Audit reporting-family target attribution (sc-2831) - #600
Merged
Conversation
The PR598 reporting-family measurement counted a FAIL from the expected lens as a hit, even when the finding described another behavior. This audit separates blocking a buggy change from identifying its intended defect, using the frozen 12-occurrence ledger and executable fixture controls. The new regression runs the original, buggy, repaired and suggested-change variants: the classifier's reported behavior is unchanged across the first three, and the suggested change still leaves the intended defect. It also demonstrates the existing scorer's expected-lens proxy without changing scoring behavior. The audit accounts for all 12 occurrences: four supported selector-target findings and eight uncertain assessments. All four repair-side findings remain unresolved because their adapted fixtures lack load-bearing input or reporting contracts. This establishes no new false-positive rate or model-quality improvement. Corpus rows, reviewer prompts and historical checkpoints remain unchanged. Validation: focused repair-control suite (10 tests), corpus validation (243 rows, zero problems), benchmark tracking and benchmark TypeScript checks pass. Commit-guard found no actionable issue; semantic retrieval was unavailable because this checkout has no usable source index. Full-suite result will be recorded below once complete. Research methods and limitations are documented in docs/benchmarks/reporting-family-audit-2026-09-05.md, including links to MalPR-Bench, c-CRAB and SWE-Review. Full historical claims remain in private evidence artifacts; the report contains occurrence hashes and public-fixture behavior. Shortcut: https://app.shortcut.com/benordlabs/story/2831
Contributor
|
Warning Review limit reachedNext included review available in 23 minutes. View limit detailsLimit details: You’ve used the included review currently available. You've used all free OSS reviews for now. Wait for the free limit to reset to keep reviewing this public repository. Review configuration: ⚙️ Run configurationConfiguration used: defaults Review profile: CHILL Plan: Team Run ID: 📒 Files selected for processing (3)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The PR598 reporting-family measurement counted a FAIL from the expected lens as a hit, even when the finding described another behavior. This audit separates blocking a buggy change from identifying its intended defect, using the frozen 12-occurrence ledger and executable fixture controls.
The new regression runs the original, buggy, repaired and suggested-change variants: the classifier's reported behavior is unchanged across the first three, and the suggested change still leaves the intended defect. It also demonstrates the existing scorer's expected-lens proxy without changing scoring behavior.
The audit accounts for all 12 occurrences: four supported selector-target findings and eight uncertain assessments. All four repair-side findings remain unresolved because their adapted fixtures lack load-bearing input or reporting contracts. This establishes no new false-positive rate or model-quality improvement. Corpus rows, reviewer prompts and historical checkpoints remain unchanged.
Validation: focused repair-control suite (10 tests), corpus validation (243 rows, zero problems), benchmark tracking and benchmark TypeScript checks pass. Commit-guard found no actionable issue; semantic retrieval was unavailable because this checkout has no usable source index. The local full suite was interrupted with SIGINT (exit 130) after about 15 minutes under heavy machine load; it is incomplete, not a pass. Before interruption it reported failures in three unchanged ship integration tests (receipt resume, absent local base branch, identical untracked symlink). The final assertion details were not emitted, so their causes are unconfirmed. CI runs the full suite on this PR commit.
Research methods and limitations are documented in docs/benchmarks/reporting-family-audit-2026-09-05.md, including links to MalPR-Bench, c-CRAB and SWE-Review. Full historical claims remain in private evidence artifacts; the report contains occurrence hashes and public-fixture behavior.
Shortcut: https://app.shortcut.com/benordlabs/story/2831
Shipping: all deterministic gates, all four correctness lenses, API security, backend performance, conventions, commit-guard and completeness passed. No gate bypass was used. Advisory limits: semantic source index unavailable, pre-existing corpus holdout-floor warnings and two orphan skill-projection files.