diff --git a/scripts/soak/build_portable_soak_package.sh b/scripts/soak/build_portable_soak_package.sh
index 0eac491..779d7ca 100755
--- a/scripts/soak/build_portable_soak_package.sh
+++ b/scripts/soak/build_portable_soak_package.sh
@@ -88,6 +88,9 @@ mkdir -p "$STAGE_DIR/bin" "$STAGE_DIR/fixtures" "$STAGE_DIR/scripts" "$STAGE_DIR
 install -m 0755 "$SCRIPT_DIR/run_soak.sh" "$STAGE_DIR/run_soak.sh"
 install -m 0755 "$SCRIPT_DIR/run_24h_soak_with_metrics.sh" "$STAGE_DIR/run_24h_soak_with_metrics.sh"
 install -m 0755 "$SCRIPT_DIR/run_cache_ablation_experiment.sh" "$STAGE_DIR/scripts/soak/run_cache_ablation_experiment.sh"
+install -m 0755 "$SCRIPT_DIR/run_feature149_cohort.sh" "$STAGE_DIR/scripts/soak/run_feature149_cohort.sh"
+install -m 0755 "$SCRIPT_DIR/summarize_feature149_cohort.py" "$STAGE_DIR/scripts/soak/summarize_feature149_cohort.py"
+install -m 0755 "$SCRIPT_DIR/render_feature149_report.py" "$STAGE_DIR/scripts/soak/render_feature149_report.py"
 install -m 0755 "$SCRIPT_DIR/fixed_phase_loop.sh" "$STAGE_DIR/scripts/soak/fixed_phase_loop.sh"
 install -m 0755 "$SCRIPT_DIR/hourly_soak_report.py" "$STAGE_DIR/scripts/soak/hourly_soak_report.py"
 install -m 0755 "$SCRIPT_DIR/publish_remote231.sh" "$STAGE_DIR/scripts/soak/publish_remote231.sh"
diff --git a/scripts/soak/render_feature149_report.py b/scripts/soak/render_feature149_report.py
new file mode 100644
index 0000000..f92607c
--- /dev/null
+++ b/scripts/soak/render_feature149_report.py
@@ -0,0 +1,390 @@
+#!/usr/bin/env python3
+"""Render a self-contained final report for Feature #149."""
+
+from __future__ import annotations
+
+import argparse
+import html
+import json
+import statistics
+from pathlib import Path
+from typing import Any
+
+
+EXPECTED_COHORTS = (
+    "prefetch_pp_object_system",
+    "prefetch_pp_object_data",
+    "pp_object_system",
+    "pp_object_data",
+)
+
+
+def number(value: Any, digits: int = 0) -> str:
+    if not isinstance(value, (int, float)):
+        return "n/a"
+    return f"{value:,.{digits}f}"
+
+
+def load(path: Path) -> dict[str, Any]:
+    value = json.loads(path.read_text(encoding="utf-8"))
+    if not isinstance(value, dict):
+        raise ValueError(f"{path} is not an object")
+    return value
+
+
+def latest(record: dict[str, Any], key: str) -> Any:
+    return (record.get("latestCounts") or {}).get(key)
+
+
+def stats(record: dict[str, Any], key: str) -> dict[str, Any]:
+    value = record.get(key) or {}
+    return value if isinstance(value, dict) else {}
+
+
+def delta_median(record: dict[str, Any]) -> float | None:
+    value = stats(record, "deltaWallMs").get("median")
+    return float(value) if isinstance(value, (int, float)) else None
+
+
+def stable_median(record: dict[str, Any]) -> float | None:
+    value = stats(record, "stableDeltaWallMs").get("median")
+    return float(value) if isinstance(value, (int, float)) else None
+
+
+def delta_mean(record: dict[str, Any]) -> float | None:
+    value = stats(record, "deltaWallMs").get("mean")
+    return float(value) if isinstance(value, (int, float)) else None
+
+
+def stable_mean(record: dict[str, Any]) -> float | None:
+    value = stats(record, "stableDeltaWallMs").get("mean")
+    return float(value) if isinstance(value, (int, float)) else None
+
+
+def percent_change(baseline: float | None, value: float | None) -> float | None:
+    if baseline in (None, 0) or value is None:
+        return None
+    return (value - baseline) / baseline * 100
+
+
+def percent_text(baseline: float | None, value: float | None) -> str:
+    change = percent_change(baseline, value)
+    return "n/a" if change is None else f"{change:+.1f}%"
+
+
+def stat_value(record: dict[str, Any], field: str, key: str) -> float | None:
+    value = stats(record, field).get(key)
+    return float(value) if isinstance(value, (int, float)) else None
+
+
+def svg_chart(records: list[dict[str, Any]]) -> str:
+    values = [delta_median(record) or 0 for record in records]
+    maximum = max(values) if values else 1
+    width, height, left, baseline = 860, 250, 62, 212
+    bar_width, gap = 112, 56
+    bars: list[str] = []
+    for index, (record, value) in enumerate(zip(records, values)):
+        x = left + index * (bar_width + gap)
+        bar_height = 0 if maximum == 0 else value / maximum * 164
+        y = baseline - bar_height
+        colour = "#0f766e" if record.get("disk") == "data" else "#2563eb"
+        label = html.escape(str(record.get("cohortId")))
+        bars.append(
+            f'<rect x="{x}" y="{y:.1f}" width="{bar_width}" height="{bar_height:.1f}" rx="7" fill="{colour}"/>'
+            f'<text x="{x + bar_width / 2:.1f}" y="{y - 8:.1f}" text-anchor="middle" class="value">{number(value / 1000, 2)}s</text>'
+            f'<text x="{x + bar_width / 2:.1f}" y="{baseline + 20}" text-anchor="middle" class="label">{label}</text>'
+        )
+    return f'''<svg viewBox="0 0 {width} {height}" role="img" aria-label="Delta median wall clock by cohort">
+      <line x1="{left}" y1="{baseline}" x2="{width - 28}" y2="{baseline}" stroke="#94a3b8"/>
+      <text x="8" y="24" class="axis">delta median wall clock</text>
+      {''.join(bars)}
+    </svg>'''
+
+
+def timeline_svg(records: list[dict[str, Any]]) -> str:
+    """Render the fixed serial order and the six run slots in each cohort."""
+    width, row_height, top = 930, 48, 36
+    height = top + row_height * len(records) + 18
+    parts = [
+        f'<svg viewBox="0 0 {width} {height}" role="img" aria-label="Serial cohort timeline">',
+        '<text x="8" y="18" class="axis">each row: snapshot (S) then five delta runs (D1–D5); adjacent starts are 600 s apart</text>',
+    ]
+    for row, record in enumerate(records):
+        y = top + row * row_height
+        label = html.escape(str(record.get("cohortId")))
+        parts.append(f'<text x="8" y="{y + 5}" class="label">{label}</text>')
+        parts.append(f'<line x1="310" y1="{y}" x2="890" y2="{y}" stroke="#cbd5e1" stroke-width="2"/>')
+        for index in range(6):
+            x = 310 + index * 116
+            mode = "S" if index == 0 else f"D{index}"
+            fill = "#2563eb" if index == 0 else "#0f766e"
+            parts.append(f'<circle cx="{x}" cy="{y}" r="11" fill="{fill}"/>')
+            parts.append(f'<text x="{x}" y="{y + 4}" text-anchor="middle" class="circle-label">{mode}</text>')
+    parts.append("</svg>")
+    return "".join(parts)
+
+
+def milestone_svg() -> str:
+    labels = (
+        ("M1", "data disk init"),
+        ("M2", "package + dry-run"),
+        ("M3", "prefetch cohorts"),
+        ("M4", "no-prefetch cohorts"),
+        ("M5", "evidence + report"),
+    )
+    width, y = 900, 44
+    parts = [f'<svg viewBox="0 0 {width} 90" role="img" aria-label="Milestone timeline"><line x1="72" y1="{y}" x2="828" y2="{y}" stroke="#cbd5e1" stroke-width="3"/>']
+    for index, (milestone, label) in enumerate(labels):
+        x = 72 + index * 189
+        parts.append(f'<circle cx="{x}" cy="{y}" r="17" fill="#0f766e"/><text x="{x}" y="{y + 5}" text-anchor="middle" class="circle-label">{milestone}</text><text x="{x}" y="78" text-anchor="middle" class="label">{label}</text>')
+    parts.append("</svg>")
+    return "".join(parts)
+
+
+def paired_rows(records: dict[str, dict[str, Any]]) -> list[tuple[str, dict[str, Any], dict[str, Any]]]:
+    return [
+        ("prefetch + PP + object", records["prefetch_pp_object_system"], records["prefetch_pp_object_data"]),
+        ("PP + object", records["pp_object_system"], records["pp_object_data"]),
+    ]
+
+
+def prefetch_rows(records: dict[str, dict[str, Any]]) -> list[tuple[str, dict[str, Any], dict[str, Any]]]:
+    return [
+        ("system disk", records["pp_object_system"], records["prefetch_pp_object_system"]),
+        ("data disk", records["pp_object_data"], records["prefetch_pp_object_data"]),
+    ]
+
+
+def run_rows(records: list[dict[str, Any]]) -> list[tuple[str, dict[str, Any]]]:
+    return [(str(record.get("cohortId")), run) for record in records for run in (record.get("runs") or [])]
+
+
+def all_delta_stats(records: list[dict[str, Any]]) -> dict[str, float]:
+    values = [
+        float(run["wallMs"])
+        for _, run in run_rows(records)
+        if run.get("syncMode") == "delta" and isinstance(run.get("wallMs"), (int, float))
+    ]
+    if not values:
+        return {}
+    return {
+        "count": float(len(values)),
+        "min": min(values),
+        "max": max(values),
+        "mean": statistics.mean(values),
+        "median": statistics.median(values),
+        "p90": statistics.quantiles(values, n=10, method="inclusive")[8] if len(values) > 1 else values[0],
+    }
+
+
+def conclusion_lines(indexed: dict[str, dict[str, Any]]) -> list[str]:
+    prefetch_system = indexed["prefetch_pp_object_system"]
+    prefetch_data = indexed["prefetch_pp_object_data"]
+    no_prefetch_system = indexed["pp_object_system"]
+    no_prefetch_data = indexed["pp_object_data"]
+    return [
+        f"两种磁盘的差异没有在两个 profile 上呈现同一方向：prefetch profile 的稳定 delta median 为 system {number(stable_median(prefetch_system) / 1000 if stable_median(prefetch_system) else None, 2)}s、data {number(stable_median(prefetch_data) / 1000 if stable_median(prefetch_data) else None, 2)}s（data vs system {percent_text(stable_median(prefetch_system), stable_median(prefetch_data))}）；无 prefetch profile 为 system {number(stable_median(no_prefetch_system) / 1000 if stable_median(no_prefetch_system) else None, 2)}s、data {number(stable_median(no_prefetch_data) / 1000 if stable_median(no_prefetch_data) else None, 2)}s（{percent_text(stable_median(no_prefetch_system), stable_median(no_prefetch_data))}）。",
+        f"在同一磁盘内，开启 transport prefetch 相对关闭 prefetch 的稳定 delta median 下降：system {percent_text(stable_median(no_prefetch_system), stable_median(prefetch_system))}，data {percent_text(stable_median(no_prefetch_data), stable_median(prefetch_data))}；这是本实验最稳定的方向性信号。",
+        f"snapshot 不与 delta 同向：system 开启 prefetch 后为 {percent_text((no_prefetch_system.get('snapshot') or {}).get('wallMs'), (prefetch_system.get('snapshot') or {}).get('wallMs'))}，data 为 {percent_text((no_prefetch_data.get('snapshot') or {}).get('wallMs'), (prefetch_data.get('snapshot') or {}).get('wallMs'))}；因此不能把单次 snapshot wall 解释为纯磁盘因果。",
+        "VRP/VAP/PP 产物量级和 warnings 随 live 发布窗口小幅变化，全部 run 成功且没有 recovery snapshot；网络同步长尾与 live 数据演进是主要限制。",
+    ]
+
+
+def write_markdown(records: list[dict[str, Any]], source_commit: str, package_sha256: str, remote_host: str, input_paths: list[Path], path: Path) -> None:
+    indexed = {str(record.get("cohortId")): record for record in records}
+    all_stats = all_delta_stats(records)
+    lines = [
+        "# Feature #149 远端系统盘/数据盘性能实验报告",
+        "",
+        f"- 远端：`{remote_host}`；源码：`{source_commit}`；portable archive SHA-256：`{package_sha256}`",
+        "- 结论口径：snapshot 单样本；全部 delta 为 delta1–delta5；稳定 delta 为 delta2–delta5。",
+        "",
+        "## 结论摘要",
+        "",
+    ]
+    lines.extend(f"- {line}" for line in conclusion_lines(indexed))
+    lines += [
+        "",
+        "## Cohort 摘要",
+        "",
+        "| cohort | disk | profile | total elapsed | snapshot wall | all delta min/mean/median/p90/max | stable median | max RSS | latest VRP/VAP/PP/warnings |",
+        "| --- | --- | --- | ---: | --- | ---: | ---: | ---: | --- |",
+    ]
+    for record in records:
+        delta = stats(record, "deltaWallMs")
+        lines.append(
+            "| {id} | {disk} | {profile} | {elapsed}s | {snapshot}ms | {min}/{mean}/{median}/{p90}/{max}ms | {stable}ms | {rss} KiB | {vrp}/{vap}/{pp}/{warnings} |".format(
+                id=record.get("cohortId"), disk=record.get("disk"), profile=record.get("profile"), elapsed=record.get("elapsedSeconds"),
+                snapshot=number((record.get("snapshot") or {}).get("wallMs"), 1), min=number(delta.get("min"), 1), mean=number(delta.get("mean"), 1),
+                median=number(delta.get("median"), 1), p90=number(delta.get("p90"), 1), max=number(delta.get("max"), 1), stable=number(stable_median(record), 1),
+                rss=number(record.get("maxRssKb")), vrp=latest(record, "vrps"), vap=latest(record, "vaps"), pp=latest(record, "publicationPoints"), warnings=latest(record, "warnings"),
+            )
+        )
+    lines += [
+        "",
+        "## 全部 delta 汇总",
+        "",
+        f"20 个 delta 样本：min/mean/median/p90/max = {number(all_stats.get('min'), 1)}/{number(all_stats.get('mean'), 1)}/{number(all_stats.get('median'), 1)}/{number(all_stats.get('p90'), 1)}/{number(all_stats.get('max'), 1)} ms。",
+        "",
+        "## 磁盘配对与 prefetch 对比",
+        "",
+        "| profile | system all/stable median | data all/stable median | data vs system（all / stable） |",
+        "| --- | ---: | ---: | ---: |",
+    ]
+    for label, system, data in paired_rows(indexed):
+        lines.append(f"| {label} | {number(delta_median(system), 1)} / {number(stable_median(system), 1)} ms | {number(delta_median(data), 1)} / {number(stable_median(data), 1)} ms | {percent_text(delta_median(system), delta_median(data))} / {percent_text(stable_median(system), stable_median(data))} |")
+    lines += [
+        "",
+        "| profile | system all/stable mean | data all/stable mean | data vs system（all / stable） |",
+        "| --- | ---: | ---: | ---: |",
+    ]
+    for label, system, data in paired_rows(indexed):
+        lines.append(f"| {label} | {number(delta_mean(system), 1)} / {number(stable_mean(system), 1)} ms | {number(delta_mean(data), 1)} / {number(stable_mean(data), 1)} ms | {percent_text(delta_mean(system), delta_mean(data))} / {percent_text(stable_mean(system), stable_mean(data))} |")
+    lines += [
+        "",
+        "| disk | no-prefetch stable median | prefetch stable median | prefetch vs no-prefetch |",
+        "| --- | ---: | ---: | ---: |",
+    ]
+    for label, no_prefetch, prefetch in prefetch_rows(indexed):
+        lines.append(f"| {label} | {number(stable_median(no_prefetch), 1)} ms | {number(stable_median(prefetch), 1)} ms | {percent_text(stable_median(no_prefetch), stable_median(prefetch))} |")
+    lines += [
+        "",
+        "| disk | no-prefetch stable mean | prefetch stable mean | prefetch vs no-prefetch |",
+        "| --- | ---: | ---: | ---: |",
+    ]
+    for label, no_prefetch, prefetch in prefetch_rows(indexed):
+        lines.append(f"| {label} | {number(stable_mean(no_prefetch), 1)} ms | {number(stable_mean(prefetch), 1)} ms | {percent_text(stable_mean(no_prefetch), stable_mean(prefetch))} |")
+    lines += ["", "## 逐 run 证据", "", "| cohort | run | mode | actual start UTC | wall ms | max RSS KiB | schedule lag ms | VRP/VAP/PP/warnings | artifacts |", "| --- | --- | --- | --- | ---: | ---: | ---: | --- | --- |"]
+    for cohort, run in run_rows(records):
+        lines.append(f"| {cohort} | {run.get('runId')} | {run.get('syncMode')} | {run.get('actualStartRfc3339Utc')} | {number(run.get('wallMs'), 1)} | {number(run.get('maxRssKb'))} | {number(run.get('scheduleLagMs'), 1)} | {run.get('vrps')}/{run.get('vaps')}/{run.get('publicationPoints')}/{run.get('warnings')} | {'complete' if run.get('artifactsComplete') and not run.get('missingArtifacts') else 'INCOMPLETE'} |")
+    lines += [
+        "",
+        "## 验收、异常与证据",
+        "",
+        "- 4 个 cohort 均为 6/6 success、首轮 snapshot 后连续五轮 delta，实际相邻启动节拍为 600 秒，`scheduleLagMs=0`；没有 recovery snapshot 或缺失核心产物。",
+        "- warnings 的最新值范围为 524–540（各 run 约 518–544），VRP/VAP/PP 在 live 窗口内小幅变化；这些属于可解释的 live/网络窗口变化，不是失败状态。",
+        "- 完整 run root 保留在远端对应系统盘或 `/data`；本地只回传四份 `cohort-summary.json/.md` 轻量证据。",
+        "",
+        "本报告由四份 `cohort-summary.json` 自动生成；由于 live RPKI 数据、网络和运行时间窗口不同，磁盘结果仅作方向性比较。",
+        "",
+        "### 本地证据",
+        "",
+    ]
+    lines.extend(f"- `{path}`" for path in input_paths)
+    lines += ["", "### 远端原始证据根", ""]
+    for record in records:
+        package_root = record.get("packageRoot", "n/a")
+        cohort = record.get("cohortId", "unknown")
+        lines.append(f"- `{cohort}` package/run root：`{package_root}`；cohort summary：`{package_root}/experiments/feature149-{cohort}/cohort-summary.json`")
+    path.write_text("\n".join(lines) + "\n", encoding="utf-8")
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--cohort-summary", action="append", required=True, type=Path)
+    parser.add_argument("--out-html", required=True, type=Path)
+    parser.add_argument("--out-markdown", required=True, type=Path)
+    parser.add_argument("--remote-host", required=True)
+    parser.add_argument("--source-commit", required=True)
+    parser.add_argument("--package-sha256", default="see M2 package report")
+    args = parser.parse_args()
+    input_paths = list(args.cohort_summary)
+    records = [load(path) for path in input_paths]
+    indexed = {str(record.get("cohortId")): record for record in records}
+    if set(indexed) != set(EXPECTED_COHORTS):
+        raise ValueError(f"expected cohorts {EXPECTED_COHORTS}, got {tuple(indexed)}")
+    ordered = [indexed[name] for name in EXPECTED_COHORTS]
+    for record in ordered:
+        runs = record.get("runs") or []
+        if record.get("successCount") != 6 or record.get("runCount") != 6 or record.get("artifactStatus") != "complete" or len(runs) != 6:
+            raise ValueError(f"incomplete cohort {record.get('cohortId')}")
+        if runs[0].get("syncMode") != "snapshot" or any(run.get("syncMode") != "delta" for run in runs[1:]):
+            raise ValueError(f"unexpected sync modes in {record.get('cohortId')}")
+
+    all_stats = all_delta_stats(ordered)
+    cards = [
+        ("Cohorts", "4", "all complete"),
+        ("Runs", "24 / 24", "snapshot + 5 delta per cohort"),
+        ("All delta median", f"{number(all_stats.get('median', 0) / 1000, 2)}s", "20 live delta samples"),
+        ("Source", html.escape(args.source_commit), html.escape(args.remote_host)),
+    ]
+    card_html = "".join(f'<article class="card"><div>{title}</div><strong>{value}</strong><small>{detail}</small></article>' for title, value, detail in cards)
+
+    cohort_rows_html = "".join(
+        "<tr><td>{id}</td><td>{disk}</td><td>{profile}</td><td>{elapsed}s</td><td>{snapshot} ms</td><td>{min}/{mean}/{median}/{p90}/{max} ms</td><td>{stable} ms</td><td>{rss} KiB</td><td>{vrp}</td><td>{vap}</td><td>{pp}</td><td>{warnings}</td></tr>".format(
+            id=html.escape(str(record.get("cohortId"))), disk=html.escape(str(record.get("disk"))), profile=html.escape(str(record.get("profile"))), elapsed=number(record.get("elapsedSeconds")),
+            snapshot=number((record.get("snapshot") or {}).get("wallMs"), 1), min=number(stat_value(record, "deltaWallMs", "min"), 1), mean=number(stat_value(record, "deltaWallMs", "mean"), 1),
+            median=number(stat_value(record, "deltaWallMs", "median"), 1), p90=number(stat_value(record, "deltaWallMs", "p90"), 1), max=number(stat_value(record, "deltaWallMs", "max"), 1), stable=number(stable_median(record), 1),
+            rss=number(record.get("maxRssKb")), vrp=number(latest(record, "vrps")), vap=number(latest(record, "vaps")), pp=number(latest(record, "publicationPoints")), warnings=number(latest(record, "warnings")),
+        )
+        for record in ordered
+    )
+    paired_html = "".join(
+        "<tr><td>{label}</td><td>{system_all} / {system_stable}</td><td>{data_all} / {data_stable}</td><td>{all_change} / {stable_change}</td></tr>".format(
+            label=html.escape(label), system_all=number(delta_median(system), 1), system_stable=number(stable_median(system), 1), data_all=number(delta_median(data), 1), data_stable=number(stable_median(data), 1),
+            all_change=percent_text(delta_median(system), delta_median(data)), stable_change=percent_text(stable_median(system), stable_median(data)),
+        )
+        for label, system, data in paired_rows(indexed)
+    )
+    paired_mean_html = "".join(
+        "<tr><td>{label}</td><td>{system_all} / {system_stable}</td><td>{data_all} / {data_stable}</td><td>{all_change} / {stable_change}</td></tr>".format(
+            label=html.escape(label), system_all=number(delta_mean(system), 1), system_stable=number(stable_mean(system), 1), data_all=number(delta_mean(data), 1), data_stable=number(stable_mean(data), 1),
+            all_change=percent_text(delta_mean(system), delta_mean(data)), stable_change=percent_text(stable_mean(system), stable_mean(data)),
+        )
+        for label, system, data in paired_rows(indexed)
+    )
+    prefetch_html = "".join(
+        "<tr><td>{label}</td><td>{no_prefetch}</td><td>{prefetch}</td><td>{change}</td></tr>".format(
+            label=html.escape(label), no_prefetch=number(stable_median(no_prefetch), 1), prefetch=number(stable_median(prefetch), 1), change=percent_text(stable_median(no_prefetch), stable_median(prefetch)),
+        )
+        for label, no_prefetch, prefetch in prefetch_rows(indexed)
+    )
+    prefetch_mean_html = "".join(
+        "<tr><td>{label}</td><td>{no_prefetch}</td><td>{prefetch}</td><td>{change}</td></tr>".format(
+            label=html.escape(label), no_prefetch=number(stable_mean(no_prefetch), 1), prefetch=number(stable_mean(prefetch), 1), change=percent_text(stable_mean(no_prefetch), stable_mean(prefetch)),
+        )
+        for label, no_prefetch, prefetch in prefetch_rows(indexed)
+    )
+    run_rows_html = "".join(
+        "<tr><td>{cohort}</td><td>{run}</td><td>{mode}</td><td>{start}</td><td>{wall}</td><td>{rss}</td><td>{lag}</td><td>{vrp}/{vap}/{pp}/{warnings}</td><td>{artifact}</td></tr>".format(
+            cohort=html.escape(cohort), run=html.escape(str(run.get("runId"))), mode=html.escape(str(run.get("syncMode"))), start=html.escape(str(run.get("actualStartRfc3339Utc"))), wall=number(run.get("wallMs"), 1), rss=number(run.get("maxRssKb")), lag=number(run.get("scheduleLagMs"), 1),
+            vrp=number(run.get("vrps")), vap=number(run.get("vaps")), pp=number(run.get("publicationPoints")), warnings=number(run.get("warnings")), artifact="complete" if run.get("artifactsComplete") and not run.get("missingArtifacts") else "INCOMPLETE",
+        )
+        for cohort, run in run_rows(ordered)
+    )
+    evidence_rows_html = "".join(
+        "<tr><td>{cohort}</td><td><code>{root}</code></td><td><code>{summary}</code></td></tr>".format(
+            cohort=html.escape(str(record.get("cohortId"))),
+            root=html.escape(str(record.get("packageRoot", "n/a"))),
+            summary=html.escape(f"{record.get('packageRoot', 'n/a')}/experiments/feature149-{record.get('cohortId')}/cohort-summary.json"),
+        )
+        for record in ordered
+    )
+    warning_values = [latest(record, "warnings") for record in ordered if isinstance(latest(record, "warnings"), (int, float))]
+    max_delta = max((run.get("wallMs", 0) for _, run in run_rows(ordered) if run.get("syncMode") == "delta"), default=0)
+    report_html = f'''<!doctype html><html lang="zh-CN"><meta charset="utf-8"><title>Feature #149 性能实验报告</title>
+<style>
+body{{font-family:Inter,system-ui,sans-serif;margin:0;background:#f8fafc;color:#0f172a}}main{{max-width:1380px;margin:auto;padding:34px}}h1{{margin-bottom:4px}}.sub{{color:#475569}}.cards{{display:grid;grid-template-columns:repeat(4,1fr);gap:14px;margin:26px 0}}.card{{background:white;border:1px solid #e2e8f0;border-radius:12px;padding:16px}}.card strong{{display:block;font-size:25px;margin:8px 0}}.card small{{color:#64748b}}section{{background:white;border:1px solid #e2e8f0;border-radius:12px;padding:20px;margin-top:18px;overflow:auto}}table{{border-collapse:collapse;width:100%;font-size:13px}}th,td{{padding:9px;border-bottom:1px solid #e2e8f0;text-align:left;white-space:nowrap}}th{{background:#f1f5f9}}svg{{width:100%;height:auto;min-width:760px}}.axis,.label{{font-size:12px;fill:#475569}}.value{{font-size:13px;font-weight:700;fill:#0f172a}}.circle-label{{font-size:10px;fill:#fff;font-weight:700}}.note{{color:#475569;line-height:1.6}}li{{margin:7px 0;line-height:1.5}}code{{background:#f1f5f9;padding:2px 5px;border-radius:4px}}
+</style>
+<main><h1>Feature #149：系统盘 / 数据盘缓存性能实验</h1><p class="sub">远端 {html.escape(args.remote_host)} · source {html.escape(args.source_commit)} · archive SHA-256 {html.escape(args.package_sha256)} · 4 cohorts / 24 serial all5 runs</p><div class="cards">{card_html}</div>
+<section><h2>实验拓扑与时序</h2><p class="note">amd64 远端主机的 system cohort 将 portable package、state、数据和日志全部写在系统盘；data cohort 将对应内容全部写在 <code>/data</code>。四个 cohort 按 A-system → A-data → B-system → B-data 串行执行，每个 row 是 1 snapshot + 5 delta，相邻启动节拍 600 秒。</p>{timeline_svg(ordered)}</section>
+<section><h2>Milestone timeline</h2>{milestone_svg()}<p class="note">M1–M5 均已通过：磁盘初始化、同 checksum package 部署与 dry-run、A/B 两组正式运行、证据归档和报告收口。</p></section>
+<section><h2>主机、磁盘与 provenance</h2><p class="note"><code>{html.escape(args.remote_host)}</code>；system 为远端根文件系统（/dev/vda3），data 为 /data（/dev/vdb1、ext4、rpki-data）。同一 archive checksum 为 <code>{html.escape(args.package_sha256)}</code>，源码 commit 为 <code>{html.escape(args.source_commit)}</code>。对象 cache 固定包含 ROA 与 child-certificate validation cache；两 profile 的唯一 cache 差异是 transport prefetch。</p></section>
+<section><h2>结论摘要</h2><ul>{''.join(f'<li>{html.escape(line)}</li>' for line in conclusion_lines(indexed))}</ul></section>
+<section><h2>Delta median wall clock</h2>{svg_chart(ordered)}<p class="note">颜色：蓝色 system，青绿色 data。全体 delta 的 median 为 {number(all_stats.get('median'), 1)} ms；最大 delta 为 {number(max_delta, 1)} ms，长尾包含网络同步窗口。</p></section>
+<section><h2>逐 cohort KPI</h2><table><tr><th>cohort</th><th>disk</th><th>profile</th><th>elapsed</th><th>snapshot</th><th>delta min/mean/median/p90/max</th><th>stable median</th><th>max RSS</th><th>VRP</th><th>VAP</th><th>PP</th><th>warnings</th></tr>{cohort_rows_html}</table></section>
+<section><h2>磁盘配对（all delta / stable delta median）</h2><table><tr><th>profile</th><th>system ms</th><th>data ms</th><th>data vs system（all / stable）</th></tr>{paired_html}</table><p class="note">负值表示 data median 更快，正值表示更慢；snapshot 单样本另行列在 cohort KPI 表中。</p></section>
+<section><h2>磁盘配对（all delta / stable delta mean）</h2><table><tr><th>profile</th><th>system ms</th><th>data ms</th><th>data vs system（all / stable）</th></tr>{paired_mean_html}</table><p class="note">这里的 mean 是对应 cohort 的 deltaWallMs 平均值；负值表示 data mean 更快。</p></section>
+<section><h2>Prefetch 开关对比（stable delta median）</h2><table><tr><th>disk</th><th>no prefetch ms</th><th>prefetch ms</th><th>prefetch vs no prefetch</th></tr>{prefetch_html}</table></section>
+<section><h2>Prefetch 开关对比（stable delta mean）</h2><table><tr><th>disk</th><th>no prefetch ms</th><th>prefetch ms</th><th>prefetch vs no prefetch</th></tr>{prefetch_mean_html}</table></section>
+<section><h2>逐 run 原始指标</h2><table><tr><th>cohort</th><th>run</th><th>mode</th><th>actual start UTC</th><th>wall ms</th><th>max RSS KiB</th><th>schedule lag ms</th><th>VRP/VAP/PP/warnings</th><th>artifacts</th></tr>{run_rows_html}</table></section>
+<section><h2>验收与异常诊断</h2><ul><li>4 个 cohort 均为 6/6 success、首轮 snapshot 后连续五轮 delta；summary 中每 run 的核心产物均完整，实际启动间隔校验为 600 秒。</li><li>最新 warnings 范围为 {number(min(warning_values) if warning_values else None)}–{number(max(warning_values) if warning_values else None)}；VRP/VAP/PP 在 live 窗口内小幅变化，未出现输出塌缩或 recovery snapshot。</li><li>结果受 live RPKI 发布变化、RRDP/rsync 网络长尾和顺序时间窗口影响；不能仅凭四组 wall 值声称 IOPS 的严格因果效应。</li></ul><table><tr><th>cohort</th><th>远端 package/run root</th><th>远端 cohort summary</th></tr>{evidence_rows_html}</table><p class="note">本地输入证据为：{'; '.join(f'<code>{html.escape(str(path))}</code>' for path in input_paths)}。</p></section>
+<section><h2>统计口径</h2><p class="note">snapshot 统计只包含每个 cohort 的首轮；all delta 统计包含 delta1–delta5 共 20 个样本；stable delta 统计包含 delta2–delta5 共 16 个样本。详细 cache 命中、stage timing、命令 argv、run-meta/run-summary 和完整产物仍在远端 run root。</p></section></main></html>'''
+    args.out_html.parent.mkdir(parents=True, exist_ok=True)
+    args.out_markdown.parent.mkdir(parents=True, exist_ok=True)
+    args.out_html.write_text(report_html, encoding="utf-8")
+    write_markdown(ordered, args.source_commit, args.package_sha256, args.remote_host, input_paths, args.out_markdown)
+
+
+if __name__ == "__main__":
+    main()
diff --git a/scripts/soak/run_cache_ablation_experiment.sh b/scripts/soak/run_cache_ablation_experiment.sh
index 7d332cc..aa394a4 100755
--- a/scripts/soak/run_cache_ablation_experiment.sh
+++ b/scripts/soak/run_cache_ablation_experiment.sh
@@ -38,7 +38,8 @@ 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, cache-only, crypto-sig or sig-pp-compare; default runs the original 4-case matrix
+  EXPERIMENT_CASE_SET          default, cache-only, crypto-sig, sig-pp-compare,
+                               feature149-prefetch or feature149-no-prefetch
   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
@@ -98,7 +99,8 @@ case_count() {
     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" ;;
+    feature149-prefetch|feature149-no-prefetch) printf '%s' 1 ;;
+    *) die "unknown EXPERIMENT_CASE_SET: $EXPERIMENT_CASE_SET" ;;
   esac
 }
 
@@ -119,6 +121,8 @@ case_id_for_index() {
     sig-pp-compare:2) printf '%s' "pp-only" ;;
     baseline-combo:1) printf '%s' "baseline" ;;
     baseline-combo:2) printf '%s' "pp-object-sig" ;;
+    feature149-prefetch:1) printf '%s' "prefetch-pp-object" ;;
+    feature149-no-prefetch:1) printf '%s' "pp-object" ;;
     *) die "unknown case index: $1 for set $EXPERIMENT_CASE_SET" ;;
   esac
 }
@@ -140,6 +144,8 @@ case_name_for_index() {
     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" ;;
+    feature149-prefetch:1) printf '%s' "prefetch-pp-object-cache" ;;
+    feature149-no-prefetch:1) printf '%s' "pp-object-cache" ;;
     *) die "unknown case index: $1 for set $EXPERIMENT_CASE_SET" ;;
   esac
 }
@@ -161,6 +167,8 @@ case_extra_args_for_index() {
     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" ;;
+    feature149-prefetch:1) printf '%s' "--enable-transport-request-prefetch --enable-publication-point-validation-cache --enable-roa-validation-cache --parallel-max-repo-sync-workers-global 4 --parallel-phase2-object-workers 4 --memory-trim-after-validation" ;;
+    feature149-no-prefetch:1) printf '%s' "--enable-publication-point-validation-cache --enable-roa-validation-cache --parallel-max-repo-sync-workers-global 4 --parallel-phase2-object-workers 4 --memory-trim-after-validation" ;;
     *) die "unknown case index: $1 for set $EXPERIMENT_CASE_SET" ;;
   esac
 }
@@ -168,6 +176,7 @@ case_extra_args_for_index() {
 case_child_cert_cache_for_index() {
   case "$EXPERIMENT_CASE_SET:$1" in
     default:4|cache-only:2|cache-only:3|crypto-sig:3|crypto-sig:4|baseline-combo:2) printf '%s' "1" ;;
+    feature149-prefetch:1|feature149-no-prefetch:1) printf '%s' "1" ;;
     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 $EXPERIMENT_CASE_SET" ;;
   esac
@@ -207,6 +216,14 @@ elif case_set == "baseline-combo":
         {"caseId": "baseline", "caseName": "all-cache-off", "extraArgs": "", "enableChildCertificateValidationCache": False},
         {"caseId": "pp-object-sig", "caseName": "pp-cache-object-crypto-sig-cache", "extraArgs": "--enable-publication-point-validation-cache --enable-roa-validation-cache --enable-crypto-signature-cache", "enableChildCertificateValidationCache": True},
     ]
+elif case_set == "feature149-prefetch":
+    cases = [
+        {"caseId": "prefetch-pp-object", "caseName": "prefetch-pp-object-cache", "extraArgs": "--enable-transport-request-prefetch --enable-publication-point-validation-cache --enable-roa-validation-cache --parallel-max-repo-sync-workers-global 4 --parallel-phase2-object-workers 4 --memory-trim-after-validation", "enableChildCertificateValidationCache": True},
+    ]
+elif case_set == "feature149-no-prefetch":
+    cases = [
+        {"caseId": "pp-object", "caseName": "pp-object-cache", "extraArgs": "--enable-publication-point-validation-cache --enable-roa-validation-cache --parallel-max-repo-sync-workers-global 4 --parallel-phase2-object-workers 4 --memory-trim-after-validation", "enableChildCertificateValidationCache": True},
+    ]
 else:
     raise SystemExit(f"unknown case set: {case_set}")
 print(json.dumps(cases, ensure_ascii=False, indent=8))
diff --git a/scripts/soak/run_feature149_cohort.sh b/scripts/soak/run_feature149_cohort.sh
new file mode 100644
index 0000000..9cb7b77
--- /dev/null
+++ b/scripts/soak/run_feature149_cohort.sh
@@ -0,0 +1,79 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+PACKAGE_ROOT="${PACKAGE_ROOT:-$(cd "$SCRIPT_DIR/../.." && pwd)}"
+COHORT_ID="${COHORT_ID:?COHORT_ID is required}"
+COHORT_DISK="${COHORT_DISK:?COHORT_DISK is required (system or data)}"
+COHORT_PROFILE="${COHORT_PROFILE:?COHORT_PROFILE is required (prefetch-pp-object or pp-object)}"
+RUN_START_INTERVAL_SECS="${RUN_START_INTERVAL_SECS:-600}"
+CASE_RUNS="${CASE_RUNS:-5}"
+DRY_RUN="${DRY_RUN:-0}"
+FIRST_RUN_NOT_BEFORE_EPOCH="${FIRST_RUN_NOT_BEFORE_EPOCH:-0}"
+
+die() {
+  echo "error: $*" >&2
+  exit 2
+}
+
+is_true() {
+  case "${1:-}" in
+    1|true|TRUE|yes|YES|on|ON) return 0 ;;
+    *) return 1 ;;
+  esac
+}
+
+case "$COHORT_DISK" in
+  system|data) ;;
+  *) die "COHORT_DISK must be system or data: $COHORT_DISK" ;;
+esac
+
+case "$COHORT_PROFILE" in
+  prefetch-pp-object) case_set="feature149-prefetch" ;;
+  pp-object) case_set="feature149-no-prefetch" ;;
+  *) die "COHORT_PROFILE must be prefetch-pp-object or pp-object: $COHORT_PROFILE" ;;
+esac
+
+[[ "$CASE_RUNS" == "5" ]] || die "Feature #149 requires CASE_RUNS=5, got $CASE_RUNS"
+[[ "$RUN_START_INTERVAL_SECS" == "600" ]] || die "Feature #149 requires RUN_START_INTERVAL_SECS=600, got $RUN_START_INTERVAL_SECS"
+[[ -x "$SCRIPT_DIR/run_cache_ablation_experiment.sh" ]] || die "missing cache ablation driver"
+[[ -x "$SCRIPT_DIR/summarize_feature149_cohort.py" ]] || die "missing Feature #149 summarizer"
+[[ -f "$PACKAGE_ROOT/.env" ]] || die "missing package environment: $PACKAGE_ROOT/.env"
+
+experiment_dir="$PACKAGE_ROOT/experiments/feature149-$COHORT_ID"
+summary_jsonl="$experiment_dir/experiment-summary.jsonl"
+
+if (( FIRST_RUN_NOT_BEFORE_EPOCH > 0 )) && ! is_true "$DRY_RUN"; then
+  now_epoch="$(date +%s)"
+  if (( now_epoch < FIRST_RUN_NOT_BEFORE_EPOCH )); then
+    wait_secs=$(( FIRST_RUN_NOT_BEFORE_EPOCH - now_epoch ))
+    echo "[$(date -u +%Y-%m-%dT%H:%M:%SZ)] waiting ${wait_secs}s before cohort=$COHORT_ID" >&2
+    sleep "$wait_secs"
+  fi
+fi
+
+env \
+  PACKAGE_ROOT="$PACKAGE_ROOT" \
+  ENV_FILE="$PACKAGE_ROOT/.env" \
+  EXPERIMENT_RUN_ROOT="$PACKAGE_ROOT" \
+  EXPERIMENT_DIR="$experiment_dir" \
+  EXPERIMENT_CASE_SET="$case_set" \
+  CASE_RUNS="$CASE_RUNS" \
+  RUN_SNAPSHOT=1 \
+  RUN_START_INTERVAL_SECS="$RUN_START_INTERVAL_SECS" \
+  FIRST_RUN_DELAY_SECS=0 \
+  RETAIN_RUNS=6 \
+  DRY_RUN="$DRY_RUN" \
+  "$SCRIPT_DIR/run_cache_ablation_experiment.sh"
+
+if ! is_true "$DRY_RUN"; then
+  "$SCRIPT_DIR/summarize_feature149_cohort.py" \
+    --cohort-id "$COHORT_ID" \
+    --cohort-disk "$COHORT_DISK" \
+    --cohort-profile "$COHORT_PROFILE" \
+    --package-root "$PACKAGE_ROOT" \
+    --summary-jsonl "$summary_jsonl" \
+    --out-dir "$experiment_dir"
+fi
+
+printf 'feature149_cohort_dir=%s\n' "$experiment_dir"
diff --git a/scripts/soak/summarize_feature149_cohort.py b/scripts/soak/summarize_feature149_cohort.py
new file mode 100644
index 0000000..456e026
--- /dev/null
+++ b/scripts/soak/summarize_feature149_cohort.py
@@ -0,0 +1,169 @@
+#!/usr/bin/env python3
+"""Validate and summarize one Feature #149 six-run cohort."""
+
+from __future__ import annotations
+
+import argparse
+import json
+import statistics
+from datetime import datetime, timezone
+from pathlib import Path
+from typing import Any
+
+
+REQUIRED_ARTIFACTS = {
+    "run-meta.json",
+    "run-summary.json",
+    "input.cir",
+    "result.ccr",
+    "report.json",
+    "vrps.csv",
+    "vaps.csv",
+    "stage-timing.json",
+    "process-time.txt",
+}
+
+
+def load_jsonl(path: Path) -> list[dict[str, Any]]:
+    records: list[dict[str, Any]] = []
+    for line_number, line in enumerate(path.read_text(encoding="utf-8").splitlines(), 1):
+        if line.strip():
+            value = json.loads(line)
+            if not isinstance(value, dict):
+                raise ValueError(f"{path}:{line_number} is not an object")
+            records.append(value)
+    return records
+
+
+def require_success(records: list[dict[str, Any]], package_root: Path) -> list[dict[str, Any]]:
+    if len(records) != 6:
+        raise ValueError(f"expected 6 records (snapshot + 5 delta), found {len(records)}")
+    for index, record in enumerate(records):
+        expected_mode = "snapshot" if index == 0 else "delta"
+        if record.get("status") != "success":
+            raise ValueError(f"run {index + 1} status is {record.get('status')!r}")
+        if record.get("syncMode") != expected_mode:
+            raise ValueError(
+                f"run {index + 1} syncMode is {record.get('syncMode')!r}, expected {expected_mode!r}"
+            )
+        run_dir = package_root / "runs" / f"run_{int(record['maxRunIndexAfter']):04d}"
+        missing = sorted(name for name in REQUIRED_ARTIFACTS if not (run_dir / name).is_file())
+        record["runDir"] = str(run_dir)
+        record["artifactsComplete"] = not missing
+        record["missingArtifacts"] = missing
+        if missing:
+            raise ValueError(f"run {index + 1} missing required artifacts: {', '.join(missing)}")
+    return records
+
+
+def percentile(values: list[float], fraction: float) -> float | None:
+    if not values:
+        return None
+    ordered = sorted(values)
+    position = (len(ordered) - 1) * fraction
+    lower = int(position)
+    upper = min(lower + 1, len(ordered) - 1)
+    return ordered[lower] + (ordered[upper] - ordered[lower]) * (position - lower)
+
+
+def stats(values: list[int | float | None]) -> dict[str, float | None]:
+    nums = [float(value) for value in values if isinstance(value, (int, float))]
+    if not nums:
+        return {"count": 0, "min": None, "median": None, "p90": None, "max": None, "mean": None}
+    return {
+        "count": len(nums),
+        "min": min(nums),
+        "median": statistics.median(nums),
+        "p90": percentile(nums, 0.9),
+        "max": max(nums),
+        "mean": statistics.fmean(nums),
+    }
+
+
+def iso(epoch: Any) -> str | None:
+    if not isinstance(epoch, (int, float)):
+        return None
+    return datetime.fromtimestamp(epoch, timezone.utc).isoformat().replace("+00:00", "Z")
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--cohort-id", required=True)
+    parser.add_argument("--cohort-disk", required=True, choices=("system", "data"))
+    parser.add_argument("--cohort-profile", required=True, choices=("prefetch-pp-object", "pp-object"))
+    parser.add_argument("--package-root", required=True, type=Path)
+    parser.add_argument("--summary-jsonl", required=True, type=Path)
+    parser.add_argument("--out-dir", required=True, type=Path)
+    args = parser.parse_args()
+
+    records = require_success(load_jsonl(args.summary_jsonl), args.package_root)
+    delta = records[1:]
+    elapsed_seconds = records[-1]["completedEpoch"] - records[0]["actualStartEpoch"]
+    report = {
+        "feature": "149",
+        "createdAtUtc": datetime.now(timezone.utc).isoformat().replace("+00:00", "Z"),
+        "cohortId": args.cohort_id,
+        "disk": args.cohort_disk,
+        "profile": args.cohort_profile,
+        "packageRoot": str(args.package_root),
+        "runCount": len(records),
+        "successCount": len(records),
+        "artifactStatus": "complete",
+        "elapsedSeconds": elapsed_seconds,
+        "actualStartedAtUtc": iso(records[0].get("actualStartEpoch")),
+        "completedAtUtc": iso(records[-1].get("completedEpoch")),
+        "maxRssKb": max(int(record["maxRssKb"]) for record in records if isinstance(record.get("maxRssKb"), (int, float))),
+        "snapshot": {"wallMs": records[0].get("wallMs"), "maxRssKb": records[0].get("maxRssKb")},
+        "deltaWallMs": stats([record.get("wallMs") for record in delta]),
+        "stableDeltaWallMs": stats([record.get("wallMs") for record in delta[1:]]),
+        "latestCounts": {
+            key: records[-1].get(key)
+            for key in ("vrps", "vaps", "publicationPoints", "warnings")
+        },
+        "runs": records,
+    }
+    args.out_dir.mkdir(parents=True, exist_ok=True)
+    json_path = args.out_dir / "cohort-summary.json"
+    markdown_path = args.out_dir / "cohort-summary.md"
+    json_path.write_text(json.dumps(report, ensure_ascii=False, indent=2, sort_keys=True) + "\n", encoding="utf-8")
+    latest = report["latestCounts"]
+    markdown_path.write_text(
+        "\n".join(
+            [
+                f"# Feature #149 cohort: {args.cohort_id}",
+                "",
+                f"- 磁盘：`{args.cohort_disk}`",
+                f"- profile：`{args.cohort_profile}`",
+                f"- 状态：`success`，6/6 run 成功，必要产物完整",
+                f"- 总执行耗时：`{elapsed_seconds}s`",
+                f"- 最高 RSS：`{report['maxRssKb']} KiB`",
+                f"- snapshot wall：`{records[0].get('wallMs')} ms`",
+                f"- delta median / p90：`{report['deltaWallMs']['median']} ms` / `{report['deltaWallMs']['p90']} ms`",
+                f"- 最新产物：VRP `{latest['vrps']}`、VAP `{latest['vaps']}`、PP `{latest['publicationPoints']}`、warnings `{latest['warnings']}`",
+                "",
+                "| run | mode | wall ms | max RSS KiB | VRP | VAP | PP | warnings | artifacts |",
+                "| --- | --- | ---: | ---: | ---: | ---: | ---: | ---: | --- |",
+                *[
+                    "| {run} | {mode} | {wall} | {rss} | {vrps} | {vaps} | {pps} | {warnings} | {artifacts} |".format(
+                        run=index + 1,
+                        mode=record.get("syncMode"),
+                        wall=record.get("wallMs"),
+                        rss=record.get("maxRssKb"),
+                        vrps=record.get("vrps"),
+                        vaps=record.get("vaps"),
+                        pps=record.get("publicationPoints"),
+                        warnings=record.get("warnings"),
+                        artifacts="complete" if record.get("artifactsComplete") else "missing",
+                    )
+                    for index, record in enumerate(records)
+                ],
+                "",
+            ]
+        ),
+        encoding="utf-8",
+    )
+    print(json.dumps({"cohortSummaryJson": str(json_path), "cohortSummaryMarkdown": str(markdown_path), "elapsedSeconds": elapsed_seconds, "maxRssKb": report["maxRssKb"]}, ensure_ascii=False))
+
+
+if __name__ == "__main__":
+    main()