@unotest/judge 0.23.0 → 0.24.0

package/CHANGELOG.md

@@ -1,5 +1,73 @@
 # @unotest/judge
 
+## [0.24.0] - 2026-08-26
+
+### Minor Changes
+
+- Judge: rendered text, one env namespace, preamble and majority vote.
+
+  **Breaking (judge env, no aliases).** Every `JUDGE_*` variable of the
+  `@unotest/judge` service is now `UNOTEST_JUDGE_*`, matching the client-side
+  names `@unotest/web` already used. `@unotest/judge` shipped a day before
+  this change with no known installs outside the monorepo, so the rename lands
+  without a deprecation window — the window would never be cheaper. Migration:
+
+  | Old                         | New                                         |
+  | --------------------------- | ------------------------------------------- |
+  | `JUDGE_PROVIDER`            | `UNOTEST_JUDGE_PROVIDER`                    |
+  | `JUDGE_MODEL`               | `UNOTEST_JUDGE_MODEL`                       |
+  | `JUDGE_RETRIES`             | `UNOTEST_JUDGE_RETRIES`                     |
+  | `JUDGE_TIMEOUT_MS`          | `UNOTEST_JUDGE_CALL_TIMEOUT_MS`             |
+  | `JUDGE_ACCESS_TOKEN`        | `UNOTEST_JUDGE_ACCESS_TOKEN`                |
+  | `JUDGE_CLAUDE_BIN`          | `UNOTEST_JUDGE_CLAUDE_BIN`                  |
+  | `JUDGE_HOST` / `JUDGE_PORT` | `UNOTEST_JUDGE_HOST` / `UNOTEST_JUDGE_PORT` |
+  | `JUDGE_TOKEN`               | `UNOTEST_JUDGE_TOKEN`                       |
+
+  Two names were deliberately not a straight prefixing. The service's per-call
+  budget became `UNOTEST_JUDGE_CALL_TIMEOUT_MS` because `UNOTEST_JUDGE_TIMEOUT_MS`
+  was already taken by the client's whole-request budget — same family, different
+  question. And the service's bearer token merged into the client's
+  `UNOTEST_JUDGE_TOKEN`: it is one secret, and in a shared overlay the two ends
+  have to carry the same value anyway.
+
+  **Behavioral fix: rendered text.** `assertJudge` and the DSL query
+  `getInnerText(loc)` now read `innerText`, not `textContent`. Both promised
+  rendered text and delivered the raw source-order concatenation: line breaks
+  gone, `display:none` subtrees and `<script>` contents folded in. A rubric
+  about structure ("the answer is three bullets") was unjudgeable, and
+  `getInnerText` returned text no user can see. Scenarios that relied on
+  hidden text reaching these two functions will now see it excluded.
+
+  **Vertex fixes.** `google-auth-library` moved from optional peer to a regular
+  dependency: as a peer it did not resolve under `npx @unotest/judge` at all,
+  which broke the advertised keyless path for anyone who never installs the
+  package locally. It is still loaded lazily, so non-vertex runs pay nothing.
+  Expired Application Default Credentials now produce
+  "run `gcloud auth application-default login`" instead of a raw Google OAuth
+  blob, read off the structured response body rather than message text.
+
+  **New: `UNOTEST_JUDGE_PREAMBLE`** (web side) — text prepended to every rubric
+  in the project, for standing context ("the app under test is a support bot,
+  Russian UI") that would otherwise be copy-pasted into each rubric. It applies
+  in both local and remote mode; a remote judge stays project-agnostic.
+
+  **New: `UNOTEST_JUDGE_VOTE=N`** (service side, odd N, default 1 = off) — N
+  independent calls decided by majority. It replaces `UNOTEST_JUDGE_RETRIES`
+  rather than stacking with it: re-asking is a deliberate bias toward `pass`,
+  a vote is deliberately symmetric, and running both would quietly restore the
+  bias the vote was chosen to remove. Ballots run concurrently, so a vote costs
+  N provider calls but roughly one call of wall-clock.
+
+  **Wire protocol.** New error code `config` on the judge's error responses:
+  the service's own setup is broken (bad env, expired credentials) and the
+  message carries the fix. `@unotest/web` surfaces it as a `ConfigError`
+  instead of burying it as a generic "judge unavailable" fault.
+
+### Patch Changes
+
+- Updated dependencies
+  - @unotest/protocol@0.24.0
+
 ## [0.23.0] - 2026-08-26
 
 ### Minor Changes

package/README.md

@@ -15,7 +15,7 @@ output.
 ## Run
 
 ```
-JUDGE_PROVIDER=vertex \
+UNOTEST_JUDGE_PROVIDER=vertex \
 GOOGLE_CLOUD_PROJECT=my-project \
 GOOGLE_CLOUD_LOCATION=europe-west1 \
 npx @unotest/judge
@@ -24,19 +24,24 @@ npx @unotest/judge
 The service listens on `127.0.0.1:8790` by default. Point `@unotest/web`
 at it with `UNOTEST_JUDGE_MODE=remote` + `UNOTEST_JUDGE_URL=http://127.0.0.1:8790`,
 or skip the HTTP hop entirely with `UNOTEST_JUDGE_MODE=local` (in-process,
-requires `@unotest/judge` installed in the project).
+requires `@unotest/judge` installed in the project — a project that runs
+unotest through `npx`, with nothing installed locally, can only use
+`remote`).
 
 ## Providers
 
 - **`vertex`** — Google Vertex AI via Application Default Credentials.
   No API keys: auth comes from ADC (`gcloud auth application-default login`,
   workload identity, or `GOOGLE_APPLICATION_CREDENTIALS`). Requires the
-  optional peer `google-auth-library` (or a static `JUDGE_ACCESS_TOKEN`).
-  Temperature is pinned to 0.
+  `google-auth-library` (a dependency of this package, loaded only when
+  this provider runs) — or a static `UNOTEST_JUDGE_ACCESS_TOKEN` from
+  `gcloud auth print-access-token`. Temperature is pinned to 0.
 - **`claude`** — the local Claude Code CLI (`claude -p`), spawned per
   judgement without shell interpretation. Auth comes from your Claude Code
-  session, so a subscription works with no API key. `JUDGE_MODEL` is passed
-  as `--model` (aliases like `sonnet` work); unset uses the CLI's default.
+  session, so a subscription works with no API key — this is the provider
+  to reach for when no cloud credentials are at hand. `UNOTEST_JUDGE_MODEL`
+  is passed as `--model` (aliases like `sonnet` work); unset uses the CLI's
+  default.
 - **`gemini`** — the Gemini API with `GEMINI_API_KEY`. Same wire format as
   `vertex`, temperature pinned to 0.
 - **`openai`** — OpenAI chat completions with `OPENAI_API_KEY`
@@ -57,24 +62,27 @@ zero provider SDKs.
 
 | Variable | Meaning | Default |
 | --- | --- | --- |
-| `JUDGE_PROVIDER` | `fake` \| `vertex` \| `claude` \| `gemini` \| `openai` \| `anthropic` | required |
-| `JUDGE_MODEL` | model id | vertex/gemini `gemini-2.5-flash`, openai `gpt-5-mini`, anthropic `claude-haiku-4-5`, claude: CLI default |
-| `JUDGE_RETRIES` | extra provider calls on `fail`; first `pass` wins | `1` |
-| `JUDGE_TIMEOUT_MS` | per-call budget, ms | `30000` (`120000` for claude) |
+| `UNOTEST_JUDGE_PROVIDER` | `fake` \| `vertex` \| `claude` \| `gemini` \| `openai` \| `anthropic` | required |
+| `UNOTEST_JUDGE_MODEL` | model id | vertex/gemini `gemini-2.5-flash`, openai `gpt-5-mini`, anthropic `claude-haiku-4-5`, claude: CLI default |
+| `UNOTEST_JUDGE_RETRIES` | extra provider calls on `fail`; first `pass` wins | `1` |
+| `UNOTEST_JUDGE_VOTE` | odd N; N independent calls, the majority decides. Replaces retries rather than stacking with them | `1` (off) |
+| `UNOTEST_JUDGE_CALL_TIMEOUT_MS` | budget for ONE provider call, ms | `30000` (`120000` for claude) |
 | `GOOGLE_CLOUD_PROJECT` | vertex: ADC project | required for vertex |
 | `GOOGLE_CLOUD_LOCATION` | vertex: ADC location | required for vertex |
-| `JUDGE_ACCESS_TOKEN` | vertex: static bearer override (skips ADC) | — |
+| `UNOTEST_JUDGE_ACCESS_TOKEN` | vertex: static bearer override (skips ADC) | — |
 | `GEMINI_API_KEY` | gemini: API key | required for gemini |
 | `OPENAI_API_KEY` | openai: API key | required for openai |
 | `ANTHROPIC_API_KEY` | anthropic: API key | required for anthropic |
-| `JUDGE_CLAUDE_BIN` | claude: binary override | `claude` |
-| `JUDGE_HOST` / `JUDGE_PORT` | bind address | `127.0.0.1` / `8790` |
-| `JUDGE_TOKEN` | bearer token required on every `/judge` call | off |
+| `UNOTEST_JUDGE_CLAUDE_BIN` | claude: binary override | `claude` |
+| `UNOTEST_JUDGE_HOST` / `UNOTEST_JUDGE_PORT` | bind address | `127.0.0.1` / `8790` |
+| `UNOTEST_JUDGE_TOKEN` | bearer token required on every `/judge` call. `@unotest/web` reads the SAME name to send it | off |
 
 ## HTTP API
 
 - `POST /judge` — body `{"rubric": "...", "text": "..."}` → a
-  `JudgeVerdict` (above). Errors: `{"error": "...", "code": "unauthorized" | "bad-request" | "provider-error" | "internal"}`.
+  `JudgeVerdict` (above). Errors: `{"error": "...", "code": "unauthorized" | "bad-request" | "provider-error" | "config" | "internal"}`.
+  `config` means the operator must fix this service's setup (bad env,
+  expired ADC) and the message says what to run.
 - `GET /health` — `{"ok": true}`.
 
 Wire types are shared through `@unotest/protocol` (`JudgeRequest`,

package/dist/{chunk-4VUHPR26.js → chunk-74FL5RWE.js}

@@ -33,7 +33,7 @@ var FakeJudgeProvider = class {
     if (constraints.length === 0) {
       return {
         pass: false,
-        reasoning: 'fake provider: rubric has no parseable constraints \u2014 use lines like "must contain: <substring>" / "must not contain: <substring>", or switch to a real provider (JUDGE_PROVIDER=vertex|claude|gemini|openai|anthropic)',
+        reasoning: 'fake provider: rubric has no parseable constraints \u2014 use lines like "must contain: <substring>" / "must not contain: <substring>", or switch to a real provider (UNOTEST_JUDGE_PROVIDER=vertex|claude|gemini|openai|anthropic)',
         model: FAKE_MODEL_ID
       };
     }
@@ -89,6 +89,55 @@ function parseVerdictReply(text, model) {
   };
 }
 
+// src/providers/adc-auth.ts
+async function buildAdcTokenSource() {
+  const specifier = "google-auth-library";
+  let mod;
+  try {
+    mod = await import(specifier);
+  } catch (e) {
+    throw new JudgeConfigError(
+      `Vertex provider could not load "google-auth-library" (a dependency of this package \u2014 a broken install?): ${e instanceof Error ? e.message : String(e)}. Workaround: pass a token via UNOTEST_JUDGE_ACCESS_TOKEN (\`gcloud auth print-access-token\`)`
+    );
+  }
+  const auth = new mod.GoogleAuth({ scopes: ["https://www.googleapis.com/auth/cloud-platform"] });
+  return async () => {
+    let token;
+    try {
+      token = await auth.getAccessToken();
+    } catch (e) {
+      throw describeAdcFailure(e);
+    }
+    if (!token) {
+      throw new JudgeConfigError(
+        "ADC produced no access token \u2014 run `gcloud auth application-default login` or set GOOGLE_APPLICATION_CREDENTIALS"
+      );
+    }
+    return token;
+  };
+}
+var REAUTH_CODES = /* @__PURE__ */ new Set(["invalid_grant", "invalid_rapt"]);
+function describeAdcFailure(e) {
+  const body = oauthErrorBody(e);
+  if (body && (REAUTH_CODES.has(body.error) || REAUTH_CODES.has(body.error_subtype ?? ""))) {
+    return new JudgeConfigError(
+      `Application Default Credentials need re-authentication (${body.error}${body.error_subtype ? `/${body.error_subtype}` : ""}) \u2014 run \`gcloud auth application-default login\``,
+      { oauthError: body.error }
+    );
+  }
+  const detail = body?.error_description ?? body?.error ?? (e instanceof Error ? e.message : String(e));
+  return new JudgeConfigError(
+    `Application Default Credentials failed: ${detail} \u2014 check GOOGLE_CLOUD_PROJECT / GOOGLE_CLOUD_LOCATION and \`gcloud auth application-default login\``
+  );
+}
+function oauthErrorBody(e) {
+  if (typeof e !== "object" || e === null) return void 0;
+  const data = e.response?.data;
+  if (typeof data !== "object" || data === null) return void 0;
+  const body = data;
+  return typeof body.error === "string" ? body : void 0;
+}
+
 // src/providers/generate-content.ts
 function generateContentBody(request) {
   return {
@@ -175,27 +224,6 @@ var VertexJudgeProvider = class {
     return this.tokenSource();
   }
 };
-async function buildAdcTokenSource() {
-  const specifier = "google-auth-library";
-  let mod;
-  try {
-    mod = await import(specifier);
-  } catch {
-    throw new JudgeConfigError(
-      'Vertex provider needs Application Default Credentials \u2014 install the optional peer "google-auth-library" (`npm i google-auth-library`), or pass a token via JUDGE_ACCESS_TOKEN'
-    );
-  }
-  const auth = new mod.GoogleAuth({ scopes: ["https://www.googleapis.com/auth/cloud-platform"] });
-  return async () => {
-    const token = await auth.getAccessToken();
-    if (!token) {
-      throw new JudgeConfigError(
-        "ADC produced no access token \u2014 run `gcloud auth application-default login` or set GOOGLE_APPLICATION_CREDENTIALS"
-      );
-    }
-    return token;
-  };
-}
 
 // src/providers/gemini.ts
 var GeminiJudgeProvider = class {
@@ -356,7 +384,7 @@ function toTypedError(e, bin, timeoutMs) {
   const err = e;
   if (err.code === "ENOENT") {
     return new JudgeConfigError(
-      `Claude Code CLI not found ("${bin}") \u2014 install it (https://claude.com/claude-code) or point JUDGE_CLAUDE_BIN at the binary`,
+      `Claude Code CLI not found ("${bin}") \u2014 install it (https://claude.com/claude-code) or point UNOTEST_JUDGE_CLAUDE_BIN at the binary`,
       { bin }
     );
   }
@@ -400,8 +428,23 @@ async function judgeWithRetries(provider, request, retries) {
   }
   return last;
 }
+async function judgeWithVote(provider, request, votes) {
+  const ballots = await Promise.all(
+    Array.from({ length: votes }, () => provider.judgeOnce(request))
+  );
+  const passes = ballots.filter((b) => b.pass);
+  const winners = passes.length * 2 > votes ? passes : ballots.filter((b) => !b.pass);
+  const winner = winners[0];
+  return {
+    verdict: winner.pass ? "pass" : "fail",
+    reasoning: `majority ${winners.length}/${votes}: ${winner.reasoning}`,
+    model: winner.model,
+    attempts: votes
+  };
+}
 
 // src/env.ts
+import { JUDGE_ENV } from "@unotest/protocol";
 var PROVIDER_KINDS = [
   "fake",
   "vertex",
@@ -436,57 +479,66 @@ function intEnv(env, name, fallback, min, max) {
   return n;
 }
 function resolveJudgeServiceEnv(env) {
-  const provider = env.JUDGE_PROVIDER;
+  const provider = env[JUDGE_ENV.provider];
   if (!provider || !PROVIDER_KINDS.includes(provider)) {
     const list = PROVIDER_KINDS.map((k) => `"${k}"`).join(" | ");
     throw new JudgeConfigError(
-      provider === void 0 || provider === "" ? `JUDGE_PROVIDER is not set \u2014 one of ${list} ("fake" is deterministic and CI-safe)` : `JUDGE_PROVIDER must be one of ${list}, got "${provider}"`
+      provider === void 0 || provider === "" ? `${JUDGE_ENV.provider} is not set \u2014 one of ${list} ("fake" is deterministic and CI-safe)` : `${JUDGE_ENV.provider} must be one of ${list}, got "${provider}"`
+    );
+  }
+  const votes = intEnv(env, JUDGE_ENV.vote, 1, 1, 9);
+  if (votes % 2 === 0) {
+    throw new JudgeConfigError(
+      `${JUDGE_ENV.vote} must be odd (1, 3, 5, 7, 9) \u2014 an even vote can tie, and a tie has no verdict; got "${votes}"`
     );
   }
   const resolved = {
     provider,
-    retries: intEnv(env, "JUDGE_RETRIES", DEFAULT_RETRIES, 0, 10),
+    votes,
+    retries: intEnv(env, JUDGE_ENV.retries, DEFAULT_RETRIES, 0, 10),
     timeoutMs: intEnv(
       env,
-      "JUDGE_TIMEOUT_MS",
+      JUDGE_ENV.callTimeoutMs,
       provider === "claude" ? DEFAULT_CLAUDE_TIMEOUT_MS : DEFAULT_TIMEOUT_MS,
       1e3,
       6e5
     )
   };
-  const model = env.JUDGE_MODEL || DEFAULT_MODEL[provider];
+  const model = env[JUDGE_ENV.model] || DEFAULT_MODEL[provider];
   if (model) resolved.model = model;
   if (provider === "vertex") {
     const project = env.GOOGLE_CLOUD_PROJECT;
     const location = env.GOOGLE_CLOUD_LOCATION;
     if (!project || !location) {
       throw new JudgeConfigError(
-        "JUDGE_PROVIDER=vertex requires GOOGLE_CLOUD_PROJECT and GOOGLE_CLOUD_LOCATION (ADC auth, no API keys)"
+        `${JUDGE_ENV.provider}=vertex requires GOOGLE_CLOUD_PROJECT and GOOGLE_CLOUD_LOCATION (ADC auth, no API keys)`
       );
     }
     resolved.project = project;
     resolved.location = location;
-    if (env.JUDGE_ACCESS_TOKEN) resolved.accessToken = env.JUDGE_ACCESS_TOKEN;
+    const accessToken = env[JUDGE_ENV.accessToken];
+    if (accessToken) resolved.accessToken = accessToken;
   }
   const keyEnvName = API_KEY_ENV[provider];
   if (keyEnvName) {
     const apiKey = env[keyEnvName];
     if (!apiKey) {
-      throw new JudgeConfigError(`JUDGE_PROVIDER=${provider} requires ${keyEnvName}`);
+      throw new JudgeConfigError(`${JUDGE_ENV.provider}=${provider} requires ${keyEnvName}`);
     }
     resolved.apiKey = apiKey;
   }
   if (provider === "claude") {
-    resolved.claudeBin = env.JUDGE_CLAUDE_BIN || "claude";
+    resolved.claudeBin = env[JUDGE_ENV.claudeBin] || "claude";
   }
   return resolved;
 }
 function resolveJudgeServerEnv(env) {
   const resolved = {
-    host: env.JUDGE_HOST || "127.0.0.1",
-    port: intEnv(env, "JUDGE_PORT", 8790, 1, 65535)
+    host: env[JUDGE_ENV.host] || "127.0.0.1",
+    port: intEnv(env, JUDGE_ENV.port, 8790, 1, 65535)
   };
-  if (env.JUDGE_TOKEN) resolved.token = env.JUDGE_TOKEN;
+  const token = env[JUDGE_ENV.token];
+  if (token) resolved.token = token;
   return resolved;
 }
 
@@ -532,7 +584,7 @@ function buildProvider(env) {
 function createJudgeService(env, provider) {
   const p = provider ?? buildProvider(env);
   return {
-    judge: (request) => judgeWithRetries(p, request, env.retries)
+    judge: (request) => env.votes > 1 ? judgeWithVote(p, request, env.votes) : judgeWithRetries(p, request, env.retries)
   };
 }
 function createJudgeServiceFromEnv(env) {
@@ -541,7 +593,10 @@ function createJudgeServiceFromEnv(env) {
 
 // src/server.ts
 import { createServer } from "http";
-import { JUDGE_ROUTES } from "@unotest/protocol";
+import {
+  JUDGE_ENV as JUDGE_ENV2,
+  JUDGE_ROUTES
+} from "@unotest/protocol";
 var MAX_BODY_BYTES = 1024 * 1024;
 function startJudgeServer(service, env) {
   const server = createServer((req, res) => {
@@ -571,7 +626,7 @@ async function route(service, env, req, res) {
     return;
   }
   if (env.token && req.headers.authorization !== `Bearer ${env.token}`) {
-    sendError(res, 401, "missing or wrong bearer token (JUDGE_TOKEN)", "unauthorized");
+    sendError(res, 401, `missing or wrong bearer token (${JUDGE_ENV2.token})`, "unauthorized");
     return;
   }
   let body;
@@ -587,7 +642,7 @@ async function route(service, env, req, res) {
     if (e instanceof JudgeProviderError) {
       sendError(res, 502, e.message, "provider-error");
     } else if (e instanceof JudgeConfigError) {
-      sendError(res, 500, e.message, "internal");
+      sendError(res, 500, e.message, "config");
     } else {
       sendError(res, 500, e instanceof Error ? e.message : String(e), "internal");
     }
@@ -657,4 +712,4 @@ export {
   createJudgeServiceFromEnv,
   startJudgeServer
 };
-//# sourceMappingURL=chunk-4VUHPR26.js.map
+//# sourceMappingURL=chunk-74FL5RWE.js.map

package/dist/cli.js

@@ -3,7 +3,7 @@ import {
   resolveJudgeServerEnv,
   resolveJudgeServiceEnv,
   startJudgeServer
-} from "./chunk-4VUHPR26.js";
+} from "./chunk-74FL5RWE.js";
 
 // src/cli.ts
 async function main(argv = process.argv.slice(2)) {
@@ -39,23 +39,26 @@ function printHelp() {
       "Usage: npx @unotest/judge [--help] [--version]",
       "",
       "Env:",
-      "  JUDGE_PROVIDER          fake (deterministic, CI-safe) | vertex (ADC) |",
-      "                          claude (Claude Code CLI, subscription auth) |",
-      "                          gemini | openai | anthropic (API keys)",
-      "  JUDGE_MODEL             model id; defaults per provider:",
-      "                          vertex/gemini gemini-2.5-flash, openai gpt-5-mini,",
-      "                          anthropic claude-haiku-4-5, claude: CLI default",
-      "  JUDGE_RETRIES           extra provider calls on fail (default 1)",
-      "  JUDGE_TIMEOUT_MS        per-call budget, ms (default 30000; claude 120000)",
-      "  GOOGLE_CLOUD_PROJECT    vertex: ADC project",
-      "  GOOGLE_CLOUD_LOCATION   vertex: ADC location",
-      "  JUDGE_ACCESS_TOKEN      vertex: static bearer override (skips ADC)",
-      "  GEMINI_API_KEY          gemini: API key",
-      "  OPENAI_API_KEY          openai: API key",
-      "  ANTHROPIC_API_KEY       anthropic: API key",
-      "  JUDGE_CLAUDE_BIN        claude: binary override (default claude)",
-      "  JUDGE_HOST / JUDGE_PORT bind address (default 127.0.0.1:8790)",
-      "  JUDGE_TOKEN             optional bearer required on every /judge call"
+      "  UNOTEST_JUDGE_PROVIDER        fake (deterministic, CI-safe) | vertex (ADC) |",
+      "                                claude (Claude Code CLI, subscription auth) |",
+      "                                gemini | openai | anthropic (API keys)",
+      "  UNOTEST_JUDGE_MODEL           model id; defaults per provider:",
+      "                                vertex/gemini gemini-2.5-flash,",
+      "                                openai gpt-5-mini, anthropic claude-haiku-4-5,",
+      "                                claude: CLI default",
+      "  UNOTEST_JUDGE_RETRIES         extra provider calls on fail (default 1)",
+      "  UNOTEST_JUDGE_VOTE            N-call majority vote (default 1 = off)",
+      "  UNOTEST_JUDGE_CALL_TIMEOUT_MS per-call budget, ms (default 30000;",
+      "                                claude 120000)",
+      "  GOOGLE_CLOUD_PROJECT          vertex: ADC project",
+      "  GOOGLE_CLOUD_LOCATION         vertex: ADC location",
+      "  UNOTEST_JUDGE_ACCESS_TOKEN    vertex: static bearer override (skips ADC)",
+      "  GEMINI_API_KEY                gemini: API key",
+      "  OPENAI_API_KEY                openai: API key",
+      "  ANTHROPIC_API_KEY             anthropic: API key",
+      "  UNOTEST_JUDGE_CLAUDE_BIN      claude: binary override (default claude)",
+      "  UNOTEST_JUDGE_HOST / _PORT    bind address (default 127.0.0.1:8790)",
+      "  UNOTEST_JUDGE_TOKEN           optional bearer required on every /judge call"
     ].join("\n")
   );
 }

package/dist/index.d.ts

@@ -40,7 +40,7 @@ interface VertexProviderOptions {
     project: string;
     location: string;
     model: string;
-    /** Static bearer token override (JUDGE_ACCESS_TOKEN). Absent → ADC. */
+    /** Static bearer token override (UNOTEST_JUDGE_ACCESS_TOKEN). Absent → ADC. */
     accessToken?: string;
     /** Per-call wall-clock budget, ms. */
     timeoutMs: number;
@@ -110,7 +110,7 @@ type ExecFileFn = (file: string, args: string[], opts: {
     maxBuffer: number;
 }) => Promise<ExecFileResult>;
 interface ClaudeCliProviderOptions {
-    /** Binary to spawn (JUDGE_CLAUDE_BIN, default "claude"). */
+    /** Binary to spawn (UNOTEST_JUDGE_CLAUDE_BIN, default "claude"). */
     bin: string;
     /** Passed as `--model` only when set — otherwise the CLI's default. */
     model?: string;
@@ -137,11 +137,14 @@ declare function judgeWithRetries(provider: JudgeProvider, request: JudgeRequest
 type JudgeProviderKind = "fake" | "vertex" | "claude" | "gemini" | "openai" | "anthropic";
 interface JudgeServiceEnv {
     provider: JudgeProviderKind;
-    /** Model id. Absent only for `claude` without JUDGE_MODEL (CLI default)
+    /** Model id. Absent only for `claude` without UNOTEST_JUDGE_MODEL (CLI default)
      *  and for `fake` (no model behind it). */
     model?: string;
-    /** Extra provider calls allowed on `fail` (first `pass` wins). */
+    /** Extra provider calls allowed on `fail` (first `pass` wins). Ignored
+     *  when `votes` > 1 — see policy.ts. */
     retries: number;
+    /** Independent calls whose majority decides the verdict. 1 = off. */
+    votes: number;
     /** Per-provider-call wall-clock budget, ms. */
     timeoutMs: number;
     /** Vertex ADC target. */

package/dist/index.js

@@ -19,7 +19,7 @@ import {
   resolveJudgeServerEnv,
   resolveJudgeServiceEnv,
   startJudgeServer
-} from "./chunk-4VUHPR26.js";
+} from "./chunk-74FL5RWE.js";
 
 // src/index.ts
 import { JUDGE_ROUTES } from "@unotest/protocol";

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@unotest/judge",
-  "version": "0.23.0",
+  "version": "0.24.0",
   "description": "LLM-judge service for the @unotest ecosystem: judges free-form text (chat replies, generated content) against a natural-language rubric and returns a structured pass/fail verdict with reasoning. Runs as a small HTTP service (`npx @unotest/judge`) or in-process. Providers: deterministic `fake` (CI-safe), Google Vertex AI via ADC, the local Claude Code CLI (subscription auth, no API key), and Gemini / OpenAI / Anthropic APIs via keys — all raw HTTP or a local process, zero provider SDKs. Backs the `assertJudge` DSL assertion of @unotest/web.",
   "license": "MIT",
   "publishConfig": {
@@ -32,15 +32,8 @@
     "README.md"
   ],
   "dependencies": {
-    "@unotest/protocol": "^0.23.0"
-  },
-  "peerDependencies": {
-    "google-auth-library": "^9.0.0 || ^10.0.0"
-  },
-  "peerDependenciesMeta": {
-    "google-auth-library": {
-      "optional": true
-    }
+    "google-auth-library": "^10.0.0",
+    "@unotest/protocol": "^0.24.0"
   },
   "devDependencies": {
     "@types/node": "^22.10.0",