FastTool

Reproducible lab · Synthetic metadata

Test the release boundary, not a fabricated Google result

This checked-in benchmark exercises the real metadata comparison engine against safe, blocking, uncertain, malformed and adversarial inputs. It reports engine behavior and export integrity only.

Worked synthetic benchmarkResults below come from reserved .test fixtures and the checked-in test command. They are not live Search results, Google measurements, production traffic or independent human certification.
39 / 39engine assertions passed
3explicit verdict classes
0false PASS observations

Method

  1. Read exact UTF-8 fixture bytes and tokenize supported head metadata without executing HTML.
  2. Run complete, blocking and uncertain comparisons through fasttool-metadata-release-gate/1.2.0.
  3. Preserve browser-first duplicate attributes, use HTML ASCII whitespace, and fail closed on malformed comments, raw-text/head-exit wrappers or parser coverage gaps.
  4. Reject malformed, NUL-containing and greater-than-64-KiB head input before artifact generation.
  5. Export active metadata only for PASS; other verdicts receive a non-executable review plan.
  6. Generate all text artifacts twice and compare exact bytes and SHA-256 hashes.
node scripts/snippet_release_gate_engine_test.mjs
node scripts/build_snippet_benchmark_receipt_20260804.mjs

Expected versus observed

ScenarioExpectedObservedEvidence
Complete supported changePASSPASSSupported metadata and parser coverage complete
Noindex + duplicate canonicalBLOCKBLOCKrobots_noindex and duplicate_canonical retained
Approximate title boundaryINDETERMINATEINDETERMINATEFixed local mobile-width threshold crossed
Duplicate content attribute decoyBLOCKBLOCKFirst HTML attribute remains authoritative; duplicate coverage is disclosed
Unclosed comment decoyRejectRejectedNo artifact is minted from an unfinished comment scope
RCDATA/RAWTEXT and head-exit wrappersNot PASSRejected or BLOCKMetadata-looking text outside the real head scope is never trusted
NBSP parser divergenceNot PASSBLOCK or INDETERMINATEOnly HTML ASCII whitespace separates tag, attribute and rel tokens
Malformed or oversized headRejectRejectedNo artifacts produced outside the input boundary
BLOCK implementation outputNon-executableNon-executableReview comments only; dangerous metadata is not emitted

What the boundary means

Measured

Parser coverage, explicit metadata policy, verdict precedence, deterministic glyph-class estimates, export escaping and exact artifact bytes.

Not measured

Google title-link or snippet selection, device-specific Search UI, rewrite frequency, index status, rank, CTR, traffic or revenue.

The width model exists to make a release review repeatable. It deliberately produces INDETERMINATE when the local estimate crosses a stated threshold; it never claims that Google will truncate the same text.

Fixture provenance and SHA-256

Downloadable benchmark evidence

Machine results

Download JSON6416d644d335bf8dfa6a64e016b1bef84533a395d66c3c63b6432b975500949e

Scenario ledger

Download CSV4e0c7f48d8b5de0b601b037e55a264f31766c52b8e28764c349f825e210b06cb

Methodology

Download Markdownb1d93ccdae6ed4ff350a5dc468c845d268ee3d09566bc5ebb0bb78cbbea363ff

Receipt v1

Download receipt4cd6d5e612be6d4d4e21c59dd2dd2c5ed61f97842b26d2719285079eb610d833

Receipt-core SHA-256

904bb052475b354cb868567e71c1d41ba35f785689b7fa8cd5e9b3f709bc5c62

Sources, limits and correction

FastTool is independent from Google. Report a fixture, method, hash or result defect to [email protected]. Include the artifact hash, runtime and observed assertion.

Read the paired workflow · Verify receipt and artifacts