feat(webfetch): publish reviewed six-method 430-case report

.agents/skills/benchmark-editorial/SKILL.md

@@ -17,7 +17,7 @@ Treat feedback as editorial input, not publishable copy. Identify the reader pro
 
 ## Evidence and argument
 
-Read the supplied reports and relevant reference articles, including their charts, cases and methodology. Track material that could not be inspected privately; do not claim full coverage from a landing page. Study how reference articles develop their reasoning as well as how they look. Their facts are not evidence for this benchmark.
+Read the supplied reports and relevant reference articles, including their charts, cases and methodology. Track material that could not be inspected privately; do not claim full coverage from a landing page. Study how reference articles develop their reasoning as well as how they look. Their facts are not evidence for this benchmark. When reproducing a supplied reference, compare its actual opening, result charts, specimen interactions and conclusion with the delivered page. Reuse the owning presentation components where available; matching headings and factual claims alone does not establish presentation parity.
 
 For this repository, use `evaluation/current-report.json`, the normalized persisted results and the current scoring implementation. Read [benchmark-report](../benchmark-report/SKILL.md) when building or auditing them. Keep a private ledger for consequential claims: evidence, calculation, interpretation and limits. Do not rerun providers, alter scores or select better attempts to improve a narrative.
 

.agents/skills/benchmark-report/SKILL.md

@@ -5,7 +5,7 @@ description: Design, run, audit, revise, and release evidence-backed benchmark r
 
 # Benchmark report
 
-Produce a report whose conclusions can be traced to frozen inputs, measured outputs, scoring rules, and an explicit result review. Optimize for a reader deciding what the benchmark means, not for preserving the author's workflow.
+Produce a report whose conclusions can be traced to frozen inputs, measured outputs, scoring rules, and the declared result-review coverage. Optimize for a reader deciding what the benchmark means, not for preserving the author's workflow.
 
 Read the repository's `AGENTS.md` first. Project rules override this general skill. Before changing a benchmark, read [evaluation design](references/evaluation-design.md). Before reviewing scores or deciding whether to rerun, read [audit and revision](references/audit-and-revision.md). Before building or publishing a report, read [report and release](references/report-and-release.md).
 
@@ -18,13 +18,15 @@ An explicitly requested diagnostic report may be delivered separately from a cer
 Public report CLIs default to English where supported and expose `--language en|zh`. Preserve the selected language on rebuild. English builds retain the matching Chinese edition; keep measured material in its source language. Historical Chinese-only renderers must disclose that limit and reject unsupported requests. Bind generated languages and the selected entry in report metadata. English navigation opens a not-generated notice when absent; never publish stale English pages. Validate each generated edition and retain route, value and evidence-link parity checks when both exist.
 
 1. Establish the decision being supported, the measured unit, cohort, versions, execution location, freshness policy, and intended deliverable. Inspect the actual dataset, runner, outputs, scorer, prior review evidence, and current report. Do not infer them from screenshots or prose.
-2. Freeze a machine-readable evaluation contract before expensive calls. It must identify the exact cases and methods, run count, timeout, concurrency policy, input and output contract, scoring provenance, reference state, and publication target when one exists. For source-dependent tasks, fail closed: verify and freeze references immediately before calls, verify the same declared scope afterward, and certify the cohort only when both checks pass. Partial or uncertified cohorts remain diagnostic evidence and cannot enter a release.
+2. Freeze a machine-readable evaluation contract before expensive calls. It must identify the exact cases and methods, run count, timeout, concurrency policy, input and output contract, scoring provenance, reference state, and publication target when one exists. Existing compiled reference answers are frozen by default. Source metadata, capture dates, and absent newer acceptance-contract fields do not imply a live-reference task. Only tasks explicitly declaring `source_verification.reference_mode: live_exact` require live checks. For those tasks, fail closed: verify and freeze references immediately before calls, verify the same declared scope afterward, and certify the cohort only when both checks pass. An unchanged compiled acceptance-contract dataset is already the benchmark input: freeze its task and rule bytes for execution, without requiring historical source-qualification raw files or re-auditing cohort admission. Qualify sources when adding tasks or changing references/acceptance rules; do not turn an ordinary version rerun into dataset reconstruction. Partial or uncertified cohorts remain diagnostic evidence and cannot enter a release.
 3. Let each method use its strongest documented route to produce the required output. Equalize the task, inputs, output contract, call budget, and scoring. Do not require identical internals. Record material asymmetries such as local versus remote execution, cold versus warm start, bundled conversion, network transit, account tier, and billable time.
 4. Preserve evidence as an append-only chain: task definition, reference, provider call, raw artifact, normalized artifact, score, review, revision, and release. Keep provider output and call identity immutable. A presentation or scoring change must not masquerade as a new provider result.
 5. Separate retrieval success, task acceptance, content quality, latency, cost, and freshness. A missing or partial reference makes the affected metric unavailable or scoped; it does not prove success or failure. Do not derive binary acceptance from an unrelated aggregate similarity score.
 6. For heterogeneous same-window latency evidence, preserve the full predeclared task-by-method matrix. Show all-call latency on that common denominator separately from successful-call latency on each method's conditional subset. Treat the all-method shared-pass intersection as a diagnostic, never as a post-outcome selector or ranking sample.
 7. At aggregate and publication boundaries, copy only explicit hash-bound evidence and public audit files. Reject symlinks and exclude worker trees, host paths, credentials, and internal run logs. Verify primary, auxiliary, delivered-file, and retained retry artifacts before release.
 8. Independent result review is opt-in: do not start it unless the user explicitly requests review. When requested without a broader scope, review machine failures only; review machine passes or every attempt only when the user explicitly asks for full review. Keep the initial verdict blind, but do not equate agreement with correctness: run the anomaly investigation in [audit and revision](references/audit-and-revision.md) for triggered results before calling the review complete. For a large independent review that is safely batchable, prefer resumable Codex CLI jobs with an exact model ID, read-only sandbox, strict output schema, non-overlapping packets, and bounded concurrency; persist each completed packet so a retry never repeats substantive review. Record the exact reviewed/total coverage, accepted unreviewed passes, reviewer, disagreements, adjudications, anomalies, and unresolved items. A review summary never replaces per-attempt evidence.
+For an explicitly requested full review, set `independent_result_review_scope: all` in the report profile and carry it into the assembled dataset comparison. The release boundary must reject failure-only review, missing attempts and accepted unreviewed passes. Reuse an existing content judgment only when the raw answer and acceptance contract are unchanged; verify each new call identity separately. For failed calls, inspect the current runtime evidence even when the answer is unchanged or empty: the same empty output can hide different failures. Changed evidence needs supplemental review, not another review of the unchanged population.
+
 9. Rerun only calls whose inputs, provider route/version, runtime contract, or source validity changed. Recompute downstream scores and reports without provider calls when only scoring, annotations, or presentation changed. Never retry merely to obtain a better result. A reference or verdict correction must identify the shared root cause and add executable positive/negative controls through the production scorer; bind those regression IDs in the revision receipt instead of patching stored verdicts.
 10. Build the public report from the frozen evidence. Every page should answer a reader question. Show task and run identifiers, the exact expected-versus-actual evidence behind a verdict, denominators and unavailable metrics, method differences, limits, concise reproduction instructions, and a compact review history.
 11. Release only after code tests, evidence integrity, report links, responsive/visual checks, secret scans, target-site checks, and deployed-output verification pass. Preserve earlier sites unless replacement was explicitly requested.

.agents/skills/benchmark-report/references/audit-and-revision.md

@@ -18,7 +18,7 @@ release manifest + deployed bytes
 
 Hash and bind transitions where accidental mixing is plausible. Validate exact cardinality and uniqueness, not only totals. Treat raw outputs, timestamps, request status, method version, and call IDs as immutable measured evidence.
 
-Source-dependent calls require a fail-closed boundary: capture and freeze the declared reference scope immediately before execution, repeat the independent observation after execution, and bind both checks to the complete cohort. A process exit, partial attempt count, changed reference, or missing post-check makes the cohort diagnostic rather than publishable; never salvage its successful rows into the release matrix.
+Live exact-reference calls require a fail-closed boundary: capture and freeze the declared reference scope immediately before execution, repeat the independent observation after execution, and bind both checks to the complete cohort. A process exit, partial attempt count, changed reference, or missing post-check makes the cohort diagnostic rather than publishable; never salvage its successful rows into the release matrix.
 
 ## Independent result review
 
@@ -49,6 +49,10 @@ Use dependency impact, not convenience:
 
 Never overwrite old evidence or select a later response because it scores better. Assemble a new complete evidence view atomically, record inherited versus rerun attempts, and verify all downstream bindings.
 
+Before a cohort-wide offline recomputation, finish the paired controls and compare the known affected output family with its independent judgments. Cover equivalent serializers and adjacent invalid forms together, including visible text outside a link, image alternatives, and record ownership. Batch the shared corrections before the expensive sweep; do not rerun unchanged browser calls. Bound both worker count and queued payloads when saved outputs or logs are large.
+
+A subresource network error does not establish a main-navigation failure. Failure categories must agree with the reviewed cause and its actual stage; keep missing embedded content, unsupported direct-file conversion, empty output, and process crashes distinct. Interpretation pages show complete distributions and representative evidence. Full per-result diagnoses belong in the existing result inventory and details.
+
 ## Revision receipts
 
 A score-only revision must record old and new dataset/result hashes, changed fields, decision changes, provider call count of zero, reviewer binding, and predecessor history hash. A new provider call must receive a new call identity and must not be described as a scoring correction.
@@ -78,3 +82,4 @@ private evidence. Never rescue an empty declared body from a nonempty display
 message. Reproject retained full responses without new product calls when only
 this selection changed; missing historical envelopes require fresh calls across
 the full declared correction scope before claiming revised benchmark scores.
+Reruns of unchanged compiled acceptance-contract tasks freeze the existing executable dataset. They do not require recovering historical qualification raw files or repeating admission review. Source qualification is part of task authoring and reference/rule correction, not a prerequisite for every provider-version comparison. Preserve historical receipts without applying their evidence-file requirements to a new execution protocol.

.agents/skills/benchmark-report/references/evaluation-design.md

@@ -43,3 +43,7 @@ Ground truth must be independent of the provider under test. Retain source locat
 ## Machine-readable current profile
 
 Keep one current profile that a fresh session can validate before running, scoring, rebuilding, or publishing. Include exact case IDs or their canonical digest, the ordered method/version set, run count, timeout, concurrency policy, output contract, reference state, review requirement, and release target. Fail closed on drift instead of reconstructing intent from old reports.
+
+For native-content browser comparisons, use task-appropriate official interfaces and score the returned deliverable. A final URL or HTML language tag is diagnostic unless the benchmark explicitly evaluates that metadata. Do not repair missing text using JSON DOM conversion or external PDF/DOCX parsers. The declared Chromium HTML-to-Markdown exception does not authorize document parsers. Bind native script versions, generate the interface matrix from the runner route selector, and require route validation at aggregation and publication. Keep negative controls for wrong content alongside positive controls for correct content with missing metadata.
+
+Keep browser engines distinct from execution configurations. Released and source-build variants use separate method identities in the execution registry and share the engine's native route constraints. Derive interface columns and participant counts from that registry. A strict native-mode marker selects default completion; it supplies no per-case selector or wait and must survive both dataset conversion and runner validation. Resource projection must preserve single URL values and tokenize only srcset candidates, including literal commas inside URLs. Protect these boundaries with runner, report and actual-browser controls.