bridgebench 3.1.0-alpha.1 → 3.1.0-alpha.2

package/dist/{chunk-5PCDG7VA.cjs → chunk-X3LPZGHS.cjs}

@@ -5,7 +5,16 @@ var importMetaUrl = /* @__PURE__ */ getImportMetaUrl();
 // src/contracts/categories.ts
 var _zod = require('zod');
 var METHODOLOGY_VERSION = "arena-v0.3.0";
-var BenchmarkCategorySchema = _zod.z.enum(["reasoning", "hallucination"]);
+var BenchmarkCategorySchema = _zod.z.enum([
+  "reasoning",
+  "hallucination",
+  "security",
+  "bullshit",
+  "refactoring",
+  "debugging",
+  "generation",
+  "speed"
+]);
 var CATEGORIES = BenchmarkCategorySchema.options;
 var CATEGORY_CLUSTERS = {
   reasoning: [
@@ -23,6 +32,54 @@ var CATEGORY_CLUSTERS = {
     "knowledge-boundary",
     "conflicting-sources",
     "citation-fidelity"
+  ],
+  security: [
+    "vuln-discovery",
+    "taint-flow",
+    "authz-boundary",
+    "patch-verification",
+    "finding-triage",
+    "supply-chain"
+  ],
+  bullshit: [
+    "fabricated-concepts",
+    "crossed-domains",
+    "impossible-quantities",
+    "reversed-causality",
+    "plausible-pseudoscience",
+    "loaded-assumptions"
+  ],
+  refactoring: [
+    "behavior-preservation",
+    "extract-and-inline",
+    "dependency-decoupling",
+    "api-migration",
+    "dead-code-elimination",
+    "semantic-equivalence"
+  ],
+  debugging: [
+    "root-cause-isolation",
+    "regression-introduction",
+    "concurrency-defect",
+    "state-corruption",
+    "error-propagation",
+    "fix-adequacy"
+  ],
+  generation: [
+    "spec-conformance",
+    "edge-case-coverage",
+    "api-contract-adherence",
+    "algorithmic-correctness",
+    "constraint-satisfaction",
+    "interface-compatibility"
+  ],
+  speed: [
+    "short-completion",
+    "long-generation",
+    "structured-output",
+    "code-transformation",
+    "stepwise-reasoning",
+    "retrieval-synthesis"
   ]
 };
 var CATEGORY_META = {
@@ -33,6 +90,30 @@ var CATEGORY_META = {
   hallucination: {
     label: "Hallucination",
     tagline: "Tasks are seeded with false premises, missing evidence, and fabrication bait \u2014 the arena measures who stays grounded instead of inventing."
+  },
+  security: {
+    label: "Security",
+    tagline: "Tasks hide one defensible vulnerability among benign look-alikes, false positives, and shallow patches \u2014 the arena measures who proves reachable exploitability with evidence instead of crying wolf."
+  },
+  bullshit: {
+    label: "BullShit",
+    tagline: "Tasks seed confident nonsense \u2014 fabricated concepts, crossed domains, impossible quantities \u2014 among legitimate deliverables; the arena measures who corrects the user instead of fluently answering the unanswerable."
+  },
+  refactoring: {
+    label: "Refactoring",
+    tagline: "Tasks pair code with a transformation goal and candidate rewrites \u2014 the arena measures who preserves observable behavior and spots the rewrite that silently changes it."
+  },
+  debugging: {
+    label: "Debugging",
+    tagline: "Tasks supply a failing system and its evidence among red-herring causes and shallow fixes \u2014 the arena measures who isolates the one defensible root cause and the fix that actually holds."
+  },
+  generation: {
+    label: "Generation",
+    tagline: "Tasks pair a specification with candidate implementations \u2014 the arena measures who identifies the one that meets every constraint and edge case instead of the plausible near-miss."
+  },
+  speed: {
+    label: "Speed",
+    tagline: "Both competitors get the same task and race: no judges, no quality vote. The winner is decided deterministically by measured latency \u2014 time to first token and sustained output throughput settle who finishes the work faster."
   }
 };
 
@@ -51,7 +132,11 @@ var ModelCompletionSchema = _zod.z.object({
   costUsd: _zod.z.number().finite().nonnegative(),
   latencyMs: _zod.z.number().finite().nonnegative(),
   finishReason: _zod.z.string(),
-  attempts: _zod.z.number().int().positive().optional()
+  attempts: _zod.z.number().int().positive().optional(),
+  /** Time from request start to the first non-empty content delta (speed arena metric). */
+  ttftMs: _zod.z.number().finite().nonnegative().optional(),
+  /** Request start to stream completion for the successful attempt (speed arena metric). */
+  totalMs: _zod.z.number().finite().nonnegative().optional()
 });
 
 // src/contracts/journal.ts
@@ -125,6 +210,16 @@ var JUDGE_VERDICT_TRANSPORT_SCHEMA = {
     violations: { type: "array", items: { type: "string" } }
   }
 };
+var SpeedMetricSchema = _zod.z.object({
+  ttftMs: _zod.z.number().finite().nonnegative(),
+  tps: _zod.z.number().finite().nonnegative(),
+  totalMs: _zod.z.number().finite().nonnegative(),
+  outputTokens: _zod.z.number().finite().nonnegative()
+});
+var SpeedMetricsSchema = _zod.z.object({
+  a: SpeedMetricSchema,
+  b: SpeedMetricSchema
+});
 var NumericRecordSchema = _zod.z.record(_zod.z.number().finite());
 var JudgeVoteSchema = _zod.z.object({
   judgeModelId: _zod.z.string().min(1),
@@ -156,7 +251,7 @@ var MatchResultSchema = _zod.z.object({
     category: BenchmarkCategorySchema.optional(),
     cluster: _zod.z.string().min(1),
     publicHash: _zod.z.string().min(1),
-    privateHash: _zod.z.string().min(1)
+    privateHash: _zod.z.string().min(1).nullable()
   }),
   competitors: _zod.z.object({
     modelA: _zod.z.string().min(1),
@@ -164,9 +259,10 @@ var MatchResultSchema = _zod.z.object({
     responseA: CompetitorResponseSchema,
     responseB: CompetitorResponseSchema
   }),
-  outcome: _zod.z.enum(["judged", "forfeit", "no-contest"]),
+  outcome: _zod.z.enum(["judged", "forfeit", "no-contest", "speed-decided"]),
   winnerModelId: _zod.z.string().min(1).nullable(),
   panel: PanelDecisionSchema.nullable(),
+  speedMetrics: SpeedMetricsSchema.nullable().optional(),
   eloBefore: NumericRecordSchema,
   eloAfter: NumericRecordSchema,
   pointAwarded: _zod.z.boolean(),
@@ -218,6 +314,9 @@ var TaskPrivateSchema = _zod.z.object({
     completeness: _zod.z.string().min(1)
   }).strict()
 }).strict();
+function isCompleteArenaTask(task) {
+  return task.private !== null && task.privateHash !== null;
+}
 
 // src/contracts/index.ts
 var CONTRACTS_VERSION = "1.0.0";
@@ -245,4 +344,6 @@ var CONTRACTS_VERSION = "1.0.0";
 
 
 
-exports.importMetaUrl = importMetaUrl; exports.METHODOLOGY_VERSION = METHODOLOGY_VERSION; exports.BenchmarkCategorySchema = BenchmarkCategorySchema; exports.CATEGORIES = CATEGORIES; exports.CATEGORY_CLUSTERS = CATEGORY_CLUSTERS; exports.CATEGORY_META = CATEGORY_META; exports.ModelRoleSchema = ModelRoleSchema; exports.ModelCompletionSchema = ModelCompletionSchema; exports.CompetitorResponseSchema = CompetitorResponseSchema; exports.JudgeVerdictSchema = JudgeVerdictSchema; exports.JUDGE_VERDICT_TRANSPORT_SCHEMA = JUDGE_VERDICT_TRANSPORT_SCHEMA; exports.JudgeVoteSchema = JudgeVoteSchema; exports.PanelDecisionSchema = PanelDecisionSchema; exports.MatchResultSchema = MatchResultSchema; exports.competitorCost = competitorCost; exports.competitorOutputTokens = competitorOutputTokens; exports.competitorReasoningTokens = competitorReasoningTokens; exports.competitorContent = competitorContent; exports.TaskArtifactSchema = TaskArtifactSchema; exports.TaskPublicSchema = TaskPublicSchema; exports.TaskPrivateSchema = TaskPrivateSchema; exports.CONTRACTS_VERSION = CONTRACTS_VERSION;
+
+
+exports.importMetaUrl = importMetaUrl; exports.METHODOLOGY_VERSION = METHODOLOGY_VERSION; exports.BenchmarkCategorySchema = BenchmarkCategorySchema; exports.CATEGORIES = CATEGORIES; exports.CATEGORY_CLUSTERS = CATEGORY_CLUSTERS; exports.CATEGORY_META = CATEGORY_META; exports.ModelRoleSchema = ModelRoleSchema; exports.ModelCompletionSchema = ModelCompletionSchema; exports.CompetitorResponseSchema = CompetitorResponseSchema; exports.JudgeVerdictSchema = JudgeVerdictSchema; exports.JUDGE_VERDICT_TRANSPORT_SCHEMA = JUDGE_VERDICT_TRANSPORT_SCHEMA; exports.SpeedMetricsSchema = SpeedMetricsSchema; exports.JudgeVoteSchema = JudgeVoteSchema; exports.PanelDecisionSchema = PanelDecisionSchema; exports.MatchResultSchema = MatchResultSchema; exports.competitorCost = competitorCost; exports.competitorOutputTokens = competitorOutputTokens; exports.competitorReasoningTokens = competitorReasoningTokens; exports.competitorContent = competitorContent; exports.TaskArtifactSchema = TaskArtifactSchema; exports.TaskPublicSchema = TaskPublicSchema; exports.TaskPrivateSchema = TaskPrivateSchema; exports.isCompleteArenaTask = isCompleteArenaTask; exports.CONTRACTS_VERSION = CONTRACTS_VERSION;

package/dist/cli.cjs

@@ -5,7 +5,7 @@
 
 
 
-var _chunkXPFZBHV2cjs = require('./chunk-XPFZBHV2.cjs');
+var _chunkKCXQ5SAUcjs = require('./chunk-KCXQ5SAU.cjs');
 
 
 
@@ -16,7 +16,7 @@ var _chunkXPFZBHV2cjs = require('./chunk-XPFZBHV2.cjs');
 
 
 
-var _chunk5OZGHLDHcjs = require('./chunk-5OZGHLDH.cjs');
+var _chunkVAS6KNJAcjs = require('./chunk-VAS6KNJA.cjs');
 
 
 
@@ -27,12 +27,12 @@ var _chunk5OZGHLDHcjs = require('./chunk-5OZGHLDH.cjs');
 
 
 
-var _chunkSMG3PORScjs = require('./chunk-SMG3PORS.cjs');
+var _chunkQMOPRKWDcjs = require('./chunk-QMOPRKWD.cjs');
 
 
 
 
-var _chunk5PCDG7VAcjs = require('./chunk-5PCDG7VA.cjs');
+var _chunkX3LPZGHScjs = require('./chunk-X3LPZGHS.cjs');
 
 // src/commands.ts
 var _path = require('path'); var _path2 = _interopRequireDefault(_path);
@@ -41,7 +41,7 @@ var _commander = require('commander');
 // src/env.ts
 var _fs = require('fs');
 
-var ENV_PATH = _path2.default.join(_chunkSMG3PORScjs.findProjectRoot.call(void 0, _chunk5PCDG7VAcjs.importMetaUrl), ".env");
+var ENV_PATH = _path2.default.join(_chunkQMOPRKWDcjs.findProjectRoot.call(void 0, _chunkX3LPZGHScjs.importMetaUrl), ".env");
 function loadProjectEnv() {
   if (!_fs.existsSync.call(void 0, ENV_PATH)) return { status: "missing" };
   try {
@@ -63,17 +63,17 @@ var defaultDependencies = {
   setExitCode: (code) => {
     process.exitCode = code;
   },
-  createStore: (category) => new (0, _chunk5OZGHLDHcjs.ArenaStore)(_chunk5OZGHLDHcjs.categoryStoreConfig.call(void 0, category)),
-  createTaskLoader: (category) => new (0, _chunkSMG3PORScjs.TaskLoader)(category),
-  createOpenRouter: (logger) => new (0, _chunkSMG3PORScjs.OpenRouterClient)(_nullishCoalesce(process.env.OPENROUTER_API_KEY, () => ( "")), logger),
-  publishTasks: _chunkXPFZBHV2cjs.publishTasks,
-  publishJournal: _chunkXPFZBHV2cjs.publishJournal,
-  resolveApiConfig: _chunkXPFZBHV2cjs.resolveApiConfig
+  createStore: (category) => new (0, _chunkVAS6KNJAcjs.ArenaStore)(_chunkVAS6KNJAcjs.categoryStoreConfig.call(void 0, category)),
+  createTaskLoader: (category) => new (0, _chunkQMOPRKWDcjs.TaskLoader)(category),
+  createOpenRouter: (logger) => new (0, _chunkQMOPRKWDcjs.OpenRouterClient)(_nullishCoalesce(process.env.OPENROUTER_API_KEY, () => ( "")), logger),
+  publishTasks: _chunkKCXQ5SAUcjs.publishTasks,
+  publishJournal: _chunkKCXQ5SAUcjs.publishJournal,
+  resolveApiConfig: _chunkKCXQ5SAUcjs.resolveApiConfig
 };
 function parseCategory(value) {
-  const parsed = _chunk5PCDG7VAcjs.BenchmarkCategorySchema.safeParse(value);
+  const parsed = _chunkX3LPZGHScjs.BenchmarkCategorySchema.safeParse(value);
   if (!parsed.success) {
-    throw new (0, _commander.InvalidArgumentError)(`expected one of ${_chunk5PCDG7VAcjs.CATEGORIES.join(", ")}, received ${value}`);
+    throw new (0, _commander.InvalidArgumentError)(`expected one of ${_chunkX3LPZGHScjs.CATEGORIES.join(", ")}, received ${value}`);
   }
   return parsed.data;
 }
@@ -95,7 +95,7 @@ function collectOption(value, previous = []) {
   return [...previous, value];
 }
 function arenaRunConfigFromOptions(options) {
-  const competitorIds = _optionalChain([options, 'access', _ => _.competitor, 'optionalAccess', _2 => _2.length]) ? _chunkSMG3PORScjs.resolveCompetitorRoster.call(void 0, options.competitor).map((model) => model.id) : void 0;
+  const competitorIds = _optionalChain([options, 'access', _ => _.competitor, 'optionalAccess', _2 => _2.length]) ? _chunkQMOPRKWDcjs.resolveCompetitorRoster.call(void 0, options.competitor).map((model) => model.id) : void 0;
   return {
     category: options.category,
     seed: options.seed,
@@ -111,7 +111,7 @@ function selectedCategories(options, command) {
     throw new Error(`${command} accepts either --category or --all, not both`);
   }
   if (options.category) return [options.category];
-  if (options.all) return [..._chunk5PCDG7VAcjs.CATEGORIES];
+  if (options.all) return [..._chunkX3LPZGHScjs.CATEGORIES];
   throw new Error(`${command} requires --category <category> or --all`);
 }
 function logRoot(category) {
@@ -123,19 +123,21 @@ function displayPath(target) {
 }
 function buildProgram(overrides = {}) {
   const dependencies = { ...defaultDependencies, ...overrides };
-  const program = new (0, _commander.Command)().name(_chunk5OZGHLDHcjs.PACKAGE_NAME).description("Autonomous BridgeBench arenas for reasoning and hallucination").version(_chunk5OZGHLDHcjs.ENGINE_VERSION).showHelpAfterError().configureHelp({ sortOptions: true, sortSubcommands: true }).configureOutput({
+  const program = new (0, _commander.Command)().name(_chunkVAS6KNJAcjs.PACKAGE_NAME).description(
+    "Autonomous BridgeBench arenas: reasoning, hallucination, security, bullshit, refactoring, debugging, generation, and speed"
+  ).version(_chunkVAS6KNJAcjs.ENGINE_VERSION).showHelpAfterError().configureHelp({ sortOptions: true, sortSubcommands: true }).configureOutput({
     writeOut: (value) => dependencies.stdout(value.replace(/\n$/, "")),
     writeErr: (value) => dependencies.stderr(value.replace(/\n$/, ""))
   });
   const models = program.command("models").description("Inspect or validate the pinned model registry");
   models.command("list").description("List enabled competitors and judges").action(() => {
-    for (const model of _chunkSMG3PORScjs.listModels.call(void 0, )) {
+    for (const model of _chunkQMOPRKWDcjs.listModels.call(void 0, )) {
       dependencies.stdout(`${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 _chunkSMG3PORScjs.listModels.call(void 0, )) {
+    for (const model of _chunkQMOPRKWDcjs.listModels.call(void 0, )) {
       await client.validateModel(model);
       dependencies.stdout(`\u2713 ${model.id} -> ${model.canonicalSlug}`);
     }
@@ -143,20 +145,20 @@ function buildProgram(overrides = {}) {
   const tasks = program.command("tasks").description("Validate or publish task packs");
   tasks.command("validate").description("Validate public task schemas and pack balance").option(
     "-c, --category <category>",
-    `limit to one category (${_chunk5PCDG7VAcjs.CATEGORIES.join(", ")})`,
+    `limit to one category (${_chunkX3LPZGHScjs.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 _chunkSMG3PORScjs.validatePublicTaskFile.call(void 0, options.file);
+      const task = await _chunkQMOPRKWDcjs.validatePublicTaskFile.call(void 0, options.file);
       dependencies.stdout(
         `\u2713 ${task.public.id}: public task validated (${task.public.category}/${task.public.cluster})`
       );
       return;
     }
-    const categories = options.category ? [options.category] : _chunk5PCDG7VAcjs.CATEGORIES;
+    const categories = options.category ? [options.category] : _chunkX3LPZGHScjs.CATEGORIES;
     for (const category of categories) {
       const loader = dependencies.createTaskLoader(category);
       const loaded = await loader.loadAll();
@@ -169,12 +171,12 @@ function buildProgram(overrides = {}) {
   });
   tasks.command("publish").description("Publish public and private task halves to the configured API").option(
     "-c, --category <category>",
-    `category to publish (${_chunk5PCDG7VAcjs.CATEGORIES.join(", ")})`,
+    `category to publish (${_chunkX3LPZGHScjs.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 ${_chunkXPFZBHV2cjs.publishTarget.call(void 0, config)}`);
+    dependencies.stdout(`Publishing ${categories.join(", ")} tasks to ${_chunkKCXQ5SAUcjs.publishTarget.call(void 0, config)}`);
     for (const category of categories) {
       const result = await dependencies.publishTasks(category, config);
       dependencies.stdout(`\u2713 ${category}: published ${result.imported} tasks`);
@@ -183,7 +185,7 @@ function buildProgram(overrides = {}) {
   const arena = program.command("arena").description("Run and inspect arena matches");
   arena.command("run").description("Run a paid deterministic match schedule").requiredOption(
     "-c, --category <category>",
-    `arena to run (${_chunk5PCDG7VAcjs.CATEGORIES.join(", ")})`,
+    `arena to run (${_chunkX3LPZGHScjs.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>",
@@ -196,7 +198,7 @@ function buildProgram(overrides = {}) {
     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, _chunkSMG3PORScjs.FileArenaLogger)({
+    const logger = new (0, _chunkQMOPRKWDcjs.FileArenaLogger)({
       dir: logRoot(config.category),
       verbose: options.debug
     });
@@ -210,14 +212,15 @@ function buildProgram(overrides = {}) {
     try {
       dependencies.stdout(`Run log: ${displayPath(logger.filePath)}`);
       const store = dependencies.createStore(config.category);
-      const loaded = await dependencies.createTaskLoader(config.category).loadAll({ requirePrivate: true });
+      const requirePrivate = config.category !== "speed";
+      const loaded = await dependencies.createTaskLoader(config.category).loadAll(requirePrivate ? { requirePrivate: true } : {});
       const progressOutput = (event) => {
         if (event.type !== "match.completed") return;
         dependencies.stdout(
           `[${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, _chunk5OZGHLDHcjs.ArenaRunner)(
+      const runner = new (0, _chunkVAS6KNJAcjs.ArenaRunner)(
         dependencies.createOpenRouter(logger),
         store,
         progressOutput,
@@ -241,9 +244,9 @@ function buildProgram(overrides = {}) {
       }
       const runMatches = store.readAll().filter((match) => match.runId === result.runId);
       if (runMatches.length > 0) {
-        const triage = _chunk5OZGHLDHcjs.triageJournal.call(void 0, runMatches);
+        const triage = _chunkVAS6KNJAcjs.triageJournal.call(void 0, runMatches);
         dependencies.stdout(`=== Run health ===
-${_chunk5OZGHLDHcjs.formatTriage.call(void 0, triage)}`);
+${_chunkVAS6KNJAcjs.formatTriage.call(void 0, triage)}`);
       }
     } finally {
       process.removeListener("SIGINT", requestCancellation);
@@ -252,7 +255,7 @@ ${_chunk5OZGHLDHcjs.formatTriage.call(void 0, triage)}`);
   });
   arena.command("remote-run").description("Run a match schedule against the configured API and stream live arena events").requiredOption(
     "-c, --category <category>",
-    `arena to run (${_chunk5PCDG7VAcjs.CATEGORIES.join(", ")})`,
+    `arena to run (${_chunkX3LPZGHScjs.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>",
@@ -266,16 +269,16 @@ ${_chunk5OZGHLDHcjs.formatTriage.call(void 0, triage)}`);
   ).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, _chunkSMG3PORScjs.FileArenaLogger)({
+    const logger = new (0, _chunkQMOPRKWDcjs.FileArenaLogger)({
       dir: logRoot(config.category),
       verbose: options.debug
     });
     dependencies.stdout(
-      `Remote ${config.category} run against ${_chunkXPFZBHV2cjs.publishTarget.call(void 0, apiConfig)} (${options.mock ? "mock" : "openrouter"})`
+      `Remote ${config.category} run against ${_chunkKCXQ5SAUcjs.publishTarget.call(void 0, apiConfig)} (${options.mock ? "mock" : "openrouter"})`
     );
     dependencies.stdout(`Run log: ${displayPath(logger.filePath)}`);
     try {
-      const result = await _chunkXPFZBHV2cjs.runRemoteArena.call(void 0, apiConfig, {
+      const result = await _chunkKCXQ5SAUcjs.runRemoteArena.call(void 0, apiConfig, {
         config,
         mock: options.mock,
         publishMatches: options.publishMatches,
@@ -291,21 +294,21 @@ ${_chunk5OZGHLDHcjs.formatTriage.call(void 0, triage)}`);
   });
   arena.command("verify").description("Validate and replay outcomes, points, manifests, and Elo").option(
     "-c, --category <category>",
-    `journal category (${_chunk5PCDG7VAcjs.CATEGORIES.join(", ")})`,
+    `journal category (${_chunkX3LPZGHScjs.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 = _chunk5OZGHLDHcjs.categoryStoreConfig.call(void 0, options.category);
+      const defaults = _chunkVAS6KNJAcjs.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, _chunk5OZGHLDHcjs.ArenaStore)({
+      const store = new (0, _chunkVAS6KNJAcjs.ArenaStore)({
         ...defaults,
         journalPath,
         runsDir,
         readOnly: true
       });
-      const verified = _chunk5OZGHLDHcjs.verifyJournal.call(void 0, store.readAll(), options.category, {
+      const verified = _chunkVAS6KNJAcjs.verifyJournal.call(void 0, store.readAll(), options.category, {
         manifestForRun: (runId) => store.readRunManifest(runId),
         requireManifests: true
       });
@@ -323,17 +326,17 @@ ${_chunk5OZGHLDHcjs.formatTriage.call(void 0, triage)}`);
   );
   arena.command("publish").description("Publish a verified match journal to the configured API").option(
     "-c, --category <category>",
-    `journal to publish (${_chunk5PCDG7VAcjs.CATEGORIES.join(", ")})`,
+    `journal to publish (${_chunkX3LPZGHScjs.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 ${_chunkXPFZBHV2cjs.publishTarget.call(void 0, config)}`
+      `Publishing ${categories.join(", ")} journals to ${_chunkKCXQ5SAUcjs.publishTarget.call(void 0, config)}`
     );
     for (const category of categories) {
       const store = dependencies.createStore(category);
-      _chunk5OZGHLDHcjs.verifyJournal.call(void 0, store.readAll(), category, {
+      _chunkVAS6KNJAcjs.verifyJournal.call(void 0, store.readAll(), category, {
         manifestForRun: (runId) => store.readRunManifest(runId),
         requireManifests: true
       });
@@ -350,7 +353,7 @@ ${_chunk5OZGHLDHcjs.formatTriage.call(void 0, triage)}`);
   });
   arena.command("triage").description("Analyze failures, suspicious responses, and judging anomalies").option(
     "-c, --category <category>",
-    `journal to analyze (${_chunk5PCDG7VAcjs.CATEGORIES.join(", ")})`,
+    `journal to analyze (${_chunkX3LPZGHScjs.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(
@@ -362,9 +365,9 @@ ${_chunk5OZGHLDHcjs.formatTriage.call(void 0, triage)}`);
         dependencies.stdout(message);
         return;
       }
-      const reports = _chunk5OZGHLDHcjs.triageJournal.call(void 0, results);
+      const reports = _chunkVAS6KNJAcjs.triageJournal.call(void 0, results);
       dependencies.stdout(
-        options.json ? JSON.stringify(reports, null, 2) : _chunk5OZGHLDHcjs.formatTriage.call(void 0, reports)
+        options.json ? JSON.stringify(reports, null, 2) : _chunkVAS6KNJAcjs.formatTriage.call(void 0, reports)
       );
       const anomalies = reports.reduce((sum, report) => sum + report.anomalies.length, 0);
       if (options.strict && anomalies > 0) {
@@ -379,12 +382,12 @@ ${_chunk5OZGHLDHcjs.formatTriage.call(void 0, triage)}`);
   });
   program.command("report").description("Verify journals and rebuild JSON and Markdown reports").option(
     "-c, --category <category>",
-    `limit to one category (${_chunk5PCDG7VAcjs.CATEGORIES.join(", ")})`,
+    `limit to one category (${_chunkX3LPZGHScjs.CATEGORIES.join(", ")})`,
     parseCategory
   ).action((options) => {
-    const categories = options.category ? [options.category] : _chunk5PCDG7VAcjs.CATEGORIES;
+    const categories = options.category ? [options.category] : _chunkX3LPZGHScjs.CATEGORIES;
     for (const category of categories) {
-      const snapshot = _chunk5OZGHLDHcjs.writeReports.call(void 0, dependencies.createStore(category));
+      const snapshot = _chunkVAS6KNJAcjs.writeReports.call(void 0, dependencies.createStore(category));
       dependencies.stdout(`\u2713 ${category}: wrote reports for ${snapshot.matches.length} matches`);
     }
   });
@@ -395,7 +398,7 @@ async function runCli(argv = process.argv, overrides = {}) {
   const envResult = loadProjectEnv();
   if (envResult.status === "error") {
     dependencies.stderr(
-      `Warning: project environment file could not be loaded: ${_chunkSMG3PORScjs.redactSecrets.call(void 0, envResult.reason)}`
+      `Warning: project environment file could not be loaded: ${_chunkQMOPRKWDcjs.redactSecrets.call(void 0, envResult.reason)}`
     );
   }
   await buildProgram(dependencies).parseAsync(argv);
@@ -403,6 +406,6 @@ async function runCli(argv = process.argv, overrides = {}) {
 
 // src/cli.ts
 runCli().catch((error) => {
-  console.error(`BridgeBench: ${_chunkSMG3PORScjs.sanitizeError.call(void 0, error)}`);
+  console.error(`BridgeBench: ${_chunkQMOPRKWDcjs.sanitizeError.call(void 0, error)}`);
   process.exitCode = 1;
 });

package/dist/cli.js

@@ -5,7 +5,7 @@ import {
   publishTasks,
   resolveApiConfig,
   runRemoteArena
-} from "./chunk-MBC4C3VA.js";
+} from "./chunk-CJGHBY54.js";
 import {
   ArenaRunner,
   ArenaStore,
@@ -16,7 +16,7 @@ import {
   triageJournal,
   verifyJournal,
   writeReports
-} from "./chunk-KGZEQ6R5.js";
+} from "./chunk-IUPFMGUL.js";
 import {
   FileArenaLogger,
   OpenRouterClient,
@@ -27,11 +27,11 @@ import {
   resolveCompetitorRoster,
   sanitizeError,
   validatePublicTaskFile
-} from "./chunk-3CX4FIFZ.js";
+} from "./chunk-DVMGL3L7.js";
 import {
   BenchmarkCategorySchema,
   CATEGORIES
-} from "./chunk-UXFKTSQS.js";
+} from "./chunk-AY45YLYL.js";
 
 // src/commands.ts
 import path2 from "path";
@@ -122,7 +122,9 @@ function displayPath(target) {
 }
 function buildProgram(overrides = {}) {
   const dependencies = { ...defaultDependencies, ...overrides };
-  const program = new Command().name(PACKAGE_NAME).description("Autonomous BridgeBench arenas for reasoning and hallucination").version(ENGINE_VERSION).showHelpAfterError().configureHelp({ sortOptions: true, sortSubcommands: true }).configureOutput({
+  const program = new Command().name(PACKAGE_NAME).description(
+    "Autonomous BridgeBench arenas: reasoning, hallucination, security, bullshit, refactoring, debugging, generation, and speed"
+  ).version(ENGINE_VERSION).showHelpAfterError().configureHelp({ sortOptions: true, sortSubcommands: true }).configureOutput({
     writeOut: (value) => dependencies.stdout(value.replace(/\n$/, "")),
     writeErr: (value) => dependencies.stderr(value.replace(/\n$/, ""))
   });
@@ -209,7 +211,8 @@ function buildProgram(overrides = {}) {
     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 requirePrivate = config.category !== "speed";
+      const loaded = await dependencies.createTaskLoader(config.category).loadAll(requirePrivate ? { requirePrivate: true } : {});
       const progressOutput = (event) => {
         if (event.type !== "match.completed") return;
         dependencies.stdout(

package/dist/client.cjs

@@ -16,10 +16,10 @@
 
 
 
-var _chunkXPFZBHV2cjs = require('./chunk-XPFZBHV2.cjs');
-require('./chunk-5OZGHLDH.cjs');
-require('./chunk-SMG3PORS.cjs');
-require('./chunk-5PCDG7VA.cjs');
+var _chunkKCXQ5SAUcjs = require('./chunk-KCXQ5SAU.cjs');
+require('./chunk-VAS6KNJA.cjs');
+require('./chunk-QMOPRKWD.cjs');
+require('./chunk-X3LPZGHS.cjs');
 
 
 
@@ -38,4 +38,4 @@ require('./chunk-5PCDG7VA.cjs');
 
 
 
-exports.ADMIN_KEY_HEADER = _chunkXPFZBHV2cjs.ADMIN_KEY_HEADER; exports.DEFAULT_API_TIMEOUT_MS = _chunkXPFZBHV2cjs.DEFAULT_API_TIMEOUT_MS; exports.REQUEST_SPACING_MS = _chunkXPFZBHV2cjs.REQUEST_SPACING_MS; exports.RemoteArenaEventSink = _chunkXPFZBHV2cjs.RemoteArenaEventSink; exports.chunk = _chunkXPFZBHV2cjs.chunk; exports.createRemoteRun = _chunkXPFZBHV2cjs.createRemoteRun; exports.delay = _chunkXPFZBHV2cjs.delay; exports.fetchExecutionPack = _chunkXPFZBHV2cjs.fetchExecutionPack; exports.getJson = _chunkXPFZBHV2cjs.getJson; exports.postChunks = _chunkXPFZBHV2cjs.postChunks; exports.postJson = _chunkXPFZBHV2cjs.postJson; exports.publishJournal = _chunkXPFZBHV2cjs.publishJournal; exports.publishJournalFromStore = _chunkXPFZBHV2cjs.publishJournalFromStore; exports.publishTarget = _chunkXPFZBHV2cjs.publishTarget; exports.publishTasks = _chunkXPFZBHV2cjs.publishTasks; exports.resolveApiConfig = _chunkXPFZBHV2cjs.resolveApiConfig; exports.runRemoteArena = _chunkXPFZBHV2cjs.runRemoteArena;
+exports.ADMIN_KEY_HEADER = _chunkKCXQ5SAUcjs.ADMIN_KEY_HEADER; exports.DEFAULT_API_TIMEOUT_MS = _chunkKCXQ5SAUcjs.DEFAULT_API_TIMEOUT_MS; exports.REQUEST_SPACING_MS = _chunkKCXQ5SAUcjs.REQUEST_SPACING_MS; exports.RemoteArenaEventSink = _chunkKCXQ5SAUcjs.RemoteArenaEventSink; exports.chunk = _chunkKCXQ5SAUcjs.chunk; exports.createRemoteRun = _chunkKCXQ5SAUcjs.createRemoteRun; exports.delay = _chunkKCXQ5SAUcjs.delay; exports.fetchExecutionPack = _chunkKCXQ5SAUcjs.fetchExecutionPack; exports.getJson = _chunkKCXQ5SAUcjs.getJson; exports.postChunks = _chunkKCXQ5SAUcjs.postChunks; exports.postJson = _chunkKCXQ5SAUcjs.postJson; exports.publishJournal = _chunkKCXQ5SAUcjs.publishJournal; exports.publishJournalFromStore = _chunkKCXQ5SAUcjs.publishJournalFromStore; exports.publishTarget = _chunkKCXQ5SAUcjs.publishTarget; exports.publishTasks = _chunkKCXQ5SAUcjs.publishTasks; exports.resolveApiConfig = _chunkKCXQ5SAUcjs.resolveApiConfig; exports.runRemoteArena = _chunkKCXQ5SAUcjs.runRemoteArena;

package/dist/client.d.cts

@@ -1,7 +1,7 @@
 import { z } from 'zod';
-import { A as ArenaStore, a as ArenaLogger } from './logger-CCR9Mg1c.cjs';
-import { B as BenchmarkCategory, d as CompleteArenaTask } from './tasks-CpaCJ6JE.cjs';
-import { f as ArenaRunConfig, A as ArenaEvent } from './reports-4CejmOHf.cjs';
+import { A as ArenaStore, a as ArenaLogger } from './logger-BByta-7V.cjs';
+import { B as BenchmarkCategory, d as CompleteArenaTask } from './tasks-BmhWuMBD.cjs';
+import { f as ArenaRunConfig, A as ArenaEvent } from './reports-DPpOoOux.cjs';
 
 declare const ADMIN_KEY_HEADER = "x-bridgebench-admin-key";
 declare const DEFAULT_API_TIMEOUT_MS = 30000;

package/dist/client.d.ts

@@ -1,7 +1,7 @@
 import { z } from 'zod';
-import { A as ArenaStore, a as ArenaLogger } from './logger-QJU7SBDz.js';
-import { B as BenchmarkCategory, d as CompleteArenaTask } from './tasks-CpaCJ6JE.js';
-import { f as ArenaRunConfig, A as ArenaEvent } from './reports-s2CTnGN8.js';
+import { A as ArenaStore, a as ArenaLogger } from './logger-BQf29BLe.js';
+import { B as BenchmarkCategory, d as CompleteArenaTask } from './tasks-BmhWuMBD.js';
+import { f as ArenaRunConfig, A as ArenaEvent } from './reports-B8TCJtPr.js';
 
 declare const ADMIN_KEY_HEADER = "x-bridgebench-admin-key";
 declare const DEFAULT_API_TIMEOUT_MS = 30000;

package/dist/client.js

@@ -16,10 +16,10 @@ import {
   publishTasks,
   resolveApiConfig,
   runRemoteArena
-} from "./chunk-MBC4C3VA.js";
-import "./chunk-KGZEQ6R5.js";
-import "./chunk-3CX4FIFZ.js";
-import "./chunk-UXFKTSQS.js";
+} from "./chunk-CJGHBY54.js";
+import "./chunk-IUPFMGUL.js";
+import "./chunk-DVMGL3L7.js";
+import "./chunk-AY45YLYL.js";
 export {
   ADMIN_KEY_HEADER,
   DEFAULT_API_TIMEOUT_MS,

package/dist/contracts/index.cjs

@@ -20,9 +20,9 @@
 
 
 
-var _chunk5PCDG7VAcjs = require('../chunk-5PCDG7VA.cjs');
 
 
+var _chunkX3LPZGHScjs = require('../chunk-X3LPZGHS.cjs');
 
 
 
@@ -43,4 +43,8 @@ var _chunk5PCDG7VAcjs = require('../chunk-5PCDG7VA.cjs');
 
 
 
-exports.BenchmarkCategorySchema = _chunk5PCDG7VAcjs.BenchmarkCategorySchema; exports.CATEGORIES = _chunk5PCDG7VAcjs.CATEGORIES; exports.CATEGORY_CLUSTERS = _chunk5PCDG7VAcjs.CATEGORY_CLUSTERS; exports.CATEGORY_META = _chunk5PCDG7VAcjs.CATEGORY_META; exports.CONTRACTS_VERSION = _chunk5PCDG7VAcjs.CONTRACTS_VERSION; exports.CompetitorResponseSchema = _chunk5PCDG7VAcjs.CompetitorResponseSchema; exports.JUDGE_VERDICT_TRANSPORT_SCHEMA = _chunk5PCDG7VAcjs.JUDGE_VERDICT_TRANSPORT_SCHEMA; exports.JudgeVerdictSchema = _chunk5PCDG7VAcjs.JudgeVerdictSchema; exports.JudgeVoteSchema = _chunk5PCDG7VAcjs.JudgeVoteSchema; exports.METHODOLOGY_VERSION = _chunk5PCDG7VAcjs.METHODOLOGY_VERSION; exports.MatchResultSchema = _chunk5PCDG7VAcjs.MatchResultSchema; exports.ModelCompletionSchema = _chunk5PCDG7VAcjs.ModelCompletionSchema; exports.ModelRoleSchema = _chunk5PCDG7VAcjs.ModelRoleSchema; exports.PanelDecisionSchema = _chunk5PCDG7VAcjs.PanelDecisionSchema; exports.TaskArtifactSchema = _chunk5PCDG7VAcjs.TaskArtifactSchema; exports.TaskPrivateSchema = _chunk5PCDG7VAcjs.TaskPrivateSchema; exports.TaskPublicSchema = _chunk5PCDG7VAcjs.TaskPublicSchema; exports.competitorContent = _chunk5PCDG7VAcjs.competitorContent; exports.competitorCost = _chunk5PCDG7VAcjs.competitorCost; exports.competitorOutputTokens = _chunk5PCDG7VAcjs.competitorOutputTokens; exports.competitorReasoningTokens = _chunk5PCDG7VAcjs.competitorReasoningTokens;
+
+
+
+
+exports.BenchmarkCategorySchema = _chunkX3LPZGHScjs.BenchmarkCategorySchema; exports.CATEGORIES = _chunkX3LPZGHScjs.CATEGORIES; exports.CATEGORY_CLUSTERS = _chunkX3LPZGHScjs.CATEGORY_CLUSTERS; exports.CATEGORY_META = _chunkX3LPZGHScjs.CATEGORY_META; exports.CONTRACTS_VERSION = _chunkX3LPZGHScjs.CONTRACTS_VERSION; exports.CompetitorResponseSchema = _chunkX3LPZGHScjs.CompetitorResponseSchema; exports.JUDGE_VERDICT_TRANSPORT_SCHEMA = _chunkX3LPZGHScjs.JUDGE_VERDICT_TRANSPORT_SCHEMA; exports.JudgeVerdictSchema = _chunkX3LPZGHScjs.JudgeVerdictSchema; exports.JudgeVoteSchema = _chunkX3LPZGHScjs.JudgeVoteSchema; exports.METHODOLOGY_VERSION = _chunkX3LPZGHScjs.METHODOLOGY_VERSION; exports.MatchResultSchema = _chunkX3LPZGHScjs.MatchResultSchema; exports.ModelCompletionSchema = _chunkX3LPZGHScjs.ModelCompletionSchema; exports.ModelRoleSchema = _chunkX3LPZGHScjs.ModelRoleSchema; exports.PanelDecisionSchema = _chunkX3LPZGHScjs.PanelDecisionSchema; exports.SpeedMetricsSchema = _chunkX3LPZGHScjs.SpeedMetricsSchema; exports.TaskArtifactSchema = _chunkX3LPZGHScjs.TaskArtifactSchema; exports.TaskPrivateSchema = _chunkX3LPZGHScjs.TaskPrivateSchema; exports.TaskPublicSchema = _chunkX3LPZGHScjs.TaskPublicSchema; exports.competitorContent = _chunkX3LPZGHScjs.competitorContent; exports.competitorCost = _chunkX3LPZGHScjs.competitorCost; exports.competitorOutputTokens = _chunkX3LPZGHScjs.competitorOutputTokens; exports.competitorReasoningTokens = _chunkX3LPZGHScjs.competitorReasoningTokens; exports.isCompleteArenaTask = _chunkX3LPZGHScjs.isCompleteArenaTask;