, Default, PartialEq, Eq, Serialize)]
+pub struct CryptoSigVerifyPointStats {
+    pub calls: u64,
+    /// Key already present in the cache (positive conclusion available).
+    pub would_hit: u64,
+    /// Key not present; after a successful verification it is inserted.
+    pub new_keys: u64,
+    /// Real cryptographic verification executed (always equals `calls` in observe-only
+    /// mode; equals `new_keys` plus hit-but-observe calls otherwise).
+    pub verify_executed: u64,
+    /// Real verification skipped because of a cache hit (only > 0 when reuse is enabled).
+    pub verify_skipped: u64,
+    pub verify_nanos_on_would_hit: u64,
+    pub verify_nanos_on_new_key: u64,
+}
+
+#[derive(Clone, Debug, PartialEq, Eq, Serialize)]
+pub struct CryptoSigCacheSummary {
+    pub schema_version: u32,
+    pub cache_file: String,
+    pub reuse_enabled: bool,
+    pub capacity: u64,
+    pub entries_loaded: u64,
+    pub entries_total: u64,
+    pub evictions_total: u64,
+    pub per_point: BTreeMap<String, CryptoSigVerifyPointStats>,
+}
+
+#[derive(Default)]
+struct PointCounters {
+    calls: AtomicU64,
+    would_hit: AtomicU64,
+    new_keys: AtomicU64,
+    verify_executed: AtomicU64,
+    verify_skipped: AtomicU64,
+    verify_nanos_on_would_hit: AtomicU64,
+    verify_nanos_on_new_key: AtomicU64,
+}
+
+fn shard_index(key: &[u8; 32]) -> usize {
+    (key[0] as usize) % SHARD_COUNT
+}
+
+/// The persistent positive-conclusion signature cache plus per-run statistics.
+///
+/// See the module docs for storage format, eviction, and failure semantics.
+pub struct CryptoSigCache {
+    cache_file: PathBuf,
+    reuse_enabled: bool,
+    capacity: usize,
+    entries_loaded: u64,
+    shards: Vec<Mutex<HashMap<[u8; 32], CacheValue>>>,
+    len: AtomicUsize,
+    evictions: AtomicU64,
+    eviction_lock: Mutex<()>,
+    points: [PointCounters; 5],
+}
+
+impl CryptoSigCache {
+    /// Load the cache from `cache_file`; start empty when the file is missing, corrupt, or
+    /// has an unknown schema header. `reuse_enabled` selects the M2 hit-skips-verify
+    /// behavior; when false the cache is observe-only (M1 semantics).
+    pub fn load_or_rebuild(cache_file: PathBuf, reuse_enabled: bool) -> Self {
+        Self::load_or_rebuild_with_capacity(cache_file, reuse_enabled, DEFAULT_CAPACITY)
+    }
+
+    pub fn load_or_rebuild_with_capacity(
+        cache_file: PathBuf,
+        reuse_enabled: bool,
+        capacity: usize,
+    ) -> Self {
+        let entries = load_cache_file(&cache_file).unwrap_or_default();
+        let mut shards = Vec::with_capacity(SHARD_COUNT);
+        for _ in 0..SHARD_COUNT {
+            shards.push(Mutex::new(HashMap::new()));
+        }
+        let mut len = 0usize;
+        for (key, value) in entries {
+            // Respect the cap even when loading a foreign/oversized file: keep the first
+            // `capacity` records in file order.
+            if len >= capacity {
+                break;
+            }
+            shards[shard_index(&key)]
+                .lock()
+                .expect("crypto sig shard lock")
+                .insert(key, value);
+            len += 1;
+        }
+        Self {
+            cache_file,
+            reuse_enabled,
+            capacity,
+            entries_loaded: len as u64,
+            shards,
+            len: AtomicUsize::new(len),
+            evictions: AtomicU64::new(0),
+            eviction_lock: Mutex::new(()),
+            points: Default::default(),
+        }
+    }
+
+    pub fn reuse_enabled(&self) -> bool {
+        self.reuse_enabled
+    }
+
+    fn lookup(&self, key: &[u8; 32]) -> bool {
+        self.shards[shard_index(key)]
+            .lock()
+            .expect("crypto sig shard lock")
+            .contains_key(key)
+    }
+
+    #[cfg(test)]
+    pub fn contains(&self, key: &[u8; 32]) -> bool {
+        self.lookup(key)
+    }
+
+    /// Insert a positive conclusion (used by the verify path and by tests).
+    pub fn insert_positive(&self, key: [u8; 32]) {
+        let value = CacheValue {
+            verified_at_unix_secs: time::OffsetDateTime::now_utc().unix_timestamp(),
+            crypto_impl_version: CRYPTO_IMPL_VERSION,
+        };
+        if self.len.load(Ordering::Relaxed) >= self.capacity {
+            self.evict_all();
+        }
+        let mut shard = self.shards[shard_index(&key)]
+            .lock()
+            .expect("crypto sig shard lock");
+        if shard.insert(key, value).is_none() {
+            self.len.fetch_add(1, Ordering::Relaxed);
+        }
+    }
+
+    /// Full-rebuild eviction: clear every shard. See the module docs for the semantics.
+    fn evict_all(&self) {
+        let _guard = self.eviction_lock.lock().expect("crypto sig eviction lock");
+        if self.len.load(Ordering::Relaxed) < self.capacity {
+            // Another thread already evicted.
+            return;
+        }
+        for shard in &self.shards {
+            shard.lock().expect("crypto sig shard lock").clear();
+        }
+        self.len.store(0, Ordering::Relaxed);
+        self.evictions.fetch_add(1, Ordering::Relaxed);
+    }
+
+    /// Cache-aware verification: observe-only or reuse depending on `reuse_enabled`.
+    /// The `verify` closure must perform exactly the real cryptographic verification.
+    pub fn verify<E>(
+        &self,
+        point: CryptoSigVerifyPoint,
+        input_bytes: &[u8],
+        signature_bytes: &[u8],
+        verifier_spki_der: &[u8],
+        verify: impl FnOnce() -> Result<(), E>,
+    ) -> Result<(), E> {
+        let key = compute_cache_key(point, input_bytes, signature_bytes, verifier_spki_der);
+        let hit = self.lookup(&key);
+        let counters = &self.points[(point.tag() - 1) as usize];
+        counters.calls.fetch_add(1, Ordering::Relaxed);
+        if hit {
+            counters.would_hit.fetch_add(1, Ordering::Relaxed);
+            if self.reuse_enabled {
+                counters.verify_skipped.fetch_add(1, Ordering::Relaxed);
+                return Ok(());
+            }
+        } else {
+            counters.new_keys.fetch_add(1, Ordering::Relaxed);
+        }
+        counters.verify_executed.fetch_add(1, Ordering::Relaxed);
+        let started = std::time::Instant::now();
+        let result = verify();
+        let nanos = started.elapsed().as_nanos().min(u128::from(u64::MAX)) as u64;
+        if hit {
+            counters
+                .verify_nanos_on_would_hit
+                .fetch_add(nanos, Ordering::Relaxed);
+        } else {
+            counters
+                .verify_nanos_on_new_key
+                .fetch_add(nanos, Ordering::Relaxed);
+            if result.is_ok() {
+                self.insert_positive(key);
+            }
+        }
+        result
+    }
+
+    /// Persist the cache atomically (write-then-rename).
+    pub fn persist(&self) -> Result<(), String> {
+        persist_cache_file(&self.cache_file, &self.shards)
+    }
+
+    pub fn summary(&self) -> CryptoSigCacheSummary {
+        let mut per_point = BTreeMap::new();
+        for (index, counters) in self.points.iter().enumerate() {
+            let stats = CryptoSigVerifyPointStats {
+                calls: counters.calls.load(Ordering::Relaxed),
+                would_hit: counters.would_hit.load(Ordering::Relaxed),
+                new_keys: counters.new_keys.load(Ordering::Relaxed),
+                verify_executed: counters.verify_executed.load(Ordering::Relaxed),
+                verify_skipped: counters.verify_skipped.load(Ordering::Relaxed),
+                verify_nanos_on_would_hit: counters
+                    .verify_nanos_on_would_hit
+                    .load(Ordering::Relaxed),
+                verify_nanos_on_new_key: counters
+                    .verify_nanos_on_new_key
+                    .load(Ordering::Relaxed),
+            };
+            if stats.calls > 0 {
+                per_point.insert(
+                    CryptoSigVerifyPoint::from_index(index).name().to_string(),
+                    stats,
+                );
+            }
+        }
+        CryptoSigCacheSummary {
+            schema_version: CACHE_SCHEMA_VERSION,
+            cache_file: self.cache_file.to_string_lossy().into_owned(),
+            reuse_enabled: self.reuse_enabled,
+            capacity: self.capacity as u64,
+            entries_loaded: self.entries_loaded,
+            entries_total: self.len.load(Ordering::Relaxed) as u64,
+            evictions_total: self.evictions.load(Ordering::Relaxed),
+            per_point,
+        }
+    }
+}
+
+/// Default cache file location: a sibling of the work DB directory, mirroring
+/// `default_pp_cache_index_dir` (`<work-db-name>.crypto-sig-cache`).
+pub fn default_cache_file_path(db_path: &Path) -> PathBuf {
+    let file_name = db_path
+        .file_name()
+        .and_then(|name| name.to_str())
+        .unwrap_or("work-db");
+    db_path.with_file_name(format!("{file_name}.crypto-sig-cache"))
+}
+
+fn load_cache_file(path: &Path) -> Option<HashMap<[u8; 32], CacheValue>> {
+    let bytes = std::fs::read(path).ok()?;
+    let header_len = CACHE_FILE_MAGIC.len() + 8;
+    if bytes.len() < header_len || &bytes[..CACHE_FILE_MAGIC.len()] != CACHE_FILE_MAGIC {
+        return None;
+    }
+    let mut count_bytes = [0u8; 8];
+    count_bytes.copy_from_slice(&bytes[CACHE_FILE_MAGIC.len()..header_len]);
+    let count = u64::from_le_bytes(count_bytes) as usize;
+    let body = &bytes[header_len..];
+    if body.len() != count.saturating_mul(RECORD_BYTES) {
+        return None;
+    }
+    let mut entries = HashMap::with_capacity(count);
+    for chunk in body.chunks_exact(RECORD_BYTES) {
+        let mut key = [0u8; 32];
+        key.copy_from_slice(&chunk[..32]);
+        let mut secs_bytes = [0u8; 8];
+        secs_bytes.copy_from_slice(&chunk[32..40]);
+        let mut version_bytes = [0u8; 4];
+        version_bytes.copy_from_slice(&chunk[40..44]);
+        entries.insert(
+            key,
+            CacheValue {
+                verified_at_unix_secs: i64::from_le_bytes(secs_bytes),
+                crypto_impl_version: u32::from_le_bytes(version_bytes),
+            },
+        );
+    }
+    Some(entries)
+}
+
+fn persist_cache_file(
+    path: &Path,
+    shards: &[Mutex<HashMap<[u8; 32], CacheValue>>],
+) -> Result<(), String> {
+    let mut bytes = Vec::new();
+    bytes.extend_from_slice(CACHE_FILE_MAGIC);
+    // Placeholder for the record count; patched after collecting.
+    bytes.extend_from_slice(&[0u8; 8]);
+    let mut count = 0u64;
+    for shard in shards {
+        let map = shard.lock().expect("crypto sig shard lock");
+        for (key, value) in map.iter() {
+            bytes.extend_from_slice(key);
+            bytes.extend_from_slice(&value.verified_at_unix_secs.to_le_bytes());
+            bytes.extend_from_slice(&value.crypto_impl_version.to_le_bytes());
+            bytes.extend_from_slice(&[0u8; 4]);
+            count += 1;
+        }
+    }
+    bytes[CACHE_FILE_MAGIC.len()..CACHE_FILE_MAGIC.len() + 8]
+        .copy_from_slice(&count.to_le_bytes());
+    if let Some(parent) = path.parent() {
+        std::fs::create_dir_all(parent)
+            .map_err(|e| format!("create cache-file dir failed: {}: {e}", parent.display()))?;
+    }
+    let tmp_path = path.with_extension("tmp");
+    std::fs::write(&tmp_path, &bytes)
+        .map_err(|e| format!("write cache file failed: {}: {e}", tmp_path.display()))?;
+    std::fs::rename(&tmp_path, path).map_err(|e| {
+        format!(
+            "rename cache file failed: {} -> {}: {e}",
+            tmp_path.display(),
+            path.display()
+        )
+    })?;
+    Ok(())
+}
+
+static GLOBAL_CACHE: RwLock<Option<Arc<CryptoSigCache>>> = RwLock::new(None);
+
+/// Install the process-global cache (called once from the CLI run path when
+/// `--crypto-signature-cache-observe-only` and/or `--enable-crypto-signature-cache` is set).
+pub fn install_global(cache: Arc<CryptoSigCache>) {
+    *GLOBAL_CACHE.write().expect("crypto sig cache lock") = Some(cache);
+}
+
+/// Remove the process-global cache. Used by tests and by the CLI after the run finished.
+pub fn clear_global() {
+    *GLOBAL_CACHE.write().expect("crypto sig cache lock") = None;
+}
+
+pub fn global_cache() -> Option<Arc<CryptoSigCache>> {
+    GLOBAL_CACHE
+        .read()
+        .expect("crypto sig cache lock")
+        .clone()
+}
+
+/// Choke-point hook: cache-aware wrapper around a real signature verification.
+///
+/// With no cache installed this is a single read-lock check plus the verification itself;
+/// verification semantics never change either way (a hit only skips the crypto primitive,
+/// never the surrounding revocation/time/policy checks).
+pub fn verify_with_cache<E>(
+    point: CryptoSigVerifyPoint,
+    input_bytes: &[u8],
+    signature_bytes: &[u8],
+    verifier_spki_der: &[u8],
+    verify: impl FnOnce() -> Result<(), E>,
+) -> Result<(), E> {
+    let Some(cache) = global_cache() else {
+        return verify();
+    };
+    cache.verify(
+        point,
+        input_bytes,
+        signature_bytes,
+        verifier_spki_der,
+        verify,
+    )
+}
+
+#[cfg(test)]
+pub(crate) static GLOBAL_TEST_LOCK: Mutex<()> = Mutex::new(());
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    fn temp_cache(reuse_enabled: bool) -> (tempfile::TempDir, CryptoSigCache) {
+        let dir = tempfile::tempdir().expect("tempdir");
+        let cache = CryptoSigCache::load_or_rebuild(
+            dir.path().join("work-db.crypto-sig-cache"),
+            reuse_enabled,
+        );
+        (dir, cache)
+    }
+
+    #[test]
+    fn cache_key_is_deterministic_and_domain_separated() {
+        let a = compute_cache_key(
+            CryptoSigVerifyPoint::CmsSignedObject,
+            b"input",
+            b"sig",
+            b"spki",
+        );
+        let b = compute_cache_key(
+            CryptoSigVerifyPoint::CmsSignedObject,
+            b"input",
+            b"sig",
+            b"spki",
+        );
+        assert_eq!(a, b);
+
+        // Different verify point.
+        assert_ne!(
+            a,
+            compute_cache_key(CryptoSigVerifyPoint::Crl, b"input", b"sig", b"spki")
+        );
+        // Different input bytes.
+        assert_ne!(
+            a,
+            compute_cache_key(
+                CryptoSigVerifyPoint::CmsSignedObject,
+                b"input2",
+                b"sig",
+                b"spki"
+            )
+        );
+        // Different signature bytes over the same input must not share a key.
+        assert_ne!(
+            a,
+            compute_cache_key(
+                CryptoSigVerifyPoint::CmsSignedObject,
+                b"input",
+                b"sig2",
+                b"spki"
+            )
+        );
+        // Different verifier key.
+        assert_ne!(
+            a,
+            compute_cache_key(
+                CryptoSigVerifyPoint::CmsSignedObject,
+                b"input",
+                b"sig",
+                b"spki2"
+            )
+        );
+    }
+
+    #[test]
+    fn miss_verifies_and_writes_then_hit_skips_verify() {
+        let (_dir, cache) = temp_cache(true);
+        let mut executed = 0u64;
+        // First call: miss -> real verification runs and passes -> entry written.
+        cache
+            .verify(
+                CryptoSigVerifyPoint::EeCertFast,
+                b"tbs",
+                b"sig",
+                b"spki",
+                || -> Result<(), String> {
+                    executed += 1;
+                    Ok(())
+                },
+            )
+            .expect("verify ok");
+        assert_eq!(executed, 1);
+        // Second call: hit -> verification skipped entirely.
+        cache
+            .verify(
+                CryptoSigVerifyPoint::EeCertFast,
+                b"tbs",
+                b"sig",
+                b"spki",
+                || -> Result<(), String> {
+                    executed += 1;
+                    Ok(())
+                },
+            )
+            .expect("verify ok");
+        assert_eq!(executed, 1, "cache hit must skip the real verification");
+        let summary = cache.summary();
+        let stats = summary.per_point.get("ee_cert_fast").expect("stats present");
+        assert_eq!(stats.calls, 2);
+        assert_eq!(stats.new_keys, 1);
+        assert_eq!(stats.would_hit, 1);
+        assert_eq!(stats.verify_executed, 1);
+        assert_eq!(stats.verify_skipped, 1);
+        assert!(summary.reuse_enabled);
+    }
+
+    #[test]
+    fn preset_positive_key_skips_verify_that_would_fail() {
+        let (_dir, cache) = temp_cache(true);
+        // Tampered object bytes whose signature can never verify; pre-seed the positive
+        // conclusion directly (as if a previous run had verified the untampered object).
+        let key = compute_cache_key(
+            CryptoSigVerifyPoint::CmsSignedObject,
+            b"tampered-input",
+            b"tampered-sig",
+            b"spki",
+        );
+        cache.insert_positive(key);
+        let mut executed = 0u64;
+        let result = cache.verify(
+            CryptoSigVerifyPoint::CmsSignedObject,
+            b"tampered-input",
+            b"tampered-sig",
+            b"spki",
+            || -> Result<(), String> {
+                executed += 1;
+                Err("signature does not verify".to_string())
+            },
+        );
+        assert_eq!(result, Ok(()), "cache hit must return Ok without verifying");
+        assert_eq!(executed, 0, "real verification must not run on a hit");
+    }
+
+    #[test]
+    fn failed_verify_is_not_cached_and_reruns_next_time() {
+        let (_dir, cache) = temp_cache(true);
+        let mut executed = 0u64;
+        for _ in 0..2 {
+            let result = cache.verify(
+                CryptoSigVerifyPoint::Crl,
+                b"tbs",
+                b"sig",
+                b"spki",
+                || -> Result<(), String> {
+                    executed += 1;
+                    Err("bad signature".to_string())
+                },
+            );
+            assert!(result.is_err());
+        }
+        assert_eq!(executed, 2, "no negative caching: every call re-verifies");
+        let key = compute_cache_key(CryptoSigVerifyPoint::Crl, b"tbs", b"sig", b"spki");
+        assert!(!cache.contains(&key));
+        let summary = cache.summary();
+        let stats = summary.per_point.get("crl").expect("stats present");
+        assert_eq!(stats.new_keys, 2);
+        assert_eq!(stats.verify_executed, 2);
+        assert_eq!(stats.verify_skipped, 0);
+    }
+
+    #[test]
+    fn observe_only_mode_counts_would_hit_but_still_verifies() {
+        let (_dir, cache) = temp_cache(false);
+        let key = compute_cache_key(CryptoSigVerifyPoint::ChildCaCert, b"tbs", b"sig", b"spki");
+        cache.insert_positive(key);
+        let mut executed = 0u64;
+        cache
+            .verify(
+                CryptoSigVerifyPoint::ChildCaCert,
+                b"tbs",
+                b"sig",
+                b"spki",
+                || -> Result<(), String> {
+                    executed += 1;
+                    Ok(())
+                },
+            )
+            .expect("verify ok");
+        assert_eq!(executed, 1, "observe-only mode must still run the verification");
+        let summary = cache.summary();
+        assert!(!summary.reuse_enabled);
+        let stats = summary
+            .per_point
+            .get("child_ca_cert")
+            .expect("stats present");
+        assert_eq!(stats.would_hit, 1);
+        assert_eq!(stats.verify_executed, 1);
+        assert_eq!(stats.verify_skipped, 0);
+    }
+
+    #[test]
+    fn cache_file_roundtrips_entries_across_reload() {
+        let dir = tempfile::tempdir().expect("tempdir");
+        let path = dir.path().join("work-db.crypto-sig-cache");
+        let cache = CryptoSigCache::load_or_rebuild(path.clone(), true);
+        let key = compute_cache_key(CryptoSigVerifyPoint::EeCertFast, b"tbs", b"sig", b"spki");
+        cache.insert_positive(key);
+        cache.persist().expect("persist");
+
+        let reloaded = CryptoSigCache::load_or_rebuild(path, true);
+        assert!(reloaded.contains(&key));
+        let summary = reloaded.summary();
+        assert_eq!(summary.entries_loaded, 1);
+        assert_eq!(summary.entries_total, 1);
+        assert_eq!(summary.schema_version, CACHE_SCHEMA_VERSION);
+
+        // Reloaded entry drives a skip.
+        let mut executed = 0u64;
+        reloaded
+            .verify(
+                CryptoSigVerifyPoint::EeCertFast,
+                b"tbs",
+                b"sig",
+                b"spki",
+                || -> Result<(), String> {
+                    executed += 1;
+                    Ok(())
+                },
+            )
+            .expect("verify ok");
+        assert_eq!(executed, 0);
+    }
+
+    #[test]
+    fn cache_file_rebuilds_on_missing_corrupt_or_mismatched_schema() {
+        let dir = tempfile::tempdir().expect("tempdir");
+        let path = dir.path().join("work-db.crypto-sig-cache");
+
+        // Missing file.
+        let cache = CryptoSigCache::load_or_rebuild(path.clone(), true);
+        assert_eq!(cache.summary().entries_loaded, 0);
+
+        // Garbage content.
+        std::fs::write(&path, b"not a valid cache file").expect("write garbage");
+        let cache = CryptoSigCache::load_or_rebuild(path.clone(), true);
+        assert_eq!(cache.summary().entries_loaded, 0);
+
+        // M1 seen-set format (RPKICSC1) must not load as a v2 cache: cold start.
+        let mut bytes = Vec::new();
+        bytes.extend_from_slice(b"RPKICSC1");
+        bytes.extend_from_slice(&1u64.to_le_bytes());
+        bytes.extend_from_slice(&[7u8; 32]);
+        std::fs::write(&path, &bytes).expect("write v1 file");
+        let cache = CryptoSigCache::load_or_rebuild(path.clone(), true);
+        assert_eq!(cache.summary().entries_loaded, 0);
+
+        // Valid v2 header but truncated body.
+        let mut bytes = Vec::new();
+        bytes.extend_from_slice(CACHE_FILE_MAGIC);
+        bytes.extend_from_slice(&2u64.to_le_bytes());
+        bytes.extend_from_slice(&[9u8; RECORD_BYTES]);
+        std::fs::write(&path, &bytes).expect("write truncated");
+        let cache = CryptoSigCache::load_or_rebuild(path, true);
+        assert_eq!(cache.summary().entries_loaded, 0);
+    }
+
+    #[test]
+    fn size_cap_triggers_full_rebuild_eviction() {
+        let dir = tempfile::tempdir().expect("tempdir");
+        let cache = CryptoSigCache::load_or_rebuild_with_capacity(
+            dir.path().join("work-db.crypto-sig-cache"),
+            true,
+            2,
+        );
+        let k1 = compute_cache_key(CryptoSigVerifyPoint::Crl, b"a", b"s", b"k");
+        let k2 = compute_cache_key(CryptoSigVerifyPoint::Crl, b"b", b"s", b"k");
+        let k3 = compute_cache_key(CryptoSigVerifyPoint::Crl, b"c", b"s", b"k");
+        cache.insert_positive(k1);
+        cache.insert_positive(k2);
+        assert!(cache.contains(&k1) && cache.contains(&k2));
+        // Third insert exceeds the cap -> full rebuild, then insert.
+        cache.insert_positive(k3);
+        assert!(!cache.contains(&k1));
+        assert!(!cache.contains(&k2));
+        assert!(cache.contains(&k3));
+        let summary = cache.summary();
+        assert_eq!(summary.entries_total, 1);
+        assert_eq!(summary.evictions_total, 1);
+    }
+
+    #[test]
+    fn default_cache_file_path_mirrors_pp_cache_index_placement() {
+        let db = Path::new("/state/db/work-db");
+        assert_eq!(
+            default_cache_file_path(db),
+            PathBuf::from("/state/db/work-db.crypto-sig-cache")
+        );
+    }
+
+    #[test]
+    fn verify_with_cache_without_global_cache_runs_verify_and_collects_nothing() {
+        let _guard = GLOBAL_TEST_LOCK.lock().expect("test lock");
+        clear_global();
+        assert!(global_cache().is_none());
+        let mut executed = 0u64;
+        let result: Result<u64, String> = (|| {
+            verify_with_cache(
+                CryptoSigVerifyPoint::CmsSignedObject,
+                b"input",
+                b"sig",
+                b"spki",
+                || -> Result<(), String> {
+                    executed += 1;
+                    Ok(())
+                },
+            )?;
+            Ok(42)
+        })();
+        assert_eq!(result, Ok(42));
+        assert_eq!(executed, 1);
+        assert!(global_cache().is_none());
+    }
+
+    #[test]
+    fn verify_with_cache_preserves_result_through_global_install() {
+        let _guard = GLOBAL_TEST_LOCK.lock().expect("test lock");
+        clear_global();
+        let dir = tempfile::tempdir().expect("tempdir");
+        let cache = Arc::new(CryptoSigCache::load_or_rebuild(
+            dir.path().join("work-db.crypto-sig-cache"),
+            true,
+        ));
+        install_global(Arc::clone(&cache));
+
+        let ok: Result<(), String> = verify_with_cache(
+            CryptoSigVerifyPoint::ChildCaCert,
+            b"tbs",
+            b"sig",
+            b"spki",
+            || Ok(()),
+        );
+        assert_eq!(ok, Ok(()));
+        let err: Result<(), String> = verify_with_cache(
+            CryptoSigVerifyPoint::ChildCaCert,
+            b"tbs2",
+            b"sig",
+            b"spki",
+            || Err("bad signature".to_string()),
+        );
+        assert_eq!(err, Err("bad signature".to_string()));
+        let summary = cache.summary();
+        let stats = summary
+            .per_point
+            .get("child_ca_cert")
+            .expect("stats present");
+        assert_eq!(stats.calls, 2);
+        assert_eq!(stats.new_keys, 2);
+        clear_global();
+        assert!(global_cache().is_none());
+    }
+}

src/current_repo_index.rs

@@ -1,5 +1,5 @@
 use std::collections::{HashMap, HashSet};
-use std::sync::{Arc, Mutex};
+use std::sync::{Arc, RwLock};
 
 use crate::storage::{RepositoryViewEntry, RepositoryViewState};
 
@@ -25,7 +25,12 @@ pub struct CurrentRepoIndex {
     by_uri: HashMap<String, CurrentRepoEntry>,
 }
 
-pub type CurrentRepoIndexHandle = Arc<Mutex<CurrentRepoIndex>>;
+/// Shared handle to the run-wide current repository index. Readers (phase 2
+/// publication point staging, cache lookups, output snapshots) take
+/// `.read()` and no longer exclude each other; writers (repo sync transports
+/// applying repository view entries, run-state reset) take `.write()` and
+/// stay exclusive.
+pub type CurrentRepoIndexHandle = Arc<RwLock<CurrentRepoIndex>>;
 
 impl CurrentRepoIndex {
     pub fn new() -> Self {
@@ -33,7 +38,7 @@ impl CurrentRepoIndex {
     }
 
     pub fn shared() -> CurrentRepoIndexHandle {
-        Arc::new(Mutex::new(Self::new()))
+        Arc::new(RwLock::new(Self::new()))
     }
 
     pub fn get_by_uri(&self, rsync_uri: &str) -> Option<&CurrentRepoEntry> {
@@ -205,7 +210,7 @@ mod tests {
     #[test]
     fn current_repo_index_snapshot_objects_and_counts_are_sorted() {
         let handle = CurrentRepoIndex::shared();
-        let mut index = handle.lock().expect("lock index");
+        let mut index = handle.write().expect("write-lock index");
         index
             .apply_repository_view_entries(&[
                 present(

src/data_model/crl.rs

@@ -285,8 +285,16 @@ impl RpkixCrl {
         if !rem.is_empty() {
             return Err(CrlVerifyError::CrlTrailingBytes(rem.len()));
         }
+        crate::crypto_sig_cache::verify_with_cache(
+            crate::crypto_sig_cache::CryptoSigVerifyPoint::Crl,
+            crl.tbs_cert_list.as_ref(),
+            crl.signature_value.data.as_ref(),
+            issuer_spki.raw,
+            || {
                 crl.verify_signature(issuer_spki)
                     .map_err(|e| CrlVerifyError::InvalidSignature(e.to_string()))
+            },
+        )
     }
 
     /// Verify the cryptographic signature on this CRL using a DER-encoded SubjectPublicKeyInfo.

src/data_model/signed_object.rs

@@ -358,7 +358,19 @@ impl RpkiSignedObject {
     /// Verify the CMS signature using the embedded EE certificate public key.
     pub fn verify_signature(&self) -> Result<(), SignedObjectVerifyError> {
         let ee = &self.signed_data.certificates[0];
-        self.verify_signature_with_rsa_components(&ee.rsa_public_modulus, &ee.rsa_public_exponent)
+        let signer = &self.signed_data.signer_infos[0];
+        crate::crypto_sig_cache::verify_with_cache(
+            crate::crypto_sig_cache::CryptoSigVerifyPoint::CmsSignedObject,
+            &signer.signed_attrs_der_for_signature,
+            &signer.signature,
+            &ee.spki_der,
+            || {
+                self.verify_signature_with_rsa_components(
+                    &ee.rsa_public_modulus,
+                    &ee.rsa_public_exponent,
+                )
+            },
+        )
     }
 
     /// Verify the CMS signature using a DER-encoded SubjectPublicKeyInfo.
@@ -394,7 +406,14 @@ impl RpkiSignedObject {
             _ => return Err(SignedObjectVerifyError::UnsupportedEePublicKeyAlgorithm),
         };
 
-        self.verify_signature_with_rsa_components(n.as_slice(), e.as_slice())
+        let signer = &self.signed_data.signer_infos[0];
+        crate::crypto_sig_cache::verify_with_cache(
+            crate::crypto_sig_cache::CryptoSigVerifyPoint::CmsSignedObject,
+            &signer.signed_attrs_der_for_signature,
+            &signer.signature,
+            ee_spki.raw,
+            || self.verify_signature_with_rsa_components(n.as_slice(), e.as_slice()),
+        )
     }
 
     fn verify_signature_with_rsa_components(

src/fetch/current_repository.rs

@@ -1,24 +1,33 @@
 use std::sync::Arc;
 
-use crate::fetch::rsync::{
+use crate::fetch::rsync::{RsyncFetchError, RsyncFetchResult, RsyncFetcher};
-    RsyncFetchError, RsyncFetchResult, RsyncFetcher, normalize_rsync_base_uri,
+use crate::fetch::rsync_system::{
+    RsyncScopePolicy, scoped_rsync_failure_dedup_key, scoped_rsync_fetch_uri,
 };
 use crate::storage::{RepositoryViewState, RocksStore};
 
 #[derive(Clone)]
 pub struct CurrentRepositoryViewRsyncFetcher {
     store: Arc<RocksStore>,
+    scope_policy: RsyncScopePolicy,
 }
 
 impl CurrentRepositoryViewRsyncFetcher {
-    pub fn new(store: Arc<RocksStore>) -> Self {
+    pub fn new(store: Arc<RocksStore>, scope_policy: RsyncScopePolicy) -> Self {
-        Self { store }
+        Self {
+            store,
+            scope_policy,
+        }
     }
 }
 
 impl RsyncFetcher for CurrentRepositoryViewRsyncFetcher {
     fn fetch_objects(&self, rsync_base_uri: &str) -> RsyncFetchResult<Vec<(String, Vec<u8>)>> {
-        let base = normalize_rsync_base_uri(rsync_base_uri);
+        // The source run's live fetcher may have fetched a module root rather
+        // than the publication-point URI that triggered it. Read that same
+        // frozen-view prefix here so transport prefetch request identities and
+        // their materialized object sets stay equivalent.
+        let base = scoped_rsync_fetch_uri(self.scope_policy, rsync_base_uri);
         let entries = self
             .store
             .list_repository_view_entries_with_prefix(&base)
@@ -75,7 +84,11 @@ impl RsyncFetcher for CurrentRepositoryViewRsyncFetcher {
     }
 
     fn dedup_key(&self, rsync_base_uri: &str) -> String {
-        normalize_rsync_base_uri(rsync_base_uri)
+        scoped_rsync_fetch_uri(self.scope_policy, rsync_base_uri)
+    }
+
+    fn failure_dedup_key(&self, rsync_base_uri: &str) -> Option<String> {
+        scoped_rsync_failure_dedup_key(self.scope_policy, rsync_base_uri)
     }
 }
 
@@ -107,7 +120,7 @@ mod tests {
                 })
                 .expect("put view");
         }
-        let fetcher = CurrentRepositoryViewRsyncFetcher::new(store);
+        let fetcher = CurrentRepositoryViewRsyncFetcher::new(store, RsyncScopePolicy::default());
         let objects = fetcher
             .fetch_objects("rsync://example.test/repo/")
             .expect("fetch objects");
@@ -168,11 +181,54 @@ mod tests {
                 current_hash: Some(missing_hash),
             })
             .expect("put missing view");
-        let fetcher = CurrentRepositoryViewRsyncFetcher::new(store);
+        let fetcher = CurrentRepositoryViewRsyncFetcher::new(store, RsyncScopePolicy::default());
         let error = fetcher
             .fetch_objects("rsync://example.test/repo/")
             .unwrap_err()
             .to_string();
         assert!(error.contains("blob bytes missing"));
     }
+
+    #[test]
+    fn current_repository_fetcher_replays_module_scope_and_failure_dedup() {
+        let dir = tempfile::tempdir().expect("tempdir");
+        let store = Arc::new(RocksStore::open(dir.path()).expect("open store"));
+        for (uri, bytes) in [
+            ("rsync://example.test/repo/ca/a.mft", b"a".as_slice()),
+            ("rsync://example.test/repo/other/b.roa", b"b".as_slice()),
+        ] {
+            let hash = hex::encode(crate::cir::sha256(bytes));
+            store
+                .put_blob_bytes_batch(&[(hash.clone(), bytes.to_vec())])
+                .expect("put blob");
+            store
+                .put_repository_view_entry(&RepositoryViewEntry {
+                    rsync_uri: uri.to_string(),
+                    repository_source: Some("fixture".to_string()),
+                    object_type: Some("fixture".to_string()),
+                    state: RepositoryViewState::Present,
+                    current_hash: Some(hash),
+                })
+                .expect("put view");
+        }
+
+        let module_fetcher = CurrentRepositoryViewRsyncFetcher::new(
+            Arc::clone(&store),
+            RsyncScopePolicy::ModuleRoot,
+        );
+        let objects = module_fetcher
+            .fetch_objects("rsync://example.test/repo/ca/")
+            .expect("fetch module scope");
+        assert_eq!(objects.len(), 2);
+        assert_eq!(
+            module_fetcher.dedup_key("rsync://example.test/repo/ca/"),
+            "rsync://example.test/repo/"
+        );
+
+        let host_fetcher = CurrentRepositoryViewRsyncFetcher::new(store, RsyncScopePolicy::Host);
+        assert_eq!(
+            host_fetcher.failure_dedup_key("rsync://example.test/repo/ca/"),
+            Some("rsync://example.test/".to_string())
+        );
+    }
 }

src/fetch/http.rs

@@ -20,6 +20,28 @@ pub fn with_scoped_http_timeout_override<R>(timeout: Duration, f: impl FnOnce()
     })
 }
 
+/// Default User-Agent sent with every outgoing HTTP request (RRDP
+/// notification/snapshot/delta, TAL/TA downloads, daemon probes).
+pub const DEFAULT_HTTP_USER_AGENT: &str = "panda-rpki/0.2";
+
+/// Environment variable that overrides the HTTP User-Agent. Unset, empty, or
+/// values containing characters that are illegal in a header value fall back
+/// to [`DEFAULT_HTTP_USER_AGENT`].
+pub const HTTP_USER_AGENT_ENV: &str = "RPKI_HTTP_USER_AGENT";
+
+fn resolve_http_user_agent(env_value: Option<String>) -> String {
+    let Some(value) = env_value else {
+        return DEFAULT_HTTP_USER_AGENT.to_string();
+    };
+    let trimmed = value.trim();
+    let valid = !trimmed.is_empty() && trimmed.bytes().all(|b| (0x20..=0x7e).contains(&b));
+    if valid {
+        trimmed.to_string()
+    } else {
+        DEFAULT_HTTP_USER_AGENT.to_string()
+    }
+}
+
 #[derive(Clone, Debug)]
 pub struct HttpFetcherConfig {
     /// Connection-establishment timeout for HTTP requests.
@@ -39,7 +61,7 @@ impl Default for HttpFetcherConfig {
             connect_timeout: Duration::from_secs(15),
             timeout: Duration::from_secs(30),
             large_body_timeout: Duration::from_secs(180),
-            user_agent: "rpki-dev/0.1 (stage2)".to_string(),
+            user_agent: resolve_http_user_agent(std::env::var(HTTP_USER_AGENT_ENV).ok()),
             extra_root_certificates_pem: Vec::new(),
         }
     }
@@ -84,7 +106,7 @@ impl BlockingHttpFetcher {
             &config,
             Duration::from_secs(1),
             Duration::from_secs(1),
-            "rpki-dev/0.1 (stage2)".to_string(),
+            config.user_agent.clone(),
         )?
         .build()
         .map_err(|e| e.to_string())?;
@@ -532,4 +554,99 @@ mod tests {
         assert_eq!(profile_large, "large_body");
         assert_eq!(timeout_large, Duration::from_secs(9));
     }
+
+    #[test]
+    fn resolve_http_user_agent_falls_back_to_default() {
+        assert_eq!(resolve_http_user_agent(None), DEFAULT_HTTP_USER_AGENT);
+        assert_eq!(
+            resolve_http_user_agent(Some(String::new())),
+            DEFAULT_HTTP_USER_AGENT
+        );
+        assert_eq!(
+            resolve_http_user_agent(Some("   ".to_string())),
+            DEFAULT_HTTP_USER_AGENT
+        );
+        // Control characters are illegal in header values; fall back instead
+        // of letting reqwest fail to build the client.
+        assert_eq!(
+            resolve_http_user_agent(Some("bad\u{1}ua".to_string())),
+            DEFAULT_HTTP_USER_AGENT
+        );
+        assert_eq!(
+            resolve_http_user_agent(Some("non-ascii-ua-中".to_string())),
+            DEFAULT_HTTP_USER_AGENT
+        );
+    }
+
+    #[test]
+    fn resolve_http_user_agent_accepts_custom_value() {
+        assert_eq!(
+            resolve_http_user_agent(Some("panda-rpki/9.9 (test)".to_string())),
+            "panda-rpki/9.9 (test)"
+        );
+        // Surrounding whitespace is trimmed.
+        assert_eq!(
+            resolve_http_user_agent(Some("  custom/1.0  ".to_string())),
+            "custom/1.0"
+        );
+    }
+
+    fn spawn_user_agent_capture_server() -> (String, std::sync::mpsc::Receiver<String>) {
+        let listener = TcpListener::bind(("127.0.0.1", 0)).expect("bind");
+        let addr = listener.local_addr().expect("addr");
+        let (tx, rx) = std::sync::mpsc::channel();
+        thread::spawn(move || {
+            let (mut stream, _) = listener.accept().expect("accept");
+            let mut buf = [0u8; 4096];
+            let read = stream.read(&mut buf).expect("read request");
+            let request = String::from_utf8_lossy(&buf[..read]).to_string();
+            let body = b"ok";
+            let hdr = format!(
+                "HTTP/1.1 200 OK\r\nContent-Length: {}\r\nConnection: close\r\n\r\n",
+                body.len()
+            );
+            stream.write_all(hdr.as_bytes()).expect("write hdr");
+            stream.write_all(body).expect("write body");
+            let _ = tx.send(request);
+        });
+        (format!("http://{}/", addr), rx)
+    }
+
+    #[test]
+    fn fetch_bytes_sends_configured_user_agent_header() {
+        let (url, rx) = spawn_user_agent_capture_server();
+        let http = BlockingHttpFetcher::new(HttpFetcherConfig {
+            timeout: Duration::from_secs(2),
+            user_agent: "panda-rpki/0.2".to_string(),
+            ..HttpFetcherConfig::default()
+        })
+        .expect("http");
+        http.fetch_bytes(&url).expect("fetch");
+        let request = rx.recv().expect("request captured");
+        assert!(
+            request
+                .to_ascii_lowercase()
+                .contains("user-agent: panda-rpki/0.2"),
+            "{request}"
+        );
+    }
+
+    #[test]
+    fn fetch_bytes_sends_custom_user_agent_header() {
+        let (url, rx) = spawn_user_agent_capture_server();
+        let http = BlockingHttpFetcher::new(HttpFetcherConfig {
+            timeout: Duration::from_secs(2),
+            user_agent: "custom-marker/7.7".to_string(),
+            ..HttpFetcherConfig::default()
+        })
+        .expect("http");
+        http.fetch_bytes(&url).expect("fetch");
+        let request = rx.recv().expect("request captured");
+        assert!(
+            request
+                .to_ascii_lowercase()
+                .contains("user-agent: custom-marker/7.7"),
+            "{request}"
+        );
+    }
 }

src/fetch/rsync_system.rs

@@ -13,7 +13,8 @@ use crate::fetch::rsync::{
     RsyncFetchError, RsyncFetchResult, RsyncFetcher, normalize_rsync_base_uri,
 };
 
-#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+#[derive(Clone, Copy, Debug, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
+#[serde(rename_all = "kebab-case")]
 pub enum RsyncScopePolicy {
     Host,
     PublicationPoint,
@@ -297,13 +298,7 @@ impl SystemRsyncFetcher {
     }
 
     fn scope_fetch_uri(&self, rsync_base_uri: &str) -> String {
-        match self.config.scope_policy {
+        scoped_rsync_fetch_uri(self.config.scope_policy, rsync_base_uri)
-            RsyncScopePolicy::Host | RsyncScopePolicy::PublicationPoint => {
-                normalize_rsync_base_uri(rsync_base_uri)
-            }
-            RsyncScopePolicy::ModuleRoot => rsync_module_root_uri(rsync_base_uri)
-                .unwrap_or_else(|| normalize_rsync_base_uri(rsync_base_uri)),
-        }
     }
 }
 
@@ -382,11 +377,36 @@ impl RsyncFetcher for SystemRsyncFetcher {
     }
 
     fn failure_dedup_key(&self, rsync_base_uri: &str) -> Option<String> {
-        match self.config.scope_policy {
+        scoped_rsync_failure_dedup_key(self.config.scope_policy, rsync_base_uri)
+    }
+}
+
+/// Return the exact successful-fetch scope used by the live rsync fetcher.
+///
+/// Verification-only reuses this function while reading the frozen repository
+/// view, so its request keys and object projection scope remain equivalent to
+/// the source run without granting it network access.
+pub fn scoped_rsync_fetch_uri(scope_policy: RsyncScopePolicy, rsync_base_uri: &str) -> String {
+    match scope_policy {
+        RsyncScopePolicy::Host | RsyncScopePolicy::PublicationPoint => {
+            normalize_rsync_base_uri(rsync_base_uri)
+        }
+        RsyncScopePolicy::ModuleRoot => rsync_module_root_uri(rsync_base_uri)
+            .unwrap_or_else(|| normalize_rsync_base_uri(rsync_base_uri)),
+    }
+}
+
+/// Return the live fetcher's failure-deduplication key for the configured
+/// scope. This is part of the transport request identity, even though host
+/// scope deliberately does not widen the successful fetch URI.
+pub fn scoped_rsync_failure_dedup_key(
+    scope_policy: RsyncScopePolicy,
+    rsync_base_uri: &str,
+) -> Option<String> {
+    match scope_policy {
         RsyncScopePolicy::Host => rsync_host_scope_uri(rsync_base_uri),
         RsyncScopePolicy::PublicationPoint | RsyncScopePolicy::ModuleRoot => None,
     }
-    }
 }
 
 fn rsync_host_scope_uri(rsync_base_uri: &str) -> Option<String> {

src/lib.rs

@@ -1,5 +1,6 @@
 pub mod ccr;
 pub mod cir;
+pub mod crypto_sig_cache;
 pub mod data_model;
 
 #[cfg(feature = "full")]

src/parallel/config.rs

@@ -1,8 +1,13 @@
+use crate::parallel::dead_repo_blacklist::DeadRepoBlacklistConfig;
+
 #[derive(Clone, Debug, PartialEq, Eq)]
 pub struct ParallelPhase1Config {
     pub max_repo_sync_workers_global: usize,
     pub max_inflight_snapshot_bytes_global: usize,
     pub max_pending_repo_results: usize,
+    /// Dead-repo transport blacklist (#141). `None` disables the feature
+    /// entirely (default, behavior unchanged).
+    pub dead_repo_blacklist: Option<DeadRepoBlacklistConfig>,
 }
 
 #[derive(Clone, Debug, PartialEq, Eq)]
@@ -15,6 +20,9 @@ pub struct ParallelPhase2Config {
     pub publication_point_finalize_batch_size: usize,
     pub publication_point_finalize_wall_time_budget_ms: u64,
     pub publication_point_finalize_queue_capacity: usize,
+    /// Experimental: workers of the ready publication point stage pool. `0`
+    /// disables the pool and keeps the inline compute+apply staging path.
+    pub stage_workers: usize,
 }
 
 impl Default for ParallelPhase2Config {
@@ -28,6 +36,7 @@ impl Default for ParallelPhase2Config {
             publication_point_finalize_batch_size: 256,
             publication_point_finalize_wall_time_budget_ms: 100,
             publication_point_finalize_queue_capacity: 32768,
+            stage_workers: 0,
         }
     }
 }
@@ -38,6 +47,7 @@ impl Default for ParallelPhase1Config {
             max_repo_sync_workers_global: 4,
             max_inflight_snapshot_bytes_global: 512 * 1024 * 1024,
             max_pending_repo_results: 1024,
+            dead_repo_blacklist: None,
         }
     }
 }

src/parallel/dead_repo_blacklist.rs

@@ -0,0 +1,376 @@
+//! Persistent dead-repo transport blacklist (#141).
+//!
+//! Tracks per-(repo, transport) consecutive transport-layer fetch failures
+//! across runs. Once an entry reaches the configured failure threshold it is
+//! admitted to the blacklist; the live transport scheduler then skips the
+//! dead transport (rrdp-blacklisted -> straight to rsync, dual-blacklisted ->
+//! immediate terminal failure). Entries are removed either by a successful
+//! fetch (self-heal) or by the daemon-side health probe.
+
+use std::collections::HashMap;
+use std::fs;
+use std::io;
+use std::path::{Path, PathBuf};
+
+use serde::{Deserialize, Serialize};
+
+use crate::parallel::types::{RepoTransportErrorClass, RepoTransportMode};
+
+pub const DEAD_REPO_BLACKLIST_SCHEMA_VERSION: u32 = 1;
+pub const DEAD_REPO_BLACKLIST_DEFAULT_CAPACITY: usize = 256;
+/// Probe backoff is capped at interval * 2^PROBE_BACKOFF_CAP_SHIFT.
+pub const PROBE_BACKOFF_CAP_SHIFT: u32 = 4;
+
+#[derive(Clone, Debug, PartialEq, Eq)]
+pub struct DeadRepoBlacklistConfig {
+    pub path: PathBuf,
+    pub fail_threshold: u32,
+    pub capacity: usize,
+}
+
+#[derive(Clone, Debug, PartialEq, Eq, Serialize, Deserialize)]
+pub struct DeadRepoBlacklistEntry {
+    pub transport: RepoTransportMode,
+    pub uri: String,
+    /// Consecutive runs with transport-class fetch failure. Reset on success.
+    pub consecutive_failures: u32,
+    /// True once `consecutive_failures` reached the admission threshold.
+    #[serde(default)]
+    pub blacklisted: bool,
+    #[serde(default)]
+    pub first_failure_at_unix: u64,
+    #[serde(default)]
+    pub last_failure_at_unix: u64,
+    /// When the entry was admitted to the blacklist.
+    #[serde(default)]
+    pub added_at_unix: u64,
+    #[serde(default)]
+    pub last_probe_at_unix: Option<u64>,
+    #[serde(default)]
+    pub probe_failures: u32,
+}
+
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+pub enum DeadRepoFailureOutcome {
+    /// Failure counted, entry not (or not yet) blacklisted.
+    Counted { consecutive_failures: u32 },
+    /// Entry just reached the threshold and was admitted.
+    Admitted,
+}
+
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+pub enum DeadRepoSuccessOutcome {
+    /// A counting (not yet admitted) entry was reset.
+    CounterReset,
+    /// A blacklisted entry was removed (repo revived).
+    BlacklistRemoved,
+}
+
+#[derive(Serialize, Deserialize)]
+struct DeadRepoBlacklistFile {
+    schema_version: u32,
+    updated_at_unix: u64,
+    #[serde(default)]
+    entries: Vec<DeadRepoBlacklistEntry>,
+}
+
+/// In-memory blacklist state for one writer at a time (child run, or daemon
+/// supervisor between runs). Lookups go through a HashMap keyed by
+/// (transport, uri).
+#[derive(Clone, Debug, Default)]
+pub struct DeadRepoBlacklist {
+    entries: HashMap<(RepoTransportMode, String), DeadRepoBlacklistEntry>,
+}
+
+impl DeadRepoBlacklist {
+    pub fn new() -> Self {
+        Self::default()
+    }
+
+    /// Load from disk. A missing file or an unreadable/incompatible file
+    /// degrades to an empty blacklist (logged by the caller via `warning`).
+    pub fn load(path: &Path) -> (Self, Option<String>) {
+        let bytes = match fs::read(path) {
+            Ok(bytes) => bytes,
+            Err(err) if err.kind() == io::ErrorKind::NotFound => return (Self::new(), None),
+            Err(err) => {
+                return (
+                    Self::new(),
+                    Some(format!(
+                        "dead repo blacklist {} unreadable ({err}); starting empty",
+                        path.display()
+                    )),
+                );
+            }
+        };
+        match serde_json::from_slice::<DeadRepoBlacklistFile>(&bytes) {
+            Ok(file) if file.schema_version == DEAD_REPO_BLACKLIST_SCHEMA_VERSION => {
+                let entries = file
+                    .entries
+                    .into_iter()
+                    .map(|entry| ((entry.transport, entry.uri.clone()), entry))
+                    .collect();
+                (Self { entries }, None)
+            }
+            Ok(file) => (
+                Self::new(),
+                Some(format!(
+                    "dead repo blacklist {} schema_version {} != {}; starting empty",
+                    path.display(),
+                    file.schema_version,
+                    DEAD_REPO_BLACKLIST_SCHEMA_VERSION
+                )),
+            ),
+            Err(err) => (
+                Self::new(),
+                Some(format!(
+                    "dead repo blacklist {} unparsable ({err}); starting empty",
+                    path.display()
+                )),
+            ),
+        }
+    }
+
+    /// Persist atomically (tmp file + rename).
+    pub fn store_atomic(&self, path: &Path, now_unix: u64) -> io::Result<()> {
+        if let Some(parent) = path.parent() {
+            fs::create_dir_all(parent)?;
+        }
+        let mut entries: Vec<&DeadRepoBlacklistEntry> = self.entries.values().collect();
+        entries.sort_by(|a, b| a.uri.cmp(&b.uri).then(a.transport.cmp(&b.transport)));
+        let file = DeadRepoBlacklistFile {
+            schema_version: DEAD_REPO_BLACKLIST_SCHEMA_VERSION,
+            updated_at_unix: now_unix,
+            entries: entries.into_iter().cloned().collect(),
+        };
+        let bytes = serde_json::to_vec_pretty(&file)
+            .map_err(|err| io::Error::new(io::ErrorKind::InvalidData, err))?;
+        let mut tmp_name = path.file_name().map(|name| name.to_os_string()).unwrap_or_default();
+        tmp_name.push(".tmp");
+        let tmp_path = path.with_file_name(tmp_name);
+        fs::write(&tmp_path, &bytes)?;
+        // Best-effort durability before the rename.
+        if let Ok(file) = fs::File::open(&tmp_path) {
+            let _ = file.sync_all();
+        }
+        fs::rename(&tmp_path, path)?;
+        Ok(())
+    }
+
+    pub fn is_blacklisted(&self, transport: RepoTransportMode, uri: &str) -> bool {
+        self.entries
+            .get(&(transport, uri.to_string()))
+            .map(|entry| entry.blacklisted)
+            .unwrap_or(false)
+    }
+
+    /// Record a transport-class fetch failure. Returns whether the failure
+    /// newly admitted the entry to the blacklist.
+    pub fn record_transport_failure(
+        &mut self,
+        transport: RepoTransportMode,
+        uri: &str,
+        now_unix: u64,
+        fail_threshold: u32,
+        capacity: usize,
+    ) -> DeadRepoFailureOutcome {
+        let key = (transport, uri.to_string());
+        if !self.entries.contains_key(&key) {
+            self.evict_if_full(capacity);
+        }
+        let threshold = fail_threshold.max(1);
+        let entry = self.entries.entry(key).or_insert_with(|| DeadRepoBlacklistEntry {
+            transport,
+            uri: uri.to_string(),
+            consecutive_failures: 0,
+            blacklisted: false,
+            first_failure_at_unix: now_unix,
+            last_failure_at_unix: now_unix,
+            added_at_unix: 0,
+            last_probe_at_unix: None,
+            probe_failures: 0,
+        });
+        entry.consecutive_failures = entry.consecutive_failures.saturating_add(1);
+        entry.last_failure_at_unix = now_unix;
+        if entry.first_failure_at_unix == 0 {
+            entry.first_failure_at_unix = now_unix;
+        }
+        if !entry.blacklisted && entry.consecutive_failures >= threshold {
+            entry.blacklisted = true;
+            entry.added_at_unix = now_unix;
+            // A fresh admission restarts probe bookkeeping.
+            entry.last_probe_at_unix = None;
+            entry.probe_failures = 0;
+            return DeadRepoFailureOutcome::Admitted;
+        }
+        DeadRepoFailureOutcome::Counted {
+            consecutive_failures: entry.consecutive_failures,
+        }
+    }
+
+    /// Record a successful fetch: reset a counting entry, or remove a
+    /// blacklisted entry entirely (self-heal).
+    pub fn record_transport_success(
+        &mut self,
+        transport: RepoTransportMode,
+        uri: &str,
+    ) -> Option<DeadRepoSuccessOutcome> {
+        match self.entries.remove(&(transport, uri.to_string())) {
+            Some(entry) if entry.blacklisted => Some(DeadRepoSuccessOutcome::BlacklistRemoved),
+            Some(_) => Some(DeadRepoSuccessOutcome::CounterReset),
+            None => None,
+        }
+    }
+
+    /// Blacklisted entries whose next probe is due at `now_unix`.
+    pub fn probe_due_entries(
+        &self,
+        now_unix: u64,
+        probe_interval_secs: u64,
+    ) -> Vec<DeadRepoBlacklistEntry> {
+        let mut due: Vec<DeadRepoBlacklistEntry> = self
+            .entries
+            .values()
+            .filter(|entry| entry.blacklisted)
+            .filter(|entry| {
+                let interval = probe_backoff_secs(probe_interval_secs, entry.probe_failures);
+                match entry.last_probe_at_unix {
+                    None => true,
+                    Some(last) => now_unix.saturating_sub(last) >= interval,
+                }
+            })
+            .cloned()
+            .collect();
+        due.sort_by(|a, b| a.uri.cmp(&b.uri));
+        due
+    }
+
+    /// Probe succeeded: remove the entry. Returns true if it was present.
+    pub fn record_probe_success(&mut self, transport: RepoTransportMode, uri: &str) -> bool {
+        self.entries.remove(&(transport, uri.to_string())).is_some()
+    }
+
+    pub fn record_probe_failure(&mut self, transport: RepoTransportMode, uri: &str, now_unix: u64) {
+        if let Some(entry) = self.entries.get_mut(&(transport, uri.to_string())) {
+            entry.last_probe_at_unix = Some(now_unix);
+            entry.probe_failures = entry.probe_failures.saturating_add(1);
+        }
+    }
+
+    pub fn len(&self) -> usize {
+        self.entries.len()
+    }
+
+    pub fn is_empty(&self) -> bool {
+        self.entries.is_empty()
+    }
+
+    pub fn blacklisted_len(&self) -> usize {
+        self.entries.values().filter(|entry| entry.blacklisted).count()
+    }
+
+    /// Deterministic snapshot of all entries (for status/observability).
+    pub fn entries_sorted(&self) -> Vec<DeadRepoBlacklistEntry> {
+        let mut entries: Vec<DeadRepoBlacklistEntry> = self.entries.values().cloned().collect();
+        entries.sort_by(|a, b| a.uri.cmp(&b.uri).then(a.transport.cmp(&b.transport)));
+        entries
+    }
+
+    fn evict_if_full(&mut self, capacity: usize) {
+        let capacity = capacity.max(1);
+        if self.entries.len() < capacity {
+            return;
+        }
+        // Prefer evicting the stalest counting entry; fall back to the oldest
+        // blacklisted entry.
+        let victim = self
+            .entries
+            .values()
+            .filter(|entry| !entry.blacklisted)
+            .min_by_key(|entry| entry.last_failure_at_unix)
+            .map(|entry| (entry.transport, entry.uri.clone()))
+            .or_else(|| {
+                self.entries
+                    .values()
+                    .min_by_key(|entry| entry.added_at_unix)
+                    .map(|entry| (entry.transport, entry.uri.clone()))
+            });
+        if let Some(key) = victim {
+            self.entries.remove(&key);
+        }
+    }
+}
+
+pub fn probe_backoff_secs(base_interval_secs: u64, probe_failures: u32) -> u64 {
+    // First retry after a probe failure still uses the base interval;
+    // exponential backoff kicks in from the second consecutive failure.
+    let shift = probe_failures.saturating_sub(1).min(PROBE_BACKOFF_CAP_SHIFT);
+    base_interval_secs.saturating_mul(1u64 << shift)
+}
+
+/// Classify an HTTP fetch error string (formats from
+/// `fetch/http.rs::BlockingHttpFetcher::fetch_bytes`).
+pub fn classify_http_fetch_error(detail: &str) -> RepoTransportErrorClass {
+    if detail.starts_with("http status") {
+        // The peer answered: transport is fine, the failure is protocol-level.
+        RepoTransportErrorClass::Protocol
+    } else if detail.starts_with("http request failed") || detail.starts_with("http read body failed") {
+        RepoTransportErrorClass::TransportFetch
+    } else {
+        RepoTransportErrorClass::Unknown
+    }
+}
+
+/// Classify an rsync fetch error string using the same host-level substring
+/// set as the run-local rsync failure scope short-circuit.
+pub fn classify_rsync_fetch_error(detail: &str) -> RepoTransportErrorClass {
+    if crate::parallel::repo_scheduler::is_host_level_rsync_failure(detail)
+        // Stall failures are transport death too: the fail-fast mechanism
+        // gives up when wall-clock windows pass with no (additional) progress
+        // (#141 M9). Content-level errors fail immediately and never carry
+        // this marker, so they still do not count.
+        || detail.contains("rsync fail-fast gave up")
+    {
+        RepoTransportErrorClass::TransportFetch
+    } else {
+        // Content/protocol-level rsync failures (digest mismatch, vanished
+        // files, partial transfers) never count toward the blacklist.
+        RepoTransportErrorClass::Unknown
+    }
+}
+
+/// Map a structured RRDP sync error to a transport error class.
+pub fn classify_rrdp_sync_error(err: &crate::sync::rrdp::RrdpSyncError) -> RepoTransportErrorClass {
+    match err {
+        crate::sync::rrdp::RrdpSyncError::Fetch(detail) => classify_http_fetch_error(detail),
+        crate::sync::rrdp::RrdpSyncError::Rrdp(_) => RepoTransportErrorClass::Protocol,
+        crate::sync::rrdp::RrdpSyncError::Storage(_) => RepoTransportErrorClass::Storage,
+    }
+}
+
+/// Map a structured rsync-side repo sync error to a transport error class.
+pub fn classify_rsync_repo_sync_error(err: &crate::sync::repo::RepoSyncError) -> RepoTransportErrorClass {
+    match err {
+        crate::sync::repo::RepoSyncError::Rsync(crate::fetch::rsync::RsyncFetchError::Fetch(detail)) => {
+            classify_rsync_fetch_error(detail)
+        }
+        crate::sync::repo::RepoSyncError::Storage(_) => RepoTransportErrorClass::Storage,
+        _ => RepoTransportErrorClass::Protocol,
+    }
+}
+
+impl RepoTransportMode {
+    pub fn as_str(self) -> &'static str {
+        match self {
+            RepoTransportMode::Rrdp => "rrdp",
+            RepoTransportMode::Rsync => "rsync",
+        }
+    }
+}
+
+// Keep the public surface free of internal key types; the (transport, uri)
+// tuple key is an implementation detail of the lookup map.
+
+#[cfg(test)]
+#[path = "dead_repo_blacklist_tests.rs"]
+mod tests;

src/parallel/dead_repo_blacklist_tests.rs

@@ -0,0 +1,360 @@
+use super::*;
+use crate::parallel::types::RepoTransportMode;
+
+const RRDP_URI: &str = "https://dead.example.com/notification.xml";
+const RSYNC_URI: &str = "rsync://dead.example.com/repo/";
+
+fn now() -> u64 {
+    1_700_000_000
+}
+
+#[test]
+fn admission_requires_threshold_consecutive_failures() {
+    let mut bl = DeadRepoBlacklist::new();
+    assert!(!bl.is_blacklisted(RepoTransportMode::Rrdp, RRDP_URI));
+
+    let out1 = bl.record_transport_failure(RepoTransportMode::Rrdp, RRDP_URI, now(), 3, 256);
+    assert_eq!(
+        out1,
+        DeadRepoFailureOutcome::Counted {
+            consecutive_failures: 1
+        }
+    );
+    assert!(!bl.is_blacklisted(RepoTransportMode::Rrdp, RRDP_URI));
+
+    let out2 = bl.record_transport_failure(RepoTransportMode::Rrdp, RRDP_URI, now() + 1, 3, 256);
+    assert_eq!(
+        out2,
+        DeadRepoFailureOutcome::Counted {
+            consecutive_failures: 2
+        }
+    );
+    assert!(!bl.is_blacklisted(RepoTransportMode::Rrdp, RRDP_URI));
+
+    let out3 = bl.record_transport_failure(RepoTransportMode::Rrdp, RRDP_URI, now() + 2, 3, 256);
+    assert_eq!(out3, DeadRepoFailureOutcome::Admitted);
+    assert!(bl.is_blacklisted(RepoTransportMode::Rrdp, RRDP_URI));
+
+    // Further failures keep counting but do not re-admit.
+    let out4 = bl.record_transport_failure(RepoTransportMode::Rrdp, RRDP_URI, now() + 3, 3, 256);
+    assert!(matches!(
+        out4,
+        DeadRepoFailureOutcome::Counted {
+            consecutive_failures: 4
+        }
+    ));
+}
+
+#[test]
+fn threshold_of_one_admits_immediately() {
+    let mut bl = DeadRepoBlacklist::new();
+    let out = bl.record_transport_failure(RepoTransportMode::Rsync, RSYNC_URI, now(), 1, 256);
+    assert_eq!(out, DeadRepoFailureOutcome::Admitted);
+}
+
+#[test]
+fn success_resets_counting_entry() {
+    let mut bl = DeadRepoBlacklist::new();
+    bl.record_transport_failure(RepoTransportMode::Rrdp, RRDP_URI, now(), 3, 256);
+    assert_eq!(bl.len(), 1);
+    let out = bl.record_transport_success(RepoTransportMode::Rrdp, RRDP_URI);
+    assert_eq!(out, Some(DeadRepoSuccessOutcome::CounterReset));
+    assert!(bl.is_empty());
+    // Next failure starts from zero again.
+    let out = bl.record_transport_failure(RepoTransportMode::Rrdp, RRDP_URI, now(), 3, 256);
+    assert_eq!(
+        out,
+        DeadRepoFailureOutcome::Counted {
+            consecutive_failures: 1
+        }
+    );
+}
+
+#[test]
+fn success_on_blacklisted_entry_self_heals() {
+    let mut bl = DeadRepoBlacklist::new();
+    for i in 0..3 {
+        bl.record_transport_failure(RepoTransportMode::Rsync, RSYNC_URI, now() + i, 3, 256);
+    }
+    assert!(bl.is_blacklisted(RepoTransportMode::Rsync, RSYNC_URI));
+    let out = bl.record_transport_success(RepoTransportMode::Rsync, RSYNC_URI);
+    assert_eq!(out, Some(DeadRepoSuccessOutcome::BlacklistRemoved));
+    assert!(!bl.is_blacklisted(RepoTransportMode::Rsync, RSYNC_URI));
+    assert!(bl.is_empty());
+}
+
+#[test]
+fn success_on_unknown_entry_is_none() {
+    let mut bl = DeadRepoBlacklist::new();
+    assert_eq!(
+        bl.record_transport_success(RepoTransportMode::Rrdp, RRDP_URI),
+        None
+    );
+}
+
+#[test]
+fn entries_are_per_transport() {
+    let mut bl = DeadRepoBlacklist::new();
+    for i in 0..3 {
+        bl.record_transport_failure(RepoTransportMode::Rrdp, RRDP_URI, now() + i, 3, 256);
+    }
+    assert!(bl.is_blacklisted(RepoTransportMode::Rrdp, RRDP_URI));
+    assert!(!bl.is_blacklisted(RepoTransportMode::Rsync, RRDP_URI));
+}
+
+#[test]
+fn capacity_evicts_stale_counting_entries_first() {
+    let mut bl = DeadRepoBlacklist::new();
+    // One blacklisted entry (old) and two counting entries.
+    for i in 0..3 {
+        bl.record_transport_failure(RepoTransportMode::Rrdp, "https://a.example/n.xml", 100 + i, 1, 3);
+    }
+    bl.record_transport_failure(RepoTransportMode::Rrdp, "https://b.example/n.xml", 200, 3, 3);
+    bl.record_transport_failure(RepoTransportMode::Rrdp, "https://c.example/n.xml", 300, 3, 3);
+    assert_eq!(bl.len(), 3);
+    assert_eq!(bl.blacklisted_len(), 1);
+
+    // Inserting a fourth entry evicts the stalest counting entry (b).
+    bl.record_transport_failure(RepoTransportMode::Rrdp, "https://d.example/n.xml", 400, 3, 3);
+    assert_eq!(bl.len(), 3);
+    assert!(bl.entries_sorted().iter().any(|e| e.uri == "https://a.example/n.xml"));
+    assert!(!bl.entries_sorted().iter().any(|e| e.uri == "https://b.example/n.xml"));
+    assert!(bl.entries_sorted().iter().any(|e| e.uri == "https://c.example/n.xml"));
+    assert!(bl.entries_sorted().iter().any(|e| e.uri == "https://d.example/n.xml"));
+}
+
+#[test]
+fn capacity_evicts_oldest_blacklisted_when_no_counting_entries() {
+    let mut bl = DeadRepoBlacklist::new();
+    bl.record_transport_failure(RepoTransportMode::Rrdp, "https://a.example/n.xml", 100, 1, 2);
+    bl.record_transport_failure(RepoTransportMode::Rrdp, "https://b.example/n.xml", 200, 1, 2);
+    assert_eq!(bl.blacklisted_len(), 2);
+    bl.record_transport_failure(RepoTransportMode::Rrdp, "https://c.example/n.xml", 300, 1, 2);
+    assert_eq!(bl.len(), 2);
+    assert!(!bl.entries_sorted().iter().any(|e| e.uri == "https://a.example/n.xml"));
+    assert!(bl.entries_sorted().iter().any(|e| e.uri == "https://b.example/n.xml"));
+    assert!(bl.entries_sorted().iter().any(|e| e.uri == "https://c.example/n.xml"));
+}
+
+#[test]
+fn store_load_roundtrip() {
+    let dir = std::env::temp_dir().join(format!(
+        "dead_repo_blacklist_test_roundtrip_{}_{}",
+        std::process::id(),
+        now()
+    ));
+    let path = dir.join("state").join("dead-repo-blacklist.json");
+    let mut bl = DeadRepoBlacklist::new();
+    for i in 0..3 {
+        bl.record_transport_failure(RepoTransportMode::Rrdp, RRDP_URI, now() + i, 3, 256);
+    }
+    bl.record_transport_failure(RepoTransportMode::Rsync, RSYNC_URI, now(), 3, 256);
+    bl.store_atomic(&path, now() + 10).unwrap();
+
+    let (loaded, warning) = DeadRepoBlacklist::load(&path);
+    assert!(warning.is_none());
+    assert!(loaded.is_blacklisted(RepoTransportMode::Rrdp, RRDP_URI));
+    assert!(!loaded.is_blacklisted(RepoTransportMode::Rsync, RSYNC_URI));
+    assert_eq!(loaded.len(), 2);
+    let entry = loaded
+        .entries_sorted()
+        .into_iter()
+        .find(|e| e.uri == RRDP_URI)
+        .unwrap();
+    assert_eq!(entry.consecutive_failures, 3);
+    assert!(entry.blacklisted);
+    assert_eq!(entry.first_failure_at_unix, now());
+
+    let _ = std::fs::remove_dir_all(&dir);
+}
+
+#[test]
+fn load_missing_file_is_empty_without_warning() {
+    let path = std::env::temp_dir().join(format!(
+        "dead_repo_blacklist_test_missing_{}_{}.json",
+        std::process::id(),
+        now()
+    ));
+    let (bl, warning) = DeadRepoBlacklist::load(&path);
+    assert!(bl.is_empty());
+    assert!(warning.is_none());
+}
+
+#[test]
+fn load_corrupted_file_degrades_to_empty_with_warning() {
+    let path = std::env::temp_dir().join(format!(
+        "dead_repo_blacklist_test_corrupt_{}_{}.json",
+        std::process::id(),
+        now()
+    ));
+    std::fs::write(&path, b"{not json").unwrap();
+    let (bl, warning) = DeadRepoBlacklist::load(&path);
+    assert!(bl.is_empty());
+    assert!(warning.is_some());
+    let _ = std::fs::remove_file(&path);
+}
+
+#[test]
+fn load_wrong_schema_version_degrades_to_empty() {
+    let path = std::env::temp_dir().join(format!(
+        "dead_repo_blacklist_test_schema_{}_{}.json",
+        std::process::id(),
+        now()
+    ));
+    std::fs::write(
+        &path,
+        r#"{"schema_version": 99, "updated_at_unix": 0, "entries": []}"#,
+    )
+    .unwrap();
+    let (bl, warning) = DeadRepoBlacklist::load(&path);
+    assert!(bl.is_empty());
+    assert!(warning.unwrap().contains("schema_version"));
+    let _ = std::fs::remove_file(&path);
+}
+
+#[test]
+fn probe_due_uses_exponential_backoff() {
+    let mut bl = DeadRepoBlacklist::new();
+    for i in 0..3 {
+        bl.record_transport_failure(RepoTransportMode::Rrdp, RRDP_URI, now() + i, 3, 256);
+    }
+    // Never probed: due immediately.
+    let due = bl.probe_due_entries(now() + 100, 600);
+    assert_eq!(due.len(), 1);
+
+    // Probe failure: not due until base interval has passed.
+    bl.record_probe_failure(RepoTransportMode::Rrdp, RRDP_URI, now() + 100);
+    assert!(bl.probe_due_entries(now() + 100 + 599, 600).is_empty());
+    assert_eq!(bl.probe_due_entries(now() + 100 + 600, 600).len(), 1);
+
+    // Second probe failure: backoff doubles to 1200s.
+    bl.record_probe_failure(RepoTransportMode::Rrdp, RRDP_URI, now() + 700);
+    assert!(bl.probe_due_entries(now() + 700 + 1199, 600).is_empty());
+    assert_eq!(bl.probe_due_entries(now() + 700 + 1200, 600).len(), 1);
+}
+
+#[test]
+fn probe_success_removes_entry() {
+    let mut bl = DeadRepoBlacklist::new();
+    for i in 0..3 {
+        bl.record_transport_failure(RepoTransportMode::Rsync, RSYNC_URI, now() + i, 3, 256);
+    }
+    assert!(bl.record_probe_success(RepoTransportMode::Rsync, RSYNC_URI));
+    assert!(bl.is_empty());
+    assert!(!bl.record_probe_success(RepoTransportMode::Rsync, RSYNC_URI));
+}
+
+#[test]
+fn counting_entries_are_not_probed() {
+    let mut bl = DeadRepoBlacklist::new();
+    bl.record_transport_failure(RepoTransportMode::Rrdp, RRDP_URI, now(), 3, 256);
+    assert!(bl.probe_due_entries(now() + 10_000, 600).is_empty());
+}
+
+#[test]
+fn probe_backoff_caps_at_max_shift() {
+    assert_eq!(probe_backoff_secs(600, 0), 600);
+    assert_eq!(probe_backoff_secs(600, 1), 600);
+    assert_eq!(probe_backoff_secs(600, 2), 1200);
+    assert_eq!(probe_backoff_secs(600, 5), 9600);
+    assert_eq!(probe_backoff_secs(600, 100), 9600);
+}
+
+#[test]
+fn http_fetch_error_classification() {
+    use crate::parallel::types::RepoTransportErrorClass;
+    assert_eq!(
+        classify_http_fetch_error("http request failed: reqwest::Error { kind: Connect, .. }"),
+        RepoTransportErrorClass::TransportFetch
+    );
+    assert_eq!(
+        classify_http_fetch_error("http read body failed: connection closed; status=200"),
+        RepoTransportErrorClass::TransportFetch
+    );
+    assert_eq!(
+        classify_http_fetch_error("http status 404 Not Found; content_type=; .."),
+        RepoTransportErrorClass::Protocol
+    );
+    assert_eq!(
+        classify_http_fetch_error("something unexpected"),
+        RepoTransportErrorClass::Unknown
+    );
+}
+
+#[test]
+fn rsync_fetch_error_classification() {
+    use crate::parallel::types::RepoTransportErrorClass;
+    assert_eq!(
+        classify_rsync_fetch_error("rsync error: timeout waiting for daemon connection"),
+        RepoTransportErrorClass::TransportFetch
+    );
+    assert_eq!(
+        classify_rsync_fetch_error("rsync: failed to connect to host: Connection refused"),
+        RepoTransportErrorClass::TransportFetch
+    );
+    assert_eq!(
+        classify_rsync_fetch_error("temporary failure in name resolution"),
+        RepoTransportErrorClass::TransportFetch
+    );
+    // Content-level failures must not count.
+    assert_eq!(
+        classify_rsync_fetch_error("rsync file digest mismatch after download"),
+        RepoTransportErrorClass::Unknown
+    );
+    // Stall failures (#141 M9): fail-fast give-up means the host delivered no
+    // bytes inside the wall-clock window — transport death, must count.
+    assert_eq!(
+        classify_rsync_fetch_error(
+            "rsync fail-fast gave up after 1 attempts with no progress: rsync wall-clock window expired"
+        ),
+        RepoTransportErrorClass::TransportFetch
+    );
+    assert_eq!(
+        classify_rsync_fetch_error(
+            "rsync fail-fast gave up after 3 attempts with no additional progress: rsync wall-clock window expired"
+        ),
+        RepoTransportErrorClass::TransportFetch
+    );
+    // Content-level give-up text without the fail-fast marker stays unknown.
+    assert_eq!(
+        classify_rsync_fetch_error("rsync error: some files vanished before transfer"),
+        RepoTransportErrorClass::Unknown
+    );
+}
+
+#[test]
+fn rrdp_sync_error_classification() {
+    use crate::parallel::types::RepoTransportErrorClass;
+    use crate::sync::rrdp::RrdpSyncError;
+    assert_eq!(
+        classify_rrdp_sync_error(&RrdpSyncError::Fetch(
+            "http request failed: connect timeout".to_string()
+        )),
+        RepoTransportErrorClass::TransportFetch
+    );
+    assert_eq!(
+        classify_rrdp_sync_error(&RrdpSyncError::Fetch("http status 500".to_string())),
+        RepoTransportErrorClass::Protocol
+    );
+    assert_eq!(
+        classify_rrdp_sync_error(&RrdpSyncError::Storage("db full".to_string())),
+        RepoTransportErrorClass::Storage
+    );
+}
+
+#[test]
+fn readmission_after_probe_removal_starts_fresh() {
+    let mut bl = DeadRepoBlacklist::new();
+    for i in 0..3 {
+        bl.record_transport_failure(RepoTransportMode::Rrdp, RRDP_URI, now() + i, 3, 256);
+    }
+    assert!(bl.record_probe_success(RepoTransportMode::Rrdp, RRDP_URI));
+    // No half-open: needs the full threshold again.
+    let out = bl.record_transport_failure(RepoTransportMode::Rrdp, RRDP_URI, now() + 10, 3, 256);
+    assert_eq!(
+        out,
+        DeadRepoFailureOutcome::Counted {
+            consecutive_failures: 1
+        }
+    );
+}

src/parallel/mod.rs

@@ -1,4 +1,5 @@
 pub mod config;
+pub mod dead_repo_blacklist;
 pub mod object_worker;
 pub mod phase2_scheduler;
 pub mod repo_runtime;

src/parallel/object_worker.rs

@@ -1,9 +1,13 @@
+use std::marker::PhantomData;
 use std::sync::Arc;
 use std::sync::mpsc::{self, Receiver, RecvTimeoutError, SyncSender, TrySendError};
-use std::thread::{self, JoinHandle};
+use std::thread::{self, JoinHandle, Scope};
 use std::time::Duration;
 
-pub trait ObjectTaskExecutor<T, R>: Send + Sync + 'static {
+/// Executor shared by every worker of a pool. The `'static` bound required to
+/// move the executor into detached threads is expressed on the pool types
+/// instead of this trait so scoped pools can borrow their environment.
+pub trait ObjectTaskExecutor<T, R>: Send + Sync {
     fn execute(&self, worker_index: usize, task: T) -> R;
 }
 
@@ -22,7 +26,7 @@ pub struct ObjectWorkerPool<T, R, E>
 where
     T: Send + 'static,
     R: Send + 'static,
-    E: ObjectTaskExecutor<T, R>,
+    E: ObjectTaskExecutor<T, R> + 'static,
 {
     task_txs: Vec<SyncSender<ObjectWorkerMessage<T>>>,
     result_rx: Receiver<R>,
@@ -35,7 +39,7 @@ impl<T, R, E> ObjectWorkerPool<T, R, E>
 where
     T: Send + 'static,
     R: Send + 'static,
-    E: ObjectTaskExecutor<T, R>,
+    E: ObjectTaskExecutor<T, R> + 'static,
 {
     pub fn new(worker_count: usize, queue_capacity: usize, executor: E) -> Result<Self, String> {
         if worker_count == 0 {
@@ -140,7 +144,7 @@ impl<T, R, E> Drop for ObjectWorkerPool<T, R, E>
 where
     T: Send + 'static,
     R: Send + 'static,
-    E: ObjectTaskExecutor<T, R>,
+    E: ObjectTaskExecutor<T, R> + 'static,
 {
     fn drop(&mut self) {
         let _ = self.shutdown_inner();
@@ -153,8 +157,8 @@ fn object_worker_loop<T, R, E>(
     result_tx: mpsc::Sender<R>,
     executor: Arc<E>,
 ) where
-    T: Send + 'static,
+    T: Send,
-    R: Send + 'static,
+    R: Send,
     E: ObjectTaskExecutor<T, R>,
 {
     loop {
@@ -170,6 +174,120 @@ fn object_worker_loop<T, R, E>(
     }
 }
 
+/// Scoped variant of `ObjectWorkerPool`: workers are spawned on a
+/// `std::thread::Scope`, so tasks, results and the executor may borrow their
+/// environment (`'env`) instead of being `'static`. The pool never sends
+/// `Shutdown`; workers exit when every task sender is dropped, which happens
+/// when the pool itself is dropped ahead of the scope join.
+pub struct ScopedObjectWorkerPool<'scope, 'env, T, R, E>
+where
+    T: Send + 'env,
+    R: Send + 'env,
+    E: ObjectTaskExecutor<T, R> + 'env,
+{
+    task_txs: Vec<SyncSender<ObjectWorkerMessage<T>>>,
+    result_rx: Receiver<R>,
+    next_worker_idx: usize,
+    _executor: Arc<E>,
+    // Join handles are intentionally not stored: dropping a `ScopedJoinHandle`
+    // detaches the worker and the enclosing scope joins it on exit, after the
+    // dropped task senders have made every worker return.
+    _marker: PhantomData<(&'scope (), &'env ())>,
+}
+
+impl<'scope, 'env, T, R, E> ScopedObjectWorkerPool<'scope, 'env, T, R, E>
+where
+    T: Send + 'env,
+    R: Send + 'env,
+    E: ObjectTaskExecutor<T, R> + 'env,
+{
+    pub fn new(
+        scope: &'scope Scope<'scope, 'env>,
+        worker_count: usize,
+        queue_capacity: usize,
+        executor: E,
+    ) -> Result<Self, String> {
+        if worker_count == 0 {
+            return Err("ScopedObjectWorkerPool requires at least one worker".to_string());
+        }
+        if queue_capacity == 0 {
+            return Err("ScopedObjectWorkerPool requires queue_capacity > 0".to_string());
+        }
+
+        let executor = Arc::new(executor);
+        let (result_tx, result_rx) = mpsc::channel::<R>();
+        let mut task_txs = Vec::with_capacity(worker_count);
+
+        for worker_index in 0..worker_count {
+            let (task_tx, task_rx) = mpsc::sync_channel::<ObjectWorkerMessage<T>>(queue_capacity);
+            let result_tx = result_tx.clone();
+            let executor = Arc::clone(&executor);
+            thread::Builder::new()
+                .name(format!("object-validation-worker-{worker_index}"))
+                .spawn_scoped(scope, move || {
+                    object_worker_loop(worker_index, task_rx, result_tx, executor)
+                })
+                .map_err(|e| format!("spawn scoped object worker failed: {e}"))?;
+            task_txs.push(task_tx);
+        }
+
+        Ok(Self {
+            task_txs,
+            result_rx,
+            next_worker_idx: 0,
+            _executor: executor,
+            _marker: PhantomData,
+        })
+    }
+
+    pub fn worker_count(&self) -> usize {
+        self.task_txs.len()
+    }
+
+    pub fn try_submit_round_robin(&mut self, task: T) -> Result<usize, ObjectWorkerSubmitError<T>> {
+        let worker_index = self.next_worker_idx % self.task_txs.len();
+        match self.task_txs[worker_index].try_send(ObjectWorkerMessage::Task(task)) {
+            Ok(()) => {
+                self.next_worker_idx = (worker_index + 1) % self.task_txs.len();
+                Ok(worker_index)
+            }
+            Err(TrySendError::Full(ObjectWorkerMessage::Task(task))) => {
+                Err(ObjectWorkerSubmitError::QueueFull { worker_index, task })
+            }
+            Err(TrySendError::Disconnected(ObjectWorkerMessage::Task(task))) => {
+                Err(ObjectWorkerSubmitError::Disconnected { worker_index, task })
+            }
+            Err(TrySendError::Full(ObjectWorkerMessage::Shutdown))
+            | Err(TrySendError::Disconnected(ObjectWorkerMessage::Shutdown)) => {
+                unreachable!("shutdown is never submitted via try_submit_round_robin")
+            }
+        }
+    }
+
+    pub fn recv_result_timeout(&self, timeout: Duration) -> Result<Option<R>, String> {
+        match self.result_rx.recv_timeout(timeout) {
+            Ok(result) => Ok(Some(result)),
+            Err(RecvTimeoutError::Timeout) => Ok(None),
+            Err(RecvTimeoutError::Disconnected) => {
+                Err("scoped object worker result channel disconnected".to_string())
+            }
+        }
+    }
+}
+
+impl<'scope, 'env, T, R, E> Drop for ScopedObjectWorkerPool<'scope, 'env, T, R, E>
+where
+    T: Send + 'env,
+    R: Send + 'env,
+    E: ObjectTaskExecutor<T, R> + 'env,
+{
+    fn drop(&mut self) {
+        // Close every worker input queue so blocked `recv` calls return and
+        // the scoped workers exit before the enclosing scope joins them.
+        self.task_txs.clear();
+    }
+}
+
 #[cfg(test)]
 mod tests {
     use super::{ObjectTaskExecutor, ObjectWorkerPool, ObjectWorkerSubmitError};
@@ -284,4 +402,84 @@ mod tests {
             Some(2)
         );
     }
+
+    struct BorrowingEchoExecutor<'a> {
+        base: &'a u32,
+    }
+
+    impl<'a> ObjectTaskExecutor<u32, u32> for BorrowingEchoExecutor<'a> {
+        fn execute(&self, _worker_index: usize, task: u32) -> u32 {
+            task + *self.base
+        }
+    }
+
+    #[test]
+    fn scoped_object_worker_pool_borrows_environment_and_processes_tasks() {
+        let base = 100u32;
+        std::thread::scope(|scope| {
+            let mut pool = super::ScopedObjectWorkerPool::new(
+                scope,
+                2,
+                2,
+                BorrowingEchoExecutor { base: &base },
+            )
+            .expect("scoped pool");
+            assert_eq!(pool.worker_count(), 2);
+            pool.try_submit_round_robin(1).expect("submit 1");
+            pool.try_submit_round_robin(2).expect("submit 2");
+            let mut results = Vec::new();
+            for _ in 0..2 {
+                results.push(
+                    pool.recv_result_timeout(Duration::from_secs(1))
+                        .expect("result channel")
+                        .expect("result"),
+                );
+            }
+            results.sort();
+            assert_eq!(results, vec![101, 102]);
+            // Dropping the pool inside the scope closes the task queues; the
+            // workers exit on their own and the scope join below must not hang.
+            drop(pool);
+        });
+    }
+
+    #[test]
+    fn scoped_object_worker_pool_reports_full_queue() {
+        std::thread::scope(|scope| {
+            let barrier = Arc::new(Barrier::new(2));
+            let started = Arc::new(AtomicBool::new(false));
+            let mut pool = super::ScopedObjectWorkerPool::new(
+                scope,
+                1,
+                1,
+                BlockingExecutor {
+                    barrier: Arc::clone(&barrier),
+                    started: Arc::clone(&started),
+                },
+            )
+            .expect("scoped pool");
+            pool.try_submit_round_robin(1).expect("first task");
+            let deadline = std::time::Instant::now() + Duration::from_secs(1);
+            while !started.load(Ordering::SeqCst) {
+                assert!(
+                    std::time::Instant::now() < deadline,
+                    "scoped worker did not start first task"
+                );
+                std::thread::sleep(Duration::from_millis(1));
+            }
+            pool.try_submit_round_robin(2).expect("queued task");
+            match pool.try_submit_round_robin(3) {
+                Err(ObjectWorkerSubmitError::QueueFull { worker_index, task }) => {
+                    assert_eq!(worker_index, 0);
+                    assert_eq!(task, 3);
+                }
+                other => panic!("expected queue full, got {other:?}"),
+            }
+            // Dropping the pool closes the task queues; releasing the barrier
+            // afterwards lets the blocked worker finish task 1, observe the
+            // closed result channel, and exit before the scope join.
+            drop(pool);
+            barrier.wait();
+        });
+    }
 }

src/parallel/repo_runtime.rs

@@ -103,6 +103,17 @@ pub trait RepoSyncRuntime: Send + Sync {
     ) -> Result<TransportPrefetchDispatchStats, String>;
 
     fn transport_prefetch_snapshot(&self) -> TransportPrefetchSnapshot;
+
+    /// Dead-repo blacklist working copy (#141) for run-end persistence.
+    /// `None` when the feature is disabled.
+    fn dead_repo_blacklist_state(
+        &self,
+    ) -> Option<(
+        crate::parallel::dead_repo_blacklist::DeadRepoBlacklistConfig,
+        crate::parallel::dead_repo_blacklist::DeadRepoBlacklist,
+    )> {
+        None
+    }
 }
 
 pub struct Phase1RepoSyncRuntime<E: RepoTransportExecutor> {
@@ -634,6 +645,18 @@ impl<E: RepoTransportExecutor> RepoSyncRuntime for Phase1RepoSyncRuntime<E> {
             })
             .unwrap_or_else(|| TransportPrefetchSnapshot::new(self.sync_preference, Vec::new()))
     }
+
+    fn dead_repo_blacklist_state(
+        &self,
+    ) -> Option<(
+        crate::parallel::dead_repo_blacklist::DeadRepoBlacklistConfig,
+        crate::parallel::dead_repo_blacklist::DeadRepoBlacklist,
+    )> {
+        let coordinator = self.coordinator.lock().expect("coordinator lock poisoned");
+        let config = coordinator.config.dead_repo_blacklist.clone()?;
+        let blacklist = coordinator.dead_repo_blacklist().cloned()?;
+        Some((config, blacklist))
+    }
 }
 
 fn outcome_from_transport_result(
@@ -666,7 +689,9 @@ fn outcome_from_transport_result(
             }
         }
         (
-            RepoTransportResultKind::Failed { detail, warnings },
+            RepoTransportResultKind::Failed {
+                detail, warnings, ..
+            },
             RepoRuntimeState::FailedTerminal,
         ) => RepoSyncRuntimeOutcome {
             repo_sync_ok: false,
@@ -798,6 +823,7 @@ mod tests {
                         result: RepoTransportResultKind::Failed {
                             detail: "rrdp failed".to_string(),
                             warnings: vec![Warning::new("rrdp failed")],
+                            error_class: crate::parallel::types::RepoTransportErrorClass::Unknown,
                         },
                     }
                 }
@@ -842,6 +868,7 @@ mod tests {
                         result: RepoTransportResultKind::Failed {
                             detail: "rrdp failed".to_string(),
                             warnings: vec![Warning::new("rrdp failed")],
+                            error_class: crate::parallel::types::RepoTransportErrorClass::Unknown,
                         },
                     }
                 }
@@ -858,6 +885,7 @@ mod tests {
                         result: RepoTransportResultKind::Failed {
                             detail: "rsync failed".to_string(),
                             warnings: vec![Warning::new("rsync failed")],
+                            error_class: crate::parallel::types::RepoTransportErrorClass::Unknown,
                         },
                     }
                 }

src/parallel/repo_scheduler.rs

@@ -1,11 +1,14 @@
 use std::collections::{HashMap, HashSet};
 
+use crate::parallel::dead_repo_blacklist::DeadRepoBlacklist;
 use crate::parallel::types::{
     InFlightRepoEntry, RepoDedupKey, RepoIdentity, RepoKey, RepoRequester, RepoRuntimeState,
     RepoSyncResultEnvelope, RepoSyncResultKind, RepoSyncResultRef, RepoSyncTask, RepoTaskState,
-    RepoTransportMode, RepoTransportResultEnvelope, RepoTransportResultKind, RepoTransportTask,
+    RepoTransportErrorClass, RepoTransportMode, RepoTransportResultEnvelope,
+    RepoTransportResultKind, RepoTransportTask,
 };
 use crate::policy::SyncPreference;
+use crate::report::Warning;
 
 #[derive(Clone, Debug, PartialEq, Eq)]
 pub enum RepoRequestAction {
@@ -74,6 +77,9 @@ pub struct TransportStateTables {
     rsync_failure_probe_inflight: HashMap<String, String>,
     rsync_failure_scope_reachable: HashSet<String>,
     runtime_records: HashMap<RepoIdentity, RepoRuntimeRecord>,
+    /// Frozen run-start snapshot of the dead-repo blacklist (#141). Entries
+    /// admitted during this run only take effect from the next run.
+    dead_repo_blacklist: Option<DeadRepoBlacklist>,
 }
 
 impl TransportStateTables {
@@ -81,6 +87,44 @@ impl TransportStateTables {
         Self::default()
     }
 
+    pub fn set_dead_repo_blacklist(&mut self, blacklist: DeadRepoBlacklist) {
+        self.dead_repo_blacklist = Some(blacklist);
+    }
+
+    fn dead_repo_terminal_envelope(
+        identity: &RepoIdentity,
+        requesters: &[RepoRequester],
+        rsync_scope_uri: &str,
+        rsync_failure_scope_uri: Option<String>,
+    ) -> RepoTransportResultEnvelope {
+        let first_requester = requesters
+            .first()
+            .expect("blacklist terminal record must keep at least one requester");
+        RepoTransportResultEnvelope {
+            dedup_key: RepoDedupKey::RsyncScope {
+                rsync_scope_uri: rsync_scope_uri.to_string(),
+            },
+            rsync_failure_scope_uri,
+            repo_identity: identity.clone(),
+            mode: RepoTransportMode::Rsync,
+            tal_id: first_requester.tal_id.clone(),
+            rir_id: first_requester.rir_id.clone(),
+            timing_ms: 0,
+            result: RepoTransportResultKind::Failed {
+                detail: format!(
+                    "dead repo blacklist: rsync transport persistently unreachable: {}",
+                    identity.rsync_base_uri
+                ),
+                warnings: vec![Warning::new(format!(
+                    "dead_repo_blacklist_skip_all: repository {} skipped after repeated transport failures",
+                    identity.rsync_base_uri
+                ))
+                .with_context(identity.rsync_base_uri.clone())],
+                error_class: RepoTransportErrorClass::Unknown,
+            },
+        }
+    }
+
     pub fn runtime_record(&self, identity: &RepoIdentity) -> Option<&RepoRuntimeRecord> {
         self.runtime_records.get(identity)
     }
@@ -205,6 +249,82 @@ impl TransportStateTables {
             };
         }
 
+        // Dead-repo blacklist fast paths (#141), consulted only for fresh
+        // identities; existing records already encode their terminal state.
+        if let Some(blacklist) = self.dead_repo_blacklist.as_ref() {
+            let rrdp_wanted = sync_preference == SyncPreference::RrdpThenRsync
+                && identity.notification_uri.is_some();
+            let rrdp_dead = rrdp_wanted
+                && identity
+                    .notification_uri
+                    .as_deref()
+                    .map(|uri| blacklist.is_blacklisted(RepoTransportMode::Rrdp, uri))
+                    .unwrap_or(false);
+            let rsync_dead = blacklist.is_blacklisted(
+                RepoTransportMode::Rsync,
+                identity.rsync_base_uri.as_str(),
+            );
+            if rsync_dead && (!rrdp_wanted || rrdp_dead) {
+                crate::progress_log::emit(
+                    "dead_repo_blacklist_skip_all",
+                    serde_json::json!({
+                        "repo_key_notification_uri": identity.notification_uri,
+                        "repo_key_rsync_base_uri": identity.rsync_base_uri,
+                    }),
+                );
+                let envelope = Self::dead_repo_terminal_envelope(
+                    &identity,
+                    std::slice::from_ref(&requester),
+                    &rsync_scope_uri,
+                    rsync_failure_scope_uri.clone(),
+                );
+                self.runtime_records.insert(
+                    identity.clone(),
+                    RepoRuntimeRecord {
+                        identity,
+                        state: RepoRuntimeState::FailedTerminal,
+                        rrdp_notification_key: None,
+                        rsync_scope_key: rsync_scope_uri,
+                        rsync_failure_scope_key: rsync_failure_scope_uri,
+                        requesters: vec![requester],
+                        validation_time,
+                        priority,
+                        last_success: None,
+                        terminal_failure: Some(envelope.clone()),
+                    },
+                );
+                return TransportRequestAction::ReusedTerminalFailure(envelope);
+            }
+            if rrdp_dead {
+                crate::progress_log::emit(
+                    "dead_repo_blacklist_skip_rrdp",
+                    serde_json::json!({
+                        "repo_key_notification_uri": identity.notification_uri,
+                        "repo_key_rsync_base_uri": identity.rsync_base_uri,
+                    }),
+                );
+                let mut action = self.register_rsync_request(
+                    identity,
+                    requester,
+                    validation_time,
+                    priority,
+                    rsync_scope_uri,
+                    rsync_failure_scope_uri,
+                );
+                // A blacklisted rrdp transport is a known-persistent failure,
+                // so the rsync fallback must keep the short retry profile
+                // (#141): the prefetch layer records no real rrdp failure in
+                // skipped runs, which would otherwise silently drop the task
+                // back to the default 15s window from the second skipped run
+                // on. The coordinator wrapper only ever sets the flag to true,
+                // so setting it here survives registration.
+                if let TransportRequestAction::Enqueue(task) = &mut action {
+                    task.retry_short_timeout = true;
+                }
+                return action;
+            }
+        }
+
         if sync_preference == SyncPreference::RrdpThenRsync {
             if let Some(notification_uri) = identity.notification_uri.clone() {
                 if let Some(entry) = self.rrdp_inflight.get_mut(&notification_uri) {
@@ -512,6 +632,7 @@ impl TransportStateTables {
         rsync_failure_by_scope: &HashMap<String, RepoTransportResultEnvelope>,
         rsync_failure_probe_inflight: &mut HashMap<String, String>,
         rsync_failure_scope_reachable: &HashSet<String>,
+        dead_repo_blacklist: Option<&DeadRepoBlacklist>,
         follow_up_tasks: &mut Vec<RepoTransportTask>,
     ) {
         let rsync_scope_uri = record.rsync_scope_key.clone();
@@ -548,6 +669,34 @@ impl TransportStateTables {
             }
         }
 
+        // Dead-repo blacklist (#141): the rsync transport of this repo is
+        // persistently unreachable; terminate instantly instead of enqueueing
+        // another doomed fetch (rrdp was already attempted or skipped).
+        if let Some(blacklist) = dead_repo_blacklist {
+            if blacklist.is_blacklisted(
+                RepoTransportMode::Rsync,
+                record.identity.rsync_base_uri.as_str(),
+            ) {
+                crate::progress_log::emit(
+                    "dead_repo_blacklist_skip_all",
+                    serde_json::json!({
+                        "repo_key_notification_uri": record.identity.notification_uri,
+                        "repo_key_rsync_base_uri": record.identity.rsync_base_uri,
+                        "after_rrdp_failure": true,
+                    }),
+                );
+                let envelope = Self::dead_repo_terminal_envelope(
+                    &record.identity,
+                    &record.requesters,
+                    &rsync_scope_uri,
+                    record.rsync_failure_scope_key.clone(),
+                );
+                record.state = RepoRuntimeState::FailedTerminal;
+                record.terminal_failure = Some(envelope);
+                return;
+            }
+        }
+
         let first_requester = record
             .requesters
             .first()
@@ -666,6 +815,7 @@ impl TransportStateTables {
                             &self.rsync_failure_by_scope,
                             &mut self.rsync_failure_probe_inflight,
                             &self.rsync_failure_scope_reachable,
+                            self.dead_repo_blacklist.as_ref(),
                             &mut follow_up_tasks,
                         );
                     }
@@ -718,6 +868,7 @@ impl TransportStateTables {
                                 &self.rsync_failure_by_scope,
                                 &mut self.rsync_failure_probe_inflight,
                                 &self.rsync_failure_scope_reachable,
+                                self.dead_repo_blacklist.as_ref(),
                                 &mut follow_up_tasks,
                             );
                         }
@@ -783,6 +934,7 @@ impl TransportStateTables {
                                     &self.rsync_failure_by_scope,
                                     &mut self.rsync_failure_probe_inflight,
                                     &self.rsync_failure_scope_reachable,
+                                    self.dead_repo_blacklist.as_ref(),
                                     &mut follow_up_tasks,
                                 );
                             }
@@ -814,7 +966,7 @@ fn reusable_rsync_failure_scope(result: &RepoTransportResultEnvelope) -> Option<
     }
 }
 
-fn is_host_level_rsync_failure(detail: &str) -> bool {
+pub(crate) fn is_host_level_rsync_failure(detail: &str) -> bool {
     let lower = detail.to_ascii_lowercase();
     lower.contains("timeout waiting for daemon connection")
         || lower.contains("failed to connect")
@@ -1462,6 +1614,7 @@ mod transport_tests {
                     result: RepoTransportResultKind::Failed {
                         detail: "rrdp timeout".to_string(),
                         warnings: Vec::new(),
+                        error_class: crate::parallel::types::RepoTransportErrorClass::Unknown,
                     },
                 },
                 time::OffsetDateTime::UNIX_EPOCH,
@@ -1567,6 +1720,7 @@ mod transport_tests {
                     result: RepoTransportResultKind::Failed {
                         detail: "rrdp timeout".to_string(),
                         warnings: Vec::new(),
+                        error_class: crate::parallel::types::RepoTransportErrorClass::Unknown,
                     },
                 },
                 time::OffsetDateTime::UNIX_EPOCH,
@@ -1597,6 +1751,7 @@ mod transport_tests {
                     result: RepoTransportResultKind::Failed {
                         detail: "rsync error: timeout waiting for daemon connection".to_string(),
                         warnings: Vec::new(),
+                        error_class: crate::parallel::types::RepoTransportErrorClass::Unknown,
                     },
                 },
                 time::OffsetDateTime::UNIX_EPOCH,
@@ -1659,6 +1814,7 @@ mod transport_tests {
                     result: RepoTransportResultKind::Failed {
                         detail: "rsync timeout".to_string(),
                         warnings: Vec::new(),
+                        error_class: crate::parallel::types::RepoTransportErrorClass::Unknown,
                     },
                 },
                 time::OffsetDateTime::UNIX_EPOCH,
@@ -1920,6 +2076,7 @@ mod transport_tests {
                     result: RepoTransportResultKind::Failed {
                         detail: "rsync file digest mismatch after download".to_string(),
                         warnings: Vec::new(),
+                        error_class: crate::parallel::types::RepoTransportErrorClass::Unknown,
                     },
                 },
                 time::OffsetDateTime::UNIX_EPOCH,
@@ -1972,6 +2129,7 @@ mod transport_tests {
                     result: RepoTransportResultKind::Failed {
                         detail: "rsync error: failed to connect to daemon".to_string(),
                         warnings: Vec::new(),
+                        error_class: crate::parallel::types::RepoTransportErrorClass::Unknown,
                     },
                 },
                 time::OffsetDateTime::UNIX_EPOCH,
@@ -2031,6 +2189,7 @@ mod transport_tests {
                     result: RepoTransportResultKind::Failed {
                         detail: "temporary failure in name resolution".to_string(),
                         warnings: Vec::new(),
+                        error_class: crate::parallel::types::RepoTransportErrorClass::Unknown,
                     },
                 },
                 time::OffsetDateTime::UNIX_EPOCH,

src/parallel/repo_worker.rs

@@ -115,7 +115,10 @@ impl<H: Fetcher + 'static> RepoTransportExecutor for LiveRrdpTransportExecutor<H
                     warnings: Vec::new(),
                 },
             },
-            Err(err) => RepoTransportResultEnvelope {
+            Err(err) => {
+                let error_class =
+                    crate::parallel::dead_repo_blacklist::classify_rrdp_sync_error(&err);
+                RepoTransportResultEnvelope {
                     dedup_key: task.dedup_key,
                     rsync_failure_scope_uri: task.rsync_failure_scope_uri,
                     repo_identity: task.repo_identity,
@@ -126,8 +129,10 @@ impl<H: Fetcher + 'static> RepoTransportExecutor for LiveRrdpTransportExecutor<H
                     result: RepoTransportResultKind::Failed {
                         detail: err.to_string(),
                         warnings: Vec::new(),
+                        error_class,
                     },
-            },
+                }
+            }
         }
     }
 }
@@ -206,7 +211,10 @@ impl<R: RsyncFetcher + 'static> RepoTransportExecutor for LiveRsyncTransportExec
                     warnings: Vec::new(),
                 },
             },
-            Err(err) => RepoTransportResultEnvelope {
+            Err(err) => {
+                let error_class =
+                    crate::parallel::dead_repo_blacklist::classify_rsync_repo_sync_error(&err);
+                RepoTransportResultEnvelope {
                     dedup_key: task.dedup_key,
                     rsync_failure_scope_uri: task.rsync_failure_scope_uri,
                     repo_identity: task.repo_identity,
@@ -217,8 +225,10 @@ impl<R: RsyncFetcher + 'static> RepoTransportExecutor for LiveRsyncTransportExec
                     result: RepoTransportResultKind::Failed {
                         detail: err.to_string(),
                         warnings: Vec::new(),
+                        error_class,
                     },
-            },
+                }
+            }
         }
     }
 }
@@ -1160,7 +1170,7 @@ mod tests {
             matches!(result.result, RepoTransportResultKind::Success { .. }),
             "{result:?}"
         );
-        let index = current_repo_index.lock().expect("index lock");
+        let index = current_repo_index.read().expect("index read lock");
         assert!(
             index
                 .get_by_uri("rsync://example.test/repo/a.roa")
@@ -1217,7 +1227,7 @@ mod tests {
             result.result,
             RepoTransportResultKind::Success { .. }
         ));
-        let index = current_repo_index.lock().expect("index lock");
+        let index = current_repo_index.read().expect("index read lock");
         assert!(
             index
                 .get_by_uri("rsync://example.test/repo/a.roa")

src/parallel/run_coordinator.rs

@@ -2,6 +2,9 @@ use std::collections::VecDeque;
 
 use crate::current_repo_index::{CurrentRepoIndex, CurrentRepoIndexHandle};
 use crate::parallel::config::ParallelPhase1Config;
+use crate::parallel::dead_repo_blacklist::{
+    DeadRepoBlacklist, DeadRepoFailureOutcome, DeadRepoSuccessOutcome,
+};
 use crate::parallel::repo_scheduler::{
     InFlightRepoTable, RepoCompletion, RepoRequestAction, TransportCompletion,
     TransportRequestAction, TransportStateTables,
@@ -9,7 +12,8 @@ use crate::parallel::repo_scheduler::{
 use crate::parallel::stats::ParallelRunStats;
 use crate::parallel::types::{
     RepoDedupKey, RepoIdentity, RepoKey, RepoRequester, RepoSyncResultEnvelope, RepoSyncTask,
-    RepoTransportResultEnvelope, RepoTransportTask, TalInputSpec,
+    RepoTransportErrorClass, RepoTransportMode, RepoTransportResultEnvelope,
+    RepoTransportResultKind, RepoTransportTask, TalInputSpec,
 };
 use crate::policy::SyncPreference;
 
@@ -22,19 +26,42 @@ pub struct GlobalRunCoordinator {
     pub pending_repo_tasks: VecDeque<RepoSyncTask>,
     pub pending_transport_tasks: VecDeque<RepoTransportTask>,
     pub stats: ParallelRunStats,
+    /// Mutable working copy of the dead-repo blacklist (#141): failure/success
+    /// counters update during the run and persist at run end. Skip decisions
+    /// inside `transport_tables` use the frozen run-start snapshot.
+    pub dead_repo_blacklist: Option<DeadRepoBlacklist>,
 }
 
 impl GlobalRunCoordinator {
     pub fn new(config: ParallelPhase1Config, tal_inputs: Vec<TalInputSpec>) -> Self {
+        let dead_repo_blacklist = config
+            .dead_repo_blacklist
+            .as_ref()
+            .map(|blacklist_config| {
+                let (blacklist, warning) = DeadRepoBlacklist::load(&blacklist_config.path);
+                if let Some(warning) = warning {
+                    crate::progress_log::emit(
+                        "dead_repo_blacklist_load_warning",
+                        serde_json::json!({ "warning": warning }),
+                    );
+                }
+                blacklist
+            });
+        let mut transport_tables = TransportStateTables::new();
+        if let Some(blacklist) = dead_repo_blacklist.as_ref() {
+            // Frozen run-start snapshot for skip decisions.
+            transport_tables.set_dead_repo_blacklist(blacklist.clone());
+        }
         Self {
             config,
             tal_inputs,
             current_repo_index: CurrentRepoIndex::shared(),
             in_flight_repos: InFlightRepoTable::new(),
-            transport_tables: TransportStateTables::new(),
+            transport_tables,
             pending_repo_tasks: VecDeque::new(),
             pending_transport_tasks: VecDeque::new(),
             stats: ParallelRunStats::default(),
+            dead_repo_blacklist,
         }
     }
 
@@ -193,9 +220,92 @@ impl GlobalRunCoordinator {
         {
             self.stats.repo_tasks_failed += 1;
         }
+        self.update_dead_repo_blacklist(&result, finished_at);
         Ok(completion)
     }
 
+    /// Update dead-repo blacklist counters (#141) from a real worker result.
+    /// Runs on the single pump thread; synthesized terminal envelopes never
+    /// pass through here, so no double counting is possible.
+    fn update_dead_repo_blacklist(
+        &mut self,
+        result: &RepoTransportResultEnvelope,
+        finished_at: time::OffsetDateTime,
+    ) {
+        let Some(blacklist_config) = self.config.dead_repo_blacklist.clone() else {
+            return;
+        };
+        let Some(blacklist) = self.dead_repo_blacklist.as_mut() else {
+            return;
+        };
+        let now_unix = finished_at.unix_timestamp().max(0) as u64;
+        let key: Option<(RepoTransportMode, String)> = match (&result.dedup_key, &result.result) {
+            (RepoDedupKey::RrdpNotify { notification_uri }, RepoTransportResultKind::Success { .. }) => {
+                Some((RepoTransportMode::Rrdp, notification_uri.clone()))
+            }
+            (RepoDedupKey::RsyncScope { .. }, RepoTransportResultKind::Success { .. }) => {
+                Some((RepoTransportMode::Rsync, result.repo_identity.rsync_base_uri.clone()))
+            }
+            (
+                RepoDedupKey::RrdpNotify { notification_uri },
+                RepoTransportResultKind::Failed {
+                    error_class: RepoTransportErrorClass::TransportFetch,
+                    ..
+                },
+            ) => Some((RepoTransportMode::Rrdp, notification_uri.clone())),
+            (
+                RepoDedupKey::RsyncScope { .. },
+                RepoTransportResultKind::Failed {
+                    error_class: RepoTransportErrorClass::TransportFetch,
+                    ..
+                },
+            ) => Some((RepoTransportMode::Rsync, result.repo_identity.rsync_base_uri.clone())),
+            _ => None,
+        };
+        let Some((transport, uri)) = key else {
+            return;
+        };
+        match &result.result {
+            RepoTransportResultKind::Success { .. } => {
+                if let Some(outcome) = blacklist.record_transport_success(transport, &uri) {
+                    crate::progress_log::emit(
+                        "dead_repo_blacklist_remove",
+                        serde_json::json!({
+                            "transport": transport.as_str(),
+                            "uri": uri,
+                            "reason": match outcome {
+                                DeadRepoSuccessOutcome::CounterReset => "counter_reset",
+                                DeadRepoSuccessOutcome::BlacklistRemoved => "fetch_success",
+                            },
+                        }),
+                    );
+                }
+            }
+            RepoTransportResultKind::Failed { .. } => {
+                if let DeadRepoFailureOutcome::Admitted = blacklist.record_transport_failure(
+                    transport,
+                    &uri,
+                    now_unix,
+                    blacklist_config.fail_threshold,
+                    blacklist_config.capacity,
+                ) {
+                    crate::progress_log::emit(
+                        "dead_repo_blacklist_admit",
+                        serde_json::json!({
+                            "transport": transport.as_str(),
+                            "uri": uri,
+                            "fail_threshold": blacklist_config.fail_threshold,
+                        }),
+                    );
+                }
+            }
+        }
+    }
+
+    pub fn dead_repo_blacklist(&self) -> Option<&DeadRepoBlacklist> {
+        self.dead_repo_blacklist.as_ref()
+    }
+
     pub fn runtime_record(
         &self,
         identity: &RepoIdentity,
@@ -225,7 +335,7 @@ impl GlobalRunCoordinator {
         self.pending_repo_tasks.clear();
         self.pending_transport_tasks.clear();
         self.stats = ParallelRunStats::default();
-        if let Ok(mut index) = self.current_repo_index.lock() {
+        if let Ok(mut index) = self.current_repo_index.write() {
             index.clear();
         }
     }
@@ -322,7 +432,7 @@ mod tests {
         assert_eq!(coordinator.pending_transport_tasks.len(), 1);
 
         {
-            let mut index = coordinator.current_repo_index.lock().expect("index lock");
+            let mut index = coordinator.current_repo_index.write().expect("index write lock");
             index
                 .apply_repository_view_entries(&[RepositoryViewEntry {
                     rsync_uri: "rsync://example.test/repo/a.roa".to_string(),
@@ -343,8 +453,8 @@ mod tests {
         assert_eq!(
             coordinator
                 .current_repo_index
-                .lock()
+                .read()
-                .expect("index lock")
+                .expect("index read lock")
                 .active_uri_count(),
             0
         );
@@ -465,4 +575,288 @@ mod tests {
         assert_eq!(coordinator.stats.repo_tasks_total, 1);
         assert_eq!(coordinator.stats.repo_tasks_reused, 1);
     }
+
+    // ---- Dead-repo blacklist (#141) integration tests ----
+
+    use crate::parallel::dead_repo_blacklist::{DeadRepoBlacklist, DeadRepoBlacklistConfig};
+    use crate::parallel::types::{
+        RepoDedupKey, RepoTransportErrorClass, RepoTransportMode, RepoTransportResultEnvelope,
+        RepoTransportResultKind,
+    };
+
+    const DEAD_NOTIFY: &str = "https://dead.example/notification.xml";
+    const DEAD_BASE: &str = "rsync://dead.example/repo/";
+
+    fn blacklist_config(fail_threshold: u32) -> DeadRepoBlacklistConfig {
+        DeadRepoBlacklistConfig {
+            path: std::env::temp_dir().join(format!(
+                "dead_repo_blacklist_coordinator_test_{}.json",
+                std::process::id()
+            )),
+            fail_threshold,
+            capacity: 256,
+        }
+    }
+
+    fn coordinator_with_blacklist(
+        fail_threshold: u32,
+        seed: DeadRepoBlacklist,
+    ) -> GlobalRunCoordinator {
+        let mut config = ParallelPhase1Config::default();
+        config.dead_repo_blacklist = Some(blacklist_config(fail_threshold));
+        let mut coordinator = GlobalRunCoordinator::new(
+            config,
+            vec![TalInputSpec::from_url("https://example.test/arin.tal")],
+        );
+        coordinator.dead_repo_blacklist = Some(seed.clone());
+        coordinator.transport_tables.set_dead_repo_blacklist(seed);
+        coordinator
+    }
+
+    fn dead_identity() -> RepoIdentity {
+        RepoIdentity::new(Some(DEAD_NOTIFY.to_string()), DEAD_BASE)
+    }
+
+    fn dead_requester() -> RepoRequester {
+        requester("arin", "arin", "rsync://dead.example/repo/root.mft")
+    }
+
+    fn transport_envelope(
+        mode: RepoTransportMode,
+        error_class: RepoTransportErrorClass,
+    ) -> RepoTransportResultEnvelope {
+        let (dedup_key, detail) = match mode {
+            RepoTransportMode::Rrdp => (
+                RepoDedupKey::RrdpNotify {
+                    notification_uri: DEAD_NOTIFY.to_string(),
+                },
+                "http request failed: connect timeout".to_string(),
+            ),
+            RepoTransportMode::Rsync => (
+                RepoDedupKey::RsyncScope {
+                    rsync_scope_uri: DEAD_BASE.to_string(),
+                },
+                "rsync error: timeout waiting for daemon connection".to_string(),
+            ),
+        };
+        RepoTransportResultEnvelope {
+            dedup_key,
+            rsync_failure_scope_uri: None,
+            repo_identity: dead_identity(),
+            mode,
+            tal_id: "arin".to_string(),
+            rir_id: "arin".to_string(),
+            timing_ms: 1,
+            result: RepoTransportResultKind::Failed {
+                detail,
+                warnings: Vec::new(),
+                error_class,
+            },
+        }
+    }
+
+    fn register_dead(
+        coordinator: &mut GlobalRunCoordinator,
+    ) -> crate::parallel::repo_scheduler::TransportRequestAction {
+        coordinator.register_transport_request(
+            dead_identity(),
+            dead_requester(),
+            time::OffsetDateTime::UNIX_EPOCH,
+            0,
+            DEAD_BASE.to_string(),
+            None,
+            SyncPreference::RrdpThenRsync,
+            false,
+        )
+    }
+
+    #[test]
+    fn blacklist_admission_after_threshold_transport_failures() {
+        let mut coordinator = coordinator_with_blacklist(2, DeadRepoBlacklist::new());
+        for _ in 0..2 {
+            let action = register_dead(&mut coordinator);
+            assert!(matches!(
+                action,
+                crate::parallel::repo_scheduler::TransportRequestAction::Enqueue(_)
+            ));
+            coordinator
+                .complete_transport_result(
+                    transport_envelope(
+                        RepoTransportMode::Rrdp,
+                        RepoTransportErrorClass::TransportFetch,
+                    ),
+                    time::OffsetDateTime::UNIX_EPOCH,
+                )
+                .expect("complete");
+            // Simulate the next run: run-local scheduler state resets, the
+            // persistent blacklist working copy carries over.
+            coordinator.reset_run_state();
+        }
+        let blacklist = coordinator.dead_repo_blacklist().expect("blacklist");
+        assert!(blacklist.is_blacklisted(RepoTransportMode::Rrdp, DEAD_NOTIFY));
+        // Admission mid-run does not affect this run's frozen snapshot: the
+        // scheduler still enqueues rrdp for a fresh identity registration.
+        coordinator.reset_run_state();
+        let action = register_dead(&mut coordinator);
+        assert!(matches!(
+            action,
+            crate::parallel::repo_scheduler::TransportRequestAction::Enqueue(_)
+        ));
+    }
+
+    #[test]
+    fn blacklist_ignores_protocol_and_unknown_failures() {
+        let mut coordinator = coordinator_with_blacklist(1, DeadRepoBlacklist::new());
+        for error_class in [
+            RepoTransportErrorClass::Protocol,
+            RepoTransportErrorClass::Unknown,
+            RepoTransportErrorClass::Storage,
+        ] {
+            let action = register_dead(&mut coordinator);
+            assert!(matches!(
+                action,
+                crate::parallel::repo_scheduler::TransportRequestAction::Enqueue(_)
+            ));
+            coordinator
+                .complete_transport_result(
+                    transport_envelope(RepoTransportMode::Rrdp, error_class),
+                    time::OffsetDateTime::UNIX_EPOCH,
+                )
+                .expect("complete");
+            coordinator.reset_run_state();
+        }
+        assert!(
+            coordinator
+                .dead_repo_blacklist()
+                .expect("blacklist")
+                .is_empty()
+        );
+    }
+
+    #[test]
+    fn blacklist_skip_rrdp_routes_straight_to_rsync() {
+        let mut seed = DeadRepoBlacklist::new();
+        seed.record_transport_failure(RepoTransportMode::Rrdp, DEAD_NOTIFY, 100, 1, 256);
+        let mut coordinator = coordinator_with_blacklist(3, seed);
+        let action = register_dead(&mut coordinator);
+        let task = match action {
+            crate::parallel::repo_scheduler::TransportRequestAction::Enqueue(task) => task,
+            other => panic!("expected rsync enqueue, got {other:?}"),
+        };
+        assert_eq!(task.mode, RepoTransportMode::Rsync);
+        // register_dead passes retry_short_timeout=false (mirroring a prefetch
+        // snapshot that recorded no real rrdp failure for skipped runs); the
+        // skip path must still keep the short retry profile (#141).
+        assert!(task.retry_short_timeout);
+    }
+
+    #[test]
+    fn blacklist_dual_dead_terminates_immediately() {
+        let mut seed = DeadRepoBlacklist::new();
+        seed.record_transport_failure(RepoTransportMode::Rrdp, DEAD_NOTIFY, 100, 1, 256);
+        seed.record_transport_failure(RepoTransportMode::Rsync, DEAD_BASE, 100, 1, 256);
+        let mut coordinator = coordinator_with_blacklist(3, seed);
+        let action = register_dead(&mut coordinator);
+        let envelope = match action {
+            crate::parallel::repo_scheduler::TransportRequestAction::ReusedTerminalFailure(
+                envelope,
+            ) => envelope,
+            other => panic!("expected terminal failure, got {other:?}"),
+        };
+        match &envelope.result {
+            RepoTransportResultKind::Failed {
+                detail, warnings, ..
+            } => {
+                assert!(detail.contains("dead repo blacklist"));
+                assert!(warnings
+                    .iter()
+                    .any(|warning| warning.message.contains("dead_repo_blacklist_skip_all")));
+            }
+            other => panic!("expected failed result, got {other:?}"),
+        }
+        let record = coordinator
+            .runtime_record(&dead_identity())
+            .expect("record");
+        assert_eq!(
+            record.state,
+            crate::parallel::types::RepoRuntimeState::FailedTerminal
+        );
+        assert_eq!(coordinator.stats.repo_tasks_total, 0);
+        assert_eq!(coordinator.stats.repo_tasks_reused, 1);
+    }
+
+    #[test]
+    fn blacklist_rsync_dead_terminates_after_rrdp_failure() {
+        // Only rsync blacklisted: rrdp is attempted; when it fails, the rsync
+        // fallback is short-circuited into an instant terminal failure.
+        let mut seed = DeadRepoBlacklist::new();
+        seed.record_transport_failure(RepoTransportMode::Rsync, DEAD_BASE, 100, 1, 256);
+        let mut coordinator = coordinator_with_blacklist(3, seed);
+        let action = register_dead(&mut coordinator);
+        assert!(matches!(
+            action,
+            crate::parallel::repo_scheduler::TransportRequestAction::Enqueue(_)
+        ));
+        let completion = coordinator
+            .complete_transport_result(
+                transport_envelope(RepoTransportMode::Rrdp, RepoTransportErrorClass::TransportFetch),
+                time::OffsetDateTime::UNIX_EPOCH,
+            )
+            .expect("complete");
+        assert!(completion.follow_up_tasks.is_empty());
+        let record = coordinator
+            .runtime_record(&dead_identity())
+            .expect("record");
+        assert_eq!(
+            record.state,
+            crate::parallel::types::RepoRuntimeState::FailedTerminal
+        );
+        let terminal = record.terminal_failure.as_ref().expect("terminal failure");
+        match &terminal.result {
+            RepoTransportResultKind::Failed { detail, .. } => {
+                assert!(detail.contains("dead repo blacklist"));
+            }
+            other => panic!("expected failed result, got {other:?}"),
+        }
+    }
+
+    #[test]
+    fn blacklist_success_resets_failure_counter() {
+        let mut coordinator = coordinator_with_blacklist(3, DeadRepoBlacklist::new());
+        // One transport failure starts counting.
+        let action = register_dead(&mut coordinator);
+        assert!(matches!(
+            action,
+            crate::parallel::repo_scheduler::TransportRequestAction::Enqueue(_)
+        ));
+        coordinator
+            .complete_transport_result(
+                transport_envelope(RepoTransportMode::Rrdp, RepoTransportErrorClass::TransportFetch),
+                time::OffsetDateTime::UNIX_EPOCH,
+            )
+            .expect("complete");
+        assert_eq!(coordinator.dead_repo_blacklist().expect("blacklist").len(), 1);
+
+        // A later success clears the entry.
+        coordinator.reset_run_state();
+        let action = register_dead(&mut coordinator);
+        assert!(matches!(
+            action,
+            crate::parallel::repo_scheduler::TransportRequestAction::Enqueue(_)
+        ));
+        let mut success = transport_envelope(RepoTransportMode::Rrdp, RepoTransportErrorClass::Unknown);
+        success.result = RepoTransportResultKind::Success {
+            source: "rrdp".to_string(),
+            warnings: Vec::new(),
+        };
+        coordinator
+            .complete_transport_result(success, time::OffsetDateTime::UNIX_EPOCH)
+            .expect("complete");
+        assert!(
+            coordinator
+                .dead_repo_blacklist()
+                .expect("blacklist")
+                .is_empty()
+        );
+    }
 }

src/parallel/transport_prefetch.rs

@@ -484,6 +484,7 @@ mod tests {
             result: crate::parallel::types::RepoTransportResultKind::Failed {
                 detail: "timeout".to_string(),
                 warnings: Vec::new(),
+                error_class: crate::parallel::types::RepoTransportErrorClass::Unknown,
             },
         });
         recorder.record_result(&crate::parallel::types::RepoTransportResultEnvelope {
@@ -499,6 +500,7 @@ mod tests {
             result: crate::parallel::types::RepoTransportResultKind::Failed {
                 detail: "timeout".to_string(),
                 warnings: Vec::new(),
+                error_class: crate::parallel::types::RepoTransportErrorClass::Unknown,
             },
         });
 

src/parallel/types.rs

@@ -1,5 +1,7 @@
 use std::path::{Path, PathBuf};
 
+use serde::{Deserialize, Serialize};
+
 use crate::policy::SyncPreference;
 use crate::report::Warning;
 
@@ -96,7 +98,8 @@ pub enum RepoDedupKey {
     RsyncScope { rsync_scope_uri: String },
 }
 
-#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+#[derive(Clone, Copy, Debug, PartialEq, Eq, PartialOrd, Ord, Hash, Serialize, Deserialize)]
+#[serde(rename_all = "snake_case")]
 pub enum RepoTransportMode {
     Rrdp,
     Rsync,
@@ -116,6 +119,26 @@ pub struct RepoTransportTask {
     pub requesters: Vec<RepoRequester>,
 }
 
+#[derive(Clone, Copy, Debug, PartialEq, Eq, Serialize, Deserialize)]
+#[serde(rename_all = "snake_case")]
+pub enum RepoTransportErrorClass {
+    /// Transport-layer fetch failure (connect timeout/refused/TLS/DNS/broken
+    /// connection). Only this class counts toward the dead-repo blacklist.
+    TransportFetch,
+    /// Protocol-level failure (HTTP status, XML/parse, content mismatch).
+    Protocol,
+    /// Local storage failure.
+    Storage,
+    /// Unclassified (synthesized or legacy results); never counted.
+    Unknown,
+}
+
+impl Default for RepoTransportErrorClass {
+    fn default() -> Self {
+        RepoTransportErrorClass::Unknown
+    }
+}
+
 #[derive(Clone, Debug, PartialEq, Eq)]
 pub enum RepoTransportResultKind {
     Success {
@@ -125,6 +148,7 @@ pub enum RepoTransportResultKind {
     Failed {
         detail: String,
         warnings: Vec<Warning>,
+        error_class: RepoTransportErrorClass,
     },
 }
 
@@ -479,6 +503,7 @@ mod tests {
             result: RepoTransportResultKind::Failed {
                 detail: "timeout".to_string(),
                 warnings: vec![Warning::new("timeout")],
+                error_class: super::RepoTransportErrorClass::Unknown,
             },
         };
         assert!(matches!(ok.result, RepoTransportResultKind::Success { .. }));

src/policy.rs

@@ -1,4 +1,5 @@
 use serde::{Deserialize, Serialize};
+use std::collections::BTreeSet;
 
 #[derive(Clone, Copy, Debug, PartialEq, Eq, Serialize, Deserialize)]
 #[serde(rename_all = "snake_case")]
@@ -127,6 +128,11 @@ pub struct Policy {
     pub signed_object_failure_policy: SignedObjectFailurePolicy,
     pub resource_validation_mode: ResourceValidationMode,
     pub strict: StrictPolicy,
+    /// Ephemeral verification-only override.  This is deliberately excluded
+    /// from policy files, contracts, and policy fingerprints: it only selects
+    /// the already contract-bound VCIR fallback branch while replaying a run.
+    #[serde(skip)]
+    pub verification_forced_vcir_reuse_manifest_uris: BTreeSet<String>,
 }
 
 impl Default for Policy {
@@ -137,6 +143,7 @@ impl Default for Policy {
             signed_object_failure_policy: SignedObjectFailurePolicy::default(),
             resource_validation_mode: ResourceValidationMode::default(),
             strict: StrictPolicy::default(),
+            verification_forced_vcir_reuse_manifest_uris: BTreeSet::new(),
         }
     }
 }

src/query/report_stream.rs

@@ -176,9 +176,14 @@ pub fn list_report_objects_filtered(
         start_pp_index,
         start_object_index,
     );
-    ObjectScanSeed { state: &mut state }
+    let scan = ObjectScanSeed { state: &mut state }.deserialize(&mut deserializer);
-        .deserialize(&mut deserializer)
+    match scan {
-        .map_err(QueryDbError::from)?;
+        Ok(()) => {}
+        // A full page plus its next cursor is all the caller needs; the
+        // visitor aborts the remaining scan with a sentinel error.
+        Err(err) if is_object_scan_complete(&err) => {}
+        Err(err) => return Err(QueryDbError::from(err)),
+    }
     Ok(QueryPage {
         data: state.data,
         next_cursor: state.next_cursor,
@@ -215,6 +220,16 @@ pub fn object_cursor(pp_index: u64, object_index: u64) -> String {
     format!("r1:{pp_index}:{object_index}")
 }
 
+/// Sentinel message used to abort the streaming object scan early once a list
+/// page (plus its next cursor) has been collected. The deserializer is left
+/// mid-document and dropped, which is fine because the caller only needs the
+/// state accumulated so far.
+const OBJECT_SCAN_COMPLETE_MSG: &str = "rpki report object scan complete: page collected";
+
+fn is_object_scan_complete(err: &serde_json::Error) -> bool {
+    err.is_data() && err.to_string().starts_with(OBJECT_SCAN_COMPLETE_MSG)
+}
+
 fn parse_object_cursor(cursor: Option<&str>) -> QueryDbResult<(u64, u64)> {
     let Some(cursor) = cursor else {
         return Ok((0, 0));
@@ -598,15 +613,15 @@ impl<'de> Deserialize<'de> for ObjectStatsSeq {
                     stats.count += 1;
                     if object.result == "error" {
                         stats.rejected += 1;
+                        if let Some(detail) = object.detail.filter(|value| !value.is_empty()) {
+                            bump_limited_count(&mut stats.reason_counts, &detail, 512);
+                        }
                     }
                     *stats
                         .object_type_counts
                         .entry(object.object_type)
                         .or_default() += 1;
                     *stats.result_counts.entry(object.result).or_default() += 1;
-                    if let Some(detail) = object.detail.filter(|value| !value.is_empty()) {
-                        bump_limited_count(&mut stats.reason_counts, &detail, 512);
-                    }
                 }
                 Ok(ObjectStatsSeq(stats))
             }
@@ -732,6 +747,15 @@ impl<'a> ObjectScanState<'a> {
             scanned_objects: 0,
         }
     }
+
+    /// True once a list page is full and its next cursor is fixed; everything
+    /// after this point cannot alter the page (cursor semantics: the cursor
+    /// already points at the first item of the next page).
+    fn list_page_complete(&self) -> bool {
+        matches!(self.mode, ObjectScanMode::List)
+            && self.data.len() >= self.limit
+            && self.next_cursor.is_some()
+    }
 }
 
 #[derive(Clone, Debug, PartialEq, Eq)]
@@ -807,7 +831,13 @@ impl<'de> Visitor<'de> for ObjectScanPpsVisitor<'_, '_> {
         A: SeqAccess<'de>,
     {
         let mut pp_index = 0u64;
-        while let Some(raw_pp) = seq.next_element::<Box<RawValue>>()? {
+        loop {
+            if self.state.list_page_complete() {
+                return Err(de::Error::custom(OBJECT_SCAN_COMPLETE_MSG));
+            }
+            let Some(raw_pp) = seq.next_element::<Box<RawValue>>()? else {
+                break;
+            };
             process_raw_publication_point(self.state, pp_index, raw_pp.get())
                 .map_err(de::Error::custom)?;
             pp_index += 1;
@@ -822,6 +852,20 @@ fn process_raw_publication_point(
     raw_pp: &str,
 ) -> QueryDbResult<()> {
     state.scanned_publication_points += 1;
+    if matches!(state.mode, ObjectScanMode::List) && pp_index < state.start_pp_index {
+        // Every object in this PP sits before the cursor, so none of them can
+        // be part of the page: skip materializing the subtree entirely.
+        return Ok(());
+    }
+    if state.scope != ObjectScope::All {
+        // Scope is decidable from PP-level fields alone; probe them before
+        // materializing the (potentially huge) objects subtree.
+        let probe: PublicationPointScopeProbe = serde_json::from_str(raw_pp)?;
+        let ctx = PpObjectContext::from_pp(state.run_id, &probe.into_summary());
+        if !state.scope.matches(&ctx) {
+            return Ok(());
+        }
+    }
     let pp: PublicationPointForObjects = serde_json::from_str(raw_pp)?;
     let ctx = PpObjectContext::from_pp(state.run_id, &pp.summary);
     if !state.scope.matches(&ctx) {
@@ -909,6 +953,30 @@ impl<'de> Visitor<'de> for PublicationPointForObjectsVisitor {
     }
 }
 
+/// Lightweight probe of the PP-level fields needed to decide `ObjectScope`
+/// before the (potentially huge) objects subtree is materialized.
+#[derive(Clone, Debug, Default, Deserialize)]
+struct PublicationPointScopeProbe {
+    node_id: Option<u64>,
+    rsync_base_uri: Option<String>,
+    manifest_rsync_uri: Option<String>,
+    publication_point_rsync_uri: Option<String>,
+    rrdp_notification_uri: Option<String>,
+}
+
+impl PublicationPointScopeProbe {
+    fn into_summary(self) -> PublicationPointSummary {
+        PublicationPointSummary {
+            node_id: self.node_id,
+            rsync_base_uri: self.rsync_base_uri,
+            manifest_rsync_uri: self.manifest_rsync_uri,
+            publication_point_rsync_uri: self.publication_point_rsync_uri,
+            rrdp_notification_uri: self.rrdp_notification_uri,
+            ..PublicationPointSummary::default()
+        }
+    }
+}
+
 #[derive(Clone, Debug, Default)]
 struct ReportObject {
     uri: String,
@@ -1307,7 +1375,7 @@ mod tests {
                 "repo_terminal_state":"fresh",
                 "warnings":[],
                 "objects":[
-                  {"rsync_uri":"rsync://repo.example/rpki/m.mft","sha256_hex":"11","kind":"manifest","result":"ok"},
+                  {"rsync_uri":"rsync://repo.example/rpki/m.mft","sha256_hex":"11","kind":"manifest","result":"ok","detail":"restored child CA instance from publication-point cache"},
                   {"rsync_uri":"rsync://repo.example/rpki/a.roa","sha256_hex":"22","kind":"roa","result":"error","detail":"bad roa"}
                 ]
               }],
@@ -1333,6 +1401,13 @@ mod tests {
         assert_eq!(summary.aspas_count, 1);
         assert_eq!(summary.stats.object_type_counts["roa"], 1);
         assert_eq!(summary.stats.reason_counts["bad roa"], 1);
+        // #133: informational detail on non-error objects must not be counted
+        // as reject reasons.
+        assert_eq!(summary.stats.reason_counts.len(), 1);
+        assert!(!summary
+            .stats
+            .reason_counts
+            .contains_key("restored child CA instance from publication-point cache"));
 
         let page =
             list_report_objects(&report_path, "run_0001", ObjectScope::All, 1, None).expect("page");
@@ -1360,4 +1435,195 @@ mod tests {
         assert_eq!(found.object.result, "error");
         assert_eq!(found.resolution.mode, "report_scan");
     }
+
+    /// Three PPs with two objects each; used to anchor paging and scope
+    /// behavior of the early-stop object scan.
+    fn write_multi_pp_report(report_path: &std::path::Path) {
+        fs::write(
+            report_path,
+            r#"{
+              "meta":{"validation_time_rfc3339_utc":"2026-06-16T00:00:00Z"},
+              "tree":{"warnings":[]},
+              "publication_points":[
+                {
+                  "node_id":1,
+                  "rsync_base_uri":"rsync://repo-a.example/rpki/",
+                  "manifest_rsync_uri":"rsync://repo-a.example/rpki/m.mft",
+                  "publication_point_rsync_uri":"rsync://repo-a.example/rpki/",
+                  "rrdp_notification_uri":"https://repo-a.example/rrdp/notification.xml",
+                  "objects":[
+                    {"rsync_uri":"rsync://repo-a.example/rpki/m.mft","sha256_hex":"a1","kind":"manifest","result":"ok"},
+                    {"rsync_uri":"rsync://repo-a.example/rpki/a.roa","sha256_hex":"a2","kind":"roa","result":"ok"}
+                  ]
+                },
+                {
+                  "node_id":2,
+                  "rsync_base_uri":"rsync://repo-b.example/rpki/",
+                  "manifest_rsync_uri":"rsync://repo-b.example/rpki/m.mft",
+                  "publication_point_rsync_uri":"rsync://repo-b.example/rpki/",
+                  "objects":[
+                    {"rsync_uri":"rsync://repo-b.example/rpki/m.mft","sha256_hex":"b1","kind":"manifest","result":"ok"},
+                    {"rsync_uri":"rsync://repo-b.example/rpki/b.roa","sha256_hex":"b2","kind":"roa","result":"error","detail":"bad roa"}
+                  ]
+                },
+                {
+                  "node_id":3,
+                  "rsync_base_uri":"rsync://repo-c.example/rpki/",
+                  "manifest_rsync_uri":"rsync://repo-c.example/rpki/m.mft",
+                  "publication_point_rsync_uri":"rsync://repo-c.example/rpki/",
+                  "rrdp_notification_uri":"https://repo-c.example/rrdp/notification.xml",
+                  "objects":[
+                    {"rsync_uri":"rsync://repo-c.example/rpki/m.mft","sha256_hex":"c1","kind":"manifest","result":"ok"},
+                    {"rsync_uri":"rsync://repo-c.example/rpki/c.roa","sha256_hex":"c2","kind":"roa","result":"ok"}
+                  ]
+                }
+              ]
+            }"#,
+        )
+        .expect("report");
+    }
+
+    #[test]
+    fn object_listing_pages_across_publication_points_with_stable_cursors() {
+        let temp = tempfile::tempdir().expect("tempdir");
+        let report_path = temp.path().join("report.json");
+        write_multi_pp_report(&report_path);
+
+        // Page 1: fills inside PP 0; the cursor must resume at PP 0 object 2.
+        let page1 = list_report_objects(&report_path, "run_0001", ObjectScope::All, 2, None)
+            .expect("page1");
+        assert_eq!(page1.data.len(), 2);
+        assert_eq!(page1.data[0].uri, "rsync://repo-a.example/rpki/m.mft");
+        assert_eq!(page1.data[1].uri, "rsync://repo-a.example/rpki/a.roa");
+        assert_eq!(page1.next_cursor.as_deref(), Some("r1:0:2"));
+
+        // Page 2: PP 0 is exhausted; the two PP 1 objects fill the page.
+        let page2 = list_report_objects(
+            &report_path,
+            "run_0001",
+            ObjectScope::All,
+            2,
+            page1.next_cursor.as_deref(),
+        )
+        .expect("page2");
+        assert_eq!(page2.data.len(), 2);
+        assert_eq!(page2.data[0].uri, "rsync://repo-b.example/rpki/m.mft");
+        assert_eq!(page2.data[1].uri, "rsync://repo-b.example/rpki/b.roa");
+        assert_eq!(page2.next_cursor.as_deref(), Some("r1:1:2"));
+
+        // Page 3: last PP, page fills at the final object.
+        let page3 = list_report_objects(
+            &report_path,
+            "run_0001",
+            ObjectScope::All,
+            2,
+            page2.next_cursor.as_deref(),
+        )
+        .expect("page3");
+        assert_eq!(page3.data.len(), 2);
+        assert_eq!(page3.next_cursor.as_deref(), Some("r1:2:2"));
+
+        // Page 4: nothing left.
+        let page4 = list_report_objects(
+            &report_path,
+            "run_0001",
+            ObjectScope::All,
+            2,
+            page3.next_cursor.as_deref(),
+        )
+        .expect("page4");
+        assert!(page4.data.is_empty());
+        assert_eq!(page4.next_cursor, None);
+
+        // A page larger than the remaining tail must not produce a cursor.
+        let tail = list_report_objects(&report_path, "run_0001", ObjectScope::All, 10, None)
+            .expect("tail");
+        assert_eq!(tail.data.len(), 6);
+        assert_eq!(tail.next_cursor, None);
+    }
+
+    #[test]
+    fn object_listing_scope_skips_non_matching_publication_points() {
+        let temp = tempfile::tempdir().expect("tempdir");
+        let report_path = temp.path().join("report.json");
+        write_multi_pp_report(&report_path);
+
+        // PP scope: only PP 1 (node_id 2) objects, cursor resumes within PP 1.
+        let page1 = list_report_objects(
+            &report_path,
+            "run_0001",
+            ObjectScope::PublicationPoint("node_2".to_string()),
+            1,
+            None,
+        )
+        .expect("page1");
+        assert_eq!(page1.data.len(), 1);
+        assert_eq!(page1.data[0].uri, "rsync://repo-b.example/rpki/m.mft");
+        assert_eq!(page1.next_cursor.as_deref(), Some("r1:1:1"));
+        let page2 = list_report_objects(
+            &report_path,
+            "run_0001",
+            ObjectScope::PublicationPoint("node_2".to_string()),
+            10,
+            page1.next_cursor.as_deref(),
+        )
+        .expect("page2");
+        assert_eq!(page2.data.len(), 1);
+        assert_eq!(page2.data[0].uri, "rsync://repo-b.example/rpki/b.roa");
+        assert_eq!(page2.next_cursor, None);
+
+        // Repo scope derived from the PP without rrdp uri (repo-b).
+        let probe_pp = list_report_objects(
+            &report_path,
+            "run_0001",
+            ObjectScope::PublicationPoint("node_2".to_string()),
+            10,
+            None,
+        )
+        .expect("probe");
+        let repo_id = probe_pp.data[0].repo_id.clone();
+        let repo_page = list_report_objects(
+            &report_path,
+            "run_0001",
+            ObjectScope::Repo(repo_id),
+            10,
+            None,
+        )
+        .expect("repo page");
+        assert_eq!(repo_page.data.len(), 2);
+        assert!(
+            repo_page
+                .data
+                .iter()
+                .all(|object| object.uri.contains("repo-b.example"))
+        );
+
+        // Unknown scope id: empty page, no cursor, no error.
+        let empty = list_report_objects(
+            &report_path,
+            "run_0001",
+            ObjectScope::PublicationPoint("node_999".to_string()),
+            10,
+            None,
+        )
+        .expect("empty");
+        assert!(empty.data.is_empty());
+        assert_eq!(empty.next_cursor, None);
+
+        // Filtered listing keeps working across skipped PPs.
+        let mut filter = ObjectFilter::default();
+        filter.query = Some("c.roa".to_string());
+        let filtered = list_report_objects_filtered(
+            &report_path,
+            "run_0001",
+            ObjectScope::All,
+            &filter,
+            10,
+            None,
+        )
+        .expect("filtered");
+        assert_eq!(filtered.data.len(), 1);
+        assert_eq!(filtered.data[0].uri, "rsync://repo-c.example/rpki/c.roa");
+        assert_eq!(filtered.next_cursor, None);
+    }
 }

src/query_db.rs

@@ -60,6 +60,8 @@ pub enum QueryDbError {
     MissingColumnFamily(&'static str),
     #[error("invalid run artifact: {0}")]
     InvalidArtifact(String),
+    #[error("report data for run {0} is no longer available (retention cleanup)")]
+    RunDataGone(String),
     #[error("CIR decode error: {0}")]
     CirDecode(String),
 }
@@ -160,6 +162,24 @@ pub struct RepositoryRecord {
     pub terminal_states: BTreeMap<String, u64>,
 }
 
+/// Sort order for `QueryDb::list_repos_sorted`.
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+pub enum RepoSortOrder {
+    ObjectsDesc,
+    ObjectsAsc,
+}
+
+/// Parse a `repos:<offset>` cursor used by sorted repository listings.
+fn parse_repos_offset_cursor(cursor: Option<&str>) -> QueryDbResult<usize> {
+    match cursor {
+        None => Ok(0),
+        Some(raw) => raw
+            .strip_prefix("repos:")
+            .and_then(|value| value.parse::<usize>().ok())
+            .ok_or_else(|| QueryDbError::InvalidArtifact(format!("invalid repos cursor: {raw}"))),
+    }
+}
+
 #[derive(Clone, Debug, PartialEq, Eq, Serialize, Deserialize)]
 #[serde(rename_all = "camelCase")]
 pub struct PublicationPointRecord {
@@ -289,6 +309,7 @@ impl QueryDb {
         opts.create_if_missing(true);
         opts.create_missing_column_families(true);
         opts.set_compression_type(rocksdb::DBCompressionType::Lz4);
+        opts.set_max_open_files(512);
         let descriptors = QUERY_DB_COLUMN_FAMILIES
             .iter()
             .map(|name| ColumnFamilyDescriptor::new(*name, cf_options()))
@@ -310,6 +331,7 @@ impl QueryDb {
         opts.create_if_missing(false);
         opts.create_missing_column_families(false);
         opts.set_compression_type(rocksdb::DBCompressionType::Lz4);
+        opts.set_max_open_files(512);
         let descriptors = QUERY_DB_COLUMN_FAMILIES
             .iter()
             .map(|name| ColumnFamilyDescriptor::new(*name, cf_options()))
@@ -351,8 +373,10 @@ impl QueryDb {
     pub fn resolve_run_id(&self, run_id: &str) -> QueryDbResult<Option<String>> {
         if run_id == "latest" || run_id == "latest_run" {
             self.latest_ready_run()
-        } else {
+        } else if self.get_run(run_id)?.is_some() {
             Ok(Some(run_id.to_string()))
+        } else {
+            Ok(None)
         }
     }
 
@@ -361,7 +385,38 @@ impl QueryDb {
         limit: usize,
         cursor: Option<&str>,
     ) -> QueryDbResult<QueryPage<RunRecord>> {
-        self.list_json_by_prefix(CF_RUNS, "run/", limit, cursor)
+        // Runs are listed newest first: zero-padded run ids sort by age, so a
+        // reverse scan over the `run/` prefix yields the latest run on page 1.
+        // Cursor semantics mirror `list_json_by_prefix`: the cursor is the
+        // first key of the next page (inclusive).
+        let limit = limit.clamp(1, 1000);
+        let cf = self.cf(CF_RUNS)?;
+        let prefix = "run/";
+        let start = match cursor {
+            Some(cursor) => cursor.as_bytes().to_vec(),
+            None => prefix_range_end(prefix.as_bytes())
+                .unwrap_or_else(|| prefix.as_bytes().to_vec()),
+        };
+        let mut data = Vec::new();
+        let mut next_cursor = None;
+        let mode = IteratorMode::From(&start, rocksdb::Direction::Reverse);
+        for item in self.db.iterator_cf(cf, mode) {
+            let (key, value) = item?;
+            let key_str = String::from_utf8_lossy(&key);
+            if !key_str.starts_with(prefix) {
+                break;
+            }
+            if data.len() >= limit {
+                next_cursor = Some(key_str.to_string());
+                break;
+            }
+            data.push(serde_json::from_slice(&value)?);
+        }
+        Ok(QueryPage {
+            data,
+            next_cursor,
+            limit,
+        })
     }
 
     pub fn list_repos(
@@ -373,6 +428,58 @@ impl QueryDb {
         self.list_json_by_prefix(CF_REPOS, &format!("repo/{run_id}/"), limit, cursor)
     }
 
+    /// List repositories sorted by object count. The sort happens at query
+    /// time over the run's repo set (hundreds of records at most), so no
+    /// extra index is needed. Pagination uses a `repos:<offset>` cursor,
+    /// independent from the index-order key cursor used by `list_repos`.
+    pub fn list_repos_sorted(
+        &self,
+        run_id: &str,
+        limit: usize,
+        cursor: Option<&str>,
+        order: RepoSortOrder,
+    ) -> QueryDbResult<QueryPage<RepositoryRecord>> {
+        let cf = self.cf(CF_REPOS)?;
+        let prefix = format!("repo/{run_id}/");
+        let mut records = Vec::new();
+        for item in self.db.iterator_cf(
+            cf,
+            IteratorMode::From(prefix.as_bytes(), rocksdb::Direction::Forward),
+        ) {
+            let (key, value) = item?;
+            if !key.starts_with(prefix.as_bytes()) {
+                break;
+            }
+            records.push(serde_json::from_slice::<RepositoryRecord>(&value)?);
+        }
+        records.sort_by(|left, right| {
+            let by_objects = match order {
+                RepoSortOrder::ObjectsDesc => right.objects.cmp(&left.objects),
+                RepoSortOrder::ObjectsAsc => left.objects.cmp(&right.objects),
+            };
+            // Stable, deterministic order for equal object counts.
+            by_objects.then_with(|| left.repo_id.cmp(&right.repo_id))
+        });
+        let offset = parse_repos_offset_cursor(cursor)?;
+        let bounded_limit = limit.clamp(1, 1000);
+        let total = records.len();
+        let data = records
+            .into_iter()
+            .skip(offset)
+            .take(bounded_limit)
+            .collect::<Vec<_>>();
+        let next_cursor = if offset + bounded_limit < total {
+            Some(format!("repos:{}", offset + bounded_limit))
+        } else {
+            None
+        };
+        Ok(QueryPage {
+            data,
+            next_cursor,
+            limit: bounded_limit,
+        })
+    }
+
     pub fn lookup_vrps(
         &self,
         run_id: &str,
@@ -691,9 +798,17 @@ impl QueryDb {
     }
 
     fn report_path_for_run(&self, run_id: &str) -> QueryDbResult<Option<PathBuf>> {
-        Ok(self
+        let Some(run) = self.get_run(run_id)? else {
-            .get_run(run_id)?
+            return Ok(None);
-            .map(|run| Path::new(&run.run_dir).join("report.json")))
+        };
+        let path = Path::new(&run.run_dir).join("report.json");
+        if !path.exists() {
+            // The run is still indexed but its report.json was removed by
+            // retention cleanup; callers must be able to tell this apart
+            // from a genuine io failure (mapped to 404 upstream).
+            return Err(QueryDbError::RunDataGone(run_id.to_string()));
+        }
+        Ok(Some(path))
     }
 
     pub fn get_object_projection(
@@ -1094,6 +1209,7 @@ impl QueryDb {
 fn cf_options() -> Options {
     let mut opts = Options::default();
     opts.set_compression_type(rocksdb::DBCompressionType::Lz4);
+    opts.set_max_open_files(512);
     opts
 }
 
@@ -1951,15 +2067,17 @@ mod tests {
             db.resolve_run_id("run_0001").unwrap().as_deref(),
             Some("run_0001")
         );
+        assert_eq!(db.resolve_run_id("run_9999").unwrap(), None);
 
         let first_page = db.list_runs(1, None).expect("runs");
         assert_eq!(first_page.data.len(), 1);
-        assert_eq!(first_page.data[0].run_id, "run_0001");
+        assert_eq!(first_page.data[0].run_id, "run_0002");
         let second_page = db
             .list_runs(1, first_page.next_cursor.as_deref())
             .expect("second runs");
         assert_eq!(second_page.data.len(), 1);
-        assert_eq!(second_page.data[0].run_id, "run_0002");
+        assert_eq!(second_page.data[0].run_id, "run_0001");
+        assert!(second_page.next_cursor.is_none());
 
         let repos = db.list_repos("run_0002", 10, None).expect("repos");
         assert_eq!(repos.data.len(), 1);
@@ -2105,6 +2223,43 @@ mod tests {
         );
     }
 
+    #[test]
+    fn missing_report_file_is_reported_as_run_data_gone() {
+        let temp = tempfile::tempdir().expect("tempdir");
+        let run1 = temp.path().join("runs/run_0001");
+        fs::create_dir_all(&run1).expect("run1");
+        write_sample_run(&run1, "run_0001", 1);
+
+        let query_db_path = temp.path().join("query-db");
+        index_artifacts(&ArtifactIndexerConfig {
+            query_db_path: query_db_path.clone(),
+            run_root: Some(temp.path().to_path_buf()),
+            run_dir: None,
+            repo_bytes_db_path: None,
+            projection_entry_limit: 50,
+            min_run_seq: None,
+            retain_indexed_runs: None,
+        })
+        .expect("index");
+        let db = QueryDb::open(&query_db_path).expect("open query db");
+        assert_eq!(db.list_objects("run_0001", 10, None).unwrap().data.len(), 2);
+
+        // Simulate retention cleanup removing the run directory while the
+        // index record survives: object listing/lookup must fail with a
+        // distinguishable error instead of a bare io error.
+        fs::remove_dir_all(&run1).expect("remove run dir");
+        let err = db.list_objects("run_0001", 10, None).unwrap_err();
+        assert!(
+            matches!(err, QueryDbError::RunDataGone(ref run_id) if run_id == "run_0001"),
+            "unexpected error: {err}"
+        );
+        let err = db.get_object_by_sha256("run_0001", "22").unwrap_err();
+        assert!(
+            matches!(err, QueryDbError::RunDataGone(_)),
+            "unexpected error: {err}"
+        );
+    }
+
     #[test]
     fn repeated_indexing_does_not_move_latest_backwards() {
         let temp = tempfile::tempdir().expect("tempdir");
@@ -2201,6 +2356,163 @@ mod tests {
         .expect("report");
     }
 
+    #[test]
+    fn query_db_list_repos_sorted_orders_by_object_count() {
+        let temp = tempfile::tempdir().expect("tempdir");
+        let run1 = temp.path().join("runs/run_0001");
+        fs::create_dir_all(&run1).expect("run1");
+        // Index order (a, b, c) deliberately differs from object-count order.
+        write_sample_run_with_repos(
+            &run1,
+            "run_0001",
+            1,
+            &[
+                ("repo-a.example", 2),
+                ("repo-b.example", 5),
+                ("repo-c.example", 1),
+            ],
+        );
+
+        let query_db_path = temp.path().join("query-db");
+        index_artifacts(&ArtifactIndexerConfig {
+            query_db_path: query_db_path.clone(),
+            run_root: Some(temp.path().to_path_buf()),
+            run_dir: None,
+            repo_bytes_db_path: None,
+            projection_entry_limit: 50,
+            min_run_seq: None,
+            retain_indexed_runs: None,
+        })
+        .expect("index");
+
+        let db = QueryDb::open(&query_db_path).expect("open query db");
+        let hosts = |page: QueryPage<RepositoryRecord>| {
+            page.data
+                .iter()
+                .map(|repo| (repo.host.clone(), repo.objects))
+                .collect::<Vec<_>>()
+        };
+
+        let desc = db
+            .list_repos_sorted("run_0001", 10, None, RepoSortOrder::ObjectsDesc)
+            .expect("desc");
+        assert_eq!(
+            hosts(desc),
+            vec![
+                ("repo-b.example".to_string(), 5),
+                ("repo-a.example".to_string(), 2),
+                ("repo-c.example".to_string(), 1),
+            ]
+        );
+
+        let asc = db
+            .list_repos_sorted("run_0001", 10, None, RepoSortOrder::ObjectsAsc)
+            .expect("asc");
+        assert_eq!(
+            hosts(asc),
+            vec![
+                ("repo-c.example".to_string(), 1),
+                ("repo-a.example".to_string(), 2),
+                ("repo-b.example".to_string(), 5),
+            ]
+        );
+
+        // Offset cursor pagination over the sorted listing.
+        let first = db
+            .list_repos_sorted("run_0001", 2, None, RepoSortOrder::ObjectsDesc)
+            .expect("first page");
+        assert_eq!(first.data.len(), 2);
+        assert_eq!(first.data[0].host, "repo-b.example");
+        assert_eq!(first.next_cursor.as_deref(), Some("repos:2"));
+        let second = db
+            .list_repos_sorted(
+                "run_0001",
+                2,
+                first.next_cursor.as_deref(),
+                RepoSortOrder::ObjectsDesc,
+            )
+            .expect("second page");
+        assert_eq!(second.data.len(), 1);
+        assert_eq!(second.data[0].host, "repo-c.example");
+        assert!(second.next_cursor.is_none());
+
+        // A key-style cursor from the index-order listing is rejected.
+        assert!(db
+            .list_repos_sorted("run_0001", 2, Some("repo/run_0001/x"), RepoSortOrder::ObjectsDesc)
+            .is_err());
+    }
+
+    /// Sample run with one publication point per repo host, each holding
+    /// `object_count` ok objects, so repo records get distinct object counts.
+    fn write_sample_run_with_repos(
+        run_dir: &Path,
+        run_id: &str,
+        run_seq: u64,
+        repo_specs: &[(&str, usize)],
+    ) {
+        let publication_points = repo_specs
+            .iter()
+            .enumerate()
+            .map(|(index, (host, object_count))| {
+                let objects = (0..*object_count)
+                    .map(|i| {
+                        json!({
+                            "rsync_uri": format!("rsync://{host}/rpki/obj{i}.roa"),
+                            "sha256_hex": format!("{index:02x}{i:02x}"),
+                            "kind": "roa",
+                            "result": "ok"
+                        })
+                    })
+                    .collect::<Vec<_>>();
+                json!({
+                    "node_id": 10 + index,
+                    "rsync_base_uri": format!("rsync://{host}/rpki/"),
+                    "manifest_rsync_uri": format!("rsync://{host}/rpki/m.mft"),
+                    "publication_point_rsync_uri": format!("rsync://{host}/rpki/"),
+                    "rrdp_notification_uri": format!("https://{host}/rrdp/notification.xml"),
+                    "source": "rrdp",
+                    "repo_sync_source": "rrdp",
+                    "repo_sync_phase": "rrdp_snapshot",
+                    "repo_sync_duration_ms": 100,
+                    "repo_terminal_state": "fresh",
+                    "warnings": [],
+                    "objects": objects,
+                })
+            })
+            .collect::<Vec<_>>();
+        let report = json!({
+            "format_version": 2,
+            "meta": {"validation_time_rfc3339_utc": "2026-06-15T00:00:00Z"},
+            "tree": {"warnings": []},
+            "publication_points": publication_points,
+            "vrps": [],
+            "aspas": [],
+            "downloads": [],
+            "download_stats": {},
+            "repo_sync_stats": {}
+        });
+        fs::write(
+            run_dir.join("report.json"),
+            serde_json::to_vec(&report).unwrap(),
+        )
+        .expect("report");
+        let summary = json!({
+            "status": "success",
+            "runId": run_id,
+            "runSeq": run_seq,
+            "startedAtRfc3339Utc": "2026-06-15T00:00:00Z",
+            "finishedAtRfc3339Utc": "2026-06-15T00:01:00Z",
+            "wallMs": 60000,
+            "reportCounts": {"vrps": 0, "aspas": 0, "publicationPoints": repo_specs.len(), "warnings": 0}
+        });
+        fs::write(
+            run_dir.join("run-summary.json"),
+            serde_json::to_vec(&summary).unwrap(),
+        )
+        .expect("summary");
+        fs::write(run_dir.join("stage-timing.json"), b"{}").expect("stage");
+    }
+
     fn write_sample_run(run_dir: &Path, run_id: &str, run_seq: u64) {
         let report = json!({
             "format_version": 2,

src/storage.rs

@@ -613,6 +613,12 @@ pub struct VcirFailedFetchReuseIdentity {
     pub effective_until: PackTime,
 }
 
+#[derive(Clone, Debug, Default, PartialEq, Eq, Serialize)]
+pub struct VcirReuseRecordsCleared {
+    pub vcir_records: u64,
+    pub failed_fetch_identity_records: u64,
+}
+
 impl VcirFailedFetchReuseIdentity {
     pub fn validate_internal(&self) -> StorageResult<()> {
         let effective_not_before = parse_time(
@@ -1660,6 +1666,9 @@ pub struct PublicationPointCacheObject {
     pub object_type: Option<String>,
     #[serde(rename = "s")]
     pub validation_status: VcirArtifactValidationStatus,
+    /// See `VcirRelatedArtifact::reject_reason`.
+    #[serde(rename = "e", default, skip_serializing_if = "Option::is_none")]
+    pub reject_reason: Option<String>,
 }
 
 impl PublicationPointCacheObject {
@@ -1671,6 +1680,7 @@ impl PublicationPointCacheObject {
             sha256: artifact.sha256.clone(),
             object_type: artifact.object_type.clone(),
             validation_status: artifact.validation_status,
+            reject_reason: artifact.reject_reason.clone(),
         }
     }
 
@@ -1682,6 +1692,7 @@ impl PublicationPointCacheObject {
             sha256: self.sha256.clone(),
             object_type: self.object_type.clone(),
             validation_status: self.validation_status,
+            reject_reason: self.reject_reason.clone(),
         }
     }
 
@@ -2157,6 +2168,11 @@ pub struct VcirRelatedArtifact {
     pub object_type: Option<String>,
     #[serde(rename = "s")]
     pub validation_status: VcirArtifactValidationStatus,
+    /// Reject reason captured at fresh validation time for `Rejected` artifacts.
+    /// `None` for accepted/warning artifacts and for cache entries written
+    /// before this field existed.
+    #[serde(rename = "e", default, skip_serializing_if = "Option::is_none")]
+    pub reject_reason: Option<String>,
 }
 
 impl VcirRelatedArtifact {
@@ -4523,6 +4539,36 @@ impl RocksStore {
         Ok(out)
     }
 
+    /// Remove every reusable VCIR and its failed-fetch identity while retaining
+    /// manifest replay metadata.  The replay metadata is a fresh-validation
+    /// guard, not a reusable validation result; removing it changes the normal
+    /// traversal semantics for manifests that are not selected for replay.
+    ///
+    /// This is intentionally narrower than [`Self::delete_vcir`].  It is used
+    /// by verification-only after cloning a work-db so that it can import an
+    /// explicit, contract-bound allowlist of fallback projections.
+    pub fn clear_vcir_reuse_records(&self) -> StorageResult<VcirReuseRecordsCleared> {
+        let vcir_cf = self.cf(CF_VCIR)?;
+        let identity_cf = self.cf(CF_VCIR_FAILED_FETCH_REUSE_IDENTITY)?;
+        let mut batch = WriteBatch::default();
+        let mut summary = VcirReuseRecordsCleared::default();
+
+        for entry in self.db.iterator_cf(vcir_cf, IteratorMode::Start) {
+            let (key, _value) = entry.map_err(|error| StorageError::RocksDb(error.to_string()))?;
+            batch.delete_cf(vcir_cf, key);
+            summary.vcir_records += 1;
+        }
+        for entry in self.db.iterator_cf(identity_cf, IteratorMode::Start) {
+            let (key, _value) = entry.map_err(|error| StorageError::RocksDb(error.to_string()))?;
+            batch.delete_cf(identity_cf, key);
+            summary.failed_fetch_identity_records += 1;
+        }
+        if summary.vcir_records > 0 || summary.failed_fetch_identity_records > 0 {
+            self.write_batch(batch)?;
+        }
+        Ok(summary)
+    }
+
     pub fn summarize_vcir_storage(&self) -> StorageResult<VcirStorageSummary> {
         let cf = self.cf(CF_VCIR)?;
         let mode = IteratorMode::Start;

src/storage/tests.rs

@@ -257,6 +257,7 @@ fn sample_vcir(manifest_rsync_uri: &str) -> ValidatedCaInstanceResult {
                 sha256: sha256_hex(b"manifest-object"),
                 object_type: Some("mft".to_string()),
                 validation_status: VcirArtifactValidationStatus::Accepted,
+                reject_reason: None,
             },
             VcirRelatedArtifact {
                 artifact_role: VcirArtifactRole::CurrentCrl,
@@ -265,6 +266,7 @@ fn sample_vcir(manifest_rsync_uri: &str) -> ValidatedCaInstanceResult {
                 sha256: sha256_hex(b"current-crl"),
                 object_type: Some("crl".to_string()),
                 validation_status: VcirArtifactValidationStatus::Accepted,
+                reject_reason: None,
             },
         ],
         summary: VcirSummary {
@@ -1308,6 +1310,111 @@ fn vcir_roundtrip_and_validation_failures_are_reported() {
     );
 }
 
+#[test]
+fn clear_vcir_reuse_records_preserves_manifest_replay_meta() {
+    let td = tempfile::tempdir().expect("tempdir");
+    let store = RocksStore::open(td.path()).expect("open rocksdb");
+    let vcir = sample_vcir("rsync://example.test/repo/current.mft");
+    store.put_vcir(&vcir).expect("put vcir");
+    let replay_meta = store
+        .get_manifest_replay_meta(&vcir.manifest_rsync_uri)
+        .expect("get replay meta")
+        .expect("replay meta exists");
+
+    let cleared = store
+        .clear_vcir_reuse_records()
+        .expect("clear reusable VCIR records");
+    assert_eq!(cleared.vcir_records, 1);
+    assert_eq!(cleared.failed_fetch_identity_records, 0);
+    assert!(store
+        .get_vcir(&vcir.manifest_rsync_uri)
+        .expect("get cleared vcir")
+        .is_none());
+    assert_eq!(
+        store
+            .get_manifest_replay_meta(&vcir.manifest_rsync_uri)
+            .expect("get preserved replay meta"),
+        Some(replay_meta)
+    );
+}
+
+#[test]
+fn verification_only_materializes_only_contract_selected_fallback_vcirs() {
+    fn reuse_identity() -> VcirFailedFetchReuseIdentity {
+        VcirFailedFetchReuseIdentity {
+            current_ca_sha256: [0x11; 32],
+            ta_context_digest: [0x22; 32],
+            ca_validation_context_digest: [0x33; 32],
+            policy_fingerprint: [0x44; 32],
+            effective_not_before: pack_time(0),
+            effective_until: pack_time(12),
+        }
+    }
+
+    let td = tempfile::tempdir().expect("tempdir");
+    let store = RocksStore::open(td.path()).expect("open rocksdb");
+    let selected = sample_vcir("rsync://example.test/repo/selected.mft");
+    let unselected = sample_vcir("rsync://example.test/repo/unselected.mft");
+    store
+        .put_vcir_with_failed_fetch_reuse_identity(&selected, &reuse_identity())
+        .expect("put selected fallback vcir");
+    store
+        .put_vcir_with_failed_fetch_reuse_identity(&unselected, &reuse_identity())
+        .expect("put unselected fallback vcir");
+    let selection = crate::verification_only::fallback_replay_selection_for_manifest(
+        &store,
+        &selected.manifest_rsync_uri,
+    )
+    .expect("select fallback vcir");
+    let normal_audits = vec![crate::audit::PublicationPointAudit {
+        manifest_rsync_uri: selected.manifest_rsync_uri.clone(),
+        source: "vcir_current_instance".to_string(),
+        ..crate::audit::PublicationPointAudit::default()
+    }];
+    assert_eq!(
+        crate::verification_only::collect_fallback_replay_selections(&store, &normal_audits)
+            .expect("collect normal fallback selection"),
+        vec![selection.clone()]
+    );
+
+    let mut tampered = selection.clone();
+    tampered.vcir_sha256 = "00".repeat(32);
+    assert!(crate::verification_only::materialize_fallback_replay_selections(
+        &store,
+        &[tampered]
+    )
+    .expect_err("tampered contract hash must fail before cleanup")
+    .contains("hash mismatch"));
+    assert_eq!(store.list_vcirs().expect("VCIRs remain after hash rejection").len(), 2);
+
+    let summary = crate::verification_only::materialize_fallback_replay_selections(
+        &store,
+        &[selection.clone()],
+    )
+    .expect("materialize contract fallback selection");
+    assert_eq!(summary.contract_selection_count, 1);
+    assert_eq!(summary.source_vcir_records_cleared, 2);
+    assert_eq!(summary.source_reuse_identity_records_cleared, 2);
+    assert_eq!(summary.imported_selection_count, 1);
+    assert_eq!(summary.scratch_vcir_count, 1);
+    assert_eq!(
+        crate::verification_only::fallback_replay_selection_for_manifest(
+            &store,
+            &selected.manifest_rsync_uri,
+        )
+        .expect("selected fallback remains"),
+        selection
+    );
+    assert!(store
+        .get_vcir(&unselected.manifest_rsync_uri)
+        .expect("get unselected fallback")
+        .is_none());
+    assert!(store
+        .get_manifest_replay_meta(&unselected.manifest_rsync_uri)
+        .expect("get unselected replay meta")
+        .is_some());
+}
+
 #[test]
 fn transport_prefetch_snapshot_roundtrips() {
     use crate::parallel::transport_prefetch::{
@@ -2394,3 +2501,35 @@ fn repository_blob_verification_detects_missing_external_blob() {
         .expect_err("missing blob must fail");
     assert!(error.to_string().contains("missing.roa"), "{error}");
 }
+
+#[test]
+fn vcir_related_artifact_reject_reason_serde_backward_compatible() {
+    // Legacy cache JSON written before the reject_reason ("e") field existed.
+    let legacy = r#"{"r":"manifest","k":"mft","u":"rsync://example.test/a.mft","h":"0000000000000000000000000000000000000000000000000000000000000000","t":"mft","s":"accepted"}"#;
+    let artifact: VcirRelatedArtifact = serde_json::from_str(legacy).expect("legacy artifact");
+    assert_eq!(artifact.reject_reason, None);
+
+    // None reason is skipped on serialization, keeping cache entries compact.
+    let json_none = serde_json::to_string(&artifact).expect("serialize none");
+    assert!(!json_none.contains("\"e\""));
+
+    // A recorded reason round-trips through the "e" field.
+    let mut with_reason = artifact.clone();
+    with_reason.validation_status = VcirArtifactValidationStatus::Rejected;
+    with_reason.reject_reason = Some("bad object".to_string());
+    let json = serde_json::to_string(&with_reason).expect("serialize");
+    assert!(json.contains("\"e\":\"bad object\""));
+    let decoded: VcirRelatedArtifact = serde_json::from_str(&json).expect("deserialize");
+    assert_eq!(decoded.reject_reason.as_deref(), Some("bad object"));
+
+    // PublicationPointCacheObject conversion carries the reason both ways.
+    let cache_object = PublicationPointCacheObject::from_related_artifact(&with_reason);
+    assert_eq!(cache_object.reject_reason.as_deref(), Some("bad object"));
+    let back = cache_object.to_related_artifact();
+    assert_eq!(back.reject_reason.as_deref(), Some("bad object"));
+
+    // Legacy PublicationPointCacheObject JSON without "e" still deserializes.
+    let legacy_object: PublicationPointCacheObject =
+        serde_json::from_str(legacy).expect("legacy cache object");
+    assert_eq!(legacy_object.reject_reason, None);
+}

src/sync/repo.rs

@@ -672,7 +672,7 @@ fn rsync_sync_into_current_store(
         .map_err(|e| RepoSyncError::Storage(e.to_string()))?;
     if let Some(index) = current_repo_index {
         index
-            .lock()
+            .write()
             .map_err(|_| RepoSyncError::Storage("current repo index lock poisoned".to_string()))?
             .apply_repository_view_entries(&repository_view_entries)
             .map_err(RepoSyncError::Storage)?;

src/sync/rrdp.rs

@@ -1029,7 +1029,7 @@ fn hydrate_current_repo_index_from_rrdp_members(
     }
 
     index
-        .lock()
+        .write()
         .map_err(|_| RrdpSyncError::Storage("current repo index lock poisoned".to_string()))?
         .apply_repository_view_entries(&entries)
         .map_err(RrdpSyncError::Storage)?;
@@ -1184,7 +1184,7 @@ fn apply_delta(
                         &current_hash,
                     );
                     index
-                        .lock()
+                        .write()
                         .map_err(|_| {
                             RrdpSyncError::Storage("current repo index lock poisoned".to_string())
                         })?
@@ -1240,7 +1240,7 @@ fn apply_delta(
                             Some(previous_hash.clone()),
                         );
                         index
-                            .lock()
+                            .write()
                             .map_err(|_| {
                                 RrdpSyncError::Storage(
                                     "current repo index lock poisoned".to_string(),

src/sync/rrdp/snapshot_apply.rs

@@ -306,7 +306,7 @@ pub(super) fn apply_snapshot_from_bufread<R: BufRead>(
         .map_err(|e| RrdpSyncError::Storage(e.to_string()))?;
     if let Some(index) = current_repo_index {
         index
-            .lock()
+            .write()
             .map_err(|_| RrdpSyncError::Storage("current repo index lock poisoned".to_string()))?
             .apply_repository_view_entries(&repository_view_entries)
             .map_err(RrdpSyncError::Storage)?;
@@ -365,7 +365,7 @@ fn flush_snapshot_publish_batch(
         .map_err(|e| RrdpSyncError::Storage(e.to_string()))?;
     if let Some(index) = current_repo_index {
         index
-            .lock()
+            .write()
             .map_err(|_| RrdpSyncError::Storage("current repo index lock poisoned".to_string()))?
             .apply_repository_view_entries(&repository_view_entries)
             .map_err(RrdpSyncError::Storage)?;

src/sync/rrdp/tests.rs

@@ -1025,7 +1025,7 @@ fn sync_from_notification_same_serial_hydrates_current_repo_index() {
     .expect("same serial no-op");
     assert_eq!(applied, 0);
 
-    let index = index.lock().expect("lock index");
+    let index = index.read().expect("read-lock index");
     assert_eq!(index.active_uri_count(), 2);
     assert!(index.get_by_uri(uri_a).is_some());
     assert!(index.get_by_uri(uri_b).is_some());
@@ -1089,7 +1089,7 @@ fn sync_from_notification_delta_hydrates_unchanged_current_repo_entries() {
     .expect("delta sync");
     assert_eq!(applied, 1);
 
-    let index = index.lock().expect("lock index");
+    let index = index.read().expect("read-lock index");
     assert_eq!(index.active_uri_count(), 3);
     assert!(
         index.get_by_uri(uri_a).is_some(),

src/tools/rpki_artifact_metrics.rs

@@ -433,6 +433,8 @@ struct LatestRunMetrics {
     download_bytes: Option<u64>,
     artifact_sizes: BTreeMap<String, u64>,
     state_path_sizes: BTreeMap<String, PathSize>,
+    dead_repo_blacklist_entries: Option<u64>,
+    dead_repo_blacklist_blacklisted: Option<u64>,
 }
 
 #[derive(Clone, Debug, Default, Serialize)]
@@ -1377,6 +1379,10 @@ fn build_latest_metrics(record: &RunRecord, snapshot: &mut MetricsSnapshot) {
         latest.download_bytes = json_u64(summary, &["stageTiming", "download_bytes_total"]);
         latest.artifact_sizes = extract_artifact_sizes(summary.get("artifacts"));
         latest.state_path_sizes = extract_path_sizes(summary.get("pathStats"));
+        latest.dead_repo_blacklist_entries =
+            json_u64(summary, &["deadRepoBlacklist", "entries"]);
+        latest.dead_repo_blacklist_blacklisted =
+            json_u64(summary, &["deadRepoBlacklist", "blacklisted"]);
     }
 
     parse_report(&record.path.join("report.json"), snapshot, &mut latest);
@@ -2419,6 +2425,22 @@ fn render_latest_metrics(writer: &mut PromWriter<'_>, instance: &str, latest: &L
             value as f64,
         );
     }
+    if let Some(value) = latest.dead_repo_blacklist_entries {
+        writer.gauge(
+            "ours_rp_dead_repo_blacklist_entries",
+            "Dead-repo transport blacklist entries (latest run)",
+            &[label("instance", instance), label("state", "total")],
+            value as f64,
+        );
+    }
+    if let Some(value) = latest.dead_repo_blacklist_blacklisted {
+        writer.gauge(
+            "ours_rp_dead_repo_blacklist_entries",
+            "Dead-repo transport blacklist entries (latest run)",
+            &[label("instance", instance), label("state", "blacklisted")],
+            value as f64,
+        );
+    }
     for (stage, value) in &latest.stage_seconds {
         writer.gauge(
             "ours_rp_run_stage_duration_seconds",
@@ -3893,4 +3915,37 @@ mod tests {
         });
         encode_content_info(&ci).expect("encode ccr")
     }
+
+    #[test]
+    fn dead_repo_blacklist_gauges_render_from_latest_summary() {
+        let td = TempDir::new().expect("tempdir");
+        write_success_run(td.path(), "run_0001", 1);
+        // Rewrite the summary with the blacklist section present.
+        let run = td.path().join("runs").join("run_0001");
+        fs::write(
+            run.join("run-summary.json"),
+            r#"{"runSeq":1,"runId":"run_0001","startedAtRfc3339Utc":"2026-07-21T00:00:00Z","finishedAtRfc3339Utc":"2026-07-21T00:00:10Z","wallMs":10000,"status":"success","exitCode":0,"deadRepoBlacklist":{"path":"/state/dead-repo-blacklist.json","entries":5,"blacklisted":3}}"#,
+        )
+        .expect("summary");
+        let snapshot = scan_run_root(td.path(), "test").expect("scan");
+        let latest = snapshot.latest_run.as_ref().expect("latest run");
+        assert_eq!(latest.dead_repo_blacklist_entries, Some(5));
+        assert_eq!(latest.dead_repo_blacklist_blacklisted, Some(3));
+        let metrics = render_metrics(&snapshot);
+        assert!(metrics.contains(
+            r#"ours_rp_dead_repo_blacklist_entries{instance="test",state="total"} 5"#
+        ));
+        assert!(metrics.contains(
+            r#"ours_rp_dead_repo_blacklist_entries{instance="test",state="blacklisted"} 3"#
+        ));
+    }
+
+    #[test]
+    fn dead_repo_blacklist_gauges_absent_without_section() {
+        let td = TempDir::new().expect("tempdir");
+        write_success_run(td.path(), "run_0001", 1);
+        let snapshot = scan_run_root(td.path(), "test").expect("scan");
+        let metrics = render_metrics(&snapshot);
+        assert!(!metrics.contains("ours_rp_dead_repo_blacklist_entries"));
+    }
 }

src/tools/rpki_daemon.rs

@@ -6,6 +6,9 @@ use std::path::{Path, PathBuf};
 use std::process::{Command, Stdio};
 use std::time::Duration;
 
+use crate::parallel::dead_repo_blacklist::DeadRepoBlacklist;
+use crate::parallel::types::RepoTransportMode;
+
 #[derive(Clone, Debug, PartialEq, Eq)]
 struct Args {
     state_root: PathBuf,
@@ -21,6 +24,12 @@ struct Args {
     db_stats_bin: Option<PathBuf>,
     db_stats_exact_every: Option<u64>,
     time_bin: Option<PathBuf>,
+    /// Dead-repo blacklist (#141): enables daemon-side health checks and
+    /// auto-injects the child `--dead-repo-blacklist` flag when absent.
+    dead_repo_blacklist: Option<PathBuf>,
+    dead_repo_blacklist_fail_threshold: Option<u32>,
+    dead_repo_health_check_interval_secs: u64,
+    dead_repo_probe_rsync_bin: PathBuf,
     child_args: Vec<String>,
 }
 
@@ -52,6 +61,23 @@ struct DaemonStatus {
     current_run_seq: Option<u64>,
     current_run_id: Option<String>,
     last_run_id: Option<String>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    dead_repo_blacklist: Option<DeadRepoBlacklistStatus>,
+}
+
+#[derive(Clone, Debug, Serialize, PartialEq, Eq)]
+#[serde(rename_all = "camelCase")]
+struct DeadRepoBlacklistStatus {
+    path: String,
+    entries: usize,
+    blacklisted: usize,
+    last_health_check_at_rfc3339_utc: Option<String>,
+}
+
+/// Mutable daemon-side health check bookkeeping (#141).
+#[derive(Clone, Debug, Default, PartialEq, Eq)]
+struct DeadRepoHealthRuntime {
+    last_health_check_at: Option<time::OffsetDateTime>,
 }
 
 #[derive(Clone, Debug, Serialize, PartialEq, Eq)]
@@ -145,6 +171,16 @@ struct RunSummary {
     db_stats: Vec<DbStatsSummary>,
     retention_deleted_runs: Vec<String>,
     artifacts: Vec<ArtifactInfo>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    dead_repo_blacklist: Option<DeadRepoBlacklistRunSummary>,
+}
+
+#[derive(Clone, Debug, Serialize, PartialEq, Eq)]
+#[serde(rename_all = "camelCase")]
+struct DeadRepoBlacklistRunSummary {
+    path: String,
+    entries: usize,
+    blacklisted: usize,
 }
 
 fn usage() -> String {
@@ -170,6 +206,17 @@ Options:
                             Run db_stats --exact every n runs (default: disabled)
   --time-bin <path>         GNU time binary for child process metrics (default: /usr/bin/time when present)
   --no-time-wrapper         Disable GNU time wrapper
+  --dead-repo-blacklist <path>
+                            Enable the dead-repo transport blacklist (#141): daemon probes
+                            blacklisted entries between runs and removes revived repos;
+                            auto-injects the child flag when not already present
+  --dead-repo-blacklist-fail-threshold <n>
+                            Admission threshold forwarded to the child when its args do not
+                            already set it (default: child default 3)
+  --dead-repo-health-check-interval-secs <n>
+                            Minimum seconds between health check sweeps (default: 600)
+  --dead-repo-probe-rsync-bin <path>
+                            rsync binary used for health probes (default: rsync)
   --help                    Show this help
 
 Child argument placeholders:
@@ -211,12 +258,16 @@ fn parse_args(argv: &[String]) -> Result<Args, String> {
     let mut db_stats_exact_every = None;
     let mut time_bin = default_time_bin();
     let mut no_time_wrapper = false;
+    let mut dead_repo_blacklist: Option<PathBuf> = None;
+    let mut dead_repo_blacklist_fail_threshold: Option<u32> = None;
+    let mut dead_repo_health_check_interval_secs = 600u64;
+    let mut dead_repo_probe_rsync_bin = PathBuf::from("rsync");
 
     let mut i = 1usize;
     while i < argv.len() {
         match argv[i].as_str() {
             "--" => {
-                let child_args = argv[i + 1..].to_vec();
+                let mut child_args = argv[i + 1..].to_vec();
                 let state_root =
                     state_root.ok_or_else(|| format!("--state-root is required\n\n{}", usage()))?;
                 let work_db = work_db.unwrap_or_else(|| state_root.join("state").join("work-db"));
@@ -225,6 +276,23 @@ fn parse_args(argv: &[String]) -> Result<Args, String> {
                 if no_time_wrapper {
                     time_bin = None;
                 }
+                // Dead-repo blacklist (#141): forward enablement to the child
+                // unless its args already carry the flag explicitly.
+                if let Some(path) = dead_repo_blacklist.as_ref() {
+                    if !child_args.iter().any(|arg| arg == "--dead-repo-blacklist") {
+                        child_args.push("--dead-repo-blacklist".to_string());
+                        child_args.push(path_string(path));
+                    }
+                    if let Some(threshold) = dead_repo_blacklist_fail_threshold {
+                        if !child_args
+                            .iter()
+                            .any(|arg| arg == "--dead-repo-blacklist-fail-threshold")
+                        {
+                            child_args.push("--dead-repo-blacklist-fail-threshold".to_string());
+                            child_args.push(threshold.to_string());
+                        }
+                    }
+                }
                 let args = Args {
                     state_root,
                     rpki_bin: rpki_bin
@@ -240,6 +308,10 @@ fn parse_args(argv: &[String]) -> Result<Args, String> {
                     db_stats_bin,
                     db_stats_exact_every,
                     time_bin,
+                    dead_repo_blacklist,
+                    dead_repo_blacklist_fail_threshold,
+                    dead_repo_health_check_interval_secs,
+                    dead_repo_probe_rsync_bin,
                     child_args,
                 };
                 return validate_args(args);
@@ -315,6 +387,33 @@ fn parse_args(argv: &[String]) -> Result<Args, String> {
             "--no-time-wrapper" => {
                 no_time_wrapper = true;
             }
+            "--dead-repo-blacklist" => {
+                i += 1;
+                dead_repo_blacklist = Some(PathBuf::from(value_at(argv, i, "--dead-repo-blacklist")?));
+            }
+            "--dead-repo-blacklist-fail-threshold" => {
+                i += 1;
+                let parsed = parse_u64(
+                    value_at(argv, i, "--dead-repo-blacklist-fail-threshold")?,
+                    "--dead-repo-blacklist-fail-threshold",
+                )?;
+                if parsed == 0 || parsed > u32::MAX as u64 {
+                    return Err("--dead-repo-blacklist-fail-threshold must be in 1..=u32::MAX".to_string());
+                }
+                dead_repo_blacklist_fail_threshold = Some(parsed as u32);
+            }
+            "--dead-repo-health-check-interval-secs" => {
+                i += 1;
+                dead_repo_health_check_interval_secs = parse_u64(
+                    value_at(argv, i, "--dead-repo-health-check-interval-secs")?,
+                    "--dead-repo-health-check-interval-secs",
+                )?;
+            }
+            "--dead-repo-probe-rsync-bin" => {
+                i += 1;
+                dead_repo_probe_rsync_bin =
+                    PathBuf::from(value_at(argv, i, "--dead-repo-probe-rsync-bin")?);
+            }
             other => return Err(format!("unknown argument: {other}\n\n{}", usage())),
         }
         i += 1;
@@ -330,6 +429,11 @@ fn validate_args(args: Args) -> Result<Args, String> {
             usage()
         ));
     }
+    if args.dead_repo_blacklist_fail_threshold.is_some() && args.dead_repo_blacklist.is_none() {
+        return Err(
+            "--dead-repo-blacklist-fail-threshold requires --dead-repo-blacklist".to_string(),
+        );
+    }
     Ok(args)
 }
 
@@ -387,10 +491,19 @@ fn write_json_pretty<T: Serialize>(path: &Path, value: &T) -> Result<(), String>
         fs::create_dir_all(parent)
             .map_err(|e| format!("create parent dir failed: {}: {e}", parent.display()))?;
     }
-    let file =
+    let bytes = serde_json::to_vec_pretty(value)
-        File::create(path).map_err(|e| format!("create json failed: {}: {e}", path.display()))?;
+        .map_err(|e| format!("serialize json failed: {}: {e}", path.display()))?;
-    serde_json::to_writer_pretty(file, value)
+    // Atomic write (tmp + rename) so concurrent readers never see a torn file.
-        .map_err(|e| format!("write json failed: {}: {e}", path.display()))
+    let mut tmp_name = path
+        .file_name()
+        .map(|name| name.to_os_string())
+        .unwrap_or_default();
+    tmp_name.push(".tmp");
+    let tmp_path = path.with_file_name(tmp_name);
+    fs::write(&tmp_path, &bytes)
+        .map_err(|e| format!("write json tmp failed: {}: {e}", tmp_path.display()))?;
+    fs::rename(&tmp_path, path)
+        .map_err(|e| format!("rename json tmp failed: {} -> {}: {e}", tmp_path.display(), path.display()))
 }
 
 fn append_json_line<T: Serialize>(path: &Path, value: &T) -> Result<(), String> {
@@ -409,12 +522,30 @@ fn append_json_line<T: Serialize>(path: &Path, value: &T) -> Result<(), String>
         .map_err(|e| format!("flush jsonl failed: {}: {e}", path.display()))
 }
 
+fn dead_repo_blacklist_status(
+    args: &Args,
+    health: &DeadRepoHealthRuntime,
+) -> Option<DeadRepoBlacklistStatus> {
+    let path = args.dead_repo_blacklist.as_ref()?;
+    let (blacklist, _) = DeadRepoBlacklist::load(path);
+    let last_health_check_at_rfc3339_utc = health
+        .last_health_check_at
+        .and_then(|t| format_rfc3339(t).ok());
+    Some(DeadRepoBlacklistStatus {
+        path: path_string(path),
+        entries: blacklist.len(),
+        blacklisted: blacklist.blacklisted_len(),
+        last_health_check_at_rfc3339_utc,
+    })
+}
+
 fn write_status(
     args: &Args,
     state: DaemonState,
     runs_completed: u64,
     current: Option<&RunContext>,
     last_run_id: Option<String>,
+    health: &DeadRepoHealthRuntime,
 ) -> Result<(), String> {
     let updated_at_rfc3339_utc = format_rfc3339(utc_now())?;
     let status = DaemonStatus {
@@ -425,10 +556,131 @@ fn write_status(
         current_run_seq: current.map(|ctx| ctx.seq),
         current_run_id: current.map(|ctx| ctx.run_id.clone()),
         last_run_id,
+        dead_repo_blacklist: dead_repo_blacklist_status(args, health),
     };
     write_json_pretty(&status_path(args), &status)
 }
 
+/// Short-timeout HTTP GET of the RRDP notification file. Any HTTP status
+/// (even 4xx) proves the transport is alive; only transport errors keep the
+/// entry blacklisted.
+fn probe_rrdp_transport(uri: &str) -> bool {
+    let config = crate::fetch::http::HttpFetcherConfig {
+        connect_timeout: Duration::from_secs(3),
+        timeout: Duration::from_secs(6),
+        large_body_timeout: Duration::from_secs(6),
+        ..Default::default()
+    };
+    let fetcher = match crate::fetch::http::BlockingHttpFetcher::new(config) {
+        Ok(fetcher) => fetcher,
+        Err(_) => return false,
+    };
+    match fetcher.fetch_bytes(uri) {
+        Ok(_) => true,
+        Err(err) => err.starts_with("http status"),
+    }
+}
+
+/// rsync list-only probe: connecting to the daemon and listing the module
+/// root is enough to prove the transport is alive.
+fn probe_rsync_transport(rsync_bin: &Path, base_uri: &str) -> bool {
+    Command::new(rsync_bin)
+        .arg("--contimeout=5")
+        .arg("--timeout=8")
+        .arg(base_uri)
+        .stdout(Stdio::null())
+        .stderr(Stdio::null())
+        .status()
+        .map(|status| status.success())
+        .unwrap_or(false)
+}
+
+/// Probe due blacklist entries between runs (#141). Runs in the supervisor
+/// loop while no child is active, preserving the single-writer assumption.
+fn maybe_run_dead_repo_health_check(args: &Args, health: &mut DeadRepoHealthRuntime) {
+    let Some(path) = args.dead_repo_blacklist.clone() else {
+        return;
+    };
+    let now = utc_now();
+    if let Some(last) = health.last_health_check_at {
+        let elapsed = (now - last).whole_seconds();
+        if elapsed >= 0 && (elapsed as u64) < args.dead_repo_health_check_interval_secs {
+            return;
+        }
+    }
+    let (mut blacklist, warning) = DeadRepoBlacklist::load(&path);
+    if let Some(warning) = warning {
+        eprintln!("[dead-repo-health] {warning}");
+    }
+    let now_unix = now.unix_timestamp().max(0) as u64;
+    let due = blacklist.probe_due_entries(now_unix, args.dead_repo_health_check_interval_secs);
+    if due.is_empty() {
+        health.last_health_check_at = Some(now);
+        return;
+    }
+    // Probe in parallel: in the docker soak the daemon runs once per run
+    // (--max-runs 1), so a serial sweep of ~12 dead entries would add up to a
+    // minute of latency to every run.
+    let rsync_bin = args.dead_repo_probe_rsync_bin.clone();
+    let probe_results: Vec<(RepoTransportMode, String, bool)> = due
+        .iter()
+        .map(|entry| (entry.transport, entry.uri.clone()))
+        .collect::<Vec<_>>()
+        .into_iter()
+        .map(|(transport, uri)| {
+            let rsync_bin = rsync_bin.clone();
+            std::thread::spawn(move || {
+                let alive = match transport {
+                    RepoTransportMode::Rrdp => probe_rrdp_transport(&uri),
+                    RepoTransportMode::Rsync => probe_rsync_transport(&rsync_bin, &uri),
+                };
+                (transport, uri, alive)
+            })
+        })
+        .filter_map(|handle| handle.join().ok())
+        .collect();
+    let mut removed = 0usize;
+    let mut still_dead = 0usize;
+    for (transport, uri, alive) in &probe_results {
+        if *alive {
+            blacklist.record_probe_success(*transport, uri);
+            removed += 1;
+            eprintln!("[dead-repo-health] removed {} {} (probe ok)", transport.as_str(), uri);
+            crate::progress_log::emit(
+                "dead_repo_blacklist_remove",
+                serde_json::json!({
+                    "transport": transport.as_str(),
+                    "uri": uri,
+                    "reason": "health_probe_ok",
+                }),
+            );
+        } else {
+            blacklist.record_probe_failure(*transport, uri, now_unix);
+            still_dead += 1;
+            eprintln!(
+                "[dead-repo-health] still dead {} {} (probe failed)",
+                transport.as_str(),
+                uri
+            );
+        }
+    }
+    if let Err(err) = blacklist.store_atomic(&path, now_unix) {
+        eprintln!(
+            "[dead-repo-health] persist blacklist failed: {}: {err}",
+            path.display()
+        );
+    }
+    crate::progress_log::emit(
+        "dead_repo_health_check",
+        serde_json::json!({
+            "probed": probe_results.len(),
+            "removed": removed,
+            "still_dead": still_dead,
+        }),
+    );
+    health.last_health_check_at = Some(now);
+}
+
 fn render_child_args(args: &[String], daemon_args: &Args, ctx: &RunContext) -> Vec<String> {
     args.iter()
         .map(|arg| {
@@ -634,6 +886,7 @@ fn run_child_once(args: &Args, ctx: &RunContext) -> Result<RunSummary, String> {
         db_stats: Vec::new(),
         retention_deleted_runs: Vec::new(),
         artifacts,
+        dead_repo_blacklist: None,
     };
     Ok(summary)
 }
@@ -1031,6 +1284,14 @@ fn collect_post_run_metrics(args: &Args, ctx: &RunContext, summary: &mut RunSumm
     summary.path_stats = collect_state_path_stats(args, ctx);
     summary.db_stats = collect_db_stats(args, ctx);
     summary.artifacts = collect_artifacts(&ctx.run_dir).unwrap_or_default();
+    if let Some(path) = args.dead_repo_blacklist.as_ref() {
+        let (blacklist, _) = DeadRepoBlacklist::load(path);
+        summary.dead_repo_blacklist = Some(DeadRepoBlacklistRunSummary {
+            path: path_string(path),
+            entries: blacklist.len(),
+            blacklisted: blacklist.blacklisted_len(),
+        });
+    }
 }
 
 fn run_daemon(args: &Args) -> Result<(), String> {
@@ -1050,13 +1311,17 @@ fn run_daemon(args: &Args) -> Result<(), String> {
     let mut runs_completed = 0u64;
     let mut next_seq = 1u64;
     let mut last_run_id = None;
+    let mut health = DeadRepoHealthRuntime::default();
     write_status(
         args,
         DaemonState::Starting,
         runs_completed,
         None,
         last_run_id.clone(),
+        &health,
     )?;
+    // Probe once at startup so revived repos are removed before the first run.
+    maybe_run_dead_repo_health_check(args, &mut health);
 
     loop {
         if args.max_runs.is_some_and(|max| runs_completed >= max) {
@@ -1069,6 +1334,7 @@ fn run_daemon(args: &Args) -> Result<(), String> {
             runs_completed,
             None,
             last_run_id.clone(),
+            &health,
         )?;
         let ctx = make_run_context(args, next_seq, utc_now());
         write_status(
@@ -1077,6 +1343,7 @@ fn run_daemon(args: &Args) -> Result<(), String> {
             runs_completed,
             Some(&ctx),
             last_run_id.clone(),
+            &health,
         )?;
         let mut summary = run_child_once(args, &ctx)?;
         write_status(
@@ -1085,6 +1352,7 @@ fn run_daemon(args: &Args) -> Result<(), String> {
             runs_completed,
             Some(&ctx),
             last_run_id.clone(),
+            &health,
         )?;
         collect_post_run_metrics(args, &ctx, &mut summary);
         let removed = apply_retention(&args.state_root.join("runs"), args.retain_runs)?;
@@ -1099,19 +1367,29 @@ fn run_daemon(args: &Args) -> Result<(), String> {
         if args.max_runs.is_some_and(|max| runs_completed >= max) {
             break;
         }
+        // Health probes run only while no child is active (single-writer).
+        maybe_run_dead_repo_health_check(args, &mut health);
         write_status(
             args,
             DaemonState::Sleeping,
             runs_completed,
             None,
             last_run_id.clone(),
+            &health,
         )?;
         if args.interval_secs > 0 {
             std::thread::sleep(Duration::from_secs(args.interval_secs));
         }
     }
 
-    write_status(args, DaemonState::Exited, runs_completed, None, last_run_id)
+    write_status(
+        args,
+        DaemonState::Exited,
+        runs_completed,
+        None,
+        last_run_id,
+        &health,
+    )
 }
 
 pub fn main_entry() -> i32 {
@@ -1154,6 +1432,10 @@ mod tests {
             db_stats_bin: None,
             db_stats_exact_every: None,
             time_bin: None,
+            dead_repo_blacklist: None,
+            dead_repo_blacklist_fail_threshold: None,
+            dead_repo_health_check_interval_secs: 600,
+            dead_repo_probe_rsync_bin: PathBuf::from("rsync"),
             child_args: vec!["--version".to_string()],
         }
     }
@@ -1330,6 +1612,10 @@ mod tests {
             db_stats_bin: None,
             db_stats_exact_every: None,
             time_bin: None,
+            dead_repo_blacklist: None,
+            dead_repo_blacklist_fail_threshold: None,
+            dead_repo_health_check_interval_secs: 600,
+            dead_repo_probe_rsync_bin: PathBuf::from("rsync"),
             child_args: vec![
                 "{state_root}/state/work-db".to_string(),
                 "{run_out}/result.ccr".to_string(),
@@ -1652,6 +1938,10 @@ mod tests {
             db_stats_bin: None,
             db_stats_exact_every: None,
             time_bin: None,
+            dead_repo_blacklist: None,
+            dead_repo_blacklist_fail_threshold: None,
+            dead_repo_health_check_interval_secs: 600,
+            dead_repo_probe_rsync_bin: PathBuf::from("rsync"),
             child_args: vec![
                 "-c".to_string(),
                 "echo stdout-{run_seq}; echo stderr-{run_seq} >&2; echo marker > {run_out}/marker.txt".to_string(),
@@ -1743,6 +2033,10 @@ mod tests {
             db_stats_bin: Some(db_stats_bin),
             db_stats_exact_every: Some(1),
             time_bin: None,
+            dead_repo_blacklist: None,
+            dead_repo_blacklist_fail_threshold: None,
+            dead_repo_health_check_interval_secs: 600,
+            dead_repo_probe_rsync_bin: PathBuf::from("rsync"),
             child_args: vec![
                 "-c".to_string(),
                 "mkdir -p {state_root}/state/work-db {state_root}/state/repo-bytes.db; \
@@ -1781,4 +2075,130 @@ mod tests {
                 .any(|item| item["label"] == "work_db" && item["exists"] == true)
         );
     }
+
+    #[test]
+    fn dead_repo_blacklist_flag_is_injected_into_child_args() {
+        let argv = vec![
+            "rpki_daemon".to_string(),
+            "--state-root".to_string(),
+            "/tmp/daemon".to_string(),
+            "--rpki-bin".to_string(),
+            "/bin/echo".to_string(),
+            "--dead-repo-blacklist".to_string(),
+            "/tmp/daemon/state/dead-repo-blacklist.json".to_string(),
+            "--dead-repo-blacklist-fail-threshold".to_string(),
+            "2".to_string(),
+            "--".to_string(),
+            "--db".to_string(),
+            "{state_root}/state/work-db".to_string(),
+        ];
+        let args = parse_args(&argv).expect("parse");
+        assert_eq!(
+            args.child_args,
+            vec![
+                "--db".to_string(),
+                "{state_root}/state/work-db".to_string(),
+                "--dead-repo-blacklist".to_string(),
+                "/tmp/daemon/state/dead-repo-blacklist.json".to_string(),
+                "--dead-repo-blacklist-fail-threshold".to_string(),
+                "2".to_string(),
+            ]
+        );
+    }
+
+    #[test]
+    fn dead_repo_blacklist_injection_respects_existing_child_flags() {
+        let argv = vec![
+            "rpki_daemon".to_string(),
+            "--state-root".to_string(),
+            "/tmp/daemon".to_string(),
+            "--rpki-bin".to_string(),
+            "/bin/echo".to_string(),
+            "--dead-repo-blacklist".to_string(),
+            "/tmp/daemon/state/dead-repo-blacklist.json".to_string(),
+            "--".to_string(),
+            "--dead-repo-blacklist".to_string(),
+            "/custom/path.json".to_string(),
+        ];
+        let args = parse_args(&argv).expect("parse");
+        assert_eq!(
+            args.child_args,
+            vec![
+                "--dead-repo-blacklist".to_string(),
+                "/custom/path.json".to_string(),
+            ]
+        );
+    }
+
+    #[test]
+    fn dead_repo_threshold_without_blacklist_is_rejected() {
+        let argv = vec![
+            "rpki_daemon".to_string(),
+            "--state-root".to_string(),
+            "/tmp/daemon".to_string(),
+            "--rpki-bin".to_string(),
+            "/bin/echo".to_string(),
+            "--dead-repo-blacklist-fail-threshold".to_string(),
+            "2".to_string(),
+            "--".to_string(),
+            "--version".to_string(),
+        ];
+        let err = parse_args(&argv).expect_err("parse should fail");
+        assert!(err.contains("requires --dead-repo-blacklist"), "{err}");
+    }
+
+    #[test]
+    fn health_check_keeps_dead_entry_and_removes_revived_entry() {
+        use std::io::{Read as _, Write as _};
+        use std::net::TcpListener;
+
+        // Revived RRDP host: one-shot HTTP server answering 200.
+        let listener = TcpListener::bind("127.0.0.1:0").expect("bind");
+        let port = listener.local_addr().expect("local addr").port();
+        std::thread::spawn(move || {
+            if let Ok((mut stream, _)) = listener.accept() {
+                let mut buf = [0u8; 1024];
+                let _ = stream.read(&mut buf);
+                let body = b"<notification/>";
+                let _ = stream.write_all(
+                    format!(
+                        "HTTP/1.1 200 OK\r\nContent-Length: {}\r\nConnection: close\r\n\r\n",
+                        body.len()
+                    )
+                    .as_bytes(),
+                );
+                let _ = stream.write_all(body);
+            }
+        });
+
+        let td = tempfile::tempdir().expect("tempdir");
+        let blacklist_path = td.path().join("dead-repo-blacklist.json");
+        let revived_uri = format!("http://127.0.0.1:{port}/notification.xml");
+        let dead_uri = "http://127.0.0.1:1/notification.xml".to_string();
+        let mut blacklist = DeadRepoBlacklist::new();
+        for uri in [&revived_uri, &dead_uri] {
+            blacklist.record_transport_failure(RepoTransportMode::Rrdp, uri, 100, 1, 256);
+        }
+        blacklist
+            .store_atomic(&blacklist_path, 100)
+            .expect("store blacklist");
+
+        let mut args = test_args(td.path().join("daemon"));
+        args.dead_repo_blacklist = Some(blacklist_path.clone());
+        args.dead_repo_health_check_interval_secs = 0;
+        let mut health = DeadRepoHealthRuntime::default();
+        maybe_run_dead_repo_health_check(&args, &mut health);
+
+        let (after, warning) = DeadRepoBlacklist::load(&blacklist_path);
+        assert!(warning.is_none());
+        assert!(!after.is_blacklisted(RepoTransportMode::Rrdp, &revived_uri));
+        assert!(after.is_blacklisted(RepoTransportMode::Rrdp, &dead_uri));
+        assert!(health.last_health_check_at.is_some());
+
+        // Status JSON exposes the blacklist section.
+        let status = dead_repo_blacklist_status(&args, &health).expect("status");
+        assert_eq!(status.entries, 1);
+        assert_eq!(status.blacklisted, 1);
+        assert!(status.last_health_check_at_rfc3339_utc.is_some());
+    }
 }

src/validation/ca_path.rs

@@ -485,9 +485,17 @@ fn verify_child_signature(
     child: &X509Certificate<'_>,
     issuer_spki: &SubjectPublicKeyInfo<'_>,
 ) -> Result<(), CaPathError> {
+    crate::crypto_sig_cache::verify_with_cache(
+        crate::crypto_sig_cache::CryptoSigVerifyPoint::ChildCaCert,
+        child.tbs_certificate.as_ref(),
+        child.signature_value.data.as_ref(),
+        issuer_spki.raw,
+        || {
             child
                 .verify_signature(Some(issuer_spki))
                 .map_err(|e| CaPathError::ChildSignatureInvalid(e.to_string()))
+        },
+    )
 }
 
 fn validate_child_aki_matches_issuer_ski(
@@ -2118,6 +2126,53 @@ mod tests {
         );
     }
 
+    #[test]
+    fn signature_cache_stores_real_verify_success_and_skips_on_hit() {
+        let _guard = crate::crypto_sig_cache::GLOBAL_TEST_LOCK.lock().expect("test lock");
+        crate::crypto_sig_cache::clear_global();
+        let td = tempfile::tempdir().expect("tempdir");
+        let (issuer, child, other) = gen_issuer_and_child_der(td.path());
+        let issuer_cert = parse_x509_cert(&issuer).expect("x509 parse issuer");
+        let child_cert = parse_x509_cert(&child).expect("x509 parse child");
+        let other_cert = parse_x509_cert(&other).expect("x509 parse other");
+
+        let cache_dir = tempfile::tempdir().expect("tempdir");
+        let cache = std::sync::Arc::new(crate::crypto_sig_cache::CryptoSigCache::load_or_rebuild(
+            cache_dir.path().join("work-db.crypto-sig-cache"),
+            true,
+        ));
+        crate::crypto_sig_cache::install_global(std::sync::Arc::clone(&cache));
+
+        // Miss -> real verification runs and passes -> positive conclusion written.
+        verify_child_signature(&child_cert, &issuer_cert.tbs_certificate.subject_pki)
+            .expect("signature ok");
+        // Hit -> skipped (a second execution would also pass, so assert via stats).
+        verify_child_signature(&child_cert, &issuer_cert.tbs_certificate.subject_pki)
+            .expect("signature ok");
+        // Wrong issuer: real verification fails, nothing is written (no negative caching),
+        // so a retry executes the real verification again instead of inheriting anything.
+        for _ in 0..2 {
+            let err = verify_child_signature(&child_cert, &other_cert.tbs_certificate.subject_pki)
+                .unwrap_err();
+            assert!(
+                matches!(err, CaPathError::ChildSignatureInvalid(_)),
+                "{err}"
+            );
+        }
+
+        let summary = cache.summary();
+        let stats = summary
+            .per_point
+            .get("child_ca_cert")
+            .expect("child_ca_cert stats");
+        assert_eq!(stats.calls, 4);
+        assert_eq!(stats.would_hit, 1);
+        assert_eq!(stats.new_keys, 3);
+        assert_eq!(stats.verify_executed, 3);
+        assert_eq!(stats.verify_skipped, 1);
+        crate::crypto_sig_cache::clear_global();
+    }
+
     #[test]
     fn issuer_effective_resources_index_and_indexed_resolvers_cover_success_and_failure_paths() {
         use crate::data_model::rc::{AsIdOrRange, IpPrefix};

src/validation/cert_path.rs

@@ -385,8 +385,16 @@ fn verify_ee_cert_signature(
     ee: &X509Certificate<'_>,
     issuer_spki: &SubjectPublicKeyInfo<'_>,
 ) -> Result<(), CertPathError> {
+    crate::crypto_sig_cache::verify_with_cache(
+        crate::crypto_sig_cache::CryptoSigVerifyPoint::EeCertX509,
+        ee.tbs_certificate.as_ref(),
+        ee.signature_value.data.as_ref(),
+        issuer_spki.raw,
+        || {
             ee.verify_signature(Some(issuer_spki))
                 .map_err(|e| CertPathError::EeSignatureInvalid(e.to_string()))
+        },
+    )
 }
 
 fn verify_ee_cert_signature_fast(
@@ -394,12 +402,20 @@ fn verify_ee_cert_signature_fast(
     ee_signature_bytes: &[u8],
     issuer_spki: &SubjectPublicKeyInfo<'_>,
 ) -> Result<(), CertPathError> {
+    crate::crypto_sig_cache::verify_with_cache(
+        crate::crypto_sig_cache::CryptoSigVerifyPoint::EeCertFast,
+        ee_tbs_der,
+        ee_signature_bytes,
+        issuer_spki.raw,
+        || {
             let key = signature::UnparsedPublicKey::new(
                 &signature::RSA_PKCS1_2048_8192_SHA256,
                 &issuer_spki.subject_public_key.data,
             );
             key.verify(ee_tbs_der, ee_signature_bytes)
                 .map_err(|e| CertPathError::EeSignatureInvalid(e.to_string()))
+        },
+    )
 }
 
 fn validate_ee_aki_matches_issuer_ski(
@@ -720,4 +736,145 @@ mod tests {
         validate_ee_crldp_contains_issuer_crl_uri(&ee, "rsync://example.test/issuer.crl")
             .expect("crldp ok");
     }
+
+    #[test]
+    fn signature_cache_hit_does_not_skip_crl_revocation_check() {
+        use crate::crypto_sig_cache::{
+            CryptoSigCache, CryptoSigVerifyPoint, compute_cache_key,
+        };
+        use crate::data_model::common::{Asn1TimeEncoding, Asn1TimeUtc};
+        use crate::data_model::crl::CrlExtensions;
+
+        let _guard = crate::crypto_sig_cache::GLOBAL_TEST_LOCK.lock().expect("test lock");
+        crate::crypto_sig_cache::clear_global();
+
+        let now = time::OffsetDateTime::now_utc();
+        let mut issuer = dummy_cert(
+            ResourceCertKind::Ca,
+            "CN=issuer",
+            "CN=issuer",
+            Some(vec![1]),
+            None,
+            None,
+            None,
+        );
+        issuer.tbs.validity_not_before = now - time::Duration::days(1);
+        issuer.tbs.validity_not_after = now + time::Duration::days(1);
+        let mut ee_rc = dummy_cert(
+            ResourceCertKind::Ee,
+            "CN=ee",
+            "CN=issuer",
+            Some(vec![2]),
+            Some(vec![1]),
+            Some(vec!["rsync://example.test/issuer.cer"]),
+            Some(vec!["rsync://example.test/issuer.crl"]),
+        );
+        ee_rc.tbs.validity_not_before = now - time::Duration::days(1);
+        ee_rc.tbs.validity_not_after = now + time::Duration::days(1);
+
+        // Tampered "object": these bytes could never pass the real RSA verification, so an
+        // Ok outcome below proves the crypto step was skipped by the cache hit.
+        let ee = ResourceEeCertificate {
+            raw_der: Vec::new(),
+            subject_key_identifier: vec![2],
+            spki_der: Vec::new(),
+            rsa_public_modulus: Vec::new(),
+            rsa_public_exponent: Vec::new(),
+            tbs_certificate_der: b"tampered-tbs-bytes".to_vec(),
+            signature_bytes: b"tampered-signature-bytes".to_vec(),
+            key_usage_summary: EeKeyUsageSummary::DigitalSignatureOnly,
+            sia_signed_object_uris: Vec::new(),
+            resource_cert: ee_rc,
+        };
+
+        // Any real RSA SPKI works as the verifier identity (the TA fixture is a valid RSA
+        // certificate); the SPKI DER only feeds the cache key here.
+        let ta_der = std::fs::read(concat!(
+            env!("CARGO_MANIFEST_DIR"),
+            "/tests/fixtures/ta/apnic-ta.cer"
+        ))
+        .expect("read ta fixture");
+        let (_, ta_x509) = X509Certificate::from_der(&ta_der).expect("parse ta fixture");
+        let issuer_spki = &ta_x509.tbs_certificate.subject_pki;
+
+        let crl = RpkixCrl {
+            raw_der: Vec::new(),
+            version: 2,
+            issuer_dn: "CN=issuer".to_string(),
+            signature_algorithm_oid: "1.2.840.113549.1.1.11".to_string(),
+            this_update: Asn1TimeUtc {
+                utc: now - time::Duration::hours(1),
+                encoding: Asn1TimeEncoding::UtcTime,
+            },
+            next_update: Asn1TimeUtc {
+                utc: now + time::Duration::hours(1),
+                encoding: Asn1TimeEncoding::UtcTime,
+            },
+            revoked_certs: Vec::new(),
+            extensions: CrlExtensions {
+                authority_key_identifier: vec![1],
+                crl_number: BigUnsigned { bytes_be: vec![1] },
+            },
+        };
+
+        let dir = tempfile::tempdir().expect("tempdir");
+        let cache = std::sync::Arc::new(CryptoSigCache::load_or_rebuild(
+            dir.path().join("work-db.crypto-sig-cache"),
+            true,
+        ));
+        cache.insert_positive(compute_cache_key(
+            CryptoSigVerifyPoint::EeCertFast,
+            b"tampered-tbs-bytes",
+            b"tampered-signature-bytes",
+            issuer_spki.raw,
+        ));
+        crate::crypto_sig_cache::install_global(std::sync::Arc::clone(&cache));
+
+        // EE serial is 1 (dummy_cert); revoke it. The revocation check runs before the
+        // signature step in this code path, so the result must be EeRevoked — never Ok —
+        // even though a positive signature conclusion is cached.
+        let revoked_serials: HashSet<Vec<u8>> = HashSet::from([vec![1u8]]);
+        let err = validate_signed_object_ee_cert_path_fast(
+            &ee,
+            &issuer,
+            issuer_spki,
+            &crl,
+            &revoked_serials,
+            Some("rsync://example.test/issuer.cer"),
+            Some("rsync://example.test/issuer.crl"),
+            now,
+        )
+        .unwrap_err();
+        assert!(
+            matches!(err, CertPathError::EeRevoked),
+            "cached signature conclusion must not bypass revocation: {err}"
+        );
+
+        // Not revoked: the cached positive conclusion lets validation pass even though the
+        // tampered bytes could never verify — proving the crypto step was skipped.
+        let not_revoked: HashSet<Vec<u8>> = HashSet::new();
+        validate_signed_object_ee_cert_path_fast(
+            &ee,
+            &issuer,
+            issuer_spki,
+            &crl,
+            &not_revoked,
+            Some("rsync://example.test/issuer.cer"),
+            Some("rsync://example.test/issuer.crl"),
+            now,
+        )
+        .expect("cache hit skips the impossible verification");
+
+        let summary = cache.summary();
+        let stats = summary
+            .per_point
+            .get("ee_cert_fast")
+            .expect("ee_cert_fast stats");
+        assert_eq!(stats.verify_skipped, 1);
+        assert_eq!(
+            stats.verify_executed, 0,
+            "real verification must never run on cache hits"
+        );
+        crate::crypto_sig_cache::clear_global();
+    }
 }

src/validation/manifest.rs

@@ -135,6 +135,11 @@ impl PublicationPointData for PublicationPointSnapshot {
 
 #[derive(Debug, thiserror::Error)]
 pub enum ManifestFreshError {
+    #[error(
+        "verification-only fallback replay selected current-instance VCIR: {manifest_rsync_uri}"
+    )]
+    VerificationOnlyForcedCurrentInstanceVcir { manifest_rsync_uri: String },
+
     #[error("repo sync failed: {detail} (RFC 8182 §3.4.5; RFC 9286 §6.6)")]
     RepoSyncFailed { detail: String },
 
@@ -222,7 +227,8 @@ impl ManifestFreshError {
     pub(crate) fn should_warn_when_current_instance_reused(&self) -> bool {
         !matches!(
             self,
-            ManifestFreshError::RepoSyncFailed { .. }
+            ManifestFreshError::VerificationOnlyForcedCurrentInstanceVcir { .. }
+                | ManifestFreshError::RepoSyncFailed { .. }
                 | ManifestFreshError::MissingManifest { .. }
                 | ManifestFreshError::MissingFile { .. }
         )
@@ -633,7 +639,7 @@ pub(crate) fn try_build_fresh_publication_point_with_timing(
 > {
     let mut timing = FreshPublicationPointTimingBreakdown::default();
     let current_index_lock_started = std::time::Instant::now();
-    let current_index_guard = current_repo_index.and_then(|handle| handle.lock().ok());
+    let current_index_guard = current_repo_index.and_then(|handle| handle.read().ok());
     timing.current_index_lock_ms = current_index_lock_started.elapsed().as_millis() as u64;
 
     if !rsync_uri_is_under_publication_point(manifest_rsync_uri, publication_point_rsync_uri) {
@@ -1199,6 +1205,7 @@ mod tests {
                     sha256: manifest_sha256.to_string(),
                     object_type: Some("mft".to_string()),
                     validation_status: VcirArtifactValidationStatus::Accepted,
+                    reject_reason: None,
                 },
                 VcirRelatedArtifact {
                     artifact_role: VcirArtifactRole::SignedObject,
@@ -1207,6 +1214,7 @@ mod tests {
                     sha256: locked_object_sha256.to_string(),
                     object_type: Some("roa".to_string()),
                     validation_status: VcirArtifactValidationStatus::Accepted,
+                    reject_reason: None,
                 },
             ],
             summary: VcirSummary {
@@ -1437,8 +1445,8 @@ mod tests {
         }
 
         current_index
-            .lock()
+            .write()
-            .expect("index lock")
+            .expect("index write lock")
             .apply_repository_view_entries(&entries)
             .expect("apply current index");
 

src/validation/objects.rs

@@ -4023,6 +4023,7 @@ mod tests {
                     sha256: sha256_hex(issuer_der),
                     object_type: Some("cer".to_string()),
                     validation_status: VcirArtifactValidationStatus::Accepted,
+                    reject_reason: None,
                 },
                 VcirRelatedArtifact {
                     artifact_role: VcirArtifactRole::CurrentCrl,
@@ -4031,6 +4032,7 @@ mod tests {
                     sha256: sha256_hex_from_32(&crl_hash),
                     object_type: Some("crl".to_string()),
                     validation_status: VcirArtifactValidationStatus::Accepted,
+                    reject_reason: None,
                 },
             ],
             summary: VcirSummary {
@@ -4543,6 +4545,7 @@ mod tests {
                 sha256: "00".repeat(32),
                 object_type: Some("cer".to_string()),
                 validation_status: VcirArtifactValidationStatus::Rejected,
+                reject_reason: None,
             },
         );
         vcir.local_outputs.push(VcirLocalOutput {

src/validation/run_tree_from_tal.rs

@@ -119,7 +119,7 @@ fn snapshot_current_repo_objects(
         return Vec::new();
     }
     current_repo_index
-        .and_then(|handle| handle.lock().ok().map(|idx| idx.snapshot_objects()))
+        .and_then(|handle| handle.read().ok().map(|idx| idx.snapshot_objects()))
         .unwrap_or_default()
 }
 
@@ -328,20 +328,73 @@ fn persist_transport_request_prefetch_snapshot(
     }
     let snapshot = runtime.transport_prefetch_snapshot();
     let recorded = snapshot.requests.len() as u64;
+    let persist_started = std::time::Instant::now();
     store
         .put_transport_prefetch_snapshot(&snapshot)
         .map_err(|err| TreeRunError::Runner(err.to_string()))?;
+    let persist_ms = persist_started.elapsed().as_millis() as u64;
     record_transport_prefetch_count(timing, "transport_prefetch_recorded_requests", recorded);
     crate::progress_log::emit(
         "phase1_repo_prefetch_persisted",
         serde_json::json!({
             "recorded_requests": recorded,
             "schema_version": snapshot.schema_version,
+            "persist_ms": persist_ms,
         }),
     );
     Ok(())
 }
 
+/// Persist the dead-repo blacklist working copy at run end (#141). Failures
+/// degrade to a progress event; a broken blacklist file must never fail a run.
+fn persist_dead_repo_blacklist(
+    runtime: &Arc<dyn RepoSyncRuntime>,
+    timing: Option<&TimingHandle>,
+) -> Result<(), RunTreeFromTalError> {
+    let Some((blacklist_config, blacklist)) = runtime.dead_repo_blacklist_state() else {
+        return Ok(());
+    };
+    let persist_started = std::time::Instant::now();
+    let now_unix = std::time::SystemTime::now()
+        .duration_since(std::time::UNIX_EPOCH)
+        .map(|duration| duration.as_secs())
+        .unwrap_or(0);
+    match blacklist.store_atomic(&blacklist_config.path, now_unix) {
+        Ok(()) => {
+            let persist_ms = persist_started.elapsed().as_millis() as u64;
+            record_transport_prefetch_count(
+                timing,
+                "dead_repo_blacklist_entries",
+                blacklist.len() as u64,
+            );
+            record_transport_prefetch_count(
+                timing,
+                "dead_repo_blacklist_blacklisted",
+                blacklist.blacklisted_len() as u64,
+            );
+            crate::progress_log::emit(
+                "dead_repo_blacklist_persisted",
+                serde_json::json!({
+                    "path": blacklist_config.path.display().to_string(),
+                    "entries": blacklist.len(),
+                    "blacklisted": blacklist.blacklisted_len(),
+                    "persist_ms": persist_ms,
+                }),
+            );
+        }
+        Err(err) => {
+            crate::progress_log::emit(
+                "dead_repo_blacklist_persist_error",
+                serde_json::json!({
+                    "path": blacklist_config.path.display().to_string(),
+                    "error": err.to_string(),
+                }),
+            );
+        }
+    }
+    Ok(())
+}
+
 fn root_discovery_from_tal_input(
     tal_input: &TalInputSpec,
     http_fetcher: &dyn Fetcher,
@@ -810,6 +863,7 @@ where
         run_tree_serial_audit(root, &runner, config)?
     };
     persist_transport_request_prefetch_snapshot(store.as_ref(), &runtime, config, timing.as_ref())?;
+    persist_dead_repo_blacklist(&runtime, timing.as_ref())?;
     let downloads = download_log.snapshot_events();
     let download_stats = DownloadLogHandle::stats_from_events(&downloads);
     Ok(RunTreeFromTalAuditOutput {
@@ -933,6 +987,7 @@ where
         run_tree_serial_audit_multi_root(root_handles, &runner, config)?
     };
     persist_transport_request_prefetch_snapshot(store.as_ref(), &runtime, config, timing.as_ref())?;
+    persist_dead_repo_blacklist(&runtime, timing.as_ref())?;
     let downloads = download_log.snapshot_events();
     let download_stats = DownloadLogHandle::stats_from_events(&downloads);
     Ok(RunTreeFromTalAuditOutput {
@@ -2392,8 +2447,8 @@ mod multi_tal_tests {
     fn snapshot_current_repo_objects_is_on_demand() {
         let handle = CurrentRepoIndex::shared();
         handle
-            .lock()
+            .write()
-            .expect("lock index")
+            .expect("write-lock index")
             .apply_repository_view_entries(&[RepositoryViewEntry {
                 rsync_uri: "rsync://example.test/repo/a.roa".to_string(),
                 current_hash: Some("11".repeat(32)),

src/validation/tree_runner.rs

@@ -511,7 +511,7 @@ impl<'a> Rpkiv1PublicationPointRunner<'a> {
 
     fn current_hash_for_uri(&self, uri: &str) -> Option<[u8; 32]> {
         if let Some(index) = self.current_repo_index.as_ref() {
-            if let Ok(index) = index.lock() {
+            if let Ok(index) = index.read() {
                 if let Some(entry) = index.get_by_uri(uri) {
                     return Some(entry.current_hash);
                 }
@@ -760,7 +760,7 @@ impl<'a> Rpkiv1PublicationPointRunner<'a> {
 
     fn current_manifest_hash_hex_for_audit(&self, ca: &CaInstanceHandle) -> Option<String> {
         if let Some(index_handle) = self.current_repo_index.as_ref()
-            && let Ok(index) = index_handle.lock()
+            && let Ok(index) = index_handle.read()
             && let Some(entry) = index.get_by_uri(&ca.manifest_rsync_uri)
         {
             return Some(entry.current_hash_hex.clone());
@@ -807,6 +807,21 @@ impl<'a> Rpkiv1PublicationPointRunner<'a> {
         repo_sync_ok: bool,
         repo_sync_err: Option<&str>,
     ) -> Result<FreshPublicationPointStage, FreshPublicationPointStageError> {
+        if self
+            .policy
+            .verification_forced_vcir_reuse_manifest_uris
+            .contains(&ca.manifest_rsync_uri)
+        {
+            if let Some(timing) = self.timing.as_ref() {
+                timing.record_count("verification_forced_vcir_reuse_publication_points", 1);
+            }
+            return Err(FreshPublicationPointStageError {
+                error: ManifestFreshError::VerificationOnlyForcedCurrentInstanceVcir {
+                    manifest_rsync_uri: ca.manifest_rsync_uri.clone(),
+                },
+                snapshot_prepare_ms: 0,
+            });
+        }
         let snapshot_prepare_started = std::time::Instant::now();
         let issuer_ca_der = ca_certificate_der_for_validation(ca, self.store, self.timing.as_ref())
             .map_err(|detail| FreshPublicationPointStageError {
@@ -3840,6 +3855,28 @@ fn audit_result_from_vcir_status(status: VcirArtifactValidationStatus) -> AuditO
     }
 }
 
+/// Fallback detail for cached artifacts rejected by an earlier run whose cache
+/// entry predates reject-reason recording (field added 2026-07-27).
+const CACHED_REJECT_REASON_NOT_RECORDED: &str =
+    "rejected in an earlier validation run (reject reason not recorded in cache)";
+
+/// Detail for an audit entry rebuilt from a cached artifact: the real reject
+/// reason when the cache recorded one, an explicit fallback for legacy cache
+/// entries, and `None` for non-rejected artifacts.
+fn audit_detail_from_vcir_status(
+    status: VcirArtifactValidationStatus,
+    reject_reason: Option<&str>,
+) -> Option<String> {
+    match status {
+        VcirArtifactValidationStatus::Rejected => Some(
+            reject_reason
+                .map(str::to_string)
+                .unwrap_or_else(|| CACHED_REJECT_REASON_NOT_RECORDED.to_string()),
+        ),
+        VcirArtifactValidationStatus::Accepted | VcirArtifactValidationStatus::WarningOnly => None,
+    }
+}
+
 fn build_publication_point_audit_from_snapshot(
     ca: &CaInstanceHandle,
     source: PublicationPointSource,
@@ -4059,7 +4096,10 @@ fn build_publication_point_audit_from_vcir(
                 sha256_hex: artifact.sha256.clone(),
                 kind: kind_from_vcir_artifact_kind(artifact.artifact_kind),
                 result: audit_result_from_vcir_status(artifact.validation_status),
-                detail: None,
+                detail: audit_detail_from_vcir_status(
+                    artifact.validation_status,
+                    artifact.reject_reason.as_deref(),
+                ),
             },
         );
     }
@@ -4170,7 +4210,10 @@ fn build_publication_point_audit_from_publication_point_cache_projection(
                 sha256_hex: artifact.sha256.clone(),
                 kind: kind_from_vcir_artifact_kind(artifact.artifact_kind),
                 result: audit_result_from_vcir_status(artifact.validation_status),
-                detail: None,
+                detail: audit_detail_from_vcir_status(
+                    artifact.validation_status,
+                    artifact.reject_reason.as_deref(),
+                ),
             },
         );
     }
@@ -6003,9 +6046,9 @@ fn build_vcir_related_artifacts(
     objects: &crate::validation::objects::ObjectsOutput,
     child_audits: &[ObjectAuditEntry],
 ) -> Vec<VcirRelatedArtifact> {
-    let mut audit_by_uri: HashMap<&str, AuditObjectResult> = HashMap::new();
+    let mut audit_by_uri: HashMap<&str, &ObjectAuditEntry> = HashMap::new();
     for entry in child_audits.iter().chain(objects.audit.iter()) {
-        audit_by_uri.insert(entry.rsync_uri.as_str(), entry.result.clone());
+        audit_by_uri.insert(entry.rsync_uri.as_str(), entry);
     }
 
     let mut artifacts = Vec::with_capacity(pack.files.len() + 2);
@@ -6016,6 +6059,7 @@ fn build_vcir_related_artifacts(
         sha256: sha256_hex(&pack.manifest_bytes),
         object_type: Some("mft".to_string()),
         validation_status: VcirArtifactValidationStatus::Accepted,
+        reject_reason: None,
     });
     artifacts.push(VcirRelatedArtifact {
         artifact_role: if ca.parent_manifest_rsync_uri.is_none() {
@@ -6035,13 +6079,20 @@ fn build_vcir_related_artifacts(
             }),
         object_type: Some("cer".to_string()),
         validation_status: VcirArtifactValidationStatus::Accepted,
+        reject_reason: None,
     });
 
     for file in &pack.files {
-        let result = audit_by_uri
+        let audit_entry = audit_by_uri.get(file.rsync_uri.as_str()).copied();
-            .get(file.rsync_uri.as_str())
+        let result = audit_entry
-            .cloned()
+            .map(|entry| entry.result.clone())
             .unwrap_or(AuditObjectResult::Ok);
+        let validation_status = audit_result_to_vcir_status(&result);
+        let reject_reason = if validation_status == VcirArtifactValidationStatus::Rejected {
+            audit_entry.and_then(|entry| entry.detail.clone())
+        } else {
+            None
+        };
         let (artifact_role, artifact_kind) = artifact_role_and_kind(file, current_crl_rsync_uri);
         artifacts.push(VcirRelatedArtifact {
             artifact_role,
@@ -6049,7 +6100,8 @@ fn build_vcir_related_artifacts(
             uri: Some(file.rsync_uri.clone()),
             sha256: sha256_hex_from_32(&file.sha256),
             object_type: object_type_from_uri(file.rsync_uri.as_str()),
-            validation_status: audit_result_to_vcir_status(&result),
+            validation_status,
+            reject_reason,
         });
     }
 

src/validation/tree_runner/tests.rs

@@ -708,6 +708,7 @@ fn sample_vcir_for_projection(
                 sha256: manifest_hash,
                 object_type: Some("mft".to_string()),
                 validation_status: VcirArtifactValidationStatus::Accepted,
+                reject_reason: None,
             },
             VcirRelatedArtifact {
                 artifact_role: VcirArtifactRole::CurrentCrl,
@@ -716,6 +717,7 @@ fn sample_vcir_for_projection(
                 sha256: current_crl_hash,
                 object_type: Some("crl".to_string()),
                 validation_status: VcirArtifactValidationStatus::Accepted,
+                reject_reason: None,
             },
             VcirRelatedArtifact {
                 artifact_role: VcirArtifactRole::ChildCaCert,
@@ -724,6 +726,7 @@ fn sample_vcir_for_projection(
                 sha256: child_cert_hash.to_string(),
                 object_type: Some("cer".to_string()),
                 validation_status: VcirArtifactValidationStatus::Accepted,
+                reject_reason: None,
             },
             VcirRelatedArtifact {
                 artifact_role: VcirArtifactRole::SignedObject,
@@ -732,6 +735,7 @@ fn sample_vcir_for_projection(
                 sha256: roa_hash,
                 object_type: Some("roa".to_string()),
                 validation_status: VcirArtifactValidationStatus::Accepted,
+                reject_reason: None,
             },
             VcirRelatedArtifact {
                 artifact_role: VcirArtifactRole::SignedObject,
@@ -740,6 +744,7 @@ fn sample_vcir_for_projection(
                 sha256: aspa_hash,
                 object_type: Some("aspa".to_string()),
                 validation_status: VcirArtifactValidationStatus::Accepted,
+                reject_reason: None,
             },
         ],
         summary: VcirSummary {
@@ -1455,10 +1460,12 @@ fn build_vcir_related_artifacts_classifies_snapshot_files_and_audit_statuses() {
         && artifact.artifact_role == VcirArtifactRole::ChildCaCert));
     assert!(artifacts.iter().any(|artifact| artifact.uri.as_deref()
         == Some("rsync://example.test/repo/issuer/a.roa")
-        && artifact.validation_status == VcirArtifactValidationStatus::Rejected));
+        && artifact.validation_status == VcirArtifactValidationStatus::Rejected
+        && artifact.reject_reason.as_deref() == Some("bad roa")));
     assert!(artifacts.iter().any(|artifact| artifact.uri.as_deref()
         == Some("rsync://example.test/repo/issuer/a.asa")
-        && artifact.validation_status == VcirArtifactValidationStatus::WarningOnly));
+        && artifact.validation_status == VcirArtifactValidationStatus::WarningOnly
+        && artifact.reject_reason.is_none()));
     assert!(artifacts.iter().any(|artifact| artifact.uri.as_deref()
         == Some("rsync://example.test/repo/issuer/a.gbr")
         && artifact.artifact_kind == VcirArtifactKind::Gbr));
@@ -3419,6 +3426,45 @@ fn runner_when_repo_sync_fails_uses_current_instance_vcir_and_keeps_children_emp
             .any(|w| w.message.contains("repo sync failed")),
         "expected warning about repo sync failure"
     );
+
+    // Verification-only replays have the frozen repository bytes available,
+    // so a fresh read would succeed and never naturally enter the fallback
+    // branch.  A contract-selected URI must therefore select the same VCIR
+    // projection without attempting fresh manifest/object validation.
+    let mut forced_policy = policy.clone();
+    forced_policy
+        .verification_forced_vcir_reuse_manifest_uris
+        .insert(manifest_rsync_uri.clone());
+    let forced_runner = Rpkiv1PublicationPointRunner {
+        store: &store,
+        policy: &forced_policy,
+        http_fetcher: &NeverHttpFetcher,
+        rsync_fetcher: &LocalDirRsyncFetcher::new(&fixture_dir),
+        validation_time,
+        timing: None,
+        download_log: None,
+        replay_archive_index: None,
+        replay_delta_index: None,
+        rrdp_dedup: false,
+        rrdp_repo_cache: Mutex::new(HashMap::new()),
+        rsync_dedup: false,
+        rsync_repo_cache: Mutex::new(HashMap::new()),
+        current_repo_index: None,
+        repo_sync_runtime: None,
+        parallel_phase2_config: None,
+        parallel_roa_worker_pool: None,
+        ccr_accumulator: None,
+        persist_vcir: true,
+        enable_roa_validation_cache: false,
+        enable_child_certificate_validation_cache: false,
+        publication_point_cache_observe_only: false,
+        enable_publication_point_validation_cache: false,
+    };
+    let forced = forced_runner
+        .run_publication_point(&handle)
+        .expect("contract-selected fallback should reuse current-instance VCIR");
+    assert_eq!(forced.source, PublicationPointSource::VcirCurrentInstance);
+    assert!(forced.discovered_children.is_empty());
 }
 
 #[test]
@@ -4510,6 +4556,7 @@ fn build_objects_output_from_vcir_tracks_expired_and_invalid_cached_outputs() {
                 .to_string(),
             ),
             validation_status: VcirArtifactValidationStatus::Accepted,
+            reject_reason: None,
         });
     }
 
@@ -4732,6 +4779,173 @@ fn build_publication_point_audit_from_vcir_uses_vcir_metadata_and_overlays_child
     }));
 }
 
+#[test]
+fn build_publication_point_audit_from_vcir_restores_reject_reason_with_legacy_fallback() {
+    let now = time::OffsetDateTime::now_utc();
+    let child_cert_hash = sha256_hex(b"child-cert");
+    let mut vcir = sample_vcir_for_projection(now, &child_cert_hash);
+    vcir.related_artifacts.push(VcirRelatedArtifact {
+        artifact_role: VcirArtifactRole::SignedObject,
+        artifact_kind: VcirArtifactKind::Roa,
+        uri: Some("rsync://example.test/repo/issuer/rejected-with-reason.roa".to_string()),
+        sha256: sha256_hex(b"rejected-with-reason"),
+        object_type: Some("roa".to_string()),
+        validation_status: VcirArtifactValidationStatus::Rejected,
+        reject_reason: Some("EE certificate path validation failed: test".to_string()),
+    });
+    vcir.related_artifacts.push(VcirRelatedArtifact {
+        artifact_role: VcirArtifactRole::SignedObject,
+        artifact_kind: VcirArtifactKind::Roa,
+        uri: Some("rsync://example.test/repo/issuer/rejected-legacy.roa".to_string()),
+        sha256: sha256_hex(b"rejected-legacy"),
+        object_type: Some("roa".to_string()),
+        validation_status: VcirArtifactValidationStatus::Rejected,
+        reject_reason: None,
+    });
+
+    let ca = CaInstanceHandle {
+        depth: 0,
+        tal_id: "test-tal".to_string(),
+        parent_manifest_rsync_uri: None,
+        ca_certificate: CaCertificateRef::inline_der(Vec::new()),
+        ca_certificate_rsync_uri: None,
+        effective_ip_resources: None,
+        effective_as_resources: None,
+        rsync_base_uri: "rsync://example.test/repo/issuer/".to_string(),
+        manifest_rsync_uri: vcir.manifest_rsync_uri.clone(),
+        publication_point_rsync_uri: "rsync://example.test/repo/issuer/".to_string(),
+        rrdp_notification_uri: None,
+    };
+    let objects = crate::validation::objects::ObjectsOutput {
+        vrps: Vec::new(),
+        aspas: Vec::new(),
+        router_keys: Vec::new(),
+        local_outputs_cache: Vec::new(),
+        warnings: Vec::new(),
+        stats: crate::validation::objects::ObjectsStats::default(),
+        audit: Vec::new(),
+        roa_cache_stats: crate::validation::objects::RoaValidationCacheStats::default(),
+        roa_cache_object_meta: Vec::new(),
+    };
+
+    let audit = build_publication_point_audit_from_vcir(
+        &ca,
+        PublicationPointSource::VcirCurrentInstance,
+        None,
+        None,
+        None,
+        None,
+        Some(&vcir),
+        None,
+        &[],
+        &objects,
+        &[],
+        &[],
+    );
+
+    assert!(audit.objects.iter().any(|entry| {
+        entry.rsync_uri == "rsync://example.test/repo/issuer/rejected-with-reason.roa"
+            && matches!(entry.result, AuditObjectResult::Error)
+            && entry.detail.as_deref() == Some("EE certificate path validation failed: test")
+    }));
+    assert!(audit.objects.iter().any(|entry| {
+        entry.rsync_uri == "rsync://example.test/repo/issuer/rejected-legacy.roa"
+            && matches!(entry.result, AuditObjectResult::Error)
+            && entry.detail.as_deref() == Some(CACHED_REJECT_REASON_NOT_RECORDED)
+    }));
+    assert!(
+        audit.objects.iter().all(|entry| {
+            !matches!(entry.result, AuditObjectResult::Ok) || entry.detail.is_none()
+        })
+    );
+}
+
+#[test]
+fn build_publication_point_audit_from_pp_cache_projection_restores_reject_reason() {
+    let now = time::OffsetDateTime::now_utc();
+    let child_cert_hash = sha256_hex(b"child-cert");
+    let mut vcir = sample_vcir_for_projection(now, &child_cert_hash);
+    vcir.related_artifacts.push(VcirRelatedArtifact {
+        artifact_role: VcirArtifactRole::SignedObject,
+        artifact_kind: VcirArtifactKind::Roa,
+        uri: Some("rsync://example.test/repo/issuer/rejected-with-reason.roa".to_string()),
+        sha256: sha256_hex(b"rejected-with-reason"),
+        object_type: Some("roa".to_string()),
+        validation_status: VcirArtifactValidationStatus::Rejected,
+        reject_reason: Some("EE certificate path validation failed: test".to_string()),
+    });
+    vcir.related_artifacts.push(VcirRelatedArtifact {
+        artifact_role: VcirArtifactRole::SignedObject,
+        artifact_kind: VcirArtifactKind::Roa,
+        uri: Some("rsync://example.test/repo/issuer/rejected-legacy.roa".to_string()),
+        sha256: sha256_hex(b"rejected-legacy"),
+        object_type: Some("roa".to_string()),
+        validation_status: VcirArtifactValidationStatus::Rejected,
+        reject_reason: None,
+    });
+    let projection = PublicationPointCacheProjection::from_vcir_with_context(
+        &vcir,
+        "rsync://example.test/repo/issuer/".to_string(),
+        None,
+        [0x11; 32],
+        [0x22; 32],
+        [0x33; 32],
+        [0x44; 32],
+        [0x55; 32],
+    )
+    .expect("build publication point projection");
+
+    let ca = CaInstanceHandle {
+        depth: 0,
+        tal_id: "test-tal".to_string(),
+        parent_manifest_rsync_uri: None,
+        ca_certificate: CaCertificateRef::inline_der(Vec::new()),
+        ca_certificate_rsync_uri: None,
+        effective_ip_resources: None,
+        effective_as_resources: None,
+        rsync_base_uri: "rsync://example.test/repo/issuer/".to_string(),
+        manifest_rsync_uri: vcir.manifest_rsync_uri.clone(),
+        publication_point_rsync_uri: "rsync://example.test/repo/issuer/".to_string(),
+        rrdp_notification_uri: None,
+    };
+    let objects = crate::validation::objects::ObjectsOutput {
+        vrps: Vec::new(),
+        aspas: Vec::new(),
+        router_keys: Vec::new(),
+        local_outputs_cache: Vec::new(),
+        warnings: Vec::new(),
+        stats: crate::validation::objects::ObjectsStats::default(),
+        audit: Vec::new(),
+        roa_cache_stats: crate::validation::objects::RoaValidationCacheStats::default(),
+        roa_cache_object_meta: Vec::new(),
+    };
+
+    let audit = build_publication_point_audit_from_publication_point_cache_projection(
+        &ca,
+        PublicationPointSource::PublicationPointCache,
+        None,
+        None,
+        None,
+        None,
+        &projection,
+        now,
+        &[],
+        &objects,
+        &[],
+    );
+
+    assert!(audit.objects.iter().any(|entry| {
+        entry.rsync_uri == "rsync://example.test/repo/issuer/rejected-with-reason.roa"
+            && matches!(entry.result, AuditObjectResult::Error)
+            && entry.detail.as_deref() == Some("EE certificate path validation failed: test")
+    }));
+    assert!(audit.objects.iter().any(|entry| {
+        entry.rsync_uri == "rsync://example.test/repo/issuer/rejected-legacy.roa"
+            && matches!(entry.result, AuditObjectResult::Error)
+            && entry.detail.as_deref() == Some(CACHED_REJECT_REASON_NOT_RECORDED)
+    }));
+}
+
 #[test]
 fn build_publication_point_audit_from_vcir_failed_no_cache_keeps_current_reject_only() {
     let now = time::OffsetDateTime::now_utc();
@@ -4948,6 +5162,7 @@ fn reconstruct_snapshot_from_vcir_reports_missing_manifest_and_related_raw_bytes
         sha256: sha256_hex(b"dup-roa-1"),
         object_type: Some("roa".to_string()),
         validation_status: VcirArtifactValidationStatus::Accepted,
+        reject_reason: None,
     });
     vcir.related_artifacts.push(VcirRelatedArtifact {
         artifact_role: VcirArtifactRole::SignedObject,
@@ -4956,6 +5171,7 @@ fn reconstruct_snapshot_from_vcir_reports_missing_manifest_and_related_raw_bytes
         sha256: sha256_hex(b"dup-roa-2"),
         object_type: Some("roa".to_string()),
         validation_status: VcirArtifactValidationStatus::Accepted,
+        reject_reason: None,
     });
     vcir.related_artifacts.push(VcirRelatedArtifact {
         artifact_role: VcirArtifactRole::IssuerCert,
@@ -4964,6 +5180,7 @@ fn reconstruct_snapshot_from_vcir_reports_missing_manifest_and_related_raw_bytes
         sha256: sha256_hex(b"issuer-cert"),
         object_type: Some("cer".to_string()),
         validation_status: VcirArtifactValidationStatus::Accepted,
+        reject_reason: None,
     });
 
     let ca = CaInstanceHandle {

src/verification_only.rs

@@ -8,8 +8,12 @@ use sha2::{Digest, Sha256};
 use time::format_description::well_known::Rfc3339;
 
 use crate::parallel::types::TalInputSpec;
-use crate::policy::Policy;
+use crate::policy::{CaFailedFetchPolicy, Policy};
-use crate::storage::{RepositoryBlobVerificationSummary, RocksStore};
+use crate::fetch::rsync_system::RsyncScopePolicy;
+use crate::storage::{
+    RepositoryBlobVerificationSummary, RocksStore, ValidatedCaInstanceResult,
+    VcirFailedFetchReuseIdentity, VcirReuseRecordsCleared,
+};
 
 pub const VALIDATION_CONTRACT_SCHEMA_VERSION: u32 = 1;
 pub const CURRENT_STATE_BINDING_SCHEMA_VERSION: u32 = 1;
@@ -21,11 +25,44 @@ pub struct ValidationCacheContract {
     pub roa: bool,
     pub child_certificate: bool,
     pub transport_prefetch: bool,
+    #[serde(default)]
+    pub crypto_signature: bool,
+}
+
+/// The normal-run fallback inputs which verification-only is permitted to
+/// replay.  The payload remains in the source work-db; the contract binds the
+/// selected VCIR and its failed-fetch identity by canonical CBOR hashes.
+#[derive(Clone, Debug, PartialEq, Eq, PartialOrd, Ord, Serialize, Deserialize)]
+#[serde(rename_all = "camelCase")]
+pub struct FallbackReplaySelection {
+    pub manifest_rsync_uri: String,
+    pub vcir_sha256: String,
+    pub reuse_identity_sha256: String,
+}
+
+#[derive(Clone, Debug, Default, PartialEq, Eq, Serialize, Deserialize)]
+#[serde(rename_all = "camelCase")]
+pub struct FallbackReplayMaterialization {
+    pub contract_selection_count: u64,
+    pub source_vcir_records_cleared: u64,
+    pub source_reuse_identity_records_cleared: u64,
+    pub imported_selection_count: u64,
+    pub scratch_vcir_count: u64,
+}
+
+#[derive(Clone, Debug, PartialEq, Eq, Serialize, Deserialize)]
+#[serde(rename_all = "camelCase")]
+pub struct FallbackReplaySelectionComparison {
+    pub matches: bool,
+    pub expected: Vec<FallbackReplaySelection>,
+    pub actual: Vec<FallbackReplaySelection>,
+    pub missing: Vec<FallbackReplaySelection>,
+    pub unexpected: Vec<FallbackReplaySelection>,
 }
 
 impl ValidationCacheContract {
     pub fn any_validation_cache_enabled(&self) -> bool {
-        self.publication_point || self.roa || self.child_certificate
+        self.publication_point || self.roa || self.child_certificate || self.crypto_signature
     }
 }
 
@@ -39,6 +76,13 @@ pub struct ValidationContract {
     pub max_ca_depth: usize,
     pub max_instances: Option<usize>,
     pub cache: ValidationCacheContract,
+    /// Source transport scope. With `transportPrefetch=true`, verification-only
+    /// replays this scope over the frozen repository view rather than deriving
+    /// a new publication-point scope.
+    #[serde(default)]
+    pub rsync_scope_policy: RsyncScopePolicy,
+    #[serde(default, skip_serializing_if = "Vec::is_empty")]
+    pub fallback_replay_selections: Vec<FallbackReplaySelection>,
 }
 
 #[derive(Clone, Debug, PartialEq, Eq, Serialize, Deserialize)]
@@ -89,6 +133,7 @@ impl ValidationContract {
         max_ca_depth: usize,
         max_instances: Option<usize>,
         cache: ValidationCacheContract,
+        rsync_scope_policy: RsyncScopePolicy,
     ) -> Result<Self, String> {
         Ok(Self {
             schema_version: VALIDATION_CONTRACT_SCHEMA_VERSION,
@@ -98,6 +143,8 @@ impl ValidationContract {
             max_ca_depth,
             max_instances,
             cache,
+            rsync_scope_policy,
+            fallback_replay_selections: Vec::new(),
         })
     }
 
@@ -119,6 +166,40 @@ impl ValidationContract {
         if self.max_ca_depth == 0 {
             return Err("validation contract maxCaDepth must be greater than zero".to_string());
         }
+        if !self.fallback_replay_selections.is_empty()
+            && self.policy.ca_failed_fetch_policy != CaFailedFetchPolicy::ReuseCurrentInstanceVcir
+        {
+            return Err(
+                "validation contract fallbackReplaySelections require caFailedFetchPolicy=reuse_current_instance_vcir"
+                    .to_string(),
+            );
+        }
+        let mut previous: Option<&str> = None;
+        for selection in &self.fallback_replay_selections {
+            if !selection.manifest_rsync_uri.starts_with("rsync://") {
+                return Err(format!(
+                    "validation contract fallbackReplaySelections manifestRsyncUri must be an rsync URI: {}",
+                    selection.manifest_rsync_uri
+                ));
+            }
+            validate_sha256_hex(
+                "validation contract fallbackReplaySelections vcirSha256",
+                &selection.vcir_sha256,
+            )?;
+            validate_sha256_hex(
+                "validation contract fallbackReplaySelections reuseIdentitySha256",
+                &selection.reuse_identity_sha256,
+            )?;
+            if let Some(previous) = previous {
+                if previous >= selection.manifest_rsync_uri.as_str() {
+                    return Err(
+                        "validation contract fallbackReplaySelections must be sorted and unique by manifestRsyncUri"
+                            .to_string(),
+                    );
+                }
+            }
+            previous = Some(&selection.manifest_rsync_uri);
+        }
         Ok(())
     }
 
@@ -195,6 +276,7 @@ pub struct PreparedVerification {
     pub cir: crate::cir::CanonicalInputRepresentation,
     pub contract: ValidationContract,
     pub blob_summary: RepositoryBlobVerificationSummary,
+    pub fallback_replay_materialization: FallbackReplayMaterialization,
     pub store: Option<Arc<RocksStore>>,
 }
 
@@ -232,6 +314,8 @@ pub struct VerificationRunMeta {
     pub repository_objects_checked: u64,
     pub repository_bytes_checked: u64,
     pub state_digest_match: bool,
+    pub fallback_replay_selection_match: bool,
+    pub fallback_replay_materialization: FallbackReplayMaterialization,
     pub scratch_retained: bool,
 }
 
@@ -303,6 +387,174 @@ pub fn cir_tal_uris(cir: &crate::cir::CanonicalInputRepresentation) -> Vec<Strin
         .collect()
 }
 
+/// Derive the exact fallback set used by a completed tree traversal and bind
+/// every selected payload to its canonical storage representation.
+pub fn collect_fallback_replay_selections(
+    store: &RocksStore,
+    publication_points: &[crate::audit::PublicationPointAudit],
+) -> Result<Vec<FallbackReplaySelection>, String> {
+    let mut selections = Vec::new();
+    for publication_point in publication_points {
+        if publication_point.source != "vcir_current_instance" {
+            continue;
+        }
+        selections.push(fallback_replay_selection_for_manifest(
+            store,
+            &publication_point.manifest_rsync_uri,
+        )?);
+    }
+    selections.sort();
+    for pair in selections.windows(2) {
+        if pair[0].manifest_rsync_uri == pair[1].manifest_rsync_uri {
+            return Err(format!(
+                "multiple vcir_current_instance publication points use the same manifest URI: {}",
+                pair[0].manifest_rsync_uri
+            ));
+        }
+    }
+    Ok(selections)
+}
+
+pub(crate) fn fallback_replay_selection_for_manifest(
+    store: &RocksStore,
+    manifest_rsync_uri: &str,
+) -> Result<FallbackReplaySelection, String> {
+    let vcir = store
+        .get_vcir(manifest_rsync_uri)
+        .map_err(|error| format!("load fallback VCIR failed for {manifest_rsync_uri}: {error}"))?
+        .ok_or_else(|| format!("fallback VCIR is missing for {manifest_rsync_uri}"))?;
+    let reuse_identity = store
+        .get_vcir_failed_fetch_reuse_identity(manifest_rsync_uri)
+        .map_err(|error| {
+            format!("load fallback VCIR reuse identity failed for {manifest_rsync_uri}: {error}")
+        })?
+        .ok_or_else(|| format!("fallback VCIR reuse identity is missing for {manifest_rsync_uri}"))?;
+    fallback_replay_selection_from_values(&vcir, &reuse_identity)
+}
+
+fn fallback_replay_selection_from_values(
+    vcir: &ValidatedCaInstanceResult,
+    reuse_identity: &VcirFailedFetchReuseIdentity,
+) -> Result<FallbackReplaySelection, String> {
+    Ok(FallbackReplaySelection {
+        manifest_rsync_uri: vcir.manifest_rsync_uri.clone(),
+        vcir_sha256: canonical_cbor_sha256("fallback VCIR", vcir)?,
+        reuse_identity_sha256: canonical_cbor_sha256("fallback VCIR reuse identity", reuse_identity)?,
+    })
+}
+
+fn canonical_cbor_sha256<T: Serialize>(label: &str, value: &T) -> Result<String, String> {
+    let bytes = serde_cbor::to_vec(value)
+        .map_err(|error| format!("serialize {label} for SHA-256 failed: {error}"))?;
+    Ok(hex::encode(Sha256::digest(bytes)))
+}
+
+pub(crate) fn materialize_fallback_replay_selections(
+    store: &RocksStore,
+    selections: &[FallbackReplaySelection],
+) -> Result<FallbackReplayMaterialization, String> {
+    let mut extracted = Vec::with_capacity(selections.len());
+    for selection in selections {
+        let actual = fallback_replay_selection_for_manifest(store, &selection.manifest_rsync_uri)?;
+        if actual != *selection {
+            return Err(format!(
+                "fallback replay selection hash mismatch for {}: contract vcir={}, work-db vcir={}, contract identity={}, work-db identity={}",
+                selection.manifest_rsync_uri,
+                selection.vcir_sha256,
+                actual.vcir_sha256,
+                selection.reuse_identity_sha256,
+                actual.reuse_identity_sha256,
+            ));
+        }
+        let vcir = store
+            .get_vcir(&selection.manifest_rsync_uri)
+            .map_err(|error| format!("reload selected fallback VCIR failed: {error}"))?
+            .expect("fallback selection was just verified as present");
+        let reuse_identity = store
+            .get_vcir_failed_fetch_reuse_identity(&selection.manifest_rsync_uri)
+            .map_err(|error| format!("reload selected fallback reuse identity failed: {error}"))?
+            .expect("fallback selection was just verified as present");
+        extracted.push((vcir, reuse_identity));
+    }
+
+    let VcirReuseRecordsCleared {
+        vcir_records,
+        failed_fetch_identity_records,
+    } = store
+        .clear_vcir_reuse_records()
+        .map_err(|error| format!("clear scratch VCIR fallback records failed: {error}"))?;
+    for (vcir, reuse_identity) in &extracted {
+        store
+            .put_vcir_with_failed_fetch_reuse_identity(vcir, reuse_identity)
+            .map_err(|error| format!("import selected fallback VCIR failed: {error}"))?;
+    }
+
+    let actual = store
+        .list_vcirs()
+        .map_err(|error| format!("list scratch fallback VCIR records failed: {error}"))?
+        .iter()
+        .map(|vcir| {
+            let reuse_identity = store
+                .get_vcir_failed_fetch_reuse_identity(&vcir.manifest_rsync_uri)
+                .map_err(|error| format!("read materialized fallback reuse identity failed: {error}"))?
+                .ok_or_else(|| {
+                    format!(
+                        "materialized fallback reuse identity is missing for {}",
+                        vcir.manifest_rsync_uri
+                    )
+                })?;
+            fallback_replay_selection_from_values(vcir, &reuse_identity)
+        })
+        .collect::<Result<Vec<_>, String>>()?;
+    if actual != selections {
+        return Err(format!(
+            "scratch fallback replay allowlist differs from validation contract: expected={}, actual={}",
+            selections.len(),
+            actual.len()
+        ));
+    }
+
+    Ok(FallbackReplayMaterialization {
+        contract_selection_count: selections.len() as u64,
+        source_vcir_records_cleared: vcir_records,
+        source_reuse_identity_records_cleared: failed_fetch_identity_records,
+        imported_selection_count: extracted.len() as u64,
+        scratch_vcir_count: actual.len() as u64,
+    })
+}
+
+pub fn compare_fallback_replay_selections(
+    expected: &[FallbackReplaySelection],
+    actual: &[FallbackReplaySelection],
+) -> FallbackReplaySelectionComparison {
+    let missing: Vec<FallbackReplaySelection> = expected
+        .iter()
+        .filter(|selection| !actual.contains(selection))
+        .cloned()
+        .collect();
+    let unexpected: Vec<FallbackReplaySelection> = actual
+        .iter()
+        .filter(|selection| !expected.contains(selection))
+        .cloned()
+        .collect();
+    FallbackReplaySelectionComparison {
+        matches: missing.is_empty() && unexpected.is_empty(),
+        expected: expected.to_vec(),
+        actual: actual.to_vec(),
+        missing,
+        unexpected,
+    }
+}
+
+pub fn write_fallback_replay_selection_comparison(
+    path: &Path,
+    comparison: &FallbackReplaySelectionComparison,
+) -> Result<(), String> {
+    let bytes = serde_json::to_vec_pretty(comparison)
+        .map_err(|error| format!("encode fallback replay selection comparison failed: {error}"))?;
+    atomic_write(path, &bytes)
+}
+
 pub fn prepare_verification(
     source_run_dir: &Path,
     source_state_root: &Path,
@@ -358,7 +610,7 @@ pub fn prepare_verification(
     contract.require_current_binary()?;
     if !contract.cache.any_validation_cache_enabled() {
         return Err(
-            "source run did not enable PP, ROA, or child-certificate validation cache".to_string(),
+            "source run did not enable PP, ROA, child-certificate, or crypto signature validation cache".to_string(),
         );
     }
     if !artifacts.source_work_db.is_dir() {
@@ -389,6 +641,10 @@ pub fn prepare_verification(
         )
         .map_err(|error| format!("open frozen verification store failed: {error}"))?,
     );
+    let fallback_replay_materialization = materialize_fallback_replay_selections(
+        store.as_ref(),
+        &contract.fallback_replay_selections,
+    )?;
     let blob_summary = store
         .verify_current_repository_blobs(1024)
         .map_err(|error| format!("verify frozen repository blobs failed: {error}"))?;
@@ -399,6 +655,7 @@ pub fn prepare_verification(
         cir,
         contract,
         blob_summary,
+        fallback_replay_materialization,
         store: Some(store),
     })
 }
@@ -645,7 +902,10 @@ mod tests {
                 roa: true,
                 child_certificate: true,
                 transport_prefetch: false,
+                crypto_signature: false,
             },
+            rsync_scope_policy: RsyncScopePolicy::default(),
+            fallback_replay_selections: Vec::new(),
         }
     }
 
@@ -751,7 +1011,9 @@ mod tests {
                 roa: false,
                 child_certificate: false,
                 transport_prefetch: true,
+                crypto_signature: false,
             },
+            RsyncScopePolicy::default(),
         )
         .expect("contract");
         write_validation_contract(&source_run.join("validation-contract.json"), &contract)
@@ -795,6 +1057,60 @@ mod tests {
         assert!(contract.validate().is_err());
     }
 
+    #[test]
+    fn fallback_replay_contract_requires_reuse_policy_and_sorted_unique_selections() {
+        let selection = FallbackReplaySelection {
+            manifest_rsync_uri: "rsync://example.test/repo/current.mft".to_string(),
+            vcir_sha256: "11".repeat(32),
+            reuse_identity_sha256: "22".repeat(32),
+        };
+        let mut contract = sample_contract();
+        contract.policy.ca_failed_fetch_policy = CaFailedFetchPolicy::StopAllOutput;
+        contract.fallback_replay_selections = vec![selection.clone()];
+        assert!(contract
+            .validate()
+            .unwrap_err()
+            .contains("caFailedFetchPolicy"));
+
+        contract.policy.ca_failed_fetch_policy = CaFailedFetchPolicy::ReuseCurrentInstanceVcir;
+        contract.fallback_replay_selections = vec![
+            FallbackReplaySelection {
+                manifest_rsync_uri: "rsync://example.test/repo/z.mft".to_string(),
+                ..selection.clone()
+            },
+            FallbackReplaySelection {
+                manifest_rsync_uri: "rsync://example.test/repo/a.mft".to_string(),
+                ..selection.clone()
+            },
+        ];
+        assert!(contract.validate().unwrap_err().contains("sorted and unique"));
+        contract.fallback_replay_selections.sort();
+        contract.validate().expect("sorted selection contract");
+    }
+
+    #[test]
+    fn fallback_replay_selection_comparison_reports_exact_differences() {
+        let expected = vec![FallbackReplaySelection {
+            manifest_rsync_uri: "rsync://example.test/repo/a.mft".to_string(),
+            vcir_sha256: "11".repeat(32),
+            reuse_identity_sha256: "22".repeat(32),
+        }];
+        let mut actual = expected.clone();
+        actual[0].vcir_sha256 = "33".repeat(32);
+        let comparison = compare_fallback_replay_selections(&expected, &actual);
+        assert!(!comparison.matches);
+        assert_eq!(comparison.missing, expected);
+        assert_eq!(comparison.unexpected, actual);
+
+        let temp = tempfile::tempdir().expect("tempdir");
+        let path = temp.path().join("fallback-replay-selection.json");
+        write_fallback_replay_selection_comparison(&path, &comparison)
+            .expect("write fallback selection comparison");
+        assert!(fs::read_to_string(path)
+            .expect("read fallback selection comparison")
+            .contains("reuseIdentitySha256"));
+    }
+
     #[test]
     fn cache_contract_distinguishes_prefetch_from_validation_caches() {
         let cache = ValidationCacheContract {
@@ -802,10 +1118,23 @@ mod tests {
             roa: false,
             child_certificate: false,
             transport_prefetch: true,
+            crypto_signature: false,
         };
         assert!(!cache.any_validation_cache_enabled());
     }
 
+    #[test]
+    fn validation_contract_defaults_missing_rsync_scope_to_module_root() {
+        let mut value = serde_json::to_value(sample_contract()).expect("serialize contract");
+        value
+            .as_object_mut()
+            .expect("contract object")
+            .remove("rsyncScopePolicy");
+        let decoded: ValidationContract =
+            serde_json::from_value(value).expect("decode older contract");
+        assert_eq!(decoded.rsync_scope_policy, RsyncScopePolicy::ModuleRoot);
+    }
+
     #[test]
     fn validation_contract_and_binding_validation_cover_rejection_paths() {
         let time = time::OffsetDateTime::parse(
@@ -819,6 +1148,7 @@ mod tests {
             32,
             Some(10),
             sample_contract().cache,
+            RsyncScopePolicy::default(),
         )
         .expect("current contract");
         assert_eq!(contract.validation_time().expect("contract time"), time);
@@ -914,6 +1244,8 @@ mod tests {
             repository_objects_checked: 3,
             repository_bytes_checked: 4,
             state_digest_match: true,
+            fallback_replay_selection_match: true,
+            fallback_replay_materialization: FallbackReplayMaterialization::default(),
             scratch_retained: false,
         };
         let meta_path = temp.path().join("meta/result.json");
@@ -1003,6 +1335,16 @@ mod tests {
         assert_eq!(prepared.blob_summary.current_objects, 0);
         assert_eq!(prepared.blob_summary.bytes_verified, 0);
         assert_eq!(prepared.contract.validation_time, "2026-07-16T00:00:00Z");
+        assert_eq!(
+            prepared
+                .fallback_replay_materialization
+                .contract_selection_count,
+            0
+        );
+        assert_eq!(
+            prepared.fallback_replay_materialization.scratch_vcir_count,
+            0
+        );
         assert_eq!(cir_tal_uris(&prepared.cir).len(), 1);
         assert!(prepared.artifacts.scratch_work_db.is_dir());
         assert!(prepared.artifacts.result_contract.is_file());

tests/test_apnic_rrdp_delta_live_20260226.rs

@@ -29,7 +29,7 @@ fn live_http_fetcher() -> BlockingHttpFetcher {
     BlockingHttpFetcher::new(HttpFetcherConfig {
         timeout: Duration::from_secs(timeout_secs),
         large_body_timeout: Duration::from_secs(timeout_secs),
-        user_agent: "rpki-dev/0.1 (stage2 live rrdp delta test)".to_string(),
+        user_agent: "panda-rpki/0.2 (live rrdp delta test)".to_string(),
         ..HttpFetcherConfig::default()
     })
     .expect("http fetcher")

tests/test_ccr_m7.rs

@@ -244,6 +244,7 @@ fn verify_against_vcir_store_matches_manifest_hashes() {
             sha256: manifest_hash.clone(),
             object_type: Some("mft".to_string()),
             validation_status: VcirArtifactValidationStatus::Accepted,
+            reject_reason: None,
         }],
         summary: VcirSummary {
             local_vrp_count: 0,

tests/test_manifest_processor_m4.rs

@@ -109,6 +109,7 @@ fn store_validated_manifest_baseline(
             sha256: manifest_sha256,
             object_type: Some("mft".to_string()),
             validation_status: VcirArtifactValidationStatus::Accepted,
+            reject_reason: None,
         }],
         summary: VcirSummary {
             local_vrp_count: 0,

tests/test_manifest_processor_repo_sync_and_cached_snapshot_cov.rs

@@ -112,6 +112,7 @@ fn store_validated_manifest_baseline(
             sha256: manifest_sha256,
             object_type: Some("mft".to_string()),
             validation_status: VcirArtifactValidationStatus::Accepted,
+            reject_reason: None,
         }],
         summary: VcirSummary {
             local_vrp_count: 0,

ui/rpki-explorer/src/api/client.ts

@@ -51,11 +51,22 @@ async function parseErrorBody(res: Response): Promise<string> {
   return `HTTP ${res.status} ${res.statusText}`.trim();
 }
 
+/** Default timeout for every API request — hung streams must not spin forever. */
+const REQUEST_TIMEOUT_MS = 30_000;
+
+/** Combine the caller's abort signal (if any) with the default timeout. */
+function requestSignal(signal?: AbortSignal | null): AbortSignal {
+  const signals = [signal, AbortSignal.timeout(REQUEST_TIMEOUT_MS)].filter(
+    (s): s is AbortSignal => s != null,
+  );
+  return AbortSignal.any(signals);
+}
+
 /** Fetch JSON from the API, throwing ApiError on non-2xx. */
 export async function apiFetch<T>(path: string, init?: RequestInit): Promise<T> {
   let res: Response;
   try {
-    res = await fetch(path, init);
+    res = await fetch(path, { ...init, signal: requestSignal(init?.signal) });
   } catch (err) {
     throw new ApiError(0, err instanceof Error ? err.message : "network error");
   }
@@ -69,7 +80,7 @@ export async function apiFetch<T>(path: string, init?: RequestInit): Promise<T>
 export async function apiFetchBlob(path: string): Promise<Blob> {
   let res: Response;
   try {
-    res = await fetch(path);
+    res = await fetch(path, { signal: requestSignal() });
   } catch (err) {
     throw new ApiError(0, err instanceof Error ? err.message : "network error");
   }

ui/rpki-explorer/src/api/schemas.ts

@@ -244,7 +244,8 @@ export const manifestFileEntrySchema = z
 export const revokedCertEntrySchema = z
   .object({
     serialNumberHex: z.string().nullish(),
-    serialNumber: z.string().nullish(),
+    // Backend serializes CRL serials as a JSON number when they fit, else a string.
+    serialNumber: z.union([z.string(), z.number()]).nullish(),
     revocationDate: z.string().nullish(),
   })
   .passthrough();

ui/rpki-explorer/src/api/service.ts

@@ -76,7 +76,8 @@ async function getList<S extends z.ZodTypeAny>(
 /* ------------------------------------------------------------------ */
 
 export function getServiceInfo(): Promise<ServiceInfo> {
-  return getData(`${API}`, serviceInfoSchema);
+  // Trailing slash: the axum router serves the info document at `/api/v1/`.
+  return getData(`${API}/`, serviceInfoSchema);
 }
 
 export function getHealth(): Promise<HealthStatus> {
@@ -116,9 +117,15 @@ export function getRunSummary(runId: string): Promise<RunSummary> {
 /* Repositories                                                        */
 /* ------------------------------------------------------------------ */
 
+export type RepoListParams = PageParams & {
+  /** Server-side sort; currently only "objects" is supported. */
+  sort?: "objects";
+  order?: "asc" | "desc";
+};
+
 export function listRepos(
   runId: string,
-  params: PageParams = {},
+  params: RepoListParams = {},
 ): Promise<ListResult<RepositoryRecord>> {
   return getList(`${runBase(runId)}/repos`, repositoryRecordSchema, params);
 }
@@ -417,6 +424,10 @@ export function getStatsObjectTypes(runId: string): Promise<Record<string, numbe
   return getData(`${runBase(runId)}/stats/object-types`, statsMapSchema);
 }
 
+export function getStatsOverview(runId: string): Promise<Record<string, number>> {
+  return getData(`${runBase(runId)}/stats/overview`, statsMapSchema);
+}
+
 export function getStatsValidation(runId: string): Promise<Record<string, number>> {
   return getData(`${runBase(runId)}/stats/validation`, statsMapSchema);
 }

ui/rpki-explorer/src/components/DataTable.tsx

@@ -86,6 +86,19 @@ export function DataTable<T>({
               key={rowKey(row)}
               className={onRowClick ? "clickable" : undefined}
               onClick={onRowClick ? () => onRowClick(row) : undefined}
+              // Clickable rows behave like links: focusable and Enter/Space-activated.
+              tabIndex={onRowClick ? 0 : undefined}
+              role={onRowClick ? "link" : undefined}
+              onKeyDown={
+                onRowClick
+                  ? (event) => {
+                      if (event.key === "Enter" || event.key === " ") {
+                        event.preventDefault();
+                        onRowClick(row);
+                      }
+                    }
+                  : undefined
+              }
             >
               {columns.map((col) => (
                 <td

ui/rpki-explorer/src/components/ObjectsTable.tsx

@@ -14,16 +14,48 @@ import { StatusPill } from "./StatusPill";
 export interface ObjectFilterValues {
   type: string;
   result: string;
+  reason: string;
   rejectedOnly: boolean;
   q: string;
 }
 export const EMPTY_OBJECT_FILTERS: ObjectFilterValues = {
   type: "",
   result: "",
+  reason: "",
   rejectedOnly: false,
   q: "",
 };
 
+/** Read object filters from the page URL (standalone pages sync them there). */
+export function objectFiltersFromParams(params: URLSearchParams): ObjectFilterValues {
+  return {
+    type: params.get("type") ?? "",
+    result: params.get("result") ?? "",
+    reason: params.get("reason") ?? "",
+    rejectedOnly: params.get("rejected") === "true",
+    q: params.get("q") ?? "",
+  };
+}
+
+/** Write object filters into a URLSearchParams copy, preserving other params. */
+export function objectFiltersToParams(
+  next: ObjectFilterValues,
+  base: URLSearchParams,
+): URLSearchParams {
+  const params = new URLSearchParams(base);
+  const write = (key: string, value: string) => {
+    if (value) params.set(key, value);
+    else params.delete(key);
+  };
+  write("type", next.type);
+  write("result", next.result);
+  write("reason", next.reason);
+  write("q", next.q);
+  if (next.rejectedOnly) params.set("rejected", "true");
+  else params.delete("rejected");
+  return params;
+}
+
 /** Report-stream object type vocabulary (backend also accepts mft/cer/asa aliases). */
 const TYPE_OPTIONS = ["roa", "manifest", "crl", "certificate", "aspa", "gbr"];
 const RESULT_OPTIONS = ["ok", "error", "skipped"];
@@ -55,7 +87,16 @@ export function ObjectsTable({
   useEffect(() => {
     setQDraft(filters.q);
   }, [filters.q]);
-  const filterKey = JSON.stringify([runId, filters.type, filters.result, filters.rejectedOnly, filters.q]);
+
+  // Auto-apply the URI draft after a short typing pause (Enter/blur still work).
+  useEffect(() => {
+    const trimmed = qDraft.trim();
+    if (trimmed === filters.q) return;
+    const timer = setTimeout(() => onFiltersChange({ ...filters, q: trimmed }), 400);
+    return () => clearTimeout(timer);
+  }, [qDraft, filters, onFiltersChange]);
+
+  const filterKey = JSON.stringify([runId, filters.type, filters.result, filters.reason, filters.rejectedOnly, filters.q]);
   const pager = useCursorPager(filterKey);
 
   const query = useQuery({
@@ -67,6 +108,7 @@ export function ObjectsTable({
         type: filters.type || undefined,
         result: filters.result || undefined,
         rejected: filters.rejectedOnly || undefined,
+        reason: filters.reason || undefined,
         q: filters.q || undefined,
       }),
     placeholderData: (prev) => prev,
@@ -101,7 +143,9 @@ export function ObjectsTable({
       {
         key: "result",
         header: "Result",
-        render: (obj) => <StatusPill status={obj.rejected ? "rejected" : obj.result} />,
+        // Raw `result` value — the Result filter options must match what the
+        // column displays (rejection is visible via the reason column/filter).
+        render: (obj) => <StatusPill status={obj.result} />,
       },
       ...(showReason
         ? [
@@ -162,6 +206,17 @@ export function ObjectsTable({
             ))}
           </select>
         </div>
+        <div className="filter-field">
+          <label htmlFor={`${queryKeyScope}-reason`}>Reason contains</label>
+          <input
+            id={`${queryKeyScope}-reason`}
+            type="search"
+            className="mono"
+            value={filters.reason}
+            onChange={(e) => onFiltersChange({ ...filters, reason: e.target.value })}
+            placeholder="substring…"
+          />
+        </div>
         <div className="filter-field">
           <label htmlFor={`${queryKeyScope}-q`}>URI contains</label>
           <input
@@ -174,7 +229,7 @@ export function ObjectsTable({
               if (e.key === "Enter") submitQ();
             }}
             onBlur={submitQ}
-            placeholder="substring…"
+            placeholder="substring… (applies as you type)"
           />
         </div>
         <label className="filter-check">
@@ -186,7 +241,7 @@ export function ObjectsTable({
           Rejected only
         </label>
         <span className="text-faint text-small" style={{ marginLeft: "auto" }}>
-          server-side filters
+          server-side filters · URI search applies as you type
         </span>
       </div>
 

ui/rpki-explorer/src/components/ProjectionView.tsx

@@ -1,5 +1,5 @@
 import { useQuery } from "@tanstack/react-query";
-import type { ReactNode } from "react";
+import { useState, type ReactNode } from "react";
 import { listManifestFiles, listRevokedCertificates } from "../api/service";
 import type { ProjectionRecord } from "../api/schemas";
 import { CopyableValue } from "./CopyableValue";
@@ -19,6 +19,147 @@ function MetaRow({ label, value, mono = false }: { label: string; value: ReactNo
   );
 }
 
+/** Format an IP address from a projection byte array (v4 dotted / v6 compressed hextets). */
+function formatIpAddr(addr: unknown, afi: unknown): string | null {
+  if (!Array.isArray(addr)) return null;
+  const bytes = addr.filter((b): b is number => typeof b === "number");
+  if (bytes.length === 0) return null;
+  if (asString(afi) === "Ipv6" || bytes.length > 4) {
+    const groups: number[] = [];
+    for (let i = 0; i + 1 < bytes.length; i += 2) {
+      groups.push((bytes[i] << 8) | bytes[i + 1]);
+    }
+    // RFC 5952-style compression of the longest run of zero groups.
+    let bestStart = -1;
+    let bestLen = 0;
+    for (let i = 0; i < groups.length; ) {
+      if (groups[i] !== 0) {
+        i += 1;
+        continue;
+      }
+      let j = i;
+      while (j < groups.length && groups[j] === 0) j += 1;
+      if (j - i > bestLen) {
+        bestStart = i;
+        bestLen = j - i;
+      }
+      i = j;
+    }
+    const hex = groups.map((g) => g.toString(16));
+    if (bestLen >= 2) {
+      const head = hex.slice(0, bestStart).join(":");
+      const tail = hex.slice(bestStart + bestLen).join(":");
+      return `${head}::${tail}`;
+    }
+    return hex.join(":");
+  }
+  return bytes.join(".");
+}
+
+/**
+ * Flatten projection ipResources to display strings.
+ * Real shape (serde of the rpki types):
+ *   { families: [ { afi, choice: { AddressesOrRanges: [ {Prefix:{addr,afi,prefix_len}} | {Range:{min,max}} ] } } ] }
+ */
+function formatIpResources(resources: unknown): string[] {
+  const out: string[] = [];
+  for (const famRaw of asArray(asRecord(resources)?.families)) {
+    const fam = asRecord(famRaw);
+    const choice = asRecord(fam?.choice);
+    if (!choice) continue;
+    if ("Inherit" in choice) {
+      out.push("inherit");
+      continue;
+    }
+    for (const entryRaw of asArray(choice.AddressesOrRanges)) {
+      const entry = asRecord(entryRaw);
+      const prefix = asRecord(entry?.Prefix);
+      if (prefix) {
+        const base = formatIpAddr(prefix.addr, prefix.afi ?? fam?.afi);
+        out.push(base ? `${base}/${asNumber(prefix.prefix_len) ?? "?"}` : JSON.stringify(entryRaw));
+        continue;
+      }
+      const range = asRecord(entry?.Range);
+      if (range) {
+        const min = formatIpAddr(asRecord(range.min)?.addr ?? range.min, fam?.afi);
+        const max = formatIpAddr(asRecord(range.max)?.addr ?? range.max, fam?.afi);
+        out.push(min && max ? `${min} - ${max}` : JSON.stringify(entryRaw));
+        continue;
+      }
+      out.push(JSON.stringify(entryRaw));
+    }
+  }
+  return out;
+}
+
+/**
+ * Flatten projection asResources to display strings.
+ * Real shape: { asnum: { AsIdsOrRanges: [ {Id: n} | {Range: {min, max}} ] } } or { "inherit"/"Inherit": … }.
+ */
+function formatAsResources(resources: unknown): string[] {
+  const rec = asRecord(resources);
+  if (!rec) return [];
+  if ("inherit" in rec || "Inherit" in rec) return ["inherit"];
+  const out: string[] = [];
+  for (const entryRaw of asArray(asRecord(rec.asnum)?.AsIdsOrRanges)) {
+    const entry = asRecord(entryRaw);
+    const id = asNumber(entry?.Id);
+    if (id != null) {
+      out.push(`AS${id}`);
+      continue;
+    }
+    const range = asRecord(entry?.Range);
+    if (range) {
+      out.push(`AS${asNumber(range.min) ?? "?"}-AS${asNumber(range.max) ?? "?"}`);
+      continue;
+    }
+    out.push(JSON.stringify(entryRaw));
+  }
+  return out;
+}
+
+/** Render a list of URIs (one per line) or null when empty. */
+function uriLines(uris: string[]): ReactNode {
+  if (!uris.length) return null;
+  return (
+    <span>
+      {uris.map((u) => (
+        <span key={u} className="mono" style={{ display: "block", wordBreak: "break-all" }}>
+          {u}
+        </span>
+      ))}
+    </span>
+  );
+}
+
+/**
+ * Chip list that collapses to `collapsedCount` entries with a clickable
+ * "+N more" toggle; expanding shows every item (plus a "Show less" toggle).
+ */
+export function ExpandableChipList({ items, collapsedCount = 12 }: { items: string[]; collapsedCount?: number }) {
+  const [expanded, setExpanded] = useState(false);
+  if (!items.length) return <>—</>;
+  const collapsible = items.length > collapsedCount;
+  const visible = expanded || !collapsible ? items : items.slice(0, collapsedCount);
+  return (
+    <span className="chip-list">
+      {visible.map((item, i) => (
+        <span className="chip" key={i}>{item}</span>
+      ))}
+      {collapsible ? (
+        <button
+          type="button"
+          className="chip chip-more"
+          onClick={() => setExpanded((prev) => !prev)}
+          aria-expanded={expanded}
+        >
+          {expanded ? "Show less" : `+${items.length - collapsedCount} more`}
+        </button>
+      ) : null}
+    </span>
+  );
+}
+
 function ManifestFiles({ runId, objectInstanceId }: { runId: string; objectInstanceId: string }) {
   const pager = useCursorPager(`${runId}:${objectInstanceId}:mft`);
   const query = useQuery({
@@ -69,12 +210,16 @@ function RevokedCerts({ runId, objectInstanceId }: { runId: string; objectInstan
     placeholderData: (prev) => prev,
   });
 
-  const columns: Column<{ serialNumberHex?: string | null; serialNumber?: string | null; revocationDate?: string | null }>[] = [
+  const columns: Column<{ serialNumberHex?: string | null; serialNumber?: string | number | null; revocationDate?: string | null }>[] = [
     {
       key: "serial",
       header: "Serial number",
       render: (row) => (
-        <CopyableValue value={row.serialNumberHex ?? row.serialNumber} max={44} label="serial number" />
+        <CopyableValue
+          value={row.serialNumberHex ?? (row.serialNumber != null ? String(row.serialNumber) : null)}
+          max={44}
+          label="serial number"
+        />
       ),
     },
     { key: "date", header: "Revocation date", render: (row) => formatUtc(row.revocationDate) },
@@ -175,12 +320,43 @@ function AspaView({ projection }: { projection: Record<string, unknown> }) {
   );
 }
 
-function CertificateView({ projection }: { projection: Record<string, unknown> }) {
+export function CertificateView({ projection }: { projection: Record<string, unknown> }) {
-  const cert = asRecord(projection.certificate) ?? asRecord(projection.cer) ?? projection;
+  // Real API shape: the certificate payload lives in `resourceCertificate`
+  // with extensions nested under `extensions`; keep the older
+  // certificate/cer + flat-field variants working as fallbacks.
+  const cert =
+    asRecord(projection.certificate) ??
+    asRecord(projection.cer) ??
+    asRecord(projection.resourceCertificate) ??
+    projection;
   const validity = asRecord(cert.validity);
-  const resources = asRecord(cert.resources);
+  const ext = asRecord(cert.extensions);
-  const ipBlocks = asArray(resources?.ipResources ?? resources?.ipv4 ?? resources?.addresses);
+  const legacyResources = asRecord(cert.resources);
-  const asBlocks = asArray(resources?.asResources ?? resources?.asns);
+
+  const siaDescriptions = asArray(asRecord(ext?.subjectInfoAccess)?.accessDescriptions);
+  const siaUris = siaDescriptions
+    .map((d) => asString(asRecord(d)?.accessLocation))
+    .filter((v): v is string => Boolean(v));
+  const aiaUris = asArray(ext?.caIssuersUris)
+    .map((u) => asString(u))
+    .filter((v): v is string => Boolean(v));
+  const crlDpUris = asArray(ext?.crlDistributionPointsUris)
+    .map((u) => asString(u))
+    .filter((v): v is string => Boolean(v));
+
+  const extIp = formatIpResources(ext?.ipResources);
+  const ipBlocks = extIp.length
+    ? extIp
+    : asArray(legacyResources?.ipResources ?? legacyResources?.ipv4 ?? legacyResources?.addresses).map((b) =>
+        typeof b === "string" ? b : JSON.stringify(b),
+      );
+  const extAs = formatAsResources(ext?.asResources);
+  const asBlocks = extAs.length
+    ? extAs
+    : asArray(legacyResources?.asResources ?? legacyResources?.asns).map((b) =>
+        typeof b === "string" || typeof b === "number" ? String(b) : JSON.stringify(b),
+      );
+
   return (
     <dl className="meta-grid">
       <MetaRow label="Serial number" value={asString(cert.serialNumberHex ?? cert.serialNumber)} mono />
@@ -188,43 +364,42 @@ function CertificateView({ projection }: { projection: Record<string, unknown> }
       <MetaRow label="Subject" value={asString(cert.subject)} mono />
       <MetaRow
         label="Validity"
-        value={`${asString(validity?.notBefore) ?? "—"} → ${asString(validity?.notAfter) ?? "—"}`}
+        value={`${formatUtc(asString(validity?.notBefore)) ?? "—"} → ${formatUtc(asString(validity?.notAfter)) ?? "—"}`}
         mono
       />
-      <MetaRow label="SKI" value={<CopyableValue value={asString(cert.ski)} max={56} label="SKI" />} />
-      <MetaRow label="AKI" value={<CopyableValue value={asString(cert.aki)} max={56} label="AKI" />} />
-      <MetaRow label="SIA" value={<CopyableValue value={asString(cert.sia)} max={72} label="SIA" />} />
-      <MetaRow label="AIA" value={<CopyableValue value={asString(cert.aia)} max={72} label="AIA" />} />
-      <MetaRow label="CRL DP" value={<CopyableValue value={asString(cert.crlDistributionPoint ?? cert.crldp)} max={72} label="CRL distribution point" />} />
       <MetaRow
-        label="IP resources"
+        label="SKI"
+        value={<CopyableValue value={asString(cert.ski) ?? asString(ext?.subjectKeyIdentifier)} max={56} label="SKI" />}
+      />
+      <MetaRow
+        label="AKI"
+        value={<CopyableValue value={asString(cert.aki) ?? asString(ext?.authorityKeyIdentifier)} max={56} label="AKI" />}
+      />
+      <MetaRow
+        label="SIA"
+        value={asString(cert.sia) ? <CopyableValue value={asString(cert.sia)} max={72} label="SIA" /> : uriLines(siaUris)}
+      />
+      <MetaRow
+        label="AIA"
+        value={asString(cert.aia) ? <CopyableValue value={asString(cert.aia)} max={72} label="AIA" /> : uriLines(aiaUris)}
+      />
+      <MetaRow
+        label="CRL DP"
         value={
-          ipBlocks.length ? (
+          asString(cert.crlDistributionPoint ?? cert.crldp) ? (
-            <span className="chip-list">
+            <CopyableValue value={asString(cert.crlDistributionPoint ?? cert.crldp)} max={72} label="CRL distribution point" />
-              {ipBlocks.slice(0, 12).map((b, i) => (
-                <span className="chip" key={i}>{typeof b === "string" ? b : JSON.stringify(b)}</span>
-              ))}
-              {ipBlocks.length > 12 ? <span className="chip">+{ipBlocks.length - 12} more</span> : null}
-            </span>
           ) : (
-            "—"
+            uriLines(crlDpUris)
           )
         }
       />
+      <MetaRow
+        label="IP resources"
+        value={<ExpandableChipList items={ipBlocks} />}
+      />
       <MetaRow
         label="AS resources"
-        value={
+        value={<ExpandableChipList items={asBlocks} />}
-          asBlocks.length ? (
-            <span className="chip-list">
-              {asBlocks.slice(0, 12).map((b, i) => (
-                <span className="chip" key={i}>{typeof b === "string" || typeof b === "number" ? String(b) : JSON.stringify(b)}</span>
-              ))}
-              {asBlocks.length > 12 ? <span className="chip">+{asBlocks.length - 12} more</span> : null}
-            </span>
-          ) : (
-            "—"
-          )
-        }
       />
       <MetaRow label="Signature algorithm" value={asString(cert.signatureAlgorithm)} mono />
     </dl>
@@ -257,12 +432,16 @@ export function ProjectionView({
 
   const projection = record.projection;
   const type = (record.objectType ?? "").toLowerCase();
+  // The query service nests the typed payload one level down:
+  //   projection = { input, object: { roa | manifest | crl | certificate | aspa, … }, schemaVersion, tool }
+  // Accept a flat payload too, so older fixtures / shape variants keep working.
+  const typed = asRecord(projection.object) ?? projection;
 
-  if (type === "roa") return <RoaView projection={projection} />;
+  if (type === "roa") return <RoaView projection={typed} />;
-  if (type === "asa" || type === "aspa") return <AspaView projection={projection} />;
+  if (type === "asa" || type === "aspa") return <AspaView projection={typed} />;
 
   if (type === "mft") {
-    const manifest = asRecord(projection.manifest);
+    const manifest = asRecord(typed.manifest);
     return (
       <>
         <dl className="meta-grid" style={{ marginBottom: 16 }}>
@@ -278,15 +457,37 @@ export function ProjectionView({
   }
 
   if (type === "crl") {
-    const crl = asRecord(projection.crl);
+    // Real API shape: CRL fields sit directly on the typed payload
+    // ({ issuer, thisUpdate, nextUpdate, extensions: { crlNumberHex, authorityKeyIdentifier }, … }),
+    // there is no `crl` wrapper; keep `typed.crl` as a legacy fallback.
+    const crl = asRecord(typed.crl) ?? typed;
+    const crlExt = asRecord(crl.extensions);
     return (
       <>
         <dl className="meta-grid" style={{ marginBottom: 16 }}>
-          <MetaRow label="Issuer" value={asString(crl?.issuer)} mono />
+          <MetaRow label="Issuer" value={asString(crl.issuer)} mono />
-          <MetaRow label="thisUpdate" value={formatUtc(asString(crl?.thisUpdate))} />
+          <MetaRow label="thisUpdate" value={formatUtc(asString(crl.thisUpdate))} />
-          <MetaRow label="nextUpdate" value={formatUtc(asString(crl?.nextUpdate))} />
+          <MetaRow label="nextUpdate" value={formatUtc(asString(crl.nextUpdate))} />
-          <MetaRow label="CRL number" value={asString(crl?.crlNumberHex ?? crl?.crlNumber)} mono />
+          <MetaRow
-          <MetaRow label="AKI" value={<CopyableValue value={asString(crl?.aki)} max={56} label="AKI" />} />
+            label="CRL number"
+            value={
+              asString(crl.crlNumberHex ?? crlExt?.crlNumberHex) ??
+              (asNumber(crl.crlNumber ?? crlExt?.crlNumber) != null
+                ? String(asNumber(crl.crlNumber ?? crlExt?.crlNumber))
+                : null)
+            }
+            mono
+          />
+          <MetaRow
+            label="AKI"
+            value={
+              <CopyableValue
+                value={asString(crl.aki) ?? asString(crlExt?.authorityKeyIdentifier)}
+                max={56}
+                label="AKI"
+              />
+            }
+          />
         </dl>
         <RevokedCerts runId={runId} objectInstanceId={objectInstanceId} />
       </>
@@ -294,12 +495,12 @@ export function ProjectionView({
   }
 
   if (type === "cer" || type === "ee" || type === "router_cert") {
-    return <CertificateView projection={projection} />;
+    return <CertificateView projection={typed} />;
   }
 
   return (
     <pre className="json-block" data-testid="projection-json">
-      {JSON.stringify(projection, null, 2)}
+      {JSON.stringify(typed, null, 2)}
     </pre>
   );
 }

ui/rpki-explorer/src/components/Shell.tsx

@@ -59,6 +59,9 @@ function RunSelector() {
     queryKey: ["runs", "selector"],
     queryFn: () => listRuns({ limit: 25 }),
     staleTime: 60_000,
+    // Soak runs keep producing new runs; poll so the dropdown picks them up
+    // without a manual page refresh.
+    refetchInterval: 30_000,
   });
   const healthQuery = useQuery({ queryKey: ["health"], queryFn: getHealth });
   const latestId = healthQuery.data?.latestReadyRun ?? runsQuery.data?.items[0]?.runId;
@@ -129,6 +132,9 @@ export function Shell({ children }: { children: ReactNode }) {
 
   return (
     <div className={`shell${collapsed ? " collapsed" : ""}`}>
+      <a href="#main-content" className="skip-link">
+        Skip to content
+      </a>
       <div className="shell-brand">
         <span className="brand-badge" aria-hidden="true">
           <ShieldCheck size={15} />
@@ -170,7 +176,9 @@ export function Shell({ children }: { children: ReactNode }) {
         <div className="nav-footer">ours RP query layer</div>
       </nav>
 
-      <main className="shell-main">{children}</main>
+      <main className="shell-main" id="main-content" tabIndex={-1}>
+        {children}
+      </main>
     </div>
   );
 }

ui/rpki-explorer/src/components/Tabs.tsx

@@ -49,7 +49,9 @@ export function Tabs({
             role="tab"
             id={`tab-${tab.id}`}
             aria-selected={selected}
-            aria-controls={`tabpanel-${tab.id}`}
+            // Only the active tab's panel is mounted — pointing aria-controls
+            // at an unmounted panel would leave a dangling reference.
+            aria-controls={selected ? `tabpanel-${tab.id}` : undefined}
             tabIndex={selected ? 0 : -1}
             onClick={() => onChange(tab.id)}
           >

ui/rpki-explorer/src/lib/cursor.ts

@@ -1,5 +1,6 @@
 /** Cursor pagination state (previous-page stack) shared by all paged tables. */
-import { useCallback, useEffect, useMemo, useState } from "react";
+import { useCallback, useEffect, useMemo, useRef, useState } from "react";
+import { useSearchParams } from "react-router-dom";
 
 export interface CursorPager {
   /** Cursor to pass to the backend for the current page (null = first page). */
@@ -40,33 +41,72 @@ export function pageNumber(state: PagerState): number {
 
 /**
  * Cursor pager that resets whenever `resetKey` changes (run id, filters…).
+ *
+ * The current cursor lives in the `?cursor=` URL param so a refresh or a
+ * shared link restores the current page; the previous-page stack stays in
+ * component state (Prev works within one mounted session). Only one pager
+ * is mounted per page, so a single shared param is unambiguous.
  */
 export function useCursorPager(resetKey: string): CursorPager {
-  const [state, setState] = useState<PagerState>(INITIAL_PAGER_STATE);
+  const [searchParams, setSearchParams] = useSearchParams();
+  const cursor = searchParams.get("cursor");
+  const [stack, setStack] = useState<(string | null)[]>([]);
 
+  const writeCursor = useCallback(
+    (next: string | null) => {
+      setSearchParams(
+        (prev) => {
+          const params = new URLSearchParams(prev);
+          if (next === null) params.delete("cursor");
+          else params.set("cursor", next);
+          return params;
+        },
+        { preventScrollReset: true },
+      );
+    },
+    [setSearchParams],
+  );
+
+  // Reset on resetKey change — but not on first mount, so a refreshed or
+  // deep-linked `?cursor=` survives.
+  const lastResetKey = useRef(resetKey);
   useEffect(() => {
-    setState(INITIAL_PAGER_STATE);
+    if (lastResetKey.current === resetKey) return;
-  }, [resetKey]);
+    lastResetKey.current = resetKey;
+    setStack([]);
+    writeCursor(null);
+  }, [resetKey, writeCursor]);
 
-  const goNext = useCallback((nextCursor: string | null) => {
+  const goNext = useCallback(
-    setState((s) => advancePager(s, nextCursor));
+    (nextCursor: string | null) => {
-  }, []);
+      if (!nextCursor) return;
+      setStack((s) => [...s, cursor]);
+      writeCursor(nextCursor);
+    },
+    [cursor, writeCursor],
+  );
 
   const goPrev = useCallback(() => {
-    setState((s) => retreatPager(s));
+    if (stack.length === 0) return;
-  }, []);
+    const prev = stack[stack.length - 1];
+    setStack(stack.slice(0, -1));
+    writeCursor(prev);
+  }, [stack, writeCursor]);
 
-  const reset = useCallback(() => setState(INITIAL_PAGER_STATE), []);
+  const reset = useCallback(() => {
+    setStack([]);
+    writeCursor(null);
+  }, [writeCursor]);
 
   return useMemo(
     () => ({
-      cursor: state.cursor,
+      cursor,
-      page: pageNumber(state),
+      page: stack.length + 1,
-      canPrev: state.stack.length > 0,
+      canPrev: stack.length > 0,
       goNext,
       goPrev,
       reset,
     }),
-    [state, goNext, goPrev, reset],
+    [cursor, stack, goNext, goPrev, reset],
   );
 }

ui/rpki-explorer/src/lib/format.test.ts

@@ -69,6 +69,14 @@ describe("formatPercent", () => {
   it("guards zero total", () => {
     expect(formatPercent(1, 0)).toBe("—");
   });
+  it("marks tiny non-zero shares instead of rounding to 0.0%", () => {
+    expect(formatPercent(1, 5000)).toBe("<0.1%");
+    expect(formatPercent(0, 5000)).toBe("0.0%");
+  });
+  it("marks near-complete shares instead of rounding to 100.0%", () => {
+    expect(formatPercent(4999, 5000)).toBe(">99.9%");
+    expect(formatPercent(5000, 5000)).toBe("100.0%");
+  });
 });
 
 describe("truncateMiddle", () => {
@@ -90,6 +98,10 @@ describe("objectTypeLabel", () => {
     expect(objectTypeLabel("mft")).toBe("Manifest");
     expect(objectTypeLabel("asa")).toBe("ASPA");
   });
+  it("maps router certificate spellings", () => {
+    expect(objectTypeLabel("router_certificate")).toBe("Router certificate");
+    expect(objectTypeLabel("router_cert")).toBe("Router certificate");
+  });
   it("uppercases unknown types", () => {
     expect(objectTypeLabel("xyz")).toBe("XYZ");
   });

ui/rpki-explorer/src/lib/format.ts

@@ -70,7 +70,12 @@ export function formatRelative(value: string | null | undefined): string {
 
 export function formatPercent(part: number, total: number): string {
   if (!total) return "—";
-  return `${((part / total) * 100).toFixed(1)}%`;
+  const pct = (part / total) * 100;
+  // One decimal is misleading at the extremes: a non-zero share rounding to
+  // "0.0%" looks empty, and 99.95% rounding to "100.0%" looks complete.
+  if (pct > 0 && pct < 0.1) return "<0.1%";
+  if (pct < 100 && pct > 99.9) return ">99.9%";
+  return `${pct.toFixed(1)}%`;
 }
 
 /** Truncate a long identifier for display, keeping head and tail. */
@@ -101,6 +106,9 @@ export function objectTypeLabel(type: string | null | undefined): string {
       return "Ghostbusters";
     case "ee":
       return "EE Certificate";
+    case "router_certificate":
+    case "router_cert":
+      return "Router certificate";
     case "other":
       return "Other";
     default:

ui/rpki-explorer/src/lib/projection.test.ts

@@ -1,5 +1,6 @@
 import { describe, expect, it } from "vitest";
-import { asArray, asNumber, asRecord, asString } from "./projection";
+import type { ProjectionRecord } from "../api/schemas";
+import { asArray, asNumber, asRecord, asString, eeCertificateProjection } from "./projection";
 
 describe("projection guards", () => {
   it("asRecord accepts plain objects only", () => {
@@ -23,3 +24,48 @@ describe("projection guards", () => {
     expect(asArray({})).toEqual([]);
   });
 });
+
+describe("eeCertificateProjection", () => {
+  const eeCert = { kind: "Ee", subject: "CN=ee" };
+  const recordWith = (object: unknown): ProjectionRecord =>
+    ({ parseStatus: "ok", projection: { object } }) as unknown as ProjectionRecord;
+
+  it("extracts the Ee certificate from the CMS wrapper", () => {
+    const record = recordWith({
+      type: "roa",
+      signedObject: {
+        signedData: {
+          certificates: [
+            { resourceCertificate: { kind: "Ca", subject: "CN=ca" } },
+            { resourceCertificate: eeCert },
+          ],
+        },
+      },
+    });
+    expect(eeCertificateProjection(record)).toEqual({ resourceCertificate: eeCert });
+  });
+
+  it("falls back to the first certificate when no Ee kind is marked", () => {
+    const record = recordWith({
+      signedObject: { signedData: { certificates: [{ resourceCertificate: eeCert }] } },
+    });
+    expect(eeCertificateProjection(record)).toEqual({ resourceCertificate: eeCert });
+  });
+
+  it("returns null for non-CMS or malformed projections", () => {
+    expect(eeCertificateProjection(null)).toBeNull();
+    expect(eeCertificateProjection(recordWith({ type: "crl" }))).toBeNull();
+    expect(eeCertificateProjection(recordWith(null))).toBeNull();
+    expect(
+      eeCertificateProjection(recordWith({ signedObject: { signedData: { certificates: [] } } })),
+    ).toBeNull();
+  });
+
+  it("returns null when the projection failed to parse", () => {
+    const record = {
+      parseStatus: "error",
+      projection: { object: { signedObject: { signedData: { certificates: [{ resourceCertificate: eeCert }] } } } },
+    } as unknown as ProjectionRecord;
+    expect(eeCertificateProjection(record)).toBeNull();
+  });
+});

ui/rpki-explorer/src/lib/projection.ts

@@ -1,4 +1,6 @@
 /** Defensive accessors for loosely-typed projection payloads. */
+import type { ProjectionRecord } from "../api/schemas";
+
 export function asRecord(value: unknown): Record<string, unknown> | null {
   return value !== null && typeof value === "object" && !Array.isArray(value)
     ? (value as Record<string, unknown>)
@@ -16,3 +18,25 @@ export function asNumber(value: unknown): number | null {
 export function asArray(value: unknown): unknown[] {
   return Array.isArray(value) ? value : [];
 }
+
+/**
+ * Extract the CMS wrapper's EE certificate from a parsed projection, if any.
+ * Real shape: projection.object.signedObject.signedData.certificates[] — each
+ * entry carries a `resourceCertificate` with `kind: "Ee"` for the signer cert.
+ * Returns a `{ resourceCertificate }` projection for CertificateView, or null
+ * when the object is not CMS-signed or has no embedded certificate.
+ */
+export function eeCertificateProjection(
+  record: ProjectionRecord | null,
+): Record<string, unknown> | null {
+  const projection = record?.projection;
+  if (!projection || record?.parseStatus === "error") return null;
+  const typed = asRecord(projection.object) ?? projection;
+  const certificates = asArray(asRecord(asRecord(typed.signedObject)?.signedData)?.certificates);
+  const certs = certificates
+    .map((entry) => asRecord(asRecord(entry)?.resourceCertificate))
+    .filter((cert): cert is Record<string, unknown> => Boolean(cert));
+  if (!certs.length) return null;
+  const ee = certs.find((cert) => asString(cert.kind) === "Ee") ?? certs[0];
+  return { resourceCertificate: ee };
+}

ui/rpki-explorer/src/lib/reasons.test.ts

@@ -0,0 +1,39 @@
+import { describe, expect, it } from "vitest";
+import { clusterReasons, reasonClusterKey } from "./reasons";
+
+describe("reasonClusterKey", () => {
+  it("takes the text before the first colon", () => {
+    expect(reasonClusterKey("manifest stale: thisUpdate in the past")).toBe("manifest stale");
+    expect(
+      reasonClusterKey("manifest is not valid at validation_time: this_update=2023-01-01"),
+    ).toBe("manifest is not valid at validation_time");
+  });
+  it("keeps reasons without a colon intact", () => {
+    expect(reasonClusterKey("certificate expired")).toBe("certificate expired");
+  });
+  it("trims whitespace around the prefix", () => {
+    expect(reasonClusterKey("bad roa : detail")).toBe("bad roa");
+  });
+});
+
+describe("clusterReasons", () => {
+  it("merges reasons sharing a prefix and sums counts", () => {
+    const clusters = clusterReasons({
+      "manifest stale: thisUpdate in the past": 3,
+      "manifest stale: nextUpdate missing": 2,
+      "certificate expired: notAfter 2026-01-01": 5,
+    });
+    expect(clusters).toEqual([
+      { key: "certificate expired", count: 5, variants: 1 },
+      { key: "manifest stale", count: 5, variants: 2 },
+    ]);
+  });
+  it("sorts by count desc, then key asc for determinism", () => {
+    const clusters = clusterReasons({ "b: 1": 4, "a: 2": 4, "c": 1 });
+    expect(clusters.map((c) => c.key)).toEqual(["a", "b", "c"]);
+  });
+  it("honours the limit", () => {
+    const clusters = clusterReasons({ "a: 1": 3, "b: 2": 2, "c: 3": 1 }, 2);
+    expect(clusters.map((c) => c.key)).toEqual(["a", "b"]);
+  });
+});

ui/rpki-explorer/src/lib/reasons.ts

@@ -0,0 +1,39 @@
+/**
+ * Reject-reason clustering for the overview dashboard.
+ *
+ * Raw reject reasons embed object-specific details after a colon
+ * (e.g. "manifest is not valid at validation_time: this_update=2023-…"),
+ * which scatters one logical cause across many rows. Grouping by the text
+ * before the first colon collapses them back into a single cluster.
+ */
+
+/** Cluster label for a raw reject reason: text before the first colon. */
+export function reasonClusterKey(reason: string): string {
+  const idx = reason.indexOf(":");
+  return (idx === -1 ? reason : reason.slice(0, idx)).trim();
+}
+
+export interface ReasonCluster {
+  /** Cluster label (prefix before the first colon, or the whole reason). */
+  key: string;
+  /** Total rejected objects across all reasons in the cluster. */
+  count: number;
+  /** Distinct raw reasons merged into the cluster. */
+  variants: number;
+}
+
+/** Aggregate a reason→count map into clusters, sorted by count desc. */
+export function clusterReasons(map: Record<string, number>, limit = 8): ReasonCluster[] {
+  const clusters = new Map<string, { count: number; variants: number }>();
+  for (const [reason, count] of Object.entries(map)) {
+    const key = reasonClusterKey(reason);
+    const entry = clusters.get(key) ?? { count: 0, variants: 0 };
+    entry.count += count;
+    entry.variants += 1;
+    clusters.set(key, entry);
+  }
+  return [...clusters.entries()]
+    .map(([key, { count, variants }]) => ({ key, count, variants }))
+    .sort((a, b) => b.count - a.count || a.key.localeCompare(b.key))
+    .slice(0, limit);
+}

ui/rpki-explorer/src/lib/useExportJob.ts

@@ -0,0 +1,34 @@
+/**
+ * Shared "start export job + poll until finished" flow used by the object,
+ * repository and publication point detail pages. The page renders the
+ * returned mutation/query state with `WorkflowStatus` and friends.
+ */
+import { useState } from "react";
+import { useMutation, useQuery } from "@tanstack/react-query";
+import { createExport, getExportJob, type ExportRequest } from "../api/service";
+
+export function useExportJob(runId: string, request: ExportRequest) {
+  const [jobId, setJobId] = useState<string | null>(null);
+
+  const startMutation = useMutation({
+    mutationFn: () => createExport(runId, request),
+    onSuccess: (job) => setJobId(job.jobId),
+  });
+
+  const jobQuery = useQuery({
+    queryKey: ["export-job", runId, jobId],
+    queryFn: () => getExportJob(runId, jobId!),
+    enabled: jobId !== null,
+    refetchInterval: (query) => (query.state.data?.status === "running" ? 2000 : false),
+  });
+
+  // Keep the trigger disabled until the job reaches a terminal state so the
+  // same export cannot be double-submitted while it is still running.
+  const running =
+    startMutation.isPending ||
+    (jobId !== null &&
+      jobQuery.data?.status !== "complete" &&
+      jobQuery.data?.status !== "failed");
+
+  return { startMutation, jobQuery, running };
+}

ui/rpki-explorer/src/lib/useRun.ts

@@ -1,8 +1,9 @@
 /** Resolve the active run record for the current `?run=` context. */
 import { useQuery, type UseQueryResult } from "@tanstack/react-query";
+import { ApiError } from "../api/client";
 import { getRun } from "../api/service";
 import type { RunRecord } from "../api/schemas";
-import { useRunId } from "./run";
+import { LATEST_RUN, useRunId } from "./run";
 
 export function useRun(): { runId: string; runQuery: UseQueryResult<RunRecord> } {
   const runId = useRunId();
@@ -13,3 +14,16 @@ export function useRun(): { runId: string; runQuery: UseQueryResult<RunRecord> }
   });
   return { runId, runQuery };
 }
+
+/**
+ * Run-aware title for the run-scope error gate: a 404 means the requested
+ * run does not exist; anything else is a service/transport failure.
+ */
+export function runErrorTitle(runId: string, error: unknown): string {
+  if (error instanceof ApiError && error.status === 404) {
+    return runId === LATEST_RUN
+      ? "No indexed run available yet"
+      : `Run ${runId} does not exist`;
+  }
+  return `Failed to load run ${runId}`;
+}

ui/rpki-explorer/src/lib/vrpQuery.test.ts

@@ -94,4 +94,34 @@ describe("classifyNetworkQuery", () => {
     expect(classifyNetworkQuery("d1488eb0e0faabba1fbe8236")).toEqual({ kind: "none" });
     expect(classifyNetworkQuery("")).toEqual({ kind: "none" });
   });
+
+  it("reads two tokens as network + ASN in either order", () => {
+    expect(classifyNetworkQuery("1.0.0.1 AS13335")).toEqual({
+      kind: "ip",
+      ip: "1.0.0.1",
+      asn: 13335,
+    });
+    expect(classifyNetworkQuery("AS13335 1.0.0.1")).toEqual({
+      kind: "ip",
+      ip: "1.0.0.1",
+      asn: 13335,
+    });
+    expect(classifyNetworkQuery("192.0.2.0/24 13335")).toEqual({
+      kind: "prefix",
+      prefix: "192.0.2.0/24",
+      asn: 13335,
+    });
+    expect(classifyNetworkQuery("  2001:db8::/32   as64496 ")).toEqual({
+      kind: "prefix",
+      prefix: "2001:db8::/32",
+      asn: 64496,
+    });
+  });
+
+  it("rejects multi-token queries without a network+ASN pair", () => {
+    expect(classifyNetworkQuery("13335 64496")).toEqual({ kind: "none" });
+    expect(classifyNetworkQuery("1.0.0.1 192.0.2.1")).toEqual({ kind: "none" });
+    expect(classifyNetworkQuery("1.0.0.1 AS13335 extra")).toEqual({ kind: "none" });
+    expect(classifyNetworkQuery("foo bar")).toEqual({ kind: "none" });
+  });
 });

ui/rpki-explorer/src/lib/vrpQuery.ts

@@ -7,8 +7,8 @@
  */
 
 export type NetworkQuery =
-  | { kind: "ip"; ip: string }
+  | { kind: "ip"; ip: string; asn?: number }
-  | { kind: "prefix"; prefix: string }
+  | { kind: "prefix"; prefix: string; asn?: number }
   | { kind: "asn"; asn: number }
   | { kind: "none" };
 
@@ -90,10 +90,30 @@ export function parsePrefix(input: string): string | null {
  * Classify a query string. Precedence: CIDR prefix → bare IP → ASN → none.
  * ("192.0.2.1" also parses as digits+ dots only; ASN requires pure digits, so
  * no overlap. A bare number is always treated as an ASN.)
+ *
+ * Two whitespace-separated tokens are read as a combined query — one network
+ * token (IP or prefix) plus one ASN token (`AS13335` or `13335`), in either
+ * order — yielding an ip/prefix query carrying an `asn` filter.
  */
 export function classifyNetworkQuery(input: string): NetworkQuery {
   const s = input.trim();
   if (!s) return { kind: "none" };
+  const tokens = s.split(/\s+/);
+  if (tokens.length === 2) {
+    for (const [netToken, asnToken] of [
+      [tokens[0], tokens[1]],
+      [tokens[1], tokens[0]],
+    ] as const) {
+      const asn = parseAsn(asnToken);
+      if (asn === null) continue;
+      const prefix = parsePrefix(netToken);
+      if (prefix) return { kind: "prefix", prefix, asn };
+      const ip = parseIpAddress(netToken);
+      if (ip) return { kind: "ip", ip, asn };
+    }
+    return { kind: "none" };
+  }
+  if (tokens.length > 2) return { kind: "none" };
   const prefix = parsePrefix(s);
   if (prefix) return { kind: "prefix", prefix };
   const ip = parseIpAddress(s);

ui/rpki-explorer/src/pages/ApiStatusPage.tsx

@@ -80,7 +80,6 @@ export default function ApiStatusPage() {
           <li>The query service has no CORS headers and no authentication — always front it with a same-origin proxy (this SPA) or an API gateway.</li>
           <li><code>/parsed</code>, <code>/raw</code> and exports require <code>--repo-bytes-db</code>.</li>
           <li>All list endpoints paginate with <code>limit</code> + <code>cursor</code>; the envelope is <code>{"{data, page, meta}"}</code>.</li>
-          <li>VRP IP/prefix/ASN lookup is tracked as backend feature #070 and is not available yet.</li>
         </ul>
       </Panel>
 

ui/rpki-explorer/src/pages/ExportsPage.tsx

@@ -5,16 +5,16 @@ import { DataTable, type Column } from "../components/DataTable";
 import { PageHeader } from "../components/PageHeader";
 import { Panel } from "../components/Panel";
 import { StatusPill } from "../components/StatusPill";
-import { Notice } from "../components/StateBlock";
+import { ErrorBlock, Notice } from "../components/StateBlock";
 import { formatBytes, formatInt, formatRelative, formatUtc } from "../lib/format";
-import { useRun } from "../lib/useRun";
+import { useRun, runErrorTitle } from "../lib/useRun";
 
 /**
  * Export job history for the active run. Jobs are held in the query service
  * process memory, so the list resets when the service restarts.
  */
 export default function ExportsPage() {
-  const { runId } = useRun();
+  const { runId, runQuery } = useRun();
 
   const exportsQuery = useQuery({
     queryKey: ["exports", runId],
@@ -25,6 +25,19 @@ export default function ExportsPage() {
     },
   });
 
+  if (runQuery.isError) {
+    return (
+      <div className="page">
+        <PageHeader title="Exports" />
+        <ErrorBlock
+          error={runQuery.error}
+          onRetry={() => runQuery.refetch()}
+          title={runErrorTitle(runId, runQuery.error)}
+        />
+      </div>
+    );
+  }
+
   const columns: Column<ExportJobRecord>[] = [
     {
       key: "job",

ui/rpki-explorer/src/pages/ObjectDetailPage.tsx

@@ -4,9 +4,7 @@ import { useMutation, useQuery } from "@tanstack/react-query";
 import { Download, PackageOpen, Play, RefreshCw } from "lucide-react";
 import { apiFetchBlob, saveBlob } from "../api/client";
 import {
-  createExport,
   explainObjectValidation,
-  getExportJob,
   getObject,
   getObjectChain,
   getObjectProjection,
@@ -18,13 +16,15 @@ import { CopyableValue } from "../components/CopyableValue";
 import { DataTable, type Column } from "../components/DataTable";
 import { PageHeader } from "../components/PageHeader";
 import { Panel } from "../components/Panel";
-import { ProjectionView } from "../components/ProjectionView";
+import { CertificateView, ProjectionView } from "../components/ProjectionView";
 import { StatusPill } from "../components/StatusPill";
 import { EmptyBlock, ErrorBlock, LoadingBlock, Notice } from "../components/StateBlock";
 import { TabPanel, Tabs } from "../components/Tabs";
 import { WorkflowStatus } from "../components/WorkflowStatus";
-import { formatBytes, formatInt, objectTypeLabel } from "../lib/format";
+import { formatInt, objectTypeLabel } from "../lib/format";
+import { eeCertificateProjection } from "../lib/projection";
 import { withRunParam } from "../lib/run";
+import { useExportJob } from "../lib/useExportJob";
 import { useRun } from "../lib/useRun";
 
 function filenameFor(uri: string | undefined, sha256: string | undefined, type: string): string {
@@ -207,17 +207,17 @@ export default function ObjectDetailPage() {
   const { runId } = useRun();
   const [searchParams, setSearchParams] = useSearchParams();
   const tab = searchParams.get("tab") ?? "parsed";
-  const [exportJobId, setExportJobId] = useState<string | null>(null);
 
   const objectQuery = useQuery({
     queryKey: ["object", runId, objectInstanceId],
     queryFn: () => getObject(runId, objectInstanceId),
   });
 
+  // Always fetch the parsed projection: the Parsed tab needs it, and the
+  // conditional "EE Parsed" tab derives from it (CMS-signed objects only).
   const projectionQuery = useQuery({
     queryKey: ["object-parsed", runId, objectInstanceId],
     queryFn: () => getObjectProjection(runId, objectInstanceId),
-    enabled: tab === "parsed",
     staleTime: 5 * 60_000,
   });
 
@@ -229,17 +229,11 @@ export default function ObjectDetailPage() {
     },
   });
 
-  const exportMutation = useMutation({
+  const {
-    mutationFn: () => createExport(runId, { scope: "object_set", objectInstanceIds: [objectInstanceId] }),
+    startMutation: exportMutation,
-    onSuccess: (job) => setExportJobId(job.jobId),
+    jobQuery: exportJobQuery,
-  });
+    running: exportRunning,
-
+  } = useExportJob(runId, { scope: "object_set", objectInstanceIds: [objectInstanceId] });
-  const exportJobQuery = useQuery({
-    queryKey: ["export-job", runId, exportJobId],
-    queryFn: () => getExportJob(runId, exportJobId!),
-    enabled: exportJobId !== null,
-    refetchInterval: (query) => (query.state.data?.status === "running" ? 2000 : false),
-  });
 
   if (objectQuery.isError) {
     return (
@@ -259,6 +253,7 @@ export default function ObjectDetailPage() {
   }
 
   const object = objectQuery.data;
+  const eeProjection = eeCertificateProjection(projectionQuery.data ?? null);
 
   const setTab = (next: string) => {
     const params = new URLSearchParams(searchParams);
@@ -298,7 +293,7 @@ export default function ObjectDetailPage() {
               type="button"
               className="btn"
               onClick={() => exportMutation.mutate()}
-              disabled={exportMutation.isPending || exportJobQuery.data?.status === "running"}
+              disabled={exportRunning}
             >
               {exportMutation.isPending ? <RefreshCw size={13} aria-hidden="true" /> : <PackageOpen size={13} aria-hidden="true" />}
               Export object set
@@ -317,6 +312,12 @@ export default function ObjectDetailPage() {
         <Notice kind="error">Export failed to start: {exportMutation.error.message}</Notice>
       ) : null}
       {exportJobQuery.data ? <WorkflowStatus job={exportJobQuery.data} runId={runId} /> : null}
+      {exportJobQuery.data?.status === "complete" ? (
+        <Notice kind="info">
+          Export complete — {formatInt(exportJobQuery.data.objectCount)} objects. Find it any
+          time on the <Link to={withRunParam("/exports", runId)}>Exports page</Link>.
+        </Notice>
+      ) : null}
 
       <Panel className="object-header-card">
         <dl className="meta-grid">
@@ -334,8 +335,6 @@ export default function ObjectDetailPage() {
           <dd>{object.rejectReason ?? <span className="text-faint">—</span>}</dd>
           <dt>Detail</dt>
           <dd>{object.detailSummary ?? <span className="text-faint">—</span>}</dd>
-          <dt>Size</dt>
-          <dd>{formatBytes(object.sizeBytes)}</dd>
           <dt>Repository</dt>
           <dd>
             <Link className="mono" to={withRunParam(`/repositories/${encodeURIComponent(object.repoId)}`, runId)}>
@@ -355,6 +354,7 @@ export default function ObjectDetailPage() {
         <Tabs
           tabs={[
             { id: "parsed", label: "Parsed" },
+            ...(eeProjection ? [{ id: "ee", label: "EE Parsed" }] : []),
             { id: "validation", label: "Validation" },
             { id: "chain", label: "Chain" },
           ]}
@@ -381,6 +381,17 @@ export default function ObjectDetailPage() {
             )}
           </div>
         </TabPanel>
+        {eeProjection ? (
+          <TabPanel id="ee" active={tab}>
+            <div style={{ padding: 16 }}>
+              <CertificateView projection={eeProjection} />
+            </div>
+          </TabPanel>
+        ) : tab === "ee" ? (
+          <div style={{ padding: 16 }}>
+            <Notice kind="info">This object has no embedded EE certificate in its CMS wrapper.</Notice>
+          </div>
+        ) : null}
         <TabPanel id="validation" active={tab}>
           <ValidationTab runId={runId} objectInstanceId={objectInstanceId} />
         </TabPanel>

ui/rpki-explorer/src/pages/ObjectsPage.tsx

@@ -4,6 +4,8 @@ import { listObjects } from "../api/service";
 import { Notice } from "../components/StateBlock";
 import {
   EMPTY_OBJECT_FILTERS,
+  objectFiltersFromParams,
+  objectFiltersToParams,
   ObjectsTable,
   type ObjectFilterValues,
 } from "../components/ObjectsTable";
@@ -11,37 +13,19 @@ import { PageHeader } from "../components/PageHeader";
 import { Panel } from "../components/Panel";
 import { useRun } from "../lib/useRun";
 
-function filtersFromParams(params: URLSearchParams): ObjectFilterValues {
-  return {
-    type: params.get("type") ?? "",
-    result: params.get("result") ?? "",
-    rejectedOnly: params.get("rejected") === "true",
-    q: params.get("q") ?? "",
-  };
-}
-
 export default function ObjectsPage() {
   const { runId } = useRun();
   const [searchParams, setSearchParams] = useSearchParams();
-  const filters = useMemo(() => filtersFromParams(searchParams), [searchParams]);
+  const filters = useMemo(() => objectFiltersFromParams(searchParams), [searchParams]);
 
   const setFilters = (next: ObjectFilterValues) => {
-    const params = new URLSearchParams(searchParams);
+    setSearchParams(objectFiltersToParams(next, searchParams), { preventScrollReset: true });
-    const write = (key: string, value: string) => {
-      if (value) params.set(key, value);
-      else params.delete(key);
-    };
-    write("type", next.type);
-    write("result", next.result);
-    write("q", next.q);
-    if (next.rejectedOnly) params.set("rejected", "true");
-    else params.delete("rejected");
-    setSearchParams(params, { preventScrollReset: true });
   };
 
   const hasFilters =
     filters.type !== EMPTY_OBJECT_FILTERS.type ||
     filters.result !== "" ||
+    filters.reason !== "" ||
     filters.rejectedOnly ||
     filters.q !== "";
 

ui/rpki-explorer/src/pages/OverviewPage.tsx

@@ -15,6 +15,7 @@ import {
 import { Clock, GitBranch, Timer } from "lucide-react";
 import {
   getStatsObjectTypes,
+  getStatsOverview,
   getStatsReasons,
   getStatsValidation,
   listRepos,
@@ -35,8 +36,9 @@ import {
   objectTypeLabel,
   truncateMiddle,
 } from "../lib/format";
+import { clusterReasons } from "../lib/reasons";
 import { withRunParam } from "../lib/run";
-import { useRun } from "../lib/useRun";
+import { useRun, runErrorTitle } from "../lib/useRun";
 
 /** Semantic colors keyed by validation result — never assigned by index. */
 const RESULT_COLORS: Record<string, string> = {
@@ -61,7 +63,7 @@ export default function OverviewPage() {
   const { runId, runQuery } = useRun();
   const navigate = useNavigate();
 
-  const [validationQuery, typesQuery, reasonsQuery, reposQuery] = useQueries({
+  const [validationQuery, typesQuery, reasonsQuery, reposQuery, overviewQuery] = useQueries({
     queries: [
       {
         queryKey: ["stats", runId, "validation"],
@@ -80,7 +82,12 @@ export default function OverviewPage() {
       },
       {
         queryKey: ["repos", runId, "top"],
-        queryFn: () => listRepos(runId, { limit: 8 }),
+        queryFn: () => listRepos(runId, { limit: 8, sort: "objects", order: "desc" }),
+        staleTime: 60_000,
+      },
+      {
+        queryKey: ["stats", runId, "overview"],
+        queryFn: () => getStatsOverview(runId),
         staleTime: 60_000,
       },
     ],
@@ -109,13 +116,10 @@ export default function OverviewPage() {
       .slice(0, 8);
   }, [typesQuery.data]);
 
-  const reasonRows = useMemo(() => {
+  const reasonRows = useMemo(
-    const map = reasonsQuery.data ?? {};
+    () => clusterReasons(reasonsQuery.data ?? {}),
-    return Object.entries(map)
+    [reasonsQuery.data],
-      .map(([reason, count]) => ({ reason, count }))
+  );
-      .sort((a, b) => b.count - a.count)
-      .slice(0, 8);
-  }, [reasonsQuery.data]);
 
   const maxReason = reasonRows[0]?.count ?? 0;
 
@@ -126,7 +130,7 @@ export default function OverviewPage() {
         <ErrorBlock
           error={runQuery.error}
           onRetry={() => runQuery.refetch()}
-          title="Failed to load the latest run"
+          title={runErrorTitle(runId, runQuery.error)}
         />
       </div>
     );
@@ -170,6 +174,17 @@ export default function OverviewPage() {
       numeric: true,
       render: (repo) => formatDurationMs(repo.syncDurationMsTotal),
     },
+    {
+      key: "terminal",
+      header: "Terminal states",
+      render: (repo) => (
+        <span className="chip-list">
+          {Object.entries(repo.terminalStates ?? {}).map(([state, count]) => (
+            <StatusPill key={state} status={state} label={`${state} ${formatInt(count)}`} />
+          ))}
+        </span>
+      ),
+    },
   ];
 
   return (
@@ -213,6 +228,11 @@ export default function OverviewPage() {
               value={formatInt(counts?.publicationPoints)}
               to={withRunParam("/publication-points", runId)}
             />
+            <KpiCard
+              label="Repositories"
+              value={formatInt(overviewQuery.data?.repos)}
+              to={withRunParam("/repositories", runId)}
+            />
             <KpiCard
               label="Rejected"
               value={formatInt(counts?.rejectedObjects)}
@@ -224,6 +244,7 @@ export default function OverviewPage() {
               value={formatInt(counts?.warnings)}
               tone="amber"
               sub="objects with warnings"
+              to={withRunParam("/validation", runId)}
             />
           </div>
 
@@ -238,7 +259,7 @@ export default function OverviewPage() {
               ) : (
                 <>
                   <div className="chart-box">
-                    <ResponsiveContainer>
+                    <ResponsiveContainer initialDimension={{ width: 400, height: 240 }}>
                       <PieChart>
                         <Pie
                           data={validationData}
@@ -286,13 +307,13 @@ export default function OverviewPage() {
                 <p className="text-muted">No object type stats recorded.</p>
               ) : (
                 <div className="chart-box">
-                  <ResponsiveContainer>
+                  <ResponsiveContainer initialDimension={{ width: 400, height: 240 }}>
                     <BarChart data={typeData} layout="vertical" margin={{ left: 12, right: 24 }}>
                       <XAxis type="number" hide />
                       <YAxis
                         type="category"
                         dataKey="name"
-                        width={104}
+                        width={150}
                         tickLine={false}
                         axisLine={false}
                         tick={{ fontSize: 12, fill: "var(--text-2)" }}
@@ -313,7 +334,7 @@ export default function OverviewPage() {
           <div className="overview-bottom">
             <Panel
               title="Top repositories"
-              subtitle="First repositories by index order"
+              subtitle="Repositories with the most objects"
               tools={
                 <Link className="btn small" to={withRunParam("/repositories", runId)}>
                   View all
@@ -341,7 +362,7 @@ export default function OverviewPage() {
 
             <Panel
               title="Top reject reasons"
-              subtitle="Click a reason to inspect matching objects"
+              subtitle="Grouped by reason prefix — click to inspect matching objects"
               tools={
                 <Link className="btn small" to={withRunParam("/validation", runId)}>
                   Validation
@@ -358,12 +379,17 @@ export default function OverviewPage() {
                 <div className="reason-list">
                   {reasonRows.map((row) => (
                     <Link
-                      key={row.reason}
+                      key={row.key}
                       className="reason-row"
-                      to={withRunParam(`/validation?reason=${encodeURIComponent(row.reason)}`, runId)}
+                      to={withRunParam(`/validation?reason=${encodeURIComponent(row.key)}`, runId)}
-                      title={row.reason}
+                      title={row.key}
                     >
-                      <span className="reason-text">{row.reason}</span>
+                      <span className="reason-text">
+                        {row.key}
+                        {row.variants > 1 ? (
+                          <span className="text-faint text-small"> ×{row.variants}</span>
+                        ) : null}
+                      </span>
                       <span className="reason-count">{formatInt(row.count)}</span>
                       <span className="reason-bar" aria-hidden="true">
                         <span style={{ width: `${maxReason ? (row.count / maxReason) * 100 : 0}%` }} />