Skip to content

Adjudicate zero-recall gold labels against pinned source - #47

Merged
MHohlios merged 2 commits into
feature/matcher-calibrationfrom
feature/gold-adjudication
Oct 2, 2026
Merged

MHohlios merged 2 commits into
feature/matcher-calibrationfrom
feature/gold-adjudication

Conversation

@MHohlios

@MHohlios MHohlios commented Oct 2, 2026 •

Copy link
Copy Markdown
Contributor

The 15 zero-correctness-recall CRB cases contain pre-existing defects, deliberate behavior, incorrect API assumptions, and mislabeled cleanup. Add pinned-source rulings to all 35 labels, preserve imported text/categories, and exclude ruled-out labels from matching and the correctness denominator without renumbering recorded matches.

The historical re-judge preserves the original outputs: 9 eligible correctness defects, 17 policy exclusions, 9 wrong or underspecified labels. One eligible defect was already reported but tiered as efficiency. Entirely excluded cases stay unscored rather than becoming expected-clean. Pinned Keycloak callers and feature-version selection exclude two initially accepted labels; recovery-credential ID loss remains eligible through the reachable account deletion path.

Validation: four gold-judge tests, four matcher tests, markdown lint, git diff --check, and source-based reproductions of partial-cache event mispairing and integer-tuple hash stability pass. This PR is stacked on #46 and does not claim current-release recall.

Fixes #13.

Signed-off-by: Michael Hohlios <michael@vyttle.com>
Signed-off-by: Michael Hohlios <michael@vyttle.com>
@MHohlios
MHohlios force-pushed the feature/gold-adjudication branch from 2317ab5 to 64b8723 Compare October 2, 2026 19:24
@MHohlios
MHohlios merged commit 3c9d691 into main Oct 2, 2026
2 checks passed
@MHohlios
MHohlios deleted the feature/gold-adjudication branch October 2, 2026 19:29
MHohlios added a commit that referenced this pull request Oct 2, 2026
Gold evaluations previously only invoked Claude candidates, preventing
the requested comparison without a Claude account. Add explicit Codex
candidate selection, pinned model/effort metadata, audit agent
enablement, and preparation of the reviewed diff before the read-only
candidate runs. Refuse to judge incomplete coverage.

Preserve the ten-case subset and diagnostic runs for #18. The attempted
comparison remains blocked: partial-clone historical objects were
unavailable and the audit host could not launch every independent
finder/verifier within its agent-thread limit. Zero cases are scored; no
recall or audit-versus-review conclusion is claimed.

Validation: two offline Codex gold tests cover both tiers, read-only
repository preservation, and rejection of incomplete coverage. Five
portability, four matcher, and four gold-judge tests pass, as do
markdown lint and git diff --check.

Stacked on #47. Related to #18; does not close it.

Signed-off-by: Michael Hohlios <michael@vyttle.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Gold sweep v0: adjudicate the 15 zero-recall CRB cases (real miss vs policy-excluded vs label-wrong)

1 participant