bridgebench 3.1.0-alpha.0

package/dist/cli.cjs

@@ -0,0 +1,409 @@
+#!/usr/bin/env node
+"use strict"; function _interopRequireDefault(obj) { return obj && obj.__esModule ? obj : { default: obj }; } function _nullishCoalesce(lhs, rhsFn) { if (lhs != null) { return lhs; } else { return rhsFn(); } } function _optionalChain(ops) { let lastAccessLHS = undefined; let value = ops[0]; let i = 1; while (i < ops.length) { const op = ops[i]; const fn = ops[i + 1]; i += 2; if ((op === 'optionalAccess' || op === 'optionalCall') && value == null) { return undefined; } if (op === 'access' || op === 'optionalAccess') { lastAccessLHS = value; value = fn(value); } else if (op === 'call' || op === 'optionalCall') { value = fn((...args) => value.call(lastAccessLHS, ...args)); lastAccessLHS = undefined; } } return value; }
+
+
+
+
+
+var _chunk7YCJSOK7cjs = require('./chunk-7YCJSOK7.cjs');
+
+
+
+
+
+
+
+
+
+
+var _chunkNJTYVNP4cjs = require('./chunk-NJTYVNP4.cjs');
+
+
+
+
+
+
+
+
+
+
+var _chunk4TWPCPRPcjs = require('./chunk-4TWPCPRP.cjs');
+
+
+
+
+var _chunkCIXITJW6cjs = require('./chunk-CIXITJW6.cjs');
+
+// src/commands.ts
+var _path = require('path'); var _path2 = _interopRequireDefault(_path);
+var _commander = require('commander');
+
+// src/env.ts
+var _fs = require('fs');
+
+var ENV_PATH = _path2.default.join(_chunk4TWPCPRPcjs.findProjectRoot.call(void 0, _chunkCIXITJW6cjs.importMetaUrl), ".env");
+function loadProjectEnv() {
+  if (!_fs.existsSync.call(void 0, ENV_PATH)) return { status: "missing" };
+  try {
+    process.loadEnvFile(ENV_PATH);
+    return { status: "loaded" };
+  } catch (error) {
+    return {
+      status: "error",
+      reason: error instanceof Error ? error.message : String(error)
+    };
+  }
+}
+
+// src/commands.ts
+var ROOT = process.cwd();
+var defaultDependencies = {
+  stdout: (message) => console.log(message),
+  stderr: (message) => console.error(message),
+  setExitCode: (code) => {
+    process.exitCode = code;
+  },
+  createStore: (category) => new (0, _chunkNJTYVNP4cjs.ArenaStore)(_chunkNJTYVNP4cjs.categoryStoreConfig.call(void 0, category)),
+  createTaskLoader: (category) => new (0, _chunk4TWPCPRPcjs.TaskLoader)(category),
+  createOpenRouter: (logger) => new (0, _chunk4TWPCPRPcjs.OpenRouterClient)(_nullishCoalesce(process.env.OPENROUTER_API_KEY, () => ( "")), logger),
+  publishTasks: _chunk7YCJSOK7cjs.publishTasks,
+  publishJournal: _chunk7YCJSOK7cjs.publishJournal,
+  resolveApiConfig: _chunk7YCJSOK7cjs.resolveApiConfig
+};
+function parseCategory(value) {
+  const parsed = _chunkCIXITJW6cjs.BenchmarkCategorySchema.safeParse(value);
+  if (!parsed.success) {
+    throw new (0, _commander.InvalidArgumentError)(`expected one of ${_chunkCIXITJW6cjs.CATEGORIES.join(", ")}, received ${value}`);
+  }
+  return parsed.data;
+}
+function positiveInteger(value) {
+  const parsed = Number(value);
+  if (!Number.isInteger(parsed) || parsed < 1) {
+    throw new (0, _commander.InvalidArgumentError)(`expected a positive integer, received ${value}`);
+  }
+  return parsed;
+}
+function positiveNumber(value) {
+  const parsed = Number(value);
+  if (!Number.isFinite(parsed) || parsed <= 0) {
+    throw new (0, _commander.InvalidArgumentError)(`expected a positive number, received ${value}`);
+  }
+  return parsed;
+}
+function collectOption(value, previous = []) {
+  return [...previous, value];
+}
+function arenaRunConfigFromOptions(options) {
+  const competitorIds = _optionalChain([options, 'access', _ => _.competitor, 'optionalAccess', _2 => _2.length]) ? _chunk4TWPCPRPcjs.resolveCompetitorRoster.call(void 0, options.competitor).map((model) => model.id) : void 0;
+  return {
+    category: options.category,
+    seed: options.seed,
+    matches: options.matches,
+    maxCostUsd: options.maxCostUsd,
+    competitorIds,
+    resume: options.resume,
+    healthStop: options.healthStop
+  };
+}
+function selectedCategories(options, command) {
+  if (options.category && options.all) {
+    throw new Error(`${command} accepts either --category or --all, not both`);
+  }
+  if (options.category) return [options.category];
+  if (options.all) return [..._chunkCIXITJW6cjs.CATEGORIES];
+  throw new Error(`${command} requires --category <category> or --all`);
+}
+function logRoot(category) {
+  return _path2.default.join(ROOT, "results", category, "logs");
+}
+function displayPath(target) {
+  const relative = _path2.default.relative(ROOT, target);
+  return relative.startsWith("..") ? _path2.default.basename(target) : relative;
+}
+function buildProgram(overrides = {}) {
+  const dependencies = { ...defaultDependencies, ...overrides };
+  const program = new (0, _commander.Command)().name(_chunkNJTYVNP4cjs.PACKAGE_NAME).description("Autonomous BridgeBench arenas for reasoning and hallucination").version(_chunkNJTYVNP4cjs.ENGINE_VERSION).showHelpAfterError().configureHelp({ sortOptions: true, sortSubcommands: true }).configureOutput({
+    writeOut: (value) => dependencies.stdout(value.replace(/\n$/, "")),
+    writeErr: (value) => dependencies.stderr(value.replace(/\n$/, ""))
+  });
+  const models = program.command("models").description("Inspect or validate the pinned model registry");
+  models.command("list").description("List enabled competitors and judges").action(() => {
+    for (const model of _chunk4TWPCPRPcjs.listModels.call(void 0, )) {
+      dependencies.stdout(`${model.role.padEnd(10)} ${model.id.padEnd(40)} ${model.displayName}`);
+    }
+  });
+  models.command("validate").description("Validate pinned model IDs and judge capabilities against OpenRouter").action(async () => {
+    const client = dependencies.createOpenRouter();
+    for (const model of _chunk4TWPCPRPcjs.listModels.call(void 0, )) {
+      await client.validateModel(model);
+      dependencies.stdout(`\u2713 ${model.id} -> ${model.canonicalSlug}`);
+    }
+  });
+  const tasks = program.command("tasks").description("Validate or publish task packs");
+  tasks.command("validate").description("Validate public task schemas and pack balance").option(
+    "-c, --category <category>",
+    `limit to one category (${_chunkCIXITJW6cjs.CATEGORIES.join(", ")})`,
+    parseCategory
+  ).option("--file <path>", "validate one public task without pack-balance checks").action(async (options) => {
+    if (options.file) {
+      if (options.category) {
+        throw new Error("tasks validate accepts --file or --category, not both");
+      }
+      const task = await _chunk4TWPCPRPcjs.validatePublicTaskFile.call(void 0, options.file);
+      dependencies.stdout(
+        `\u2713 ${task.public.id}: public task validated (${task.public.category}/${task.public.cluster})`
+      );
+      return;
+    }
+    const categories = options.category ? [options.category] : _chunkCIXITJW6cjs.CATEGORIES;
+    for (const category of categories) {
+      const loader = dependencies.createTaskLoader(category);
+      const loaded = await loader.loadAll();
+      const mode = loader.hasPrivate ? "public + private halves" : "public halves only";
+      dependencies.stdout(`\u2713 ${category}: ${loaded.length} tasks validated (${mode})`);
+      for (const task of loaded) {
+        dependencies.stdout(`  ${task.public.cluster.padEnd(28)} ${task.public.id}`);
+      }
+    }
+  });
+  tasks.command("publish").description("Publish public and private task halves to the configured API").option(
+    "-c, --category <category>",
+    `category to publish (${_chunkCIXITJW6cjs.CATEGORIES.join(", ")})`,
+    parseCategory
+  ).option("--all", "publish every category explicitly", false).action(async (options) => {
+    const categories = selectedCategories(options, "tasks publish");
+    const config = dependencies.resolveApiConfig();
+    dependencies.stdout(`Publishing ${categories.join(", ")} tasks to ${_chunk7YCJSOK7cjs.publishTarget.call(void 0, config)}`);
+    for (const category of categories) {
+      const result = await dependencies.publishTasks(category, config);
+      dependencies.stdout(`\u2713 ${category}: published ${result.imported} tasks`);
+    }
+  });
+  const arena = program.command("arena").description("Run and inspect arena matches");
+  arena.command("run").description("Run a paid deterministic match schedule").requiredOption(
+    "-c, --category <category>",
+    `arena to run (${_chunkCIXITJW6cjs.CATEGORIES.join(", ")})`,
+    parseCategory
+  ).option("-n, --matches <count>", "number of task-level matches", positiveInteger, 12).option("-s, --seed <seed>", "reproducible scheduling seed", "bridgebench-v3-mvp").option(
+    "--max-cost-usd <amount>",
+    "hard stop before the next match after this spend",
+    positiveNumber,
+    25
+  ).option(
+    "--competitor <modelId>",
+    "limit the run roster; repeat for each competitor (minimum two)",
+    collectOption
+  ).option("--resume", "skip match IDs already present in the journal", false).option("--debug", "mirror structured log entries to the console", false).option("--no-health-stop", "do not halt when most matches have failed responses").action(async (options) => {
+    const config = arenaRunConfigFromOptions(options);
+    const logger = new (0, _chunk4TWPCPRPcjs.FileArenaLogger)({
+      dir: logRoot(config.category),
+      verbose: options.debug
+    });
+    const cancellation = new AbortController();
+    const requestCancellation = () => {
+      if (cancellation.signal.aborted) return;
+      dependencies.stderr("Cancellation requested; aborting active model calls.");
+      cancellation.abort();
+    };
+    process.once("SIGINT", requestCancellation);
+    try {
+      dependencies.stdout(`Run log: ${displayPath(logger.filePath)}`);
+      const store = dependencies.createStore(config.category);
+      const loaded = await dependencies.createTaskLoader(config.category).loadAll({ requirePrivate: true });
+      const progressOutput = (event) => {
+        if (event.type !== "match.completed") return;
+        dependencies.stdout(
+          `[${event.data.completed}/${event.data.total}] ${event.data.taskId}: ${_nullishCoalesce(event.data.winnerModelId, () => ( "no contest"))} (${event.data.outcome}, $${event.data.costUsd.toFixed(4)})`
+        );
+      };
+      const runner = new (0, _chunkNJTYVNP4cjs.ArenaRunner)(
+        dependencies.createOpenRouter(logger),
+        store,
+        progressOutput,
+        logger
+      );
+      const result = await runner.run(config, loaded, { signal: cancellation.signal });
+      if (result.cancelled) {
+        dependencies.stdout(
+          `Cancelled after ${result.completed} new ${config.category} matches; journaled spend $${result.costUsd.toFixed(4)}. Completed matches remain journaled.`
+        );
+        dependencies.setExitCode(130);
+      } else {
+        dependencies.stdout(
+          `Completed ${result.completed} new ${config.category} matches; run spend $${result.costUsd.toFixed(4)}.`
+        );
+        if (result.stoppedForBudget) {
+          dependencies.stdout(
+            "Stopped at the cost boundary; rerun the same command with --resume."
+          );
+        }
+      }
+      const runMatches = store.readAll().filter((match) => match.runId === result.runId);
+      if (runMatches.length > 0) {
+        const triage = _chunkNJTYVNP4cjs.triageJournal.call(void 0, runMatches);
+        dependencies.stdout(`=== Run health ===
+${_chunkNJTYVNP4cjs.formatTriage.call(void 0, triage)}`);
+      }
+    } finally {
+      process.removeListener("SIGINT", requestCancellation);
+      dependencies.stdout(`Run log: ${displayPath(logger.filePath)}`);
+    }
+  });
+  arena.command("remote-run").description("Run a match schedule against the configured API and stream live arena events").requiredOption(
+    "-c, --category <category>",
+    `arena to run (${_chunkCIXITJW6cjs.CATEGORIES.join(", ")})`,
+    parseCategory
+  ).option("-n, --matches <count>", "number of task-level matches", positiveInteger, 12).option("-s, --seed <seed>", "reproducible scheduling seed", "bridgebench-v3-mvp").option(
+    "--max-cost-usd <amount>",
+    "hard stop before the next match after this spend",
+    positiveNumber,
+    25
+  ).option(
+    "--competitor <modelId>",
+    "limit the run roster; repeat for each competitor (minimum two)",
+    collectOption
+  ).option("--resume", "skip match IDs already present in the journal", false).option("--debug", "mirror structured log entries to the console", false).option("--no-health-stop", "do not halt when most matches have failed responses").option("--mock", "use deterministic mock model completions instead of OpenRouter", false).option("--no-publish-matches", "skip publishing match results to the API as they complete").action(async (options) => {
+    const config = arenaRunConfigFromOptions(options);
+    const apiConfig = dependencies.resolveApiConfig();
+    const logger = new (0, _chunk4TWPCPRPcjs.FileArenaLogger)({
+      dir: logRoot(config.category),
+      verbose: options.debug
+    });
+    dependencies.stdout(
+      `Remote ${config.category} run against ${_chunk7YCJSOK7cjs.publishTarget.call(void 0, apiConfig)} (${options.mock ? "mock" : "openrouter"})`
+    );
+    dependencies.stdout(`Run log: ${displayPath(logger.filePath)}`);
+    try {
+      const result = await _chunk7YCJSOK7cjs.runRemoteArena.call(void 0, apiConfig, {
+        config,
+        mock: options.mock,
+        publishMatches: options.publishMatches,
+        logger
+      });
+      dependencies.stdout(
+        `Remote run ${result.runKey}: ${result.completed} matches, $${result.costUsd.toFixed(4)}`
+      );
+      if (result.cancelled) dependencies.setExitCode(130);
+    } finally {
+      dependencies.stdout(`Run log: ${displayPath(logger.filePath)}`);
+    }
+  });
+  arena.command("verify").description("Validate and replay outcomes, points, manifests, and Elo").option(
+    "-c, --category <category>",
+    `journal category (${_chunkCIXITJW6cjs.CATEGORIES.join(", ")})`,
+    parseCategory,
+    "reasoning"
+  ).option("--journal <path>", "journal file to verify (defaults to the local category journal)").option("--manifests-dir <path>", "run-manifest directory (defaults to a runs/ sibling)").option("--json", "emit the machine-readable verification result", false).action(
+    (options) => {
+      const defaults = _chunkNJTYVNP4cjs.categoryStoreConfig.call(void 0, options.category);
+      const journalPath = options.journal ? _path2.default.resolve(options.journal) : defaults.journalPath;
+      const runsDir = options.manifestsDir ? _path2.default.resolve(options.manifestsDir) : options.journal ? _path2.default.join(_path2.default.dirname(journalPath), "runs") : defaults.runsDir;
+      const store = new (0, _chunkNJTYVNP4cjs.ArenaStore)({
+        ...defaults,
+        journalPath,
+        runsDir,
+        readOnly: true
+      });
+      const verified = _chunkNJTYVNP4cjs.verifyJournal.call(void 0, store.readAll(), options.category, {
+        manifestForRun: (runId) => store.readRunManifest(runId),
+        requireManifests: true
+      });
+      if (options.json) {
+        dependencies.stdout(JSON.stringify(verified, null, 2));
+        return;
+      }
+      dependencies.stdout(
+        `\u2713 ${options.category}: verified ${verified.matches.length} matches across ${verified.runs.length} runs`
+      );
+      for (const warning of verified.warnings) {
+        dependencies.stderr(`Warning: ${warning}`);
+      }
+    }
+  );
+  arena.command("publish").description("Publish a verified match journal to the configured API").option(
+    "-c, --category <category>",
+    `journal to publish (${_chunkCIXITJW6cjs.CATEGORIES.join(", ")})`,
+    parseCategory
+  ).option("--all", "publish every category explicitly", false).option("--allow-empty", "treat an empty journal as a successful no-op", false).action(async (options) => {
+    const categories = selectedCategories(options, "arena publish");
+    const config = dependencies.resolveApiConfig();
+    dependencies.stdout(
+      `Publishing ${categories.join(", ")} journals to ${_chunk7YCJSOK7cjs.publishTarget.call(void 0, config)}`
+    );
+    for (const category of categories) {
+      const store = dependencies.createStore(category);
+      _chunkNJTYVNP4cjs.verifyJournal.call(void 0, store.readAll(), category, {
+        manifestForRun: (runId) => store.readRunManifest(runId),
+        requireManifests: true
+      });
+      const result = await dependencies.publishJournal(category, config);
+      if (result.matches === 0 && !options.allowEmpty) {
+        throw new Error(
+          `The ${category} journal is empty; pass --allow-empty for an intentional no-op`
+        );
+      }
+      dependencies.stdout(
+        `\u2713 ${category}: ${result.imported} new, ${result.skipped} already present (of ${result.matches} lines)`
+      );
+    }
+  });
+  arena.command("triage").description("Analyze failures, suspicious responses, and judging anomalies").option(
+    "-c, --category <category>",
+    `journal to analyze (${_chunkCIXITJW6cjs.CATEGORIES.join(", ")})`,
+    parseCategory,
+    "reasoning"
+  ).option("--run <runId>", "limit the report to one run ID").option("--json", "emit the machine-readable report", false).option("--strict", "exit nonzero when any anomaly is found", false).option("--allow-empty", "treat no matching journal lines as success", false).action(
+    (options) => {
+      const results = dependencies.createStore(options.category).readAll().filter((match) => !options.run || match.runId === options.run);
+      if (results.length === 0) {
+        const message = options.run ? `No ${options.category} matches found for run ${options.run}.` : `The ${options.category} journal is empty.`;
+        if (!options.allowEmpty) throw new Error(message);
+        dependencies.stdout(message);
+        return;
+      }
+      const reports = _chunkNJTYVNP4cjs.triageJournal.call(void 0, results);
+      dependencies.stdout(
+        options.json ? JSON.stringify(reports, null, 2) : _chunkNJTYVNP4cjs.formatTriage.call(void 0, reports)
+      );
+      const anomalies = reports.reduce((sum, report) => sum + report.anomalies.length, 0);
+      if (options.strict && anomalies > 0) {
+        dependencies.stderr(`Strict mode: ${anomalies} anomalies found.`);
+        dependencies.setExitCode(1);
+      }
+    }
+  );
+  arena.command("generation <generationId>").description("Fetch OpenRouter's record for a journaled generation").action(async (generationId) => {
+    const data = await dependencies.createOpenRouter().fetchGeneration(generationId);
+    dependencies.stdout(JSON.stringify(data, null, 2));
+  });
+  program.command("report").description("Verify journals and rebuild JSON and Markdown reports").option(
+    "-c, --category <category>",
+    `limit to one category (${_chunkCIXITJW6cjs.CATEGORIES.join(", ")})`,
+    parseCategory
+  ).action((options) => {
+    const categories = options.category ? [options.category] : _chunkCIXITJW6cjs.CATEGORIES;
+    for (const category of categories) {
+      const snapshot = _chunkNJTYVNP4cjs.writeReports.call(void 0, dependencies.createStore(category));
+      dependencies.stdout(`\u2713 ${category}: wrote reports for ${snapshot.matches.length} matches`);
+    }
+  });
+  return program;
+}
+async function runCli(argv = process.argv, overrides = {}) {
+  const dependencies = { ...defaultDependencies, ...overrides };
+  const envResult = loadProjectEnv();
+  if (envResult.status === "error") {
+    dependencies.stderr(
+      `Warning: project environment file could not be loaded: ${_chunk4TWPCPRPcjs.redactSecrets.call(void 0, envResult.reason)}`
+    );
+  }
+  await buildProgram(dependencies).parseAsync(argv);
+}
+
+// src/cli.ts
+runCli().catch((error) => {
+  console.error(`BridgeBench: ${_chunk4TWPCPRPcjs.sanitizeError.call(void 0, error)}`);
+  process.exitCode = 1;
+});
+//# sourceMappingURL=cli.cjs.map

package/dist/cli.cjs.map

@@ -0,0 +1 @@
+{"version":3,"sources":["/Users/matthewmiller/Desktop/bridgemind/bridgebench/dist/cli.cjs","../src/commands.ts","../src/env.ts","../src/cli.ts"],"names":[],"mappings":"AAAA;AACA;AACE;AACA;AACA;AACA;AACA;AACF,wDAA6B;AAC7B;AACE;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACF,wDAA6B;AAC7B;AACE;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACF,wDAA6B;AAC7B;AACE;AACA;AACA;AACF,wDAA6B;AAC7B;AACA;ACpCA,wEAAiB;AAEjB,sCAA8C;ADqC9C;AACA;AExCA,wBAA2B;AAC3B;AAIO,IAAM,SAAA,EAAW,cAAA,CAAK,IAAA,CAAK,+CAAA,+BAA+B,CAAA,EAAG,MAAM,CAAA;AAWnE,SAAS,cAAA,CAAA,EAAuC;AACrD,EAAA,GAAA,CAAI,CAAC,4BAAA,QAAmB,CAAA,EAAG,OAAO,EAAE,MAAA,EAAQ,UAAU,CAAA;AACtD,EAAA,IAAI;AACF,IAAA,OAAA,CAAQ,WAAA,CAAY,QAAQ,CAAA;AAC5B,IAAA,OAAO,EAAE,MAAA,EAAQ,SAAS,CAAA;AAAA,EAC5B,EAAA,MAAA,CAAS,KAAA,EAAO;AACd,IAAA,OAAO;AAAA,MACL,MAAA,EAAQ,OAAA;AAAA,MACR,MAAA,EAAQ,MAAA,WAAiB,MAAA,EAAQ,KAAA,CAAM,QAAA,EAAU,MAAA,CAAO,KAAK;AAAA,IAC/D,CAAA;AAAA,EACF;AACF;AF6BA;AACA;AC9BA,IAAM,KAAA,EAAO,OAAA,CAAQ,GAAA,CAAI,CAAA;AA8BzB,IAAM,oBAAA,EAAuC;AAAA,EAC3C,MAAA,EAAQ,CAAC,OAAA,EAAA,GAAY,OAAA,CAAQ,GAAA,CAAI,OAAO,CAAA;AAAA,EACxC,MAAA,EAAQ,CAAC,OAAA,EAAA,GAAY,OAAA,CAAQ,KAAA,CAAM,OAAO,CAAA;AAAA,EAC1C,WAAA,EAAa,CAAC,IAAA,EAAA,GAAS;AACrB,IAAA,OAAA,CAAQ,SAAA,EAAW,IAAA;AAAA,EACrB,CAAA;AAAA,EACA,WAAA,EAAa,CAAC,QAAA,EAAA,GAAa,IAAI,iCAAA,CAAW,mDAAA,QAA4B,CAAC,CAAA;AAAA,EACvE,gBAAA,EAAkB,CAAC,QAAA,EAAA,GAAa,IAAI,iCAAA,CAAW,QAAQ,CAAA;AAAA,EACvD,gBAAA,EAAkB,CAAC,MAAA,EAAA,GAAW,IAAI,uCAAA,kBAAiB,OAAA,CAAQ,GAAA,CAAI,kBAAA,UAAsB,IAAA,EAAI,MAAM,CAAA;AAAA,EAC/F,4CAAA;AAAA,EACA,gDAAA;AAAA,EACA;AACF,CAAA;AAEA,SAAS,aAAA,CAAc,KAAA,EAAkC;AACvD,EAAA,MAAM,OAAA,EAAS,yCAAA,CAAwB,SAAA,CAAU,KAAK,CAAA;AACtD,EAAA,GAAA,CAAI,CAAC,MAAA,CAAO,OAAA,EAAS;AACnB,IAAA,MAAM,IAAI,oCAAA,CAAqB,CAAA,gBAAA,EAAmB,4BAAA,CAAW,IAAA,CAAK,IAAI,CAAC,CAAA,WAAA,EAAc,KAAK,CAAA,CAAA;AAC5F,EAAA;AACc,EAAA;AAChB;AAEgD;AACnB,EAAA;AACkB,EAAA;AACoC,IAAA;AACjF,EAAA;AACO,EAAA;AACT;AAE+C;AAClB,EAAA;AACkB,EAAA;AACmC,IAAA;AAChF,EAAA;AACO,EAAA;AACT;AAEyE;AAC7C,EAAA;AAC5B;AAEuF;AAEvC,EAAA;AAEvC,EAAA;AACa,IAAA;AACJ,IAAA;AACG,IAAA;AACG,IAAA;AACpB,IAAA;AACgB,IAAA;AACI,IAAA;AACtB,EAAA;AACF;AAE8F;AACvD,EAAA;AACsC,IAAA;AAC3E,EAAA;AAC8C,EAAA;AACR,EAAA;AAC8B,EAAA;AACtE;AAEsD;AACF,EAAA;AACpD;AAE6C;AACA,EAAA;AACgB,EAAA;AAC7D;AAEgF;AACD,EAAA;AAG9D,EAAA;AAKsD,IAAA;AACA,IAAA;AAClE,EAAA;AAIY,EAAA;AAIC,EAAA;AACsB,IAAA;AACwD,MAAA;AAC1F,IAAA;AACD,EAAA;AAGY,EAAA;AAEkC,IAAA;AACX,IAAA;AACA,MAAA;AAC6B,MAAA;AAC/D,IAAA;AACD,EAAA;AAEgF,EAAA;AAIhF,EAAA;AACC,IAAA;AAC+C,IAAA;AAC/C,IAAA;AAGM,EAAA;AACY,IAAA;AACM,MAAA;AACmD,QAAA;AACzE,MAAA;AACsD,MAAA;AACzC,MAAA;AACiE,QAAA;AAC9E,MAAA;AACA,MAAA;AACF,IAAA;AAC2D,IAAA;AACxB,IAAA;AACoB,MAAA;AACjB,MAAA;AACyB,MAAA;AACkB,MAAA;AACpD,MAAA;AACkD,QAAA;AAC7E,MAAA;AACF,IAAA;AACD,EAAA;AAIY,EAAA;AAEX,IAAA;AAC6C,IAAA;AAC7C,IAAA;AAG4C,EAAA;AACkB,IAAA;AACjB,IAAA;AAC8C,IAAA;AACxD,IAAA;AAC8B,MAAA;AACQ,MAAA;AACzE,IAAA;AACD,EAAA;AAE+E,EAAA;AAI/E,EAAA;AACC,IAAA;AACsC,IAAA;AACtC,IAAA;AAGD,EAAA;AAEC,IAAA;AACA,IAAA;AACA,IAAA;AACA,IAAA;AAED,EAAA;AACC,IAAA;AACA,IAAA;AACA,IAAA;AAGM,EAAA;AAG0C,IAAA;AACb,IAAA;AACL,MAAA;AACX,MAAA;AAClB,IAAA;AACwC,IAAA;AACD,IAAA;AACL,MAAA;AACyC,MAAA;AACvD,MAAA;AACrB,IAAA;AAC0C,IAAA;AACtC,IAAA;AAC4D,MAAA;AACR,MAAA;AAGzC,MAAA;AACqC,MAAA;AACV,QAAA;AACzB,QAAA;AACuE,UAAA;AACpF,QAAA;AACF,MAAA;AACmB,MAAA;AACmB,QAAA;AACpC,QAAA;AACA,QAAA;AACA,QAAA;AACF,MAAA;AAC+E,MAAA;AACzD,MAAA;AACP,QAAA;AAC+C,UAAA;AAC5D,QAAA;AAC4B,QAAA;AACvB,MAAA;AACQ,QAAA;AACwE,UAAA;AACrF,QAAA;AAC6B,QAAA;AACd,UAAA;AACX,YAAA;AACF,UAAA;AACF,QAAA;AACF,MAAA;AACiF,MAAA;AACtD,MAAA;AACc,QAAA;AACnB,QAAA;AAA6C;AACnE,MAAA;AACA,IAAA;AACoD,MAAA;AACU,MAAA;AAChE,IAAA;AACD,EAAA;AAIY,EAAA;AAEX,IAAA;AACsC,IAAA;AACtC,IAAA;AAGD,EAAA;AAEC,IAAA;AACA,IAAA;AACA,IAAA;AACA,IAAA;AAED,EAAA;AACC,IAAA;AACA,IAAA;AACA,IAAA;AAGM,EAAA;AAK0C,IAAA;AACA,IAAA;AACb,IAAA;AACL,MAAA;AACX,MAAA;AAClB,IAAA;AACY,IAAA;AAC0E,MAAA;AACvF,IAAA;AAC8D,IAAA;AAC1D,IAAA;AAC6C,MAAA;AAC7C,QAAA;AACc,QAAA;AACU,QAAA;AACxB,QAAA;AACD,MAAA;AACY,MAAA;AAC4E,QAAA;AACzF,MAAA;AACkD,MAAA;AAClD,IAAA;AAC8D,MAAA;AAChE,IAAA;AACD,EAAA;AAIY,EAAA;AAEX,IAAA;AAC0C,IAAA;AAC1C,IAAA;AACA,IAAA;AAE0B,EAAA;AASpB,IAAA;AACiD,MAAA;AAC0B,MAAA;AAGnE,MAAA;AAGiB,MAAA;AACxB,QAAA;AACH,QAAA;AACA,QAAA;AACU,QAAA;AACX,MAAA;AACiE,MAAA;AACV,QAAA;AACpC,QAAA;AACnB,MAAA;AACiB,MAAA;AACqC,QAAA;AACrD,QAAA;AACF,MAAA;AACa,MAAA;AACkE,QAAA;AAC/E,MAAA;AACyC,MAAA;AACE,QAAA;AAC3C,MAAA;AACF,IAAA;AACF,EAAA;AAIqE,EAAA;AAEnE,IAAA;AAC4C,IAAA;AAC5C,IAAA;AAGuB,EAAA;AAEuC,IAAA;AACjB,IAAA;AAChC,IAAA;AAC6D,MAAA;AAC1E,IAAA;AACmC,IAAA;AACc,MAAA;AACN,MAAA;AACe,QAAA;AACpC,QAAA;AACnB,MAAA;AACgE,MAAA;AAChB,MAAA;AACrC,QAAA;AACO,UAAA;AACjB,QAAA;AACF,MAAA;AACa,MAAA;AACqE,QAAA;AAClF,MAAA;AACF,IAAA;AACD,EAAA;AAIY,EAAA;AAEX,IAAA;AAC4C,IAAA;AAC5C,IAAA;AACA,IAAA;AAGgB,EAAA;AAUV,IAAA;AAIkB,MAAA;AACI,MAAA;AAEoC,QAAA;AAEZ,QAAA;AACrB,QAAA;AAC3B,QAAA;AACF,MAAA;AACqC,MAAA;AACxB,MAAA;AAC2D,QAAA;AACxE,MAAA;AACkF,MAAA;AAC7C,MAAA;AAC6B,QAAA;AACtC,QAAA;AAC5B,MAAA;AACF,IAAA;AACF,EAAA;AAIa,EAAA;AAEoE,IAAA;AAC9B,IAAA;AAClD,EAAA;AAImE,EAAA;AAElE,IAAA;AAC+C,IAAA;AAC/C,IAAA;AAEqD,EAAA;AACM,IAAA;AACxB,IAAA;AAC+B,MAAA;AACe,MAAA;AACjF,IAAA;AACD,EAAA;AAEI,EAAA;AACT;AAKiB;AAC6C,EAAA;AAC3B,EAAA;AACC,EAAA;AACnB,IAAA;AAC8E,MAAA;AAC3F,IAAA;AACF,EAAA;AACgD,EAAA;AAClD;ADxH+F;AACA;AG9YrE;AAC4B,EAAA;AACjC,EAAA;AACpB","file":"/Users/matthewmiller/Desktop/bridgemind/bridgebench/dist/cli.cjs","sourcesContent":[null,"import path from 'node:path';\n\nimport { Command, InvalidArgumentError } from 'commander';\n\nimport { ArenaRunner } from './arena.js';\nimport { loadProjectEnv } from './env.js';\nimport { FileArenaLogger, redactSecrets, type ArenaLogger } from './logger.js';\nimport { listModels, resolveCompetitorRoster } from './models.js';\nimport { OpenRouterClient } from './openrouter.js';\nimport { publishJournal, publishTarget, publishTasks, resolveApiConfig } from './publish.js';\nimport { runRemoteArena } from './remote-arena.js';\nimport { writeReports } from './report.js';\nimport { ArenaStore, categoryStoreConfig } from './store.js';\nimport { TaskLoader, validatePublicTaskFile } from './tasks.js';\nimport { formatTriage, triageJournal } from './triage.js';\nimport {\n  BenchmarkCategorySchema,\n  CATEGORIES,\n  type ArenaEventSink,\n  type ArenaRunConfig,\n  type BenchmarkCategory,\n} from './types.js';\nimport { verifyJournal } from './verification.js';\nimport { ENGINE_VERSION, PACKAGE_NAME } from './version.js';\n\n// CLI logs and path display anchor to the operator's working directory,\n// matching where categoryStoreConfig writes results.\nconst ROOT = process.cwd();\n\ninterface CategorySelection {\n  category?: BenchmarkCategory;\n  all?: boolean;\n}\n\nexport interface ArenaRunCliOptions {\n  category: BenchmarkCategory;\n  matches: number;\n  seed: string;\n  maxCostUsd: number;\n  competitor?: string[];\n  resume: boolean;\n  debug: boolean;\n  healthStop: boolean;\n}\n\nexport interface CliDependencies {\n  stdout: (message: string) => void;\n  stderr: (message: string) => void;\n  setExitCode: (code: number) => void;\n  createStore: (category: BenchmarkCategory) => ArenaStore;\n  createTaskLoader: (category: BenchmarkCategory) => TaskLoader;\n  createOpenRouter: (logger?: ArenaLogger) => OpenRouterClient;\n  publishTasks: typeof publishTasks;\n  publishJournal: typeof publishJournal;\n  resolveApiConfig: typeof resolveApiConfig;\n}\n\nconst defaultDependencies: CliDependencies = {\n  stdout: (message) => console.log(message),\n  stderr: (message) => console.error(message),\n  setExitCode: (code) => {\n    process.exitCode = code;\n  },\n  createStore: (category) => new ArenaStore(categoryStoreConfig(category)),\n  createTaskLoader: (category) => new TaskLoader(category),\n  createOpenRouter: (logger) => new OpenRouterClient(process.env.OPENROUTER_API_KEY ?? '', logger),\n  publishTasks,\n  publishJournal,\n  resolveApiConfig,\n};\n\nfunction parseCategory(value: string): BenchmarkCategory {\n  const parsed = BenchmarkCategorySchema.safeParse(value);\n  if (!parsed.success) {\n    throw new InvalidArgumentError(`expected one of ${CATEGORIES.join(', ')}, received ${value}`);\n  }\n  return parsed.data;\n}\n\nfunction positiveInteger(value: string): number {\n  const parsed = Number(value);\n  if (!Number.isInteger(parsed) || parsed < 1) {\n    throw new InvalidArgumentError(`expected a positive integer, received ${value}`);\n  }\n  return parsed;\n}\n\nfunction positiveNumber(value: string): number {\n  const parsed = Number(value);\n  if (!Number.isFinite(parsed) || parsed <= 0) {\n    throw new InvalidArgumentError(`expected a positive number, received ${value}`);\n  }\n  return parsed;\n}\n\nfunction collectOption(value: string, previous: string[] = []): string[] {\n  return [...previous, value];\n}\n\nexport function arenaRunConfigFromOptions(options: ArenaRunCliOptions): ArenaRunConfig {\n  const competitorIds = options.competitor?.length\n    ? resolveCompetitorRoster(options.competitor).map((model) => model.id)\n    : undefined;\n  return {\n    category: options.category,\n    seed: options.seed,\n    matches: options.matches,\n    maxCostUsd: options.maxCostUsd,\n    competitorIds,\n    resume: options.resume,\n    healthStop: options.healthStop,\n  };\n}\n\nfunction selectedCategories(options: CategorySelection, command: string): BenchmarkCategory[] {\n  if (options.category && options.all) {\n    throw new Error(`${command} accepts either --category or --all, not both`);\n  }\n  if (options.category) return [options.category];\n  if (options.all) return [...CATEGORIES];\n  throw new Error(`${command} requires --category <category> or --all`);\n}\n\nfunction logRoot(category: BenchmarkCategory): string {\n  return path.join(ROOT, 'results', category, 'logs');\n}\n\nfunction displayPath(target: string): string {\n  const relative = path.relative(ROOT, target);\n  return relative.startsWith('..') ? path.basename(target) : relative;\n}\n\nexport function buildProgram(overrides: Partial<CliDependencies> = {}): Command {\n  const dependencies: CliDependencies = { ...defaultDependencies, ...overrides };\n  const program = new Command()\n    .name(PACKAGE_NAME)\n    .description('Autonomous BridgeBench arenas for reasoning and hallucination')\n    .version(ENGINE_VERSION)\n    .showHelpAfterError()\n    .configureHelp({ sortOptions: true, sortSubcommands: true })\n    .configureOutput({\n      writeOut: (value) => dependencies.stdout(value.replace(/\\n$/, '')),\n      writeErr: (value) => dependencies.stderr(value.replace(/\\n$/, '')),\n    });\n\n  const models = program\n    .command('models')\n    .description('Inspect or validate the pinned model registry');\n  models\n    .command('list')\n    .description('List enabled competitors and judges')\n    .action(() => {\n      for (const model of listModels()) {\n        dependencies.stdout(`${model.role.padEnd(10)} ${model.id.padEnd(40)} ${model.displayName}`);\n      }\n    });\n  models\n    .command('validate')\n    .description('Validate pinned model IDs and judge capabilities against OpenRouter')\n    .action(async () => {\n      const client = dependencies.createOpenRouter();\n      for (const model of listModels()) {\n        await client.validateModel(model);\n        dependencies.stdout(`✓ ${model.id} -> ${model.canonicalSlug}`);\n      }\n    });\n\n  const tasks = program.command('tasks').description('Validate or publish task packs');\n  tasks\n    .command('validate')\n    .description('Validate public task schemas and pack balance')\n    .option(\n      '-c, --category <category>',\n      `limit to one category (${CATEGORIES.join(', ')})`,\n      parseCategory,\n    )\n    .option('--file <path>', 'validate one public task without pack-balance checks')\n    .action(async (options: { category?: BenchmarkCategory; file?: string }) => {\n      if (options.file) {\n        if (options.category) {\n          throw new Error('tasks validate accepts --file or --category, not both');\n        }\n        const task = await validatePublicTaskFile(options.file);\n        dependencies.stdout(\n          `✓ ${task.public.id}: public task validated (${task.public.category}/${task.public.cluster})`,\n        );\n        return;\n      }\n      const categories = options.category ? [options.category] : CATEGORIES;\n      for (const category of categories) {\n        const loader = dependencies.createTaskLoader(category);\n        const loaded = await loader.loadAll();\n        const mode = loader.hasPrivate ? 'public + private halves' : 'public halves only';\n        dependencies.stdout(`✓ ${category}: ${loaded.length} tasks validated (${mode})`);\n        for (const task of loaded) {\n          dependencies.stdout(`  ${task.public.cluster.padEnd(28)} ${task.public.id}`);\n        }\n      }\n    });\n\n  tasks\n    .command('publish')\n    .description('Publish public and private task halves to the configured API')\n    .option(\n      '-c, --category <category>',\n      `category to publish (${CATEGORIES.join(', ')})`,\n      parseCategory,\n    )\n    .option('--all', 'publish every category explicitly', false)\n    .action(async (options: CategorySelection) => {\n      const categories = selectedCategories(options, 'tasks publish');\n      const config = dependencies.resolveApiConfig();\n      dependencies.stdout(`Publishing ${categories.join(', ')} tasks to ${publishTarget(config)}`);\n      for (const category of categories) {\n        const result = await dependencies.publishTasks(category, config);\n        dependencies.stdout(`✓ ${category}: published ${result.imported} tasks`);\n      }\n    });\n\n  const arena = program.command('arena').description('Run and inspect arena matches');\n  arena\n    .command('run')\n    .description('Run a paid deterministic match schedule')\n    .requiredOption(\n      '-c, --category <category>',\n      `arena to run (${CATEGORIES.join(', ')})`,\n      parseCategory,\n    )\n    .option('-n, --matches <count>', 'number of task-level matches', positiveInteger, 12)\n    .option('-s, --seed <seed>', 'reproducible scheduling seed', 'bridgebench-v3-mvp')\n    .option(\n      '--max-cost-usd <amount>',\n      'hard stop before the next match after this spend',\n      positiveNumber,\n      25,\n    )\n    .option(\n      '--competitor <modelId>',\n      'limit the run roster; repeat for each competitor (minimum two)',\n      collectOption,\n    )\n    .option('--resume', 'skip match IDs already present in the journal', false)\n    .option('--debug', 'mirror structured log entries to the console', false)\n    .option('--no-health-stop', 'do not halt when most matches have failed responses')\n    .action(async (options: ArenaRunCliOptions) => {\n      const config = arenaRunConfigFromOptions(options);\n      const logger = new FileArenaLogger({\n        dir: logRoot(config.category),\n        verbose: options.debug,\n      });\n      const cancellation = new AbortController();\n      const requestCancellation = (): void => {\n        if (cancellation.signal.aborted) return;\n        dependencies.stderr('Cancellation requested; aborting active model calls.');\n        cancellation.abort();\n      };\n      process.once('SIGINT', requestCancellation);\n      try {\n        dependencies.stdout(`Run log: ${displayPath(logger.filePath)}`);\n        const store = dependencies.createStore(config.category);\n        const loaded = await dependencies\n          .createTaskLoader(config.category)\n          .loadAll({ requirePrivate: true });\n        const progressOutput: ArenaEventSink = (event) => {\n          if (event.type !== 'match.completed') return;\n          dependencies.stdout(\n            `[${event.data.completed}/${event.data.total}] ${event.data.taskId}: ${event.data.winnerModelId ?? 'no contest'} (${event.data.outcome}, $${event.data.costUsd.toFixed(4)})`,\n          );\n        };\n        const runner = new ArenaRunner(\n          dependencies.createOpenRouter(logger),\n          store,\n          progressOutput,\n          logger,\n        );\n        const result = await runner.run(config, loaded, { signal: cancellation.signal });\n        if (result.cancelled) {\n          dependencies.stdout(\n            `Cancelled after ${result.completed} new ${config.category} matches; journaled spend $${result.costUsd.toFixed(4)}. Completed matches remain journaled.`,\n          );\n          dependencies.setExitCode(130);\n        } else {\n          dependencies.stdout(\n            `Completed ${result.completed} new ${config.category} matches; run spend $${result.costUsd.toFixed(4)}.`,\n          );\n          if (result.stoppedForBudget) {\n            dependencies.stdout(\n              'Stopped at the cost boundary; rerun the same command with --resume.',\n            );\n          }\n        }\n        const runMatches = store.readAll().filter((match) => match.runId === result.runId);\n        if (runMatches.length > 0) {\n          const triage = triageJournal(runMatches);\n          dependencies.stdout(`=== Run health ===\\n${formatTriage(triage)}`);\n        }\n      } finally {\n        process.removeListener('SIGINT', requestCancellation);\n        dependencies.stdout(`Run log: ${displayPath(logger.filePath)}`);\n      }\n    });\n\n  arena\n    .command('remote-run')\n    .description('Run a match schedule against the configured API and stream live arena events')\n    .requiredOption(\n      '-c, --category <category>',\n      `arena to run (${CATEGORIES.join(', ')})`,\n      parseCategory,\n    )\n    .option('-n, --matches <count>', 'number of task-level matches', positiveInteger, 12)\n    .option('-s, --seed <seed>', 'reproducible scheduling seed', 'bridgebench-v3-mvp')\n    .option(\n      '--max-cost-usd <amount>',\n      'hard stop before the next match after this spend',\n      positiveNumber,\n      25,\n    )\n    .option(\n      '--competitor <modelId>',\n      'limit the run roster; repeat for each competitor (minimum two)',\n      collectOption,\n    )\n    .option('--resume', 'skip match IDs already present in the journal', false)\n    .option('--debug', 'mirror structured log entries to the console', false)\n    .option('--no-health-stop', 'do not halt when most matches have failed responses')\n    .option('--mock', 'use deterministic mock model completions instead of OpenRouter', false)\n    .option('--no-publish-matches', 'skip publishing match results to the API as they complete')\n    .action(async (options: ArenaRunCliOptions & { mock: boolean; publishMatches: boolean }) => {\n      const config = arenaRunConfigFromOptions(options);\n      const apiConfig = dependencies.resolveApiConfig();\n      const logger = new FileArenaLogger({\n        dir: logRoot(config.category),\n        verbose: options.debug,\n      });\n      dependencies.stdout(\n        `Remote ${config.category} run against ${publishTarget(apiConfig)} (${options.mock ? 'mock' : 'openrouter'})`,\n      );\n      dependencies.stdout(`Run log: ${displayPath(logger.filePath)}`);\n      try {\n        const result = await runRemoteArena(apiConfig, {\n          config,\n          mock: options.mock,\n          publishMatches: options.publishMatches,\n          logger,\n        });\n        dependencies.stdout(\n          `Remote run ${result.runKey}: ${result.completed} matches, $${result.costUsd.toFixed(4)}`,\n        );\n        if (result.cancelled) dependencies.setExitCode(130);\n      } finally {\n        dependencies.stdout(`Run log: ${displayPath(logger.filePath)}`);\n      }\n    });\n\n  arena\n    .command('verify')\n    .description('Validate and replay outcomes, points, manifests, and Elo')\n    .option(\n      '-c, --category <category>',\n      `journal category (${CATEGORIES.join(', ')})`,\n      parseCategory,\n      'reasoning',\n    )\n    .option('--journal <path>', 'journal file to verify (defaults to the local category journal)')\n    .option('--manifests-dir <path>', 'run-manifest directory (defaults to a runs/ sibling)')\n    .option('--json', 'emit the machine-readable verification result', false)\n    .action(\n      (options: {\n        category: BenchmarkCategory;\n        journal?: string;\n        manifestsDir?: string;\n        json: boolean;\n      }) => {\n        const defaults = categoryStoreConfig(options.category);\n        const journalPath = options.journal ? path.resolve(options.journal) : defaults.journalPath;\n        const runsDir = options.manifestsDir\n          ? path.resolve(options.manifestsDir)\n          : options.journal\n            ? path.join(path.dirname(journalPath), 'runs')\n            : defaults.runsDir;\n        const store = new ArenaStore({\n          ...defaults,\n          journalPath,\n          runsDir,\n          readOnly: true,\n        });\n        const verified = verifyJournal(store.readAll(), options.category, {\n          manifestForRun: (runId) => store.readRunManifest(runId),\n          requireManifests: true,\n        });\n        if (options.json) {\n          dependencies.stdout(JSON.stringify(verified, null, 2));\n          return;\n        }\n        dependencies.stdout(\n          `✓ ${options.category}: verified ${verified.matches.length} matches across ${verified.runs.length} runs`,\n        );\n        for (const warning of verified.warnings) {\n          dependencies.stderr(`Warning: ${warning}`);\n        }\n      },\n    );\n\n  arena\n    .command('publish')\n    .description('Publish a verified match journal to the configured API')\n    .option(\n      '-c, --category <category>',\n      `journal to publish (${CATEGORIES.join(', ')})`,\n      parseCategory,\n    )\n    .option('--all', 'publish every category explicitly', false)\n    .option('--allow-empty', 'treat an empty journal as a successful no-op', false)\n    .action(async (options: CategorySelection & { allowEmpty: boolean }) => {\n      const categories = selectedCategories(options, 'arena publish');\n      const config = dependencies.resolveApiConfig();\n      dependencies.stdout(\n        `Publishing ${categories.join(', ')} journals to ${publishTarget(config)}`,\n      );\n      for (const category of categories) {\n        const store = dependencies.createStore(category);\n        verifyJournal(store.readAll(), category, {\n          manifestForRun: (runId) => store.readRunManifest(runId),\n          requireManifests: true,\n        });\n        const result = await dependencies.publishJournal(category, config);\n        if (result.matches === 0 && !options.allowEmpty) {\n          throw new Error(\n            `The ${category} journal is empty; pass --allow-empty for an intentional no-op`,\n          );\n        }\n        dependencies.stdout(\n          `✓ ${category}: ${result.imported} new, ${result.skipped} already present (of ${result.matches} lines)`,\n        );\n      }\n    });\n\n  arena\n    .command('triage')\n    .description('Analyze failures, suspicious responses, and judging anomalies')\n    .option(\n      '-c, --category <category>',\n      `journal to analyze (${CATEGORIES.join(', ')})`,\n      parseCategory,\n      'reasoning',\n    )\n    .option('--run <runId>', 'limit the report to one run ID')\n    .option('--json', 'emit the machine-readable report', false)\n    .option('--strict', 'exit nonzero when any anomaly is found', false)\n    .option('--allow-empty', 'treat no matching journal lines as success', false)\n    .action(\n      (options: {\n        category: BenchmarkCategory;\n        run?: string;\n        json: boolean;\n        strict: boolean;\n        allowEmpty: boolean;\n      }) => {\n        const results = dependencies\n          .createStore(options.category)\n          .readAll()\n          .filter((match) => !options.run || match.runId === options.run);\n        if (results.length === 0) {\n          const message = options.run\n            ? `No ${options.category} matches found for run ${options.run}.`\n            : `The ${options.category} journal is empty.`;\n          if (!options.allowEmpty) throw new Error(message);\n          dependencies.stdout(message);\n          return;\n        }\n        const reports = triageJournal(results);\n        dependencies.stdout(\n          options.json ? JSON.stringify(reports, null, 2) : formatTriage(reports),\n        );\n        const anomalies = reports.reduce((sum, report) => sum + report.anomalies.length, 0);\n        if (options.strict && anomalies > 0) {\n          dependencies.stderr(`Strict mode: ${anomalies} anomalies found.`);\n          dependencies.setExitCode(1);\n        }\n      },\n    );\n\n  arena\n    .command('generation <generationId>')\n    .description(\"Fetch OpenRouter's record for a journaled generation\")\n    .action(async (generationId: string) => {\n      const data = await dependencies.createOpenRouter().fetchGeneration(generationId);\n      dependencies.stdout(JSON.stringify(data, null, 2));\n    });\n\n  program\n    .command('report')\n    .description('Verify journals and rebuild JSON and Markdown reports')\n    .option(\n      '-c, --category <category>',\n      `limit to one category (${CATEGORIES.join(', ')})`,\n      parseCategory,\n    )\n    .action((options: { category?: BenchmarkCategory }) => {\n      const categories = options.category ? [options.category] : CATEGORIES;\n      for (const category of categories) {\n        const snapshot = writeReports(dependencies.createStore(category));\n        dependencies.stdout(`✓ ${category}: wrote reports for ${snapshot.matches.length} matches`);\n      }\n    });\n\n  return program;\n}\n\nexport async function runCli(\n  argv: string[] = process.argv,\n  overrides: Partial<CliDependencies> = {},\n): Promise<void> {\n  const dependencies = { ...defaultDependencies, ...overrides };\n  const envResult = loadProjectEnv();\n  if (envResult.status === 'error') {\n    dependencies.stderr(\n      `Warning: project environment file could not be loaded: ${redactSecrets(envResult.reason)}`,\n    );\n  }\n  await buildProgram(dependencies).parseAsync(argv);\n}\n","import { existsSync } from 'node:fs';\nimport path from 'node:path';\n\nimport { findProjectRoot } from './paths.js';\n\nexport const ENV_PATH = path.join(findProjectRoot(import.meta.url), '.env');\n\nexport type ProjectEnvLoadResult =\n  { status: 'missing' } | { status: 'loaded' } | { status: 'error'; reason: string };\n\n/**\n * Loads the project-root .env into process.env regardless of the process cwd\n * or launch flags. Variables already present in the environment win, matching\n * node's --env-file precedence. Safe to call repeatedly: a key added to .env\n * after startup is picked up on the next call.\n */\nexport function loadProjectEnv(): ProjectEnvLoadResult {\n  if (!existsSync(ENV_PATH)) return { status: 'missing' };\n  try {\n    process.loadEnvFile(ENV_PATH);\n    return { status: 'loaded' };\n  } catch (error) {\n    return {\n      status: 'error',\n      reason: error instanceof Error ? error.message : String(error),\n    };\n  }\n}\n","#!/usr/bin/env node\n\nimport { runCli } from './commands.js';\nimport { sanitizeError } from './openrouter.js';\n\nrunCli().catch((error) => {\n  console.error(`BridgeBench: ${sanitizeError(error)}`);\n  process.exitCode = 1;\n});\n"]}

package/dist/cli.d.cts

@@ -0,0 +1 @@
+#!/usr/bin/env node

package/dist/cli.d.ts

@@ -0,0 +1 @@
+#!/usr/bin/env node