Meta-review can attach assessments to the wrong finding and misreport review status #667

Problem

Meta-review can attach a response to the wrong finding when two findings share a rule and source line. Reports also group different unconfirmed outcomes together. A coverage finding added after meta-review can make a review phase appear to have failed even when no review was required.

Evidence

• A test with controlled model responses and two different findings on the same rule/line attached an assessment to the wrong finding.
• Per-finding review results do not consistently distinguish model disagreement, an omitted decision and a failed call.
• Findings added during finalization can be counted as work that the earlier meta-review phase should have reviewed.

The controlled-response test checks the matching logic; it does not measure model accuracy.

Expected behavior

Each assessment should stay attached to the finding it reviewed. Reports should preserve the original evidence and distinguish a completed assessment from missing or failed review. Findings added after meta-review should not create a false model-call failure. Actual gaps in review should still be reported.

Keep the existing scoring and completeness behavior. LLM disagreement or a missing review must not automatically clear a deterministic finding.

Related work

PR #616 preserves LLM finding fields; PR #663 names analyzers in status rows. These changes do not fix finding-to-assessment matching or distinguish the per-finding review outcomes described above.

Relevant code

meta_analyzer.py:76, meta_analyzer.py:391, semantic_runtime.py:191, meta_analyzer.py:617.