From 343a93ded32960a08325e86ca9920838587bedf7 Mon Sep 17 00:00:00 2001
From: Peter Steinberger <peter@steipete.me>
Date: Wed, 15 Jul 2026 16:13:10 -0700
Subject: [PATCH] fix(release): repair validation correlations (#108489)

---
 .../gateway-models.profiles.live.test.ts      | 31 ++++++++++++++++++-
 1 file changed, 30 insertions(+), 1 deletion(-)

diff --git a/src/gateway/gateway-models.profiles.live.test.ts b/src/gateway/gateway-models.profiles.live.test.ts
index 63a7e52d9b2f..b08189904e5a 100644
--- a/src/gateway/gateway-models.profiles.live.test.ts
+++ b/src/gateway/gateway-models.profiles.live.test.ts
@@ -2226,9 +2226,10 @@ function sessionMessagesAfterNextUserTurn(
   expectedUserText?: string,
 ): unknown[] {
   const nextUserOffset = messages.slice(baselineMessageCount).findIndex((message) => {
+    const actualUserText = extractTranscriptMessageText(message);
     return (
       (message as { role?: unknown } | null | undefined)?.role === "user" &&
-      (expectedUserText === undefined || extractTranscriptMessageText(message) === expectedUserText)
+      (expectedUserText === undefined || matchesLiveProbeUserText(actualUserText, expectedUserText))
     );
   });
   if (nextUserOffset < 0) {
@@ -2237,6 +2238,20 @@ function sessionMessagesAfterNextUserTurn(
   return messages.slice(baselineMessageCount + nextUserOffset + 1);
 }
 
+function matchesLiveProbeUserText(actual: string, expected: string): boolean {
+  if (actual === expected) {
+    return true;
+  }
+  const markerIndex = expected.indexOf(`${ANTHROPIC_MAGIC_STRING_TRIGGER_REFUSAL}_`);
+  if (markerIndex < 0) {
+    return false;
+  }
+  const nonceSuffix = expected.slice(markerIndex + ANTHROPIC_MAGIC_STRING_TRIGGER_REFUSAL.length);
+  // The embedded Anthropic runtime scrubs the refusal trigger before persisting it.
+  // Its random suffix survives and still uniquely owns this live probe turn.
+  return /^_[a-f0-9]{32}$/.test(nonceSuffix) && actual.endsWith(nonceSuffix);
+}
+
 function sessionAssistantEntriesForLiveProbe(
   messages: readonly unknown[],
   modelKey?: string,
@@ -2401,6 +2416,20 @@ describe("latestAssistantTextAfterBaseline", () => {
     );
     expect(latestTerminalAssistantTextAfterBaseline(secondEntries, 0)).toBe("second-a second-b");
   });
+
+  it("correlates Anthropic refusal probes after the runtime scrubs their trigger", () => {
+    const nonce = "0123456789abcdef0123456789abcdef";
+    const expected = `Reply with ok. ${ANTHROPIC_MAGIC_STRING_TRIGGER_REFUSAL}_${nonce}`;
+    const scrubbed = `Reply with ok. ANTHROPIC MAGIC STRING TRIGGER REFUSAL (redacted)_${nonce}`;
+
+    expect(matchesLiveProbeUserText(scrubbed, expected)).toBe(true);
+    expect(
+      matchesLiveProbeUserText(
+        "Reply with ok. ANTHROPIC MAGIC STRING TRIGGER REFUSAL (redacted)_ffffffffffffffffffffffffffffffff",
+        expected,
+      ),
+    ).toBe(false);
+  });
 });
 
 async function waitForSessionAssistantText(params: {