fix(providers): honor fusion config.judgeModel for final synthesis (#6455) (#6607)

The fusion single-survivor degrade path (added for #6454) returned the
lone panel answer directly whenever only one panelist succeeded, ignoring
an explicitly configured judgeModel. With default minPanel=2 and a 2-model
panel, any single flaky panelist forced this path every request, so the
configured judge never ran and the response .model reflected a panel member.

The judge is now still invoked to synthesize a lone surviving answer when
judgeModel is explicitly configured; the direct-answer shortcut is kept only
for the implicit case (no judgeModel, judge defaults to panel[0]).

CHANGELOG.md

@@ -10,6 +10,7 @@ _Living section — bullets land here as PRs merge into `release/v3.8.47` (paral
 
 ### 🐛 Bug Fixes
 
+- **fix(providers):** `fusion` combo strategy silently returned a panel member's raw answer instead of the configured `config.judgeModel` synthesis ([#6455](https://github.com/diegosouzapw/OmniRoute/issues/6455)) — `handleFusionChat()`'s single-survivor "degrade gracefully" path (added for #6454) returned the lone panel answer directly whenever only one panelist succeeded, regardless of whether an explicit `judgeModel` was configured; with the default `minPanel: 2` and a 2-model panel, any single flaky/rate-limited panelist forced this path on every request, so the configured judge (e.g. `auto/claude-opus`) was never invoked and the client-visible `.model` reflected whichever panelist happened to survive. The judge is now still invoked to synthesize a lone surviving answer whenever `judgeModel` is explicitly configured; the cheap direct-answer shortcut is kept only for the implicit case (no `judgeModel` set, where the "judge" is just `panel[0]`). Regression guard: `tests/unit/fusion-judge-model-6455.test.ts` + updated `tests/unit/combo-fusion-strategy.test.ts`. (thanks @chirag127)
 - **fix(providers):** image/diffusion models discovered from an upstream catalog (e.g. HuggingFace's live `/v1/models`) are no longer advertised as chat models ([#6457](https://github.com/diegosouzapw/OmniRoute/issues/6457)) — the chat catalog builder defaulted synced models with no modality info to `endpoints: ["chat"]`, so `huggingface/stabilityai/stable-diffusion-xl-base-1.0` showed up in the chat `/v1/models` listing and returned `400 "not a chat model"` when called. `catalog.ts` now skips any synced model already registered as an image model for that provider (via the new `isRegisteredImageModel()`), leaving `getAllImageModels()` to list it with the correct `type: "image"`. Regression guard: `tests/unit/image-model-not-in-chat-catalog-6457.test.ts`.
 - **fix(test):** replace the bare `expect(true).toBe(true)` tautology in `playground-api-tab.test.tsx`'s SSE test and close the `check:test-masking` gap that let it slip through for a full cycle ([#6404](https://github.com/diegosouzapw/OmniRoute/issues/6404)) — a prior pass (#6548) had already swapped the literal to `expect(sendBtn).toBeDefined()`, but that stayed just as vacuous: the test's fetch mock returned an empty `/v1/models` list, so `ApiTab`'s Send button is always `disabled` (`!selectedModel`) and the SSE branch never runs — the "SSE infra is verified" comment was never true. The test now mocks a real model, drives the model `<select>` to enable Send, asserts `sendBtn.disabled === false` before clicking, and asserts the streamed SSE delta (`"Hello!"`) actually reached the response editor. Root cause on the detector side: `check-test-masking.mjs`'s tautology subcheck only compares base-vs-HEAD counts within a PR's own diff (`headExtTaut > baseExtTaut`) and no-ops locally when `GITHUB_BASE_SHA`/`GITHUB_BASE_REF` are unset ("sem base ref — pulando") — so a tautology merged once, or checked with a bare local run, was invisible forever after. Added a new always-on, PR-independent absolute-floor scan (`scanBareTautologies` + `countBareTautologies`) over every git-tracked test file for the bare `expect(true).toBe(true)` / `assert.equal(1,1)` / `assert.strictEqual(1,1)` patterns specifically (deliberately excluding `assert.ok(true)`, which has ~15 pre-existing verified-legitimate try/catch-fallback uses repo-wide and stays governed by the lenient diff-only subcheck) — verified zero pre-existing hits repo-wide once this file was fixed, so the new floor is safe to enforce unconditionally. Regression guard: `tests/unit/check-test-masking.test.ts` (new `scanBareTautologies`/`countBareTautologies` cases) + `tests/unit/ui/playground-api-tab.test.tsx`. (thanks @chirag127)
 - **fix(oauth):** Codex/ChatGPT (and every other OAuth provider) connection stays stuck showing "Auth Failed" even after a genuinely successful token refresh ([#6352](https://github.com/diegosouzapw/OmniRoute/issues/6352)) — `updateProviderCredentials()` (the shared `onPersist` callback for the manual "Refresh token" route, the reactive per-request refresh in `chat.ts`, and the Codex/Claude auth-file importers) correctly reused the stored `refresh_token`, persisted the new `access_token`, and replaced a rotated `refresh_token`, but never cleared the stale `testStatus`/`lastError*`/`errorCode` fields left over from a prior expired/invalid refresh or upstream 401/403 — only the separate background health-check sweep did that clearing. A successful refresh now resets `testStatus` to `"active"` and clears `lastError`, `lastErrorAt`, `lastErrorType`, `lastErrorSource`, and `errorCode` (an explicit `testStatus` from the caller still wins). Regression guard: `tests/unit/codex-oauth-refresh-persist-6352.test.ts`.

open-sse/services/fusion.ts

@@ -250,7 +250,8 @@ export async function handleFusionChat({
   // Honor user-supplied minPanel down to 1: with 1 survivor we still degrade
   // gracefully via the answers.length===1 branch below (issue #6454).
   const minPanel = Math.min(Math.max(1, cfg.minPanel), panel.length);
-  const judge = judgeModel && judgeModel.trim() ? judgeModel.trim() : panel[0];
+  const hasExplicitJudge = Boolean(judgeModel && judgeModel.trim());
+  const judge = hasExplicitJudge ? (judgeModel as string).trim() : panel[0];
   log.info(
     "FUSION",
     `Combo "${comboName ?? ""}" | panel=${panel.length} [${panel.join(", ")}] | judge=${judge} | quorum=${minPanel}`
@@ -335,11 +336,24 @@ export async function handleFusionChat({
     );
   }
   if (answers.length === 1) {
+    // No explicit judgeModel configured: the "judge" is just panel[0], so
+    // synthesizing from a single source through itself would be redundant —
+    // answer directly with the lone survivor (issue #6454).
+    if (!hasExplicitJudge) {
+      log.info(
+        "FUSION",
+        `Only ${answers[0].model} succeeded — answering directly (no fusion)`
+      );
+      return handleSingleModel(body, answers[0].model);
+    }
+    // An explicit judgeModel IS configured: honor it even with a single
+    // surviving panel answer, rather than silently substituting the panel
+    // member for the configured judge (issue #6455). The judge still adds
+    // value reviewing/polishing a lone source per its documented contract.
     log.info(
       "FUSION",
-      `Only ${answers[0].model} succeeded — answering directly (no fusion)`
+      `Only ${answers[0].model} succeeded — judging single answer with ${judge}`
     );
-    return handleSingleModel(body, answers[0].model);
   }
 
   // 4. Judge analyzes + writes one final answer (streams to client if requested).

tests/unit/combo-fusion-strategy.test.ts

@@ -168,7 +168,7 @@ test("fusion: proceeds on quorum without waiting for a straggler (grace window)"
   assert.ok(!/slow/.test(judgeText), "straggler answer should not appear in the judge prompt");
 });
 
-test("fusion: returns the lone survivor directly when only one panel model succeeds", async () => {
+test("fusion: returns the lone survivor directly when only one panel model succeeds and no judgeModel is configured", async () => {
   const seen: string[] = [];
   const handleSingleModel = async (_b: Body, m: string) => {
     seen.push(m);
@@ -176,6 +176,37 @@ test("fusion: returns the lone survivor directly when only one panel model succe
     return errResponse(500);
   };
   await handleComboChat({
+    body: { messages: [{ role: "user", content: "Q" }] },
+    combo: fusionCombo(["p/ok", "p/bad"], {
+      // No judgeModel configured: the implicit "judge" is just panel[0], so
+      // synthesizing a single source through itself is redundant.
+      fusionTuning: { minPanel: 2, stragglerGraceMs: 50, panelHardTimeoutMs: 5000 },
+    }),
+    handleSingleModel,
+    log,
+    settings: {},
+    allCombos: [],
+  });
+  // No judge call — single answer + no explicit judge means there is nothing to fuse.
+  assert.ok(
+    !seen.includes("p/judge"),
+    "judge should not be invoked when only one panel model survives and no judgeModel is set"
+  );
+});
+
+// #6455: when an explicit judgeModel IS configured, the lone-survivor degrade
+// path used to silently return the raw panel answer, never invoking the
+// configured judge. See tests/unit/fusion-judge-model-6455.test.ts for the
+// full regression guard.
+test("fusion: honors an explicit judgeModel even with a single surviving panel answer", async () => {
+  const seen: string[] = [];
+  const handleSingleModel = async (_b: Body, m: string) => {
+    seen.push(m);
+    if (m === "p/ok") return okResponse("lone");
+    if (m === "p/judge") return okResponse("JUDGED");
+    return errResponse(500);
+  };
+  const res = await handleComboChat({
     body: { messages: [{ role: "user", content: "Q" }] },
     combo: fusionCombo(["p/ok", "p/bad"], {
       judgeModel: "p/judge",
@@ -186,11 +217,12 @@ test("fusion: returns the lone survivor directly when only one panel model succe
     settings: {},
     allCombos: [],
   });
-  // No judge call — single answer means there is nothing to fuse.
   assert.ok(
-    !seen.includes("p/judge"),
-    "judge should not be invoked when only one panel model survives"
+    seen.includes("p/judge"),
+    "explicit judgeModel should still be invoked to synthesize a single panel answer"
   );
+  assert.equal(seen[seen.length - 1], "p/judge");
+  assert.equal(res.status, 200);
 });
 
 test("fusion: returns 503 when the whole panel fails", async () => {

tests/unit/fusion-judge-model-6455.test.ts

@@ -0,0 +1,142 @@
+/**
+ * Regression guard for #6455 — fusion combo silently returning a panel member
+ * instead of the configured judgeModel's synthesis.
+ *
+ * Root cause: handleFusionChat()'s "degrade gracefully" path (added for #6454)
+ * returned the lone surviving panel answer directly whenever only one panel
+ * member succeeded — regardless of whether an explicit `config.judgeModel`
+ * was configured. With a 2-model panel and the default `minPanel: 2`, any
+ * single flaky/rate-limited panelist forces this path on *every* request, so
+ * the configured judge (e.g. "auto/claude-opus") was never invoked and the
+ * client-visible `.model` field always reflected whichever panel member
+ * happened to survive that request (case (a): judge genuinely not run).
+ *
+ * Fix: when an explicit judgeModel is configured, still route the lone
+ * surviving answer through the judge for synthesis instead of returning it
+ * raw. Only the truly implicit case (no judgeModel set, "judge" defaults to
+ * panel[0]) keeps the cheap direct-answer shortcut.
+ */
+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-fusion-6455-"));
+process.env.DATA_DIR = TEST_DATA_DIR;
+process.env.API_KEY_SECRET = process.env.API_KEY_SECRET || "fusion-6455-test-secret";
+
+const { handleComboChat } = await import("../../open-sse/services/combo.ts");
+
+type Body = Record<string, unknown>;
+
+function jsonResponse(model: string, content: string): Response {
+  const body = JSON.stringify({ model, choices: [{ message: { role: "assistant", content } }] });
+  return new Response(body, { status: 200, headers: { "Content-Type": "application/json" } });
+}
+
+function errResponse(status = 500): Response {
+  return new Response(JSON.stringify({ error: { message: "boom" } }), {
+    status,
+    headers: { "Content-Type": "application/json" },
+  });
+}
+
+const noop = () => {};
+const log = { info: noop, warn: noop, debug: noop, error: noop };
+
+function fusionCombo(models: string[], extra: Record<string, unknown> = {}) {
+  return {
+    name: "fusion-free",
+    strategy: "fusion",
+    models: models.map((m) => ({ model: m })),
+    config: extra,
+  };
+}
+
+test("6455: judge synthesizes even a single surviving panel answer when judgeModel is explicit", async () => {
+  const calls: string[] = [];
+  const handleSingleModel = async (_b: Body, m: string) => {
+    calls.push(m);
+    if (m === "gpt-5.5-panelist") return jsonResponse(m, "panel answer");
+    if (m === "sonar-pro-panelist") return errResponse(503); // flaky panel member
+    if (m === "auto/claude-opus") return jsonResponse("auto/claude-opus", "JUDGED FINAL");
+    throw new Error(`unexpected model ${m}`);
+  };
+
+  const res = await handleComboChat({
+    body: { messages: [{ role: "user", content: "hi" }] },
+    combo: fusionCombo(["gpt-5.5-panelist", "sonar-pro-panelist"], {
+      judgeModel: "auto/claude-opus",
+    }),
+    handleSingleModel,
+    log,
+    settings: {},
+    allCombos: [],
+  });
+
+  // (a) The judge model MUST be invoked to synthesize, not skipped.
+  assert.ok(
+    calls.includes("auto/claude-opus"),
+    `expected the configured judge to be invoked; calls were: ${calls.join(", ")}`
+  );
+  // The judge call is the last one — it consumes the panel answer(s).
+  assert.equal(calls[calls.length - 1], "auto/claude-opus");
+
+  // (b) The response body reflects the judge's synthesis, not the raw panelist.
+  assert.equal(res.status, 200);
+  const json = (await res.json()) as { model?: string; choices?: Array<Record<string, unknown>> };
+  assert.equal(json.model, "auto/claude-opus");
+  const message = json.choices?.[0]?.message as { content?: string } | undefined;
+  assert.equal(message?.content, "JUDGED FINAL");
+});
+
+test("6455: panel still fans out to every member before degrading (no regression)", async () => {
+  const calls: string[] = [];
+  const handleSingleModel = async (_b: Body, m: string) => {
+    calls.push(m);
+    if (m === "p/a") return jsonResponse(m, "answer-a");
+    if (m === "p/b") return jsonResponse(m, "answer-b");
+    if (m === "p/judge") return jsonResponse("p/judge", "FUSED");
+    throw new Error(`unexpected model ${m}`);
+  };
+
+  const res = await handleComboChat({
+    body: { messages: [{ role: "user", content: "hi" }] },
+    combo: fusionCombo(["p/a", "p/b"], { judgeModel: "p/judge" }),
+    handleSingleModel,
+    log,
+    settings: {},
+    allCombos: [],
+  });
+
+  assert.deepEqual(calls.slice(0, 2).sort(), ["p/a", "p/b"]);
+  assert.equal(calls[2], "p/judge");
+  assert.equal(res.status, 200);
+  const json = (await res.json()) as { model?: string };
+  assert.equal(json.model, "p/judge");
+});
+
+test("6455: implicit judge (no judgeModel configured) still short-circuits a lone survivor", async () => {
+  const calls: string[] = [];
+  const handleSingleModel = async (_b: Body, m: string) => {
+    calls.push(m);
+    if (m === "p/ok") return jsonResponse(m, "lone");
+    return errResponse(500);
+  };
+
+  await handleComboChat({
+    body: { messages: [{ role: "user", content: "hi" }] },
+    combo: fusionCombo(["p/ok", "p/bad"]), // no judgeModel — defaults to panel[0]
+    handleSingleModel,
+    log,
+    settings: {},
+    allCombos: [],
+  });
+
+  // Panel fan-out calls p/ok and p/bad; the lone survivor is then answered
+  // directly with the client's original body (not a judge-synthesis turn) —
+  // panel[0] IS the implicit judge, so there is nothing to gain by routing
+  // it through a separate synthesis call.
+  assert.deepEqual(calls, ["p/ok", "p/bad", "p/ok"]);
+});