feat(dashboard): orchestration canvas — unified model + snapshot hook (part 1/2) (#12156)

Modelo puro do Orchestration Canvas (tipos, 3 mappers, mergeSnapshot com dedupe/staleness/cap, projeções flow+overview) + hook de polling com gatilho WS. Ciclo completo: 9 tasks TDD com review por task, review final whole-branch + fixes verificados 6/6, refactor de complexity re-validado (comportamento preservado). CI: 18 pass. Testes: 31 node:test + 2 vitest. Parte 2/2 (UI /dashboard/orchestration) na sequência.

changelog.d/features/12155-weighted-score-router-strategy.md

@@ -0,0 +1 @@
+- **feat(routing):** add a `score` Auto router strategy that selects the highest configured weighted score and reuses `explorationRate`.

changelog.d/fixes/11761-embedding-registry-width-and-type.md

@@ -0,0 +1,5 @@
+- Keep the embedding registry's vector width and `embedding` type on models when a synced model exists
+  for the same id, so `/v1/models` no longer reports registry-described embedding models widthless or
+  untyped (#11761)
+- Correct `google/gemini-embedding-001` on the OpenRouter route to 3072 dimensions, the width it
+  returns when `dimensions` is not sent (#11761)

changelog.d/fixes/abort-guard-absorb-request-signal-aborted.md

@@ -0,0 +1 @@
+- Absorb `Error [AbortError]: request_signal_aborted` and DOMException AbortError shapes in the process-level client-abort crash guard so routine client disconnects no longer kill the server (exit code 7).

config/quality/eslint-suppressions.json

@@ -829,12 +829,6 @@
   "src/app/(dashboard)/dashboard/HomePageClient.tsx": {
     "@typescript-eslint/no-unused-vars": {
       "count": 2
-    },
-    "react-hooks/exhaustive-deps": {
-      "count": 1
-    },
-    "react-hooks/set-state-in-effect": {
-      "count": 2
     }
   },
   "src/app/(dashboard)/dashboard/a2a/page.tsx": {
@@ -850,62 +844,16 @@
       "count": 1
     }
   },
-  "src/app/(dashboard)/dashboard/activity/ActivityFeedClient.tsx": {
-    "react-hooks/purity": {
-      "count": 1
-    },
-    "react-hooks/refs": {
-      "count": 1
-    },
-    "react-hooks/set-state-in-effect": {
-      "count": 1
-    }
-  },
-  "src/app/(dashboard)/dashboard/analytics/CacheHealthTab.tsx": {
-    "react-hooks/set-state-in-effect": {
-      "count": 1
-    }
-  },
-  "src/app/(dashboard)/dashboard/analytics/ComboHealthTab.tsx": {
-    "react-hooks/set-state-in-effect": {
-      "count": 1
-    }
-  },
   "src/app/(dashboard)/dashboard/analytics/CompressionAnalyticsTab.tsx": {
     "@typescript-eslint/no-unused-vars": {
       "count": 1
     }
   },
-  "src/app/(dashboard)/dashboard/analytics/ProviderUtilizationTab.tsx": {
-    "react-hooks/set-state-in-effect": {
-      "count": 1
-    }
-  },
-  "src/app/(dashboard)/dashboard/analytics/RouteExplainabilityTab.tsx": {
-    "react-hooks/set-state-in-effect": {
-      "count": 2
-    }
-  },
   "src/app/(dashboard)/dashboard/api-manager/ApiManagerPageClient.tsx": {
     "@typescript-eslint/no-unused-vars": {
       "count": 1
     }
   },
-  "src/app/(dashboard)/dashboard/audit/A2aAuditTab.tsx": {
-    "react-hooks/set-state-in-effect": {
-      "count": 1
-    }
-  },
-  "src/app/(dashboard)/dashboard/audit/ComplianceTab.tsx": {
-    "react-hooks/set-state-in-effect": {
-      "count": 1
-    }
-  },
-  "src/app/(dashboard)/dashboard/audit/McpAuditTab.tsx": {
-    "react-hooks/set-state-in-effect": {
-      "count": 2
-    }
-  },
   "src/app/(dashboard)/dashboard/batch/components/wizard/CostEstimateStep.tsx": {
     "react-hooks/set-state-in-effect": {
       "count": 1
@@ -926,24 +874,6 @@
       "count": 1
     }
   },
-  "src/app/(dashboard)/dashboard/cache/components/CacheEntriesTab.tsx": {
-    "react-hooks/set-state-in-effect": {
-      "count": 1
-    }
-  },
-  "src/app/(dashboard)/dashboard/cache/components/ReasoningCacheTab.tsx": {
-    "react-hooks/purity": {
-      "count": 1
-    },
-    "react-hooks/set-state-in-effect": {
-      "count": 1
-    }
-  },
-  "src/app/(dashboard)/dashboard/cache/page.tsx": {
-    "react-hooks/set-state-in-effect": {
-      "count": 1
-    }
-  },
   "src/app/(dashboard)/dashboard/cli-agents/CliAgentsPageClient.tsx": {
     "no-restricted-syntax": {
       "count": 4
@@ -1060,31 +990,6 @@
       "count": 2
     }
   },
-  "src/app/(dashboard)/dashboard/memory/components/EditMemoryModal.tsx": {
-    "react-hooks/set-state-in-effect": {
-      "count": 1
-    }
-  },
-  "src/app/(dashboard)/dashboard/memory/components/QdrantConfigCard.tsx": {
-    "react-hooks/set-state-in-effect": {
-      "count": 1
-    }
-  },
-  "src/app/(dashboard)/dashboard/memory/components/tabs/MemoriesTab.tsx": {
-    "react-hooks/set-state-in-effect": {
-      "count": 1
-    }
-  },
-  "src/app/(dashboard)/dashboard/memory/hooks/useEngineStatus.ts": {
-    "react-hooks/set-state-in-effect": {
-      "count": 1
-    }
-  },
-  "src/app/(dashboard)/dashboard/memory/hooks/useMemorySettings.ts": {
-    "react-hooks/set-state-in-effect": {
-      "count": 1
-    }
-  },
   "src/app/(dashboard)/dashboard/onboarding/page.tsx": {
     "@typescript-eslint/no-unused-vars": {
       "count": 1
@@ -1185,29 +1090,6 @@
       "count": 1
     }
   },
-  "src/app/(dashboard)/dashboard/radar/RadarCatalogTable.tsx": {
-    "react-hooks/set-state-in-effect": {
-      "count": 1
-    }
-  },
-  "src/app/(dashboard)/dashboard/radar/intel/page.tsx": {
-    "react-hooks/set-state-in-effect": {
-      "count": 1
-    }
-  },
-  "src/app/(dashboard)/dashboard/radar/page.tsx": {
-    "react-hooks/set-state-in-effect": {
-      "count": 2
-    }
-  },
-  "src/app/(dashboard)/dashboard/radar/setup/page.tsx": {
-    "react-hooks/preserve-manual-memoization": {
-      "count": 1
-    },
-    "react-hooks/set-state-in-effect": {
-      "count": 1
-    }
-  },
   "src/app/(dashboard)/dashboard/relay/RelayProxyClient.tsx": {
     "react-hooks/set-state-in-effect": {
       "count": 1
@@ -1376,11 +1258,6 @@
       "count": 1
     }
   },
-  "src/app/(dashboard)/dashboard/usage/components/EvalsTab.tsx": {
-    "react-hooks/set-state-in-effect": {
-      "count": 2
-    }
-  },
   "src/app/(dashboard)/dashboard/usage/components/ProviderLimits/ProviderLimitCard.tsx": {
     "@typescript-eslint/no-unused-vars": {
       "count": 3
@@ -1391,11 +1268,6 @@
       "count": 1
     }
   },
-  "src/app/(dashboard)/dashboard/usage/components/ProviderLimits/useCodexResetCreditRedemption.ts": {
-    "react-hooks/immutability": {
-      "count": 2
-    }
-  },
   "src/app/(dashboard)/dashboard/usage/components/ProviderLimits/utils.tsx": {
     "@typescript-eslint/no-unused-vars": {
       "count": 1
@@ -1404,17 +1276,11 @@
   "src/app/(dashboard)/dashboard/usage/components/RateLimitStatus.tsx": {
     "@typescript-eslint/no-unused-vars": {
       "count": 1
-    },
-    "react-hooks/set-state-in-effect": {
-      "count": 1
     }
   },
   "src/app/(dashboard)/dashboard/usage/components/SessionsTab.tsx": {
     "@typescript-eslint/no-unused-vars": {
       "count": 1
-    },
-    "react-hooks/set-state-in-effect": {
-      "count": 1
     }
   },
   "src/app/(dashboard)/dashboard/webhooks/WebhooksPageClient.tsx": {
@@ -1437,17 +1303,6 @@
       "count": 1
     }
   },
-  "src/app/(dashboard)/home/ProviderQuotaWidget.tsx": {
-    "react-hooks/purity": {
-      "count": 1
-    },
-    "react-hooks/refs": {
-      "count": 1
-    },
-    "react-hooks/set-state-in-effect": {
-      "count": 1
-    }
-  },
   "src/app/api/assess/route.ts": {
     "@typescript-eslint/no-unused-vars": {
       "count": 1
@@ -2298,9 +2153,6 @@
   "src/shared/components/RequestLoggerV2.tsx": {
     "@typescript-eslint/no-unused-vars": {
       "count": 3
-    },
-    "react-hooks/exhaustive-deps": {
-      "count": 6
     }
   },
   "src/shared/components/RequestTimeline.tsx": {

docs/routing/AUTO-COMBO.md

@@ -414,6 +414,8 @@ Persisted `strategy: "auto"` combos can set `config.routerStrategy` (or legacy
 `config.auto.routerStrategy`) to one of:
 
 - `rules` — default weighted scoring
+- `score` — selects the highest configured weighted score. Exact ties preserve configured
+  candidate order; the existing `explorationRate` samples from the full ranked pool.
 - `cost` / `eco` — cheapest healthy provider
 - `latency` / `fast` — lowest p95 latency with reliability penalty
 - `sla-aware` / `sla` — prefer candidates that satisfy p95 latency, error-rate, and optional
@@ -422,7 +424,7 @@ Persisted `strategy: "auto"` combos can set `config.routerStrategy` (or legacy
 
 ### Router strategies in detail
 
-The auto-combo engine exposes 5 pluggable **RouterStrategy** implementations that
+The auto-combo engine exposes 6 pluggable **RouterStrategy** implementations that
 you can swap via `config.routerStrategy` (or the legacy `config.auto.routerStrategy`).
 Each strategy picks one provider from the candidate pool, given a `RoutingContext`
 (task type, tool/vision hints, token estimate, optional SLA policy, optional

open-sse/config/codexIdentity.ts

@@ -570,3 +570,37 @@ export function isVerifiedNativeCodexRequest(
 ): boolean {
   return isCodexOriginatedHeaders(headers) && hasNativeCodexTurnBinding(body);
 }
+
+/**
+ * Detect the Claude Code CLI as the request *client* from request headers.
+ * Used to auto-enable model echo so session restores work when the resolved
+ * upstream model (e.g. `oc/nemotron-3-ultra-free`) is not recognized by the
+ * Claude Code client on `--resume`.
+ */
+export function isClaudeCodeOriginatedHeaders(
+  headers: Headers | Record<string, unknown> | null | undefined
+): boolean {
+  const getHeader = (name: string): string => {
+    if (headers instanceof Headers) {
+      return headers.get(name)?.toLowerCase() ?? "";
+    }
+    if (headers && typeof headers === "object") {
+      for (const [key, value] of Object.entries(headers as Record<string, unknown>)) {
+        if (key.toLowerCase() === name && typeof value === "string") {
+          return value.toLowerCase();
+        }
+      }
+    }
+    return "";
+  };
+
+  // Claude Code identifies itself via the user-agent header
+  const userAgent = getHeader("user-agent");
+  if (userAgent.includes("claude-code") || userAgent.includes("anthropic-ai/claude-code")) {
+    return true;
+  }
+  // Also check originator if present
+  const originator = getHeader("originator");
+  if (originator.startsWith("claude-code")) return true;
+  return false;
+}

open-sse/config/embeddingRegistry.ts

@@ -239,7 +239,7 @@ export const EMBEDDING_PROVIDERS: Record<string, EmbeddingProvider> = {
       {
         id: "google/gemini-embedding-001",
         name: "Gemini Embedding 001 (OpenRouter)",
-        dimensions: 768,
+        dimensions: 3072,
       },
       {
         id: "google/gemini-embedding-2",

open-sse/executors/credential.ts

@@ -0,0 +1,54 @@
+import { assertCommonChatGptWebProviderAvailable } from "@/shared/constants/chatgptWebRetirement";
+import { assertMicrosoftDesignerWebProviderAvailable } from "@/shared/constants/designerWebRetirement";
+import { assertRuntimeProviderAvailable } from "@/shared/constants/providerRetirement";
+import type { BaseExecutor } from "./base.ts";
+import { getDefaultExecutor } from "./defaultResolver.ts";
+
+type CredentialExecutorLoader = () => Promise<BaseExecutor>;
+
+const specializedCredentialExecutors: Record<string, CredentialExecutorLoader> = {
+  antigravity: () => import("./antigravity.ts").then((m) => new m.AntigravityExecutor()),
+  agy: () => import("./antigravity.ts").then((m) => new m.AntigravityExecutor()),
+  github: () => import("./github.ts").then((m) => new m.GithubExecutor()),
+  "ghe-copilot": () => import("./ghe-copilot.ts").then((m) => new m.GheCopilotExecutor()),
+  kiro: () => import("./kiro.ts").then((m) => new m.KiroExecutor()),
+  "amazon-q": () => import("./kiro.ts").then((m) => new m.KiroExecutor("amazon-q")),
+  codex: () => import("./codex.ts").then((m) => new m.CodexExecutor()),
+  cursor: () => import("./cursor.ts").then((m) => new m.CursorExecutor()),
+  cu: () => import("./cursor.ts").then((m) => new m.CursorExecutor()),
+  "cursor-api": () => import("./cursor.ts").then((m) => new m.CursorExecutor("cursor-api")),
+  cua: () => import("./cursor.ts").then((m) => new m.CursorExecutor("cursor-api")),
+  trae: () => import("./trae.ts").then((m) => new m.TraeExecutor()),
+  gitlab: () => import("./gitlab.ts").then((m) => new m.GitlabExecutor()),
+  "gitlab-duo": () => import("./gitlab.ts").then((m) => new m.GitlabExecutor("gitlab-duo")),
+  "zed-hosted": () => import("./zed-hosted.ts").then((m) => new m.ZedHostedExecutor()),
+  "grok-cli": () => import("./grok-cli.ts").then((m) => new m.GrokCliExecutor()),
+  gc: () => import("./grok-cli.ts").then((m) => new m.GrokCliExecutor()),
+  auggie: () => import("./auggie.ts").then((m) => new m.AuggieExecutor()),
+  xai: () => import("./xai.ts").then((m) => new m.XaiExecutor()),
+  "xai-oauth": () => import("./xai.ts").then((m) => new m.XaiExecutor("xai-oauth")),
+  xao: () => import("./xai.ts").then((m) => new m.XaiExecutor("xai-oauth")),
+};
+
+const credentialExecutorCache = new Map<string, Promise<BaseExecutor>>();
+
+/** Resolve only executors with credential-refresh behavior, without loading the chat registry. */
+export async function getCredentialRefreshExecutor(provider: string): Promise<BaseExecutor> {
+  assertMicrosoftDesignerWebProviderAvailable(provider);
+  assertRuntimeProviderAvailable(provider);
+  assertCommonChatGptWebProviderAvailable(provider);
+
+  let executor = credentialExecutorCache.get(provider);
+  if (!executor) {
+    const specializedLoader = specializedCredentialExecutors[provider];
+    executor = specializedLoader
+      ? specializedLoader()
+      : Promise.resolve(getDefaultExecutor(provider));
+    executor = executor.catch((error) => {
+      credentialExecutorCache.delete(provider);
+      throw error;
+    });
+    credentialExecutorCache.set(provider, executor);
+  }
+  return executor;
+}

open-sse/executors/defaultResolver.ts

@@ -0,0 +1,13 @@
+import { DefaultExecutor } from "./default.ts";
+
+const defaultExecutorCache = new Map<string, DefaultExecutor>();
+
+/** Resolve the shared fallback executor without initializing the specialized executor registry. */
+export function getDefaultExecutor(provider: string): DefaultExecutor {
+  let executor = defaultExecutorCache.get(provider);
+  if (!executor) {
+    executor = new DefaultExecutor(provider);
+    defaultExecutorCache.set(provider, executor);
+  }
+  return executor;
+}

open-sse/executors/index.ts

@@ -9,7 +9,7 @@ import {
 } from "./registry.ts";
 // Type-only: pulls no runtime code, keeps DefaultExecutor the only eager class.
 import type { BaseExecutor } from "./base.ts";
-import { DefaultExecutor } from "./default.ts";
+import { getDefaultExecutor } from "./defaultResolver.ts";
 
 // R0.3 — declarative built-in table, made LAZY by #11220.
 //
@@ -207,8 +207,6 @@ for (const [alias, load] of Object.entries(lazyExecutors)) {
   registerLazyExecutor(alias, load);
 }
 
-const defaultCache = new Map();
-
 // #6699 — providers that exist ONLY as Cloud Agent task-API entries
 // (CLOUD_AGENT_PROVIDERS / staticModels "Available Models" catalog) and have no
 // chat-completions REGISTRY entry anywhere in open-sse/. Without this guard,
@@ -251,8 +249,7 @@ export async function getExecutor(provider: string): Promise<BaseExecutor> {
     (err as Error & { status?: number }).status = 400;
     throw err;
   }
-  if (!defaultCache.has(provider)) defaultCache.set(provider, new DefaultExecutor(provider));
-  return defaultCache.get(provider)!;
+  return getDefaultExecutor(provider);
 }
 
 export function hasSpecializedExecutor(provider: string): boolean {

open-sse/handlers/chatCore.ts

@@ -77,7 +77,7 @@ import {
   isStripReasoningRequested,
 } from "./chatCore/headers.ts";
 import { markCodexScopeRateLimited } from "./chatCore/codexFailover.ts";
-import { getCodexClientSessionId, isCodexOriginatedHeaders } from "../config/codexIdentity.ts";
+import { getCodexClientSessionId, isCodexOriginatedHeaders, isClaudeCodeOriginatedHeaders } from "../config/codexIdentity.ts";
 import {
   noteCodexTurnStateProvenance,
   readCodexTurnStateHeader,
@@ -981,8 +981,14 @@ export async function handleChatCore({
   const isCodexResponsesEcho =
     (isResponsesEndpoint || sourceFormat === FORMATS.OPENAI_RESPONSES) &&
     isCodexOriginatedHeaders(clientRawRequest?.headers);
+
+  // Detect Claude Code CLI so we can auto-enable model echo — this prevents
+  // session restore failures when the resolved upstream model (e.g.
+  // `oc/nemotron-3-ultra-free`) is not recognized by the client on `--resume`.
+  const isClaudeCodeClient = isClaudeCodeOriginatedHeaders(clientRawRequest?.headers);
+
   let echoModel =
-    (settings.echoRequestedModelName === true || isCodexResponsesEcho) &&
+    (settings.echoRequestedModelName === true || isCodexResponsesEcho || isClaudeCodeClient) &&
     typeof requestedModel === "string" &&
     requestedModel
       ? requestedModel
@@ -5465,6 +5471,7 @@ export async function handleChatCore({
 
   const streamReadiness = await ensureStreamReadiness(providerResponse, {
     timeoutMs: streamReadinessPolicy.timeoutMs,
+    maxTimeoutMs: streamReadinessPolicy.maxTimeoutMs,
     provider,
     model,
     log,

open-sse/services/autoCombo/routerStrategy.ts

@@ -4,13 +4,14 @@
  * Inspired by ClawRouter commit 14c83c258 "refactor: extract routing into pluggable RouterStrategy system".
  * Provides a RouterStrategy interface and built-in implementations:
  *   - RulesStrategy (default): wraps the existing 15-factor scoring engine
+ *   - ScoreStrategy: highest configured weighted score, with explicit exploration
  *   - CostStrategy: always picks cheapest available model
  *   - LatencyStrategy: prioritizes low p95 latency with reliability weighting
  *   - SLAStrategy: prefers candidates that satisfy latency/error/cost SLOs
  *   - LKGPStrategy: tries last known good provider first
  */
 
-import type { ProviderCandidate, ScoredProvider } from "./scoring.ts";
+import type { ProviderCandidate, ScoredProvider, ScoringWeights } from "./scoring.ts";
 import { scorePool } from "./scoring.ts";
 import { getTaskFitness } from "./taskFitness.ts";
 import { clamp01 } from "../../utils/number.ts";
@@ -32,6 +33,8 @@ export interface RoutingContext {
   lastKnownGoodProvider?: string;
   lkgpEnabled?: boolean;
   sla?: SlaRoutingPolicy;
+  weights?: ScoringWeights;
+  explorationRate?: number;
 }
 
 export interface RoutingDecision {
@@ -108,6 +111,38 @@ class RulesStrategyImpl implements RouterStrategy {
   }
 }
 
+// ── ScoreStrategy: configured score wins, with explicit exploration ──────────
+
+class ScoreStrategyImpl implements RouterStrategy {
+  readonly name = "score";
+  readonly description = "Selects the highest configured weighted score, with explicit exploration";
+
+  select(pool: ProviderCandidate[], context: RoutingContext): RoutingDecision {
+    const eligible = pool.filter((candidate) => candidate.circuitBreakerState !== "OPEN");
+    const ranked = scorePool(
+      eligible.length > 0 ? eligible : pool,
+      context.taskType,
+      context.weights,
+      getTaskFitness
+    );
+    if (ranked.length === 0) throw new Error("[ScoreStrategy] No candidates to score");
+
+    const explorationRate = Math.min(1, Math.max(0, context.explorationRate ?? 0));
+    const isExploration = Math.random() < explorationRate && ranked.length > 1;
+    const selected = isExploration ? ranked[Math.floor(Math.random() * ranked.length)] : ranked[0];
+
+    return {
+      provider: selected.provider,
+      model: selected.model,
+      strategy: this.name,
+      reason: `ScoreStrategy: score=${selected.score.toFixed(3)}${isExploration ? " (exploration)" : ""}`,
+      candidatesConsidered: ranked.length,
+      finalScore: selected.score,
+      connectionId: selected.connectionId,
+    };
+  }
+}
+
 // ── CostStrategy: always picks cheapest healthy provider ─────────────────────
 
 class CostStrategyImpl implements RouterStrategy {
@@ -337,12 +372,14 @@ class LKGPStrategyImpl implements RouterStrategy {
 const strategyRegistry = new Map<string, RouterStrategy>();
 
 const rulesStrategy = new RulesStrategyImpl();
+const scoreStrategy = new ScoreStrategyImpl();
 const costStrategy = new CostStrategyImpl();
 const latencyStrategy = new LatencyStrategyImpl();
 const slaStrategy = new SLAStrategyImpl();
 const lkgpStrategy = new LKGPStrategyImpl();
 
 strategyRegistry.set("rules", rulesStrategy);
+strategyRegistry.set("score", scoreStrategy);
 strategyRegistry.set("cost", costStrategy);
 strategyRegistry.set("eco", costStrategy); // alias
 strategyRegistry.set("latency", latencyStrategy);

open-sse/services/combo/resolveAutoStrategy.ts

@@ -377,6 +377,8 @@ export async function resolveAutoStrategyOrder(
               boolean | undefined,
             estimatedInputTokens,
             sla: slaPolicy,
+            weights,
+            explorationRate,
           },
           routingStrategy
         );

open-sse/services/reasoningInputPolicy.ts

@@ -43,7 +43,19 @@ export function resolveReasoningTransport(
 ): ReasoningTransport {
   const normalized = typeof provider === "string" ? provider.trim().toLowerCase() : "";
   const transport = REASONING_TRANSPORTS.get(normalized);
-  return transport ?? (preserveEncryptedReasoning ? "opaque" : "plaintext");
+  if (transport) return transport;
+  // #12128: Generic Responses-protocol endpoints (e.g. openai-compatible-responses-*,
+  // custom-openai-responses, proxy backends) implement the OpenAI/Codex Responses API
+  // where reasoning input items cannot accept plaintext content (maxItems: 0).
+  if (
+    normalized.startsWith("openai-compatible-responses") ||
+    normalized.startsWith("custom-openai-responses") ||
+    normalized.includes("codex") ||
+    normalized.includes("responses")
+  ) {
+    return "opaque";
+  }
+  return preserveEncryptedReasoning ? "opaque" : "plaintext";
 }
 
 function asRecord(value: unknown): JsonRecord | null {

open-sse/translator/index.ts

@@ -40,6 +40,7 @@ import {
   normalizeResponsesReasoningEffort,
   RESPONSES_STORE_MARKER,
 } from "./request/openai-responses/helpers.ts";
+import { applyReasoningInputPolicy } from "../services/reasoningInputPolicy.ts";
 
 bootstrapTranslatorRegistry();
 export { register } from "./registry.ts";
@@ -575,6 +576,14 @@ export function translateRequest(
   // Normalize openai-responses input shape for providers that require list input.
   if (targetFormat === FORMATS.OPENAI_RESPONSES) {
     result = normalizeOpenAIResponsesRequest(result);
+    // #12128: Sanitize reasoning input items for Responses targets (strip plaintext content for opaque backends)
+    applyReasoningInputPolicy(result as Record<string, unknown>, "responses", {
+      provider,
+      preserveEncryptedReasoning:
+        (credentials as { providerSpecificData?: { preserveEncryptedReasoning?: boolean } } | null)
+          ?.providerSpecificData?.preserveEncryptedReasoning === true,
+      onIncompatibleReasoning: "drop",
+    });
   }
 
   // Second role normalization: only for OPENAI_RESPONSES. Here messages are built from input

open-sse/translator/request/antigravity-to-openai.ts

@@ -220,12 +220,15 @@ function preserveRequired(obj: unknown): void {
     return;
   }
   const record = obj as JsonRecord;
-  if (Array.isArray(record.required) && record.properties && typeof record.properties === "object") {
+  if (
+    Array.isArray(record.required) &&
+    record.properties &&
+    typeof record.properties === "object"
+  ) {
     const properties = record.properties as JsonRecord;
     const valid = (record.required as unknown[]).filter(
       (field) =>
-        typeof field === "string" &&
-        Object.prototype.hasOwnProperty.call(properties, field)
+        typeof field === "string" && Object.prototype.hasOwnProperty.call(properties, field)
     );
     if (valid.length === 0) {
       delete record.required;
@@ -298,12 +301,13 @@ function convertContent(content) {
 
     // Function response → collect all, each becomes a separate tool message
     if (part.functionResponse) {
+      const resp = part.functionResponse.response;
+      const resultPayload =
+        resp && typeof resp === "object" && "result" in resp ? resp.result : (resp ?? {});
       toolResults.push({
         role: "tool",
         tool_call_id: part.functionResponse.id || part.functionResponse.name,
-        content: JSON.stringify(
-          part.functionResponse.response?.result || part.functionResponse.response || {}
-        ),
+        content: JSON.stringify(resultPayload),
       });
     }
   }
@@ -316,9 +320,7 @@ function convertContent(content) {
       const assistantMsg: JsonRecord = { role: "assistant" };
       if (textParts.length > 0) {
         assistantMsg.content =
-          textParts.length === 1 && textParts[0].type === "text"
-            ? textParts[0].text
-            : textParts;
+          textParts.length === 1 && textParts[0].type === "text" ? textParts[0].text : textParts;
       }
       if (reasoningContent) {
         assistantMsg.reasoning_content = reasoningContent;

open-sse/translator/request/gemini-to-openai.ts

@@ -147,12 +147,13 @@ function convertGeminiContent(content) {
     }
 
     if (part.functionResponse) {
+      const resp = part.functionResponse.response;
+      const resultPayload =
+        resp && typeof resp === "object" && "result" in resp ? resp.result : (resp ?? {});
       return {
         role: "tool",
         tool_call_id: part.functionResponse.id || part.functionResponse.name,
-        content: JSON.stringify(
-          part.functionResponse.response?.result || part.functionResponse.response || {}
-        ),
+        content: JSON.stringify(resultPayload),
       };
     }
   }

open-sse/utils/streamReadiness.ts

@@ -471,6 +471,9 @@ export async function ensureStreamReadiness(
   response: Response,
   options: {
     timeoutMs: number;
+    /** Hard ceiling for liveness-extended deadlines. When omitted, no hard ceiling
+     *  is applied beyond `timeoutMs`. */
+    maxTimeoutMs?: number;
     provider?: string | null;
     model?: string | null;
     log?: StreamReadinessLogger | null;
@@ -489,7 +492,14 @@ export async function ensureStreamReadiness(
   };
   const startedAt = Date.now();
   const effectiveTimeoutMs = Math.max(0, Math.floor(options.timeoutMs));
-  const deadline = startedAt + effectiveTimeoutMs;
+  // Hard ceiling: the deadline may extend on liveness signals (bytes arriving),
+  // but never past this absolute maximum.  When maxTimeoutMs is omitted the
+  // initial timeoutMs itself acts as the ceiling (no extension).
+  const maxDeadline =
+    options.maxTimeoutMs != null
+      ? startedAt + Math.max(effectiveTimeoutMs, Math.floor(options.maxTimeoutMs))
+      : startedAt + effectiveTimeoutMs;
+  let deadline = startedAt + effectiveTimeoutMs;
   let handedOffReader = false;
 
   const buildReadyResponse = () =>
@@ -500,7 +510,7 @@ export async function ensureStreamReadiness(
     });
 
   const timeoutReason = () =>
-    `Stream produced no non-ping SSE event within ${effectiveTimeoutMs}ms`;
+    `Stream produced no non-ping SSE event within ${deadline - startedAt}ms (max=${maxDeadline - startedAt}ms)`;
 
   try {
     while (true) {
@@ -593,6 +603,22 @@ export async function ensureStreamReadiness(
       chunks.push(readResult.value);
       const decodedChunk = decoder.decode(readResult.value, { stream: true });
 
+      // Liveness extension: bytes arrived → connection is alive, not dead.
+      // Reset the deadline so slow-but-alive upstreams (reasoning warm-ups,
+      // keepalive-only phases) are not aborted.  The hard ceiling (maxDeadline)
+      // prevents unbounded waits and preserves the operator's fast-fail intent
+      // for truly dead connections.
+      const now = Date.now();
+      if (deadline < maxDeadline) {
+        deadline = Math.min(now + effectiveTimeoutMs, maxDeadline);
+        if (now - startedAt > effectiveTimeoutMs) {
+          options.log?.debug?.(
+            "STREAM",
+            `readiness deadline extended to ${deadline - startedAt}ms (liveness signal) (${options.provider || "provider"}/${options.model || "unknown"})`
+          );
+        }
+      }
+
       if (appendStreamReadinessSignal(readinessState, decodedChunk)) {
         options.log?.debug?.(
           "STREAM",

open-sse/utils/streamReadinessPolicy.ts

@@ -14,6 +14,7 @@ export type StreamReadinessPolicyInput = {
 export type StreamReadinessPolicyResult = {
   timeoutMs: number;
   baseTimeoutMs: number;
+  maxTimeoutMs: number;
   reasons: string[];
 };
 
@@ -121,7 +122,7 @@ export function resolveStreamReadinessTimeout(
 ): StreamReadinessPolicyResult {
   const baseTimeoutMs = Math.max(0, Math.floor(input.baseTimeoutMs || 0));
   if (baseTimeoutMs <= 0) {
-    return { timeoutMs: baseTimeoutMs, baseTimeoutMs, reasons: ["disabled"] };
+    return { timeoutMs: baseTimeoutMs, baseTimeoutMs, maxTimeoutMs: baseTimeoutMs, reasons: ["disabled"] };
   }
 
   const maxTimeoutMs = Math.max(baseTimeoutMs, input.maxTimeoutMs ?? DEFAULT_MAX_TIMEOUT_MS);
@@ -197,5 +198,5 @@ export function resolveStreamReadinessTimeout(
   timeoutMs = Math.min(timeoutMs, maxTimeoutMs);
   if (timeoutMs === baseTimeoutMs) reasons.push("base");
 
-  return { timeoutMs, baseTimeoutMs, reasons };
+  return { timeoutMs, baseTimeoutMs, maxTimeoutMs, reasons };
 }

scripts/dev/run-next.mjs

@@ -137,6 +137,11 @@ function createNextApp() {
   });
 }
 
+// The custom HTTP server owns process exit. Application instrumentation still
+// registers its cleanup function, but must not install a competing signal
+// listener that can race this runner's async server/Next teardown.
+globalThis.__omnirouteCustomServerOwnsShutdown = true;
+
 let nextApp = createNextApp();
 
 // Best-effort self-heal for a corrupted Turbopack persistent dev cache (#6289):
@@ -231,6 +236,7 @@ async function start() {
     systemdNotifier.stopping();
     try {
       await new Promise((resolve) => server.close(resolve));
+      await globalThis.__omnirouteRequestShutdown?.(signal);
       await nextApp.close();
     } catch (error) {
       console.error("[SHUTDOWN] Failed during signal:", signal, error);

src/app/(dashboard)/dashboard/HomePageClient.tsx

@@ -2,7 +2,7 @@
 
 import { useTranslations } from "next-intl";
 
-import { useState, useEffect, useMemo, useCallback, useRef } from "react";
+import { useState, useEffect, useMemo, useCallback, useRef, useSyncExternalStore } from "react";
 import Link from "next/link";
 import { useRouter } from "next/navigation";
 import { Card, CardSkeleton, Button, Modal } from "@/shared/components";
@@ -106,6 +106,12 @@ const INLINE_LINK = "text-primary hover:underline";
 const DOCS_LINK =
   "hidden sm:inline-flex items-center gap-1.5 px-3 py-1.5 rounded-lg text-xs font-medium border border-border text-text-muted hover:text-text-main hover:bg-bg-subtle transition-colors";
 
+// Stable no-op subscription for useSyncExternalStore reads of never-changing
+// browser globals (location.origin does not change without a full navigation).
+function emptySubscribe() {
+  return () => {};
+}
+
 export default function HomePageClient({ machineId }: HomePageClientProps) {
   const router = useRouter();
   const isElectron = useIsElectron();
@@ -115,7 +121,13 @@ export default function HomePageClient({ machineId }: HomePageClientProps) {
   const [providerConnections, setProviderConnections] = useState([]);
   const [models, setModels] = useState([]);
   const [loading, setLoading] = useState(true);
-  const [baseUrl, setBaseUrl] = useState("/v1");
+  // useSyncExternalStore keeps SSR/hydration consistent ("/v1" on the server,
+  // the real origin after hydration) without a setState-in-effect round-trip.
+  const baseUrl = useSyncExternalStore(
+    emptySubscribe,
+    () => `${globalThis.location.origin}/v1`,
+    () => "/v1"
+  );
   const [selectedProvider, setSelectedProvider] = useState(null);
   const [providerMetrics, setProviderMetrics] = useState<Record<string, ProviderMetricSummary>>({});
   const [providerTopology, setProviderTopology] = useState({ lastProvider: "", errorProvider: "" });
@@ -135,36 +147,39 @@ export default function HomePageClient({ machineId }: HomePageClientProps) {
   // Platform detection and download links for Electron
   const platform =
     typeof globalThis.window === "undefined" ? undefined : globalThis.window.electronAPI?.platform;
+  // Destructured to locals: `versionInfo?.current` in a dependency array trips
+  // the lint heuristic that treats any `.current` access as a mutable ref read.
+  const installedVersion = versionInfo?.current || "";
+  const latestVersion = versionInfo?.latest || "";
   const electronDownload = useMemo(() => {
-    const latest = versionInfo?.latest || "";
-    const cleanLatest = latest.replace(/^v/, "");
+    const cleanLatest = latestVersion.replace(/^v/, "");
     if (platform === "darwin") {
       return {
         label: t("downloadDmg"),
         url: `https://github.com/diegosouzapw/OmniRoute/releases/download/v${cleanLatest}/OmniRoute-${cleanLatest}.dmg`,
-        desc: t("downloadDmgDescription", { version: versionInfo?.current || "" }),
+        desc: t("downloadDmgDescription", { version: installedVersion }),
       };
     }
     if (platform === "win32") {
       return {
         label: t("downloadExe"),
         url: `https://github.com/diegosouzapw/OmniRoute/releases/download/v${cleanLatest}/OmniRoute.Setup.${cleanLatest}.exe`,
-        desc: t("downloadExeDescription", { version: versionInfo?.current || "" }),
+        desc: t("downloadExeDescription", { version: installedVersion }),
       };
     }
     if (platform === "linux") {
       return {
         label: t("downloadAppImage"),
         url: `https://github.com/diegosouzapw/OmniRoute/releases/download/v${cleanLatest}/OmniRoute-${cleanLatest}.AppImage`,
-        desc: t("downloadAppImageDescription", { version: versionInfo?.current || "" }),
+        desc: t("downloadAppImageDescription", { version: installedVersion }),
       };
     }
     return {
       label: t("downloadUpdate"),
       url: `https://github.com/diegosouzapw/OmniRoute/releases/tag/v${cleanLatest}`,
-      desc: t("downloadUpdateDescription", { version: versionInfo?.current || "" }),
+      desc: t("downloadUpdateDescription", { version: installedVersion }),
     };
-  }, [platform, t, versionInfo?.latest, versionInfo?.current]);
+  }, [platform, t, latestVersion, installedVersion]);
 
   // Electron internal auto-updater state and listeners
   const [electronUpdateStatus, setElectronUpdateStatus] = useState<{
@@ -234,12 +249,6 @@ export default function HomePageClient({ machineId }: HomePageClientProps) {
       });
   }, []);
 
-  useEffect(() => {
-    if (typeof globalThis.window !== "undefined") {
-      setBaseUrl(`${globalThis.location.origin}/v1`);
-    }
-  }, []);
-
   const fetchData = useCallback(async () => {
     try {
       const [provRes, modelsRes, versionRes] = await Promise.all([
@@ -267,7 +276,9 @@ export default function HomePageClient({ machineId }: HomePageClientProps) {
   }, []);
 
   useEffect(() => {
-    fetchData();
+    void (async () => {
+      await fetchData();
+    })();
   }, [fetchData]);
 
   // Fetch provider nodes for display labels (compat providers)

src/app/(dashboard)/dashboard/activity/ActivityFeedClient.tsx

@@ -1,6 +1,6 @@
 "use client";
 
-import { useState, useEffect, useCallback, useRef } from "react";
+import { useState, useEffect, useCallback } from "react";
 import { useTranslations } from "next-intl";
 import type { AuditLogEntry } from "@/lib/compliance/index";
 import ActivityFeed from "./components/ActivityFeed";
@@ -14,7 +14,9 @@ export default function ActivityFeedClient() {
   const [loading, setLoading] = useState(true);
   const [error, setError] = useState<string | null>(null);
   const [category, setCategory] = useState<EventCategory>("all");
-  const referenceNowMs = useRef<number>(Date.now());
+  // State (not a ref) because it is rendered: refs cannot be read during
+  // render, and Date.now() cannot run there either — the fetch settles it.
+  const [referenceNowMs, setReferenceNowMs] = useState<number>(0);
 
   const fetchEntries = useCallback(async () => {
     setLoading(true);
@@ -30,7 +32,7 @@ export default function ActivityFeedClient() {
       }
       const data = (await res.json()) as AuditLogEntry[];
       // Reset reference time on fresh load so relative timestamps are stable
-      referenceNowMs.current = Date.now();
+      setReferenceNowMs(Date.now());
       setAllEntries(Array.isArray(data) ? data : []);
     } catch (err: unknown) {
       const msg = err instanceof Error ? err.message : t("fetchFailed");
@@ -41,7 +43,9 @@ export default function ActivityFeedClient() {
   }, [t]);
 
   useEffect(() => {
-    fetchEntries();
+    void (async () => {
+      await fetchEntries();
+    })();
   }, [fetchEntries]);
 
   const filtered =
@@ -114,7 +118,7 @@ export default function ActivityFeedClient() {
             <span className="text-sm">{t("loadingActivity")}</span>
           </div>
         ) : (
-          <ActivityFeed entries={filtered} referenceNowMs={referenceNowMs.current} />
+          <ActivityFeed entries={filtered} referenceNowMs={referenceNowMs} />
         )}
       </div>
     </div>

src/app/(dashboard)/dashboard/analytics/CacheHealthTab.tsx

@@ -95,7 +95,9 @@ export default function CacheHealthTab() {
   }, []);
 
   useEffect(() => {
-    void load(range);
+    void (async () => {
+      await load(range);
+    })();
   }, [load, range]);
 
   if (loading) return <Skeleton className="h-64 w-full" />;
@@ -179,8 +181,8 @@ export default function CacheHealthTab() {
                   {text(t, "cacheHealthConcentration", "Where the writes are concentrated")}
                 </h3>
                 <span className="text-xs text-text-muted">
-                  {text(t, "cacheHealthThreshold", "outlier above")} {compact(data.heavyWriteThreshold)}{" "}
-                  {text(t, "cacheHealthTokens", "tokens")}
+                  {text(t, "cacheHealthThreshold", "outlier above")}{" "}
+                  {compact(data.heavyWriteThreshold)} {text(t, "cacheHealthTokens", "tokens")}
                 </span>
               </div>
               <p className="text-sm text-text-main">
@@ -216,7 +218,9 @@ export default function CacheHealthTab() {
               <table className="w-full min-w-[560px] text-sm">
                 <thead>
                   <tr className="border-b border-border text-left text-xs uppercase text-text-muted">
-                    <th className="pb-2 pr-4 font-medium">{text(t, "cacheHealthModel", "Model")}</th>
+                    <th className="pb-2 pr-4 font-medium">
+                      {text(t, "cacheHealthModel", "Model")}
+                    </th>
                     <th className="pb-2 pr-4 text-right font-medium">
                       {text(t, "cacheHealthCalls", "Calls")}
                     </th>

src/app/(dashboard)/dashboard/analytics/ComboHealthTab.tsx

@@ -852,7 +852,9 @@ export default function ComboHealthTab() {
 
   useEffect(() => {
     const controller = new AbortController();
-    fetchData(controller, false);
+    void (async () => {
+      await fetchData(controller, false);
+    })();
     return () => controller.abort();
   }, [fetchData]);
 

src/app/(dashboard)/dashboard/analytics/ProviderUtilizationTab.tsx

@@ -131,7 +131,9 @@ export default function ProviderUtilizationTab() {
   useEffect(() => {
     const controller = new AbortController();
 
-    fetchUtilization(range, aggregateBy, controller.signal);
+    void (async () => {
+      await fetchUtilization(range, aggregateBy, controller.signal);
+    })();
 
     return () => controller.abort();
   }, [fetchUtilization, range, aggregateBy]);
@@ -340,12 +342,8 @@ export default function ProviderUtilizationTab() {
                           <ProviderIcon providerId={providerPart} size={22} />
                         </div>
                         <div>
-                          <p className="text-sm font-semibold text-text-main">
-                            {cardTitle}
-                          </p>
-                          <p className="text-xs text-text-muted">
-                            {cardSubtitle}
-                          </p>
+                          <p className="text-sm font-semibold text-text-main">{cardTitle}</p>
+                          <p className="text-xs text-text-muted">{cardSubtitle}</p>
                         </div>
                       </div>
                       <span

src/app/(dashboard)/dashboard/analytics/RouteExplainabilityTab.tsx

@@ -522,14 +522,18 @@ export default function RouteExplainabilityTab({
 
   useEffect(() => {
     const controller = new AbortController();
-    fetchLogs(controller.signal);
+    void (async () => {
+      await fetchLogs(controller.signal);
+    })();
     return () => controller.abort();
   }, [fetchLogs]);
 
   useEffect(() => {
     if (!selectedId) return;
     const controller = new AbortController();
-    fetchExplanation(selectedId, controller.signal);
+    void (async () => {
+      await fetchExplanation(selectedId, controller.signal);
+    })();
     return () => controller.abort();
   }, [fetchExplanation, selectedId]);
 

src/app/(dashboard)/dashboard/audit/A2aAuditTab.tsx

@@ -64,7 +64,9 @@ export default function A2aAuditTab() {
   }, [offset, skillFilter, stateFilter]);
 
   useEffect(() => {
-    void fetchTasks();
+    void (async () => {
+      await fetchTasks();
+    })();
   }, [fetchTasks]);
 
   return (

src/app/(dashboard)/dashboard/audit/ComplianceTab.tsx

@@ -110,7 +110,9 @@ export default function ComplianceTab() {
   }, [actor, eventType, from, offset, t, to]);
 
   useEffect(() => {
-    void fetchEntries();
+    void (async () => {
+      await fetchEntries();
+    })();
   }, [fetchEntries]);
 
   const visibleEntries = useMemo(() => {
@@ -330,7 +332,9 @@ export default function ComplianceTab() {
                       </td>
                       <td className="px-4 py-3">
                         <span className="rounded-md border border-border bg-surface px-2 py-1 font-mono text-xs text-text-main">
-                          {t.has(`eventTypes.${entry.action}`) ? t(`eventTypes.${entry.action}`) : entry.action}
+                          {t.has(`eventTypes.${entry.action}`)
+                            ? t(`eventTypes.${entry.action}`)
+                            : entry.action}
                         </span>
                       </td>
                       <td className="px-4 py-3">

src/app/(dashboard)/dashboard/audit/McpAuditTab.tsx

@@ -56,7 +56,9 @@ export default function McpAuditTab() {
   }, []);
 
   useEffect(() => {
-    void fetchStats();
+    void (async () => {
+      await fetchStats();
+    })();
   }, [fetchStats]);
 
   const fetchAudit = useCallback(async () => {
@@ -86,7 +88,9 @@ export default function McpAuditTab() {
   }, [offset, successFilter, t, toolFilter]);
 
   useEffect(() => {
-    void fetchAudit();
+    void (async () => {
+      await fetchAudit();
+    })();
   }, [fetchAudit]);
 
   return (

src/app/(dashboard)/dashboard/cache/components/CacheEntriesTab.tsx

@@ -63,7 +63,9 @@ export default function CacheEntriesTab() {
   );
 
   useEffect(() => {
-    fetchEntries();
+    void (async () => {
+      await fetchEntries();
+    })();
   }, [fetchEntries]);
 
   const handleDelete = async (signature: string) => {

src/app/(dashboard)/dashboard/cache/components/ReasoningCacheTab.tsx

@@ -130,9 +130,12 @@ export default function ReasoningCacheTab() {
   const [loading, setLoading] = useState(true);
   const [clearing, setClearing] = useState(false);
   const [expandedId, setExpandedId] = useState<string | null>(null);
+  // Snapshot of "now" taken when the data lands (never during render — the
+  // purity rule bars Date.now() there); entries only render after a fetch.
+  const [nowMs, setNowMs] = useState(0);
 
   const timeAgo = (dateStr: string): string => {
-    const diff = Date.now() - new Date(dateStr).getTime();
+    const diff = nowMs - new Date(dateStr).getTime();
     const minutes = Math.floor(diff / 60000);
     if (minutes < 1) return t("justNow");
     if (minutes < 60) return t("minutesAgo", { minutes });
@@ -147,6 +150,7 @@ export default function ReasoningCacheTab() {
       const res = await fetch("/api/cache/reasoning");
       if (res.ok) {
         const json: ReasoningCacheData = await res.json();
+        setNowMs(Date.now());
         setData(json);
       }
     } catch (error) {
@@ -157,7 +161,9 @@ export default function ReasoningCacheTab() {
   }, []);
 
   useEffect(() => {
-    void fetchData();
+    void (async () => {
+      await fetchData();
+    })();
     const id = setInterval(() => void fetchData(), REFRESH_INTERVAL_MS);
     return () => clearInterval(id);
   }, [fetchData]);

src/app/(dashboard)/dashboard/cache/page.tsx

@@ -374,7 +374,9 @@ export default function CachePage() {
   }, []);
 
   useEffect(() => {
-    void fetchStats();
+    void (async () => {
+      await fetchStats();
+    })();
     const id = setInterval(() => void fetchStats(), REFRESH_INTERVAL_MS);
     return () => clearInterval(id);
   }, [fetchStats]);

src/app/(dashboard)/dashboard/memory/components/EditMemoryModal.tsx

@@ -1,6 +1,6 @@
 "use client";
 
-import { useState, useEffect } from "react";
+import { useState } from "react";
 import { Modal, Button, Input, Select } from "@/shared/components";
 import { useTranslations } from "next-intl";
 
@@ -29,7 +29,14 @@ export default function EditMemoryModal({ memory, isOpen, onClose, onSaved }: Pr
   const [isSaving, setIsSaving] = useState(false);
   const [error, setError] = useState("");
 
-  useEffect(() => {
+  // Adjust-during-render (React docs pattern): when the modal (re)opens for a
+  // memory, seed the form fields from it before painting — no effect round-trip.
+  const [prevSync, setPrevSync] = useState<{ memory: Memory | null; isOpen: boolean }>({
+    memory: null,
+    isOpen: false,
+  });
+  if (memory !== prevSync.memory || isOpen !== prevSync.isOpen) {
+    setPrevSync({ memory, isOpen });
     if (memory && isOpen) {
       setType(memory.type);
       setKey(memory.key);
@@ -38,7 +45,7 @@ export default function EditMemoryModal({ memory, isOpen, onClose, onSaved }: Pr
       setMetadataError("");
       setError("");
     }
-  }, [memory, isOpen]);
+  }
 
   const handleMetadataChange = (value: string) => {
     setMetadataStr(value);
@@ -151,9 +158,7 @@ export default function EditMemoryModal({ memory, isOpen, onClose, onSaved }: Pr
               metadataError ? "border-red-500" : "border-border"
             }`}
           />
-          {metadataError && (
-            <p className="text-xs text-red-400 mt-1">{metadataError}</p>
-          )}
+          {metadataError && <p className="text-xs text-red-400 mt-1">{metadataError}</p>}
         </div>
       </div>
     </Modal>

src/app/(dashboard)/dashboard/memory/components/QdrantConfigCard.tsx

@@ -40,7 +40,8 @@ export default function QdrantConfigCard() {
     collection?: { exists: boolean; vectorSize?: number; vectorName?: string | null };
   } | null>(null);
   const [searchValidated, setSearchValidated] = useState(false);
-  const [tutorialOpen, setTutorialOpen] = useState(false);  const [checking, setChecking] = useState(false);
+  const [tutorialOpen, setTutorialOpen] = useState(false);
+  const [checking, setChecking] = useState(false);
   const [searchQuery, setSearchQuery] = useState("");
   const [searching, setSearching] = useState(false);
   const [searchResults, setSearchResults] = useState<
@@ -109,7 +110,8 @@ export default function QdrantConfigCard() {
       // invalidate in-flight checks so they cannot overwrite the new state.
       healthSeqRef.current += 1;
       setHealth(null);
-      setSearchValidated(false);      setQdrant(next);
+      setSearchValidated(false);
+      setQdrant(next);
       setSaving(true);
       setSaveStatus("");
       try {
@@ -153,7 +155,8 @@ export default function QdrantConfigCard() {
         setSaving(false);
       }
     },
-    [qdrant, checkHealth]  );
+    [qdrant, checkHealth]
+  );
 
   // Auto-check on mount once settings load: without this the status badge
   // renders red after a page refresh because `health` starts as null and the
@@ -161,7 +164,9 @@ export default function QdrantConfigCard() {
   // connection button still drives the same check manually.
   useEffect(() => {
     if (!loading && qdrant.enabled && health === null) {
-      void checkHealth();
+      void (async () => {
+        await checkHealth();
+      })();
     }
   }, [loading, qdrant.enabled, health, checkHealth]);
 
@@ -245,7 +250,8 @@ export default function QdrantConfigCard() {
                 ? "text-text-muted"
                 : health.ok
                   ? "text-emerald-500"
-                  : "text-red-500"          }`}
+                  : "text-red-500"
+          }`}
         >
           <span
             className={`inline-block w-2.5 h-2.5 rounded-full ${

src/app/(dashboard)/dashboard/memory/components/tabs/MemoriesTab.tsx

@@ -190,9 +190,7 @@ export default function MemoriesTab() {
         else skipped++;
       }
       fetchMemories();
-      setImportStatus(
-        t("importResult", { imported, skipped }),
-      );
+      setImportStatus(t("importResult", { imported, skipped }));
     } catch {
       setImportStatus(t("importError"));
     } finally {
@@ -239,7 +237,9 @@ export default function MemoriesTab() {
   // Auto-run health check on mount + poll every 30s, so the indicator reflects
   // engine health without requiring a manual click.
   useEffect(() => {
-    void checkHealth();
+    void (async () => {
+      await checkHealth();
+    })();
     const id = setInterval(() => {
       void checkHealth();
     }, 30_000);
@@ -260,8 +260,9 @@ export default function MemoriesTab() {
         body: JSON.stringify({ dryRun: true, olderThanDays: 30 }),
       });
       const data = await res.json().catch(() => null);
-      const candidates: string[] =
-        Array.isArray(data?.candidates) ? data.candidates.map((c: { key?: string }) => c?.key ?? String(c)) : [];
+      const candidates: string[] = Array.isArray(data?.candidates)
+        ? data.candidates.map((c: { key?: string }) => c?.key ?? String(c))
+        : [];
       setSummarizeCandidates(candidates);
       setSummarizeDialogOpen(true);
     } catch {
@@ -289,8 +290,7 @@ export default function MemoriesTab() {
     }
   };
 
-  const showHitRate =
-    (stats.cacheStats?.hits ?? 0) + (stats.cacheStats?.misses ?? 0) > 0;
+  const showHitRate = (stats.cacheStats?.hits ?? 0) + (stats.cacheStats?.misses ?? 0) > 0;
 
   if (isLoading) {
     return (
@@ -401,9 +401,7 @@ export default function MemoriesTab() {
                   info
                 </span>
               </div>
-              <div className="text-2xl font-bold">
-                {((stats.hitRate ?? 0) * 100).toFixed(1)}%
-              </div>
+              <div className="text-2xl font-bold">{((stats.hitRate ?? 0) * 100).toFixed(1)}%</div>
             </div>
           </Card>
         )}
@@ -448,12 +446,8 @@ export default function MemoriesTab() {
               <span className="material-symbols-outlined text-[40px] text-text-muted mb-3">
                 psychology
               </span>
-              <p className="text-sm font-medium text-text-main mb-1">
-                {t("emptyState.title")}
-              </p>
-              <p className="text-xs text-text-muted max-w-xs">
-                {t("emptyState.description")}
-              </p>
+              <p className="text-sm font-medium text-text-main mb-1">{t("emptyState.title")}</p>
+              <p className="text-xs text-text-muted max-w-xs">{t("emptyState.description")}</p>
               <Button className="mt-4" size="sm" onClick={() => setAddDialogOpen(true)}>
                 {t("addMemory")}
               </Button>
@@ -477,7 +471,9 @@ export default function MemoriesTab() {
                         <td className="py-2 px-4">
                           <Badge
                             variant={getTypeColor(memory.type)}
-                            title={t(TYPE_TOOLTIPS[memory.type]?.replace("memory.", "") ?? memory.type)}
+                            title={t(
+                              TYPE_TOOLTIPS[memory.type]?.replace("memory.", "") ?? memory.type
+                            )}
                           >
                             {t(memory.type)}
                           </Badge>
@@ -665,7 +661,10 @@ export default function MemoriesTab() {
               </p>
               <ul className="space-y-1 max-h-48 overflow-y-auto">
                 {summarizeCandidates.map((key, i) => (
-                  <li key={i} className="text-xs font-mono text-text-main truncate px-2 py-1 bg-surface/30 rounded">
+                  <li
+                    key={i}
+                    className="text-xs font-mono text-text-main truncate px-2 py-1 bg-surface/30 rounded"
+                  >
                     {key}
                   </li>
                 ))}

src/app/(dashboard)/dashboard/memory/hooks/useEngineStatus.ts

@@ -39,7 +39,9 @@ export function useEngineStatus(refreshIntervalMs = 5000): UseEngineStatusResult
 
   useEffect(() => {
     mounted.current = true;
-    void fetchOnce();
+    void (async () => {
+      await fetchOnce();
+    })();
     if (!refreshIntervalMs || refreshIntervalMs <= 0) {
       return () => {
         mounted.current = false;

src/app/(dashboard)/dashboard/memory/hooks/useMemorySettings.ts

@@ -40,7 +40,9 @@ export function useMemorySettings(): UseMemorySettingsResult {
 
   useEffect(() => {
     mounted.current = true;
-    void fetchOnce();
+    void (async () => {
+      await fetchOnce();
+    })();
     return () => {
       mounted.current = false;
     };

src/app/(dashboard)/dashboard/orchestration/hooks/useOrchestrationSnapshot.ts

@@ -0,0 +1,157 @@
+"use client";
+/** Polls the 3 agent sources (allSettled), listens to the `requests` WS channel as a refetch trigger. */
+import { useCallback, useEffect, useMemo, useRef, useState } from "react";
+import { useLiveDashboard } from "@/hooks/useLiveDashboard";
+import type { CloudAgentTask } from "@/lib/cloudAgent/types";
+import type { A2ATask } from "@/lib/a2a/taskManager";
+import type { FleetSnapshot } from "@/lib/conductor/hubProxy";
+import { fromCloudAgent } from "../model/fromCloudAgent";
+import { fromA2A } from "../model/fromA2A";
+import { fromConductor } from "../model/fromConductor";
+import { mergeSnapshot } from "../model/mergeSnapshot";
+import type { OrchSnapshot, SourceStatus } from "../model/orchestrationTypes";
+
+export const POLL_MS = 5_000;
+export const WS_REFETCH_DEBOUNCE_MS = 1_000;
+
+interface Raw {
+  cloudAgent: CloudAgentTask[];
+  a2a: A2ATask[];
+  conductor: FleetSnapshot;
+}
+const EMPTY_RAW: Raw = {
+  cloudAgent: [],
+  a2a: [],
+  conductor: { offline: true, runners: [], tasks: [] },
+};
+
+async function fetchJson<T>(url: string, signal: AbortSignal): Promise<T> {
+  const res = await fetch(url, { signal, cache: "no-store" });
+  if (!res.ok) throw new Error(`HTTP ${res.status}`);
+  return res.json() as Promise<T>;
+}
+
+/** Builds the 3-source status list from a `Promise.allSettled` triple. */
+function buildSourceStatuses(
+  ca: PromiseSettledResult<{ data: CloudAgentTask[] }>,
+  a2a: PromiseSettledResult<{ tasks: A2ATask[] }>,
+  cond: PromiseSettledResult<FleetSnapshot>,
+  nowIso: string
+): SourceStatus[] {
+  const next: SourceStatus[] = [];
+  if (ca.status === "fulfilled") next.push({ source: "cloud-agent", ok: true });
+  else
+    next.push({
+      source: "cloud-agent",
+      ok: false,
+      error: String(ca.reason),
+      staleSince: nowIso,
+    });
+  if (a2a.status === "fulfilled") next.push({ source: "a2a", ok: true });
+  else next.push({ source: "a2a", ok: false, error: String(a2a.reason), staleSince: nowIso });
+  if (cond.status === "fulfilled") {
+    next.push({ source: "conductor", ok: true, offline: cond.value.offline });
+  } else
+    next.push({
+      source: "conductor",
+      ok: false,
+      error: String(cond.reason),
+      staleSince: nowIso,
+    });
+  return next;
+}
+
+export function useOrchestrationSnapshot() {
+  // `raw` and `polledAt` are React state (not refs) so the merge below reads them
+  // during render like any other state — a ref read during render trips the
+  // `react-hooks/refs` lint rule, and computing `Date.now()` inline in the memo
+  // factory trips `react-hooks/purity`. Sampling `Date.now()` once per poll (inside
+  // the effect, not during render) keeps `mergeSnapshot`'s staleness math correct
+  // without either violation.
+  const [raw, setRaw] = useState<Raw>(EMPTY_RAW);
+  const [statuses, setStatuses] = useState<SourceStatus[]>([]);
+  // Lazy initializer (not a literal 0) so the pre-first-poll render already has a
+  // real timestamp — with `0` the very first `mergeSnapshot` call stamped
+  // `generatedAt` as the 1970 epoch. Safe: with EMPTY_RAW there is nothing to
+  // staleness-filter at mount, so seeding `Date.now()` here changes no behavior.
+  const [polledAt, setPolledAt] = useState<number>(() => Date.now());
+  const [isLoading, setIsLoading] = useState(true);
+  const [showCompleted, setShowCompleted] = useState(false);
+  const debounceRef = useRef<ReturnType<typeof setTimeout> | null>(null);
+  // Populated by the mount effect below; lets `refetch()` (and the WS debounce
+  // trigger) reach the same poll loop without hoisting it out of the effect —
+  // hoisting to a top-level `useCallback` invoked from the effect body trips
+  // `react-hooks/set-state-in-effect`.
+  const pollRef = useRef<() => void>(() => {});
+
+  useEffect(() => {
+    const controller = new AbortController();
+
+    const poll = async () => {
+      const [ca, a2a, cond] = await Promise.allSettled([
+        fetchJson<{ data: CloudAgentTask[] }>("/api/v1/agents/tasks?limit=100", controller.signal),
+        fetchJson<{ tasks: A2ATask[] }>("/api/a2a/tasks?limit=200", controller.signal),
+        fetchJson<FleetSnapshot>("/api/conductor/fleet", controller.signal),
+      ]);
+      if (controller.signal.aborted) return;
+      const nowMs = Date.now();
+      const nowIso = new Date(nowMs).toISOString();
+      const next = buildSourceStatuses(ca, a2a, cond, nowIso);
+
+      // Failed sources keep the previously stored slice — only overwrite what
+      // actually resolved this round ("last good data" contract from the brief).
+      setRaw((prev) => ({
+        cloudAgent: ca.status === "fulfilled" ? ca.value.data : prev.cloudAgent,
+        a2a: a2a.status === "fulfilled" ? a2a.value.tasks : prev.a2a,
+        conductor: cond.status === "fulfilled" ? cond.value : prev.conductor,
+      }));
+      setStatuses(next);
+      setPolledAt(nowMs);
+      setIsLoading(false);
+    };
+
+    pollRef.current = () => void poll();
+    void poll();
+    const id = setInterval(() => void poll(), POLL_MS);
+    return () => {
+      clearInterval(id);
+      controller.abort();
+      if (debounceRef.current) {
+        clearTimeout(debounceRef.current);
+        debounceRef.current = null;
+      }
+    };
+  }, []);
+
+  const refetch = useCallback(() => {
+    pollRef.current();
+  }, []);
+
+  useLiveDashboard({
+    channels: ["requests"],
+    onEvent: (payload) => {
+      if (payload.channel !== "requests") return;
+      if (debounceRef.current) return; // debounce burst → one refetch
+      debounceRef.current = setTimeout(() => {
+        debounceRef.current = null;
+        refetch();
+      }, WS_REFETCH_DEBOUNCE_MS);
+    },
+  });
+
+  const snapshot: OrchSnapshot = useMemo(
+    () =>
+      mergeSnapshot(
+        {
+          cloudAgent: fromCloudAgent(raw.cloudAgent),
+          a2a: fromA2A(raw.a2a),
+          conductor: fromConductor(raw.conductor),
+        },
+        statuses,
+        { now: polledAt, showCompleted }
+      ),
+    [raw, statuses, showCompleted, polledAt]
+  );
+
+  return { snapshot, isLoading, showCompleted, setShowCompleted, refetch };
+}

src/app/(dashboard)/dashboard/orchestration/model/fromA2A.ts

@@ -0,0 +1,53 @@
+/** A2A tasks → unified orchestration nodes. Pure. */
+import type { A2ATask } from "@/lib/a2a/taskManager";
+import type { OrchEdge, OrchNode, OrchState } from "./orchestrationTypes";
+
+const STATE_MAP: Record<string, OrchState> = {
+  submitted: "queued",
+  working: "running",
+  completed: "succeeded",
+  failed: "failed",
+  cancelled: "cancelled",
+};
+const TERMINAL: ReadonlySet<OrchState> = new Set(["succeeded", "failed", "cancelled"]);
+
+function truncate(s: string, n = 60): string {
+  return s.length > n ? `${s.slice(0, n - 1)}…` : s;
+}
+
+export function fromA2A(tasks: A2ATask[]): { nodes: OrchNode[]; edges: OrchEdge[] } {
+  if (tasks.length === 0) return { nodes: [], edges: [] };
+  const nodes: OrchNode[] = [];
+  const edges: OrchEdge[] = [];
+  const counts: Partial<Record<OrchState, number>> = {};
+
+  for (const t of tasks) {
+    const mapped = STATE_MAP[t.state];
+    const state: OrchState = mapped ?? "failed";
+    counts[state] = (counts[state] ?? 0) + 1;
+    const id = `a2a:${t.id}`;
+    const firstUser = t.input.messages.find((m) => m.role === "user")?.content ?? "";
+    nodes.push({
+      id,
+      kind: "work",
+      source: "a2a",
+      state,
+      label: t.skill,
+      sublabel: mapped ? truncate(firstUser) : `unknown state: ${String(t.state)}`,
+      startedAt: t.createdAt,
+      updatedAt: t.updatedAt,
+      endedAt: TERMINAL.has(state) ? t.updatedAt : undefined,
+      raw: t,
+    });
+    edges.push({
+      id: `e:source:a2a→${id}`,
+      from: "source:a2a",
+      to: id,
+      kind: "owns",
+      active: state === "running",
+    });
+  }
+
+  nodes.unshift({ id: "source:a2a", kind: "source", source: "a2a", label: "A2A", counts });
+  return { nodes, edges };
+}

src/app/(dashboard)/dashboard/orchestration/model/fromCloudAgent.ts

@@ -0,0 +1,75 @@
+/** Cloud Agent tasks → unified orchestration nodes. Pure. */
+import type { CloudAgentTask } from "@/lib/cloudAgent/types";
+import type { OrchEdge, OrchNode, OrchState } from "./orchestrationTypes";
+
+const STATE_MAP: Record<string, OrchState> = {
+  queued: "queued",
+  running: "running",
+  awaiting_approval: "waiting_approval",
+  completed: "succeeded",
+  failed: "failed",
+  cancelled: "cancelled",
+};
+
+function truncate(s: string, n = 60): string {
+  return s.length > n ? `${s.slice(0, n - 1)}…` : s;
+}
+
+export function fromCloudAgent(tasks: CloudAgentTask[]): { nodes: OrchNode[]; edges: OrchEdge[] } {
+  if (tasks.length === 0) return { nodes: [], edges: [] };
+  const nodes: OrchNode[] = [];
+  const edges: OrchEdge[] = [];
+  const counts: Partial<Record<OrchState, number>> = {};
+
+  for (const t of tasks) {
+    const mapped = STATE_MAP[t.status];
+    const state: OrchState = mapped ?? "failed";
+    counts[state] = (counts[state] ?? 0) + 1;
+    const id = `cloud-agent:${t.id}`;
+    const active = state === "running";
+    nodes.push({
+      id,
+      kind: "work",
+      source: "cloud-agent",
+      state,
+      label: truncate(t.prompt),
+      sublabel: mapped ? t.providerId : `${t.providerId} — unknown status: ${String(t.status)}`,
+      startedAt: t.createdAt,
+      updatedAt: t.updatedAt,
+      endedAt: t.completedAt,
+      cost: t.result?.cost,
+      raw: t,
+    });
+    edges.push({
+      id: `e:source:cloud-agent→${id}`,
+      from: "source:cloud-agent",
+      to: id,
+      kind: "owns",
+      active,
+    });
+
+    const last = t.activities[t.activities.length - 1];
+    if (active && last) {
+      const actId = `${id}:activity`;
+      nodes.push({
+        id: actId,
+        kind: "activity",
+        source: "cloud-agent",
+        state,
+        label: truncate(last.content),
+        sublabel: last.type,
+        updatedAt: last.timestamp,
+      });
+      edges.push({ id: `e:${id}→${actId}`, from: id, to: actId, kind: "owns", active: true });
+    }
+  }
+
+  nodes.unshift({
+    id: "source:cloud-agent",
+    kind: "source",
+    source: "cloud-agent",
+    label: "Cloud Agent",
+    counts,
+  });
+  return { nodes, edges };
+}

src/app/(dashboard)/dashboard/orchestration/model/fromConductor.ts

@@ -0,0 +1,148 @@
+/** Conductor fleet snapshot → unified orchestration nodes. Pure. */
+import type { FleetRunner, FleetSnapshot, FleetTask } from "@/lib/conductor/hubProxy";
+import type { OrchEdge, OrchNode, OrchState } from "./orchestrationTypes";
+
+const TERMINAL: ReadonlySet<OrchState> = new Set(["succeeded", "failed", "cancelled"]);
+
+function mapHubStatus(status: string): OrchState | null {
+  const s = status.toLowerCase();
+  if (s === "queued" || s === "pending") return "queued";
+  if (s === "running" || s === "working" || s === "scheduled") return "running";
+  if (s === "done" || s === "completed" || s === "succeeded") return "succeeded";
+  if (s === "failed" || s === "error") return "failed";
+  if (s === "cancelled" || s === "canceled") return "cancelled";
+  return null;
+}
+
+function taskNode(t: FleetTask, kind: "work" | "activity"): OrchNode {
+  const mapped = mapHubStatus(t.status);
+  const state: OrchState = mapped ?? "failed";
+  return {
+    id: `conductor:task:${t.id}`,
+    kind,
+    source: "conductor",
+    state,
+    label: t.summary ?? t.id,
+    sublabel: mapped ? (t.repo ?? t.mode) : `unknown status: ${t.status}`,
+    updatedAt: t.updated_at ?? undefined,
+    // FleetTask has no dedicated completion timestamp — updated_at is the closest
+    // proxy, same pattern as fromA2A.ts (A2ATask has no completedAt either).
+    endedAt: TERMINAL.has(state) ? (t.updated_at ?? undefined) : undefined,
+    raw: t,
+  };
+}
+
+/**
+ * Tasks whose runner actually exists in `snap.runners` AND is currently "running" — those
+ * get "absorbed" into that runner's ActivityNode instead of getting their own work node.
+ * A running task pointing at a runner id that has since deregistered falls through to the
+ * normal work-node loop instead of being silently skipped as "already an activity".
+ */
+function computeActiveByRunner(snap: FleetSnapshot): Map<string, FleetTask> {
+  const runnerIds = new Set(snap.runners.map((r) => r.id));
+  const activeByRunner = new Map<string, FleetTask>();
+  for (const t of snap.tasks) {
+    if (t.runner && runnerIds.has(t.runner) && mapHubStatus(t.status) === "running") {
+      activeByRunner.set(t.runner, t);
+    }
+  }
+  return activeByRunner;
+}
+
+function runnerState(r: FleetRunner, activeTask: FleetTask | undefined): OrchState {
+  if (!r.online) return "failed";
+  if (r.draining) return "cancelled";
+  return activeTask ? "running" : "queued";
+}
+
+/** One work node per runner, plus an activity node for its currently-active task. */
+function runnerWorkNodes(
+  snap: FleetSnapshot,
+  activeByRunner: Map<string, FleetTask>,
+  bump: (s: OrchState) => void
+): { nodes: OrchNode[]; edges: OrchEdge[] } {
+  const nodes: OrchNode[] = [];
+  const edges: OrchEdge[] = [];
+  for (const r of snap.runners) {
+    const id = `conductor:runner:${r.id}`;
+    const activeTask = activeByRunner.get(r.id);
+    const state = runnerState(r, activeTask);
+    bump(state);
+    nodes.push({
+      id,
+      kind: "work",
+      source: "conductor",
+      state,
+      label: r.name,
+      sublabel: r.clis.join(", "),
+      raw: r,
+    });
+    edges.push({
+      id: `e:source:conductor→${id}`,
+      from: "source:conductor",
+      to: id,
+      kind: "owns",
+      active: state === "running",
+    });
+    if (activeTask) {
+      nodes.push(taskNode(activeTask, "activity"));
+      edges.push({
+        id: `e:${id}→conductor:task:${activeTask.id}`,
+        from: id,
+        to: `conductor:task:${activeTask.id}`,
+        kind: "owns",
+        active: true,
+      });
+    }
+  }
+  return { nodes, edges };
+}
+
+/** Work nodes for tasks not already absorbed as a runner's activity node. */
+function remainingTaskWorkNodes(
+  snap: FleetSnapshot,
+  activeByRunner: Map<string, FleetTask>,
+  bump: (s: OrchState) => void
+): { nodes: OrchNode[]; edges: OrchEdge[] } {
+  const nodes: OrchNode[] = [];
+  const edges: OrchEdge[] = [];
+  for (const t of snap.tasks) {
+    if (t.runner && activeByRunner.get(t.runner)?.id === t.id) continue; // already an activity
+    const node = taskNode(t, "work");
+    bump(node.state as OrchState);
+    nodes.push(node);
+    edges.push({
+      id: `e:source:conductor→${node.id}`,
+      from: "source:conductor",
+      to: node.id,
+      kind: "owns",
+      active: node.state === "running",
+    });
+  }
+  return { nodes, edges };
+}
+
+export function fromConductor(snap: FleetSnapshot): { nodes: OrchNode[]; edges: OrchEdge[] } {
+  if (snap.offline || (snap.runners.length === 0 && snap.tasks.length === 0)) {
+    return { nodes: [], edges: [] };
+  }
+  const counts: Partial<Record<OrchState, number>> = {};
+  const bump = (s: OrchState) => {
+    counts[s] = (counts[s] ?? 0) + 1;
+  };
+
+  const activeByRunner = computeActiveByRunner(snap);
+  const runners = runnerWorkNodes(snap, activeByRunner, bump);
+  const tasks = remainingTaskWorkNodes(snap, activeByRunner, bump);
+  const nodes = [...runners.nodes, ...tasks.nodes];
+  const edges = [...runners.edges, ...tasks.edges];
+
+  nodes.unshift({
+    id: "source:conductor",
+    kind: "source",
+    source: "conductor",
+    label: "Conductor",
+    counts,
+  });
+  return { nodes, edges };
+}

src/app/(dashboard)/dashboard/orchestration/model/mergeSnapshot.ts

@@ -0,0 +1,217 @@
+/** Merge the three source mappers into one snapshot: root, dedupe, staleness filter, cap. Pure. */
+import {
+  MAX_WORK_NODES,
+  STALE_COMPLETED_MS,
+  type OrchEdge,
+  type OrchNode,
+  type OrchSnapshot,
+  type OrchSource,
+  type OrchState,
+  type SourceStatus,
+} from "./orchestrationTypes";
+
+const TERMINAL: ReadonlySet<OrchState> = new Set(["succeeded", "failed", "cancelled"]);
+
+export interface MergeOptions {
+  now: number;
+  showCompleted?: boolean;
+}
+
+interface Part {
+  nodes: OrchNode[];
+  edges: OrchEdge[];
+}
+
+interface NodesAndEdges {
+  nodes: OrchNode[];
+  edges: OrchEdge[];
+}
+
+function conductorMirrorId(node: OrchNode): string | null {
+  const raw = node.raw as { metadata?: { conductor?: { task_id?: unknown } } } | undefined;
+  const id = raw?.metadata?.conductor?.task_id;
+  return typeof id === "string" ? id : null;
+}
+
+/**
+ * (2) Conductor↔A2A dedupe — key verified in src/lib/conductor/bridge.ts::ensureMirrored.
+ * Mutates `dropped` in place; returns the (possibly patched) nodes/edges.
+ */
+function dedupeConductorMirrors(
+  nodes: OrchNode[],
+  edges: OrchEdge[],
+  dropped: Set<string>
+): NodesAndEdges {
+  const conductorTaskIds = new Set(
+    nodes
+      .filter((n) => n.source === "conductor" && n.id.startsWith("conductor:task:"))
+      .map((n) => n.id.slice("conductor:task:".length))
+  );
+  const nextNodes = [...nodes];
+  const nextEdges = [...edges];
+  for (const n of nextNodes) {
+    if (n.source !== "a2a" || n.kind !== "work") continue;
+    const mirror = conductorMirrorId(n);
+    if (mirror && conductorTaskIds.has(mirror)) {
+      dropped.add(n.id);
+      const cIndex = nextNodes.findIndex((c) => c.id === `conductor:task:${mirror}`);
+      if (cIndex !== -1) {
+        // Copy rather than mutate — the original object is still referenced by
+        // parts.conductor.nodes, and this function's contract is Pure.
+        const cNode: OrchNode = { ...nextNodes[cIndex], mirrorOf: n.id };
+        nextNodes[cIndex] = cNode;
+        nextEdges.push({
+          id: `e:mirror:${cNode.id}`,
+          from: cNode.id,
+          to: "source:a2a",
+          kind: "mirror",
+          active: false,
+        });
+      }
+    }
+  }
+  return { nodes: nextNodes, edges: nextEdges };
+}
+
+/** (3) staleness filter — adds stale terminal work/activity node ids to `dropped`. */
+function markStaleCompleted(nodes: OrchNode[], now: number, dropped: Set<string>): void {
+  for (const n of nodes) {
+    if (n.kind !== "work" && n.kind !== "activity") continue;
+    if (
+      n.state &&
+      TERMINAL.has(n.state) &&
+      n.endedAt &&
+      now - Date.parse(n.endedAt) > STALE_COMPLETED_MS
+    ) {
+      dropped.add(n.id);
+    }
+  }
+}
+
+/** One source's overflow placeholder node, or null when it fits under `budgetPer`. */
+function overflowNodeForSource(
+  source: OrchSource,
+  list: OrchNode[],
+  budgetPer: number,
+  dropped: Set<string>
+): OrchNode | null {
+  if (list.length <= budgetPer) return null;
+  list.sort((a, b) => Date.parse(b.updatedAt ?? "0") - Date.parse(a.updatedAt ?? "0"));
+  const excess = list.slice(budgetPer);
+  const counts: Partial<Record<OrchState, number>> = {};
+  for (const n of excess) {
+    dropped.add(n.id);
+    if (n.state) counts[n.state] = (counts[n.state] ?? 0) + 1;
+  }
+  return {
+    id: `overflow:${source}`,
+    kind: "overflow",
+    source,
+    label: `+${excess.length} more`,
+    counts,
+    // Additive: lets overviewProjection fold true per-state totals into its
+    // counters even though these nodes no longer render on the canvas
+    // (operator ruling — spec governs, counters must show TRUE totals).
+    droppedByState: counts,
+  };
+}
+
+/** (4) cap with per-source overflow, newest kept. */
+function capWorkNodesWithOverflow(
+  nodes: OrchNode[],
+  edges: OrchEdge[],
+  dropped: Set<string>
+): NodesAndEdges {
+  const works = nodes.filter((n) => n.kind === "work");
+  if (works.length <= MAX_WORK_NODES) return { nodes, edges };
+
+  const bySource = new Map<OrchSource, OrchNode[]>();
+  for (const w of works) {
+    const list = bySource.get(w.source as OrchSource) ?? [];
+    list.push(w);
+    bySource.set(w.source as OrchSource, list);
+  }
+  const budgetPer = Math.max(1, Math.floor(MAX_WORK_NODES / bySource.size));
+  const overflowNodes: OrchNode[] = [];
+  for (const [source, list] of bySource) {
+    const overflow = overflowNodeForSource(source, list, budgetPer, dropped);
+    if (overflow) overflowNodes.push(overflow);
+  }
+
+  const nextNodes = nodes.filter((n) => !dropped.has(n.id)).concat(overflowNodes);
+  const nextEdges = edges.filter((e) => !dropped.has(e.from) && !dropped.has(e.to));
+  for (const o of overflowNodes) {
+    nextEdges.push({
+      id: `e:source:${o.source}→${o.id}`,
+      from: `source:${o.source}`,
+      to: o.id,
+      kind: "owns",
+      active: false,
+    });
+  }
+  return { nodes: nextNodes, edges: nextEdges };
+}
+
+/**
+ * (1) root — link every present SourceNode, plus failed sources so the UI can show them
+ * stale. Returns nodes with the root prepended.
+ */
+function buildRootAndSourceEdges(
+  nodes: OrchNode[],
+  edges: OrchEdge[],
+  sources: SourceStatus[]
+): NodesAndEdges {
+  const root: OrchNode = { id: "orchestrator", kind: "orchestrator", label: "OmniRoute" };
+  const nextNodes = [...nodes];
+  const nextEdges = [...edges];
+  const sourceIds = new Set(nextNodes.filter((n) => n.kind === "source").map((n) => n.id));
+  for (const s of sources) {
+    // `!s.ok` covers hard failures; `s.offline` also materializes a placeholder
+    // for a source that reported ok:true but offline:true (e.g. Conductor with
+    // no hub configured) — otherwise that source never gets a SourceNode at all
+    // and its "offline" sublabel can never render.
+    if ((!s.ok || s.offline) && !sourceIds.has(`source:${s.source}`) && s.source !== "routing") {
+      nextNodes.push({
+        id: `source:${s.source}`,
+        kind: "source",
+        source: s.source,
+        label: s.source,
+        sublabel: s.offline ? "offline" : "error",
+      });
+      sourceIds.add(`source:${s.source}`);
+    }
+  }
+  for (const id of sourceIds) {
+    nextEdges.push({
+      id: `e:orchestrator→${id}`,
+      from: "orchestrator",
+      to: id,
+      kind: "owns",
+      active: false,
+    });
+  }
+  return { nodes: [root, ...nextNodes], edges: nextEdges };
+}
+
+export function mergeSnapshot(
+  parts: { cloudAgent: Part; a2a: Part; conductor: Part },
+  sources: SourceStatus[],
+  opts: MergeOptions
+): OrchSnapshot {
+  let nodes: OrchNode[] = [...parts.cloudAgent.nodes, ...parts.a2a.nodes, ...parts.conductor.nodes];
+  let edges: OrchEdge[] = [...parts.cloudAgent.edges, ...parts.a2a.edges, ...parts.conductor.edges];
+
+  const dropped = new Set<string>();
+  ({ nodes, edges } = dedupeConductorMirrors(nodes, edges, dropped));
+
+  if (!opts.showCompleted) {
+    markStaleCompleted(nodes, opts.now, dropped);
+  }
+  nodes = nodes.filter((n) => !dropped.has(n.id));
+  edges = edges.filter((e) => !dropped.has(e.from) && !dropped.has(e.to));
+
+  ({ nodes, edges } = capWorkNodesWithOverflow(nodes, edges, dropped));
+  ({ nodes, edges } = buildRootAndSourceEdges(nodes, edges, sources));
+
+  return { nodes, edges, sources, generatedAt: new Date(opts.now).toISOString() };
+}

src/app/(dashboard)/dashboard/orchestration/model/orchestrationToFlow.ts

@@ -0,0 +1,58 @@
+/** OrchSnapshot → @xyflow nodes/edges with a deterministic shallow 3-layer layout. Pure. */
+import type { Edge, Node } from "@xyflow/react";
+import { edgeStyle } from "@/shared/components/flow/edgeStyles";
+import type { OrchNodeKind, OrchSnapshot } from "./orchestrationTypes";
+
+const LAYER_Y: Record<OrchNodeKind, number> = {
+  orchestrator: 0,
+  source: 150,
+  work: 320,
+  overflow: 320,
+  activity: 470,
+};
+const X_GAP = 260;
+
+export function orchestrationToFlow(snap: OrchSnapshot): {
+  nodes: Node[];
+  edges: Edge[];
+  fitKey: string;
+} {
+  const byLayer = new Map<number, string[]>();
+  for (const n of [...snap.nodes].sort((a, b) => a.id.localeCompare(b.id))) {
+    const y = LAYER_Y[n.kind];
+    const ids = byLayer.get(y) ?? [];
+    ids.push(n.id);
+    byLayer.set(y, ids);
+  }
+  const pos = new Map<string, { x: number; y: number }>();
+  for (const [y, ids] of byLayer) {
+    const width = (ids.length - 1) * X_GAP;
+    ids.forEach((id, i) => pos.set(id, { x: i * X_GAP - width / 2, y }));
+  }
+
+  const stateOf = new Map(snap.nodes.map((n) => [n.id, n.state]));
+  const nodes: Node[] = snap.nodes.map((n) => ({
+    id: n.id,
+    type: n.kind,
+    position: pos.get(n.id)!,
+    data: n as unknown as Record<string, unknown>,
+  }));
+  const edges: Edge[] = snap.edges.map((e) => {
+    const target = stateOf.get(e.to);
+    const style = edgeStyle(e.active, false, target === "failed", target === "succeeded");
+    return {
+      id: e.id,
+      source: e.from,
+      target: e.to,
+      animated: e.active,
+      style: e.kind === "mirror" ? { ...style, strokeDasharray: "6 4" } : style,
+    };
+  });
+
+  const fitKey = snap.nodes
+    .filter((n) => n.kind === "work")
+    .map((n) => n.id)
+    .sort()
+    .join("|");
+  return { nodes, edges, fitKey };
+}

src/app/(dashboard)/dashboard/orchestration/model/orchestrationTypes.ts

@@ -0,0 +1,79 @@
+/**
+ * Pure domain vocabulary for the Orchestration Canvas — no React, no side effects.
+ * Spec: _tasks/superpowers/specs/2026-08-30-orchestration-canvas-design.md
+ */
+import { STATUS_HEX } from "@/shared/constants/statusColors";
+
+export type OrchState =
+  "queued" | "running" | "waiting_approval" | "succeeded" | "failed" | "cancelled";
+export type OrchSource = "cloud-agent" | "a2a" | "conductor" | "routing";
+export type OrchNodeKind = "orchestrator" | "source" | "work" | "activity" | "overflow";
+
+export interface OrchNode {
+  id: string; // `${source}:${sourceId}` for work nodes
+  kind: OrchNodeKind;
+  source?: OrchSource;
+  state?: OrchState;
+  label: string;
+  sublabel?: string;
+  startedAt?: string;
+  updatedAt?: string;
+  endedAt?: string;
+  cost?: number;
+  counts?: Partial<Record<OrchState, number>>;
+  // Overflow nodes only: per-state counts of the work nodes folded into this
+  // overflow node when the MAX_WORK_NODES cap engages. overviewProjection folds
+  // this into its `counts` totals (never into `columns`) so operators still see
+  // TRUE totals even when the canvas caps the rendered node count.
+  droppedByState?: Partial<Record<OrchState, number>>;
+  mirrorOf?: string;
+  raw?: unknown;
+}
+
+export interface OrchEdge {
+  id: string;
+  from: string;
+  to: string;
+  kind: "owns" | "mirror";
+  active: boolean; // true while the target work is `running`
+}
+
+export interface SourceStatus {
+  source: OrchSource;
+  ok: boolean;
+  offline?: boolean;
+  error?: string;
+  staleSince?: string;
+}
+
+export interface OrchSnapshot {
+  nodes: OrchNode[];
+  edges: OrchEdge[];
+  sources: SourceStatus[];
+  generatedAt: string;
+}
+
+export const ORCH_STATES = [
+  "queued",
+  "running",
+  "waiting_approval",
+  "succeeded",
+  "failed",
+  "cancelled",
+] as const satisfies readonly OrchState[];
+
+const STATE_HEX: Record<OrchState, string> = {
+  queued: STATUS_HEX.muted,
+  running: STATUS_HEX.warning,
+  waiting_approval: STATUS_HEX.approval,
+  succeeded: STATUS_HEX.success,
+  failed: STATUS_HEX.error,
+  cancelled: STATUS_HEX.muted,
+};
+
+export function orchStateColor(state: OrchState): string {
+  return STATE_HEX[state];
+}
+
+export const STALE_COMPLETED_MS = 600_000; // completed >10 min ago drop out of the live view
+export const MAX_WORK_NODES = 40; // beyond this, per-source overflow nodes take over

src/app/(dashboard)/dashboard/orchestration/model/overviewProjection.ts

@@ -0,0 +1,50 @@
+/** OrchSnapshot → overview counters + kanban columns. Pure. */
+import {
+  ORCH_STATES,
+  type OrchNode,
+  type OrchSnapshot,
+  type OrchState,
+} from "./orchestrationTypes";
+
+export interface OverviewData {
+  counts: Record<OrchState, number>;
+  columns: {
+    queued: OrchNode[];
+    running: OrchNode[];
+    waiting_approval: OrchNode[];
+    done: OrchNode[];
+  };
+}
+
+export function overviewProjection(snap: OrchSnapshot, comboActive: number): OverviewData {
+  const counts = Object.fromEntries(ORCH_STATES.map((s) => [s, 0])) as Record<OrchState, number>;
+  const columns: OverviewData["columns"] = {
+    queued: [],
+    running: [],
+    waiting_approval: [],
+    done: [],
+  };
+
+  for (const n of snap.nodes) {
+    // Overflow nodes (MAX_WORK_NODES cap) fold their dropped work nodes' true
+    // per-state counts into `counts` only — never into `columns`, since those
+    // nodes are not rendered on the canvas. Counters must show TRUE totals
+    // even when the canvas caps the rendered node count (operator ruling).
+    if (n.kind === "overflow" && n.droppedByState) {
+      for (const s of ORCH_STATES) {
+        counts[s] += n.droppedByState[s] ?? 0;
+      }
+      continue;
+    }
+    if (n.kind !== "work" || !n.state) continue;
+    counts[n.state] += 1;
+    if (n.state === "queued" || n.state === "running" || n.state === "waiting_approval") {
+      columns[n.state].push(n);
+    } else {
+      columns.done.push(n);
+    }
+  }
+  columns.done.sort((a, b) => Date.parse(b.updatedAt ?? "0") - Date.parse(a.updatedAt ?? "0"));
+  counts.running += comboActive;
+  return { counts, columns };
+}

src/app/(dashboard)/dashboard/radar/RadarCatalogTable.tsx

@@ -95,7 +95,9 @@ export function RadarCatalogTable({ entries, refreshCatalog, onError }: RadarCat
   }, [onError, t]);
 
   useEffect(() => {
-    void loadState();
+    void (async () => {
+      await loadState();
+    })();
   }, [loadState]);
 
   const stateByKey = useMemo(

src/app/(dashboard)/dashboard/radar/intel/page.tsx

@@ -67,9 +67,15 @@ export default function RadarIntelPage() {
   }, [load, t]);
 
   useEffect(() => {
-    load()
-      .catch(() => setError(t("loadFailed")))
-      .finally(() => setLoading(false));
+    void (async () => {
+      try {
+        await load();
+      } catch {
+        setError(t("loadFailed"));
+      } finally {
+        setLoading(false);
+      }
+    })();
   }, [load, t]);
 
   if (flagOff) notFound();

src/app/(dashboard)/dashboard/radar/page.tsx

@@ -201,7 +201,9 @@ export default function RadarPage() {
   }, [fetchCatalog, fetchReferrals]);
 
   useEffect(() => {
-    fetchSettings();
+    void (async () => {
+      await fetchSettings();
+    })();
   }, [fetchSettings]);
 
   // Sync (defined before handleActivate which depends on it)
@@ -242,7 +244,9 @@ export default function RadarPage() {
     if (loading || syncing || optIn !== true || autoSyncFiredRef.current) return;
     if (!shouldAutoSyncOnOpen(meta?.fetchedAt ?? null, Date.now())) return;
     autoSyncFiredRef.current = true;
-    void handleSync();
+    void (async () => {
+      await handleSync();
+    })();
   }, [loading, syncing, optIn, meta, handleSync]);
 
   // Activate opt-in

src/app/(dashboard)/dashboard/radar/setup/page.tsx

@@ -55,7 +55,16 @@ export default function RadarSetupPage() {
   const provider = searchParams.get("provider");
 
   const [setupData, setSetupData] = useState<ProviderSetupData | null>(null);
-  const [loading, setLoading] = useState(true);
+  const [loading, setLoading] = useState(provider !== null);
+
+  // Adjust-during-render when the provider query param changes (React docs
+  // pattern): a null provider has nothing to load, any other transition
+  // restarts the loading state before the fetch effect fires.
+  const [prevProvider, setPrevProvider] = useState(provider);
+  if (provider !== prevProvider) {
+    setPrevProvider(provider);
+    setLoading(provider !== null);
+  }
   const [error, setError] = useState("");
   const [testing, setTesting] = useState(false);
   const [testResult, setTestResult] = useState<{ ok: boolean; message: string } | null>(null);
@@ -63,7 +72,6 @@ export default function RadarSetupPage() {
   // Fetch catalog to find the provider's setup data
   useEffect(() => {
     if (!provider) {
-      setLoading(false);
       return;
     }
 
@@ -123,12 +131,13 @@ export default function RadarSetupPage() {
   }, [provider, t]);
 
   // Test connection — uses the EXISTING connection-test endpoint
+  const connectionId = setupData?.connectionId ?? null;
   const handleTestConnection = useCallback(async () => {
-    if (!setupData?.connectionId) return;
+    if (!connectionId) return;
     setTesting(true);
     setTestResult(null);
     try {
-      const res = await fetch(`/api/providers/${encodeURIComponent(setupData.connectionId)}/test`, {
+      const res = await fetch(`/api/providers/${encodeURIComponent(connectionId)}/test`, {
         method: "POST",
         headers: { "Content-Type": "application/json" },
         body: JSON.stringify({}),
@@ -147,7 +156,7 @@ export default function RadarSetupPage() {
     } finally {
       setTesting(false);
     }
-  }, [setupData?.connectionId, t]);
+  }, [connectionId, t]);
 
   if (!provider) {
     return (

src/app/(dashboard)/dashboard/usage/components/EvalsTab.tsx

@@ -490,18 +490,18 @@ export default function EvalsTab() {
     // eslint-disable-next-line react-hooks/exhaustive-deps
   }, []);
 
-  useEffect(() => {
-    if (targetOptions.length === 0) return;
-    if (targetOptions.some((option) => option.key === selectedTargetKey)) return;
+  // Adjust-during-render (React docs pattern): keep the selected target inside
+  // the current option set, and never compare a target against itself. Both
+  // guards self-extinguish after their setState, so the re-render settles.
+  if (
+    targetOptions.length > 0 &&
+    !targetOptions.some((option) => option.key === selectedTargetKey)
+  ) {
     setSelectedTargetKey(targetOptions[0]?.key || "suite-default:__default__");
-  }, [selectedTargetKey, targetOptions]);
-
-  useEffect(() => {
-    if (!compareTargetKey) return;
-    if (compareTargetKey === selectedTargetKey) {
-      setCompareTargetKey("");
-    }
-  }, [compareTargetKey, selectedTargetKey]);
+  }
+  if (compareTargetKey && compareTargetKey === selectedTargetKey) {
+    setCompareTargetKey("");
+  }
 
   const filteredSuites = !search.trim()
     ? suites
@@ -1846,7 +1846,11 @@ export default function EvalsTab() {
   );
 }
 
-const HeroSection = memo(function HeroSection({ t }: { t: (key: string, values?: Record<string, unknown>) => string }) {
+const HeroSection = memo(function HeroSection({
+  t,
+}: {
+  t: (key: string, values?: Record<string, unknown>) => string;
+}) {
   return (
     <Card className="p-0 overflow-hidden">
       <div

src/app/(dashboard)/dashboard/usage/components/ProviderLimits/useCodexResetCreditRedemption.ts

@@ -41,6 +41,23 @@ interface ResetCreditRequestState {
   tr: TranslateUsage;
 }
 
+// Module-level so the ref-store mutation stays outside any hook body — the
+// immutability rule bars in-callback writes to `state.idempotencyKeysRef.current`.
+function resetIdempotencyKeys(keys: React.MutableRefObject<Record<string, string>>): void {
+  keys.current = {};
+}
+
+function ensureIdempotencyKey(
+  keys: React.MutableRefObject<Record<string, string>>,
+  selectionToken: string
+): string {
+  const existing = keys.current[selectionToken];
+  if (existing) return existing;
+  const created = createIdempotencyKey();
+  keys.current[selectionToken] = created;
+  return created;
+}
+
 function createIdempotencyKey(): string {
   return typeof crypto !== "undefined" && typeof crypto.randomUUID === "function"
     ? crypto.randomUUID()
@@ -112,9 +129,7 @@ function useRedeemCodexResetCredit(state: ResetCreditRequestState) {
     async (selectionToken: string) => {
       const picker = state.resetCreditPicker;
       if (!picker || state.redeemingResetCreditId || !selectionToken) return;
-      const idempotencyKey =
-        state.idempotencyKeysRef.current[selectionToken] ??
-        (state.idempotencyKeysRef.current[selectionToken] = createIdempotencyKey());
+      const idempotencyKey = ensureIdempotencyKey(state.idempotencyKeysRef, selectionToken);
       state.setRedeemingResetCreditId(picker.connectionId);
       state.setErrors((prev) => ({ ...prev, [picker.connectionId]: null }));
       try {
@@ -145,7 +160,7 @@ function useRedeemCodexResetCredit(state: ResetCreditRequestState) {
           [picker.connectionId]: new Date().toISOString(),
         }));
         state.setResetCreditPicker(null);
-        state.idempotencyKeysRef.current = {};
+        resetIdempotencyKeys(state.idempotencyKeysRef);
         notify.success(state.tr("resetCreditRedeemed", "Reset redeemed"));
       } catch (error) {
         const message = getRequestErrorMessage(

src/app/(dashboard)/dashboard/usage/components/RateLimitStatus.tsx

@@ -22,7 +22,9 @@ export default function RateLimitStatus() {
   }, []);
 
   useEffect(() => {
-    load();
+    void (async () => {
+      await load();
+    })();
     const interval = setInterval(load, 10000);
     return () => clearInterval(interval);
   }, [load]);

src/app/(dashboard)/dashboard/usage/components/SessionsTab.tsx

@@ -49,7 +49,9 @@ export default function SessionsTab() {
   }, []);
 
   useEffect(() => {
-    loadSessions();
+    void (async () => {
+      await loadSessions();
+    })();
     const interval = setInterval(loadSessions, 5000);
     return () => clearInterval(interval);
   }, [loadSessions]);

src/app/(dashboard)/home/ProviderQuotaWidget.tsx

@@ -161,7 +161,9 @@ export default function ProviderQuotaWidget({
   const [refreshingAll, setRefreshingAll] = useState(false);
   const [updatedAt, setUpdatedAt] = useState<number | null>(null);
   const refreshingAllRef = useRef(false);
-  const lastRefreshAllAtRef = useRef(Date.now());
+  // State (not a ref): the countdown renders it, and refs cannot be read during
+  // render nor initialized with Date.now() (purity rule).
+  const [lastRefreshAllAt, setLastRefreshAllAt] = useState(() => Date.now());
   const autoRefreshIntervalMs = autoRefreshInterval > 0 ? autoRefreshInterval * 1000 : 0;
   const [autoRefreshClock, setAutoRefreshClock] = useState(() => Date.now());
 
@@ -188,14 +190,16 @@ export default function ProviderQuotaWidget({
   }, []);
 
   useEffect(() => {
-    void loadData();
+    void (async () => {
+      await loadData();
+    })();
   }, [loadData]);
 
   const refreshAll = useCallback(async () => {
     if (refreshingAllRef.current) return;
     refreshingAllRef.current = true;
     const now = Date.now();
-    lastRefreshAllAtRef.current = now;
+    setLastRefreshAllAt(now);
     setAutoRefreshClock(now);
     setRefreshingAll(true);
     try {
@@ -235,10 +239,12 @@ export default function ProviderQuotaWidget({
     if (document.visibilityState !== "visible") return;
     if (refreshingAllRef.current) return;
 
-    if (autoRefreshClock - lastRefreshAllAtRef.current >= autoRefreshIntervalMs) {
-      void refreshAll();
+    if (autoRefreshClock - lastRefreshAllAt >= autoRefreshIntervalMs) {
+      void (async () => {
+        await refreshAll();
+      })();
     }
-  }, [autoRefreshClock, autoRefreshIntervalMs, refreshAll]);
+  }, [autoRefreshClock, lastRefreshAllAt, autoRefreshIntervalMs, refreshAll]);
 
   const providerGroups = useMemo(() => {
     const groups = new Map<string, Connection[]>();
@@ -286,10 +292,7 @@ export default function ProviderQuotaWidget({
             ? tr("refreshing", "Refreshing")
             : autoRefreshIntervalMs > 0
               ? `${tr("autoRefreshing", "Auto-refreshing")} ${formatAutoRefreshCountdown(
-                  Math.max(
-                    0,
-                    autoRefreshIntervalMs - (autoRefreshClock - lastRefreshAllAtRef.current)
-                  )
+                  Math.max(0, autoRefreshIntervalMs - (autoRefreshClock - lastRefreshAllAt))
                 )}`
               : tr("forceRefresh", "Refresh now")}
         </button>

src/app/api/v1/models/catalog.ts

@@ -1432,13 +1432,13 @@ async function buildUnifiedModelsResponseCore(
       return activeAliases.has(alias) || activeAliases.has(provider);
     };
 
-    const hasEquivalentSpecialtyModel = (
+    const findEquivalentSpecialtyModel = (
       providerId: string,
       rawModelId: string,
       type: string,
       scopedModelId: string
     ) =>
-      models.some((model: any) => {
+      models.find((model: any) => {
         if (model?.id === scopedModelId) return true;
         if (model?.owned_by !== providerId || model?.type !== type) return false;
         const existingRoot =
@@ -1450,6 +1450,13 @@ async function buildUnifiedModelsResponseCore(
         return existingRoot === rawModelId;
       });
 
+    const hasEquivalentSpecialtyModel = (
+      providerId: string,
+      rawModelId: string,
+      type: string,
+      scopedModelId: string
+    ) => findEquivalentSpecialtyModel(providerId, rawModelId, type, scopedModelId) !== undefined;
+
     // Helper: strip the provider prefix from a specialty model ID to get the
     // provider-relative path (e.g. "openrouter/google/chirp-3" -> "google/chirp-3").
     // This is the correct key used by the hidden-model lookup — using .split("/").pop()
@@ -1464,7 +1471,22 @@ async function buildUnifiedModelsResponseCore(
       const rawModelId = getSpecialtyModelRelativeId(embModel.id, embModel.provider);
       if (!providerSupportsModel(embModel.provider, rawModelId)) continue;
       if (isModelHiddenBulk(embModel.provider, rawModelId)) continue;
-      if (hasEquivalentSpecialtyModel(embModel.provider, rawModelId, "embedding", embModel.id)) {
+      const existingEmbedding = findEquivalentSpecialtyModel(
+        embModel.provider,
+        rawModelId,
+        "embedding",
+        embModel.id
+      );
+      if (existingEmbedding) {
+        // Discovery publishes no vector width, so the registry is the authority.
+        if (embModel.dimensions !== undefined) {
+          existingEmbedding.dimensions = embModel.dimensions;
+        }
+        // A provider that does not report its endpoints leaves the model unclassified. Being in
+        // the embedding registry is that statement, so make it rather than leave it untyped.
+        if (!existingEmbedding.type) {
+          existingEmbedding.type = "embedding";
+        }
         continue;
       }
       models.push({

src/app/layout.tsx

@@ -5,7 +5,7 @@ import { NextIntlClientProvider } from "next-intl";
 import { getMessages, getLocale, getTranslations } from "next-intl/server";
 import { RTL_LOCALES } from "@/i18n/config";
 import { normalizeComplianceEventTypes } from "@/i18n/request";
-import { getSettings } from "@/lib/db/settings";
+import { getRootLayoutSettings } from "@/lib/db/rootLayoutSettings";
 import type { Viewport } from "next";
 import { PwaRegister } from "@/shared/components/PwaRegister";
 import { LocaleAutoDetect } from "@/shared/components/LocaleAutoDetect";
@@ -22,9 +22,9 @@ export const viewport: Viewport = {
 };
 
 export async function generateMetadata() {
-  const settings = await getSettings();
-  const instanceName = settings?.instanceName || "OmniRoute";
-  const customFaviconUrl = settings?.customFaviconUrl || settings?.customFaviconBase64;
+  const settings = await getRootLayoutSettings();
+  const instanceName = settings.instanceName;
+  const customFaviconUrl = settings.customFaviconUrl || settings.customFaviconBase64;
 
   return {
     title: `${instanceName} — AI Gateway for Multi-Provider LLMs`,

src/instrumentation-node.ts

@@ -319,7 +319,7 @@ export async function registerNodejs(): Promise<void> {
   process.title = renameProcessTitle(process.title);
 
   // Initialize proxy fetch patch FIRST (before any HTTP requests)
-  await import("@omniroute/open-sse/index.ts");
+  await import("@omniroute/open-sse/utils/proxyFetch.ts");
   console.log("[STARTUP] Global fetch proxy patch initialized");
 
   // Register quota fetchers early so combo routing can use real quota-aware

src/lib/batches/dispatch.ts

@@ -1,29 +1,6 @@
 import type { SupportedBatchEndpoint } from "@/shared/constants/batchEndpoints";
-
-type BatchRouteHandler = (request: Request) => Promise<Response> | Response;
-
-const handlerLoaders: Record<SupportedBatchEndpoint, () => Promise<BatchRouteHandler>> = {
-  "/v1/responses": async () => (await import("@/app/api/v1/responses/route")).POST,
-  "/v1/chat/completions": async () => (await import("@/app/api/v1/chat/completions/route")).POST,
-  "/v1/embeddings": async () => (await import("@/app/api/v1/embeddings/route")).POST,
-  "/v1/completions": async () => (await import("@/app/api/v1/completions/route")).POST,
-  "/v1/moderations": async () => (await import("@/app/api/v1/moderations/route")).POST,
-  "/v1/images/generations": async () =>
-    (await import("@/app/api/v1/images/generations/route")).POST,
-  "/v1/videos/generations": async () =>
-    (await import("@/app/api/v1/videos/generations/route")).POST,
-};
-
-const handlerCache = new Map<SupportedBatchEndpoint, BatchRouteHandler>();
-
-async function getHandler(endpoint: SupportedBatchEndpoint): Promise<BatchRouteHandler> {
-  const cached = handlerCache.get(endpoint);
-  if (cached) return cached;
-
-  const handler = await handlerLoaders[endpoint]();
-  handlerCache.set(endpoint, handler);
-  return handler;
-}
+import { getRuntimePorts } from "@/lib/runtime/ports";
+import { normalizeBasePath } from "@/shared/utils/basePath";
 
 async function dispatchBatchApiRequest({
   endpoint,
@@ -39,13 +16,17 @@ async function dispatchBatchApiRequest({
     headers.set("Authorization", `Bearer ${apiKey}`);
   }
 
-  const handler = await getHandler(endpoint);
-  const request = new Request(`http://localhost${endpoint}`, {
+  const { dashboardPort } = getRuntimePorts();
+  const basePath = normalizeBasePath(process.env.OMNIROUTE_BASE_PATH);
+  const url = `http://127.0.0.1:${dashboardPort}${basePath}${endpoint}`;
+
+  return await globalThis.fetch(url, {
     method: "POST",
     headers,
     body: JSON.stringify(body),
+    // Never follow a redirect while carrying the stored batch API key.
+    redirect: "error",
   });
-  return await handler(request);
 }
 
 export const dispatch = {

src/lib/combos/intelligentRouting.ts

@@ -67,6 +67,7 @@ export const MODE_PACK_OPTIONS = [
 
 export const ROUTER_STRATEGY_OPTIONS = [
   { id: "rules", label: "Rules (6-Factor Scoring)" },
+  { id: "score", label: "Highest Weighted Score" },
   { id: "cost", label: "Cost Optimized" },
   { id: "latency", label: "Latency Optimized" },
   { id: "sla-aware", label: "SLA-aware" },

src/lib/db/core.ts

@@ -40,6 +40,7 @@ import { invalidateDbCache } from "./readCache";
 import { rowToCamel } from "./caseMapping";
 import { isAutomatedTestProcess } from "@/shared/utils/testProcess";
 import { parseModelAccessMode } from "./apiKeys/modelAccessMode";
+import { getExistingDbInstance as getDb, setDbInstance as setDb } from "./singleton";
 // Re-exported so existing call sites that pull these helpers off the core module keep working.
 export { toSnakeCase, toCamelCase, objToSnake, rowToCamel, cleanNulls } from "./caseMapping";
 import {
@@ -514,7 +515,6 @@ const SCHEMA_SQL = `
 // Module-level `let` resets on every webpack recompile, causing connection leaks.
 
 declare global {
-  var __omnirouteDb: SqliteAdapter | undefined;
   // Cycle-breaker counter for the probe-failed/restore cascade. Survives
   // Next.js HMR re-evaluations so concurrent subsystems all see the same
   // count and we abort with a clear error instead of looping forever.
@@ -529,18 +529,6 @@ declare global {
   var __omnirouteDbOomFailureCount: number | undefined;
 }
 
-function getDb(): SqliteDatabase | null {
-  return globalThis.__omnirouteDb ?? null;
-}
-
-function setDb(db: SqliteDatabase | null): void {
-  if (db) {
-    globalThis.__omnirouteDb = db;
-  } else {
-    delete globalThis.__omnirouteDb;
-  }
-}
-
 function checkpointDb(db: SqliteDatabase, mode: CheckpointMode = "TRUNCATE"): boolean {
   if (isCloud || isBuildPhase || !SQLITE_FILE) return false;
   db.pragma(`wal_checkpoint(${mode})`);

src/lib/db/rootLayoutSettings.ts

@@ -0,0 +1,62 @@
+import { getExistingDbInstance } from "./singleton";
+
+export interface RootLayoutSettings {
+  instanceName: string;
+  customFaviconUrl: string;
+  customFaviconBase64: string;
+}
+
+type RootLayoutSettingKey = keyof RootLayoutSettings;
+type SettingsRow = {
+  key?: unknown;
+  value?: unknown;
+};
+
+const ROOT_LAYOUT_SETTING_KEYS = [
+  "instanceName",
+  "customFaviconUrl",
+  "customFaviconBase64",
+] as const satisfies readonly RootLayoutSettingKey[];
+
+const ROOT_LAYOUT_SETTING_KEY_SET = new Set<string>(ROOT_LAYOUT_SETTING_KEYS);
+
+const DEFAULT_ROOT_LAYOUT_SETTINGS: RootLayoutSettings = {
+  instanceName: "OmniRoute",
+  customFaviconUrl: "",
+  customFaviconBase64: "",
+};
+
+function parseStoredString(value: unknown): string | null {
+  if (typeof value !== "string") return null;
+
+  try {
+    const parsed = JSON.parse(value) as unknown;
+    return typeof parsed === "string" ? parsed : null;
+  } catch {
+    return null;
+  }
+}
+
+/** Read only the settings needed while compiling and rendering the root layout. */
+export async function getRootLayoutSettings(): Promise<RootLayoutSettings> {
+  const db = getExistingDbInstance();
+  if (!db) return { ...DEFAULT_ROOT_LAYOUT_SETTINGS };
+
+  const rows = db
+    .prepare(
+      `SELECT key, value FROM key_value
+       WHERE namespace = 'settings' AND key IN (?, ?, ?)`
+    )
+    .all(...ROOT_LAYOUT_SETTING_KEYS) as SettingsRow[];
+  const settings = { ...DEFAULT_ROOT_LAYOUT_SETTINGS };
+
+  for (const row of rows) {
+    if (typeof row.key !== "string" || !ROOT_LAYOUT_SETTING_KEY_SET.has(row.key)) continue;
+    const value = parseStoredString(row.value);
+    if (value === null) continue;
+    settings[row.key as RootLayoutSettingKey] = value;
+  }
+
+  if (!settings.instanceName) settings.instanceName = DEFAULT_ROOT_LAYOUT_SETTINGS.instanceName;
+  return settings;
+}

src/lib/db/singleton.ts

@@ -0,0 +1,19 @@
+import type { SqliteAdapter } from "./adapters/types";
+
+declare global {
+  var __omnirouteDb: SqliteAdapter | undefined;
+}
+
+/** Read the process-wide DB handle without initializing storage. */
+export function getExistingDbInstance(): SqliteAdapter | null {
+  return globalThis.__omnirouteDb ?? null;
+}
+
+/** Replace the process-wide DB handle while preserving it across Next.js HMR. */
+export function setDbInstance(db: SqliteAdapter | null): void {
+  if (db) {
+    globalThis.__omnirouteDb = db;
+  } else {
+    delete globalThis.__omnirouteDb;
+  }
+}

src/lib/gracefulShutdown.ts

@@ -19,7 +19,15 @@ const SHUTDOWN_TIMEOUT_MS = parseInt(process.env.SHUTDOWN_TIMEOUT_MS || "30000",
 
 declare global {
   var __omnirouteShutdown:
-    { init: boolean; shuttingDown: boolean; activeRequests: number } | undefined;
+    | {
+        init: boolean;
+        shuttingDown: boolean;
+        activeRequests: number;
+        shutdownPromise?: Promise<void>;
+      }
+    | undefined;
+  var __omnirouteRequestShutdown: ((signal: string) => Promise<void>) | undefined;
+  var __omnirouteCustomServerOwnsShutdown: boolean | undefined;
 }
 
 function getShutdownState() {
@@ -102,12 +110,14 @@ async function cleanup(): Promise<void> {
       { closeDbInstance },
       { flushSpendBatchWriter },
       { closeLogRotation },
+      { closeSharedLoggerResource },
       { closeCallLogSaves },
     ] = await Promise.all([
       import("@omniroute/open-sse/mcp-server/audit.ts"),
       import("@/lib/db/core"),
       import("@/lib/spend/batchWriter"),
       import("@/lib/logRotation"),
+      import("@/shared/utils/loggerResource"),
       import("@/lib/usage/callLogs"),
     ]);
     const flushResult = await flushSpendBatchWriter();
@@ -123,9 +133,6 @@ async function cleanup(): Promise<void> {
     if (closeDbInstance()) {
       console.log("[Shutdown] SQLite database checkpointed and closed.");
     }
-    closeLogRotation();
-    console.log("[Shutdown] Log rotation timer stopped.");
-
     // Tear down any persistent VNC login browser containers so they don't leak
     // past the server process. Best-effort; no-op if the feature was never used
     // or the docker CLI is unavailable.
@@ -147,41 +154,62 @@ async function cleanup(): Promise<void> {
     } catch {
       /* feature unused */
     }
+
+    await closeSharedLoggerResource();
+    closeLogRotation();
+    console.log("[Shutdown] Logger transport and log rotation stopped.");
   } catch (err) {
     console.error("[Shutdown] Error during cleanup:", (err as Error).message);
   }
 }
 
+/**
+ * Start the process-wide shutdown sequence, or join the sequence already in progress.
+ */
+export function requestGracefulShutdown(signal: string): Promise<void> {
+  const state = getShutdownState();
+  if (state.shutdownPromise) return state.shutdownPromise;
+
+  state.shuttingDown = true;
+  markServerStopping();
+  state.shutdownPromise = (async () => {
+    console.log(`\n[Shutdown] Received ${signal}. Draining ${state.activeRequests} request(s)...`);
+
+    await waitForDrain();
+    await cleanup();
+
+    console.log("[Shutdown] Bye.");
+  })();
+
+  return state.shutdownPromise;
+}
+
 /**
  * Initialize graceful shutdown handlers.
  * Should be called once during server startup.
  */
 export function initGracefulShutdown(): void {
   const state = getShutdownState();
+  globalThis.__omnirouteRequestShutdown ??= requestGracefulShutdown;
   if (state.init) return;
   state.init = true;
 
-  const shutdown = async (signal: string) => {
-    if (state.shuttingDown) return;
-    state.shuttingDown = true;
-    markServerStopping();
+  if (globalThis.__omnirouteCustomServerOwnsShutdown) {
+    console.log("[Shutdown] Cleanup registered with the custom server shutdown owner.");
+    return;
+  }
 
-    console.log(`\n[Shutdown] Received ${signal}. Draining ${state.activeRequests} request(s)...`);
-
-    await waitForDrain();
-    await cleanup();
-
-    console.log("[Shutdown] Bye.");
-    process.exit(0);
+  const shutdown = (signal: string) => {
+    void globalThis.__omnirouteRequestShutdown?.(signal).then(() => process.exit(0));
   };
 
-  process.on("SIGTERM", () => shutdown("SIGTERM"));
-  process.on("SIGINT", () => shutdown("SIGINT"));
+  process.on("SIGTERM", () => void shutdown("SIGTERM"));
+  process.on("SIGINT", () => void shutdown("SIGINT"));
   // #8045: on Windows, closing the console window delivers CTRL_CLOSE_EVENT, which
   // Node/libuv maps to a JS-visible "SIGHUP" event — without this listener, closing
   // the window never runs cleanup() (WAL checkpoint + closeDbInstance()), leaving
   // storage.sqlite's WAL un-checkpointed for the next launch.
-  process.on("SIGHUP", () => shutdown("SIGHUP"));
+  process.on("SIGHUP", () => void shutdown("SIGHUP"));
 
   console.log("[Shutdown] Graceful shutdown handlers registered.");
 }

src/lib/logRotation.ts

@@ -42,8 +42,18 @@ export function getAppLogRotationCheckInterval(): number {
   );
 }
 
-/** Module-level timer handle — cleared by closeLogRotation(). */
-let rotationTimer: ReturnType<typeof setInterval> | null = null;
+interface LogRotationState {
+  timer: ReturnType<typeof setInterval> | null;
+}
+
+declare global {
+  var __omnirouteLogRotationState: LogRotationState | undefined;
+}
+
+/** Process-wide state survives Next.js development HMR and split server chunks. */
+function getLogRotationState(): LogRotationState {
+  return (globalThis.__omnirouteLogRotationState ??= { timer: null });
+}
 
 export function getLogConfig() {
   const logToFile = getAppLogToFile();
@@ -172,6 +182,9 @@ export function cleanupOverflowLogs(logFilePath: string, maxFiles: number): void
  * Call closeLogRotation() during application shutdown to clear the timer.
  */
 export function initLogRotation(): void {
+  const state = getLogRotationState();
+  if (state.timer !== null) return;
+
   const config = getLogConfig();
   if (!config.logToFile) return;
 
@@ -181,7 +194,7 @@ export function initLogRotation(): void {
   cleanupOverflowLogs(config.logFilePath, config.maxFiles);
 
   const intervalMs = getAppLogRotationCheckInterval();
-  rotationTimer = setInterval(
+  state.timer = setInterval(
     (filePath: string, maxSize: number, maxFiles: number) => {
       rotateIfNeeded(filePath, maxSize);
       cleanupOverflowLogs(filePath, maxFiles);
@@ -191,7 +204,7 @@ export function initLogRotation(): void {
     config.maxFileSize,
     config.maxFiles
   );
-  rotationTimer.unref?.();
+  state.timer.unref?.();
 }
 
 /**
@@ -199,8 +212,9 @@ export function initLogRotation(): void {
  * Idempotent — safe to call multiple times.
  */
 export function closeLogRotation(): void {
-  if (rotationTimer !== null) {
-    clearInterval(rotationTimer);
-    rotationTimer = null;
+  const state = getLogRotationState();
+  if (state.timer !== null) {
+    clearInterval(state.timer);
+    state.timer = null;
   }
 }

src/lib/providers/validation/anthropicFormat.ts

@@ -9,6 +9,7 @@ import {
   joinClaudeCodeCompatibleUrl,
   joinBaseUrlAndPath,
 } from "@omniroute/open-sse/services/claudeCodeCompatible.ts";
+import { getDefaultExecutor } from "@omniroute/open-sse/executors/defaultResolver.ts";
 import {
   addModelsSuffix,
   normalizeAnthropicBaseUrl,
@@ -137,8 +138,7 @@ export async function validateClaudeOAuthInline({
     typeof override === "string" && override ? override : modelId || "claude-haiku-4-5-20251001";
 
   try {
-    const { getExecutor } = await import("@omniroute/open-sse/executors/index.ts");
-    const executed = await (await getExecutor("claude")).execute({
+    const executed = await getDefaultExecutor("claude").execute({
       model: testModelId,
       body: {
         model: testModelId,

src/lib/services/quotaAutoPing.ts

@@ -22,13 +22,12 @@
 
 import { logger } from "@omniroute/open-sse/utils/logger.ts";
 import { sanitizeErrorMessage } from "@omniroute/open-sse/utils/error.ts";
-import { getExecutor } from "@omniroute/open-sse/executors/index.ts";
 import type { BaseExecutor } from "@omniroute/open-sse/executors/base";
 import { getCodexUsage } from "@omniroute/open-sse/services/usage/codex.ts";
 import { getSettings } from "@/lib/db/settings";
 import { getProviderConnections, updateProviderConnection } from "@/lib/db/providers";
 import { isConnectionUnavailableToAuxiliaryActivity } from "@/lib/exclusiveLeaseIsolation";
-import { refreshAndUpdateCredentials } from "@/lib/usage/providerLimits";
+import { refreshAndUpdateCredentialsWithResolver } from "@/lib/usage/providerLimits/credentialRefresh";
 import { getCircuitBreaker } from "@/shared/utils/circuitBreaker";
 import {
   QUOTA_AUTOPING_FAILURE_COOLDOWN_MS,
@@ -63,8 +62,11 @@ export interface QuotaAutoPingDeps {
   refreshAndUpdateCredentials: (
     connection: QuotaAutoPingConnection
   ) => Promise<{ connection: QuotaAutoPingConnection }>;
-  getCodexUsage: (accessToken?: string, providerSpecificData?: JsonRecord) => Promise<JsonRecord>;
-  getExecutor: (provider: string) => Promise<BaseExecutor>;
+  getCodexUsage: (
+    accessToken?: string,
+    providerSpecificData?: JsonRecord
+  ) => Promise<JsonRecord>;
+  getExecutor: (provider: "codex") => Promise<BaseExecutor>;
   canExecuteProvider: (provider: string) => boolean;
   isConnectionUnavailableToAuxiliaryActivity: (connectionId: string) => Promise<boolean>;
 }
@@ -79,15 +81,33 @@ export function createQuotaAutoPingState(): QuotaAutoPingState {
   return { running: false, resetCache: {}, failureCache: {} };
 }
 
+let codexExecutorPromise: Promise<BaseExecutor> | null = null;
+
+async function loadQuotaAutoPingExecutor(provider: string): Promise<BaseExecutor> {
+  if (provider !== "codex") {
+    throw new Error(`Quota auto-ping does not support provider "${provider}"`);
+  }
+
+  try {
+    codexExecutorPromise ??= import("@omniroute/open-sse/executors/codex.ts").then(
+      ({ CodexExecutor }) => new CodexExecutor()
+    );
+    return await codexExecutorPromise;
+  } catch (error) {
+    codexExecutorPromise = null;
+    throw error;
+  }
+}
+
 export function createDefaultQuotaAutoPingDeps(): QuotaAutoPingDeps {
   return {
     getSettings,
     getProviderConnections,
     updateProviderConnection,
     refreshAndUpdateCredentials: async (connection) =>
-      refreshAndUpdateCredentials(connection as never),
+      refreshAndUpdateCredentialsWithResolver(connection, loadQuotaAutoPingExecutor),
     getCodexUsage,
-    getExecutor,
+    getExecutor: loadQuotaAutoPingExecutor,
     canExecuteProvider: (provider) => getCircuitBreaker(provider).canExecute(),
     isConnectionUnavailableToAuxiliaryActivity,
   };

src/lib/usage/providerLimits.ts

@@ -18,13 +18,10 @@ import { clearRecoveredProviderState } from "@/sse/services/auth";
 import { getMachineId } from "@/shared/utils/machine";
 import { USAGE_SUPPORTED_PROVIDERS } from "@/shared/constants/providers";
 import { mergeProviderLimitsCacheEntry, toProviderLimitsCacheEntry } from "./providerLimitsCache";
-import { getExecutor } from "@omniroute/open-sse/executors/index.ts";
+import { getCredentialRefreshExecutor } from "@omniroute/open-sse/executors/credential.ts";
 import { getUsageForProvider } from "@omniroute/open-sse/services/usage.ts";
 import { cooldownUntilMs } from "@omniroute/open-sse/services/accountFallback.ts";
-import {
-  rotationGroupFor,
-  serializeRefresh,
-} from "@omniroute/open-sse/services/refreshSerializer.ts";
+import { rotationGroupFor } from "@omniroute/open-sse/services/refreshSerializer.ts";
 import {
   extractCodeAssistOnboardTierId,
   extractCodeAssistSubscriptionTier,
@@ -42,29 +39,15 @@ import {
   sanitizeUsageQuotasForProvider,
 } from "./providerLimits/quotaNormalize";
 import { syncInChunksWithSpacing } from "./providerLimits/chunkedSpacingSync";
+import {
+  refreshAndUpdateCredentialsWithResolver,
+  type CredentialRefreshOptions,
+  type ProviderConnectionLike,
+} from "./providerLimits/credentialRefresh";
+export { shouldAttemptRotatingRefresh } from "./providerLimits/credentialRefresh";
 type JsonRecord = Record<string, unknown>;
 type SyncSource = "manual" | "scheduled";
 
-interface ProviderConnectionLike {
-  id: string;
-  provider: string;
-  authType?: string;
-  accessToken?: string;
-  refreshToken?: string;
-  expiresAt?: string;
-  tokenExpiresAt?: string;
-  providerSpecificData?: JsonRecord;
-  testStatus?: string;
-  isActive?: boolean;
-  lastError?: string | null;
-  lastErrorAt?: string | null;
-  lastErrorType?: string | null;
-  lastErrorSource?: string | null;
-  errorCode?: string | number | null;
-  rateLimitedUntil?: string | null;
-  backoffLevel?: number;
-}
-
 const PROVIDER_LIMITS_APIKEY_PROVIDERS = new Set([
   "glm",
   "glm-cn",
@@ -218,122 +201,15 @@ async function syncToCloudIfEnabled() {
   }
 }
 
-/**
- * Whether the quota path may refresh this provider's token. Exported for testing.
- *
- * Rotating-refresh providers (Codex/OpenAI share one Auth0 client_id, etc.) mint a
- * single-use refresh_token on every refresh. The BULK quota-sync path runs many
- * connections concurrently; refreshing sibling accounts in parallel makes Auth0
- * revoke the whole token family (openai/codex#9648) and kills every account but
- * the last (#3019). So the bulk path never refreshes rotating providers
- * (`allowRotatingRefresh` falsy). The on-demand, per-connection path opts in and
- * is made safe by `serializeRefresh` (one token mint at a time per rotation group,
- * so even N concurrent per-account requests can never refresh siblings in
- * parallel). Non-rotating providers are always eligible.
- */
-export function shouldAttemptRotatingRefresh(
-  provider: string,
-  allowRotatingRefresh: boolean | undefined
-): boolean {
-  if (rotationGroupFor(provider) === null) return true;
-  return allowRotatingRefresh === true;
-}
-
 export async function refreshAndUpdateCredentials(
   connection: ProviderConnectionLike,
-  opts: { allowRotatingRefresh?: boolean; force?: boolean } = {}
+  opts: CredentialRefreshOptions = {}
 ) {
-  if (!shouldAttemptRotatingRefresh(connection.provider, opts.allowRotatingRefresh)) {
-    return { connection, refreshed: false };
-  }
-  const executor = await getExecutor(connection.provider);
-  const credentials = {
-    connectionId: connection.id,
-    accessToken: connection.accessToken,
-    refreshToken: connection.refreshToken,
-    expiresAt: connection.tokenExpiresAt || connection.expiresAt || null,
-    providerSpecificData: connection.providerSpecificData,
-    copilotToken: connection.providerSpecificData?.copilotToken,
-    copilotTokenExpiresAt: connection.providerSpecificData?.copilotTokenExpiresAt,
-  };
-
-  // `force` is used ONLY on the reactive 401 recovery path (a usage fetch came
-  // back unauthorized) — it bypasses the proactive `needsRefresh` heuristic so
-  // imported accounts (expiresAt=null, where needsRefresh is always false) can
-  // still re-mint. The mint stays serialized per rotation group; this never
-  // refreshes proactively from the bulk path (#3019 guard above is unchanged).
-  if (!opts.force && !executor.needsRefresh(credentials)) {
-    return { connection, refreshed: false };
-  }
-
-  // Serialize the actual token mint per rotation group so two sibling accounts
-  // never hit Auth0 concurrently (passthrough for non-rotating providers).
-  const refreshResult = (await serializeRefresh(connection.provider, () =>
-    executor.refreshCredentials(credentials, console)
-  )) as
-    | (JsonRecord & {
-        accessToken?: string;
-        refreshToken?: string;
-        expiresIn?: number;
-        expiresAt?: string;
-        copilotToken?: string;
-        copilotTokenExpiresAt?: string;
-      })
-    | null;
-
-  if (!refreshResult) {
-    // Refresh failed but we still have an accessToken — fall back to the
-    // existing token for ANY OAuth provider (graceful degradation) instead of
-    // hard-failing. Previously this was qualified to `provider === "github"`,
-    // which left every other provider stuck on a transient refresh failure even
-    // when a usable access token was still on hand.
-    if (connection.accessToken) {
-      return { connection, refreshed: false };
-    }
-    throw withStatus(
-      new Error("Failed to refresh credentials. Please re-authorize the connection."),
-      401
-    );
-  }
-
-  const updateData: JsonRecord = {
-    updatedAt: new Date().toISOString(),
-  };
-
-  if (refreshResult.accessToken) {
-    updateData.accessToken = refreshResult.accessToken;
-  }
-  if (refreshResult.refreshToken) {
-    updateData.refreshToken = refreshResult.refreshToken;
-  }
-  if (refreshResult.expiresIn) {
-    const expiresAt = new Date(Date.now() + refreshResult.expiresIn * 1000).toISOString();
-    updateData.expiresAt = expiresAt;
-    updateData.tokenExpiresAt = expiresAt;
-  } else if (refreshResult.expiresAt) {
-    updateData.expiresAt = refreshResult.expiresAt;
-    updateData.tokenExpiresAt = refreshResult.expiresAt;
-  }
-  if (refreshResult.copilotToken || refreshResult.copilotTokenExpiresAt) {
-    updateData.providerSpecificData = {
-      ...(connection.providerSpecificData || {}),
-      copilotToken: refreshResult.copilotToken,
-      copilotTokenExpiresAt: refreshResult.copilotTokenExpiresAt,
-    };
-  }
-
-  await updateProviderConnection(connection.id, updateData);
-
-  return {
-    connection: {
-      ...connection,
-      ...updateData,
-      providerSpecificData:
-        (updateData.providerSpecificData as JsonRecord | undefined) ||
-        connection.providerSpecificData,
-    },
-    refreshed: true,
-  };
+  return refreshAndUpdateCredentialsWithResolver(
+    connection,
+    getCredentialRefreshExecutor,
+    opts
+  );
 }
 
 function isUsageAuthError(message: unknown): boolean {

src/lib/usage/providerLimits/credentialRefresh.ts

@@ -0,0 +1,150 @@
+import { updateProviderConnection } from "@/lib/db/providers";
+import type { BaseExecutor } from "@omniroute/open-sse/executors/base";
+import {
+  rotationGroupFor,
+  serializeRefresh,
+} from "@omniroute/open-sse/services/refreshSerializer.ts";
+
+type JsonRecord = Record<string, unknown>;
+
+type CredentialRefreshResult = JsonRecord & {
+  accessToken?: string;
+  refreshToken?: string;
+  expiresIn?: number;
+  expiresAt?: string;
+  copilotToken?: string;
+  copilotTokenExpiresAt?: string;
+};
+
+export interface ProviderConnectionLike {
+  id: string;
+  provider: string;
+  authType?: string;
+  accessToken?: string;
+  refreshToken?: string;
+  expiresAt?: string | null;
+  tokenExpiresAt?: string | null;
+  providerSpecificData?: JsonRecord;
+  testStatus?: string;
+  isActive?: boolean;
+  lastError?: string | null;
+  lastErrorAt?: string | null;
+  lastErrorType?: string | null;
+  lastErrorSource?: string | null;
+  errorCode?: string | number | null;
+  rateLimitedUntil?: string | null;
+  backoffLevel?: number;
+}
+
+export interface CredentialRefreshOptions {
+  allowRotatingRefresh?: boolean;
+  force?: boolean;
+}
+
+export type CredentialExecutorResolver = (provider: string) => Promise<BaseExecutor>;
+
+function withStatus(error: Error, status: number): Error & { status: number } {
+  return Object.assign(error, { status });
+}
+
+/**
+ * Whether the quota path may refresh this provider's token.
+ *
+ * Rotating-refresh providers mint a single-use refresh token on every refresh,
+ * so bulk quota sync must not refresh siblings concurrently. The on-demand path
+ * explicitly opts in and remains serialized per rotation group.
+ */
+export function shouldAttemptRotatingRefresh(
+  provider: string,
+  allowRotatingRefresh: boolean | undefined
+): boolean {
+  if (rotationGroupFor(provider) === null) return true;
+  return allowRotatingRefresh === true;
+}
+
+function buildCredentialUpdateData(
+  connection: ProviderConnectionLike,
+  refreshResult: CredentialRefreshResult
+): JsonRecord {
+  const updateData: JsonRecord = {
+    updatedAt: new Date().toISOString(),
+  };
+
+  if (refreshResult.accessToken) {
+    updateData.accessToken = refreshResult.accessToken;
+  }
+  if (refreshResult.refreshToken) {
+    updateData.refreshToken = refreshResult.refreshToken;
+  }
+  if (refreshResult.expiresIn) {
+    const expiresAt = new Date(Date.now() + refreshResult.expiresIn * 1000).toISOString();
+    updateData.expiresAt = expiresAt;
+    updateData.tokenExpiresAt = expiresAt;
+  } else if (refreshResult.expiresAt) {
+    updateData.expiresAt = refreshResult.expiresAt;
+    updateData.tokenExpiresAt = refreshResult.expiresAt;
+  }
+  if (refreshResult.copilotToken || refreshResult.copilotTokenExpiresAt) {
+    updateData.providerSpecificData = {
+      ...(connection.providerSpecificData || {}),
+      copilotToken: refreshResult.copilotToken,
+      copilotTokenExpiresAt: refreshResult.copilotTokenExpiresAt,
+    };
+  }
+
+  return updateData;
+}
+
+/** Refresh and persist credentials using a caller-supplied executor resolver. */
+export async function refreshAndUpdateCredentialsWithResolver(
+  connection: ProviderConnectionLike,
+  resolveExecutor: CredentialExecutorResolver,
+  opts: CredentialRefreshOptions = {}
+) {
+  if (!shouldAttemptRotatingRefresh(connection.provider, opts.allowRotatingRefresh)) {
+    return { connection, refreshed: false };
+  }
+  const executor = await resolveExecutor(connection.provider);
+  const credentials = {
+    connectionId: connection.id,
+    accessToken: connection.accessToken,
+    refreshToken: connection.refreshToken,
+    expiresAt: connection.tokenExpiresAt || connection.expiresAt || null,
+    providerSpecificData: connection.providerSpecificData,
+    copilotToken: connection.providerSpecificData?.copilotToken,
+    copilotTokenExpiresAt: connection.providerSpecificData?.copilotTokenExpiresAt,
+  };
+
+  if (!opts.force && !executor.needsRefresh(credentials)) {
+    return { connection, refreshed: false };
+  }
+
+  const refreshResult = (await serializeRefresh(connection.provider, () =>
+    executor.refreshCredentials(credentials, console)
+  )) as CredentialRefreshResult | null;
+
+  if (!refreshResult) {
+    if (connection.accessToken) {
+      return { connection, refreshed: false };
+    }
+    throw withStatus(
+      new Error("Failed to refresh credentials. Please re-authorize the connection."),
+      401
+    );
+  }
+
+  const updateData = buildCredentialUpdateData(connection, refreshResult);
+
+  await updateProviderConnection(connection.id, updateData);
+
+  return {
+    connection: {
+      ...connection,
+      ...updateData,
+      providerSpecificData:
+        (updateData.providerSpecificData as JsonRecord | undefined) ||
+        connection.providerSpecificData,
+    },
+    refreshed: true,
+  };
+}

src/shared/components/RequestLoggerV2.tsx

@@ -55,6 +55,16 @@ import {
 // Reduced from 300 → 50 to avoid browser freeze and network saturation.
 const PAGE_SIZE = 50;
 
+// Column sort toggle mapping: clicking a column header toggles asc/desc.
+const COLUMN_SORT_MAP = {
+  status: { desc: "status_desc", asc: "status_asc" },
+  model: { desc: "model_desc", asc: "model_asc" },
+  tokens: { desc: "tokens_desc", asc: "tokens_asc" },
+  tps: { desc: "tps_desc", asc: "tps_asc" },
+  duration: { desc: "duration_desc", asc: "duration_asc" },
+  time: { desc: "newest", asc: "oldest" },
+} as const;
+
 function getLogTotalTokens(log) {
   return (log?.tokens?.in || 0) + (log?.tokens?.out || 0);
 }
@@ -147,17 +157,8 @@ const RequestLoggerV2 = forwardRef<RequestLoggerV2Handle, { initialSelectedId?:
     const [groupedView, setGroupedView] = useState(false);
     const [detailLoading, setDetailLoading] = useState(false);
 
-    // Column sort toggle: clicking a column header toggles asc/desc
-    const columnSortMap = {
-      status: { desc: "status_desc", asc: "status_asc" },
-      model: { desc: "model_desc", asc: "model_asc" },
-      tokens: { desc: "tokens_desc", asc: "tokens_asc" },
-      tps: { desc: "tps_desc", asc: "tps_asc" },
-      duration: { desc: "duration_desc", asc: "duration_asc" },
-      time: { desc: "newest", asc: "oldest" },
-    };
     const toggleSort = useCallback((column: string) => {
-      const mapping = columnSortMap[column as keyof typeof columnSortMap];
+      const mapping = COLUMN_SORT_MAP[column as keyof typeof COLUMN_SORT_MAP];
       if (!mapping) return;
       setSortBy((prev) => {
         if (prev === mapping.desc) return mapping.asc;
@@ -166,7 +167,7 @@ const RequestLoggerV2 = forwardRef<RequestLoggerV2Handle, { initialSelectedId?:
     }, []);
     const getSortIndicator = useCallback(
       (column: string) => {
-        const mapping = columnSortMap[column as keyof typeof columnSortMap];
+        const mapping = COLUMN_SORT_MAP[column as keyof typeof COLUMN_SORT_MAP];
         if (!mapping) return "";
         if (sortBy === mapping.desc) return " ↓";
         if (sortBy === mapping.asc) return " ↑";
@@ -535,89 +536,7 @@ const RequestLoggerV2 = forwardRef<RequestLoggerV2Handle, { initialSelectedId?:
     // endpoint until the row appears.
     const router = useRouter();
 
-    const openDetail = async (logEntry) => {
-      // Guard: if no valid id provided, close instead of opening an empty modal
-      if (!logEntry?.id) {
-        try {
-          closeDetail();
-        } catch {}
-        return;
-      }
-
-      const requestToken = `${logEntry.id}:${Date.now()}:${Math.random()}`;
-      detailRequestRef.current = requestToken;
-      const isCurrentDetailRequest = () => detailRequestRef.current === requestToken;
-
-      setSelectedLog(logEntry);
-      try {
-        const url = new URL(globalThis.location.href);
-        url.searchParams.set("id", logEntry.id);
-        router.replace(url.pathname + url.search, { scroll: false });
-      } catch (e) {
-        // ignore navigation errors
-      }
-      setDetailLoading(true);
-      setDetailData(null);
-      try {
-        const res = await fetch(`/api/logs/${logEntry.id}`, { cache: "no-store" });
-        if (res.ok) {
-          const data = await res.json();
-          if (!isCurrentDetailRequest()) return;
-          const dataHasPipeline =
-            data?.pipelinePayloads && Object.keys(data.pipelinePayloads || {}).length > 0;
-          setDetailData((prev: { pipelinePayloads: any }) => ({
-            ...prev,
-            ...data,
-            pipelinePayloads: dataHasPipeline ? data.pipelinePayloads : prev?.pipelinePayloads,
-          }));
-          // ensure the modal summary reflects the fetched call log summary
-          if (data && typeof data === "object") {
-            setSelectedLog((prev: any) => ({
-              ...prev,
-              ...data,
-              active: data.active === true,
-            }));
-          }
-        } else {
-          // A deep-linked id can legitimately 404 while the request is still
-          // finalizing. Keep the modal open and poll /api/logs/[id] instead of
-          // falling back to an in-memory active-request endpoint.
-          if (!isCurrentDetailRequest()) return;
-          if (res.status === 404) {
-            if (logEntry.pendingLookup || logEntry.active) {
-              setSelectedLog((prev: { method: any; path: any }) => ({
-                ...prev,
-                id: logEntry.id,
-                status: 0,
-                method: prev?.method,
-                path: prev?.path || "",
-              }));
-              setDetailData({ detailState: "pending" });
-              return;
-            }
-            try {
-              console.warn("Log not found:", logEntry.id);
-            } catch {}
-            try {
-              closeDetail();
-            } catch {}
-            return;
-          }
-          // other errors: show a minimal error indicator by setting detailData to an error object
-          try {
-            const body = await res.text().catch(() => null);
-            if (!isCurrentDetailRequest()) return;
-            setDetailData({ error: `Failed to fetch log (status ${res.status})`, body });
-          } catch {}
-        }
-      } catch (error) {
-        console.error("Failed to fetch log detail:", error);
-      } finally {
-        if (isCurrentDetailRequest()) setDetailLoading(false);
-      }
-    };
-
-    const closeDetail = () => {
+    const closeDetail = useCallback(() => {
       detailRequestRef.current = "";
       setSelectedLog(null);
       setDetailData(null);
@@ -629,7 +548,92 @@ const RequestLoggerV2 = forwardRef<RequestLoggerV2Handle, { initialSelectedId?:
       } catch (e) {
         // ignore navigation errors
       }
-    };
+    }, [router]);
+
+    const openDetail = useCallback(
+      async (logEntry) => {
+        // Guard: if no valid id provided, close instead of opening an empty modal
+        if (!logEntry?.id) {
+          try {
+            closeDetail();
+          } catch {}
+          return;
+        }
+
+        const requestToken = `${logEntry.id}:${Date.now()}:${Math.random()}`;
+        detailRequestRef.current = requestToken;
+        const isCurrentDetailRequest = () => detailRequestRef.current === requestToken;
+
+        setSelectedLog(logEntry);
+        try {
+          const url = new URL(globalThis.location.href);
+          url.searchParams.set("id", logEntry.id);
+          router.replace(url.pathname + url.search, { scroll: false });
+        } catch (e) {
+          // ignore navigation errors
+        }
+        setDetailLoading(true);
+        setDetailData(null);
+        try {
+          const res = await fetch(`/api/logs/${logEntry.id}`, { cache: "no-store" });
+          if (res.ok) {
+            const data = await res.json();
+            if (!isCurrentDetailRequest()) return;
+            const dataHasPipeline =
+              data?.pipelinePayloads && Object.keys(data.pipelinePayloads || {}).length > 0;
+            setDetailData((prev: { pipelinePayloads: any }) => ({
+              ...prev,
+              ...data,
+              pipelinePayloads: dataHasPipeline ? data.pipelinePayloads : prev?.pipelinePayloads,
+            }));
+            // ensure the modal summary reflects the fetched call log summary
+            if (data && typeof data === "object") {
+              setSelectedLog((prev: any) => ({
+                ...prev,
+                ...data,
+                active: data.active === true,
+              }));
+            }
+          } else {
+            // A deep-linked id can legitimately 404 while the request is still
+            // finalizing. Keep the modal open and poll /api/logs/[id] instead of
+            // falling back to an in-memory active-request endpoint.
+            if (!isCurrentDetailRequest()) return;
+            if (res.status === 404) {
+              if (logEntry.pendingLookup || logEntry.active) {
+                setSelectedLog((prev: { method: any; path: any }) => ({
+                  ...prev,
+                  id: logEntry.id,
+                  status: 0,
+                  method: prev?.method,
+                  path: prev?.path || "",
+                }));
+                setDetailData({ detailState: "pending" });
+                return;
+              }
+              try {
+                console.warn("Log not found:", logEntry.id);
+              } catch {}
+              try {
+                closeDetail();
+              } catch {}
+              return;
+            }
+            // other errors: show a minimal error indicator by setting detailData to an error object
+            try {
+              const body = await res.text().catch(() => null);
+              if (!isCurrentDetailRequest()) return;
+              setDetailData({ error: `Failed to fetch log (status ${res.status})`, body });
+            } catch {}
+          }
+        } catch (error) {
+          console.error("Failed to fetch log detail:", error);
+        } finally {
+          if (isCurrentDetailRequest()) setDetailLoading(false);
+        }
+      },
+      [closeDetail, router]
+    );
 
     const sortedLogsForNav = useMemo(() => sortedLogs, [sortedLogs]);
 
@@ -654,7 +658,7 @@ const RequestLoggerV2 = forwardRef<RequestLoggerV2Handle, { initialSelectedId?:
             console.error("Failed to open initial log id:", error_);
           });
       }
-    }, [initialSelectedId]);
+    }, [initialSelectedId, openDetail]);
 
     useEffect(() => {
       const isActive = selectedLog?.active === true;
@@ -765,7 +769,7 @@ const RequestLoggerV2 = forwardRef<RequestLoggerV2Handle, { initialSelectedId?:
         pendingBoundaryNavRef.current = "prev";
         fetchLogs(false);
       }
-    }, [currentLogIndex, sortedLogsForNav, fetchLogs]);
+    }, [currentLogIndex, sortedLogsForNav, fetchLogs, openDetail]);
 
     const handleNext = useCallback(() => {
       const idx = currentLogIndex;
@@ -781,7 +785,7 @@ const RequestLoggerV2 = forwardRef<RequestLoggerV2Handle, { initialSelectedId?:
         pendingBoundaryNavRef.current = "next";
         fetchLogs(false);
       }
-    }, [currentLogIndex, sortedLogsForNav, fetchLogs]);
+    }, [currentLogIndex, sortedLogsForNav, fetchLogs, openDetail]);
 
     // Resolves a pending boundary nav (see handlePrev/handleNext) once a
     // triggered fetchLogs() resync has landed in sortedLogsForNav. Only fires

src/shared/constants/routingStrategies.ts

@@ -37,6 +37,7 @@ export type AnyRoutingStrategyValue = RoutingStrategyValue | InternalRoutingStra
 
 export const AUTO_ROUTING_STRATEGY_VALUES = [
   "rules",
+  "score",
   "cost",
   "eco",
   "latency",

src/shared/constants/statusColors.ts

@@ -14,4 +14,6 @@ export const STATUS_HEX = {
   warning: "#f59e0b",
   error: "#ef4444",
   muted: "#6b7280",
+  /** Human-approval gate (waiting_approval) — violet, matching the industry de-facto palette. */
+  approval: "#8b5cf6",
 } as const;

src/shared/utils/httpClientAbortGuard.mjs

@@ -42,6 +42,13 @@ export function isClientAbortError(err) {
   const e = /** @type {NodeJS.ErrnoException} */ (err);
   // Node emits `Error: aborted` (no code) from http.Server#abortIncoming.
   if (e.message === "aborted" || e.message === "Aborted") return true;
+  // OmniRoute's SSE teardown aborts in-flight legs with
+  // `Error [AbortError]: request_signal_aborted` on client disconnects
+  // (open-sse/utils/streamHandler.ts), and fetch/DOM cancellation surfaces as
+  // `AbortError` with an abort-flavoured message. Same benign class as
+  // `Error: aborted` — an emitter-left 'error' event on any of these used to
+  // kill the process (#fix-dev-server-aborted).
+  if (e.name === "AbortError" && /abort/i.test(String(e.message))) return true;
   switch (e.code) {
     case "ERR_STREAM_PREMATURE_CLOSE":
     case "ECONNRESET":

src/shared/utils/logger.ts

@@ -18,6 +18,10 @@ import { resolve } from "path";
 import { getLogConfig, initLogRotation } from "@/lib/logRotation";
 import { getAppLogLevel } from "@/lib/logEnv";
 import { redactLogArgs } from "@/shared/utils/logRedaction";
+import {
+  getOrCreateSharedLoggerResource,
+  type SharedLoggerResource,
+} from "@/shared/utils/loggerResource";
 
 const isDev = process.env.NODE_ENV !== "production";
 
@@ -61,7 +65,7 @@ function getTransportCompatibleConfig(): pino.LoggerOptions {
  * vanished, so failed writes are dropped (best-effort stderr notice) instead of
  * escalating.
  */
-function buildFileTransportStream(targets: NonNullable<pino.TransportMultiOptions["targets"]>) {
+function buildTransportStream(targets: NonNullable<pino.TransportMultiOptions["targets"]>) {
   const stream = pino.transport({ targets });
   stream.on("error", (err: unknown) => {
     try {
@@ -75,11 +79,65 @@ function buildFileTransportStream(targets: NonNullable<pino.TransportMultiOption
   return stream;
 }
 
+interface OwnedLogStream {
+  flushSync?: () => void;
+  end?: () => void;
+  once?: (event: string, listener: () => void) => unknown;
+}
+
+async function closeOwnedStream(stream: OwnedLogStream | null): Promise<void> {
+  if (!stream) return;
+
+  try {
+    stream.flushSync?.();
+  } catch {
+    // Best-effort shutdown: a missing log destination must not block process exit.
+  }
+
+  if (!stream.end) return;
+
+  await new Promise<void>((resolveClose) => {
+    let resolved = false;
+    const finish = () => {
+      if (resolved) return;
+      resolved = true;
+      resolveClose();
+    };
+    const fallback = setTimeout(finish, 1_000);
+
+    stream.once?.("close", () => {
+      clearTimeout(fallback);
+      finish();
+    });
+
+    try {
+      stream.end?.();
+      if (!stream.once) {
+        clearTimeout(fallback);
+        finish();
+      }
+    } catch {
+      clearTimeout(fallback);
+      finish();
+    }
+  });
+}
+
+function createLoggerResource(
+  logger: pino.Logger,
+  stream: OwnedLogStream | null
+): SharedLoggerResource {
+  return {
+    logger,
+    close: () => closeOwnedStream(stream),
+  };
+}
+
 /**
  * Build the logger with optional file transport.
  * Uses pino transport targets for all destinations.
  */
-function buildLogger(): pino.Logger {
+function buildLoggerResource(): SharedLoggerResource {
   const logConfig = getLogConfig();
   const logLevel = (baseConfig.level as string) || "info";
   const transportConfig = getTransportCompatibleConfig();
@@ -95,7 +153,7 @@ function buildLogger(): pino.Logger {
 
       if (isDev) {
         // Dev: pino-pretty → stdout, JSON → file
-        const stream = buildFileTransportStream([
+        const stream = buildTransportStream([
           {
             target: "pino-pretty",
             options: {
@@ -113,12 +171,12 @@ function buildLogger(): pino.Logger {
             level: logLevel,
           },
         ]);
-        return pino(transportConfig, stream);
+        return createLoggerResource(pino(transportConfig, stream), stream);
       }
 
       // Production: JSON → stdout + JSON → file
       {
-        const stream = buildFileTransportStream([
+        const stream = buildTransportStream([
           {
             target: "pino/file",
             options: { destination: 1 }, // stdout
@@ -130,7 +188,7 @@ function buildLogger(): pino.Logger {
             level: logLevel,
           },
         ]);
-        return pino(transportConfig, stream);
+        return createLoggerResource(pino(transportConfig, stream), stream);
       }
     } catch (err) {
       // Log the actual error for diagnostics (issue #165)
@@ -156,12 +214,15 @@ function buildLogger(): pino.Logger {
         });
 
         // Production fallback: JSON to both stdout and file via multistream
-        return pino(
-          baseConfig,
-          pino.multistream([
-            { stream: process.stdout, level: logLevel as pino.Level },
-            { stream: fileDestination, level: logLevel as pino.Level },
-          ])
+        return createLoggerResource(
+          pino(
+            baseConfig,
+            pino.multistream([
+              { stream: process.stdout, level: logLevel as pino.Level },
+              { stream: fileDestination, level: logLevel as pino.Level },
+            ])
+          ),
+          fileDestination
         );
       } catch (fallbackErr) {
         try {
@@ -175,9 +236,8 @@ function buildLogger(): pino.Logger {
 
   // Console-only (no file logging)
   if (isDev) {
-    return pino({
-      ...baseConfig,
-      transport: {
+    const stream = buildTransportStream([
+      {
         target: "pino-pretty",
         options: {
           colorize: true,
@@ -185,14 +245,18 @@ function buildLogger(): pino.Logger {
           ignore: "pid,hostname,service",
           messageFormat: "[{module}] {msg}",
         },
+        level: logLevel,
       },
-    });
+    ]);
+    return createLoggerResource(pino(transportConfig, stream), stream);
   }
 
-  return pino(baseConfig);
+  return createLoggerResource(pino(baseConfig), null);
 }
 
-export const logger = buildLogger();
+const sharedLoggerResource = getOrCreateSharedLoggerResource(buildLoggerResource);
+
+export const logger = sharedLoggerResource.logger;
 
 /**
  * Create a child logger with a module tag.

src/shared/utils/loggerResource.ts

@@ -0,0 +1,32 @@
+import type { Logger } from "pino";
+
+export interface SharedLoggerResource {
+  logger: Logger;
+  close: () => Promise<void>;
+}
+
+declare global {
+  var __omnirouteLoggerResource: SharedLoggerResource | undefined;
+}
+
+/**
+ * Return the process-wide logger resource, creating it only once.
+ *
+ * Next.js development HMR can evaluate logger.ts in more than one server chunk.
+ * Keeping the resource on globalThis prevents each evaluation from spawning a
+ * new pino worker transport.
+ */
+export function getOrCreateSharedLoggerResource(
+  create: () => SharedLoggerResource
+): SharedLoggerResource {
+  return (globalThis.__omnirouteLoggerResource ??= create());
+}
+
+/** Close and forget the shared transport. Idempotent across HMR module copies. */
+export async function closeSharedLoggerResource(): Promise<void> {
+  const resource = globalThis.__omnirouteLoggerResource;
+  if (!resource) return;
+
+  delete globalThis.__omnirouteLoggerResource;
+  await resource.close();
+}

tests/unit/11759-embedding-registry-width-and-type.test.ts

@@ -0,0 +1,155 @@
+/**
+ * Regression test for #11759 — `/v1/models` dropped the vector width and the
+ * `embedding` type from embedding models that `embeddingRegistry.ts` describes,
+ * whenever a synced model existed for the same id.
+ *
+ * `embeddingRegistry.ts` is the only machine-readable source of two facts: a model's
+ * vector width, and that it is an embedding model at all. No upstream provider
+ * publishes either — OpenRouter's catalogue carries no `dimensions`, and OpenAI's
+ * `/v1/models` returns ids with no capability information.
+ *
+ * Two paths lost them:
+ *
+ *  - Width. The registry loop skipped its entry outright when discovery had already
+ *    produced the model (`hasEquivalentSpecialtyModel(...) continue`), so the surviving
+ *    entry was the discovered one — typed, but with no `dimensions`.
+ *
+ *  - Type. Classification comes from `sm.supportedEndpoints`, which falls back to
+ *    `["chat"]`. A provider that does not report endpoints therefore yielded
+ *    `modelType === undefined` and the `type` key was omitted entirely, so an embedding
+ *    model read as a chat model.
+ *
+ * A consumer that stores vectors needs both: collections are keyed on width, and an
+ * untyped model cannot be identified as an embedding model. These tests assert the
+ * merged `/v1/models` output, not the registry in isolation — the registry was always
+ * correct; the loss happened in `catalog.ts`.
+ */
+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-11759-"));
+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 modelsDb = await import("../../src/lib/db/models.ts");
+const v1ModelsCatalog = await import("../../src/app/api/v1/models/catalog.ts");
+const embeddingRegistry = await import("../../open-sse/config/embeddingRegistry.ts");
+
+/** Both are real registry entries, so the expected widths come from the registry itself. */
+const OPENROUTER_MODEL = "qwen/qwen3-embedding-8b";
+const OPENAI_MODEL = "text-embedding-3-small";
+
+function registryWidth(providerId: string, modelId: string): number {
+  const provider = embeddingRegistry.getEmbeddingProvider(providerId);
+  assert.ok(provider, `embeddingRegistry has no provider "${providerId}"`);
+  const model = provider!.models.find((m) => m.id === modelId);
+  assert.ok(model, `embeddingRegistry has no model "${modelId}" on "${providerId}"`);
+  const { dimensions } = model!;
+  assert.equal(
+    typeof dimensions,
+    "number",
+    `this test assumes a single advertised width for "${modelId}"`
+  );
+  return dimensions as number;
+}
+
+async function resetStorage() {
+  core.resetDbInstance();
+  fs.rmSync(TEST_DATA_DIR, { recursive: true, force: true });
+  fs.mkdirSync(TEST_DATA_DIR, { recursive: true });
+  v1ModelsCatalog.__resetCatalogBuilderRunsForTest();
+}
+
+/** An active connection, which is what makes the provider's registry models eligible. */
+async function connectProvider(provider: string) {
+  return (await providersDb.createProviderConnection({
+    provider,
+    authType: "apikey",
+    name: `${provider}-conn`,
+    apiKey: "sk-test",
+    isActive: true,
+    testStatus: "active",
+  })) as { id: string };
+}
+
+async function modelEntry(id: string) {
+  const response = await v1ModelsCatalog.getUnifiedModelsResponse(
+    new Request("http://localhost/api/v1/models")
+  );
+  assert.equal(response.status, 200);
+  const body = (await response.json()) as { data: Array<Record<string, unknown>> };
+  const entry = body.data.find((model) => model.id === id);
+  assert.ok(
+    entry,
+    `expected "${id}" in /v1/models, got ${JSON.stringify(
+      body.data.filter((m) => String(m.id).includes("embedding")).map((m) => m.id)
+    )}`
+  );
+  return entry!;
+}
+
+test.beforeEach(async () => {
+  await resetStorage();
+});
+
+test.after(async () => {
+  core.resetDbInstance();
+  fs.rmSync(TEST_DATA_DIR, { recursive: true, force: true });
+});
+
+test("#11759: a synced embedding model keeps the width the registry states", async () => {
+  const connection = await connectProvider("openrouter");
+
+  // What discovery produces for OpenRouter: typed as an embedding model, and carrying no
+  // width, because the upstream catalogue does not publish one.
+  await modelsDb.replaceSyncedAvailableModelsForConnection("openrouter", connection.id, [
+    {
+      id: OPENROUTER_MODEL,
+      name: "Qwen3 Embedding 8B",
+      source: "imported",
+      supportedEndpoints: ["embeddings"],
+    },
+  ]);
+
+  const entry = await modelEntry(`openrouter/${OPENROUTER_MODEL}`);
+  const expected = registryWidth("openrouter", OPENROUTER_MODEL);
+
+  assert.equal(
+    entry.dimensions,
+    expected,
+    `the registry states ${expected} for "${OPENROUTER_MODEL}"; the synced entry must not drop it — got ${JSON.stringify(entry.dimensions)}`
+  );
+  assert.equal(entry.type, "embedding", "a synced embedding model must stay typed as one");
+});
+
+test("#11759: a synced model the embedding registry names is typed as an embedding model", async () => {
+  const connection = await connectProvider("openai");
+
+  // What discovery produces for OpenAI: no `supportedEndpoints`, because `/v1/models`
+  // returns ids with no capability information, so classification falls back to `["chat"]`
+  // and the model is emitted with no `type` at all.
+  await modelsDb.replaceSyncedAvailableModelsForConnection("openai", connection.id, [
+    {
+      id: OPENAI_MODEL,
+      name: "Text Embedding 3 Small",
+      source: "imported",
+    },
+  ]);
+
+  const entry = await modelEntry(`openai/${OPENAI_MODEL}`);
+
+  assert.equal(
+    entry.type,
+    "embedding",
+    `"${OPENAI_MODEL}" is in the embedding registry, so it must not be published untyped — got ${JSON.stringify(entry.type)}`
+  );
+  assert.equal(
+    entry.dimensions,
+    registryWidth("openai", OPENAI_MODEL),
+    "an embedding model published without its width cannot be indexed by a consumer"
+  );
+});

tests/unit/batch-dispatch-instrumentation-boundary-12074.test.ts

@@ -0,0 +1,69 @@
+import assert from "node:assert/strict";
+import { readFileSync } from "node:fs";
+import { join } from "node:path";
+import test from "node:test";
+
+const dispatchSourcePath = join(import.meta.dirname, "../../src/lib/batches/dispatch.ts");
+
+const originalFetch = globalThis.fetch;
+const originalEnv = {
+  OMNIROUTE_PORT: process.env.OMNIROUTE_PORT,
+  PORT: process.env.PORT,
+  DASHBOARD_PORT: process.env.DASHBOARD_PORT,
+  OMNIROUTE_BASE_PATH: process.env.OMNIROUTE_BASE_PATH,
+};
+
+function restoreEnv(): void {
+  for (const key of Object.keys(originalEnv) as Array<keyof typeof originalEnv>) {
+    const value = originalEnv[key];
+    if (value === undefined) {
+      delete process.env[key];
+    } else {
+      process.env[key] = value;
+    }
+  }
+}
+
+test.afterEach(() => {
+  globalThis.fetch = originalFetch;
+  restoreEnv();
+});
+
+test("batch dispatch does not pull API route modules into the instrumentation graph", () => {
+  const source = readFileSync(dispatchSourcePath, "utf8");
+
+  assert.doesNotMatch(source, /@\/app\/api\/v1\/.+\/route/);
+  assert.doesNotMatch(source, /handlerLoaders|BatchRouteHandler/);
+});
+
+test("batch dispatch posts to the active dashboard loopback listener", async () => {
+  process.env.OMNIROUTE_PORT = "24120";
+  process.env.PORT = "24121";
+  process.env.DASHBOARD_PORT = "24122";
+  process.env.OMNIROUTE_BASE_PATH = "/omniroute/";
+
+  const calls: Array<{ input: string; init?: RequestInit }> = [];
+  const upstreamResponse = new Response("accepted", { status: 202 });
+  globalThis.fetch = async (input, init) => {
+    calls.push({ input: String(input), init });
+    return upstreamResponse;
+  };
+
+  const { dispatch } = await import("../../src/lib/batches/dispatch.ts");
+  const response = await dispatch.dispatchBatchApiRequest({
+    endpoint: "/v1/chat/completions",
+    body: { model: "provider/model", messages: [] },
+    apiKey: "batch-secret",
+  });
+
+  assert.strictEqual(response, upstreamResponse);
+  assert.equal(calls.length, 1);
+  assert.equal(calls[0].input, "http://127.0.0.1:24122/omniroute/v1/chat/completions");
+  assert.equal(calls[0].init?.method, "POST");
+  assert.equal(calls[0].init?.redirect, "error");
+  assert.equal(new Headers(calls[0].init?.headers).get("authorization"), "Bearer batch-secret");
+  assert.deepEqual(JSON.parse(String(calls[0].init?.body)), {
+    model: "provider/model",
+    messages: [],
+  });
+});

tests/unit/batch_api.test.ts

@@ -202,6 +202,13 @@ test("Batch API and Processing", async () => {
 });
 
 test("Batch handles and counts failures correctly", async () => {
+  const originalFetch = globalThis.fetch;
+  globalThis.fetch = async () =>
+    new Response(JSON.stringify({ error: { message: "Model not found" } }), {
+      status: 404,
+      headers: { "Content-Type": "application/json" },
+    });
+
   initBatchProcessor();
   try {
     // 1. Create a file with a request that will fail (invalid provider/model)
@@ -264,6 +271,7 @@ test("Batch handles and counts failures correctly", async () => {
     }
   } finally {
     stopBatchProcessor();
+    globalThis.fetch = originalFetch;
   }
 });
 
@@ -390,6 +398,21 @@ test("Batch rejects input lines whose url does not match the batch endpoint", as
 });
 
 test("Batch forces stream: false for all requests", async () => {
+  const originalFetch = globalThis.fetch;
+  let dispatchedBody: Record<string, unknown> | null = null;
+  globalThis.fetch = async (_input, init) => {
+    dispatchedBody = JSON.parse(String(init?.body));
+    return new Response(
+      JSON.stringify({
+        choices: [{ message: { role: "assistant", content: "batch response" } }],
+      }),
+      {
+        status: 200,
+        headers: { "Content-Type": "application/json" },
+      }
+    );
+  };
+
   initBatchProcessor();
   try {
     const batchItems = [
@@ -449,8 +472,10 @@ test("Batch forces stream: false for all requests", async () => {
         "Should not have JSON parsing error from SSE stream"
       );
     }
+    assert.strictEqual(dispatchedBody?.stream, false, "Batch dispatch must disable streaming");
   } finally {
     stopBatchProcessor();
+    globalThis.fetch = originalFetch;
   }
 });
 

tests/unit/codex-quota-sync-no-proactive-refresh.test.ts

@@ -27,11 +27,12 @@ import path from "node:path";
 
 process.env.DATA_DIR = fs.mkdtempSync(path.join(os.tmpdir(), "omniroute-codex-quota-"));
 
-const { getExecutor } = await import("../../open-sse/executors/index.ts");
+const { getCredentialRefreshExecutor } =
+  await import("../../open-sse/executors/credential.ts");
 const { refreshAndUpdateCredentials } = await import("../../src/lib/usage/providerLimits.ts");
 
 test("codex: quota-sync must NOT proactively rotate the refresh_token (Auth0 family-revocation cascade guard)", async () => {
-  const exec = await getExecutor("codex");
+  const exec = await getCredentialRefreshExecutor("codex");
   const origNeeds = exec.needsRefresh;
   const origRefresh = exec.refreshCredentials;
   let refreshCalls = 0;
@@ -68,7 +69,7 @@ test("codex: quota-sync must NOT proactively rotate the refresh_token (Auth0 fam
 });
 
 test("non-rotating OAuth provider is still refreshed proactively from quota-sync (gate is not over-broad)", async () => {
-  const exec = await getExecutor("cursor");
+  const exec = await getCredentialRefreshExecutor("cursor");
   const origNeeds = exec.needsRefresh;
   const origRefresh = exec.refreshCredentials;
   let refreshCalls = 0;

tests/unit/executor-registry.test.ts

@@ -15,6 +15,7 @@ const { registerExecutor, getRegisteredExecutor, hasRegisteredExecutor, listExec
   await import("../../open-sse/executors/registry.ts");
 const { getExecutor, hasSpecializedExecutor, BaseExecutor, DefaultExecutor } =
   await import("../../open-sse/executors/index.ts");
+const { getDefaultExecutor } = await import("../../open-sse/executors/defaultResolver.ts");
 
 test.after(() => {
   fs.rmSync(TEST_DATA_DIR, { recursive: true, force: true, maxRetries: 5, retryDelay: 100 });
@@ -54,3 +55,9 @@ test("registry lookup is exact — Object.prototype names are not executors", as
     assert.ok((await getExecutor(name)) instanceof DefaultExecutor, name);
   }
 });
+
+test("the registry and leaf resolver share fallback executor instances", async () => {
+  const provider = "default-resolver-test-provider";
+  assert.equal(await getExecutor(provider), getDefaultExecutor(provider));
+  assert.equal(getDefaultExecutor(provider), getDefaultExecutor(provider));
+});

tests/unit/graceful-shutdown-sighup-8045.test.ts

@@ -1,5 +1,12 @@
 import { test } from "node:test";
 import assert from "node:assert/strict";
+import { join } from "node:path";
+import { pathToFileURL } from "node:url";
+
+type GracefulShutdownModule = typeof import("../../src/lib/gracefulShutdown.ts");
+
+const gracefulShutdownUrl = pathToFileURL(join(process.cwd(), "src/lib/gracefulShutdown.ts")).href;
+const shutdownSignals = ["SIGTERM", "SIGINT", "SIGHUP"] as const;
 
 // #8045: on Windows, closing the console window delivers CTRL_CLOSE_EVENT, which
 // Node/libuv maps to a JS-visible "SIGHUP" event (confirmed: nodejs/node#10165,
@@ -7,17 +14,90 @@ import assert from "node:assert/strict";
 // closed"). Before this fix, initGracefulShutdown() only registered SIGTERM/SIGINT,
 // so the "close the window" path never ran cleanup() (WAL checkpoint(TRUNCATE) +
 // closeDbInstance()), leaving storage.sqlite's WAL un-checkpointed for the next launch.
-test("initGracefulShutdown registers a SIGHUP handler (Windows console-close path)", async () => {
-  const before = process.listenerCount("SIGHUP");
-  const { initGracefulShutdown } = await import("../../src/lib/gracefulShutdown.ts");
-  initGracefulShutdown();
-  const after = process.listenerCount("SIGHUP");
-  assert.ok(
-    after > before,
-    `Expected initGracefulShutdown() to add a SIGHUP listener (before=${before}, after=${after}).`
+test("graceful shutdown listeners remain process-singletons across HMR module instances", async () => {
+  const previousState = globalThis.__omnirouteShutdown;
+  const previousRequestShutdown = globalThis.__omnirouteRequestShutdown;
+  const previousCustomServerOwner = globalThis.__omnirouteCustomServerOwnsShutdown;
+  const listenersBefore = new Map(
+    shutdownSignals.map((signal) => [signal, process.listeners(signal)] as const)
+  );
+  delete globalThis.__omnirouteShutdown;
+  delete globalThis.__omnirouteRequestShutdown;
+  delete globalThis.__omnirouteCustomServerOwnsShutdown;
+
+  try {
+    const first = (await import(
+      `${gracefulShutdownUrl}?phase4=shutdown-a`
+    )) as GracefulShutdownModule;
+    const second = (await import(
+      `${gracefulShutdownUrl}?phase4=shutdown-b`
+    )) as GracefulShutdownModule;
+
+    first.initGracefulShutdown();
+    assert.equal(globalThis.__omnirouteRequestShutdown, first.requestGracefulShutdown);
+    for (const signal of shutdownSignals) {
+      assert.equal(process.listenerCount(signal), listenersBefore.get(signal)!.length + 1);
+    }
+
+    second.initGracefulShutdown();
+    for (const signal of shutdownSignals) {
+      assert.equal(
+        process.listenerCount(signal),
+        listenersBefore.get(signal)!.length + 1,
+        `${signal} listener must not be duplicated by HMR re-initialization`
+      );
+    }
+  } finally {
+    for (const signal of shutdownSignals) {
+      const previousListeners = listenersBefore.get(signal)!;
+      for (const listener of process.listeners(signal)) {
+        if (!previousListeners.includes(listener)) process.removeListener(signal, listener);
+      }
+    }
+
+    if (previousState === undefined) delete globalThis.__omnirouteShutdown;
+    else globalThis.__omnirouteShutdown = previousState;
+    if (previousRequestShutdown === undefined) delete globalThis.__omnirouteRequestShutdown;
+    else globalThis.__omnirouteRequestShutdown = previousRequestShutdown;
+    if (previousCustomServerOwner === undefined) {
+      delete globalThis.__omnirouteCustomServerOwnsShutdown;
+    } else {
+      globalThis.__omnirouteCustomServerOwnsShutdown = previousCustomServerOwner;
+    }
+  }
+});
+
+test("a custom server owner receives cleanup without duplicate process signal listeners", async () => {
+  const previousState = globalThis.__omnirouteShutdown;
+  const previousRequestShutdown = globalThis.__omnirouteRequestShutdown;
+  const previousCustomServerOwner = globalThis.__omnirouteCustomServerOwnsShutdown;
+  const listenerCounts = new Map(
+    shutdownSignals.map((signal) => [signal, process.listenerCount(signal)] as const)
   );
 
-  // Clean up: remove all SIGHUP listeners added by this test so it doesn't leak
-  // into other test files sharing the same process.
-  process.removeAllListeners("SIGHUP");
+  delete globalThis.__omnirouteShutdown;
+  delete globalThis.__omnirouteRequestShutdown;
+  globalThis.__omnirouteCustomServerOwnsShutdown = true;
+
+  try {
+    const shutdownModule = (await import(
+      `${gracefulShutdownUrl}?phase4=custom-owner`
+    )) as GracefulShutdownModule;
+    shutdownModule.initGracefulShutdown();
+
+    assert.equal(globalThis.__omnirouteRequestShutdown, shutdownModule.requestGracefulShutdown);
+    for (const signal of shutdownSignals) {
+      assert.equal(process.listenerCount(signal), listenerCounts.get(signal));
+    }
+  } finally {
+    if (previousState === undefined) delete globalThis.__omnirouteShutdown;
+    else globalThis.__omnirouteShutdown = previousState;
+    if (previousRequestShutdown === undefined) delete globalThis.__omnirouteRequestShutdown;
+    else globalThis.__omnirouteRequestShutdown = previousRequestShutdown;
+    if (previousCustomServerOwner === undefined) {
+      delete globalThis.__omnirouteCustomServerOwnsShutdown;
+    } else {
+      globalThis.__omnirouteCustomServerOwnsShutdown = previousCustomServerOwner;
+    }
+  }
 });

tests/unit/httpClientAbortGuard.test.mjs

@@ -3,6 +3,8 @@
 import assert from "node:assert";
 import { test } from "node:test";
 import { EventEmitter } from "node:events";
+import { spawn } from "node:child_process";
+import { fileURLToPath } from "node:url";
 import {
   isClientAbortError,
   shouldSwallowUncaught,
@@ -115,5 +117,88 @@ test("shouldSwallowUncaught preserves crash semantics for genuine errors", () =>
 
 test("installProcessCrashGuard does not throw on import and is idempotent", () => {
   assert.doesNotThrow(() => installProcessCrashGuard(() => {}));
-  assert.doesNotThrow(() => installProcessCrashGuard(() => {}));
+});
+test("isClientAbortError matches OmniRoute SSE AbortError shapes (#fix-crash-guard-logger-7)", () => {
+  // Exact production shape from the 2026-08-31 crash log:
+  //   ⨯ unhandledRejection: Error [AbortError]: request_signal_aborted
+  const sseAbort = Object.assign(new Error("request_signal_aborted"), { name: "AbortError" });
+  assert.equal(isClientAbortError(sseAbort), true, "SSE teardown AbortError must be absorbed");
+  // fetch / DOMException-style cancellation
+  const domAbort = new DOMException("This operation was aborted", "AbortError");
+  assert.equal(isClientAbortError(domAbort), true, "DOMException AbortError must be absorbed");
+  // A genuine TypeError that merely MENTIONS 'abort' must NOT be absorbed.
+  const typo = new TypeError("Cannot read properties of undefined (reading 'abort')");
+  assert.equal(isClientAbortError(typo), false);
+});
+
+test("shouldSwallowUncaught absorbs SSE AbortError rejections", () => {
+  const sseAbort = Object.assign(new Error("request_signal_aborted"), { name: "AbortError" });
+  assert.equal(shouldSwallowUncaught(sseAbort, "unhandledRejection"), true);
+});
+
+// Production crash (2026-08-25 → 08-31, ~170 restarts, exit code 7):
+// every real call site installs the guard with NO logger, so the old
+// `const logger = log ?? console` default invoked the console OBJECT as a
+// function inside the uncaughtException handler → TypeError inside
+// process._fatalException → Node exit code 7. These children run the REAL
+// production call shape; the process must survive benign aborts and still
+// crash on genuine errors.
+test("installProcessCrashGuard() with no logger swallows aborts instead of dying (exit-7 regression)", async () => {
+  const guardPath = fileURLToPath(
+    new URL("../../src/shared/utils/httpClientAbortGuard.mjs", import.meta.url)
+  );
+  const script = `
+    const { installProcessCrashGuard } = await import(process.argv[1]);
+    installProcessCrashGuard(); // production call sites pass NO logger
+    process.emit(
+      "uncaughtException",
+      Object.assign(new Error("aborted"), { code: "ECONNRESET" }),
+      "uncaughtException"
+    );
+    process.emit(
+      "unhandledRejection",
+      Object.assign(new Error("request_signal_aborted"), { name: "AbortError" }),
+      Promise.resolve()
+    );
+    console.log("ALIVE");
+    process.exit(0);
+  `;
+  const { status, stdout, stderr } = await new Promise((resolve, reject) => {
+    const child = spawn(process.execPath, ["--input-type=module", "-e", script, guardPath], {
+      stdio: ["ignore", "pipe", "pipe"],
+    });
+    let out = "";
+    let err = "";
+    child.stdout.on("data", (d) => (out += d));
+    child.stderr.on("data", (d) => (err += d));
+    child.on("close", (status) => resolve({ status, stdout: out, stderr: err }));
+    child.on("error", reject);
+  });
+  assert.equal(status, 0, `child must survive benign aborts; stderr: ${stderr}`);
+  assert.match(stdout, /ALIVE/);
+});
+
+test("installProcessCrashGuard still crashes on genuine errors (no over-swallowing)", async () => {
+  const guardPath = fileURLToPath(
+    new URL("../../src/shared/utils/httpClientAbortGuard.mjs", import.meta.url)
+  );
+  const script = `
+    const { installProcessCrashGuard } = await import(process.argv[1]);
+    installProcessCrashGuard();
+    process.emit("uncaughtException", new Error("genuine failure"), "uncaughtException");
+    console.log("SHOULD_NOT_REACH");
+  `;
+  const { status, stdout, stderr: _stderr } = await new Promise((resolve, reject) => {
+    const child = spawn(process.execPath, ["--input-type=module", "-e", script, guardPath], {
+      stdio: ["ignore", "pipe", "pipe"],
+    });
+    let out = "";
+    let err = "";
+    child.stdout.on("data", (d) => (out += d));
+    child.stderr.on("data", (d) => (err += d));
+    child.on("close", (status) => resolve({ status, stdout: out, stderr: err }));
+    child.on("error", reject);
+  });
+  assert.notEqual(status, 0, "genuine errors must keep crash semantics");
+  assert.doesNotMatch(stdout, /SHOULD_NOT_REACH/);
 });

tests/unit/instrumentation-import-graph-12074.test.ts

@@ -0,0 +1,107 @@
+import assert from "node:assert/strict";
+import fs from "node:fs";
+import path from "node:path";
+import test from "node:test";
+
+const instrumentationPath = path.join(process.cwd(), "src/instrumentation-node.ts");
+const quotaAutoPingPath = path.join(process.cwd(), "src/lib/services/quotaAutoPing.ts");
+const credentialRefreshPath = path.join(
+  process.cwd(),
+  "src/lib/usage/providerLimits/credentialRefresh.ts"
+);
+const providerLimitsPath = path.join(process.cwd(), "src/lib/usage/providerLimits.ts");
+const credentialExecutorPath = path.join(process.cwd(), "open-sse/executors/credential.ts");
+const executorDirectory = path.join(process.cwd(), "open-sse/executors");
+const anthropicValidationPath = path.join(
+  process.cwd(),
+  "src/lib/providers/validation/anthropicFormat.ts"
+);
+const defaultExecutorResolverPath = path.join(
+  process.cwd(),
+  "open-sse/executors/defaultResolver.ts"
+);
+
+test("node instrumentation loads the proxy patch leaf before quota registration", () => {
+  const source = fs.readFileSync(instrumentationPath, "utf8");
+  const proxyPatchImport = 'await import("@omniroute/open-sse/utils/proxyFetch.ts")';
+  const proxyPatchIndex = source.indexOf(proxyPatchImport);
+  const quotaRegistrationIndex = source.indexOf("await registerQuotaFetchers()");
+
+  assert.ok(proxyPatchIndex >= 0, "startup must load the proxyFetch side-effect leaf");
+  assert.ok(quotaRegistrationIndex > proxyPatchIndex, "proxy patch must run before quota setup");
+  assert.doesNotMatch(source, /import\("@omniroute\/open-sse\/index\.ts"\)/);
+});
+
+test("quota auto-ping lazily loads only the Codex executor", () => {
+  const source = fs.readFileSync(quotaAutoPingPath, "utf8");
+  const credentialRefreshSource = fs.readFileSync(credentialRefreshPath, "utf8");
+
+  assert.doesNotMatch(source, /open-sse\/executors\/index(?:\.ts)?/);
+  assert.doesNotMatch(source, /@\/lib\/usage\/providerLimits["']/);
+  assert.match(source, /import\("@omniroute\/open-sse\/executors\/codex\.ts"\)/);
+  assert.match(source, /@\/lib\/usage\/providerLimits\/credentialRefresh/);
+  assert.match(source, /getExecutor: loadQuotaAutoPingExecutor/);
+  assert.doesNotMatch(credentialRefreshSource, /open-sse\/executors\/index(?:\.ts)?/);
+});
+
+test("startup still registers bespoke, batch, and generic quota fetchers", async () => {
+  const [{ registerQuotaFetchers }, { getQuotaFetcher }] = await Promise.all([
+    import("../../src/instrumentation-node.ts"),
+    import("../../open-sse/services/quotaPreflight.ts"),
+  ]);
+
+  await registerQuotaFetchers();
+
+  for (const provider of [
+    "agentrouter",
+    "codex",
+    "bailian-coding-plan",
+    "qwen-cloud-token-plan",
+    "crof",
+    "deepseek",
+    "openrouter",
+    "opencode-go",
+    "grok-web",
+    "antigravity",
+  ]) {
+    assert.equal(typeof getQuotaFetcher(provider), "function", `${provider} quota fetcher missing`);
+  }
+});
+
+test("provider-limit startup uses the refresh-only executor resolver", () => {
+  const providerLimitsSource = fs.readFileSync(providerLimitsPath, "utf8");
+  const credentialExecutorSource = fs.readFileSync(credentialExecutorPath, "utf8");
+
+  assert.doesNotMatch(providerLimitsSource, /open-sse\/executors\/index(?:\.ts)?/);
+  assert.match(providerLimitsSource, /open-sse\/executors\/credential\.ts/);
+  assert.doesNotMatch(credentialExecutorSource, /\.\/index(?:\.ts)?/);
+  assert.match(credentialExecutorSource, /export async function getCredentialRefreshExecutor/);
+});
+
+test("credential resolver covers every executor with custom refresh behavior", () => {
+  const credentialExecutorSource = fs.readFileSync(credentialExecutorPath, "utf8");
+  const refreshOverrideFiles = fs
+    .readdirSync(executorDirectory)
+    .filter((file) => file.endsWith(".ts") && !["base.ts", "default.ts"].includes(file))
+    .filter((file) => {
+      const source = fs.readFileSync(path.join(executorDirectory, file), "utf8");
+      return /^\s*(?:async\s+)?(?:needsRefresh|refreshCredentials)\s*\(/m.test(source);
+    });
+
+  for (const file of refreshOverrideFiles) {
+    assert.ok(
+      credentialExecutorSource.includes(`import("./${file}")`),
+      `${file} must be registered in the refresh-only resolver`
+    );
+  }
+});
+
+test("Claude OAuth validation resolves the default executor without the chat registry", () => {
+  const validationSource = fs.readFileSync(anthropicValidationPath, "utf8");
+  const resolverSource = fs.readFileSync(defaultExecutorResolverPath, "utf8");
+
+  assert.doesNotMatch(validationSource, /open-sse\/executors\/index(?:\.ts)?/);
+  assert.match(validationSource, /open-sse\/executors\/defaultResolver\.ts/);
+  assert.doesNotMatch(resolverSource, /\.\/index(?:\.ts)?/);
+  assert.match(resolverSource, /export function getDefaultExecutor/);
+});

tests/unit/logger-process-singleton-12074.test.ts

@@ -0,0 +1,132 @@
+import assert from "node:assert/strict";
+import { mkdtempSync, rmSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import { pathToFileURL } from "node:url";
+import test from "node:test";
+import pino from "pino";
+
+type LoggerModule = typeof import("../../src/shared/utils/logger.ts");
+type LoggerResourceModule = typeof import("../../src/shared/utils/loggerResource.ts");
+type LogRotationModule = typeof import("../../src/lib/logRotation.ts");
+
+const loggerUrl = pathToFileURL(join(process.cwd(), "src/shared/utils/logger.ts")).href;
+const loggerResourceUrl = pathToFileURL(
+  join(process.cwd(), "src/shared/utils/loggerResource.ts")
+).href;
+const logRotationUrl = pathToFileURL(join(process.cwd(), "src/lib/logRotation.ts")).href;
+const envKeys = [
+  "NODE_ENV",
+  "APP_LOG_TO_FILE",
+  "APP_LOG_FILE_PATH",
+  "APP_LOG_LEVEL",
+  "APP_LOG_ROTATION_CHECK_INTERVAL_MS",
+] as const;
+
+function saveEnv(): Record<(typeof envKeys)[number], string | undefined> {
+  return Object.fromEntries(envKeys.map((key) => [key, process.env[key]])) as Record<
+    (typeof envKeys)[number],
+    string | undefined
+  >;
+}
+
+function restoreEnv(saved: ReturnType<typeof saveEnv>): void {
+  for (const key of envKeys) {
+    const value = saved[key];
+    if (value === undefined) delete process.env[key];
+    else process.env[key] = value;
+  }
+}
+
+function closeLoggerStream(logger: LoggerModule["logger"]): void {
+  const stream = (logger as unknown as Record<symbol, unknown>)[pino.symbols.streamSym] as
+    { flushSync?: () => void; end?: () => void } | undefined;
+  try {
+    stream?.flushSync?.();
+  } catch {}
+  try {
+    stream?.end?.();
+  } catch {}
+}
+
+test("logger transport and rotation timer remain process-singletons across HMR module instances", async () => {
+  const savedEnv = saveEnv();
+  const testDir = mkdtempSync(join(tmpdir(), "omniroute-logger-singleton-12074-"));
+  const originalSetInterval = globalThis.setInterval;
+  let firstLogger: LoggerModule | undefined;
+  let secondLogger: LoggerModule | undefined;
+  let firstLoggerResource: LoggerResourceModule | undefined;
+  let secondLoggerResource: LoggerResourceModule | undefined;
+  let firstRotation: LogRotationModule | undefined;
+  let secondRotation: LogRotationModule | undefined;
+
+  process.env.NODE_ENV = "production";
+  process.env.APP_LOG_TO_FILE = "true";
+  process.env.APP_LOG_FILE_PATH = join(testDir, "application.log");
+  process.env.APP_LOG_LEVEL = "debug";
+  process.env.APP_LOG_ROTATION_CHECK_INTERVAL_MS = "60000";
+
+  try {
+    firstRotation = (await import(`${logRotationUrl}?phase4=rotation-a`)) as LogRotationModule;
+    secondRotation = (await import(`${logRotationUrl}?phase4=rotation-b`)) as LogRotationModule;
+    firstRotation.closeLogRotation();
+    secondRotation.closeLogRotation();
+
+    let intervalCreations = 0;
+    globalThis.setInterval = ((...args: unknown[]) => {
+      intervalCreations++;
+      return Reflect.apply(originalSetInterval, globalThis, args);
+    }) as typeof setInterval;
+
+    firstRotation.initLogRotation();
+    secondRotation.initLogRotation();
+    assert.equal(intervalCreations, 1, "HMR reloads must share one log rotation timer");
+
+    globalThis.setInterval = originalSetInterval;
+    firstLoggerResource = (await import(
+      `${loggerResourceUrl}?phase4=resource-a`
+    )) as LoggerResourceModule;
+    secondLoggerResource = (await import(
+      `${loggerResourceUrl}?phase4=resource-b`
+    )) as LoggerResourceModule;
+    firstLogger = (await import(`${loggerUrl}?phase4=logger-a`)) as LoggerModule;
+    secondLogger = (await import(`${loggerUrl}?phase4=logger-b`)) as LoggerModule;
+
+    assert.equal(firstLogger.logger, secondLogger.logger, "HMR reloads must reuse one logger");
+    const firstStream = (firstLogger.logger as unknown as Record<symbol, unknown>)[
+      pino.symbols.streamSym
+    ];
+    const secondStream = (secondLogger.logger as unknown as Record<symbol, unknown>)[
+      pino.symbols.streamSym
+    ];
+    assert.equal(firstStream, secondStream, "HMR reloads must reuse one pino transport");
+
+    const resource = globalThis.__omnirouteLoggerResource;
+    assert.ok(resource, "expected the process-wide logger resource to be registered");
+    const originalClose = resource.close;
+    let closeCalls = 0;
+    resource.close = async () => {
+      closeCalls++;
+      await originalClose();
+    };
+
+    await firstLoggerResource.closeSharedLoggerResource();
+    await secondLoggerResource.closeSharedLoggerResource();
+    assert.equal(closeCalls, 1, "shared logger teardown must be idempotent across HMR modules");
+    assert.equal(globalThis.__omnirouteLoggerResource, undefined);
+  } finally {
+    globalThis.setInterval = originalSetInterval;
+    firstRotation?.closeLogRotation();
+    secondRotation?.closeLogRotation();
+    if (firstLoggerResource) {
+      await firstLoggerResource.closeSharedLoggerResource();
+    } else {
+      if (firstLogger) closeLoggerStream(firstLogger.logger);
+      if (secondLogger && secondLogger.logger !== firstLogger?.logger) {
+        closeLoggerStream(secondLogger.logger);
+      }
+    }
+    restoreEnv(savedEnv);
+    rmSync(testDir, { recursive: true, force: true, maxRetries: 5, retryDelay: 100 });
+  }
+});

tests/unit/provider-limits-accesstoken-fallback-on-refresh-failure.test.ts

@@ -22,7 +22,8 @@ process.env.DATA_DIR = fs.mkdtempSync(
   path.join(os.tmpdir(), "omniroute-accesstoken-fallback-")
 );
 
-const { getExecutor } = await import("../../open-sse/executors/index.ts");
+const { getCredentialRefreshExecutor } =
+  await import("../../open-sse/executors/credential.ts");
 const { refreshAndUpdateCredentials } = await import("../../src/lib/usage/providerLimits.ts");
 
 // `gemini` is a non-rotating (no rotation lock group), non-github OAuth provider,
@@ -39,7 +40,7 @@ function geminiConnection() {
 }
 
 test("falls back to the existing accessToken for a non-github provider when refreshCredentials returns null", async () => {
-  const exec = await getExecutor("gemini");
+  const exec = await getCredentialRefreshExecutor("gemini");
   const origNeeds = exec.needsRefresh;
   const origRefresh = exec.refreshCredentials;
   exec.needsRefresh = () => true; // force the refresh attempt
@@ -65,7 +66,7 @@ test("falls back to the existing accessToken for a non-github provider when refr
 });
 
 test("still throws when refresh fails AND there is no accessToken to fall back on", async () => {
-  const exec = await getExecutor("gemini");
+  const exec = await getCredentialRefreshExecutor("gemini");
   const origNeeds = exec.needsRefresh;
   const origRefresh = exec.refreshCredentials;
   exec.needsRefresh = () => true;

tests/unit/provider-limits-force-refresh-imported.test.ts

@@ -18,7 +18,8 @@ import path from "node:path";
 
 process.env.DATA_DIR = fs.mkdtempSync(path.join(os.tmpdir(), "omniroute-force-refresh-imported-"));
 
-const { getExecutor } = await import("../../open-sse/executors/index.ts");
+const { getCredentialRefreshExecutor } =
+  await import("../../open-sse/executors/credential.ts");
 const { refreshAndUpdateCredentials } = await import("../../src/lib/usage/providerLimits.ts");
 
 function importedCodexConnection() {
@@ -33,7 +34,7 @@ function importedCodexConnection() {
 }
 
 test("force re-mints an imported rotating account that needsRefresh would skip (#3019 reactive)", async () => {
-  const exec = await getExecutor("codex");
+  const exec = await getCredentialRefreshExecutor("codex");
   const origNeeds = exec.needsRefresh;
   const origRefresh = exec.refreshCredentials;
   let refreshCalls = 0;
@@ -65,7 +66,7 @@ test("force re-mints an imported rotating account that needsRefresh would skip (
 });
 
 test("force does NOT override the bulk #3019 guard (no allowRotatingRefresh → no mint)", async () => {
-  const exec = await getExecutor("codex");
+  const exec = await getCredentialRefreshExecutor("codex");
   const origNeeds = exec.needsRefresh;
   const origRefresh = exec.refreshCredentials;
   let refreshCalls = 0;

tests/unit/responses-plaintext-reasoning-input-12128.test.ts

@@ -0,0 +1,88 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+
+const { translateRequest } = await import("../../open-sse/translator/index.ts");
+const { resolveReasoningTransport, applyReasoningInputPolicy } = await import(
+  "../../open-sse/services/reasoningInputPolicy.ts"
+);
+const { FORMATS } = await import("../../open-sse/translator/formats.ts");
+
+test("#12128: resolveReasoningTransport identifies openai-compatible-responses and custom responses variants as opaque transport", () => {
+  assert.equal(resolveReasoningTransport("openai-compatible-responses-codex"), "opaque");
+  assert.equal(resolveReasoningTransport("custom-openai-responses"), "opaque");
+  assert.equal(resolveReasoningTransport("codex-proxy"), "opaque");
+  assert.equal(resolveReasoningTransport("openai"), "opaque");
+  assert.equal(resolveReasoningTransport("codex"), "opaque");
+});
+
+test("#12128: translateRequest to Responses target strips plaintext reasoning.content for opaque responses providers", () => {
+  const chatBody = {
+    model: "gpt-5.6-codex",
+    messages: [
+      { role: "user", content: "hello" },
+      {
+        role: "assistant",
+        content: "Hi there!",
+        reasoning_content: "Let me think about how to greet the user properly.",
+      },
+      { role: "user", content: "what is 2+2?" },
+    ],
+  };
+
+  const translated = translateRequest(
+    FORMATS.OPENAI,
+    FORMATS.OPENAI_RESPONSES,
+    "gpt-5.6-codex",
+    chatBody,
+    false,
+    null,
+    "openai-compatible-responses-codex"
+  ) as Record<string, unknown>;
+
+  assert.ok(Array.isArray(translated.input), "translated.input must be an array");
+
+  const input = translated.input as Record<string, unknown>[];
+  const reasoningItems = input.filter((item) => item.type === "reasoning");
+
+  // For opaque responses targets without encrypted continuation, orphaned plaintext reasoning items
+  // must either be stripped entirely or have content.length === 0, so strict Codex backends do not 400.
+  for (const r of reasoningItems) {
+    assert.ok(
+      !r.content || (Array.isArray(r.content) && r.content.length === 0),
+      "reasoning item must not carry non-empty content array to strict responses endpoints"
+    );
+  }
+});
+
+test("#12128: applyReasoningInputPolicy directly sanitizes replayed plaintext reasoning for openai-compatible-responses", () => {
+  const body: Record<string, unknown> = {
+    input: [
+      {
+        type: "reasoning",
+        content: [{ type: "reasoning_text", text: "step 1 plan" }],
+        summary: [],
+      },
+      {
+        type: "message",
+        role: "assistant",
+        content: [{ type: "output_text", text: "Done" }],
+      },
+    ],
+  };
+
+  applyReasoningInputPolicy(body, "responses", {
+    provider: "openai-compatible-responses-vllm",
+    onIncompatibleReasoning: "drop",
+  });
+
+  const input = body.input as Record<string, unknown>[];
+  const reasoningItems = input.filter((item) => item.type === "reasoning");
+
+  for (const r of reasoningItems) {
+    assert.equal(
+      r.content,
+      undefined,
+      "plaintext reasoning content must be dropped for opaque responses providers"
+    );
+  }
+});

tests/unit/root-layout-settings-boundary-12074.test.ts

@@ -0,0 +1,111 @@
+import assert from "node:assert/strict";
+import fs from "node:fs";
+import os from "node:os";
+import path from "node:path";
+import test from "node:test";
+
+const TEST_DATA_DIR = fs.mkdtempSync(path.join(os.tmpdir(), "omniroute-layout-settings-"));
+process.env.DATA_DIR = TEST_DATA_DIR;
+
+const ORIGINAL_INITIAL_PASSWORD = process.env.INITIAL_PASSWORD;
+const core = await import("../../src/lib/db/core.ts");
+const { getRootLayoutSettings } = await import("../../src/lib/db/rootLayoutSettings.ts");
+
+function resetStorage(): void {
+  core.resetDbInstance();
+  fs.rmSync(TEST_DATA_DIR, { recursive: true, force: true });
+  fs.mkdirSync(TEST_DATA_DIR, { recursive: true });
+}
+
+test.beforeEach(() => {
+  resetStorage();
+  delete process.env.INITIAL_PASSWORD;
+});
+
+test.after(() => {
+  core.resetDbInstance();
+  fs.rmSync(TEST_DATA_DIR, { recursive: true, force: true });
+
+  if (ORIGINAL_INITIAL_PASSWORD === undefined) {
+    delete process.env.INITIAL_PASSWORD;
+  } else {
+    process.env.INITIAL_PASSWORD = ORIGINAL_INITIAL_PASSWORD;
+  }
+});
+
+test("root layout imports the read-only settings leaf", () => {
+  const layoutSource = fs.readFileSync(path.join(process.cwd(), "src/app/layout.tsx"), "utf8");
+  const leafSource = fs.readFileSync(
+    path.join(process.cwd(), "src/lib/db/rootLayoutSettings.ts"),
+    "utf8"
+  );
+  const coreSource = fs.readFileSync(path.join(process.cwd(), "src/lib/db/core.ts"), "utf8");
+  const singletonSource = fs.readFileSync(
+    path.join(process.cwd(), "src/lib/db/singleton.ts"),
+    "utf8"
+  );
+
+  assert.match(
+    layoutSource,
+    /import \{ getRootLayoutSettings \} from "@\/lib\/db\/rootLayoutSettings";/
+  );
+  assert.doesNotMatch(layoutSource, /@\/lib\/db\/settings/);
+  assert.match(leafSource, /from "\.\/singleton"/);
+  assert.doesNotMatch(
+    leafSource,
+    /db\/settings|\.\/core|runtimeSettings|tokenHealth|providerModels/
+  );
+  assert.doesNotMatch(leafSource, /\b(?:INSERT|UPDATE|DELETE|REPLACE)\b/i);
+  assert.match(coreSource, /from "\.\/singleton"/);
+  assert.doesNotMatch(singletonSource, /db\/settings|\.\/core|readCache|runtimeSettings/);
+});
+
+test("root layout settings reader does not initialize the database", async () => {
+  process.env.INITIAL_PASSWORD = "must-not-trigger-startup";
+
+  assert.deepEqual(await getRootLayoutSettings(), {
+    instanceName: "OmniRoute",
+    customFaviconUrl: "",
+    customFaviconBase64: "",
+  });
+  assert.equal(fs.existsSync(path.join(TEST_DATA_DIR, "storage.sqlite")), false);
+});
+
+test("root layout settings reader returns only the persisted metadata fields", async () => {
+  const db = core.getDbInstance();
+  const insert = db.prepare(
+    "INSERT INTO key_value (namespace, key, value) VALUES ('settings', ?, ?)"
+  );
+
+  insert.run("instanceName", JSON.stringify("Route Lab"));
+  insert.run("customFaviconUrl", JSON.stringify("https://example.com/favicon.png"));
+  insert.run("customFaviconBase64", JSON.stringify("data:image/png;base64,AA=="));
+  insert.run("proxyEnabled", JSON.stringify(false));
+
+  assert.deepEqual(await getRootLayoutSettings(), {
+    instanceName: "Route Lab",
+    customFaviconUrl: "https://example.com/favicon.png",
+    customFaviconBase64: "data:image/png;base64,AA==",
+  });
+});
+
+test("root layout settings reader is read-only and falls back safely", async () => {
+  process.env.INITIAL_PASSWORD = "must-not-trigger-onboarding";
+  const db = core.getDbInstance();
+  db.prepare(
+    "INSERT INTO key_value (namespace, key, value) VALUES ('settings', 'instanceName', ?)"
+  ).run("not-json");
+
+  assert.deepEqual(await getRootLayoutSettings(), {
+    instanceName: "OmniRoute",
+    customFaviconUrl: "",
+    customFaviconBase64: "",
+  });
+
+  const onboardingRows = db
+    .prepare(
+      "SELECT key FROM key_value WHERE namespace = 'settings' AND key IN ('setupComplete', 'requireLogin')"
+    )
+    .all();
+  assert.deepEqual(onboardingRows, []);
+});

tests/unit/router-strategies.test.ts

@@ -15,7 +15,10 @@ import {
   listStrategies,
   type RoutingContext,
 } from "../../open-sse/services/autoCombo/routerStrategy.ts";
-import type { ProviderCandidate } from "../../open-sse/services/autoCombo/scoring.ts";
+import {
+  DEFAULT_WEIGHTS,
+  type ProviderCandidate,
+} from "../../open-sse/services/autoCombo/scoring.ts";
 
 function cand(p: Partial<ProviderCandidate> & { provider: string }): ProviderCandidate {
   return {
@@ -33,6 +36,32 @@ function cand(p: Partial<ProviderCandidate> & { provider: string }): ProviderCan
 
 const ctx: RoutingContext = { taskType: "default" };
 
+// ── score ────────────────────────────────────────────────────────────────────
+test("score — exploits the configured winner and uses explorationRate", (t) => {
+  const pool = [
+    cand({ provider: "cheap", costPer1MTokens: 1 }),
+    cand({ provider: "expensive", costPer1MTokens: 9 }),
+  ];
+  const weights = { ...DEFAULT_WEIGHTS, costInv: 1, quota: 0, health: 0, latencyInv: 0 };
+
+  t.mock.method(Math, "random", () => 0.99);
+
+  assert.equal(
+    getStrategy("score").select(pool, { ...ctx, weights, explorationRate: 0 }).provider,
+    "cheap"
+  );
+  assert.equal(
+    getStrategy("score").select(pool, { ...ctx, weights, explorationRate: 1 }).provider,
+    "expensive"
+  );
+});
+
+test("score — exact ties preserve configured candidate order", () => {
+  const pool = [cand({ provider: "first" }), cand({ provider: "second" })];
+
+  assert.equal(getStrategy("score").select(pool, { ...ctx, explorationRate: 0 }).provider, "first");
+});
+
 // ── cost ─────────────────────────────────────────────────────────────────────
 test("cost — selects the cheapest healthy candidate", () => {
   const pool = [
@@ -307,6 +336,7 @@ test("selectWithStrategy — unknown strategy silently falls back to rules", ()
 
 test("listStrategies — exposes every registered strategy + aliases", () => {
   const names = listStrategies().map((s) => s.name);
+  assert.ok(names.includes("score"), "listStrategies missing 'score'");
   for (const n of ["rules", "cost", "eco", "latency", "fast", "sla-aware", "sla", "lkgp"]) {
     assert.ok(names.includes(n), `listStrategies missing '${n}'`);
   }

tests/unit/run-next-graceful-shutdown-12074.test.ts

@@ -0,0 +1,29 @@
+import assert from "node:assert/strict";
+import { readFileSync } from "node:fs";
+import { join } from "node:path";
+import test from "node:test";
+
+const runNextSource = readFileSync(join(process.cwd(), "scripts/dev/run-next.mjs"), "utf8");
+
+test("the custom Next runner owns exit and awaits application cleanup before closing Next", () => {
+  const ownerRegistration = runNextSource.indexOf(
+    "globalThis.__omnirouteCustomServerOwnsShutdown = true"
+  );
+  const prepareCall = runNextSource.indexOf("await prepareWithHeal()");
+  assert.ok(ownerRegistration >= 0, "custom server shutdown ownership must be registered");
+  assert.ok(
+    ownerRegistration < prepareCall,
+    "shutdown ownership must exist before instrumentation"
+  );
+
+  const serverClose = runNextSource.indexOf("server.close(resolve)");
+  const applicationCleanup = runNextSource.indexOf(
+    "await globalThis.__omnirouteRequestShutdown?.(signal)"
+  );
+  const nextClose = runNextSource.indexOf("await nextApp.close()", applicationCleanup);
+  const processExit = runNextSource.indexOf("process.exit(0)", nextClose);
+
+  assert.ok(serverClose < applicationCleanup, "stop accepting requests before application cleanup");
+  assert.ok(applicationCleanup < nextClose, "application cleanup must finish before Next closes");
+  assert.ok(nextClose < processExit, "process exit must remain the final shutdown action");
+});

tests/unit/termux-android-playwright-static-import-7265.test.ts

@@ -9,8 +9,8 @@ import { dirname, join } from "node:path";
 // top-level IIFE at *require time* — merely importing the `playwright` package crashes,
 // no browser needs to be launched. `claudeTurnstileSolver.ts` used to `import { chromium }
 // from "playwright"` statically, and that module is unconditionally reachable from the
-// Next.js instrumentation hook on every boot via open-sse/executors/index.ts, so any
-// unsupported platform crashed the whole server at startup regardless of configured provider.
+// Next.js instrumentation hook used to reach it on every boot via open-sse/executors/index.ts,
+// so any unsupported platform crashed the whole server regardless of configured provider.
 const HERE = dirname(fileURLToPath(import.meta.url));
 const SOLVER = join(HERE, "../../open-sse/services/claudeTurnstileSolver.ts");
 
@@ -28,10 +28,9 @@ test("importing the real executor chain does not throw on an unsupported process
   Object.defineProperty(process, "platform", { value: "android", configurable: true });
 
   try {
-    // This is the exact reachability chain from the Next.js instrumentation hook:
-    // instrumentation-node.ts -> open-sse/index.ts -> executors/index.ts -> claude-web*.ts
-    // -> claudeTurnstileSolver.ts. Before the fix, this threw
-    // "Unsupported platform: android" purely from the static playwright import.
+    // Keep the executor-registry import safe for direct runtime consumers. Before the
+    // original fix, this threw "Unsupported platform: android" purely from the static
+    // playwright import; instrumentation no longer imports this graph at boot (#12074).
     await import("../../open-sse/executors/index.ts");
   } finally {
     Object.defineProperty(process, "platform", originalDescriptor);

tests/unit/translator-antigravity-to-openai.test.ts

@@ -162,7 +162,13 @@ test("Antigravity -> OpenAI keeps co-located function call and text but strips t
             role: "model",
             parts: [
               { text: "Let me look that up." },
-              { functionResponse: { id: "call_9", name: "lookup", response: { result: { ok: true } } } },
+              {
+                functionResponse: {
+                  id: "call_9",
+                  name: "lookup",
+                  response: { result: { ok: true } },
+                },
+              },
               { functionCall: { id: "call_10", name: "lookup", args: { q: "weather" } } },
             ],
           },
@@ -249,7 +255,7 @@ test("Antigravity -> OpenAI lowers schema types recursively", () => {
     false
   );
 
-  assert.deepEqual((result.tools[0].function as any).parameters, {
+  assert.deepEqual((result.tools[0].function as Record<string, unknown>).parameters, {
     type: "object",
     properties: {
       items: {
@@ -298,12 +304,17 @@ test("Antigravity -> OpenAI strips enumDescriptions from tool schema (top-level
     false
   );
 
-  const parameters = (result.tools[0].function as any).parameters;
+  const parameters = (result.tools[0].function as Record<string, unknown>).parameters as Record<
+    string,
+    unknown
+  >;
+  const props = parameters.properties as Record<string, unknown>;
+  const tags = props.tags as Record<string, unknown>;
 
   // enumDescriptions must be removed at every level of the schema tree...
   assert.equal("enumDescriptions" in parameters, false);
-  assert.equal("enumDescriptions" in parameters.properties.mode, false);
-  assert.equal("enumDescriptions" in parameters.properties.tags.items, false);
+  assert.equal("enumDescriptions" in (props.mode as Record<string, unknown>), false);
+  assert.equal("enumDescriptions" in (tags.items as Record<string, unknown>), false);
 
   // ...while leaving the rest of the schema (incl. enum values) intact.
   assert.deepEqual(parameters, {
@@ -349,7 +360,10 @@ test("Antigravity -> OpenAI preserves the required array on Draft 2020-12 tool s
     false
   );
 
-  const params = (result.tools[0].function as any).parameters;
+  const params = (result.tools[0].function as Record<string, unknown>).parameters as Record<
+    string,
+    unknown
+  >;
   // The required array must survive so the model treats mandatory args as mandatory.
   assert.deepEqual(params.required, ["path", "contents"]);
   // Types are still lowered and Draft 2020-12 meta keywords are stripped.
@@ -385,6 +399,104 @@ test("Antigravity -> OpenAI drops required entries that no longer exist in prope
     false
   );
 
-  const params = (result.tools[0].function as any).parameters;
+  const params = (result.tools[0].function as Record<string, unknown>).parameters as Record<
+    string,
+    unknown
+  >;
   assert.deepEqual(params.required, ["kept"]);
 });
+
+test("Antigravity -> OpenAI preserves falsy primitive results in function responses (false, 0, empty string, null)", () => {
+  const cases: Array<[unknown, string]> = [
+    [false, "false"],
+    [0, "0"],
+    ["", '""'],
+    [null, "null"],
+    [true, "true"],
+    [42, "42"],
+    ["done", '"done"'],
+  ];
+
+  for (const [inputVal, expected] of cases) {
+    const result = antigravityToOpenAIRequest(
+      "gpt-4o",
+      {
+        request: {
+          contents: [
+            {
+              role: "model",
+              parts: [
+                {
+                  functionCall: {
+                    id: "call_test",
+                    name: "check_condition",
+                    args: {},
+                  },
+                },
+              ],
+            },
+            {
+              role: "user",
+              parts: [
+                {
+                  functionResponse: {
+                    id: "call_test",
+                    name: "check_condition",
+                    response: { result: inputVal },
+                  },
+                },
+              ],
+            },
+          ],
+        },
+      },
+      false
+    );
+
+    const toolMsg = result.messages.find((m) => m.role === "tool");
+    assert.ok(toolMsg, "expected role:tool message");
+    assert.equal(toolMsg.tool_call_id, "call_test");
+    assert.equal(toolMsg.content, expected);
+  }
+});
+
+test("Antigravity -> OpenAI preserves custom response objects without result key", () => {
+  const result = antigravityToOpenAIRequest(
+    "gpt-4o",
+    {
+      request: {
+        contents: [
+          {
+            role: "model",
+            parts: [
+              {
+                functionCall: {
+                  id: "call_custom",
+                  name: "custom_op",
+                  args: {},
+                },
+              },
+            ],
+          },
+          {
+            role: "user",
+            parts: [
+              {
+                functionResponse: {
+                  id: "call_custom",
+                  name: "custom_op",
+                  response: { output: "value", success: false },
+                },
+              },
+            ],
+          },
+        ],
+      },
+    },
+    false
+  );
+
+  const toolMsg = result.messages.find((m) => m.role === "tool");
+  assert.ok(toolMsg, "expected role:tool message");
+  assert.equal(toolMsg.content, '{"output":"value","success":false}');
+});

tests/unit/translator-gemini-to-openai.test.ts

@@ -237,3 +237,70 @@ test("Gemini -> OpenAI maintains matching IDs across multi-turn tool call and re
   assert.equal(toolResponseCallId, "call_calc_456");
   assert.equal(assistantCallId, toolResponseCallId);
 });
+
+test("Gemini -> OpenAI preserves falsy primitive results in function responses (false, 0, empty string, null)", () => {
+  const cases: Array<[unknown, string]> = [
+    [false, "false"],
+    [0, "0"],
+    ["", '""'],
+    [null, "null"],
+    [true, "true"],
+    [42, "42"],
+    ["done", '"done"'],
+  ];
+
+  for (const [inputVal, expected] of cases) {
+    const result = geminiToOpenAIRequest(
+      "gpt-4o",
+      {
+        contents: [
+          {
+            role: "user",
+            parts: [
+              {
+                functionResponse: {
+                  id: "call_test",
+                  name: "check_condition",
+                  response: { result: inputVal },
+                },
+              },
+            ],
+          },
+        ],
+      },
+      false
+    );
+
+    assert.equal(result.messages.length, 1);
+    assert.equal(result.messages[0].role, "tool");
+    assert.equal(result.messages[0].tool_call_id, "call_test");
+    assert.equal(result.messages[0].content, expected);
+  }
+});
+
+test("Gemini -> OpenAI preserves custom response objects without result key", () => {
+  const result = geminiToOpenAIRequest(
+    "gpt-4o",
+    {
+      contents: [
+        {
+          role: "user",
+          parts: [
+            {
+              functionResponse: {
+                id: "call_custom",
+                name: "custom_op",
+                response: { output: "value", success: false },
+              },
+            },
+          ],
+        },
+      ],
+    },
+    false
+  );
+
+  assert.equal(result.messages.length, 1);
+  assert.equal(result.messages[0].role, "tool");
+  assert.equal(result.messages[0].content, '{"output":"value","success":false}');
+});

tests/unit/ui/orchestrationModel.test.ts

@@ -0,0 +1,372 @@
+/**
+ * tests/unit/ui/orchestrationModel.test.ts
+ * Run: node --import tsx/esm --test tests/unit/ui/orchestrationModel.test.ts
+ */
+import { describe, it } from "node:test";
+import assert from "node:assert/strict";
+import {
+  ORCH_STATES,
+  orchStateColor,
+} from "../../../src/app/(dashboard)/dashboard/orchestration/model/orchestrationTypes.ts";
+import { STATUS_HEX } from "../../../src/shared/constants/statusColors.ts";
+import { fromCloudAgent } from "../../../src/app/(dashboard)/dashboard/orchestration/model/fromCloudAgent.ts";
+import type { CloudAgentTask } from "../../../src/lib/cloudAgent/types.ts";
+import { fromA2A } from "../../../src/app/(dashboard)/dashboard/orchestration/model/fromA2A.ts";
+import type { A2ATask } from "../../../src/lib/a2a/taskManager.ts";
+import { fromConductor } from "../../../src/app/(dashboard)/dashboard/orchestration/model/fromConductor.ts";
+import type { FleetSnapshot } from "../../../src/lib/conductor/hubProxy.ts";
+import { mergeSnapshot } from "../../../src/app/(dashboard)/dashboard/orchestration/model/mergeSnapshot.ts";
+import {
+  STALE_COMPLETED_MS,
+  MAX_WORK_NODES,
+} from "../../../src/app/(dashboard)/dashboard/orchestration/model/orchestrationTypes.ts";
+
+describe("orchestrationTypes", () => {
+  it("covers all six states with a color each", () => {
+    assert.equal(ORCH_STATES.length, 6);
+    for (const s of ORCH_STATES) {
+      assert.match(orchStateColor(s), /^#[0-9a-f]{6}$/i, s);
+    }
+  });
+  it("waiting_approval maps to the new STATUS_HEX.approval violet", () => {
+    assert.equal(orchStateColor("waiting_approval"), STATUS_HEX.approval);
+    assert.equal(STATUS_HEX.approval, "#8b5cf6");
+  });
+  it("running maps to warning, succeeded to success, failed to error", () => {
+    assert.equal(orchStateColor("running"), STATUS_HEX.warning);
+    assert.equal(orchStateColor("succeeded"), STATUS_HEX.success);
+    assert.equal(orchStateColor("failed"), STATUS_HEX.error);
+  });
+});
+
+function caTask(over: Partial<CloudAgentTask>): CloudAgentTask {
+  return {
+    id: "t1",
+    providerId: "devin",
+    status: "running",
+    prompt: "Fix the flaky test in CI",
+    source: { repoName: "acme/app", repoUrl: "https://github.com/acme/app" },
+    options: {},
+    activities: [],
+    createdAt: "2026-08-30T10:00:00Z",
+    updatedAt: "2026-08-30T10:05:00Z",
+    ...over,
+  } as CloudAgentTask;
+}
+
+describe("fromCloudAgent", () => {
+  it("maps every status to the unified OrchState", () => {
+    const cases: Array<[CloudAgentTask["status"], string]> = [
+      ["queued", "queued"],
+      ["running", "running"],
+      ["awaiting_approval", "waiting_approval"],
+      ["completed", "succeeded"],
+      ["failed", "failed"],
+      ["cancelled", "cancelled"],
+    ];
+    for (const [input, expected] of cases) {
+      const { nodes } = fromCloudAgent([caTask({ status: input })]);
+      const work = nodes.find((n) => n.kind === "work");
+      assert.equal(work?.state, expected, input);
+    }
+  });
+  it("unknown status becomes failed with the raw value in sublabel", () => {
+    const { nodes } = fromCloudAgent([caTask({ status: "exploded" as CloudAgentTask["status"] })]);
+    const work = nodes.find((n) => n.kind === "work");
+    assert.equal(work?.state, "failed");
+    assert.match(work?.sublabel ?? "", /exploded/);
+  });
+  it("running task with activities gets one ActivityNode; completed does not", () => {
+    const running = caTask({
+      activities: [
+        { id: "a1", type: "command", content: "npm test", timestamp: "2026-08-30T10:04:00Z" },
+      ],
+    });
+    const done = caTask({ id: "t2", status: "completed", activities: running.activities });
+    const { nodes, edges } = fromCloudAgent([running, done]);
+    const acts = nodes.filter((n) => n.kind === "activity");
+    assert.equal(acts.length, 1);
+    assert.equal(acts[0].id, "cloud-agent:t1:activity");
+    assert.ok(
+      edges.some(
+        (e) => e.from === "cloud-agent:t1" && e.to === "cloud-agent:t1:activity" && e.active
+      )
+    );
+  });
+  it("emits a SourceNode with per-state counts and owns-edges from it", () => {
+    const { nodes, edges } = fromCloudAgent([caTask({}), caTask({ id: "t2", status: "failed" })]);
+    const src = nodes.find((n) => n.id === "source:cloud-agent");
+    assert.equal(src?.counts?.running, 1);
+    assert.equal(src?.counts?.failed, 1);
+    assert.ok(
+      edges.some(
+        (e) => e.from === "source:cloud-agent" && e.to === "cloud-agent:t1" && e.kind === "owns"
+      )
+    );
+  });
+  it("empty input emits nothing", () => {
+    const out = fromCloudAgent([]);
+    assert.equal(out.nodes.length, 0);
+    assert.equal(out.edges.length, 0);
+  });
+});
+
+function a2aTask(over: Partial<A2ATask>): A2ATask {
+  return {
+    id: "a1",
+    skill: "smart-routing",
+    state: "working",
+    input: { skill: "smart-routing", messages: [{ role: "user", content: "route this well" }] },
+    artifacts: [],
+    events: [{ timestamp: "2026-08-30T10:00:00Z", state: "submitted" }],
+    metadata: {},
+    createdAt: "2026-08-30T10:00:00Z",
+    updatedAt: "2026-08-30T10:01:00Z",
+    expiresAt: "2026-08-30T10:05:00Z",
+    ...over,
+  } as A2ATask;
+}
+
+describe("fromA2A", () => {
+  it("maps the five A2A states", () => {
+    const cases: Array<[A2ATask["state"], string]> = [
+      ["submitted", "queued"],
+      ["working", "running"],
+      ["completed", "succeeded"],
+      ["failed", "failed"],
+      ["cancelled", "cancelled"],
+    ];
+    for (const [input, expected] of cases) {
+      const { nodes } = fromA2A([a2aTask({ state: input })]);
+      assert.equal(nodes.find((n) => n.kind === "work")?.state, expected, input);
+    }
+  });
+  it("work node id is a2a:<id>, label is the skill", () => {
+    const { nodes } = fromA2A([a2aTask({})]);
+    const w = nodes.find((n) => n.kind === "work");
+    assert.equal(w?.id, "a2a:a1");
+    assert.equal(w?.label, "smart-routing");
+  });
+  it("empty input emits nothing", () => {
+    assert.equal(fromA2A([]).nodes.length, 0);
+  });
+});
+
+const baseSnap: FleetSnapshot = {
+  offline: false,
+  runners: [{ id: "r1", name: "runner-one", clis: ["claude"], online: true, draining: false }],
+  tasks: [
+    {
+      id: "ct1",
+      status: "running",
+      mode: "auto",
+      repo: "acme/app",
+      runner: "r1",
+      summary: "Refactor auth",
+      branch: null,
+      error: null,
+      updated_at: "2026-08-30T10:00:00Z",
+    },
+  ],
+};
+
+describe("fromConductor", () => {
+  it("online runner with a running task → running WorkNode + ActivityNode for the task", () => {
+    const { nodes, edges } = fromConductor(baseSnap);
+    const runner = nodes.find((n) => n.id === "conductor:runner:r1");
+    assert.equal(runner?.state, "running");
+    const act = nodes.find((n) => n.id === "conductor:task:ct1");
+    assert.equal(act?.kind, "activity");
+    assert.ok(edges.some((e) => e.from === "conductor:runner:r1" && e.to === "conductor:task:ct1"));
+  });
+  it("queued task without runner hangs directly under the source as a work node", () => {
+    const snap: FleetSnapshot = {
+      ...baseSnap,
+      runners: [],
+      tasks: [{ ...baseSnap.tasks[0], id: "ct2", status: "queued", runner: null }],
+    };
+    const { nodes, edges } = fromConductor(snap);
+    const w = nodes.find((n) => n.id === "conductor:task:ct2");
+    assert.equal(w?.kind, "work");
+    assert.equal(w?.state, "queued");
+    assert.ok(edges.some((e) => e.from === "source:conductor" && e.to === "conductor:task:ct2"));
+  });
+  it("offline snapshot emits only nothing (hook marks the source offline separately)", () => {
+    const out = fromConductor({ offline: true, runners: [], tasks: [] });
+    assert.equal(out.nodes.length, 0);
+  });
+  it("unknown hub status maps to failed with the raw value in sublabel", () => {
+    const snap: FleetSnapshot = {
+      ...baseSnap,
+      runners: [],
+      tasks: [{ ...baseSnap.tasks[0], id: "ct3", status: "vaporized", runner: null }],
+    };
+    const w = fromConductor(snap).nodes.find((n) => n.id === "conductor:task:ct3");
+    assert.equal(w?.state, "failed");
+    assert.match(w?.sublabel ?? "", /vaporized/);
+  });
+  it("running task whose runner is not present in snap.runners is emitted as a work node, not swallowed", () => {
+    const snap: FleetSnapshot = {
+      offline: false,
+      runners: [],
+      tasks: [
+        {
+          ...baseSnap.tasks[0],
+          id: "ct-orphan",
+          status: "running",
+          runner: "ghost-runner",
+        },
+      ],
+    };
+    const { nodes, edges } = fromConductor(snap);
+    const w = nodes.find((n) => n.id === "conductor:task:ct-orphan");
+    assert.equal(w?.kind, "work");
+    assert.equal(w?.state, "running");
+    assert.ok(
+      edges.some((e) => e.from === "source:conductor" && e.to === "conductor:task:ct-orphan")
+    );
+  });
+});
+
+const OK_SOURCES = [
+  { source: "cloud-agent" as const, ok: true },
+  { source: "a2a" as const, ok: true },
+  { source: "conductor" as const, ok: true },
+];
+const NOW = Date.parse("2026-08-30T12:00:00Z");
+const empty = { nodes: [], edges: [] };
+
+describe("mergeSnapshot", () => {
+  it("adds the orchestrator root and root→source edges", () => {
+    const snap = mergeSnapshot(
+      { cloudAgent: fromCloudAgent([caTask({})]), a2a: empty, conductor: empty },
+      OK_SOURCES,
+      { now: NOW }
+    );
+    assert.ok(snap.nodes.some((n) => n.id === "orchestrator"));
+    assert.ok(snap.edges.some((e) => e.from === "orchestrator" && e.to === "source:cloud-agent"));
+  });
+  it("dedupes a Conductor-mirrored A2A task into one node with a mirror edge", () => {
+    const a2a = fromA2A([
+      a2aTask({ id: "am1", skill: "conductor", metadata: { conductor: { task_id: "ct1" } } }),
+    ]);
+    const conductor = fromConductor(baseSnap); // contains conductor:task:ct1 as activity of r1
+    const snap = mergeSnapshot({ cloudAgent: empty, a2a, conductor }, OK_SOURCES, { now: NOW });
+    assert.ok(
+      !snap.nodes.some((n) => n.id === "a2a:am1"),
+      "mirrored A2A work node must be dropped"
+    );
+    const mirrorEdge = snap.edges.find((e) => e.kind === "mirror");
+    assert.equal(mirrorEdge?.to, "source:a2a");
+  });
+  it("drops terminal work older than STALE_COMPLETED_MS unless showCompleted", () => {
+    const old = caTask({
+      id: "old",
+      status: "completed",
+      completedAt: new Date(NOW - STALE_COMPLETED_MS - 1000).toISOString(),
+    });
+    const parts = { cloudAgent: fromCloudAgent([old]), a2a: empty, conductor: empty };
+    assert.ok(
+      !mergeSnapshot(parts, OK_SOURCES, { now: NOW }).nodes.some((n) => n.id === "cloud-agent:old")
+    );
+    assert.ok(
+      mergeSnapshot(parts, OK_SOURCES, { now: NOW, showCompleted: true }).nodes.some(
+        (n) => n.id === "cloud-agent:old"
+      )
+    );
+  });
+  it("caps work nodes at MAX_WORK_NODES with a per-source overflow node", () => {
+    const many = Array.from({ length: MAX_WORK_NODES + 10 }, (_, i) =>
+      caTask({ id: `m${i}`, updatedAt: new Date(NOW - i * 1000).toISOString() })
+    );
+    const snap = mergeSnapshot(
+      { cloudAgent: fromCloudAgent(many), a2a: empty, conductor: empty },
+      OK_SOURCES,
+      { now: NOW }
+    );
+    const works = snap.nodes.filter((n) => n.kind === "work");
+    assert.ok(works.length <= MAX_WORK_NODES, `got ${works.length}`);
+    const overflow = snap.nodes.find((n) => n.id === "overflow:cloud-agent");
+    assert.ok(overflow, "overflow node expected");
+  });
+  it("keeps SourceStatus[] verbatim on the snapshot", () => {
+    const src = [{ source: "conductor" as const, ok: false, offline: true }];
+    const snap = mergeSnapshot({ cloudAgent: empty, a2a: empty, conductor: empty }, src, {
+      now: NOW,
+    });
+    assert.deepEqual(snap.sources, src);
+  });
+  it("does not mutate input node objects when deduping (honors the Pure contract)", () => {
+    const a2a = fromA2A([
+      a2aTask({ id: "am2", skill: "conductor", metadata: { conductor: { task_id: "ct1" } } }),
+    ]);
+    const conductor = fromConductor(baseSnap); // contains conductor:task:ct1 as activity of r1
+    const originalNode = conductor.nodes.find((n) => n.id === "conductor:task:ct1");
+    assert.ok(originalNode, "conductor:task:ct1 must exist in the source part");
+    const snap = mergeSnapshot({ cloudAgent: empty, a2a, conductor }, OK_SOURCES, { now: NOW });
+    assert.equal(
+      "mirrorOf" in (originalNode as object),
+      false,
+      "the original conductor input node must not be mutated"
+    );
+    const mergedNode = snap.nodes.find((n) => n.id === "conductor:task:ct1");
+    assert.equal(mergedNode?.mirrorOf, "a2a:am2");
+  });
+  it("materializes an offline placeholder source node for a source reporting offline:true even when ok:true", () => {
+    const src = [{ source: "conductor" as const, ok: true, offline: true }];
+    const snap = mergeSnapshot({ cloudAgent: empty, a2a: empty, conductor: empty }, src, {
+      now: NOW,
+    });
+    const node = snap.nodes.find((n) => n.id === "source:conductor");
+    assert.ok(node, "offline placeholder source:conductor node expected");
+    assert.equal(node?.sublabel, "offline");
+    assert.ok(
+      snap.edges.some(
+        (e) => e.from === "orchestrator" && e.to === "source:conductor" && e.kind === "owns"
+      )
+    );
+  });
+  it("overflow node carries droppedByState with the per-state counts of dropped work nodes", () => {
+    const many = Array.from({ length: MAX_WORK_NODES + 5 }, (_, i) =>
+      caTask({
+        id: `ov${i}`,
+        status: i < MAX_WORK_NODES ? "running" : "failed",
+        updatedAt: new Date(NOW - i * 1000).toISOString(),
+      })
+    );
+    const snap = mergeSnapshot(
+      { cloudAgent: fromCloudAgent(many), a2a: empty, conductor: empty },
+      OK_SOURCES,
+      { now: NOW }
+    );
+    const overflow = snap.nodes.find((n) => n.id === "overflow:cloud-agent");
+    assert.ok(overflow, "overflow node expected");
+    assert.deepEqual(overflow?.droppedByState, { failed: 5 });
+  });
+  it("drops a stale terminal Conductor task older than STALE_COMPLETED_MS unless showCompleted", () => {
+    const staleSnap: FleetSnapshot = {
+      ...baseSnap,
+      runners: [],
+      tasks: [
+        {
+          ...baseSnap.tasks[0],
+          id: "ctOld",
+          status: "completed",
+          runner: null,
+          updated_at: new Date(NOW - STALE_COMPLETED_MS - 1000).toISOString(),
+        },
+      ],
+    };
+    const parts = { cloudAgent: empty, a2a: empty, conductor: fromConductor(staleSnap) };
+    assert.ok(
+      !mergeSnapshot(parts, OK_SOURCES, { now: NOW }).nodes.some(
+        (n) => n.id === "conductor:task:ctOld"
+      )
+    );
+    assert.ok(
+      mergeSnapshot(parts, OK_SOURCES, { now: NOW, showCompleted: true }).nodes.some(
+        (n) => n.id === "conductor:task:ctOld"
+      )
+    );
+  });
+});

tests/unit/ui/orchestrationToFlow.test.ts

@@ -0,0 +1,56 @@
+/** Run: node --import tsx/esm --test tests/unit/ui/orchestrationToFlow.test.ts */
+import { describe, it } from "node:test";
+import assert from "node:assert/strict";
+import { orchestrationToFlow } from "../../../src/app/(dashboard)/dashboard/orchestration/model/orchestrationToFlow.ts";
+import type { OrchSnapshot } from "../../../src/app/(dashboard)/dashboard/orchestration/model/orchestrationTypes.ts";
+
+const snap: OrchSnapshot = {
+  nodes: [
+    { id: "orchestrator", kind: "orchestrator", label: "OmniRoute" },
+    { id: "source:a2a", kind: "source", source: "a2a", label: "A2A" },
+    { id: "a2a:t1", kind: "work", source: "a2a", state: "running", label: "smart-routing" },
+    { id: "a2a:t2", kind: "work", source: "a2a", state: "failed", label: "cost-analysis" },
+  ],
+  edges: [
+    { id: "e1", from: "orchestrator", to: "source:a2a", kind: "owns", active: false },
+    { id: "e2", from: "source:a2a", to: "a2a:t1", kind: "owns", active: true },
+    { id: "e3", from: "source:a2a", to: "a2a:t2", kind: "owns", active: false },
+  ],
+  sources: [],
+  generatedAt: "2026-08-30T12:00:00Z",
+};
+
+describe("orchestrationToFlow", () => {
+  it("puts each kind on its own Y layer and is deterministic", () => {
+    const a = orchestrationToFlow(snap);
+    const b = orchestrationToFlow(snap);
+    assert.deepEqual(
+      a.nodes.map((n) => n.position),
+      b.nodes.map((n) => n.position)
+    );
+    const ys = new Map(a.nodes.map((n) => [n.id, n.position.y]));
+    assert.equal(ys.get("orchestrator"), 0);
+    assert.equal(ys.get("source:a2a"), 150);
+    assert.equal(ys.get("a2a:t1"), 320);
+  });
+  it("active edge is animated; edge to failed work is red", () => {
+    const { edges } = orchestrationToFlow(snap);
+    assert.equal(edges.find((e) => e.id === "e2")?.animated, true);
+    const failedEdge = edges.find((e) => e.id === "e3");
+    assert.equal((failedEdge?.style as { stroke?: string })?.stroke, "#ef4444");
+  });
+  it("fitKey only tracks the set of work ids", () => {
+    const k1 = orchestrationToFlow(snap).fitKey;
+    const stateChanged = {
+      ...snap,
+      nodes: snap.nodes.map((n) => (n.id === "a2a:t1" ? { ...n, state: "succeeded" as const } : n)),
+    };
+    assert.equal(orchestrationToFlow(stateChanged).fitKey, k1);
+    const nodeRemoved = {
+      ...snap,
+      nodes: snap.nodes.filter((n) => n.id !== "a2a:t2"),
+      edges: snap.edges.filter((e) => e.to !== "a2a:t2"),
+    };
+    assert.notEqual(orchestrationToFlow(nodeRemoved).fitKey, k1);
+  });
+});

tests/unit/ui/overviewProjection.test.ts

@@ -0,0 +1,67 @@
+/** Run: node --import tsx/esm --test tests/unit/ui/overviewProjection.test.ts */
+import { describe, it } from "node:test";
+import assert from "node:assert/strict";
+import { overviewProjection } from "../../../src/app/(dashboard)/dashboard/orchestration/model/overviewProjection.ts";
+import type { OrchSnapshot } from "../../../src/app/(dashboard)/dashboard/orchestration/model/orchestrationTypes.ts";
+
+const snap: OrchSnapshot = {
+  nodes: [
+    { id: "orchestrator", kind: "orchestrator", label: "OmniRoute" },
+    { id: "cloud-agent:1", kind: "work", source: "cloud-agent", state: "running", label: "a" },
+    {
+      id: "cloud-agent:2",
+      kind: "work",
+      source: "cloud-agent",
+      state: "waiting_approval",
+      label: "b",
+    },
+    {
+      id: "a2a:3",
+      kind: "work",
+      source: "a2a",
+      state: "failed",
+      label: "c",
+      updatedAt: "2026-08-30T11:00:00Z",
+    },
+    { id: "a2a:3:activity", kind: "activity", source: "a2a", state: "running", label: "noise" },
+  ],
+  edges: [],
+  sources: [],
+  generatedAt: "2026-08-30T12:00:00Z",
+};
+
+describe("overviewProjection", () => {
+  it("counts only work nodes and adds comboActive to running", () => {
+    const { counts } = overviewProjection(snap, 4);
+    assert.equal(counts.running, 1 + 4);
+    assert.equal(counts.waiting_approval, 1);
+    assert.equal(counts.failed, 1);
+    assert.equal(counts.queued, 0);
+  });
+  it("folds terminals into the done column", () => {
+    const { columns } = overviewProjection(snap, 0);
+    assert.equal(columns.done.length, 1);
+    assert.equal(columns.done[0].id, "a2a:3");
+    assert.equal(columns.running.length, 1);
+  });
+  it("folds an overflow node's droppedByState into counts but not into columns", () => {
+    const snapWithOverflow: OrchSnapshot = {
+      ...snap,
+      nodes: [
+        ...snap.nodes,
+        {
+          id: "overflow:cloud-agent",
+          kind: "overflow",
+          source: "cloud-agent",
+          label: "+7 more",
+          droppedByState: { running: 5, failed: 2 },
+        },
+      ],
+    };
+    const { counts, columns } = overviewProjection(snapWithOverflow, 0);
+    assert.equal(counts.running, 1 + 5);
+    assert.equal(counts.failed, 1 + 2);
+    assert.equal(columns.running.length, 1);
+    assert.equal(columns.done.length, 1);
+  });
+});

tests/unit/ui/useOrchestrationSnapshot.test.tsx

@@ -0,0 +1,121 @@
+// @vitest-environment jsdom
+import React, { act } from "react";
+import { createRoot } from "react-dom/client";
+import { describe, it, expect, beforeEach, afterEach, vi } from "vitest";
+
+// Capture the onEvent handler the hook registers on the requests channel.
+let capturedOnEvent: ((p: { channel: string }) => void) | null = null;
+vi.mock("@/hooks/useLiveDashboard", () => ({
+  useLiveDashboard: (opts: { onEvent?: (p: { channel: string }) => void }) => {
+    capturedOnEvent = opts.onEvent ?? null;
+    return { connection: { isConnected: true }, events: [] };
+  },
+}));
+
+import { useOrchestrationSnapshot } from "@/app/(dashboard)/dashboard/orchestration/hooks/useOrchestrationSnapshot";
+
+function HookProbe({
+  onRender,
+}: {
+  onRender: (v: ReturnType<typeof useOrchestrationSnapshot>) => void;
+}) {
+  onRender(useOrchestrationSnapshot());
+  return null;
+}
+
+const okJson = (body: unknown) =>
+  Promise.resolve({ ok: true, json: () => Promise.resolve(body) } as Response);
+
+describe("useOrchestrationSnapshot", () => {
+  let container: HTMLDivElement;
+  let root: ReturnType<typeof createRoot>;
+  beforeEach(() => {
+    vi.useFakeTimers();
+    container = document.createElement("div");
+    document.body.appendChild(container);
+    root = createRoot(container);
+  });
+  afterEach(() => {
+    act(() => root.unmount());
+    container.remove();
+    vi.useRealTimers();
+    vi.restoreAllMocks();
+  });
+
+  it("polls the three endpoints and builds a snapshot; a failed source keeps the last good data", async () => {
+    let latest: ReturnType<typeof useOrchestrationSnapshot> | null = null;
+    const task = {
+      id: "t1",
+      providerId: "devin",
+      status: "running",
+      prompt: "p",
+      source: { repoName: "r", repoUrl: "https://x" },
+      options: {},
+      activities: [],
+      createdAt: "2026-08-30T10:00:00Z",
+      updatedAt: "2026-08-30T10:00:00Z",
+    };
+    const fetchMock = vi.fn((url: string) => {
+      if (url.startsWith("/api/v1/agents/tasks")) return okJson({ data: [task] });
+      if (url.startsWith("/api/a2a/tasks"))
+        return okJson({ tasks: [], total: 0, limit: 200, offset: 0 });
+      return okJson({ offline: false, runners: [], tasks: [] });
+    });
+    vi.stubGlobal("fetch", fetchMock);
+
+    await act(async () => {
+      root.render(
+        <HookProbe
+          onRender={(v) => {
+            latest = v;
+          }}
+        />
+      );
+    });
+    await act(async () => {
+      await vi.advanceTimersByTimeAsync(10);
+    });
+    expect(latest!.snapshot.nodes.some((n) => n.id === "cloud-agent:t1")).toBe(true);
+
+    // Second tick: cloud agent fails — node must survive from the last good photo, source marked stale.
+    fetchMock.mockImplementation((url: string) => {
+      if (url.startsWith("/api/v1/agents/tasks")) return Promise.reject(new Error("boom"));
+      if (url.startsWith("/api/a2a/tasks"))
+        return okJson({ tasks: [], total: 0, limit: 200, offset: 0 });
+      return okJson({ offline: false, runners: [], tasks: [] });
+    });
+    await act(async () => {
+      await vi.advanceTimersByTimeAsync(5_100);
+    });
+    expect(latest!.snapshot.nodes.some((n) => n.id === "cloud-agent:t1")).toBe(true);
+    const st = latest!.snapshot.sources.find((s) => s.source === "cloud-agent");
+    expect(st?.ok).toBe(false);
+  });
+
+  it("a requests-channel WS event triggers a debounced immediate refetch", async () => {
+    const fetchMock = vi.fn((url: string) => {
+      if (url.startsWith("/api/v1/agents/tasks")) return okJson({ data: [] });
+      if (url.startsWith("/api/a2a/tasks"))
+        return okJson({ tasks: [], total: 0, limit: 200, offset: 0 });
+      return okJson({ offline: false, runners: [], tasks: [] });
+    });
+    vi.stubGlobal("fetch", fetchMock);
+    await act(async () => {
+      root.render(<HookProbe onRender={() => {}} />);
+    });
+    await act(async () => {
+      await vi.advanceTimersByTimeAsync(10);
+    });
+    const callsAfterMount = fetchMock.mock.calls.length;
+
+    act(() => {
+      capturedOnEvent?.({ channel: "requests" });
+      capturedOnEvent?.({ channel: "requests" });
+    });
+    await act(async () => {
+      await vi.advanceTimersByTimeAsync(1_100);
+    });
+    // Two burst events → exactly ONE extra round of 3 fetches (debounce), not two.
+    expect(fetchMock.mock.calls.length).toBe(callsAfterMount + 3);
+  });
+});