test(release): align five suites with the contracts #11933, #11919 and #11876 shipped on release/v3.8.51 (#11944)

Eleven PRs landed on release/v3.8.51 while the branch carried fifteen base reds, and
nine more red tests hid among them. None is a defect in the shipped code; each test
still encoded the contract that the merged PR deliberately replaced:

- openai-to-claude finish deferral (dd35750e5f, #11933): a finish chunk that carries no
  usage is now held until the end-of-stream flush that production performs
  (open-sse/utils/stream.ts flush -> translateResponse(..., null, state)). The drivers in
  stream-markdown-token-boundary, translator-tool-call-shim and
  gemini-malformed-function-call-finish-reason-2462 fed the finish chunk and asserted
  the terminal events immediately; they now mirror the flush. Assertions unchanged.
- authoritative live catalog (3d2832b836, #11919 fixes #11829): a synced catalog replaces
  the static registry, so model-lifecycle-integration no longer expects the static-only
  gpt-5.6-sol row to survive a sync. The #8627 contract the file guards (stale chat rows
  suppressed, typed media retained) is untouched.
- provider asset provenance (#11876): the unit shards check out with depth 1. The fixture
  pinned a historical commit as auditedCommit (absent on a shallow clone), the
  "binds auditedCommit" case relied on the repository root commit (the grafted HEAD on
  a shallow clone, which matches the physical snapshot), and the real-manifest case
  needs the audited commit fetched. The fixture now audits HEAD, the mismatch case
  builds a dangling empty-tree commit (no ref written), and the real-manifest case
  skips only on a shallow checkout that lacks the commit - the gate itself keeps
  running on both fetch-depth-0 rails, which the next test asserts.

All five files pass locally (30, 11, 38, 3 and 18 tests); lint with the frozen
suppressions is clean.

tests/unit/check-provider-asset-provenance.test.ts

@@ -40,7 +40,10 @@ function writeManifest(
     recordType: "manifest",
     schemaVersion: 1,
     expectedAssetCount: records.filter((record) => record.recordType === "asset").length,
-    auditedCommit: "091589089cd134a94df9f6cdab9ba562b2cefd18",
+    // HEAD instead of a pinned SHA: the fast-unit shards run on a shallow checkout,
+    // where a historical commit object does not exist and the gate would reject
+    // the fixture before exercising what the test is about.
+    auditedCommit: gitObjectId("HEAD"),
     auditedAt: "2026-08-26",
     legalScope:
       "Provenance records source matching only; it does not establish copyright or trademark clearance.",
@@ -98,12 +101,53 @@ function gitObjectId(revision: string) {
   return result.stdout.trim();
 }
 
-function gitRootCommit() {
-  const result = spawnSync("git", ["-C", REPO_ROOT, "rev-list", "--max-parents=0", "HEAD"], {
+function gitHasCommit(objectId: string) {
+  return (
+    spawnSync("git", ["-C", REPO_ROOT, "cat-file", "-e", `${objectId}^{commit}`], {
+      encoding: "utf8",
+    }).status === 0
+  );
+}
+
+function isShallowRepository() {
+  const result = spawnSync("git", ["-C", REPO_ROOT, "rev-parse", "--is-shallow-repository"], {
     encoding: "utf8",
   });
-  assert.equal(result.status, 0, result.stderr);
-  return result.stdout.trim().split(/\r?\n/)[0];
+  return result.status === 0 && result.stdout.trim() === "true";
+}
+
+/**
+ * A commit whose tree is empty, so every physical provider file is "missing"
+ * from its snapshot. Built as a dangling object (no ref is written) so it also
+ * works on the shallow checkouts the unit shards use, where the root commit is
+ * the grafted HEAD itself and would match the physical snapshot exactly.
+ */
+function emptyTreeCommit() {
+  const tree = spawnSync("git", ["-C", REPO_ROOT, "hash-object", "-w", "-t", "tree", "--stdin"], {
+    input: "",
+    encoding: "utf8",
+  });
+  assert.equal(tree.status, 0, tree.stderr);
+  const identity = {
+    GIT_AUTHOR_NAME: "provenance-fixture",
+    GIT_AUTHOR_EMAIL: "provenance-fixture@example.invalid",
+    GIT_COMMITTER_NAME: "provenance-fixture",
+    GIT_COMMITTER_EMAIL: "provenance-fixture@example.invalid",
+  };
+  const commit = spawnSync(
+    "git",
+    [
+      "-C",
+      REPO_ROOT,
+      "commit-tree",
+      tree.stdout.trim(),
+      "-m",
+      "provenance fixture: empty snapshot",
+    ],
+    { encoding: "utf8", env: { ...process.env, ...identity } }
+  );
+  assert.equal(commit.status, 0, commit.stderr);
+  return commit.stdout.trim();
 }
 
 function workflowJob(source: string, name: string) {
@@ -479,7 +523,7 @@ test("provider asset provenance gate binds auditedCommit to the physical provide
       .trim()
       .split("\n")
       .map((line) => JSON.parse(line));
-    records[0] = { ...records[0], auditedCommit: gitRootCommit() };
+    records[0] = { ...records[0], auditedCommit: emptyTreeCommit() };
     writeFileSync(
       fixture.manifestPath,
       `${records.map((record) => JSON.stringify(record)).join("\n")}\n`
@@ -497,11 +541,17 @@ test("provider asset provenance gate binds auditedCommit to the physical provide
   }
 });
 
-test("repository provider asset manifest covers the audited 142-file snapshot", () => {
-  const result = runGate(
-    join(REPO_ROOT, "public/providers"),
-    join(REPO_ROOT, "config/quality/provider-assets-provenance.jsonl")
-  );
+test("repository provider asset manifest covers the audited 142-file snapshot", (t) => {
+  const manifestPath = join(REPO_ROOT, "config/quality/provider-assets-provenance.jsonl");
+  const { auditedCommit } = JSON.parse(readFileSync(manifestPath, "utf8").split("\n")[0]);
+  if (!gitHasCommit(auditedCommit) && isShallowRepository()) {
+    // The real manifest pins a historical commit. The unit shards check out with
+    // depth 1, so it is not fetched there; the gate itself still runs on both
+    // blocking rails with fetch-depth 0 (asserted by the test right below).
+    t.skip(`shallow checkout without auditedCommit ${auditedCommit}`);
+    return;
+  }
+  const result = runGate(join(REPO_ROOT, "public/providers"), manifestPath);
 
   assert.equal(result.status, 0, `${result.stdout}\n${result.stderr}`);
   assert.match(

tests/unit/gemini-malformed-function-call-finish-reason-2462.test.ts

@@ -70,6 +70,12 @@ function runGeminiToClaude(geminiChunk) {
     const converted = openaiToClaudeResponse(chunk, claudeState);
     if (converted) claudeEvents.push(...converted);
   }
+  // End-of-stream flush: production calls the translator once more with `null`
+  // when the upstream stream closes (open-sse/utils/stream.ts flush →
+  // translateResponse(..., null, state)). Since dd35750e5f a finish chunk that
+  // carries no usage is deferred until that flush, so the driver must mirror it.
+  const flushed = openaiToClaudeResponse(null, claudeState);
+  if (flushed) claudeEvents.push(...flushed);
   return { openaiEvents, claudeEvents };
 }
 

tests/unit/model-lifecycle-integration.test.ts

@@ -115,7 +115,12 @@ test("unified catalog suppresses stale OpenAI chat rows but retains typed media"
   assert.equal(ids.has("openai/gpt-5.2-codex"), false);
   assert.equal(ids.has("openai/sora-2"), false);
   assert.equal(ids.has("openai/sora-2-pro"), false);
-  assert.equal(ids.has("openai/gpt-5.6-sol"), true);
+  // Since #11919 (fixes #11829) an authoritative live catalog REPLACES the static
+  // registry: a static-only row like gpt-5.6-sol that the synced catalog does not
+  // list is suppressed instead of leaking into /v1/models. The lifecycle contract
+  // this file guards (#8627: stale chat rows suppressed, typed media retained)
+  // is unchanged — only the "static rows survive a sync" expectation moved.
+  assert.equal(ids.has("openai/gpt-5.6-sol"), false);
   assert.equal(body.data.find((item) => item.id === "openai/gpt-image-2")?.type, "image");
 });
 

tests/unit/stream-markdown-token-boundary.test.ts

@@ -203,7 +203,10 @@ test("OpenAI to Claude: finish flushes a fully-held boundary before message stop
     },
     state
   );
-  const result = flatten([chunk1, chunk2]);
+  // End-of-stream flush (see dd35750e5f): a finish chunk without usage is deferred
+  // until production's null flush, so mirror it before asserting the terminal events.
+  const chunk3 = openaiToClaudeResponse(null, state);
+  const result = flatten([chunk1, chunk2, chunk3]);
 
   assert.deepEqual(getTextDeltas(result), ["`"]);
   assert.equal(state._markdownBuffer, "");
@@ -251,7 +254,10 @@ test("OpenAI to Claude: tool call flushes a fully-held boundary before tool use"
     },
     state
   );
-  const result = flatten([chunk1, chunk2]);
+  // End-of-stream flush (see dd35750e5f): a finish chunk without usage is deferred
+  // until production's null flush, so mirror it before asserting the terminal events.
+  const chunk3 = openaiToClaudeResponse(null, state);
+  const result = flatten([chunk1, chunk2, chunk3]);
   const contentEvents = result.filter((event) =>
     String((event as Record<string, unknown>).type).startsWith("content_block_")
   );

tests/unit/translator-tool-call-shim.test.ts

@@ -1,12 +1,10 @@
 import test from "node:test";
 import assert from "node:assert/strict";
 
-const { applyToolCallShimToBuffer, hasToolCallShim, __test } = await import(
-  "../../open-sse/translator/helpers/toolCallShim.ts"
-);
-const { openaiToClaudeResponse } = await import(
-  "../../open-sse/translator/response/openai-to-claude.ts"
-);
+const { applyToolCallShimToBuffer, hasToolCallShim, __test } =
+  await import("../../open-sse/translator/helpers/toolCallShim.ts");
+const { openaiToClaudeResponse } =
+  await import("../../open-sse/translator/response/openai-to-claude.ts");
 
 const { coerceToArray } = __test as { coerceToArray: (v: unknown) => unknown[] };
 
@@ -120,30 +118,21 @@ test("applyToolCallShimToBuffer: Read coerces numeric-string limit/offset", () =
 
 test("applyToolCallShimToBuffer: Read strips pages for non-PDF files", () => {
   const out = JSON.parse(
-    applyToolCallShimToBuffer(
-      "Read",
-      JSON.stringify({ file_path: "/etc/hosts", pages: "1-3" })
-    )
+    applyToolCallShimToBuffer("Read", JSON.stringify({ file_path: "/etc/hosts", pages: "1-3" }))
   );
   assert.equal("pages" in out, false);
 });
 
 test("applyToolCallShimToBuffer: Read strips malformed pages even on PDFs", () => {
   const out = JSON.parse(
-    applyToolCallShimToBuffer(
-      "Read",
-      JSON.stringify({ file_path: "/tmp/doc.pdf", pages: "abc" })
-    )
+    applyToolCallShimToBuffer("Read", JSON.stringify({ file_path: "/tmp/doc.pdf", pages: "abc" }))
   );
   assert.equal("pages" in out, false);
 });
 
 test("applyToolCallShimToBuffer: Read accepts a single page on PDFs", () => {
   const out = JSON.parse(
-    applyToolCallShimToBuffer(
-      "Read",
-      JSON.stringify({ file_path: "/tmp/doc.PDF", pages: "7" })
-    )
+    applyToolCallShimToBuffer("Read", JSON.stringify({ file_path: "/tmp/doc.PDF", pages: "7" }))
   );
   assert.equal(out.pages, "7");
 });
@@ -255,6 +244,12 @@ function streamChunks(chunks: any[], state: any): any[] {
     const out = openaiToClaudeResponse(c, state);
     if (out) all.push(...out);
   }
+  // End-of-stream flush: production calls the translator once more with `null`
+  // when the upstream stream closes (open-sse/utils/stream.ts flush →
+  // translateResponse(..., null, state)). Since dd35750e5f a finish chunk that
+  // carries no usage is deferred until that flush, so the driver must mirror it.
+  const flushed = openaiToClaudeResponse(null, state);
+  if (flushed) all.push(...flushed);
   return all;
 }