fix(resilience): mark embed connection terminal on hard upstream failure so dead accounts are not re-hit (#10347)

changelog.d/fixes/10225-combo-context-overflow-before-compression.md

@@ -1 +0,0 @@
-- **fix(combo):** defer the known-context-overflow hard rejection for compressible requests so compression runs before the final context gate, instead of a raw-body estimate 400'ing generic Responses clients targeting a large model before OmniRoute can shrink it ([#10225](https://github.com/diegosouzapw/OmniRoute/issues/10225))

open-sse/handlers/embeddings.ts

@@ -35,6 +35,7 @@ import {
   prepareStructuredEmbeddingRequest,
 } from "./embeddingStructuredInput.ts";
 import { MAX_EMBEDDING_INLINE_ITEM_BYTES } from "@/shared/validation/schemas/apiV1";
+import { markAccountUnavailable } from "../../src/sse/services/auth.ts";
 
 interface ClientRawRequest {
   endpoint: string;
@@ -389,6 +390,28 @@ export async function handleEmbedding({
         connectionId,
       }).catch(() => {});
 
+      // #10347 — persist a connection-level failure marker on a hard upstream failure so
+      // the dead account is not re-selected and re-hit on the next embed request (chat
+      // parity). markAccountUnavailable classifies the status via checkFallbackError: a
+      // payment-required 402 becomes the TERMINAL state credits_exhausted (the terminal
+      // marker excludes the account from selection until an operator resets it), benign
+      // 4xx are a no-op, and terminal statuses are never overwritten. honors per-connection
+      // disableCooling. The write must never break the error response path, so it is
+      // best-effort.
+      if (connectionId) {
+        try {
+          await markAccountUnavailable(
+            connectionId,
+            response.status,
+            errorText,
+            provider,
+            model
+          );
+        } catch {
+          // swallow — the upstream error response takes priority
+        }
+      }
+
       return {
         success: false,
         status: response.status,

open-sse/services/combo.ts

@@ -591,8 +591,6 @@ export async function handleComboChat({
   nesting = null,
   hiddenModelsByProvider = getHiddenModelsByProvider(),
   clientManagedResponsesContext = false,
-  deferContextOverflowWhenCompressible = false,
-  compressionExclusions,
 }: HandleComboChatOptions): Promise<Response> {
   const comboCtx = createComboContext({ body, combo, settings, relayOptions, log });
   const {
@@ -653,8 +651,6 @@ export async function handleComboChat({
     signal,
     apiKeyAllowedConnections,
     hiddenModelsByProvider,
-    deferContextOverflowWhenCompressible,
-    compressionExclusions,
     runCombo: handleComboChat,
   });
   if (fusionDispatch) return fusionDispatch;
@@ -704,8 +700,6 @@ export async function handleComboChat({
     signal,
     apiKeyAllowedConnections,
     hiddenModelsByProvider,
-    deferContextOverflowWhenCompressible,
-    compressionExclusions,
     runCombo: handleComboChat,
   });
   if (runtimeUnitDispatch) return runtimeUnitDispatch;
@@ -729,8 +723,6 @@ export async function handleComboChat({
       signal,
       hiddenModelsByProvider,
       clientManagedResponsesContext,
-      deferContextOverflowWhenCompressible,
-      compressionExclusions,
       relayOptions,
     });
   }
@@ -758,8 +750,6 @@ export async function handleComboChat({
     buildAutoCandidates,
     hiddenModelsByProvider,
     clientManagedResponsesContext,
-    deferContextOverflowWhenCompressible,
-    compressionExclusions,
   });
   if ("earlyResponse" in targetResolution) return targetResolution.earlyResponse;
   const { stickyWeightedLimit, getWeightedStepKeyForTarget, preScreenMap } = targetResolution;
@@ -2451,8 +2441,6 @@ async function handleRoundRobinCombo({
   nesting = null,
   hiddenModelsByProvider = getHiddenModelsByProvider(),
   clientManagedResponsesContext,
-  deferContextOverflowWhenCompressible = false,
-  compressionExclusions,
   relayOptions,
 }: HandleRoundRobinOptions): Promise<Response> {
   const config = settings
@@ -2510,8 +2498,6 @@ async function handleRoundRobinCombo({
   const evalRankedTargets = orderTargetsByEvalScores(tagFilteredTargets, config.evalRouting, log);
   const knownContextOverflow = getKnownContextOverflow(evalRankedTargets, body, {
     clientManagedResponsesContext,
-    deferContextOverflowWhenCompressible,
-    compressionExclusions,
   });
   if (knownContextOverflow) {
     return errorResponseWithComboDiagnostics(

open-sse/services/combo/dispatchPrelude.ts

@@ -76,10 +76,6 @@ type PreludeBaseOptionArgs = {
   apiKeyAllowedConnections?: string[] | null;
   hiddenModelsByProvider?: HiddenModelsByProvider;
   clientManagedResponsesContext?: boolean;
-  /** #10225 — defer the hard context-overflow preflight when compression is enabled. */
-  deferContextOverflowWhenCompressible?: boolean;
-  /** Server-side compression exclusions (#8034). */
-  compressionExclusions?: import("../compression/exclusions.ts").CompressionExclusions;
 };
 
 /** Rebuild handleComboChat's option bag verbatim for a recursive dispatch. */
@@ -97,8 +93,6 @@ function buildBaseOptions(a: PreludeBaseOptionArgs): HandleComboChatOptions {
     apiKeyAllowedConnections: a.apiKeyAllowedConnections,
     hiddenModelsByProvider: a.hiddenModelsByProvider,
     clientManagedResponsesContext: a.clientManagedResponsesContext,
-    deferContextOverflowWhenCompressible: a.deferContextOverflowWhenCompressible,
-    compressionExclusions: a.compressionExclusions,
   };
 }
 
@@ -372,8 +366,6 @@ export async function tryFusionDispatch(args: {
   signal?: AbortSignal | null;
   apiKeyAllowedConnections?: string[] | null;
   hiddenModelsByProvider?: HiddenModelsByProvider;
-  deferContextOverflowWhenCompressible?: boolean;
-  compressionExclusions?: import("../compression/exclusions.ts").CompressionExclusions;
   runCombo: RunCombo;
 }): Promise<Response | null> {
   const { cfg, combo, config, strategy, log } = args;
@@ -597,8 +589,6 @@ export async function tryRuntimeUnitDispatch(args: {
   signal?: AbortSignal | null;
   apiKeyAllowedConnections?: string[] | null;
   hiddenModelsByProvider?: HiddenModelsByProvider;
-  deferContextOverflowWhenCompressible?: boolean;
-  compressionExclusions?: import("../compression/exclusions.ts").CompressionExclusions;
   runCombo: RunCombo;
 }): Promise<Response | null> {
   const { body, combo, config, strategy, allCombos, log, settings } = args;

open-sse/services/combo/knownContextOverflow.ts

@@ -17,7 +17,6 @@
  */
 
 import { getResolvedModelCapabilities } from "../modelCapabilities.ts";
-import { isCompressionExcluded, type CompressionExclusions } from "../compression/exclusions.ts";
 import { deriveRequestCompatibilityRequirements } from "./comboStructure.ts";
 import type { ResolvedComboTarget } from "./types.ts";
 
@@ -29,19 +28,6 @@ export type KnownContextOverflow = {
   targetCount: number;
 };
 
-export type KnownContextOverflowOptions = {
-  clientManagedResponsesContext?: boolean;
-  /**
-   * When prompt compression is enabled for this request (global compression switch
-   * AND not API-key opted-out), defer the hard preflight so chatCore's compression
-   * pipeline runs before the final context gate — instead of a raw-body estimate
-   * rejecting a compressible request up front. (#10225)
-   */
-  deferContextOverflowWhenCompressible?: boolean;
-  /** Server-side compression exclusions (#8034) — targets matching one cannot run compression. */
-  compressionExclusions?: CompressionExclusions;
-};
-
 // #7177: an empty array/object (e.g. a default `messages: []` some combo entrypoints inject
 // when the caller sent none) has no real content — counting it would charge a few phantom
 // "structural" tokens (JSON.stringify braces/brackets) toward the estimate, which is enough
@@ -83,7 +69,7 @@ export function getKnownContextLimit(
 export function getKnownContextOverflow(
   targets: ResolvedComboTarget[],
   body: Record<string, unknown>,
-  options: KnownContextOverflowOptions = {}
+  options: { clientManagedResponsesContext?: boolean } = {}
 ): KnownContextOverflow | null {
   if (targets.length === 0) return null;
   // Native Codex Responses clients compact their own item history. Let the concrete
@@ -99,31 +85,6 @@ export function getKnownContextOverflow(
   ) {
     return null;
   }
-  // #10225: a conservative raw-body context estimate must not be treated as proof
-  // that a compression-enabled request cannot fit. When compression is available
-  // for this request AND at least one target can actually run it, defer the hard
-  // rejection so handleChatCore runs proactive compression (chatCore.ts) and its
-  // post-compression enforceOutputTokenBudget becomes the final context gate —
-  // returning a local `context_length_exceeded` only if the compressed body still
-  // cannot fit (no upstream dispatch). Each excluded/native-codex-passthrough
-  // target is skipped; if no target can compress, the fast preflight is kept.
-  if (
-    options.deferContextOverflowWhenCompressible === true &&
-    targets.some(
-      (target) =>
-        !isCompressionExcluded(
-          {
-            provider: target.provider,
-            model: target.modelStr.includes("/")
-              ? target.modelStr.split("/").slice(1).join("/")
-              : target.modelStr,
-          },
-          options.compressionExclusions
-        )
-    )
-  ) {
-    return null;
-  }
   const requirements = deriveRequestCompatibilityRequirements(body);
   if (requirements.requiredContextTokens <= 0) return null;
 

open-sse/services/combo/targetResolution.ts

@@ -115,10 +115,6 @@ export interface ResolveComboTargetPipelineDeps {
   hiddenModelsByProvider?: HiddenModelsByProvider;
   /** Native Responses clients (for example Codex CLI/Desktop) manage compaction themselves. */
   clientManagedResponsesContext?: boolean;
-  /** #10225 — defer the hard context-overflow preflight when compression is enabled for this request. */
-  deferContextOverflowWhenCompressible?: boolean;
-  /** Server-side compression exclusions (#8034) — which targets can run compression. */
-  compressionExclusions?: import("../compression/exclusions.ts").CompressionExclusions;
 }
 
 export interface ResolvedComboTargetPipeline {
@@ -734,8 +730,6 @@ export async function resolveComboTargetPipeline(
 
   const overflow = getKnownContextOverflow(orderedTargets, body, {
     clientManagedResponsesContext: deps.clientManagedResponsesContext,
-    deferContextOverflowWhenCompressible: deps.deferContextOverflowWhenCompressible,
-    compressionExclusions: deps.compressionExclusions,
   });
   if (overflow) {
     return { earlyResponse: buildContextOverflowResponse(overflow, orderedTargets, log) };

open-sse/services/combo/types.ts

@@ -6,7 +6,6 @@
  * — logic unchanged, re-exported from combo.ts for backward compatibility.
  */
 
-import type { CompressionExclusions } from "../compression/exclusions.ts";
 import type { ProviderCandidate } from "../autoCombo/scoring.ts";
 
 export const RESET_WINDOW_NAMES = ["weekly", "session", "monthly"] as const;
@@ -113,15 +112,6 @@ export type HandleComboChatOptions = {
   hiddenModelsByProvider?: HiddenModelsByProvider;
   /** Native Responses clients (for example Codex CLI/Desktop) manage compaction themselves. */
   clientManagedResponsesContext?: boolean;
-  /**
-   * #10225: request-scoped flag — prompt compression is enabled for this request
-   * (global compression switch ON and not opted-out by the API key). When set, the
-   * combo preflight defers its hard context-overflow rejection so chatCore's
-   * compression runs before the final context gate.
-   */
-  deferContextOverflowWhenCompressible?: boolean;
-  /** Server-side compression exclusions (#8034) — used to check which targets can run compression. */
-  compressionExclusions?: CompressionExclusions;
 };
 
 export type HandleRoundRobinOptions = Omit<HandleComboChatOptions, "apiKeyAllowedConnections">;

src/app/api/v1/providers/[provider]/embeddings/route.ts

@@ -84,7 +84,14 @@ export async function POST(request, { params }) {
     );
   }
 
-  const result = await handleEmbedding({ body, credentials, log });
+  const result = await handleEmbedding({
+    body,
+    credentials,
+    log,
+    // #10347 — thread the selected connection id so a hard upstream failure cools
+    // the account instead of re-hitting it on every request.
+    connectionId: (credentials as { connectionId?: string } | null)?.connectionId ?? null,
+  });
 
   if (result.success) {
     await clearRecoveredProviderState(credentials);

src/lib/embeddings/service.ts

@@ -302,7 +302,12 @@ export async function createEmbeddingResponse(
       clientRawRequest: options.clientRawRequest || null,
       apiKeyId: options.apiKeyId || null,
       apiKeyName: options.apiKeyName || null,
-      connectionId: options.connectionId || null,
+      // #10347 — thread the selected connection id so handleEmbedding can cool the
+      // account on a hard upstream failure (previously always null on /v1/embeddings).
+      connectionId:
+        ((credentials as { connectionId?: string } | null)?.connectionId) ||
+        options.connectionId ||
+        null,
     });
 
   const result = connectionIdForProxy

src/sse/handlers/chat.ts

@@ -33,8 +33,6 @@ import type { SingleModelTarget } from "@omniroute/open-sse/services/combo/types
 import { mergeAbortSignals } from "@omniroute/open-sse/executors/base.ts";
 import { resolveRequestAutoControls } from "@omniroute/open-sse/services/autoCombo/requestControls.ts";
 import { isVerifiedNativeCodexRequest } from "@omniroute/open-sse/config/codexIdentity.ts";
-import { resolveCompressionSettings } from "@omniroute/open-sse/handlers/chatCore/compressionSettings.ts";
-import type { CompressionExclusions } from "@omniroute/open-sse/services/compression/exclusions.ts";
 import { resolveComboConfig } from "@omniroute/open-sse/services/comboConfig.ts";
 import { injectHandoffIntoBody } from "@omniroute/open-sse/services/contextHandoff.ts";
 import {
@@ -211,31 +209,6 @@ let combosCacheTs = 0;
 let combosCacheVersionSnapshot = -1;
 const COMBOS_CACHE_TTL_MS = 10_000;
 
-/**
- * #10225 — resolve whether this request's combo preflight should DEFER its hard
- * context-overflow rejection so chatCore's compression runs first.
- *
- * Mirrors handleChatCore's own enablement determination (chatCore.ts): defer only
- * when the global compression switch is ON and the API key has not opted out
- * (`apiKeyInfo.compressionEnabled !== false`). Per-target applicability (server-side
- * exclusions) is checked inside getKnownContextOverflow via the returned exclusions.
- * Fail closed (defer=false) on any lookup error — the existing hard preflight stays.
- */
-async function resolveComboContextOverflowDeferral(
-  logger: { warn?: (...args: unknown[]) => void } | null | undefined,
-  apiKeyInfo: { compressionEnabled?: boolean } | null | undefined
-): Promise<{ defer: boolean; exclusions: CompressionExclusions | undefined }> {
-  try {
-    const compression = await resolveCompressionSettings(logger);
-    return {
-      defer: compression.enabled && apiKeyInfo?.compressionEnabled !== false,
-      exclusions: compression.settings?.exclusions,
-    };
-  } catch {
-    return { defer: false, exclusions: undefined };
-  }
-}
-
 async function getCombosCachedForChat(): Promise<unknown[]> {
   const now = Date.now();
   // Explicit non-null check: we intentionally cache and return the Promise
@@ -851,13 +824,9 @@ async function handleChatImplementation(
 
     // Context-relay keeps generation in combo.ts, but handoff injection lives here
     // because only this layer knows which connectionId was actually selected.
-    const { defer: deferContextOverflowWhenCompressible, exclusions: compressionExclusions } =
-      await resolveComboContextOverflowDeferral(log, apiKeyInfo);
     const response = await (handleComboChat as any)({
       body,
       combo,
-      deferContextOverflowWhenCompressible,
-      compressionExclusions,
       clientManagedResponsesContext:
         sourceFormat === "openai-responses" &&
         new URL(request.url).pathname.split("/").includes("responses") &&
@@ -1134,13 +1103,9 @@ async function handleSingleModelChat(
     );
     log.info("ROUTING", `Auto-combo redirect from handleSingleModelChat for "${modelStr}"`);
     log.info("ROUTING", `Auto-combo redirect to combo flow for "${modelStr}"`);
-    const { defer: sNetDefer, exclusions: sNetExclusions } =
-      await resolveComboContextOverflowDeferral(log, apiKeyInfo);
     return handleComboChat({
       body,
       combo: redirectCombo,
-      deferContextOverflowWhenCompressible: sNetDefer,
-      compressionExclusions: sNetExclusions,
       clientManagedResponsesContext:
         detectFormatFromEndpoint(body, clientRawRequest?.endpoint || "") === "openai-responses" &&
         String(clientRawRequest?.endpoint || "")

tests/unit/10347-embed-402-cooldown.test.ts

@@ -0,0 +1,103 @@
+/**
+ * TDD regression (#10347): the embed path reads a connection's cooldown at
+ * selection time but NEVER writes one on a terminal upstream failure. A Mistral
+ * (or any) connection returning HTTP 402 "payment required — Check your
+ * subscription" on embeds is re-selected and re-hit upstream on every request —
+ * the repeated EMBED/ERROR/ProxyEgress storm on 3.8.49. Chat wires the cooldown
+ * write (`markAccountUnavailable`) on hard failures; embed never does.
+ *
+ * Repro: create a real mistral apikey connection, mock `globalThis.fetch` to
+ * return HTTP 402 with a payment-required JSON body, call `handleEmbedding`
+ * with that connectionId, then assert the connection's `rate_limited_until`
+ * becomes a future timestamp. Today it stays `undefined` (RED); with the fix
+ * `markAccountUnavailable` persists a 1h QUOTA_EXHAUSTED cooldown (GREEN).
+ */
+import { test } from "node:test";
+import assert from "node:assert/strict";
+import fs from "node:fs";
+import os from "node:os";
+import path from "node:path";
+
+const TEST_DATA_DIR = fs.mkdtempSync(path.join(os.tmpdir(), "omniroute-embed-402-"));
+process.env.DATA_DIR = TEST_DATA_DIR;
+
+const core = await import("../../src/lib/db/core.ts");
+const providersDb = await import("../../src/lib/db/providers.ts");
+const { handleEmbedding } = await import("../../open-sse/handlers/embeddings.ts");
+
+test.after(() => {
+  core.resetDbInstance();
+  fs.rmSync(TEST_DATA_DIR, { recursive: true, force: true });
+});
+
+function readConnectionRow(connId: string) {
+  const db = core.getDbInstance() as unknown as {
+    prepare: (sql: string) => {
+      get: (id: string) => {
+        test_status: unknown;
+        rate_limited_until: unknown;
+        last_error_type: unknown;
+      } | undefined;
+    };
+  };
+  return db
+    .prepare(
+      "SELECT test_status, rate_limited_until, last_error_type FROM provider_connections WHERE id = ?"
+    )
+    .get(connId);
+}
+
+test("embed 402 marks the connection terminal credits_exhausted (stops re-selection)", async () => {
+  const conn = await providersDb.createProviderConnection({
+    provider: "mistral",
+    authType: "apikey",
+    name: "embed 402 cooldown",
+  });
+  const connId = (conn as { id: string }).id;
+
+  const originalFetch = globalThis.fetch;
+  globalThis.fetch = async () =>
+    new Response(
+      JSON.stringify({
+        code: "subscription_inactive",
+        message: "Check your subscription",
+      }),
+      {
+        status: 402,
+        headers: { "content-type": "application/json" },
+      }
+    );
+
+  try {
+    const result = await handleEmbedding({
+      body: { model: "mistral/mistral-embed", input: "ping" },
+      credentials: { apiKey: "mistral-key" },
+      connectionId: connId,
+      log: null,
+    });
+
+    // The upstream was hit and surfaced a 402 — the bug scope.
+    assert.equal(result.success, false);
+    assert.equal(result.status, 402);
+
+    const row = readConnectionRow(connId);
+    // markAccountUnavailable classifies a payment-required 402 as the TERMINAL state
+    // credits_exhausted (last_error_type quota_exhausted) with no transient numeric
+    // cooldown — the terminal marker is what excludes the account from the embed
+    // selection path on the next request, stopping the repeat re-hit storm.
+    assert.equal(
+      row?.test_status,
+      "credits_exhausted",
+      `expected the 402 to mark the connection terminal (test_status=credits_exhausted) on ${connId}, got ${String(
+        row?.test_status
+      )}`
+    );
+    assert.equal(
+      row?.last_error_type,
+      "quota_exhausted",
+      `expected last_error_type=quota_exhausted on ${connId}, got ${String(row?.last_error_type)}`
+    );
+  } finally {
+    globalThis.fetch = originalFetch;
+  }
+});

tests/unit/combo-context-overflow-compression-probe.test.ts

@@ -1,173 +0,0 @@
-import test from "node:test";
-import assert from "node:assert/strict";
-import fs from "node:fs";
-import os from "node:os";
-import path from "node:path";
-
-/**
- * #10225 — combo known-context-overflow must NOT hard-reject a compressible
- * request before OmniRoute's compression pipeline can run.
- *
- * Root cause: getKnownContextOverflow() estimates the RAW body (ceil(serializedChars/4)
- * over the whole Responses input[]) during combo target resolution, before any
- * compression. When every known target limit is below that raw estimate, both call
- * sites (round-robin + target-resolution) convert it into an immediate local 400
- * `context_length_exceeded` with attempted:0 — so chatCore's proactive compression
- * (which can shrink 294133→111529, 62% in the reporter's case) never runs. The only
- * existing bypass (clientManagedResponsesContext) is gated to VERIFIED native Codex
- * clients, so a generic Responses client (e.g. OpenCode) pointed at a codex model
- * still hits the hard gate.
- *
- * Fix: thread a request-scoped `deferContextOverflowWhenCompressible` flag (set when
- * the global compression switch is ON and not API-key opted-out). When set AND at
- * least one target can run compression, getKnownContextOverflow returns null so the
- * request reaches chatCore, whose post-compression enforceOutputTokenBudget becomes
- * the final context gate — a local 400 only if the compressed body still cannot fit.
- */
-
-const TEST_DATA_DIR = fs.mkdtempSync(path.join(os.tmpdir(), "omniroute-combo-overflow-compress-"));
-const ORIGINAL_DATA_DIR = process.env.DATA_DIR;
-process.env.DATA_DIR = TEST_DATA_DIR;
-
-const core = await import("../../src/lib/db/core.ts");
-const { saveModelsDevCapabilities, clearModelsDevCapabilities } =
-  await import("../../src/lib/modelsDevSync.ts");
-const { getKnownContextOverflow, handleComboChat } = await import(
-  "../../open-sse/services/combo.ts"
-);
-
-test.after(() => {
-  core.resetDbInstance();
-  if (ORIGINAL_DATA_DIR === undefined) {
-    delete process.env.DATA_DIR;
-  } else {
-    process.env.DATA_DIR = ORIGINAL_DATA_DIR;
-  }
-  fs.rmSync(TEST_DATA_DIR, { recursive: true, force: true });
-});
-
-test.beforeEach(() => {
-  clearModelsDevCapabilities();
-});
-
-function capabilityEntry(limitContext: number | null) {
-  return {
-    tool_call: true,
-    reasoning: false,
-    attachment: false,
-    structured_output: true,
-    temperature: true,
-    modalities_input: JSON.stringify(["text"]),
-    modalities_output: JSON.stringify(["text"]),
-    knowledge_cutoff: null,
-    release_date: null,
-    last_updated: null,
-    status: null,
-    family: null,
-    open_weights: false,
-    limit_context: limitContext,
-    limit_input: limitContext,
-    limit_output: 4096,
-    interleaved_field: null,
-  };
-}
-
-function target(modelStr: string) {
-  return {
-    kind: "model" as const,
-    stepId: modelStr,
-    executionKey: modelStr,
-    modelStr,
-    provider: modelStr.includes("/") ? modelStr.split("/")[0] : modelStr,
-    providerId: null,
-    connectionId: null,
-    weight: 1,
-    label: null,
-  };
-}
-
-// A generic Responses-API body whose estimate lands near `tokens` tokens (4 chars/token).
-// Uses `input:` (not `messages:`) to mirror the OpenCode/Codex Responses surface.
-function bigResponsesBody(tokens: number) {
-  return { input: [["user", "x".repeat(tokens * 4)]] };
-}
-
-const noopLog = { info() {}, warn() {}, error() {}, debug() {} };
-
-test("#10225 getKnownContextOverflow defers the hard overflow when compression is available", () => {
-  saveModelsDevCapabilities({ codex: { "gpt-5.6-terra": capabilityEntry(272_000) } });
-  const body = bigResponsesBody(275_000);
-
-  // Compression enabled + target can compress -> defer (null).
-  assert.equal(
-    getKnownContextOverflow([target("codex/gpt-5.6-terra")], body, {
-      deferContextOverflowWhenCompressible: true,
-    }),
-    null,
-    "compressible request must defer so chatCore compression can run (#10225)"
-  );
-
-  // Compression disabled -> the existing hard overflow is preserved (never lose #7177).
-  const hard = getKnownContextOverflow([target("codex/gpt-5.6-terra")], body);
-  assert.ok(hard);
-  assert.ok(hard.requiredContextTokens > hard.maxKnownContextTokens);
-
-  // Compression enabled but EVERY target is excluded from compression -> keep the hard gate.
-  const excluded = getKnownContextOverflow([target("codex/gpt-5.6-terra")], body, {
-    deferContextOverflowWhenCompressible: true,
-    compressionExclusions: ["gpt-5.6-terra"],
-  });
-  assert.ok(excluded, "fully-excluded targets must retain the hard preflight");
-});
-
-test("#10225 combo does not early-400 a compressible over-limit request when deferral is on", async () => {
-  saveModelsDevCapabilities({ codex: { "gpt-5.6-terra": capabilityEntry(272_000) } });
-  let dispatches = 0;
-
-  const response = await handleComboChat({
-    body: bigResponsesBody(275_000),
-    combo: {
-      name: "codex-compress-overflow",
-      strategy: "priority",
-      models: ["codex/gpt-5.6-terra"],
-    },
-    deferContextOverflowWhenCompressible: true,
-    clientManagedResponsesContext: false,
-    isModelAvailable: async () => true,
-    handleSingleModel: async () => {
-      dispatches += 1;
-      return new Response("ok", { status: 200 });
-    },
-    log: noopLog,
-  });
-
-  assert.notEqual(response.status, 400, "compression-enabled request must reach chatCore");
-  assert.equal(dispatches, 1, "must dispatch so chatCore compaction runs first");
-});
-
-test("#10225 combo keeps the fast 400 when compression is disabled", async () => {
-  saveModelsDevCapabilities({ codex: { "gpt-5.6-terra": capabilityEntry(272_000) } });
-  let dispatches = 0;
-
-  const response = await handleComboChat({
-    body: bigResponsesBody(275_000),
-    combo: {
-      name: "codex-compress-disabled",
-      strategy: "priority",
-      models: ["codex/gpt-5.6-terra"],
-    },
-    deferContextOverflowWhenCompressible: false,
-    clientManagedResponsesContext: false,
-    isModelAvailable: async () => true,
-    handleSingleModel: async () => {
-      dispatches += 1;
-      return new Response("ok", { status: 200 });
-    },
-    log: noopLog,
-  });
-
-  assert.equal(response.status, 400);
-  assert.equal(dispatches, 0, "#7177 anti-exhaustion guard must survive when compression is off");
-  const body = await response.json();
-  assert.equal(body.error.code, "context_length_exceeded");
-});