From 5839112c74af448037833324118c7bcb213ddbfe Mon Sep 17 00:00:00 2001
From: lanyue-llk <270302213+lanyue-llk@users.noreply.github.com>
Date: Fri, 23 Oct 2026 15:02:24 +0800
Subject: [PATCH 1/5] feat(reporting): bind reports and calls to benchmark
 editions

---
 CONTRIBUTING.md                         |   5 +
 README.md                               |   5 +
 docs/README.zh-CN.md                    |   4 +
 docs/benchmarks.md                      |  48 +++++
 docs/report-hosting.md                  |   7 +
 evaluation/lexbench-webfetch.json       |   2 +-
 harness/web_fetch/certification.py      |  19 ++
 harness/web_fetch/published_snapshot.py |  15 +-
 harness/web_fetch/report.py             |   4 +
 harness/web_fetch/report_build.py       |  12 +-
 harness/web_fetch/report_check.py       |   4 +
 harness/web_fetch/report_version.py     | 128 ++++++++++++++
 pyproject.toml                          |   2 +-
 tests/test_benchmark_release_binding.py | 225 ++++++++++++++++++++++++
 tests/test_local_audit.py               |   9 +
 tests/test_run_certified_benchmark.py   |   6 +
 uv.lock                                 |   4 +-
 17 files changed, 491 insertions(+), 8 deletions(-)
 create mode 100644 harness/web_fetch/report_version.py
 create mode 100644 tests/test_benchmark_release_binding.py

diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md
index 152938b9c..3a6bac756 100644
--- a/CONTRIBUTING.md
+++ b/CONTRIBUTING.md
@@ -27,6 +27,11 @@ Test both accepted and rejected scenarios. Fix scoring rules rather than editing
 
 Write documentation in English. Keep the [Chinese README](docs/README.zh-CN.md) aligned when setup or usage changes; other docs do not need duplicate translations. Preserve original task text, quotations and frozen acceptance inputs. Reports follow their track's language settings.
 
+Benchmark changes follow [benchmark editions](docs/benchmarks.md#benchmark-editions).
+Keep human-readable editions, immutable dataset/scorer hashes and the package
+commit together. A minor scoring correction requires common-edition rescoring;
+changed task populations or requests require a major edition.
+
 ## PR and commit titles
 
 Use `<type>(<scope>): <summary>` for PR titles and new commit subjects. The scope is optional; use the module or feature affected, such as `scoring` or `webfetch`.
diff --git a/README.md b/README.md
index 586f61ee4..be06ae515 100644
--- a/README.md
+++ b/README.md
@@ -106,6 +106,11 @@ Install an extra with `uv sync --frozen --extra NAME`. Browser binaries, Docker,
 
 ## Read and share a report
 
+Reports retain the selected benchmark edition and exact dataset/scorer hashes.
+Package upgrades leave historical results unchanged. See
+[benchmark editions](docs/benchmarks.md#benchmark-editions) for patch, minor and
+major comparison rules.
+
 | Output | What you will find |
 | --- | --- |
 | `report/en/index.html` | English comparisons, task results and evidence links |
diff --git a/docs/README.zh-CN.md b/docs/README.zh-CN.md
index c5e2ad2a0..7b2bceeb8 100644
--- a/docs/README.zh-CN.md
+++ b/docs/README.zh-CN.md
@@ -106,6 +106,10 @@ uv run --frozen browser-harness-eval run --delay 0
 
 ## 查看与分享报告
 
+报告固定绑定所选评测版本，以及题目和评分源码的精确哈希。升级软件包不会改变历史结果。
+补丁版可以直接比较；评分修正的小版本要统一重评；题目或任务要求改变的大版本分别展示。
+具体规则见[评测版本](benchmarks.md#benchmark-editions)。
+
 | 输出 | 内容 |
 | --- | --- |
 | `report/en/index.html` | 英文比较、任务结果和证据入口 |
diff --git a/docs/benchmarks.md b/docs/benchmarks.md
index cab5da6cb..71e8d5d50 100644
--- a/docs/benchmarks.md
+++ b/docs/benchmarks.md
@@ -51,6 +51,54 @@ Change task requirements in LexBench, freeze a new dataset/catalog/manifest vers
 
 The retained `harness/lexbench/` module concerns historical headless-browser comparisons. It is unrelated to the `lexbench-webfetch` input package and remains an execution/report adapter.
 
+## Benchmark editions
+
+Use a benchmark version together with exact task/reference and scoring-source
+hashes. Keep the package's full Git commit pinned separately. A package update
+does not select a new dataset for an existing report.
+
+| Change | Edition | Result comparison |
+| --- | --- | --- |
+| Documentation or presentation only | Patch | Direct when execution controls also match |
+| Reviewed scoring/reference correction with the same task inputs | Minor | Re-score both saved cohorts under one edition |
+| Added/removed cases, changed requests, output scope or routes | Major | Separate full-population results |
+
+The task package owns `lexbench-webfetch release` and `compare-releases`.
+See its [version contract](https://github.com/lexmount/lexbench-webfetch/blob/98dd3423a5d179b4662c27beccae0918b94bc2e8/docs/versions.md).
+Freeze `benchmark-release.json` beside the compiled inputs before calls.
+The certified Web Fetch runner accepts `--benchmark-release`; it also discovers
+the sidecar beside the selected dataset. Release creation records the full scorer
+commit alongside the independently computed source hash. The runner checks tasks and the active scorer
+before external work and retains the binding in execution metadata. The original
+call edition remains in benchmark-input-release.json; a report may select a newer
+scoring edition after rescoring, retaining both files.
+
+An instance profile can select an edition explicitly:
+
+```json
+{
+  "benchmark_release": {
+    "path": "benchmark-release.json",
+    "sha256": "<SHA-256 of the selected sidecar>"
+  }
+}
+```
+
+Paths are relative to the external profile. The existing build copies this
+edition into the report manifest and package and displays a link beside the
+home-page scope. Checks reject mismatched populations, acceptance rules,
+scoring sources, missing sidecars and incorrect version labels. Historical
+hash-bound reports remain usable without inventing a version label. An
+explicit historical edition preserves the saved scorer identity during a
+presentation rebuild; it does not run the current scorer over old results.
+
+Reuse unchanged saved outputs for a minor correction, including prior
+successes, then recompute scores under one edition. Use the existing revision
+receipts; do not overwrite the older report. Major editions may retain
+diagnostic task-overlap analyses, but their complete throughputs/pass rates
+are separate measurements. Latency still requires matching runtime controls
+and the existing same-window comparison contract.
+
 ## Report instances
 
 Each report track uses its maintained build, check and package entry points.
diff --git a/docs/report-hosting.md b/docs/report-hosting.md
index 848f1401e..2a539bae4 100644
--- a/docs/report-hosting.md
+++ b/docs/report-hosting.md
@@ -11,6 +11,13 @@ Make targets. A new instance reuses the same commands with different inputs.
 Historical full-cohort formats use an explicit `--authority-root` bundle that
 retains their original relative paths and hash-bound profile bytes.
 
+An optional `benchmark_release` path/hash binding in the external profile fixes
+the selected benchmark edition. Builds and packages retain
+`benchmark-release.json`; presentation rebuilds keep its dataset and scoring
+identity. See [benchmark editions](benchmarks.md#benchmark-editions) for the
+version and comparison rules. A newer task package must not silently update
+an older report.
+
 ## Report languages
 
 The Web Fetch CLI defaults to English. `run --language zh` generates Chinese only; `report <run> --language en` rebuilds both editions and opens English. A rebuild without `--language` keeps the saved choice, including Chinese-only reports created by older versions.
diff --git a/evaluation/lexbench-webfetch.json b/evaluation/lexbench-webfetch.json
index 56f98bcff..8f9d07116 100644
--- a/evaluation/lexbench-webfetch.json
+++ b/evaluation/lexbench-webfetch.json
@@ -2,7 +2,7 @@
   "schema_version": 1,
   "kind": "external_benchmark_binding",
   "repository": "https://github.com/lexmount/lexbench-webfetch",
-  "revision": "6bd263ed98030a883e66ac75a47ac4e7fbae4ad1",
+  "revision": "98dd3423a5d179b4662c27beccae0918b94bc2e8",
   "ownership": {
     "benchmark": "Task populations, task model, acceptance scoring, file checks, compiler and frozen definitions",
     "harness": "Execution, reference lifecycle, evidence and reports"
diff --git a/harness/web_fetch/certification.py b/harness/web_fetch/certification.py
index 9a46ab084..617d68045 100644
--- a/harness/web_fetch/certification.py
+++ b/harness/web_fetch/certification.py
@@ -140,6 +140,8 @@ def main(argv: list[str] | None = None) -> None:
         return
     parser = argparse.ArgumentParser(description=__doc__)
     parser.add_argument("--dataset", type=Path, default=ROOT / "evaluation/dataset.json")
+    parser.add_argument("--benchmark-release", type=Path,
+                        help="Frozen benchmark edition; checked before any external calls")
     parser.add_argument("--output", type=Path, required=True)
     parser.add_argument("--work-root", type=Path, required=True)
     parser.add_argument("--providers", required=True)
@@ -215,6 +217,16 @@ def main(argv: list[str] | None = None) -> None:
             publication_state="diagnostic",
             unqualified_reference_cases=sorted(unqualified),
         )
+    release_raw = None
+    release_path = args.benchmark_release or args.dataset.parent / 'benchmark-release.json'
+    if args.benchmark_release or release_path.exists():
+        from lexbench_webfetch.scoring import SCORING_SOURCE_SHA256
+        from lexbench_webfetch.versions import verify_release
+        if release_path.is_symlink():
+            parser.error('Benchmark release rejects symlinks')
+        release_raw = release_path.read_bytes()
+        verify_release(json.loads(release_raw), dataset, exact_dataset=False,
+                       scoring_source_sha256=SCORING_SOURCE_SHA256)
     # Reject malformed execution contracts before doing any external captures.
     from harness.web_fetch.reference_lifecycle import (
         requires_browser_observation,
@@ -320,6 +332,13 @@ def main(argv: list[str] | None = None) -> None:
     if args.provider_config:
         benchmark.extend(["--provider-config", str(args.provider_config)])
     run(benchmark_runner.main, benchmark)
+    if release_raw is not None:
+        (args.output / 'benchmark-input-release.json').write_bytes(release_raw)
+        execution_path = args.output / 'execution.json'
+        execution = json.loads(execution_path.read_text())
+        execution['aggregate']['benchmark_release'] = {
+            'path': 'benchmark-input-release.json', 'sha256': hashlib.sha256(release_raw).hexdigest()}
+        execution_path.write_text(json.dumps(execution, ensure_ascii=False, indent=2) + '\n')
 
     # Re-capture HTTP evidence after every provider call has completed, then
     # choose browser fallbacks from that post-call evidence. Reusing the
diff --git a/harness/web_fetch/published_snapshot.py b/harness/web_fetch/published_snapshot.py
index 2ece93fa0..2cef0b001 100644
--- a/harness/web_fetch/published_snapshot.py
+++ b/harness/web_fetch/published_snapshot.py
@@ -64,6 +64,11 @@ def validate_snapshot_audit(source, profile_path):
     records = load_jsonl(inputs['results'])
     receipt = json.loads(inputs['receipt'].read_text())
     audit = json.loads(inputs['review'].read_text())
+    scoring_source = SCORING_SOURCE_SHA256
+    if config.get('benchmark_release'):
+        from .report_version import load_release
+        _, release = load_release(source, dataset, records, profile=profile_path)
+        scoring_source = release['scoring_source_sha256']
     old_cases = {row['id']: row for row in baseline['cases']}
     cases = {row['id']: row for row in dataset['cases']}
     old_rows = {row['attempt_id']: row for row in previous}
@@ -75,7 +80,7 @@ def validate_snapshot_audit(source, profile_path):
     if (receipt.get('kind') != 'published_snapshot_rescore' or receipt.get('provider_calls_added') != 0
             or receipt.get('input_dataset_sha256') != file_sha256(source / 'dataset.json')
             or receipt.get('input_results_sha256') != file_sha256(source / 'results.jsonl')
-            or receipt.get('scoring_source_sha256') != SCORING_SOURCE_SHA256
+            or receipt.get('scoring_source_sha256') != scoring_source
             or receipt.get('corrected_dataset_sha256') != canonical_json_sha256(dataset)):
         raise ValueError('Snapshot audit does not bind the saved calls and current scorer')
     correction_provenance(**config['correction_provenance'])
@@ -96,7 +101,7 @@ def validate_snapshot_audit(source, profile_path):
         if (view['path'] != old['artifact_path'] or view['published_sha256'] != file_sha256(artifact)
                 or view['original_sha256'] != old['metrics']['content_sha256']
                 or row['metrics']['content_sha256'] != view['scored_sha256']
-                or row['metrics']['scoring_source_sha256'] != SCORING_SOURCE_SHA256
+                or row['metrics']['scoring_source_sha256'] != scoring_source
                 or row.get('scoring', {}).get('artifact_view') != view['view']):
             raise ValueError('Snapshot audit artifact identity changed: ' + identifier)
         expected = {'original': view['original_sha256'], 'recovered_original': view['original_sha256'],
@@ -212,6 +217,8 @@ def build_snapshot_audit(source, output, *, profile):
         (staging / 'dataset.json').write_text(json.dumps(dataset, ensure_ascii=False, indent=2)+'\n')
         shutil.copy2(inputs['results'], staging / 'results.jsonl')
         batches = json.loads((staging / 'execution.json').read_text())['batches']
+        from .report_version import bind_release
+        release_binding = bind_release(source, staging, dataset, records, profile=profile)
         generate(staging, dataset, records, batches, authored_interpretation=True, frozen_scores=True)
         write_frozen_result_summary(staging, dataset, records)
         (staging / '.generated-report').write_text('webfetch\n')
@@ -221,6 +228,8 @@ def build_snapshot_audit(source, output, *, profile):
             'results_sha256':file_sha256(staging/'results.jsonl'),
             'generated_languages':list(generated_languages(staging)),
             'score_audit':comparison['score_audit'], 'profile_sha256':file_sha256(profile)}
+        if release_binding:
+            manifest['benchmark_release'] = release_binding
         (staging/'report-manifest.json').write_text(json.dumps(manifest,ensure_ascii=False,indent=2)+'\n')
         check_snapshot_audit(staging)
         _install_report_tree(staging, output)
@@ -241,6 +250,8 @@ def check_snapshot_audit(root):
             raise ValueError('Snapshot audit input changed: '+name)
     dataset = json.loads((root/'dataset.json').read_text())
     rows = load_jsonl(root/'results.jsonl')
+    from .report_version import check_release
+    check_release(root, dataset, rows)
     if dataset.get('definition_revision'):
         revision_path=_bound(root,dataset['definition_revision'])
         historical=root/'audit/history/dataset.json'
diff --git a/harness/web_fetch/report.py b/harness/web_fetch/report.py
index f96ae5b93..d9e62a1ef 100644
--- a/harness/web_fetch/report.py
+++ b/harness/web_fetch/report.py
@@ -2215,6 +2215,10 @@ def generate(out:Path,dataset:dict,records:list,manifests:list,*,include_english
   description=f'{len(cases)} 个案例 · {len(names)} 种方法 · {track_label} 类任务。'
  if local_pass_rate_ranking(dataset):
   description=model_text(f'{len(cases)} 个案例 · {len(names)} 种方法 · 按本轮任务通过率排序。',f'{len(cases)} cases · {len(names)} methods · Ordered by task pass rate for this sample.')
+ release_path=out/'benchmark-release.json'
+ if release_path.is_file():
+  release=json.loads(release_path.read_text())
+  description+=' · <a data-benchmark-release="'+esc(release['version'])+'" href="benchmark-release.json">'+esc(release['benchmark']+' v'+release['version'])+'</a>'
  home='<div class="home-page">'+hero('Web Fetch 评测报告',description)
  from .report_explainer import render_explainer
  TITLE='本轮调用链检查与日志范围' if dataset.get('comparison',{}).get('call_chain_audit') else '本报告的任务范围与验收规则'
diff --git a/harness/web_fetch/report_build.py b/harness/web_fetch/report_build.py
index c0765fbcb..84766f889 100644
--- a/harness/web_fetch/report_build.py
+++ b/harness/web_fetch/report_build.py
@@ -250,7 +250,7 @@ def _install_report_tree(staging: Path, output: Path):
 
 def _build_frozen_presentation(source, output, presentation_binding, *,
                                include_english, language, authored_interpretation,
-                               show_local_status_banners):
+                               show_local_status_banners, profile_path=None):
     """Render a bound historical report using saved scores and public evidence."""
     from harness.report_languages import select_report_language
     from .safety import evidence_tree_identity, copy_evidence_file
@@ -291,6 +291,10 @@ def _build_frozen_presentation(source, output, presentation_binding, *,
                 if relative in before:
                     metadata["public_" + kind + "_path"] = relative
                     metadata["public_" + kind + "_sha256"] = before[relative]
+        from .report_version import bind_release
+        release_binding = bind_release(source, staging, dataset, records, profile=profile_path)
+        if release_binding:
+            manifest['benchmark_release'] = release_binding
         generate(staging, dataset, records, batches, include_english=include_english,
                  authored_interpretation=authored_interpretation,
                  show_local_status_banners=show_local_status_banners, frozen_scores=True)
@@ -380,7 +384,7 @@ def build_report(source: Path, output: Path, *, include_english: bool = False,
         return _build_frozen_presentation(
             source, output, presentation_binding, include_english=include_english,
             language=language, authored_interpretation=authored_interpretation,
-            show_local_status_banners=show_local_status_banners)
+            show_local_status_banners=show_local_status_banners, profile_path=profile_path)
     dataset_probe = json.loads(contained_path(source, 'dataset.json').read_text())
     diagnostic_preview = (diagnostic_preview and
         dataset_probe.get('comparison', {}).get('kind') == 'web_fetch_diagnostic')
@@ -771,6 +775,8 @@ def build_report(source: Path, output: Path, *, include_english: bool = False,
             }, ensure_ascii=False, indent=2))
         from .report_reading import copy_archived_reading
         copy_archived_reading(source,staging)
+        from .report_version import bind_release
+        release_binding = bind_release(source, staging, dataset, records, profile=profile_path)
         generate(staging,dataset,records,manifests,include_english=include_english,
                  authored_interpretation=authored_interpretation,
                  show_local_status_banners=show_local_status_banners)
@@ -782,6 +788,8 @@ def build_report(source: Path, output: Path, *, include_english: bool = False,
             'dataset_sha256': file_sha256(staging/'dataset.json'),
             'results_sha256': file_sha256(staging/'results.jsonl'),
             'generated_languages': generated_languages(staging)}
+        if release_binding:
+            template_manifest['benchmark_release'] = release_binding
         if dataset.get('comparison', {}).get('kind') == 'uniform_native_local_measurement':
             template_manifest.update(evidence_state='local_measurement', certified=False,
                 languages=template_manifest['generated_languages'], provider_calls_added=0)
diff --git a/harness/web_fetch/report_check.py b/harness/web_fetch/report_check.py
index c895e0914..4e007e15b 100644
--- a/harness/web_fetch/report_check.py
+++ b/harness/web_fetch/report_check.py
@@ -300,6 +300,10 @@ def check(root: Path, *, authority_root: Path, profile: Path | None = None, pres
         profile_path = (
             (repository / profile).resolve() if not profile.is_absolute() else profile.resolve()
         )
+    if (root / 'dataset.json').is_file() and (root / 'results.jsonl').is_file():
+        from .report_version import check_release
+        check_release(root, json.loads((root / 'dataset.json').read_text()),
+                      load_jsonl(root / 'results.jsonl'), profile=profile_path)
     if (root / 'report-manifest.json').is_file() and json.loads((root / 'report-manifest.json').read_text()).get('evidence_state') == 'published_snapshot_score_audit':
         from .published_snapshot import check_snapshot_audit
         if profile_path and file_sha256(profile_path) != json.loads((root / 'report-manifest.json').read_text()).get('profile_sha256'):
diff --git a/harness/web_fetch/report_version.py b/harness/web_fetch/report_version.py
new file mode 100644
index 000000000..42aba46cb
--- /dev/null
+++ b/harness/web_fetch/report_version.py
@@ -0,0 +1,128 @@
+"""Carry an explicitly selected benchmark edition through existing reports."""
+
+import hashlib
+import json
+from pathlib import Path
+
+from lexbench_webfetch.paths import contained_path
+from lexbench_webfetch.versions import verify_release
+
+FILENAME = "benchmark-release.json"
+
+
+def _input_release(root):
+    execution = root / "execution.json"
+    if not execution.is_file():
+        return None
+    binding = json.loads(execution.read_text()).get("aggregate", {}).get("benchmark_release")
+    if binding is not None:
+        path = contained_path(root, binding["path"], reject_symlinks=True)
+        if hashlib.sha256(path.read_bytes()).hexdigest() != binding.get("sha256"):
+            raise ValueError("Measured benchmark input release changed")
+    return binding
+
+
+def _profile_binding(profile):
+    if profile is None:
+        return None
+    return json.loads(Path(profile).read_text()).get("benchmark_release")
+
+
+def _verify_scores(release, dataset, records):
+    verify_release(release, dataset, exact_dataset=False)
+    if not records or any(
+        (
+            row.get("metrics", {}).get("scoring_source_sha256") != release["scoring_source_sha256"]
+            or row.get("scoring", {}).get("source_sha256", release["scoring_source_sha256"])
+            != release["scoring_source_sha256"]
+        )
+        for row in records
+    ):
+        raise ValueError("Benchmark release differs from the saved scoring sources; rescore first")
+
+
+def load_release(source, dataset, records, *, profile=None):
+    """Validate the profile's edition, otherwise the measured input's edition."""
+    binding = _profile_binding(profile)
+    if binding is not None:
+        selected = contained_path(Path(profile).parent, binding["path"], reject_symlinks=True)
+    else:
+        selected = contained_path(source, FILENAME, reject_symlinks=True)
+        if not selected.exists():
+            measured = _input_release(source)
+            if measured is None:
+                return None
+            selected = contained_path(source, measured["path"], reject_symlinks=True)
+    raw = selected.read_bytes()
+    digest = hashlib.sha256(raw).hexdigest()
+    if binding is not None and digest != binding.get("sha256"):
+        raise ValueError("Profile benchmark release changed")
+    release = json.loads(raw)
+    _verify_scores(release, dataset, records)
+    return raw, release
+
+
+def bind_release(source, output, dataset, records, *, profile=None):
+    """Carry the validated edition into the generated report."""
+    selected = load_release(source, dataset, records, profile=profile)
+    if selected is None:
+        return None
+    raw, release = selected
+    digest = hashlib.sha256(raw).hexdigest()
+    target = output / FILENAME
+    target.write_bytes(raw)
+    measured = _input_release(source)
+    if measured is not None:
+        original = contained_path(source, measured["path"], reject_symlinks=True)
+        archived = contained_path(output, measured["path"], reject_symlinks=True)
+        archived.parent.mkdir(parents=True, exist_ok=True)
+        archived.write_bytes(original.read_bytes())
+    return {
+        "path": FILENAME,
+        "sha256": digest,
+        "benchmark": release["benchmark"],
+        "version": release["version"],
+    }
+
+
+def check_release(root, dataset, records, *, profile=None):
+    """Check the edition, rendered label and scoring identity without rescoring."""
+    from bs4 import BeautifulSoup
+
+    manifest_path = root / "report-manifest.json"
+    manifest = json.loads(manifest_path.read_text()) if manifest_path.is_file() else {}
+    binding = manifest.get("benchmark_release")
+    expected = _profile_binding(profile)
+    measured = _input_release(root)
+    release_path = contained_path(root, FILENAME, reject_symlinks=True)
+    if binding is None:
+        if expected is not None or release_path.exists() or measured is not None:
+            raise ValueError("Report lost its benchmark release binding")
+        return
+    if binding.get("path") != FILENAME or not release_path.is_file():
+        raise ValueError("Report benchmark release is missing")
+    raw = release_path.read_bytes()
+    digest = hashlib.sha256(raw).hexdigest()
+    if digest != binding.get("sha256") or (
+        expected is not None and digest != expected.get("sha256")
+    ):
+        raise ValueError("Report benchmark release changed")
+    release = json.loads(raw)
+    _verify_scores(release, dataset, records)
+    if any(binding.get(key) != release[key] for key in ("benchmark", "version")):
+        raise ValueError("Report benchmark version label differs from its release")
+    for route in ("index.html", "en/index.html"):
+        path = root / route
+        if not path.is_file():
+            continue
+        page = BeautifulSoup(path.read_text(), "html.parser")
+        if page.select_one("[data-english-not-generated]"):
+            continue
+        label = page.select_one("[data-benchmark-release]")
+        if (
+            label is None
+            or label.get("href") != ("../" if route.startswith("en/") else "") + FILENAME
+            or label.get("data-benchmark-release") != release["version"]
+            or label.get_text(" ", strip=True) != release["benchmark"] + " v" + release["version"]
+        ):
+            raise ValueError("Report lost its benchmark version label: " + route)
diff --git a/pyproject.toml b/pyproject.toml
index 687aa557f..ebe908eee 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -7,7 +7,7 @@ license = "Apache-2.0"
 license-files = ["LICENSE", "NOTICE", "THIRD_PARTY.md"]
 requires-python = ">=3.11"
 dependencies = [
-  "lexbench-webfetch @ git+https://github.com/lexmount/lexbench-webfetch.git@6bd263ed98030a883e66ac75a47ac4e7fbae4ad1",
+  "lexbench-webfetch @ git+https://github.com/lexmount/lexbench-webfetch.git@98dd3423a5d179b4662c27beccae0918b94bc2e8",
   "beautifulsoup4>=4.12,<5",
   "fastcdc==1.7.0",
 ]
diff --git a/tests/test_benchmark_release_binding.py b/tests/test_benchmark_release_binding.py
new file mode 100644
index 000000000..2dc6d5c8c
--- /dev/null
+++ b/tests/test_benchmark_release_binding.py
@@ -0,0 +1,225 @@
+"""Reports preserve an edition and reject mismatched tasks, scorers or labels."""
+
+import json
+
+import pytest
+from lexbench_webfetch.versions import freeze_release
+
+from harness.record_io import file_sha256
+from harness.web_fetch.published_snapshot import build_snapshot_audit, check_snapshot_audit
+from harness.web_fetch.report_version import bind_release, check_release
+from tests.test_published_snapshot_report import example
+
+
+def configured(tmp_path):
+    source, profile = example(tmp_path)
+    data = json.loads(profile.read_text())
+    dataset = profile.parent / data["inputs"]["dataset"]["path"]
+    release = tmp_path / "edition.json"
+    freeze_release(dataset, release, benchmark="sample-benchmark", version="1.0.0")
+    data["benchmark_release"] = {"path": release.name, "sha256": file_sha256(release)}
+    profile.write_text(json.dumps(data))
+    return source, profile, release
+
+
+def test_versioned_snapshot_uses_existing_build_check_and_package(tmp_path):
+    source, profile, release = configured(tmp_path)
+    before = {p.relative_to(source): p.read_bytes() for p in source.rglob("*") if p.is_file()}
+    output = tmp_path / "report"
+    build_snapshot_audit(source, output, profile=profile)
+    assert (output / "benchmark-release.json").read_bytes() == release.read_bytes()
+    assert 'data-benchmark-release="1.0.0"' in (output / "index.html").read_text()
+    assert "2 saved calls" in check_snapshot_audit(output)
+    dataset = json.loads((output / "dataset.json").read_text())
+    rows = [json.loads(line) for line in (output / "results.jsonl").read_text().splitlines()]
+    check_release(output, dataset, rows, profile=profile)
+    from harness.web_fetch.publication_check import check
+    from harness.web_fetch.report_package import build
+
+    archive = tmp_path / "report.zip"
+    build(output, archive, snapshot_audit=True)
+    check(archive, output, snapshot_audit=True)
+    assert before == {
+        p.relative_to(source): p.read_bytes() for p in source.rglob("*") if p.is_file()
+    }
+
+
+@pytest.mark.parametrize(
+    "defect", ["missing-file", "wrong-version", "removed-label", "wrong-scorer", "dropped-binding"]
+)
+def test_report_gate_rejects_release_drift(tmp_path, defect):
+    source, profile, _ = configured(tmp_path)
+    output = tmp_path / "report"
+    build_snapshot_audit(source, output, profile=profile)
+    if defect == "missing-file":
+        (output / "benchmark-release.json").unlink()
+    elif defect in {"wrong-version", "removed-label"}:
+        page = output / "index.html"
+        old = 'data-benchmark-release="1.0.0"'
+        page.write_text(
+            page.read_text().replace(
+                old,
+                'data-benchmark-release="2.0.0"'
+                if defect == "wrong-version"
+                else 'data-removed="1.0.0"',
+            )
+        )
+    elif defect == "wrong-scorer":
+        rows = output / "results.jsonl"
+        records = [json.loads(line) for line in rows.read_text().splitlines()]
+        records[0]["metrics"]["scoring_source_sha256"] = "0" * 64
+        rows.write_text("".join(json.dumps(row) + "\n" for row in records))
+        manifest = output / "report-manifest.json"
+        data = json.loads(manifest.read_text())
+        data["results_sha256"] = file_sha256(rows)
+        manifest.write_text(json.dumps(data))
+    else:
+        manifest = output / "report-manifest.json"
+        data = json.loads(manifest.read_text())
+        del data["benchmark_release"]
+        manifest.write_text(json.dumps(data))
+    with pytest.raises(ValueError, match="[Bb]enchmark|scoring sources"):
+        check_snapshot_audit(output)
+
+
+def test_wrong_task_edition_is_rejected_before_rendering(tmp_path):
+    source, profile, release = configured(tmp_path)
+    data = json.loads(profile.read_text())
+    dataset = json.loads((profile.parent / data["inputs"]["dataset"]["path"]).read_text())
+    dataset["cases"][0]["task_prompt"] = "Return a different scope."
+    records = [
+        json.loads(line)
+        for line in (profile.parent / data["inputs"]["results"]["path"]).read_text().splitlines()
+    ]
+    with pytest.raises(ValueError, match="dataset"):
+        bind_release(source, tmp_path, dataset, records, profile=profile)
+    assert release.is_file()
+
+
+def test_legacy_hash_bound_reports_do_not_acquire_an_invented_version(tmp_path):
+    source, profile = example(tmp_path)
+    output = tmp_path / "report"
+    build_snapshot_audit(source, output, profile=profile)
+    assert not (output / "benchmark-release.json").exists()
+    assert "data-benchmark-release" not in (output / "index.html").read_text()
+    assert "2 saved calls" in check_snapshot_audit(output)
+
+
+def test_wrong_edition_stops_before_capture_or_provider_calls(tmp_path, monkeypatch):
+    from lexbench_webfetch.scoring import SCORING_SOURCE_SHA256
+
+    from harness.web_fetch.certification import main
+
+    source = tmp_path / "dataset.json"
+    data = {
+        "cases": [
+            {
+                "id": "sample",
+                "url": "https://example.test/",
+                "task_prompt": "Return the complete article.",
+            }
+        ]
+    }
+    source.write_text(json.dumps(data))
+    edition = tmp_path / "benchmark-release.json"
+    freeze_release(source, edition, benchmark="sample-benchmark", version="1.0.0")
+    data["cases"][0]["task_prompt"] = "Return only its title."
+    source.write_text(json.dumps(data))
+    calls = []
+    monkeypatch.setattr("harness.web_fetch.certification.run", lambda *args: calls.append(args))
+    with pytest.raises(ValueError, match="dataset"):
+        main(
+            [
+                "--dataset",
+                str(source),
+                "--output",
+                str(tmp_path / "run"),
+                "--work-root",
+                str(tmp_path / "work"),
+                "--providers",
+                "fixture",
+                "--case-ids",
+                "sample",
+            ]
+        )
+    assert not calls
+    assert not (tmp_path / "work").exists()
+    assert SCORING_SOURCE_SHA256
+
+
+def test_explicit_old_edition_preserves_historical_scores_with_new_renderer(tmp_path):
+    source, profile, edition = configured(tmp_path)
+    data = json.loads(profile.read_text())
+    old_source = "a" * 64
+    result_path = profile.parent / data["inputs"]["results"]["path"]
+    rows = [json.loads(line) for line in result_path.read_text().splitlines()]
+    for row in rows:
+        row["metrics"]["scoring_source_sha256"] = old_source
+        row.setdefault("scoring", {})["source_sha256"] = old_source
+    result_path.write_text("".join(json.dumps(row) + "\n" for row in rows))
+    data["inputs"]["results"]["sha256"] = file_sha256(result_path)
+    receipt_path = profile.parent / data["inputs"]["receipt"]["path"]
+    receipt = json.loads(receipt_path.read_text())
+    receipt["scoring_source_sha256"] = old_source
+    receipt_path.write_text(json.dumps(receipt))
+    data["inputs"]["receipt"]["sha256"] = file_sha256(receipt_path)
+    release = json.loads(edition.read_text())
+    release["scoring_source_sha256"] = old_source
+    edition.write_text(json.dumps(release))
+    data["benchmark_release"]["sha256"] = file_sha256(edition)
+    from harness.record_io import canonical_json_sha256
+    from harness.web_fetch.report_explainer import _headless_article_input
+
+    dataset_path = profile.parent / data["inputs"]["dataset"]["path"]
+    cases = {case["id"]: case for case in json.loads(dataset_path.read_text())["cases"]}
+    article_path = profile.parent / data["inputs"]["interpretation"]["path"]
+    article = json.loads(article_path.read_text())
+    article["evidence_sha256"] = canonical_json_sha256(
+        _headless_article_input(cases, rows, ["chromium"])
+    )
+    article_path.write_text(json.dumps(article))
+    data["inputs"]["interpretation"]["sha256"] = file_sha256(article_path)
+    profile.write_text(json.dumps(data))
+    output = tmp_path / "historical-report"
+    build_snapshot_audit(source, output, profile=profile)
+    assert (output / "results.jsonl").read_bytes() == result_path.read_bytes()
+    assert "2 saved calls" in check_snapshot_audit(output)
+
+
+def test_rescoring_edition_keeps_the_original_call_edition(tmp_path):
+    from lexbench_webfetch.versions import create_release
+
+    source, profile, edition = configured(tmp_path)
+    data = json.loads(profile.read_text())
+    dataset = json.loads((profile.parent / data["inputs"]["dataset"]["path"]).read_text())
+    rows = [
+        json.loads(line)
+        for line in (profile.parent / data["inputs"]["results"]["path"]).read_text().splitlines()
+    ]
+    original = create_release(
+        dataset, benchmark="sample-benchmark", version="1.0.0", scoring_source_sha256="f" * 64
+    )
+    original_path = source / "benchmark-input-release.json"
+    original_path.write_text(json.dumps(original))
+    execution = json.loads((source / "execution.json").read_text())
+    execution["aggregate"] = {
+        "benchmark_release": {"path": original_path.name, "sha256": file_sha256(original_path)}
+    }
+    (source / "execution.json").write_text(json.dumps(execution))
+    corrected = create_release(
+        dataset,
+        benchmark="sample-benchmark",
+        version="1.1.0",
+        previous=original,
+        change_note="Correct the saved-response scorer",
+    )
+    edition.write_text(json.dumps(corrected))
+    data["benchmark_release"]["sha256"] = file_sha256(edition)
+    profile.write_text(json.dumps(data))
+    output = tmp_path / "revised"
+    output.mkdir()
+    selected = bind_release(source, output, dataset, rows, profile=profile)
+    assert selected["version"] == "1.1.0"
+    assert json.loads((output / "benchmark-release.json").read_text()) == corrected
+    assert (output / original_path.name).read_bytes() == original_path.read_bytes()
+    assert json.loads(original_path.read_text())["version"] == "1.0.0"
diff --git a/tests/test_local_audit.py b/tests/test_local_audit.py
index a837e7f83..67346246b 100644
--- a/tests/test_local_audit.py
+++ b/tests/test_local_audit.py
@@ -285,6 +285,9 @@ def test_new_local_matrix_is_complete_and_keeps_logs_without_claiming_review(tmp
     _write(contract_path, contract)
     dataset["comparison"]["local_run_contract_sha256"] = _digest(contract_path)
     _write(evidence / "dataset.json", dataset)
+    from lexbench_webfetch.versions import freeze_release
+    freeze_release(evidence / "dataset.json", evidence / "benchmark-release.json",
+                   benchmark="local-fixture", version="1.0.0")
     monkeypatch.setattr(local_execution_contract, "ROOT", tmp_path / "repo")
     assert local_audit.require_local_audit(evidence, dataset, records) is None
     changed_contract = {**contract, 'timeout_seconds': timeout + 1}
@@ -293,6 +296,8 @@ def test_new_local_matrix_is_complete_and_keeps_logs_without_claiming_review(tmp
             evidence, dataset, records, changed_contract, {'batches': [batch]})
     output = tmp_path / "report"
     build_report(evidence, output, local_audit=True, include_english=include_english)
+    from harness.web_fetch.report_version import check_release
+    check_release(output, json.loads((output / "dataset.json").read_text()), records)
     home = (output / "index.html").read_text()
     assert "本地未认证诊断快照" in home
     assert "查看调用链问题和日志留存范围" not in home
@@ -345,6 +350,10 @@ def test_new_local_matrix_is_complete_and_keeps_logs_without_claiming_review(tmp
     (output / "results.jsonl").write_text("".join(json.dumps(row) + "\n" for row in frozen_rows))
     manifest = json.loads((output / "report-manifest.json").read_text())
     manifest["results_sha256"] = _digest(output / "results.jsonl")
+    historical_edition = json.loads((output / "benchmark-release.json").read_text())
+    historical_edition["scoring_source_sha256"] = historical_scorer
+    _write(output / "benchmark-release.json", historical_edition)
+    manifest["benchmark_release"]["sha256"] = _digest(output / "benchmark-release.json")
     manifest["template"] = {"renderer": "historical-template"}
     _write(output / "report-manifest.json", manifest)
     binding = {"comparison_kind": "uniform_native_local_measurement",
diff --git a/tests/test_run_certified_benchmark.py b/tests/test_run_certified_benchmark.py
index bde4e5353..6c768064b 100644
--- a/tests/test_run_certified_benchmark.py
+++ b/tests/test_run_certified_benchmark.py
@@ -148,6 +148,9 @@ def test_frozen_contract_execution_needs_no_historical_qualification_files(self)
                             'evidence_sha256': 'b' * 64}}
             dataset.write_text(json.dumps({'tool': 'web_fetch', 'comparison': {}, 'cases': [case, fixed, resource]}))
             original = dataset.read_bytes()
+            from lexbench_webfetch.versions import freeze_release
+            edition = root / 'benchmark-release.json'
+            freeze_release(dataset, edition, benchmark='frozen-example', version='1.0.0')
             config = root / 'providers.json'
             config.write_text(json.dumps({'providers': {'fixture': {'type': 'fixture'}}}))
             (root / 'evaluation/list-references').mkdir(parents=True)
@@ -175,6 +178,9 @@ def execute(stage, arguments):
             measured = json.loads((output / 'dataset.json').read_text())
             self.assertEqual(measured['cases'], [case, fixed, resource])
             self.assertEqual(dataset.read_bytes(), original)
+            self.assertEqual((output / 'benchmark-input-release.json').read_bytes(), edition.read_bytes())
+            release_binding = json.loads((output / 'execution.json').read_text())['aggregate']['benchmark_release']
+            self.assertEqual(release_binding['sha256'], hashlib.sha256(edition.read_bytes()).hexdigest())
             self.assertEqual(measured['comparison']['source_reference_policy'],
                              'frozen_benchmark_inputs_with_explicit_live_exact_checks')
             batches = json.loads((output / 'execution.json').read_text())['batches']
diff --git a/uv.lock b/uv.lock
index 2cc83e2d8..789cc6a0d 100644
--- a/uv.lock
+++ b/uv.lock
@@ -354,7 +354,7 @@ requires-dist = [
     { name = "jieba", marker = "extra == 'webmainbench-eval'", specifier = "==0.42.1" },
     { name = "json-repair", marker = "extra == 'browser-use-eval'", specifier = ">=0.54,<1" },
     { name = "langchain-community", marker = "extra == 'oss-eval'", specifier = "==0.3.30" },
-    { name = "lexbench-webfetch", git = "https://github.com/lexmount/lexbench-webfetch.git?rev=6bd263ed98030a883e66ac75a47ac4e7fbae4ad1" },
+    { name = "lexbench-webfetch", git = "https://github.com/lexmount/lexbench-webfetch.git?rev=98dd3423a5d179b4662c27beccae0918b94bc2e8" },
     { name = "markdown", marker = "extra == 'webmainbench-eval'", specifier = ">=3.7,<4" },
     { name = "openai", marker = "extra == 'webmainbench-eval'", specifier = ">=2.8,<4" },
     { name = "openpyxl", marker = "extra == 'oss-eval'", specifier = "==3.1.5" },
@@ -1933,7 +1933,7 @@ wheels = [
 [[package]]
 name = "lexbench-webfetch"
 version = "0.3.0"
-source = { git = "https://github.com/lexmount/lexbench-webfetch.git?rev=6bd263ed98030a883e66ac75a47ac4e7fbae4ad1#6bd263ed98030a883e66ac75a47ac4e7fbae4ad1" }
+source = { git = "https://github.com/lexmount/lexbench-webfetch.git?rev=98dd3423a5d179b4662c27beccae0918b94bc2e8#98dd3423a5d179b4662c27beccae0918b94bc2e8" }
 dependencies = [
     { name = "beautifulsoup4" },
     { name = "soupsieve" },

From 69829cec602e491c25ff0a1f7952d70e7b1bf270 Mon Sep 17 00:00:00 2001
From: lanyue-llk <270302213+lanyue-llk@users.noreply.github.com>
Date: Fri, 23 Oct 2026 16:02:20 +0800
Subject: [PATCH 2/5] build(deps): pin benchmark versioning after main
 integration

---
 evaluation/lexbench-webfetch.json | 2 +-
 pyproject.toml                    | 2 +-
 uv.lock                           | 4 ++--
 3 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/evaluation/lexbench-webfetch.json b/evaluation/lexbench-webfetch.json
index 8f9d07116..005ac5007 100644
--- a/evaluation/lexbench-webfetch.json
+++ b/evaluation/lexbench-webfetch.json
@@ -2,7 +2,7 @@
   "schema_version": 1,
   "kind": "external_benchmark_binding",
   "repository": "https://github.com/lexmount/lexbench-webfetch",
-  "revision": "98dd3423a5d179b4662c27beccae0918b94bc2e8",
+  "revision": "04c2b99e0f86aa0264976ad4159bc4e45b2b10ab",
   "ownership": {
     "benchmark": "Task populations, task model, acceptance scoring, file checks, compiler and frozen definitions",
     "harness": "Execution, reference lifecycle, evidence and reports"
diff --git a/pyproject.toml b/pyproject.toml
index ebe908eee..b05922db2 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -7,7 +7,7 @@ license = "Apache-2.0"
 license-files = ["LICENSE", "NOTICE", "THIRD_PARTY.md"]
 requires-python = ">=3.11"
 dependencies = [
-  "lexbench-webfetch @ git+https://github.com/lexmount/lexbench-webfetch.git@98dd3423a5d179b4662c27beccae0918b94bc2e8",
+  "lexbench-webfetch @ git+https://github.com/lexmount/lexbench-webfetch.git@04c2b99e0f86aa0264976ad4159bc4e45b2b10ab",
   "beautifulsoup4>=4.12,<5",
   "fastcdc==1.7.0",
 ]
diff --git a/uv.lock b/uv.lock
index 789cc6a0d..1d046ca88 100644
--- a/uv.lock
+++ b/uv.lock
@@ -354,7 +354,7 @@ requires-dist = [
     { name = "jieba", marker = "extra == 'webmainbench-eval'", specifier = "==0.42.1" },
     { name = "json-repair", marker = "extra == 'browser-use-eval'", specifier = ">=0.54,<1" },
     { name = "langchain-community", marker = "extra == 'oss-eval'", specifier = "==0.3.30" },
-    { name = "lexbench-webfetch", git = "https://github.com/lexmount/lexbench-webfetch.git?rev=98dd3423a5d179b4662c27beccae0918b94bc2e8" },
+    { name = "lexbench-webfetch", git = "https://github.com/lexmount/lexbench-webfetch.git?rev=04c2b99e0f86aa0264976ad4159bc4e45b2b10ab" },
     { name = "markdown", marker = "extra == 'webmainbench-eval'", specifier = ">=3.7,<4" },
     { name = "openai", marker = "extra == 'webmainbench-eval'", specifier = ">=2.8,<4" },
     { name = "openpyxl", marker = "extra == 'oss-eval'", specifier = "==3.1.5" },
@@ -1933,7 +1933,7 @@ wheels = [
 [[package]]
 name = "lexbench-webfetch"
 version = "0.3.0"
-source = { git = "https://github.com/lexmount/lexbench-webfetch.git?rev=98dd3423a5d179b4662c27beccae0918b94bc2e8#98dd3423a5d179b4662c27beccae0918b94bc2e8" }
+source = { git = "https://github.com/lexmount/lexbench-webfetch.git?rev=04c2b99e0f86aa0264976ad4159bc4e45b2b10ab#04c2b99e0f86aa0264976ad4159bc4e45b2b10ab" }
 dependencies = [
     { name = "beautifulsoup4" },
     { name = "soupsieve" },

From 8086c561ebb087e3fbf7ff100baeb143d7d72f22 Mon Sep 17 00:00:00 2001
From: lanyue-llk <270302213+lanyue-llk@users.noreply.github.com>
Date: Fri, 23 Oct 2026 16:07:31 +0800
Subject: [PATCH 3/5] build(deps): bind the published benchmark catalog edition

---
 evaluation/lexbench-webfetch.json | 2 +-
 pyproject.toml                    | 2 +-
 uv.lock                           | 4 ++--
 3 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/evaluation/lexbench-webfetch.json b/evaluation/lexbench-webfetch.json
index 005ac5007..af2a852a3 100644
--- a/evaluation/lexbench-webfetch.json
+++ b/evaluation/lexbench-webfetch.json
@@ -2,7 +2,7 @@
   "schema_version": 1,
   "kind": "external_benchmark_binding",
   "repository": "https://github.com/lexmount/lexbench-webfetch",
-  "revision": "04c2b99e0f86aa0264976ad4159bc4e45b2b10ab",
+  "revision": "56aca0cb9b9b9eef9995f0168e0504a72e309cd1",
   "ownership": {
     "benchmark": "Task populations, task model, acceptance scoring, file checks, compiler and frozen definitions",
     "harness": "Execution, reference lifecycle, evidence and reports"
diff --git a/pyproject.toml b/pyproject.toml
index b05922db2..0d9114012 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -7,7 +7,7 @@ license = "Apache-2.0"
 license-files = ["LICENSE", "NOTICE", "THIRD_PARTY.md"]
 requires-python = ">=3.11"
 dependencies = [
-  "lexbench-webfetch @ git+https://github.com/lexmount/lexbench-webfetch.git@04c2b99e0f86aa0264976ad4159bc4e45b2b10ab",
+  "lexbench-webfetch @ git+https://github.com/lexmount/lexbench-webfetch.git@56aca0cb9b9b9eef9995f0168e0504a72e309cd1",
   "beautifulsoup4>=4.12,<5",
   "fastcdc==1.7.0",
 ]
diff --git a/uv.lock b/uv.lock
index 1d046ca88..30b9e9a1b 100644
--- a/uv.lock
+++ b/uv.lock
@@ -354,7 +354,7 @@ requires-dist = [
     { name = "jieba", marker = "extra == 'webmainbench-eval'", specifier = "==0.42.1" },
     { name = "json-repair", marker = "extra == 'browser-use-eval'", specifier = ">=0.54,<1" },
     { name = "langchain-community", marker = "extra == 'oss-eval'", specifier = "==0.3.30" },
-    { name = "lexbench-webfetch", git = "https://github.com/lexmount/lexbench-webfetch.git?rev=04c2b99e0f86aa0264976ad4159bc4e45b2b10ab" },
+    { name = "lexbench-webfetch", git = "https://github.com/lexmount/lexbench-webfetch.git?rev=56aca0cb9b9b9eef9995f0168e0504a72e309cd1" },
     { name = "markdown", marker = "extra == 'webmainbench-eval'", specifier = ">=3.7,<4" },
     { name = "openai", marker = "extra == 'webmainbench-eval'", specifier = ">=2.8,<4" },
     { name = "openpyxl", marker = "extra == 'oss-eval'", specifier = "==3.1.5" },
@@ -1933,7 +1933,7 @@ wheels = [
 [[package]]
 name = "lexbench-webfetch"
 version = "0.3.0"
-source = { git = "https://github.com/lexmount/lexbench-webfetch.git?rev=04c2b99e0f86aa0264976ad4159bc4e45b2b10ab#04c2b99e0f86aa0264976ad4159bc4e45b2b10ab" }
+source = { git = "https://github.com/lexmount/lexbench-webfetch.git?rev=56aca0cb9b9b9eef9995f0168e0504a72e309cd1#56aca0cb9b9b9eef9995f0168e0504a72e309cd1" }
 dependencies = [
     { name = "beautifulsoup4" },
     { name = "soupsieve" },

From 4221ea42adf369f73b2d69c8d7b8cecf40a103a3 Mon Sep 17 00:00:00 2001
From: lanyue-llk <270302213+lanyue-llk@users.noreply.github.com>
Date: Fri, 23 Oct 2026 16:11:19 +0800
Subject: [PATCH 4/5] build(deps): hash-bind the published benchmark release
 manifest

---
 evaluation/lexbench-webfetch.json | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/evaluation/lexbench-webfetch.json b/evaluation/lexbench-webfetch.json
index af2a852a3..a50f6c973 100644
--- a/evaluation/lexbench-webfetch.json
+++ b/evaluation/lexbench-webfetch.json
@@ -80,6 +80,7 @@
     "evaluation/risk-control/source-inputs.json": "77c5a4b62c07d59ed57154a833e85f9e7a97bc739730f51cd4ca3a698edb9258",
     "evaluation/scoring-regressions.json": "432ddabe23d045d0ce4c23aef69ca3337ba4d240fe3c2061932cee26b7dc00dc",
     "evaluation/source-inputs.json": "c202abc4624811b938a837402bbaf0ecbe716e375fbca70567346d2edb3b1708",
-    "annotation-revisions/serialization-20261021.json": "e0fe42179e584abb73782cbc459f521cab9cb09d3442de1ec4a6ff54d1b1235a"
+    "annotation-revisions/serialization-20261021.json": "e0fe42179e584abb73782cbc459f521cab9cb09d3442de1ec4a6ff54d1b1235a",
+    "benchmark-release.json": "75fde8db11bb8b3b8b1460fea0e78385bd2a0bdddecc84fbae8287df5d25e879"
   }
 }

From c32adca4710b101415f4368a544e46d54b8615e9 Mon Sep 17 00:00:00 2001
From: lanyue-llk <270302213+lanyue-llk@users.noreply.github.com>
Date: Fri, 23 Oct 2026 17:46:11 +0800
Subject: [PATCH 5/5] fix(reporting): preserve and enforce benchmark edition
 bindings

---
 docs/benchmarks.md                    |  6 ++-
 evaluation/lexbench-webfetch.json     |  4 +-
 harness/web_fetch/report.py           |  2 +-
 harness/web_fetch/report_build.py     |  2 +
 harness/web_fetch/report_package.py   |  8 ++--
 harness/web_fetch/report_version.py   | 21 ++++++++--
 pyproject.toml                        |  2 +-
 tests/test_external_local_report.py   | 57 +++++++++++++++++++++++++++
 tests/test_webfetch_report_package.py | 51 ++++++++++++++++++++++++
 uv.lock                               |  4 +-
 10 files changed, 143 insertions(+), 14 deletions(-)

diff --git a/docs/benchmarks.md b/docs/benchmarks.md
index 71e8d5d50..113e435ff 100644
--- a/docs/benchmarks.md
+++ b/docs/benchmarks.md
@@ -57,6 +57,10 @@ Use a benchmark version together with exact task/reference and scoring-source
 hashes. Keep the package's full Git commit pinned separately. A package update
 does not select a new dataset for an existing report.
 
+Use the same `benchmark-vMAJOR.MINOR.PATCH` tag prefix in both repositories.
+Reports reference their benchmark edition; report rebuilds use the hosting
+platform's version ID rather than a separate semantic-version sequence.
+
 | Change | Edition | Result comparison |
 | --- | --- | --- |
 | Documentation or presentation only | Patch | Direct when execution controls also match |
@@ -64,7 +68,7 @@ does not select a new dataset for an existing report.
 | Added/removed cases, changed requests, output scope or routes | Major | Separate full-population results |
 
 The task package owns `lexbench-webfetch release` and `compare-releases`.
-See its [version contract](https://github.com/lexmount/lexbench-webfetch/blob/98dd3423a5d179b4662c27beccae0918b94bc2e8/docs/versions.md).
+See its [version contract](https://github.com/lexmount/lexbench-webfetch/blob/benchmark-v2.0.1/docs/versions.md).
 Freeze `benchmark-release.json` beside the compiled inputs before calls.
 The certified Web Fetch runner accepts `--benchmark-release`; it also discovers
 the sidecar beside the selected dataset. Release creation records the full scorer
diff --git a/evaluation/lexbench-webfetch.json b/evaluation/lexbench-webfetch.json
index a50f6c973..14379fbb0 100644
--- a/evaluation/lexbench-webfetch.json
+++ b/evaluation/lexbench-webfetch.json
@@ -2,7 +2,7 @@
   "schema_version": 1,
   "kind": "external_benchmark_binding",
   "repository": "https://github.com/lexmount/lexbench-webfetch",
-  "revision": "56aca0cb9b9b9eef9995f0168e0504a72e309cd1",
+  "revision": "2ea8c52df67c39eee2ea5113a63f5da34f102160",
   "ownership": {
     "benchmark": "Task populations, task model, acceptance scoring, file checks, compiler and frozen definitions",
     "harness": "Execution, reference lifecycle, evidence and reports"
@@ -81,6 +81,6 @@
     "evaluation/scoring-regressions.json": "432ddabe23d045d0ce4c23aef69ca3337ba4d240fe3c2061932cee26b7dc00dc",
     "evaluation/source-inputs.json": "c202abc4624811b938a837402bbaf0ecbe716e375fbca70567346d2edb3b1708",
     "annotation-revisions/serialization-20261021.json": "e0fe42179e584abb73782cbc459f521cab9cb09d3442de1ec4a6ff54d1b1235a",
-    "benchmark-release.json": "75fde8db11bb8b3b8b1460fea0e78385bd2a0bdddecc84fbae8287df5d25e879"
+    "benchmark-release.json": "96ac7f1044601f4a3f768be814a49a48ff46e3ad9802f7f9e17a524460b57c4b"
   }
 }
diff --git a/harness/web_fetch/report.py b/harness/web_fetch/report.py
index d9e62a1ef..b67ae17c8 100644
--- a/harness/web_fetch/report.py
+++ b/harness/web_fetch/report.py
@@ -2218,7 +2218,7 @@ def generate(out:Path,dataset:dict,records:list,manifests:list,*,include_english
  release_path=out/'benchmark-release.json'
  if release_path.is_file():
   release=json.loads(release_path.read_text())
-  description+=' · <a data-benchmark-release="'+esc(release['version'])+'" href="benchmark-release.json">'+esc(release['benchmark']+' v'+release['version'])+'</a>'
+  description='<span>'+description+'</span> · <a data-benchmark-release="'+esc(release['version'])+'" href="benchmark-release.json">'+esc(release['benchmark']+' v'+release['version'])+'</a>'
  home='<div class="home-page">'+hero('Web Fetch 评测报告',description)
  from .report_explainer import render_explainer
  TITLE='本轮调用链检查与日志范围' if dataset.get('comparison',{}).get('call_chain_audit') else '本报告的任务范围与验收规则'
diff --git a/harness/web_fetch/report_build.py b/harness/web_fetch/report_build.py
index 84766f889..b1b7fd4ff 100644
--- a/harness/web_fetch/report_build.py
+++ b/harness/web_fetch/report_build.py
@@ -1509,6 +1509,8 @@ def validate_local_presentation(output: Path, binding_path: Path):
         require_host_metadata_projection(output, binding['publication_projection'])
     dataset = json.loads((output / 'dataset.json').read_text())
     records = load_jsonl(contained_path(output, 'results.jsonl'))
+    from .report_version import check_release
+    check_release(output, dataset, records)
     if binding.get('publication_evidence_views'):
         from .publication_privacy import require_public_evidence_view_inventory
         require_public_evidence_view_inventory(output, binding['publication_evidence_views'], records)
diff --git a/harness/web_fetch/report_package.py b/harness/web_fetch/report_package.py
index 2fd9b22eb..08af4ab63 100644
--- a/harness/web_fetch/report_package.py
+++ b/harness/web_fetch/report_package.py
@@ -29,6 +29,11 @@ def _require(condition, detail):
 
 
 def validate_release(source: Path, *, presentation_binding: Path | None = None, snapshot_audit=False) -> bool:
+    from .report_version import check_release
+    dataset_path = source / "dataset.json"
+    dataset = json.loads(dataset_path.read_text()) if dataset_path.exists() else {}
+    records = load_jsonl(source / "results.jsonl")
+    check_release(source, dataset, records)
     manifest_path=source/'report-manifest.json'
     audit_state=manifest_path.is_file() and json.loads(manifest_path.read_text()).get('evidence_state')=='published_snapshot_score_audit'
     if snapshot_audit:
@@ -43,9 +48,6 @@ def validate_release(source: Path, *, presentation_binding: Path | None = None,
         from .report_build import validate_local_presentation
         validate_local_presentation(source, presentation_binding)
         return False
-    dataset_path = source / "dataset.json"
-    dataset = json.loads(dataset_path.read_text()) if dataset_path.exists() else {}
-    records = load_jsonl(source / "results.jsonl")
     from .reference_lifecycle import require_certifiable_evidence
     manifests = []
     for name in ('manifest.json', 'execution.json'):
diff --git a/harness/web_fetch/report_version.py b/harness/web_fetch/report_version.py
index 42aba46cb..d926f2474 100644
--- a/harness/web_fetch/report_version.py
+++ b/harness/web_fetch/report_version.py
@@ -12,9 +12,13 @@
 
 def _input_release(root):
     execution = root / "execution.json"
-    if not execution.is_file():
-        return None
-    binding = json.loads(execution.read_text()).get("aggregate", {}).get("benchmark_release")
+    binding = (json.loads(execution.read_text()).get("aggregate", {}).get("benchmark_release")
+               if execution.is_file() else None)
+    manifest = root / "report-manifest.json"
+    if manifest.is_file():
+        declared = (json.loads(manifest.read_text()).get("benchmark_release") or {}).get("input_release")
+        if declared is not None and declared != binding:
+            raise ValueError("Report lost its original benchmark input release binding")
     if binding is not None:
         path = contained_path(root, binding["path"], reject_symlinks=True)
         if hashlib.sha256(path.read_bytes()).hexdigest() != binding.get("sha256"):
@@ -77,12 +81,21 @@ def bind_release(source, output, dataset, records, *, profile=None):
         archived = contained_path(output, measured["path"], reject_symlinks=True)
         archived.parent.mkdir(parents=True, exist_ok=True)
         archived.write_bytes(original.read_bytes())
-    return {
+        execution_path = output / "execution.json"
+        execution = json.loads(execution_path.read_text()) if execution_path.exists() else {}
+        aggregate = execution.setdefault("aggregate", {})
+        if aggregate.get("benchmark_release") != measured:
+            aggregate["benchmark_release"] = measured
+            execution_path.write_text(json.dumps(execution, ensure_ascii=False, indent=2) + "\n")
+    binding = {
         "path": FILENAME,
         "sha256": digest,
         "benchmark": release["benchmark"],
         "version": release["version"],
     }
+    if measured is not None:
+        binding["input_release"] = measured
+    return binding
 
 
 def check_release(root, dataset, records, *, profile=None):
diff --git a/pyproject.toml b/pyproject.toml
index 0d9114012..92d2ceb82 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -7,7 +7,7 @@ license = "Apache-2.0"
 license-files = ["LICENSE", "NOTICE", "THIRD_PARTY.md"]
 requires-python = ">=3.11"
 dependencies = [
-  "lexbench-webfetch @ git+https://github.com/lexmount/lexbench-webfetch.git@56aca0cb9b9b9eef9995f0168e0504a72e309cd1",
+  "lexbench-webfetch @ git+https://github.com/lexmount/lexbench-webfetch.git@2ea8c52df67c39eee2ea5113a63f5da34f102160",
   "beautifulsoup4>=4.12,<5",
   "fastcdc==1.7.0",
 ]
diff --git a/tests/test_external_local_report.py b/tests/test_external_local_report.py
index 14904a5a2..b8cdd32f3 100644
--- a/tests/test_external_local_report.py
+++ b/tests/test_external_local_report.py
@@ -92,6 +92,16 @@ def versioned_fixture(self, case, timeout, **kwargs):
         "max_failure_retries": 0})
     dataset["comparison"]["local_run_contract_sha256"] = contract_digest
     write(evidence / "dataset.json", dataset)
+    if case_count == 5:
+        from lexbench_webfetch.versions import create_release
+        original = create_release(dataset, benchmark="synthetic", version="1.0.0",
+                                  scoring_source_sha256="f" * 64)
+        original_sha = write(evidence / "benchmark-input-release.json", original)
+        current = create_release(dataset, benchmark="synthetic", version="1.1.0",
+                                 previous=original, change_note="Correct scoring of saved outputs")
+        write(evidence / "benchmark-release.json", current)
+        write(evidence / "execution.json", {"batches": batches, "aggregate": {
+            "benchmark_release": {"path": "benchmark-input-release.json", "sha256": original_sha}}})
     output = tmp_path / "report"
     for script, args in [
         ("build_report.py", ["--local-audit", "--evidence", str(evidence),
@@ -154,6 +164,28 @@ def versioned_fixture(self, case, timeout, **kwargs):
         from harness.web_fetch.report_build import build_report
         from harness.web_fetch.report_package import build as package_report
         from harness.web_fetch.publication_check import check as check_package
+        from harness.web_fetch.report_version import check_release
+        from harness.web_fetch.report_build import validate_local_presentation
+        original_binding = json.loads((output / "execution.json").read_text())["aggregate"]["benchmark_release"]
+        assert original_binding["sha256"] == original_sha
+        ordinary_rebuild = tmp_path / "ordinary-rebuild"
+        build_report(output, ordinary_rebuild, local_audit=True, language="en", include_english=True)
+        assert json.loads((ordinary_rebuild / "execution.json").read_text())["aggregate"]["benchmark_release"] == original_binding
+        assert json.loads((ordinary_rebuild / "benchmark-release.json").read_text())["version"] == "1.1.0"
+        original_path = ordinary_rebuild / original_binding["path"]
+        original_bytes = original_path.read_bytes()
+        original_path.write_text('{}')
+        with pytest.raises(ValueError, match="input release changed"):
+            check_release(ordinary_rebuild, dataset, saved)
+        original_path.write_bytes(original_bytes)
+        execution_path = ordinary_rebuild / "execution.json"
+        execution_bytes = execution_path.read_bytes()
+        changed_execution = json.loads(execution_bytes)
+        del changed_execution["aggregate"]["benchmark_release"]
+        write(execution_path, changed_execution)
+        with pytest.raises(ValueError, match="input release binding"):
+            check_release(ordinary_rebuild, dataset, saved)
+        execution_path.write_bytes(execution_bytes)
         binding_path = tmp_path / "presentation-binding.json"
         write(binding_path, {
             "comparison_kind": "uniform_native_local_measurement",
@@ -174,6 +206,31 @@ def versioned_fixture(self, case, timeout, **kwargs):
         archive = tmp_path / "frozen-report.zip"
         package_report(rebuilt, archive, presentation_binding=binding_path)
         check_package(archive, rebuilt, presentation_binding=binding_path)
+        damaged_paths = [rebuilt / "benchmark-release.json", rebuilt / "report-manifest.json",
+                         rebuilt / "index.html"]
+        pristine = {path: path.read_bytes() for path in damaged_paths}
+        for damage in ("missing", "changed", "dropped-binding", "wrong-label"):
+            if damage == "missing":
+                damaged_paths[0].unlink()
+            elif damage == "changed":
+                changed = json.loads(pristine[damaged_paths[0]])
+                changed["version"] = "99.0.0"
+                write(damaged_paths[0], changed)
+            elif damage == "dropped-binding":
+                changed = json.loads(pristine[damaged_paths[1]])
+                del changed["benchmark_release"]
+                write(damaged_paths[1], changed)
+            else:
+                damaged_paths[2].write_text(pristine[damaged_paths[2]].decode().replace(
+                    'data-benchmark-release="1.1.0"', 'data-benchmark-release="99.0.0"'))
+            with pytest.raises(ValueError, match="[Bb]enchmark"):
+                validate_local_presentation(rebuilt, binding_path)
+            with pytest.raises(ValueError, match="[Bb]enchmark"):
+                package_report(rebuilt, archive, presentation_binding=binding_path)
+            with pytest.raises(ValueError, match="[Bb]enchmark"):
+                check_package(archive, rebuilt, presentation_binding=binding_path)
+            for path, raw in pristine.items():
+                path.write_bytes(raw)
     # Language entry normalization must not hide changes to ordinary links.
     from bs4 import BeautifulSoup
     from harness.web_fetch.report_check import check
diff --git a/tests/test_webfetch_report_package.py b/tests/test_webfetch_report_package.py
index 03b3cab8e..8fd8ddfff 100644
--- a/tests/test_webfetch_report_package.py
+++ b/tests/test_webfetch_report_package.py
@@ -131,6 +131,57 @@ def test_retained_success_and_failure_rerun_can_be_packaged_together(tmp_path):
     assert archive.is_file()
 
 
+@pytest.mark.parametrize("damage", [None, "missing", "changed", "dropped-binding", "wrong-label"])
+def test_ordinary_package_and_publication_require_the_bound_benchmark_edition(tmp_path, damage):
+    from lexbench_webfetch.scoring import score_attempt
+    from lexbench_webfetch.versions import create_release
+    from harness.web_fetch.publication_check import check
+
+    source, archive = report_tree(tmp_path, sampled=True)
+    dataset_path = source / "dataset.json"
+    dataset = json.loads(dataset_path.read_text())
+    for case in dataset["cases"]:
+        case["url"] = "https://example.test/" + case["id"]
+    dataset_path.write_text(json.dumps(dataset))
+    cases = {case["id"]: case for case in dataset["cases"]}
+    rows = [json.loads(line) for line in (source / "results.jsonl").read_text().splitlines()]
+    for row in rows:
+        row["ok"] = True
+        row["metrics"] = score_attempt(cases[row["case_id"]],
+                                       (source / row["artifact_path"]).read_bytes(), ok=True)
+    (source / "results.jsonl").write_text("".join(json.dumps(row) + "\n" for row in rows))
+    release = create_release(dataset, benchmark="sample", version="1.0.0")
+    sidecar = source / "benchmark-release.json"
+    sidecar.write_text(json.dumps(release))
+    manifest = {"benchmark_release": {"path": sidecar.name,
+        "sha256": hashlib.sha256(sidecar.read_bytes()).hexdigest(),
+        "benchmark": "sample", "version": "1.0.0"}}
+    (source / "report-manifest.json").write_text(json.dumps(manifest))
+    home = source / "index.html"
+    home.write_text(home.read_text() + '<a data-benchmark-release="1.0.0" '
+                    'href="benchmark-release.json">sample v1.0.0</a>')
+    report_package.build(source, archive)
+    check(archive, source)
+    before = archive.read_bytes()
+    if damage is None:
+        return
+    if damage == "missing":
+        sidecar.unlink()
+    elif damage == "changed":
+        release["version"] = "99.0.0"
+        sidecar.write_text(json.dumps(release))
+    elif damage == "dropped-binding":
+        (source / "report-manifest.json").write_text('{}')
+    else:
+        home.write_text(home.read_text().replace('data-benchmark-release="1.0.0"',
+                                                 'data-benchmark-release="99.0.0"'))
+    with pytest.raises(ValueError, match="[Bb]enchmark"):
+        report_package.build(source, archive)
+    assert archive.read_bytes() == before
+    with pytest.raises(ValueError, match="[Bb]enchmark"):
+        check(archive, source)
+
+
 @pytest.mark.parametrize(
     "damage",
     [
diff --git a/uv.lock b/uv.lock
index 30b9e9a1b..3c0aac23d 100644
--- a/uv.lock
+++ b/uv.lock
@@ -354,7 +354,7 @@ requires-dist = [
     { name = "jieba", marker = "extra == 'webmainbench-eval'", specifier = "==0.42.1" },
     { name = "json-repair", marker = "extra == 'browser-use-eval'", specifier = ">=0.54,<1" },
     { name = "langchain-community", marker = "extra == 'oss-eval'", specifier = "==0.3.30" },
-    { name = "lexbench-webfetch", git = "https://github.com/lexmount/lexbench-webfetch.git?rev=56aca0cb9b9b9eef9995f0168e0504a72e309cd1" },
+    { name = "lexbench-webfetch", git = "https://github.com/lexmount/lexbench-webfetch.git?rev=2ea8c52df67c39eee2ea5113a63f5da34f102160" },
     { name = "markdown", marker = "extra == 'webmainbench-eval'", specifier = ">=3.7,<4" },
     { name = "openai", marker = "extra == 'webmainbench-eval'", specifier = ">=2.8,<4" },
     { name = "openpyxl", marker = "extra == 'oss-eval'", specifier = "==3.1.5" },
@@ -1933,7 +1933,7 @@ wheels = [
 [[package]]
 name = "lexbench-webfetch"
 version = "0.3.0"
-source = { git = "https://github.com/lexmount/lexbench-webfetch.git?rev=56aca0cb9b9b9eef9995f0168e0504a72e309cd1#56aca0cb9b9b9eef9995f0168e0504a72e309cd1" }
+source = { git = "https://github.com/lexmount/lexbench-webfetch.git?rev=2ea8c52df67c39eee2ea5113a63f5da34f102160#2ea8c52df67c39eee2ea5113a63f5da34f102160" }
 dependencies = [
     { name = "beautifulsoup4" },
     { name = "soupsieve" },