Audit October feature coverage and surface recorded evidence in Lab

[Copilot]

Audit the October 5–19 changes against repository docs, the published site, and the opt-in Lab dashboard. The audit reconciles 50 merged PRs with default-branch commits and records each feature’s documentation, UI/API, and regression coverage.

• Lab evidence: Add a Recorded evidence tab for saved reliability estimates, portable evidence, scoped requirement observations, and independent prompt-judging results. Distinguish missing or invalid metadata, explicit zero credits, cached provenance, and unverified evidence; do not imply evidence contents were independently verified.
• Audit and docs: Add a linked coverage inventory and update Lab guidance, CLI/runtime wording, and October version boundaries. Add a screenshot generated from a mock-provenance fixture with a matching site mirror.
• Known gap: Malformed runtime reliability interval and pass^k values still need validation in the Lab inspector.

• Fixes Audit October 5–19 features and fixes for documentation, site, and new dashboard coverage #722

[Copilot]

Copilot wasn't able to review any files in this pull request. Check if the Files changed in this pull request are included in default exclusions.

[Copilot]

🟡 Changes recommended

Malformed judging metadata can crash the inspector, and regression fixtures contain provenance that production cannot emit.

3 open issues

• Null judging rows crash rendering instead of being withheld · New
• Test fixture combines incompatible cached and judging provenance · New
• Malformed judging metadata is misreported as absent · New

_{Review effort: Balanced}

[Copilot]

🔵 Needs a closer look

Malformed object-shaped evidence records can still crash the inspector instead of displaying a withheld state.

1 open issue

• Test fixture combines incompatible cached and judging provenance

1 previously missed issue

In code that hasn't changed since last review

Validate malformed rows before rendering header fields

web/​src/​lab/​RunInspector.tsx:74

Object-shaped malformed rows still reach the header. For example, { runNumber: {}, attempts: 1, cached: false, assessment: "invalid" } passes this check; the manifest guard returns false, but React then tries to render runNumber as a child and crashes the inspector. Validate the envelope fields before rendering them and return the existing withheld placeholder on failure. Add a regression with a malformed object beside a valid record.

2 resolved since last review

• Null judging rows crash rendering instead of being withheld
• Malformed judging metadata is misreported as absent

_{Review effort: Balanced}

[Copilot]

🔵 Needs a closer look

Malformed object-shaped evidence rows can still crash the inspector instead of displaying a withheld state.

1 open issue

• Test fixture combines incompatible cached and judging provenance

1 previously missed issue

In code that hasn't changed since last review

Validate malformed evidence envelopes before rendering

web/​src/​lab/​RunInspector.tsx:114

Object-shaped malformed rows still pass this guard. For example, { runNumber: {}, attempts: 1, cached: false, assessment: "invalid" } reaches Run {evidence.runNumber}, where React throws instead of showing the withheld placeholder. Validate the envelope fields before rendering them, and add a regression with a malformed object alongside a valid sibling so one bad row cannot hide the remaining evidence.

_{Review effort: Balanced}