diff --git a/changelog.d/fixes/release-green-complexity-outcomes.md b/changelog.d/fixes/release-green-complexity-outcomes.md
new file mode 100644
index 0000000000..d67c5c1b5e
--- /dev/null
+++ b/changelog.d/fixes/release-green-complexity-outcomes.md
@@ -0,0 +1 @@
+- **fix(quality):** prevent cancelled, failed, or incomplete complexity measurements from appearing as passing release-green metrics; retain advisory classification for complete, measured ratchet drift without changing baselines or CI workflows.
diff --git a/scripts/quality/validate-release-green.mjs b/scripts/quality/validate-release-green.mjs
index 2c5bdb0022..7dc1bc4def 100644
--- a/scripts/quality/validate-release-green.mjs
+++ b/scripts/quality/validate-release-green.mjs
@@ -243,6 +243,59 @@ export function parseCognitiveCount(out) {
   return m ? Number(m[1]) : null;
 }
 
+/** Translate the shared complexity walk into the two metric records. */
+export function evaluateComplexityRun({ code, out }, cogBase) {
+  // The shared walk emits both machine counts followed by both ratchet verdicts.
+  // Do not use the legacy generic "N violações" fallback here: a partial walk
+  // containing only the cyclomatic summary would masquerade as cognitive data.
+  const cogMatch = /^cognitiveComplexity=(\d+)\r?$/m.exec(out);
+  const cycMatch = /^complexity=(\d+)\r?$/m.exec(out);
+  const cogSummary = /^\[cognitive-complexity\] (OK|REGRESSÃO) — (\d+) violações /m.exec(out);
+  const cycSummary = /^\[complexity\] (OK|REGRESSÃO) — (\d+) violações /m.exec(out);
+  const cogCurrent = cogMatch ? Number(cogMatch[1]) : null;
+  const cycCurrent = cycMatch ? Number(cycMatch[1]) : null;
+  const cogOver = isDrift(cogCurrent, cogBase);
+  const cycRegressed = cycSummary?.[1] === "REGRESSÃO";
+  const measured =
+    [cogCurrent, cycCurrent, cogBase].every((count) => Number.isSafeInteger(count) && count >= 0) &&
+    Number(cogSummary?.[2]) === cogCurrent &&
+    Number(cycSummary?.[2]) === cycCurrent &&
+    (cogSummary?.[1] === "REGRESSÃO") === cogOver;
+  // Exit 1 is expected only for a complete, measured ratchet regression. A
+  // cancelled/crashed tool or missing metrics is execution failure, not drift.
+  const measuredDrift = code === 1 && measured && (cogOver || cycRegressed);
+  if (!measured || (code !== 0 && !measuredDrift)) {
+    return [
+      {
+        id: "complexity",
+        label: "Complexity measurement",
+        kind: "hard",
+        ok: false,
+        detail:
+          code !== 0
+            ? `Complexity process exited ${code}: ${firstFailureLine(out)}`
+            : "Complexity process did not report complete finite metrics and baseline",
+      },
+    ];
+  }
+  return [
+    {
+      id: "cognitive-complexity",
+      label: "Cognitive complexity (ratchet)",
+      kind: "drift",
+      ok: !cogOver,
+      detail: `${cogCurrent} vs baseline ${cogBase}${cogOver ? ` (+${cogCurrent - cogBase} drift → rebaseline at release)` : ""}`,
+    },
+    {
+      id: "complexity",
+      label: "Cyclomatic complexity (ratchet)",
+      kind: "drift",
+      ok: !cycRegressed,
+      detail: `complexity=${cycCurrent} (shared walk with cognitive)${cycRegressed ? " REGRESSED" : ""}`,
+    },
+  ];
+}
+
 /**
  * Drift verdict for a ratchet: a metric that grew past its committed baseline is
  * "drift" (reported, never blocking). `direction:"down"` metrics (warnings,
@@ -591,42 +644,11 @@ async function main() {
   // Complexity + cognitive (one ESLint walk; both still recorded as drift)
   {
     announce("Complexity + cognitive ratchets (shared ESLint walk)");
-    const { out } = run(npmCmd, ["run", "check:complexity-ratchets"]);
-    saveGateLog("complexity-ratchets", out);
-    const cogCurrent = parseCognitiveCount(out);
-    const cogBase = baselineValue("cognitiveComplexity");
-    const cogOver = isDrift(cogCurrent, cogBase);
-    const cycMatch = /(?:^|\n)complexity=(\d+)/.exec(out);
-    const cycOkMatch = /\[complexity\] OK — (\d+)/.exec(out);
-    const cycRegMatch = /\[complexity\] REGRESSÃO — (\d+)/.exec(out);
-    const cycCurrent = cycMatch
-      ? Number(cycMatch[1])
-      : cycOkMatch
-        ? Number(cycOkMatch[1])
-        : cycRegMatch
-          ? Number(cycRegMatch[1])
-          : null;
-    const cycRegressed = /\[complexity\] REGRESSÃO/.test(out);
-    record({
-      id: "cognitive-complexity",
-      label: "Cognitive complexity (ratchet)",
-      kind: "drift",
-      ok: !cogOver,
-      detail:
-        cogCurrent == null
-          ? "could not parse count"
-          : `${cogCurrent} vs baseline ${cogBase}${cogOver ? ` (+${cogCurrent - cogBase} drift → rebaseline at release)` : ""}`,
-    });
-    record({
-      id: "complexity",
-      label: "Cyclomatic complexity (ratchet)",
-      kind: "drift",
-      ok: !cycRegressed,
-      detail:
-        cycCurrent == null
-          ? firstFailureLine(out) || "measured via check:complexity-ratchets"
-          : `complexity=${cycCurrent} (shared walk with cognitive)${cycRegressed ? " REGRESSED" : ""}`,
-    });
+    const result = run(npmCmd, ["run", "check:complexity-ratchets"]);
+    saveGateLog("complexity-ratchets", result.out);
+    for (const metric of evaluateComplexityRun(result, baselineValue("cognitiveComplexity"))) {
+      record(metric);
+    }
   }
 
   // file-size (drift)
diff --git a/tests/unit/build/release-green-complexity-outcomes.test.ts b/tests/unit/build/release-green-complexity-outcomes.test.ts
new file mode 100644
index 0000000000..e8b0b31d86
--- /dev/null
+++ b/tests/unit/build/release-green-complexity-outcomes.test.ts
@@ -0,0 +1,105 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+import {
+  computeVerdict,
+  evaluateComplexityRun,
+} from "../../../scripts/quality/validate-release-green.mjs";
+
+test("cancelled complexity measurement cannot emit passing metrics", () => {
+  const results = evaluateComplexityRun(
+    { code: 130, out: "Gate execution CANCELLED; see command receipt" },
+    100
+  );
+  assert.equal(
+    results.some((result) => result.ok),
+    false
+  );
+  assert.equal(computeVerdict(results).releaseGreen, false);
+});
+
+function completeOutput(cognitive = 80, cyclomatic = 20, cyclomaticDrift = false) {
+  return [
+    `complexity=${cyclomatic}`,
+    `cognitiveComplexity=${cognitive}`,
+    `[complexity] ${cyclomaticDrift ? "REGRESSÃO" : "OK"} — ${cyclomatic} violações ${cyclomaticDrift ? "> baseline 10" : "(baseline 20)"}`,
+    `[cognitive-complexity] ${cognitive > 100 ? "REGRESSÃO" : "OK"} — ${cognitive} violações ${cognitive > 100 ? "> baseline 100" : "(baseline 100)"}`,
+  ].join("\n");
+}
+
+for (const code of [1, 2, 42, 124, 130, 143]) {
+  test(`complete metrics cannot hide an unexpected process exit ${code}`, () => {
+    const results = evaluateComplexityRun({ code, out: completeOutput() }, 100);
+    assert.equal(
+      results.some((result) => result.ok),
+      false
+    );
+    assert.equal(computeVerdict(results).releaseGreen, false);
+    assert.equal(results[0].kind, "hard");
+  });
+}
+
+for (const out of [
+  "",
+  "complexity=20\n[complexity] OK — 20 violações (baseline 20)",
+  "cognitiveComplexity=80\n[cognitive-complexity] OK — 80 violações (baseline 100)",
+  "complexity=20\ncognitiveComplexity=80",
+  completeOutput().replace("cognitiveComplexity=80", "cognitiveComplexity=NaN"),
+  completeOutput().replace("complexity=20", "complexity=20.5"),
+  completeOutput().replace("complexity=20", "complexity=20invalid"),
+  completeOutput().replace("complexity=20", "complexity=-20"),
+  completeOutput(Number.MAX_SAFE_INTEGER + 1),
+]) {
+  test(`incomplete or invalid output fails closed: ${JSON.stringify(out)}`, () => {
+    const results = evaluateComplexityRun({ code: 0, out }, 100);
+    assert.equal(
+      results.some((result) => result.ok),
+      false
+    );
+    assert.equal(computeVerdict(results).releaseGreen, false);
+  });
+}
+
+test("unknown baseline fails closed instead of assuming no drift", () => {
+  for (const baseline of [null, NaN, Infinity, -1]) {
+    const results = evaluateComplexityRun({ code: 0, out: completeOutput() }, baseline);
+    assert.equal(computeVerdict(results).releaseGreen, false);
+  }
+});
+
+test("completed metrics within baseline retain both passing records", () => {
+  const results = evaluateComplexityRun({ code: 0, out: completeOutput() }, 100);
+  assert.deepEqual(
+    results.map(({ id, kind, ok }) => ({ id, kind, ok })),
+    [
+      { id: "cognitive-complexity", kind: "drift", ok: true },
+      { id: "complexity", kind: "drift", ok: true },
+    ]
+  );
+  assert.equal(computeVerdict(results).releaseGreen, true);
+});
+
+for (const [cognitive, cyclomaticDrift] of [
+  [101, false],
+  [80, true],
+  [101, true],
+] as const) {
+  test(`measured drift remains advisory: cognitive=${cognitive}, cyclomatic=${cyclomaticDrift}`, () => {
+    const results = evaluateComplexityRun(
+      { code: 1, out: completeOutput(cognitive, 20, cyclomaticDrift) },
+      100
+    );
+    assert.equal(results.length, 2);
+    assert.ok(results.every((result) => result.kind === "drift"));
+    assert.equal(results[0].ok, cognitive <= 100);
+    assert.equal(results[1].ok, !cyclomaticDrift);
+    assert.equal(computeVerdict(results).releaseGreen, true);
+  });
+}
+
+test("partial regression output cannot disguise a crashed check as advisory drift", () => {
+  const results = evaluateComplexityRun(
+    { code: 1, out: "complexity=20\ncognitiveComplexity=101\nError: incomplete walk" },
+    100
+  );
+  assert.equal(computeVerdict(results).releaseGreen, false);
+});