Keep context failures from poisoning account pools

src/lib/router.js

@@ -225,6 +225,11 @@ function classifyFailure(status, errorText) {
     status === 429 ||
     /rate[ _-]?limit|too many requests|quota|usage[ _-]?limit|resource[ _-]?exhaust|capacity|overload/.test(text);
   if (quota) return { eligible: true, kind: "quota", defaultCooldownMs: COOLDOWN_MS.quota };
+  const request =
+    /context[ _-]?window|context[ _-]?length[ _-]?exceed|maximum[ _-]?context[ _-]?length|input.{0,80}(?:too[ _-]?long|too[ _-]?large|exceed.{0,40}(?:context|token))|request[ _-]?too[ _-]?large/.test(
+      text
+    );
+  if (request) return { eligible: false, kind: "request", defaultCooldownMs: 0 };
   const capability =
     (status === 400 || status === 404 || status === 422) &&
     (/(?:unsupported|invalid|unknown)[ _-]?model|model[ _-]?(?:not[ _-]?found|unsupported|unavailable)/.test(
@@ -322,13 +327,19 @@ function parseEarlyResponsesFailure(text, response) {
       /usage[ _-]?limit|rate[ _-]?limit|quota|resource[ _-]?exhaust|insufficient[ _-]?quota/i.test(
         signature
       );
+    const request =
+      /context[ _-]?window|context[ _-]?length[ _-]?exceed|maximum[ _-]?context[ _-]?length|input.{0,80}(?:too[ _-]?long|too[ _-]?large|exceed.{0,40}(?:context|token))|request[ _-]?too[ _-]?large/i.test(
+        signature
+      );
     const reportedStatus = Number(error?.status ?? data?.status);
     const status =
       Number.isInteger(reportedStatus) && reportedStatus >= 400 && reportedStatus <= 599
         ? reportedStatus
         : quota
           ? 429
-          : 502;
+          : request
+            ? 400
+            : 502;
     const message =
       error?.message || data?.message || error?.code || error?.type || "Upstream stream failed";
     return {

tests/router-fallback.test.js

@@ -126,10 +126,79 @@ describe("provider failure classification", () => {
       kind: "request",
       defaultCooldownMs: 0,
     });
+    assert.deepEqual(
+      classifyFailure(
+        502,
+        "Your input exceeds the context window of this model. Please adjust your input and try again."
+      ),
+      { eligible: false, kind: "request", defaultCooldownMs: 0 }
+    );
   });
 });
 
 describe("same-provider OAuth account fallback", () => {
+  it("does not turn context failures into cooldowns or a terminal 429", async () => {
+    const store = createStore(tmpConfig());
+    const quotaLock = {
+      until: Date.now() + 60 * 60_000,
+      status: 429,
+      kind: "quota",
+      reason: "usage limit reached",
+    };
+    store.seed({
+      providers: [
+        chatgptAccount("prov_a", "token-a", 100),
+        chatgptAccount("prov_b", "token-b", 200),
+        chatgptAccount("prov_c", "token-c", 300, { modelLocks: { "*": quotaLock } }),
+      ],
+    });
+    const calls = [];
+    const logger = captureLogger();
+    const router = createRouter({
+      store,
+      logger,
+      fetchImpl: async (_url, options) => {
+        calls.push(authToken(options));
+        return new Response(
+          [
+            "event: response.failed",
+            `data: ${JSON.stringify({
+              type: "response.failed",
+              response: {
+                error: {
+                  message:
+                    "Your input exceeds the context window of this model. Please adjust your input and try again.",
+                },
+              },
+            })}`,
+            "",
+          ].join("\n"),
+          { status: 200, headers: { "Content-Type": "text/event-stream" } }
+        );
+      },
+    });
+
+    const result = await router.chatCompletions({
+      body: {
+        model: "chatgpt/gpt-5.4",
+        messages: [{ role: "user", content: "oversized request" }],
+        stream: true,
+      },
+    });
+
+    assert.equal(result.ok, false);
+    assert.equal(result.status, 400);
+    assert.deepEqual(calls, ["token-a", "token-b"]);
+    assert.match(result.error.error.message, /ChatGPT \(oauth1\) \[400\]/);
+    assert.match(result.error.error.message, /ChatGPT \(oauth2\) \[400\]/);
+    assert.match(result.error.error.message, /ChatGPT \(oauth3\) \[429\]/);
+    const providers = store.load().providers;
+    assert.equal(providers.find((provider) => provider.id === "prov_a").modelLocks?.["gpt-5.4"], undefined);
+    assert.equal(providers.find((provider) => provider.id === "prov_b").modelLocks?.["gpt-5.4"], undefined);
+    assert.deepEqual(providers.find((provider) => provider.id === "prov_c").modelLocks?.["*"], quotaLock);
+    assert.equal(logger.entries.find((entry) => entry.meta?.event === "accounts_exhausted").meta.status, 400);
+  });
+
   it("assigns monotonic oauth aliases and advertises only shared model ids", () => {
     const configPath = tmpConfig();
     const store = createStore(configPath);