RF-07 — Warning extraction precedes the canonical outcome check

State: fixed 2026-08-27
Repair shape: small local bug fix
Severity: high for contract correctness, low implementation cost

Finding

extract_warns() extracts explicit - warn: items before it checks whether the canonical outcome is warn. PASS- or FAIL-shaped result prose can therefore produce actionable warning text even though the fix contract says the selector consumes WARN pairs.

Evidence

• extract_warns() returns actionable findings before its outcome != "warn" guard.
• The fix procedure specifies current baseline-backed WARN pairs.
• Direct execution observed PASS-shaped input returning an explicit warning; see the evidence boundary.

Why it matters

The canonical outcome is supposed to determine whether a verdict enters the fix queue. Letting prose override that field makes protocol state and operational behavior disagree and can trigger work the reviewer explicitly classified as a PASS or FAIL instead.

Provisional repair direction

Return immediately when outcome != "warn", before parsing findings. Keep any future cross-outcome informational finding feature separate from the fix queue.

Done when

• PASS and FAIL bodies containing - warn: produce no warn-selector entries.
• WARN retains explicit-finding, summary, and fallback behavior.
• A full stored-pair test exercises the scanner, not only the helper.

Resolution

extract_warns() now returns before parsing result prose unless the canonical outcome is warn. Parameterized regression tests cover PASS and FAIL at both the helper boundary and through stored-pair scanning. The complete test suite passes (591 passed).