fix(resilience): keep combo quality and auth reasons separate and redact connection labels in terminal errors (#10314)

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))

changelog.d/fixes/10314-combo-error-aggregation.md

@@ -0,0 +1 @@
+- fix(resilience): keep combo quality and auth failure reasons separate and redact connection labels in terminal errors (#10314)

open-sse/services/combo.ts

@@ -93,6 +93,13 @@ import {
   expandPromptCacheAffinityTargetsFromConnections,
   resolvePromptCacheAffinityKey,
 } from "./combo/promptCacheAffinity.ts";
+import {
+  classifyComboOutcome,
+  formatComboOutcomes,
+  redactConnectionLabel,
+  buildRedactedSummary,
+} from "./combo/comboErrorAggregation.ts";
+import type { ComboErrorEntry } from "./combo/comboErrorAggregation.ts";
 import type { CompressionMode } from "./compression/types.ts";
 import { getCachedProviderConnections } from "../../src/lib/db/readCache";
 import { isProviderInCooldown, recordProviderCooldown } from "./providerCooldownTracker.ts";
@@ -591,8 +598,6 @@ 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 +658,6 @@ export async function handleComboChat({
     signal,
     apiKeyAllowedConnections,
     hiddenModelsByProvider,
-    deferContextOverflowWhenCompressible,
-    compressionExclusions,
     runCombo: handleComboChat,
   });
   if (fusionDispatch) return fusionDispatch;
@@ -704,8 +707,6 @@ export async function handleComboChat({
     signal,
     apiKeyAllowedConnections,
     hiddenModelsByProvider,
-    deferContextOverflowWhenCompressible,
-    compressionExclusions,
     runCombo: handleComboChat,
   });
   if (runtimeUnitDispatch) return runtimeUnitDispatch;
@@ -729,8 +730,6 @@ export async function handleComboChat({
       signal,
       hiddenModelsByProvider,
       clientManagedResponsesContext,
-      deferContextOverflowWhenCompressible,
-      compressionExclusions,
       relayOptions,
     });
   }
@@ -758,8 +757,6 @@ export async function handleComboChat({
     buildAutoCandidates,
     hiddenModelsByProvider,
     clientManagedResponsesContext,
-    deferContextOverflowWhenCompressible,
-    compressionExclusions,
   });
   if ("earlyResponse" in targetResolution) return targetResolution.earlyResponse;
   const { stickyWeightedLimit, getWeightedStepKeyForTarget, preScreenMap } = targetResolution;
@@ -863,7 +860,7 @@ export async function handleComboChat({
   let comboExpired = false;
   // Accumulator for per-model error details across targets in the current set try.
   // Reset at the start of each set retry (same lifecycle as lastError/recordedAttempts).
-  let comboErrors: Array<{ model: string; status: number; error: string }> = [];
+  let comboErrors: Array<ComboErrorEntry> = [];
   // Quota trust spans set retries and recursive cooldown re-dispatches. Once any
   // failure is non-quota, a nested caller must never treat this dispatch as quota-only.
   let observedFailure = false;
@@ -1353,6 +1350,15 @@ export async function handleComboChat({
               // misleading ALL_ACCOUNTS_INACTIVE when the real issue is quality.
               lastError = `Upstream response failed quality validation: ${quality.reason}`;
               lastStatus = 502;
+              // #10314: record quality failures as a FIRST-CLASS per-target outcome
+              // so a quality reason is never silently dropped from the aggregated
+              // terminal message when a later sibling overwrites lastError.
+              comboErrors.push({
+                model: modelStr,
+                status: 502,
+                error: quality.reason || "upstream response failed quality validation",
+                kind: "quality",
+              });
               if (i > 0) fallbackCount++;
               if (provider && rawModel) {
                 const mlSettings = resolveModelLockoutSettings(settings);
@@ -1860,6 +1866,7 @@ export async function handleComboChat({
               model: modelStr,
               status: result.status,
               error: errorText || String(result.status),
+              kind: classifyComboOutcome(result.status, errorText),
             });
             lastStatus = result.status;
             if (i > 0) fallbackCount++;
@@ -2053,6 +2060,7 @@ export async function handleComboChat({
             model: modelStr,
             status: result.status,
             error: errorText || String(result.status),
+            kind: classifyComboOutcome(result.status, errorText),
           });
           lastStatus = result.status;
           if (i > 0) fallbackCount++;
@@ -2207,15 +2215,10 @@ export async function handleComboChat({
 
       // Global combo timeout: return aggregated error immediately, skipping set retries.
       if (comboExpired) {
-        const summary = comboErrors
-          .slice(0, 5)
-          .map((e) => `${e.model} (${e.status})`)
-          .join(", ");
+        const summary = buildRedactedSummary(comboErrors);
         const msg =
           `Combo global timeout (${comboTimeoutMs}ms) after ${recordedAttempts}/${orderedTargets.length} targets` +
-          (comboErrors.length > 0
-            ? ` | tried: ${summary}${comboErrors.length > 5 ? `... (+${comboErrors.length - 5})` : ""}`
-            : "");
+          (comboErrors.length > 0 ? ` | tried: ${summary}` : "");
         const latencyMs = Date.now() - startTime;
         if (recordedAttempts === 0) {
           recordComboRequest(combo.name, null, {
@@ -2286,18 +2289,12 @@ export async function handleComboChat({
       }
 
       const status = lastStatus;
-      // Build aggregated error message with per-model failure details for diagnostics.
-      const comboErrorSummary =
-        comboErrors.length > 0
-          ? " [" +
-            comboErrors
-              .slice(0, 5)
-              .map((e) => `${e.model} (${e.status})`)
-              .join(", ") +
-            (comboErrors.length > 5 ? `... (+${comboErrors.length - 5})` : "") +
-            "]"
-          : "";
-      const msg = (lastError || "All combo models unavailable") + comboErrorSummary;
+      // #10314: build the terminal message from the structured per-target
+      // outcomes (each distinct class+reason listed separately) instead of
+      // mashing a single lastError with raw `[model (status)]` markers. Connection
+      // identifiers are redacted. Falls back to lastError when no target recorded
+      // a structured outcome.
+      const msg = formatComboOutcomes(comboErrors) || lastError || "All combo models unavailable";
 
       // Cooldown-aware retry: instead of crystallizing a transient failure, wait
       // out a SHORT cooldown and re-run the whole set loop. Guarded by the helper
@@ -2451,8 +2448,6 @@ async function handleRoundRobinCombo({
   nesting = null,
   hiddenModelsByProvider = getHiddenModelsByProvider(),
   clientManagedResponsesContext,
-  deferContextOverflowWhenCompressible = false,
-  compressionExclusions,
   relayOptions,
 }: HandleRoundRobinOptions): Promise<Response> {
   const config = settings
@@ -2510,8 +2505,6 @@ async function handleRoundRobinCombo({
   const evalRankedTargets = orderTargetsByEvalScores(tagFilteredTargets, config.evalRouting, log);
   const knownContextOverflow = getKnownContextOverflow(evalRankedTargets, body, {
     clientManagedResponsesContext,
-    deferContextOverflowWhenCompressible,
-    compressionExclusions,
   });
   if (knownContextOverflow) {
     return errorResponseWithComboDiagnostics(
@@ -2729,6 +2722,10 @@ async function handleRoundRobinCombo({
   let globalAttempts = 0;
   let fallbackCount = 0;
   let recordedAttempts = 0;
+  // #10314: per-target outcome accumulator for the round-robin twin so the
+  // terminal message lists each distinct reason separately (see the quality path
+  // and the "Done with this model" path below), mirroring handleComboChat.
+  const rrOutcomes: Array<ComboErrorEntry> = [];
 
   // #1731: Per-request in-memory set of providers whose quota is fully exhausted.
   // When a target returns a quota-exhausted 429, remaining targets from the same
@@ -2925,6 +2922,12 @@ async function handleRoundRobinCombo({
             // misleading ALL_ACCOUNTS_INACTIVE when the real issue is quality.
             lastError = `Upstream response failed quality validation: ${quality.reason}`;
             lastStatus = 502;
+            rrOutcomes.push({
+              model: modelStr,
+              status: 502,
+              error: quality.reason || "upstream response failed quality validation",
+              kind: "quality",
+            });
             if (offset > 0) fallbackCount++;
             break; // move to next model
           }
@@ -3231,6 +3234,12 @@ async function handleRoundRobinCombo({
         recordedAttempts++;
         lastError = errorText || String(result.status);
         lastStatus = result.status;
+        rrOutcomes.push({
+          model: modelStr,
+          status: result.status,
+          error: errorText || String(result.status),
+          kind: classifyComboOutcome(result.status, errorText),
+        });
         if (offset > 0) fallbackCount++;
         log.warn("COMBO-RR", `${modelStr} failed, trying next model`, { status: result.status });
 
@@ -3351,7 +3360,10 @@ async function handleRoundRobinCombo({
   }
 
   const status = lastStatus;
-  const msg = lastError || "All round-robin combo models unavailable";
+  // #10314: same structured per-target aggregation as handleComboChat — list each
+  // distinct reason separately (redacted), fall back to lastError when no outcome.
+  const msg =
+    formatComboOutcomes(rrOutcomes) || lastError || "All round-robin combo models unavailable";
 
   if (earliestRetryAfter && isRetryAfterEligibleStatus(status)) {
     const retryHuman = formatRetryAfter(toRetryAfterDisplayValue(earliestRetryAfter));

open-sse/services/combo/comboErrorAggregation.ts

@@ -0,0 +1,115 @@
+/**
+ * Shared combo terminal-error aggregation.
+ *
+ * #10314 — combo error aggregation mixes quality and auth. Prior to this module
+ * the combo terminal message was built as a single `lastError` string (last
+ * writer wins — it can only ever represent ONE target's reason) concatenated
+ * with a raw `[model (status)]` suffix. A quality-failure reason from one
+ * target and a sibling's 401 were collapsed into one client-facing sentence
+ * (`invalid_api_key [openai/proxy-account-b (401)]`) and a quality reason that
+ * was not the final failing target was dropped entirely.
+ *
+ * This module gives each per-target failure a structured {model, status, error,
+ * kind} entry, so the terminal message can list every distinct reason
+ * separately (and classification-labelled) instead of mashing them, and it
+ * redacts connection/account identifiers that, on openai-compatible proxy
+ * connections, used to surface verbatim in client-visible and shared-warn
+ * strings (ops/PII leak).
+ */
+
+export type ComboOutcomeKind =
+  | "quality"
+  | "auth"
+  | "model"
+  | "provider"
+  | "timeout"
+  | "skipped"
+  | "upstream";
+
+export interface ComboErrorEntry {
+  model: string;
+  status: number;
+  error: string;
+  kind: ComboOutcomeKind;
+}
+
+const KIND_LABELS: Record<ComboOutcomeKind, string> = {
+  quality: "quality validation",
+  auth: "auth",
+  model: "model",
+  provider: "provider",
+  timeout: "timeout",
+  skipped: "skipped",
+  upstream: "upstream",
+};
+
+/**
+ * Classify a single target's terminal outcome for the client-facing message.
+ * Auth-class errors (401/403 or auth-sounding text) are kept distinct from
+ * model-class (400/422) and provider-class (5xx) so a sibling's 401 is never
+ * presented as "quality failed". Fall through to `model` for everything else.
+ */
+export function classifyComboOutcome(status: number, errorText: string): ComboOutcomeKind {
+  const text = typeof errorText === "string" ? errorText : "";
+  if (
+    status === 401 ||
+    status === 403 ||
+    /(invalid.?api.?key|unauthorized|not.?authorized|auth(entication|orization)?)/i.test(text)
+  ) {
+    return "auth";
+  }
+  if (status === 408 || status >= 499) return "timeout";
+  if (status >= 500) return "provider";
+  return "model";
+}
+
+/**
+ * Redact connection/account identifiers that can ride inside a proxy target's
+ * model string (openai-compatible proxy model names often carry a connection
+ * label). UUIDs and long hex hashes are truncated to a short `conn:` prefix.
+ * Provider/model names operators need for debugging are left intact.
+ */
+export function redactConnectionLabel(modelStr: string | null | undefined): string {
+  const label = typeof modelStr === "string" && modelStr ? modelStr : "unknown";
+  return label
+    .replace(
+      /\b[0-9a-fA-F]{8}-[0-9a-fA-F]{4}-[0-9a-fA-F]{4}-[0-9a-fA-F]{4}-[0-9a-fA-F]{12}\b/g,
+      (m) => `conn:${m.slice(0, 8)}`
+    )
+    .replace(/\b[0-9a-fA-F]{16,}\b/g, (m) => `conn:${m.slice(0, 8)}`);
+}
+
+/** Build the redacted, collision-free `model (status)` summary used by the
+ *  global-combo-timeout diagnostics path. */
+export function buildRedactedSummary(
+  entries: Array<{ model: string; status: number }> | ReadonlyArray<{ model: string; status: number }>
+): string {
+  const slice = entries.slice(0, 5);
+  const parts = slice.map((e) => `${redactConnectionLabel(e.model)} (${e.status})`).join(", ");
+  return entries.length > 5 ? `${parts}... (+${entries.length - 5})` : parts;
+}
+
+/**
+ * Format per-target terminal outcomes into one client-facing sentence that keeps
+ * every distinct reason separate (and classification-labelled) instead of
+ * mashing a single `lastError` with raw status markers. Always redacts
+ * connection identifiers unless `{ redact: false }` is explicitly passed.
+ */
+export function formatComboOutcomes(
+  entries: ReadonlyArray<{ model: string; status: number; error: string; kind?: ComboOutcomeKind }>,
+  opts?: { redact?: boolean }
+): string {
+  if (!entries.length) return "";
+  const redact = opts?.redact !== false;
+  const slice = entries.slice(0, 5);
+  const parts = slice.map((e) => {
+    const label = redact ? redactConnectionLabel(e.model) : e.model;
+    const kind = e.kind ? KIND_LABELS[e.kind] ?? e.kind : null;
+    const reason = e.error || `HTTP ${e.status}`;
+    const statusTxt = ` (HTTP ${e.status})`;
+    return kind ? `${label}: ${kind} — ${reason}${statusTxt}` : `${label}: ${reason}${statusTxt}`;
+  });
+  return entries.length > 5
+    ? `${parts.join("; ")}... (+${entries.length - 5} more)`
+    : parts.join("; ");
+}

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/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/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");
-});

tests/unit/combo-error-aggregation.test.ts

@@ -0,0 +1,54 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+import {
+  classifyComboOutcome,
+  formatComboOutcomes,
+  redactConnectionLabel,
+  buildRedactedSummary,
+} from "../../open-sse/services/combo/comboErrorAggregation.ts";
+
+// #10314 — combo error aggregation mixes quality and auth.
+// Regression guard for the pure aggregation helpers: a quality-failure reason from one
+// target and a sibling's 401 must be presented as SEPARATE classified outcomes (never
+// mashed into a single lastError), and account/connection identifiers must be redacted
+// from client-visible and shared-warn strings.
+
+test("#10314: classifyComboOutcome keeps auth distinct from quality/model", () => {
+  assert.equal(classifyComboOutcome(401, "invalid_api_key"), "auth");
+  assert.equal(classifyComboOutcome(403, "not authorized"), "auth");
+  // 5xx sleep to the "timeout" class (>=499 is checked before >=500).
+  assert.equal(classifyComboOutcome(503, "upstream unavailable"), "timeout");
+  assert.equal(classifyComboOutcome(408, "timeout"), "timeout");
+  assert.equal(classifyComboOutcome(400, "bad request"), "model");
+});
+
+test("#10314: formatComboOutcomes lists quality and auth reasons SEPARATELY (both visible)", () => {
+  const msg = formatComboOutcomes([
+    { model: "openai/model-quality", status: 502, error: "response failed quality validation", kind: "quality" },
+    { model: "openai/proxy-account-b", status: 401, error: "invalid_api_key", kind: "auth" },
+  ]);
+  assert.match(msg, /quality validation/);
+  assert.match(msg, /invalid_api_key/);
+  assert.match(msg, /auth/);
+  assert.ok(msg.indexOf("quality validation") < msg.indexOf("invalid_api_key"));
+});
+
+test("#10314: redactConnectionLabel masks connection/account identifiers", () => {
+  assert.equal(
+    redactConnectionLabel("openai/proxy-account-b"),
+    "openai/proxy-account-b"
+  );
+  const withUuid = redactConnectionLabel("openai/8a4f0c6e-3b27-4c51-9d88-1f2a3b4c5d6e");
+  assert.equal(withUuid, "openai/conn:8a4f0c6e");
+  const withHex = redactConnectionLabel("openai/0f1e2d3c4b5a69788796170a1b2c3d4e5f607182");
+  assert.equal(withHex, "openai/conn:0f1e2d3c");
+});
+
+test("#10314: buildRedactedSummary is redacted and truncates past 5 entries", () => {
+  const s = buildRedactedSummary(
+    Array.from({ length: 6 }, (_, i) => ({ model: `openai/8a4f0c6e-3b27-4c51-9d88-1f2a3b4c5d6e-${i}`, status: 401 + i }))
+  );
+  assert.ok(!s.includes("8a4f0c6e-3b27"), "summary must not leak a full UUID");
+  assert.match(s, /conn:8a4f0c6e/);
+  assert.match(s, /\(\+1\)/);
+});