fix(gan): distinguish scores from verdict thresholds

scripts/gan-harness.sh

@@ -62,15 +62,30 @@ extract_score() {
   # Extract the TOTAL weighted score from a feedback file
   local file="$1"
   awk '
-    /\*\*TOTAL\*\*/ || /Verdict:/ {
-      if (match($0, /[0-9]+[.][0-9]+/)) {
-        print substr($0, RSTART, RLENGTH)
+    /\*\*TOTAL\*\*/ {
+      total_line = $0
+      total = ""
+      while (match(total_line, /[0-9]+[.][0-9]+/)) {
+        total = substr(total_line, RSTART, RLENGTH)
+        total_line = substr(total_line, RSTART + RLENGTH)
+      }
+      if (total != "") {
+        print total
         found = 1
         exit
       }
     }
+    /Verdict:/ && /[Ss]core[[:space:]]*[:=]?[[:space:]]*[0-9]+[.][0-9]+/ {
+      verdict = $0
+      sub(/^.*[Ss]core[[:space:]]*[:=]?[[:space:]]*/, "", verdict)
+      if (match(verdict, /^[0-9]+[.][0-9]+/)) {
+        verdict = substr(verdict, RSTART, RLENGTH)
+      } else {
+        verdict = ""
+      }
+    }
     END {
-      if (!found) print "0.0"
+      if (!found) print (verdict != "" ? verdict : "0.0")
     }
   ' "$file" 2>/dev/null
 }

tests/gan-harness.test.js

@@ -82,6 +82,23 @@ const results = Object.freeze([
     assert.strictEqual(result, '9.1');
   }),
 
+  test('extract_score does not treat a Verdict threshold as a score', () => {
+    const feedback = '## Verdict: PASS / FAIL (threshold: 7.0)\n';
+    const result = extractScore(feedback);
+
+    assert.strictEqual(result, '0.0');
+  }),
+
+  test('extract_score prefers a TOTAL score after a Verdict threshold', () => {
+    const feedback = [
+      '## Verdict: PASS / FAIL (threshold: 7.0)',
+      '| **TOTAL** | **1.0** | **9.0** |',
+    ].join('\n');
+    const result = extractScore(feedback);
+
+    assert.strictEqual(result, '9.0');
+  }),
+
   test('extract_score returns the fallback when no supported score exists', () => {
     const feedback = 'Other score: 9.9\n';
     const result = extractScore(feedback);

tests/scripts/codex-hooks.test.js

@@ -43,15 +43,20 @@ function cleanup(dirPath) {
   fs.rmSync(dirPath, { recursive: true, force: true });
 }
 
+function resolveBashExecutable(env = process.env) {
+  return env.BASH_PATH
+    || (process.platform === 'win32' && fs.existsSync('C:\\Program Files\\Git\\bin\\bash.exe')
+      ? 'C:\\Program Files\\Git\\bin\\bash.exe'
+      : fs.existsSync('/bin/bash')
+        ? '/bin/bash'
+        : 'bash');
+}
+
 function runBash(
   scriptPath,
   { args = [], env = {}, cwd = repoRoot, input = undefined, preservePath = true } = {},
 ) {
-  const bash = process.platform === 'win32' && fs.existsSync('C:\\Program Files\\Git\\bin\\bash.exe')
-    ? 'C:\\Program Files\\Git\\bin\\bash.exe'
-    : fs.existsSync('/bin/bash')
-      ? '/bin/bash'
-      : 'bash';
+  const bash = resolveBashExecutable();
   return spawnSync(bash, [scriptPath, ...args], {
     cwd,
     env: {
@@ -132,6 +137,17 @@ const cacheManifestWithLocalRefs = {
 let passed = 0;
 let failed = 0;
 
+if (
+  test('shell test runner honors an explicit BASH_PATH override', () => {
+    assert.strictEqual(
+      resolveBashExecutable({ BASH_PATH: '/custom/git/bin/bash' }),
+      '/custom/git/bin/bash',
+    );
+  })
+)
+  passed++;
+else failed++;
+
 function runHermeticPrePush({
   failScript = null,
   includeCorepack = true,