review: Preserve valid false-positive adjudications

resources/review-validation-findings.md

@@ -26,9 +26,7 @@ The user message gives you a JSON object of this exact shape:
       "baseline_false_positive_challenges": [
         {
           "baseline_id": "fast-N",
-          "finding": { "problem": "<exact baseline finding>", "severity": "..." },
           "proof": {
-            "finding_claim": "<exact problem text>",
             "verified_facts": ["<specialist-verified fact>"],
             "contradiction": "<why the complete finding is impossible>",
             "conclusion": "false_positive"
@@ -93,9 +91,9 @@ For each finding, decide one of:
 adjudicate EVERY entry, even when no specialist challenged it. Assign each
 entry its host identity `fast-N`, where N is its zero-based array index. Return
 exactly one `baseline_adjudications` record per baseline entry, in the same
-order. Copy `baseline_id` and the complete finding object exactly. Every record
-must contain repository-tool-verified proof: make `proof.finding_claim` equal
-the finding's `problem` exactly, list concrete `verified_facts`, and explain in
+order. Use the exact `baseline_id` as the sole identity. The host owns the
+original finding object, so never echo or rewrite it. Every record must contain
+repository-tool-verified proof: list concrete `verified_facts`, and explain in
 `assessment` why those facts support or disprove the complete finding. Use
 `verdict: "KEEP"` with `proof.conclusion: "supported"` unless the checked-out
 tree conclusively proves the reported failure impossible. Only then use
@@ -106,6 +104,23 @@ uncertainty remains, KEEP. You MUST execute repository tools while adjudicating
 the baseline. If tools are unavailable, KEEP every entry and state the concrete
 facts available from the supplied commit material.
 
+For NULL-dereference, publication, initialization, teardown, and lifetime
+findings, validation is caller-first and lifecycle-complete. Before KEEP or DROP:
+
+1. Enumerate every actual entry path to the named reader, including inline
+   wrappers and static-branch gates in unchanged files.
+2. Locate when each gate becomes true and false relative to pointer/table
+   publication, enable failure, reader draining, and retirement.
+3. Check both forward enable ordering and reverse disable/error ordering.
+4. If the finding cites a similar caller with an explicit NULL check, compare
+   the two callers' execution phases and gates; do not assume the check proves
+   identical reachability.
+
+A local unguarded dereference is not sufficient proof of reachability. A KEEP
+assessment must name the concrete caller that crosses the unsafe lifecycle
+window. A DROP assessment must name the gate or ordering that makes every named
+path impossible.
+
 DROP a regular candidate when it reports the same underlying problem as a
 surviving baseline finding, even if the wording differs. Do not drop a candidate
 merely because it shares a location, function name, or terminology with the
@@ -151,7 +166,8 @@ Hard rules:
 - Adjudicate every `baseline_findings` entry, including entries absent from
   `baseline_false_positive_challenges`. Emit exactly one ordered record under
   `baseline_adjudications` for each entry. Omission, duplication, reordering,
-  or an inexact finding copy invalidates the complete response.
+  or an inexact baseline ID invalidates the complete response. Do not echo the
+  host-owned finding object.
 - A finding must describe a problem that remains in or is introduced by
   the reviewed commit. Do NOT keep a finding merely because the parent
   version was wrong; the final report is a review of the patch, not a
@@ -197,10 +213,8 @@ Output shape (strict):
       "baseline_adjudications": [
         {
           "baseline_id": "fast-N",
-          "finding": { "problem": "..." },
           "verdict": "KEEP|DROP",
           "proof": {
-            "finding_claim": "...",
             "verified_facts": ["..."],
             "assessment": "...",
             "conclusion": "supported|false_positive"