@tangle-network/agent-eval 0.34.1 → 0.36.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/README.md CHANGED
@@ -222,6 +222,7 @@ const next = await analyzeOptimizationResult(campaign, { researcher })
222
222
  | `@tangle-network/agent-eval/rl` | adapters, verifiable rewards, preferences, OPE, PRM, contamination, tournaments, adversarial, compute curves, auto-research |
223
223
  | `@tangle-network/agent-eval/wire` | HTTP/RPC server + schemas (same protocol the Python client speaks) |
224
224
  | `@tangle-network/agent-eval/benchmarks` | benchmark adapter contracts and reference wrappers |
225
+ | `@tangle-network/agent-eval/matrix` | N-axis cartesian runner over substrate types — see [`src/matrix/`](./src/matrix/) |
225
226
 
226
227
  The root export remains available for convenience; new code should prefer
227
228
  focused subpaths. Anything under `/rl`, `/pipelines`, `/meta-eval`, `/prm`,
@@ -294,6 +295,39 @@ and runtime. See [`examples/`](./examples/).
294
295
  - [`examples/production-loop`](./examples/production-loop/README.md):
295
296
  ingest prod traces + feedback, cluster failures, evolve, gate, open a PR.
296
297
 
298
+ ## Matrix
299
+
300
+ `@tangle-network/agent-eval/matrix` is an N-axis cartesian runner over the
301
+ substrate types you already use — `AgentProfile` from
302
+ `@tangle-network/sandbox`, `Driver` / `Validator` from
303
+ `@tangle-network/agent-runtime`, rubric records, anything. It does not wrap
304
+ substrate types; the caller passes them in axis values, the runner iterates
305
+ the cartesian, and the aggregator returns per-axis pass / score / cost /
306
+ duration summaries.
307
+
308
+ ```ts
309
+ import { runAgentMatrix } from '@tangle-network/agent-eval/matrix'
310
+
311
+ const result = await runAgentMatrix({
312
+ axes: [
313
+ { name: 'scenario', values: scenarios.map((s) => ({ id: s.id, value: s })) },
314
+ { name: 'profile', values: profiles.map((p) => ({ id: p.name, value: p })) },
315
+ { name: 'thinking', values: [
316
+ { id: 'low', value: 'low' }, { id: 'high', value: 'high' },
317
+ ] },
318
+ ],
319
+ reps: 3,
320
+ maxConcurrency: 4,
321
+ costCeiling: 5.0,
322
+ filter: (cell) => !(cell.axes.scenario.value.hard === 5 && cell.axes.thinking.id === 'low'),
323
+ runCell: async (cell) => runScenario(cell.axes.scenario.value, cell.axes.profile.value),
324
+ })
325
+
326
+ console.log(result.byAxis.profile) // per-profile passRate / meanScore / p90 / cost
327
+ ```
328
+
329
+ See [`src/matrix/`](./src/matrix/) for the full surface.
330
+
297
331
  ## Docs
298
332
 
299
333
  Read in this order:
@@ -937,6 +937,33 @@ function startServer(opts = {}) {
937
937
  console.log(`[agent-eval] serving on http://${address}:${actualPort}`);
938
938
  });
939
939
  }
940
+ function startServerAsync(opts = {}) {
941
+ const app = createApp(opts);
942
+ const port = opts.port ?? 5005;
943
+ const host = opts.host ?? "127.0.0.1";
944
+ return new Promise((resolve2, reject) => {
945
+ let settled = false;
946
+ let server;
947
+ server = serve({ fetch: app.fetch, port, hostname: host }, ({ address, port: actualPort }) => {
948
+ if (settled) return;
949
+ settled = true;
950
+ console.log(`[agent-eval] serving on http://${address}:${actualPort}`);
951
+ resolve2({
952
+ server,
953
+ port: actualPort,
954
+ host: address,
955
+ close: () => new Promise((res, rej) => {
956
+ server.close((err) => err ? rej(err) : res());
957
+ })
958
+ });
959
+ });
960
+ server.on("error", (err) => {
961
+ if (settled) return;
962
+ settled = true;
963
+ reject(err);
964
+ });
965
+ });
966
+ }
940
967
 
941
968
  export {
942
969
  RubricDimensionSchema,
@@ -972,6 +999,7 @@ export {
972
999
  runRpcOnce,
973
1000
  runRpcBatch,
974
1001
  createApp,
975
- startServer
1002
+ startServer,
1003
+ startServerAsync
976
1004
  };
977
- //# sourceMappingURL=chunk-YU3G6I7F.js.map
1005
+ //# sourceMappingURL=chunk-63EPZQUZ.js.map
@@ -1 +1 @@
1
- {"version":3,"sources":["../src/wire/schemas.ts","../src/wire/rubrics.ts","../src/wire/handlers.ts","../src/wire/openapi.ts","../src/wire/rpc.ts","../src/wire/server.ts"],"sourcesContent":["/**\n * Wire-protocol schemas.\n *\n * These Zod schemas are the contract between the agent-eval runtime and\n * any non-TypeScript client (Python, Rust, Go, …). They get rendered to\n * OpenAPI by `wire/openapi.ts` and code-generators consume that spec to\n * produce typed clients in other languages.\n *\n * Rule: if it's not in this file, it isn't on the wire. Keep names and\n * shapes self-explanatory — every field has a `.describe()` so the\n * generated docs are useful without reading the source.\n */\nimport { extendZodWithOpenApi } from '@asteasolutions/zod-to-openapi'\nimport { z } from 'zod'\n\nextendZodWithOpenApi(z)\n\n// ── Building blocks ─────────────────────────────────────────────────\n\nexport const RubricDimensionSchema = z\n .object({\n id: z\n .string()\n .min(1)\n .describe('Short stable id like \"buyer_quality\" — used as the key in scoring output.'),\n description: z\n .string()\n .min(1)\n .describe('One-line plain-English meaning. Read by humans reviewing low scores.'),\n weight: z\n .number()\n .min(0)\n .default(1)\n .describe('Relative weight in the composite score. Default 1; 0 disables.'),\n min: z.number().default(0).describe('Lower bound of valid score for this dimension.'),\n max: z.number().default(1).describe('Upper bound of valid score for this dimension.'),\n })\n .openapi('RubricDimension')\n\nexport const FailureModeSchema = z\n .object({\n id: z.string().min(1).describe('Short stable id like \"ai-cadence\" — used in detection lists.'),\n description: z.string().min(1).describe('Plain-English description of the failure pattern.'),\n })\n .openapi('FailureMode')\n\n// ── Rubric ──────────────────────────────────────────────────────────\n\nexport const RubricSchema = z\n .object({\n name: z\n .string()\n .min(1)\n .describe('Stable name like \"anti-slop\" — used by clients to invoke this rubric.'),\n description: z\n .string()\n .min(1)\n .describe('What this rubric measures. Shown in /v1/rubrics listing.'),\n systemPrompt: z\n .string()\n .min(1)\n .describe(\n 'Instructs the judging LLM. Should explain the persona (e.g. \"senior engineer reviewing voice\"), what to score on, and what to return.',\n ),\n dimensions: z\n .array(RubricDimensionSchema)\n .min(1)\n .describe('Scoring axes. The composite score is a weighted sum of these.'),\n failureModes: z\n .array(FailureModeSchema)\n .default([])\n .describe('Patterns to detect; each detected mode appears in the result.failureModes list.'),\n wins: z\n .array(FailureModeSchema)\n .default([])\n .describe('Positive patterns; each detected one appears in the result.wins list.'),\n })\n .openapi('Rubric')\n\n// ── Judge call ──────────────────────────────────────────────────────\n\nexport const JudgeRequestSchema = z\n .object({\n rubricName: z\n .string()\n .optional()\n .describe('Use a built-in rubric by name. Mutually exclusive with `rubric`.'),\n rubric: RubricSchema.optional().describe(\n 'Inline rubric definition. Mutually exclusive with `rubricName`.',\n ),\n content: z\n .string()\n .min(1)\n .describe('The text being judged — a tweet, a blog post, a code snippet, anything stringly.'),\n context: z\n .record(z.string(), z.unknown())\n .optional()\n .describe(\n 'Free-form metadata for the rubric to use — analytics, source URL, author, etc. Surfaced to the LLM.',\n ),\n model: z\n .string()\n .optional()\n .describe('Override the judge model (default routes via tcloud). e.g. \"claude-opus-4-7\".'),\n })\n .refine((v) => Boolean(v.rubricName) !== Boolean(v.rubric), {\n message: 'Provide exactly one of `rubricName` or `rubric`.',\n })\n .openapi('JudgeRequest')\n\nexport const JudgeResultSchema = z\n .object({\n composite: z\n .number()\n .min(0)\n .max(1)\n .describe('Weighted combination of dimension scores in 0..1. The single number to gate on.'),\n dimensions: z\n .record(z.string(), z.number())\n .describe('Per-dimension score, keyed by RubricDimension.id.'),\n failureModes: z\n .array(z.string())\n .default([])\n .describe('Failure-mode ids detected in the content (subset of rubric.failureModes ids).'),\n wins: z\n .array(z.string())\n .default([])\n .describe('Win ids detected in the content (subset of rubric.wins ids).'),\n rationale: z\n .string()\n .describe('Plain-English explanation of the score. Surfaced to the human reviewer.'),\n rubricVersion: z\n .string()\n .describe(\n 'Stable hash of the rubric used. Scores are only comparable across runs when this matches.',\n ),\n model: z.string().describe('Model that produced the judgement, for reproducibility.'),\n durationMs: z.number().int().nonnegative().describe('End-to-end wall time for this call.'),\n })\n .openapi('JudgeResult')\n\n// ── Rubric listing ──────────────────────────────────────────────────\n\nexport const RubricInfoSchema = z\n .object({\n name: z.string().describe('Pass this to /v1/judge as `rubricName`.'),\n description: z.string().describe('What this rubric measures.'),\n dimensions: z\n .array(z.object({ id: z.string(), description: z.string(), weight: z.number() }))\n .describe('The scoring axes this rubric uses, with weights.'),\n failureModes: z.array(z.string()).default([]).describe('Failure-mode ids this rubric detects.'),\n rubricVersion: z.string().describe('Stable hash — match this to compare scores across runs.'),\n })\n .openapi('RubricInfo')\n\nexport const ListRubricsResponseSchema = z\n .object({\n rubrics: z.array(RubricInfoSchema),\n })\n .openapi('ListRubricsResponse')\n\n// ── Version / health ────────────────────────────────────────────────\n\nexport const VersionResponseSchema = z\n .object({\n package: z.string().describe('Package name (always \"@tangle-network/agent-eval\").'),\n version: z.string().describe('Semver of the running server. Match your client to this.'),\n wireVersion: z\n .string()\n .describe(\n 'Wire-protocol semver. Bumps separately from package version when the schema changes.',\n ),\n apiSurface: z.array(z.string()).describe('List of supported method names.'),\n })\n .openapi('VersionResponse')\n\nexport const HealthResponseSchema = z\n .object({\n status: z.literal('ok'),\n uptimeSec: z.number(),\n })\n .openapi('HealthResponse')\n\n// ── Ingestion: production traces + user feedback ────────────────────\n\n/**\n * Minimal `TraceEvent` shape that the production runtime emits.\n * Matches `trace/schema.ts` `TraceEvent` but is duplicated here as a\n * wire schema so non-TypeScript clients can validate without depending\n * on internal types.\n */\nexport const TraceEventSchema = z\n .object({\n eventId: z.string().min(1).describe('Stable id for the event. Use ULID or UUID.'),\n runId: z.string().min(1).describe('Run this event belongs to.'),\n spanId: z.string().optional().describe('Span that emitted the event, if any.'),\n kind: z\n .enum([\n 'log',\n 'error',\n 'budget_decrement',\n 'budget_breach',\n 'state_mutation',\n 'policy_violation',\n 'redaction_applied',\n 'custom',\n ])\n .describe('Coarse event category — matches the TraceSchema v1 EventKind enum.'),\n timestamp: z\n .number()\n .int()\n .nonnegative()\n .describe('Unix millis. Must be monotonically non-decreasing within a span.'),\n payload: z\n .record(z.string(), z.unknown())\n .describe('Free-form payload — the runtime owns the shape.'),\n })\n .openapi('TraceEvent')\n\nexport const TracesIngestRequestSchema = z\n .object({\n events: z\n .array(TraceEventSchema)\n .min(1)\n .max(10_000)\n .describe('Batch of events. Max 10k per call — bigger streams should be chunked.'),\n })\n .openapi('TracesIngestRequest')\n\nexport const TracesIngestResponseSchema = z\n .object({\n accepted: z.number().int().nonnegative().describe('Number of events persisted.'),\n rejected: z\n .number()\n .int()\n .nonnegative()\n .describe('Number of events the store refused — see `errors[]` for reasons.'),\n errors: z\n .array(\n z.object({\n eventId: z.string().describe('Event id this error applies to.'),\n message: z.string().describe('Why the event was rejected.'),\n }),\n )\n .default([]),\n })\n .openapi('TracesIngestResponse')\n\nexport const FeedbackLabelSchema = z\n .object({\n id: z.string().optional(),\n source: z.enum(['user', 'judge', 'environment', 'metric', 'policy', 'system']),\n kind: z.enum([\n 'approve',\n 'reject',\n 'select',\n 'edit',\n 'rank',\n 'rate',\n 'comment',\n 'metric_outcome',\n 'policy_block',\n 'revision_request',\n ]),\n value: z.unknown(),\n reason: z.string().optional(),\n severity: z.enum(['info', 'warning', 'error', 'critical']).optional(),\n createdAt: z.string().describe('ISO-8601 UTC.'),\n metadata: z.record(z.string(), z.unknown()).optional(),\n })\n .openapi('FeedbackLabel')\n\nexport const FeedbackAttemptSchema = z\n .object({\n id: z.string().min(1),\n stepIndex: z.number().int().nonnegative(),\n artifactType: z.enum([\n 'text',\n 'code',\n 'plan',\n 'research',\n 'action',\n 'ui',\n 'decision',\n 'data',\n 'other',\n ]),\n artifact: z.unknown(),\n options: z.array(z.unknown()).optional(),\n proposedAction: z\n .object({\n type: z.string(),\n risk: z.enum(['low', 'medium', 'high']).optional(),\n costUsd: z.number().optional(),\n externalSideEffect: z.boolean().optional(),\n requiresApproval: z.boolean().optional(),\n metadata: z.record(z.string(), z.unknown()).optional(),\n })\n .optional(),\n feedback: z.array(FeedbackLabelSchema).optional(),\n createdAt: z.string(),\n metadata: z.record(z.string(), z.unknown()).optional(),\n })\n .openapi('FeedbackAttempt')\n\nexport const FeedbackTrajectorySchema = z\n .object({\n id: z.string().min(1).describe('Stable id; idempotency key for the trajectory.'),\n projectId: z.string().optional(),\n scenarioId: z.string().optional(),\n task: z.object({\n intent: z.string().min(1),\n context: z.unknown().optional(),\n }),\n attempts: z.array(FeedbackAttemptSchema).default([]),\n labels: z.array(FeedbackLabelSchema).default([]),\n outcome: z\n .object({\n success: z.boolean().optional(),\n score: z.number().optional(),\n metrics: z.record(z.string(), z.number()).optional(),\n costUsd: z.number().optional(),\n detail: z.string().optional(),\n observedAt: z.string().optional(),\n metadata: z.record(z.string(), z.unknown()).optional(),\n })\n .optional(),\n split: z.enum(['train', 'dev', 'test', 'holdout']).optional(),\n tags: z.record(z.string(), z.string()).optional(),\n createdAt: z.string().describe('ISO-8601 UTC.'),\n updatedAt: z.string().optional(),\n metadata: z.record(z.string(), z.unknown()).optional(),\n })\n .openapi('FeedbackTrajectory')\n\nexport const FeedbackIngestResponseSchema = z\n .object({\n id: z.string().describe('Trajectory id that was persisted.'),\n persisted: z.boolean().describe('True when the trajectory was saved (idempotent on id).'),\n })\n .openapi('FeedbackIngestResponse')\n\nexport type TraceEvent = z.infer<typeof TraceEventSchema>\nexport type TracesIngestRequest = z.infer<typeof TracesIngestRequestSchema>\nexport type TracesIngestResponse = z.infer<typeof TracesIngestResponseSchema>\nexport type FeedbackTrajectory = z.infer<typeof FeedbackTrajectorySchema>\nexport type FeedbackIngestResponse = z.infer<typeof FeedbackIngestResponseSchema>\n\n// ── Errors ──────────────────────────────────────────────────────────\n\nexport const ErrorResponseSchema = z\n .object({\n error: z\n .object({\n code: z\n .string()\n .describe(\n 'Machine-readable code: \"validation_error\", \"rubric_not_found\", \"judge_error\".',\n ),\n message: z.string().describe('Human-readable message.'),\n details: z.unknown().optional().describe('Optional structured detail.'),\n })\n .describe('Errors are always wrapped in this shape across all endpoints.'),\n })\n .openapi('ErrorResponse')\n\n// ── Type exports for callers in the same package ────────────────────\n\nexport type RubricDimension = z.infer<typeof RubricDimensionSchema>\nexport type FailureMode = z.infer<typeof FailureModeSchema>\nexport type Rubric = z.infer<typeof RubricSchema>\nexport type JudgeRequest = z.infer<typeof JudgeRequestSchema>\nexport type JudgeResult = z.infer<typeof JudgeResultSchema>\nexport type RubricInfo = z.infer<typeof RubricInfoSchema>\nexport type ListRubricsResponse = z.infer<typeof ListRubricsResponseSchema>\nexport type VersionResponse = z.infer<typeof VersionResponseSchema>\nexport type ErrorResponse = z.infer<typeof ErrorResponseSchema>\n\n// ── Wire-protocol version ───────────────────────────────────────────\n\n/**\n * Bump on any breaking change to a request/response schema.\n * Non-breaking (additive) changes don't require a bump.\n */\nexport const WIRE_VERSION = '1.0.0'\n\n/**\n * Stable hash of a rubric. Used to make scores comparable across runs:\n * if the rubricVersion matches, the rubric was identical.\n */\nexport function hashRubric(rubric: Rubric): string {\n const stable = stableStringify(rubric)\n let h = 5381\n for (let i = 0; i < stable.length; i++) {\n h = (h * 33) ^ stable.charCodeAt(i)\n }\n // Unsigned 32-bit hex, prefixed with rubric name + version slot\n return `${rubric.name}@${(h >>> 0).toString(16).padStart(8, '0')}`\n}\n\nfunction stableStringify(value: unknown): string {\n if (Array.isArray(value)) return `[${value.map((item) => stableStringify(item)).join(',')}]`\n if (value && typeof value === 'object') {\n const entries = Object.entries(value as Record<string, unknown>)\n .sort(([a], [b]) => a.localeCompare(b))\n .map(([key, item]) => `${JSON.stringify(key)}:${stableStringify(item)}`)\n return `{${entries.join(',')}}`\n }\n return JSON.stringify(value)\n}\n","/**\n * Built-in rubrics shipped with agent-eval.\n *\n * A rubric is a set of scoring axes plus a system prompt that tells the\n * judging LLM how to grade against those axes. Built-in rubrics are\n * curated for use cases that recur across Tangle projects — call them\n * by name from any client.\n *\n * Adding a rubric:\n * 1. Define the Rubric object below with a clear `description` and\n * named `dimensions`.\n * 2. Register it in `BUILTIN_RUBRICS` at the bottom.\n * 3. Add a test in `tests/wire/rubrics.test.ts`.\n *\n * Custom rubrics: callers pass `rubric` inline to /v1/judge instead of\n * `rubricName` — see schemas.ts.\n */\nimport type { Rubric } from './schemas'\nimport { hashRubric } from './schemas'\n\n// ── anti-slop ───────────────────────────────────────────────────────\n// Voice/style judge tuned for technical-buyer audiences. Used by the\n// Postiz autoresearch loop and any content-quality gate.\n\nconst ANTI_SLOP: Rubric = {\n name: 'anti-slop',\n description:\n 'Voice and signal quality for content aimed at senior engineers. Catches AI cadence, marketing tone, and engagement-bait shapes.',\n systemPrompt: `You are evaluating a piece of content written for senior engineers and technical founders.\n\nYou score three things:\n- buyer_quality (0..1): would a senior engineer in the target ICP find this worth their attention? High = specific, earned, technically interesting. Low = generic, hyped, off-target.\n- voice (0..1): does it read like a person who built the thing, or like AI/marketing copy?\n- signal (0..1): does it contain a non-obvious detail, constraint, or claim a reader couldn't get from the public docs?\n\nDetect failure modes (return ids matching):\n- ai-cadence: rule-of-three openings, em-dash flourish, \"Let me explain\", \"Here's the thing\", AI rhythm\n- marketing-tone: \"We're excited to announce\", \"thrilled\", \"delighted\", \"game-changer\", buzzword stack\n- vague-claim: technical claim without a specific component, file, or measurement\n- no-hook: opening doesn't earn attention from the target reader\n- engagement-bait: \"agree?\", \"thoughts?\", listicles, controversy-fishing, hook-detail-pitch\n- off-icp: content shape would attract motivational/grift/hype audiences instead of buyers\n- stale-claim: repeats a positioning line we've used many times this month\n\nDetect wins (return ids matching):\n- specific-component: names a real file, component, or measurement\n- earned-detail: shares a non-obvious detail not derivable from public docs\n- constraint-articulated: names a real tradeoff and the side chosen\n- honest-failure: describes a real failure mode and what was done about it\n\nReturn ONLY JSON matching the response schema. Be conservative — most content has 0-1 wins and 1-2 failure modes, not many of each.`,\n dimensions: [\n {\n id: 'buyer_quality',\n description: 'Would the target buyer find this worth attention?',\n weight: 0.5,\n min: 0,\n max: 1,\n },\n {\n id: 'voice',\n description: 'Does it sound like a builder, not AI or marketing?',\n weight: 0.3,\n min: 0,\n max: 1,\n },\n {\n id: 'signal',\n description: 'Non-obvious detail, constraint, or claim?',\n weight: 0.2,\n min: 0,\n max: 1,\n },\n ],\n failureModes: [\n { id: 'ai-cadence', description: 'AI-rhythm openings and transitions' },\n { id: 'marketing-tone', description: 'Buzzwords, hype, corporate-PR voice' },\n { id: 'vague-claim', description: 'Technical claim without specifics' },\n { id: 'no-hook', description: 'Opening fails to earn attention' },\n { id: 'engagement-bait', description: 'Listicle/controversy/agree-pattern' },\n { id: 'off-icp', description: 'Voice attracts the wrong audience' },\n { id: 'stale-claim', description: 'Reuses an over-used positioning line' },\n ],\n wins: [\n { id: 'specific-component', description: 'Names a real file/component/number' },\n { id: 'earned-detail', description: 'Detail not in public docs' },\n { id: 'constraint-articulated', description: 'Names a real tradeoff' },\n { id: 'honest-failure', description: 'Describes a real failure honestly' },\n ],\n}\n\n// ── Registry ────────────────────────────────────────────────────────\n\nexport const BUILTIN_RUBRICS: Record<string, Rubric> = {\n 'anti-slop': ANTI_SLOP,\n}\n\n/** Get a built-in rubric by name, or undefined. */\nexport function getBuiltinRubric(name: string): Rubric | undefined {\n return BUILTIN_RUBRICS[name]\n}\n\n/** List built-in rubrics with their stable versions. */\nexport function listBuiltinRubrics() {\n return Object.values(BUILTIN_RUBRICS).map((r) => ({\n name: r.name,\n description: r.description,\n dimensions: r.dimensions.map((d) => ({\n id: d.id,\n description: d.description,\n weight: d.weight,\n })),\n failureModes: r.failureModes.map((f) => f.id),\n rubricVersion: hashRubric(r),\n }))\n}\n","/**\n * Pure handler functions — the \"business logic\" behind every wire-protocol\n * method. The HTTP server (`server.ts`) and the stdio RPC (`rpc.ts`) both\n * call these. Tests call these directly without spinning a server.\n *\n * Each handler:\n * - Takes a parsed request (already Zod-validated by the transport).\n * - Returns a result that matches the response schema.\n * - Throws `WireError` for caller-fixable errors (404, 400, 422).\n * - Lets unexpected errors bubble — the transport maps them to 500.\n */\nimport type { FeedbackTrajectoryStore } from '../feedback-trajectory'\nimport { callLlmJson } from '../llm-client'\nimport type { TraceEvent as InternalTraceEvent } from '../trace/schema'\nimport type { TraceStore } from '../trace/store'\nimport { getBuiltinRubric, listBuiltinRubrics } from './rubrics'\nimport {\n type FeedbackIngestResponse,\n hashRubric,\n type JudgeRequest,\n type JudgeResult,\n type ListRubricsResponse,\n type Rubric,\n type TracesIngestRequest,\n type TracesIngestResponse,\n type VersionResponse,\n WIRE_VERSION,\n type FeedbackTrajectory as WireFeedbackTrajectory,\n} from './schemas'\n\n/** Caller-fixable error. The transport renders this to 4xx + ErrorResponse. */\nexport class WireError extends Error {\n constructor(\n public readonly code: string,\n message: string,\n public readonly status: number = 400,\n public readonly details?: unknown,\n ) {\n super(message)\n this.name = 'WireError'\n }\n}\n\n// ── judge ───────────────────────────────────────────────────────────\n\n/** The JSON schema we ask the judging LLM to fill in. */\nfunction judgeOutputSchema(rubric: Rubric) {\n return {\n name: 'JudgeOutput',\n schema: {\n type: 'object',\n additionalProperties: false,\n properties: {\n dimensions: {\n type: 'object',\n additionalProperties: false,\n properties: Object.fromEntries(\n rubric.dimensions.map((d) => [\n d.id,\n { type: 'number', minimum: d.min, maximum: d.max },\n ]),\n ),\n required: rubric.dimensions.map((d) => d.id),\n },\n failureModes: {\n type: 'array',\n items: { type: 'string', enum: rubric.failureModes.map((f) => f.id) },\n },\n wins: {\n type: 'array',\n items: { type: 'string', enum: rubric.wins.map((w) => w.id) },\n },\n rationale: { type: 'string' },\n },\n required: ['dimensions', 'rationale'],\n } as Record<string, unknown>,\n }\n}\n\ninterface JudgeOutput {\n dimensions: Record<string, number>\n failureModes?: string[]\n wins?: string[]\n rationale: string\n}\n\nfunction validateJudgeOutput(value: unknown, rubric: Rubric): JudgeOutput {\n if (!value || typeof value !== 'object') {\n throw new WireError('judge_error', 'Judge returned malformed output.', 500, value)\n }\n const raw = value as Record<string, unknown>\n const rawDimensions = raw.dimensions\n if (!rawDimensions || typeof rawDimensions !== 'object' || Array.isArray(rawDimensions)) {\n throw new WireError('judge_error', 'Judge returned malformed dimensions.', 500, value)\n }\n\n const dimensions: Record<string, number> = {}\n const dimensionRecord = rawDimensions as Record<string, unknown>\n for (const dim of rubric.dimensions) {\n const score = dimensionRecord[dim.id]\n if (\n typeof score !== 'number' ||\n !Number.isFinite(score) ||\n score < dim.min ||\n score > dim.max\n ) {\n throw new WireError(\n 'judge_error',\n `Judge returned invalid score for dimension \"${dim.id}\".`,\n 500,\n value,\n )\n }\n dimensions[dim.id] = score\n }\n\n const allowedFailures = new Set(rubric.failureModes.map((mode) => mode.id))\n const allowedWins = new Set(rubric.wins.map((win) => win.id))\n const failureModes = validateIdArray(raw.failureModes, allowedFailures, 'failureModes', value)\n const wins = validateIdArray(raw.wins, allowedWins, 'wins', value)\n if (typeof raw.rationale !== 'string' || raw.rationale.trim().length === 0) {\n throw new WireError('judge_error', 'Judge returned missing rationale.', 500, value)\n }\n\n return { dimensions, failureModes, wins, rationale: raw.rationale }\n}\n\nfunction validateIdArray(\n raw: unknown,\n allowed: Set<string>,\n field: 'failureModes' | 'wins',\n original: unknown,\n): string[] {\n if (raw === undefined) return []\n if (!Array.isArray(raw)) {\n throw new WireError('judge_error', `Judge returned non-array ${field}.`, 500, original)\n }\n const out: string[] = []\n for (const item of raw) {\n if (typeof item !== 'string' || !allowed.has(item)) {\n throw new WireError(\n 'judge_error',\n `Judge returned unknown ${field} id \"${String(item)}\".`,\n 500,\n original,\n )\n }\n out.push(item)\n }\n return out\n}\n\nfunction compositeScore(dimensions: Record<string, number>, rubric: Rubric): number {\n let weighted = 0\n let totalWeight = 0\n for (const dim of rubric.dimensions) {\n const raw = dimensions[dim.id] ?? 0\n const range = dim.max - dim.min || 1\n const normalized = Math.max(0, Math.min(1, (raw - dim.min) / range))\n weighted += normalized * dim.weight\n totalWeight += dim.weight\n }\n return totalWeight > 0 ? weighted / totalWeight : 0\n}\n\nfunction buildJudgePrompt(content: string, context: unknown): string {\n const ctx = context && Object.keys(context as object).length ? JSON.stringify(context) : ''\n return [\n `CONTENT TO JUDGE:`,\n content,\n '',\n ctx ? `CONTEXT (metadata, analytics, etc.):` : '',\n ctx ? ctx : '',\n ]\n .filter(Boolean)\n .join('\\n')\n}\n\nconst DEFAULT_JUDGE_MODEL = 'claude-sonnet-4-6'\n\nexport async function handleJudge(req: JudgeRequest): Promise<JudgeResult> {\n // Resolve rubric\n let rubric: Rubric\n if (req.rubricName) {\n const found = getBuiltinRubric(req.rubricName)\n if (!found) {\n throw new WireError('rubric_not_found', `No built-in rubric named \"${req.rubricName}\".`, 404)\n }\n rubric = found\n } else if (req.rubric) {\n rubric = req.rubric\n } else {\n // refine() in the schema should already have caught this — defense in depth\n throw new WireError('validation_error', 'Provide either `rubricName` or `rubric`.', 422)\n }\n\n const startedAt = Date.now()\n const model = req.model ?? DEFAULT_JUDGE_MODEL\n\n const { value, result } = await callLlmJson<JudgeOutput>({\n model,\n messages: [\n { role: 'system', content: rubric.systemPrompt },\n { role: 'user', content: buildJudgePrompt(req.content, req.context) },\n ],\n jsonSchema: judgeOutputSchema(rubric),\n temperature: 0.0,\n timeoutMs: 60_000,\n })\n\n const output = validateJudgeOutput(value, rubric)\n\n const composite = compositeScore(output.dimensions, rubric)\n const durationMs = Date.now() - startedAt\n\n return {\n composite,\n dimensions: output.dimensions,\n failureModes: output.failureModes ?? [],\n wins: output.wins ?? [],\n rationale: output.rationale,\n rubricVersion: hashRubric(rubric),\n model: result.model,\n durationMs,\n }\n}\n\n// ── listRubrics ─────────────────────────────────────────────────────\n\nexport function handleListRubrics(): ListRubricsResponse {\n return { rubrics: listBuiltinRubrics() }\n}\n\n// ── version ─────────────────────────────────────────────────────────\n\nimport { readFileSync } from 'node:fs'\nimport { dirname, resolve } from 'node:path'\nimport { fileURLToPath } from 'node:url'\n\nlet CACHED_VERSION: string | undefined\n\nfunction readPackageVersion(): string {\n if (CACHED_VERSION) return CACHED_VERSION\n // Walk up from this file looking for the nearest package.json.\n // In dist/ this is dist/.., in src/wire/ this is ../../package.json.\n const here = dirname(fileURLToPath(import.meta.url))\n const candidates = [\n resolve(here, '..', '..', 'package.json'), // src/wire → repo root\n resolve(here, '..', 'package.json'), // dist → repo root\n ]\n for (const path of candidates) {\n try {\n const pkg = JSON.parse(readFileSync(path, 'utf-8')) as { version?: string }\n if (pkg.version) {\n CACHED_VERSION = pkg.version\n return pkg.version\n }\n } catch {\n // try next\n }\n }\n return '0.0.0-unknown'\n}\n\nexport function handleVersion(): VersionResponse {\n return {\n package: '@tangle-network/agent-eval',\n version: readPackageVersion(),\n wireVersion: WIRE_VERSION,\n apiSurface: ['judge', 'listRubrics', 'version', 'feedback.ingest', 'traces.ingest'],\n }\n}\n\n// ── Ingestion handlers ───────────────────────────────────────────────\n\n/**\n * Pluggable stores the wire layer routes ingestion writes into. Both\n * are optional — when omitted, the corresponding endpoint returns 503.\n *\n * Production deployments wire a `FileSystemTraceStore` and\n * `FileSystemFeedbackTrajectoryStore` here. Tests substitute in-memory\n * stores.\n */\nexport interface IngestionStores {\n traceStore?: TraceStore\n feedbackStore?: FeedbackTrajectoryStore\n}\n\n/**\n * `POST /v1/traces/ingest` — accept a batch of `TraceEvent`s from the\n * production runtime. Best-effort: each event is appended independently;\n * one bad event does not poison the batch.\n *\n * Idempotency: the underlying store is append-only; consumers retrying\n * the same payload will get duplicate events. Consumers should\n * de-duplicate by `eventId` downstream — production traces frequently\n * land via at-least-once buses (Kafka, SQS) where dedup is unavoidable.\n */\nexport async function handleTracesIngest(\n req: TracesIngestRequest,\n stores: IngestionStores,\n): Promise<TracesIngestResponse> {\n if (!stores.traceStore) {\n throw new WireError(\n 'service_unavailable',\n 'No trace store configured on this server. Pass `traceStore` to `createApp`.',\n 503,\n )\n }\n const errors: Array<{ eventId: string; message: string }> = []\n let accepted = 0\n for (const event of req.events) {\n try {\n // The wire `TraceEvent` is structurally identical to the internal one.\n await stores.traceStore.appendEvent(event as InternalTraceEvent)\n accepted++\n } catch (err) {\n errors.push({\n eventId: event.eventId,\n message: err instanceof Error ? err.message : String(err),\n })\n }\n }\n return { accepted, rejected: errors.length, errors }\n}\n\n/**\n * `POST /v1/feedback` — accept a single `FeedbackTrajectory` from the\n * production runtime. Idempotent on `id`: re-posting the same trajectory\n * replaces the prior record.\n */\nexport async function handleFeedbackIngest(\n req: WireFeedbackTrajectory,\n stores: IngestionStores,\n): Promise<FeedbackIngestResponse> {\n if (!stores.feedbackStore) {\n throw new WireError(\n 'service_unavailable',\n 'No feedback store configured on this server. Pass `feedbackStore` to `createApp`.',\n 503,\n )\n }\n // The wire `FeedbackTrajectory` aligns 1:1 with the internal type;\n // cast through `unknown` since the wire schema is a Zod-inferred\n // structural type with optional fields the internal store consumes.\n await stores.feedbackStore.save(req as unknown as Parameters<FeedbackTrajectoryStore['save']>[0])\n return { id: req.id, persisted: true }\n}\n","/**\n * Build an OpenAPI spec from the wire schemas.\n *\n * The spec is the contract that other-language clients (Python, Rust,\n * Go, …) generate from. There is no hand-written client — clients are\n * derived artifacts of this file plus `schemas.ts`.\n *\n * Run `pnpm openapi` (defined in package.json) to write the spec to\n * `dist/openapi.json`. CI uses that file to regenerate the Python\n * client and gate the dual-publish workflow.\n */\nimport { OpenAPIRegistry, OpenApiGeneratorV31 } from '@asteasolutions/zod-to-openapi'\nimport type { OpenAPIObject } from 'openapi3-ts/oas31'\n\nimport {\n ErrorResponseSchema,\n FeedbackIngestResponseSchema,\n FeedbackTrajectorySchema,\n HealthResponseSchema,\n JudgeRequestSchema,\n JudgeResultSchema,\n ListRubricsResponseSchema,\n TracesIngestRequestSchema,\n TracesIngestResponseSchema,\n VersionResponseSchema,\n WIRE_VERSION,\n} from './schemas'\n\nexport function buildOpenApi(packageVersion: string): OpenAPIObject {\n const registry = new OpenAPIRegistry()\n\n // Components — each schema becomes a $ref-able component\n registry.register('JudgeRequest', JudgeRequestSchema)\n registry.register('JudgeResult', JudgeResultSchema)\n registry.register('ListRubricsResponse', ListRubricsResponseSchema)\n registry.register('VersionResponse', VersionResponseSchema)\n registry.register('HealthResponse', HealthResponseSchema)\n registry.register('ErrorResponse', ErrorResponseSchema)\n registry.register('TracesIngestRequest', TracesIngestRequestSchema)\n registry.register('TracesIngestResponse', TracesIngestResponseSchema)\n registry.register('FeedbackTrajectory', FeedbackTrajectorySchema)\n registry.register('FeedbackIngestResponse', FeedbackIngestResponseSchema)\n\n // Routes\n registry.registerPath({\n method: 'post',\n path: '/v1/judge',\n summary: 'Score a piece of content against a rubric',\n description:\n 'Runs the judging LLM with the named (or inline) rubric and returns dimension scores, detected failure modes, wins, and a composite score in 0..1.',\n request: {\n body: {\n content: {\n 'application/json': { schema: JudgeRequestSchema },\n },\n },\n },\n responses: {\n 200: {\n description: 'Successful judgement',\n content: { 'application/json': { schema: JudgeResultSchema } },\n },\n 400: {\n description: 'Validation error',\n content: { 'application/json': { schema: ErrorResponseSchema } },\n },\n 404: {\n description: 'Rubric not found',\n content: { 'application/json': { schema: ErrorResponseSchema } },\n },\n 500: {\n description: 'Judge error',\n content: { 'application/json': { schema: ErrorResponseSchema } },\n },\n },\n })\n\n registry.registerPath({\n method: 'get',\n path: '/v1/rubrics',\n summary: 'List built-in rubrics',\n description:\n 'Returns every rubric registered server-side, with their dimensions and stable rubricVersion hash.',\n responses: {\n 200: {\n description: 'Listing',\n content: { 'application/json': { schema: ListRubricsResponseSchema } },\n },\n },\n })\n\n registry.registerPath({\n method: 'get',\n path: '/v1/version',\n summary: 'Server and wire-protocol version',\n description: 'Match your client version to `version`; check `wireVersion` for compatibility.',\n responses: {\n 200: {\n description: 'Version info',\n content: { 'application/json': { schema: VersionResponseSchema } },\n },\n },\n })\n\n registry.registerPath({\n method: 'get',\n path: '/healthz',\n summary: 'Liveness check',\n responses: {\n 200: {\n description: 'OK',\n content: { 'application/json': { schema: HealthResponseSchema } },\n },\n },\n })\n\n registry.registerPath({\n method: 'post',\n path: '/v1/traces/ingest',\n summary: 'Ingest a batch of production TraceEvents',\n description:\n 'Append a batch of TraceEvents to the configured TraceStore. Accepts application/json ({events:[...]}) or application/x-ndjson (one event per line). Returns counts of accepted + rejected events.',\n request: {\n body: {\n content: {\n 'application/json': { schema: TracesIngestRequestSchema },\n 'application/x-ndjson': { schema: TracesIngestRequestSchema },\n },\n },\n },\n responses: {\n 200: {\n description: 'Ingestion summary',\n content: { 'application/json': { schema: TracesIngestResponseSchema } },\n },\n 400: {\n description: 'Validation error',\n content: { 'application/json': { schema: ErrorResponseSchema } },\n },\n 401: {\n description: 'Unauthorized (when bearer auth is configured)',\n content: { 'application/json': { schema: ErrorResponseSchema } },\n },\n 503: {\n description: 'No trace store configured',\n content: { 'application/json': { schema: ErrorResponseSchema } },\n },\n },\n })\n\n registry.registerPath({\n method: 'post',\n path: '/v1/feedback',\n summary: 'Ingest a FeedbackTrajectory from production',\n description:\n 'Persist a single FeedbackTrajectory. Idempotent on trajectory.id — re-posting replaces the prior record. Used by production runtimes to forward user 👍/👎/edits into the eval substrate.',\n request: {\n body: {\n content: {\n 'application/json': { schema: FeedbackTrajectorySchema },\n },\n },\n },\n responses: {\n 200: {\n description: 'Persisted',\n content: { 'application/json': { schema: FeedbackIngestResponseSchema } },\n },\n 400: {\n description: 'Validation error',\n content: { 'application/json': { schema: ErrorResponseSchema } },\n },\n 401: {\n description: 'Unauthorized (when bearer auth is configured)',\n content: { 'application/json': { schema: ErrorResponseSchema } },\n },\n 503: {\n description: 'No feedback store configured',\n content: { 'application/json': { schema: ErrorResponseSchema } },\n },\n },\n })\n\n const generator = new OpenApiGeneratorV31(registry.definitions)\n const doc = generator.generateDocument({\n openapi: '3.1.0',\n info: {\n title: '@tangle-network/agent-eval — wire protocol',\n version: packageVersion,\n description: `HTTP and stdio RPC interface to agent-eval. The TypeScript runtime is the source of truth; this spec is the contract that cross-language clients (Python, Rust, Go) generate from.\n\nWire-protocol version: ${WIRE_VERSION}. Bumps on breaking changes to request/response schemas.`,\n contact: { name: 'Tangle Network', url: 'https://github.com/tangle-network/agent-eval' },\n license: { name: 'MIT' },\n },\n servers: [{ url: 'http://localhost:5005', description: 'Local agent-eval serve' }],\n })\n const rubricRef = { $ref: '#/components/schemas/Rubric' } as const\n const commonJudgeFields = {\n content: { type: 'string', minLength: 1 },\n context: { type: 'object', additionalProperties: true },\n model: { type: 'string' },\n } as const\n doc.components ??= {}\n doc.components.schemas ??= {}\n doc.components.schemas.JudgeRequest = {\n oneOf: [\n {\n type: 'object',\n additionalProperties: false,\n required: ['rubricName', 'content'],\n properties: {\n rubricName: { type: 'string', minLength: 1 },\n ...commonJudgeFields,\n },\n },\n {\n type: 'object',\n additionalProperties: false,\n required: ['rubric', 'content'],\n properties: {\n rubric: rubricRef,\n ...commonJudgeFields,\n },\n },\n ],\n description: 'Judge request. Provide exactly one of rubricName or rubric.',\n }\n return doc\n}\n","/**\n * stdio RPC transport.\n *\n * For batch / cron use without a running server. The Python client falls\n * back to this when no server is reachable.\n *\n * Protocol (line-delimited JSON over stdin/stdout):\n * IN: one JSON object on stdin: {\"method\":\"judge\",\"params\":{...}}\n * OUT: one JSON object on stdout: {\"result\":{...}} or {\"error\":{...}}\n *\n * One request per process invocation. To pipeline many calls, the client\n * writes JSONL to stdin and reads JSONL from stdout — see batch mode below.\n */\nimport { handleJudge, handleListRubrics, handleVersion, WireError } from './handlers'\nimport { JudgeRequestSchema } from './schemas'\n\ninterface RpcRequest {\n method: 'judge' | 'listRubrics' | 'version'\n params?: unknown\n}\n\ninterface RpcSuccess {\n result: unknown\n}\n\ninterface RpcError {\n error: { code: string; message: string; details?: unknown }\n}\n\nexport async function dispatchRpc(req: RpcRequest): Promise<RpcSuccess | RpcError> {\n try {\n switch (req.method) {\n case 'judge': {\n const parsed = JudgeRequestSchema.safeParse(req.params)\n if (!parsed.success) {\n return {\n error: {\n code: 'validation_error',\n message: 'params did not match JudgeRequest schema.',\n details: parsed.error.issues,\n },\n }\n }\n return { result: await handleJudge(parsed.data) }\n }\n case 'listRubrics':\n return { result: handleListRubrics() }\n case 'version':\n return { result: handleVersion() }\n default:\n return {\n error: {\n code: 'unknown_method',\n message: `No such method: ${(req as { method: string }).method}`,\n },\n }\n }\n } catch (err) {\n if (err instanceof WireError) {\n return { error: { code: err.code, message: err.message, details: err.details } }\n }\n const message = err instanceof Error ? err.message : String(err)\n return { error: { code: 'internal_error', message } }\n }\n}\n\n// ── stdin/stdout driver ─────────────────────────────────────────────\n\nasync function readAll(stream: NodeJS.ReadableStream): Promise<string> {\n const chunks: Buffer[] = []\n for await (const chunk of stream) {\n chunks.push(Buffer.isBuffer(chunk) ? chunk : Buffer.from(chunk as string))\n }\n return Buffer.concat(chunks).toString('utf-8')\n}\n\n/** Read one JSON request from stdin, write one JSON response to stdout. */\nexport async function runRpcOnce(method?: string): Promise<number> {\n const raw = await readAll(process.stdin)\n let req: RpcRequest\n try {\n const body = JSON.parse(raw)\n req = method ? { method: method as RpcRequest['method'], params: body } : (body as RpcRequest)\n } catch (err) {\n process.stdout.write(\n `${JSON.stringify({\n error: {\n code: 'parse_error',\n message: `stdin was not valid JSON: ${err instanceof Error ? err.message : String(err)}`,\n },\n })}\\n`,\n )\n return 1\n }\n const out = await dispatchRpc(req)\n process.stdout.write(`${JSON.stringify(out)}\\n`)\n return 'error' in out ? 1 : 0\n}\n\n/** Read JSONL requests from stdin, write JSONL responses to stdout. */\nexport async function runRpcBatch(method?: string): Promise<number> {\n const raw = await readAll(process.stdin)\n const lines = raw.split('\\n').filter((l) => l.trim().length > 0)\n let exitCode = 0\n for (const line of lines) {\n let req: RpcRequest\n try {\n const body = JSON.parse(line)\n req = method ? { method: method as RpcRequest['method'], params: body } : (body as RpcRequest)\n } catch (err) {\n process.stdout.write(\n `${JSON.stringify({\n error: {\n code: 'parse_error',\n message: `line was not valid JSON: ${err instanceof Error ? err.message : String(err)}`,\n },\n })}\\n`,\n )\n exitCode = 1\n continue\n }\n const out = await dispatchRpc(req)\n process.stdout.write(`${JSON.stringify(out)}\\n`)\n if ('error' in out) exitCode = 1\n }\n return exitCode\n}\n","/**\n * HTTP transport for the wire protocol.\n *\n * Hono + @hono/node-server. Every endpoint:\n * 1. Validates the request against its Zod schema.\n * 2. Calls the matching handler in `handlers.ts`.\n * 3. Renders 4xx for `WireError` with structured body, 500 for unexpected.\n *\n * The server holds optional `IngestionStores` (passed to `createApp`)\n * to receive production traces and user feedback. With no stores wired,\n * the ingestion endpoints return 503 — read endpoints (`/v1/judge`,\n * `/v1/rubrics`, `/v1/version`) remain fully functional.\n *\n * Run via `agent-eval serve --port 5005`.\n */\nimport { type ServerType, serve } from '@hono/node-server'\nimport { Hono } from 'hono'\nimport { cors } from 'hono/cors'\n\nimport {\n handleFeedbackIngest,\n handleJudge,\n handleListRubrics,\n handleTracesIngest,\n handleVersion,\n type IngestionStores,\n WireError,\n} from './handlers'\nimport { buildOpenApi } from './openapi'\nimport { FeedbackTrajectorySchema, JudgeRequestSchema, TracesIngestRequestSchema } from './schemas'\n\nconst STARTED_AT = Date.now()\n\nexport interface CreateAppOptions {\n /** Stores wired to the ingestion endpoints. */\n stores?: IngestionStores\n /**\n * Bearer-token auth. When provided, every endpoint EXCEPT `/healthz`\n * and `/v1/version` requires `Authorization: Bearer <token>`. The\n * token may be a static string OR a function for time-bounded /\n * rotating tokens.\n *\n * Recommended for any server that accepts ingestion writes from the\n * public internet. Read-only deployments may omit it.\n */\n auth?: {\n bearer: string | ((token: string) => boolean | Promise<boolean>)\n }\n}\n\nconst AUTH_EXEMPT_PATHS = new Set(['/healthz', '/v1/version', '/openapi.json'])\n\nexport function createApp(opts: CreateAppOptions = {}) {\n const app = new Hono()\n\n app.use('*', cors())\n\n // Bearer-token middleware (only attached when configured).\n if (opts.auth) {\n const verify = opts.auth.bearer\n app.use('*', async (c, next) => {\n const path = new URL(c.req.url).pathname\n if (AUTH_EXEMPT_PATHS.has(path)) return next()\n const raw = c.req.header('authorization') ?? ''\n const match = raw.match(/^Bearer\\s+(.+)$/i)\n if (!match) {\n throw new WireError('unauthorized', 'Missing or malformed Authorization header.', 401)\n }\n const token = match[1] as string\n const ok = typeof verify === 'string' ? token === verify : await verify(token)\n if (!ok) {\n throw new WireError('unauthorized', 'Invalid bearer token.', 401)\n }\n return next()\n })\n }\n\n app.onError((err, c) => {\n if (err instanceof WireError) {\n const status = err.status as 400 | 401 | 404 | 422 | 500 | 503\n return c.json(\n { error: { code: err.code, message: err.message, details: err.details } },\n status,\n )\n }\n // Unexpected — log and return generic 500 without leaking internals.\n console.error('[agent-eval] unhandled error:', err)\n return c.json({ error: { code: 'internal_error', message: 'Internal server error.' } }, 500)\n })\n\n // ── Health ──\n app.get('/healthz', (c) =>\n c.json({ status: 'ok' as const, uptimeSec: (Date.now() - STARTED_AT) / 1000 }),\n )\n\n // ── Version ──\n app.get('/v1/version', (c) => c.json(handleVersion()))\n\n // ── Rubrics ──\n app.get('/v1/rubrics', (c) => c.json(handleListRubrics()))\n\n // ── Judge ──\n app.post('/v1/judge', async (c) => {\n const raw = await c.req.json().catch(() => null)\n if (raw == null) {\n throw new WireError('validation_error', 'Request body must be JSON.', 400)\n }\n const parsed = JudgeRequestSchema.safeParse(raw)\n if (!parsed.success) {\n throw new WireError(\n 'validation_error',\n 'Request did not match JudgeRequest schema.',\n 400,\n parsed.error.issues,\n )\n }\n const result = await handleJudge(parsed.data)\n return c.json(result)\n })\n\n // ── Traces ingest (NDJSON-friendly: accepts either {events:[...]} or NDJSON) ──\n app.post('/v1/traces/ingest', async (c) => {\n const contentType = c.req.header('content-type') ?? ''\n let payload: unknown\n if (contentType.includes('application/x-ndjson')) {\n const text = await c.req.text()\n const events = text\n .split('\\n')\n .map((line) => line.trim())\n .filter((line) => line.length > 0)\n .map((line) => {\n try {\n return JSON.parse(line)\n } catch {\n throw new WireError(\n 'validation_error',\n 'NDJSON line did not parse as JSON.',\n 400,\n line.slice(0, 200),\n )\n }\n })\n payload = { events }\n } else {\n payload = await c.req.json().catch(() => null)\n }\n if (payload == null) {\n throw new WireError('validation_error', 'Request body must be JSON or NDJSON.', 400)\n }\n const parsed = TracesIngestRequestSchema.safeParse(payload)\n if (!parsed.success) {\n throw new WireError(\n 'validation_error',\n 'Request did not match TracesIngestRequest schema.',\n 400,\n parsed.error.issues,\n )\n }\n const result = await handleTracesIngest(parsed.data, opts.stores ?? {})\n return c.json(result)\n })\n\n // ── Feedback ingest ──\n app.post('/v1/feedback', async (c) => {\n const raw = await c.req.json().catch(() => null)\n if (raw == null) {\n throw new WireError('validation_error', 'Request body must be JSON.', 400)\n }\n const parsed = FeedbackTrajectorySchema.safeParse(raw)\n if (!parsed.success) {\n throw new WireError(\n 'validation_error',\n 'Request did not match FeedbackTrajectory schema.',\n 400,\n parsed.error.issues,\n )\n }\n const result = await handleFeedbackIngest(parsed.data, opts.stores ?? {})\n return c.json(result)\n })\n\n // ── OpenAPI spec ──\n app.get('/openapi.json', (c) => c.json(buildOpenApi(handleVersion().version)))\n\n return app\n}\n\nexport interface ServeOptions extends CreateAppOptions {\n /** Default 5005. */\n port?: number\n /** Default '127.0.0.1'. Set to '0.0.0.0' to listen on all interfaces. */\n host?: string\n}\n\nexport function startServer(opts: ServeOptions = {}): ServerType {\n const app = createApp(opts)\n const port = opts.port ?? 5005\n const host = opts.host ?? '127.0.0.1'\n return serve({ fetch: app.fetch, port, hostname: host }, ({ address, port: actualPort }) => {\n // eslint-disable-next-line no-console\n console.log(`[agent-eval] serving on http://${address}:${actualPort}`)\n })\n}\n"],"mappings":";;;;;AAYA,SAAS,4BAA4B;AACrC,SAAS,SAAS;AAElB,qBAAqB,CAAC;AAIf,IAAM,wBAAwB,EAClC,OAAO;AAAA,EACN,IAAI,EACD,OAAO,EACP,IAAI,CAAC,EACL,SAAS,gFAA2E;AAAA,EACvF,aAAa,EACV,OAAO,EACP,IAAI,CAAC,EACL,SAAS,sEAAsE;AAAA,EAClF,QAAQ,EACL,OAAO,EACP,IAAI,CAAC,EACL,QAAQ,CAAC,EACT,SAAS,gEAAgE;AAAA,EAC5E,KAAK,EAAE,OAAO,EAAE,QAAQ,CAAC,EAAE,SAAS,gDAAgD;AAAA,EACpF,KAAK,EAAE,OAAO,EAAE,QAAQ,CAAC,EAAE,SAAS,gDAAgD;AACtF,CAAC,EACA,QAAQ,iBAAiB;AAErB,IAAM,oBAAoB,EAC9B,OAAO;AAAA,EACN,IAAI,EAAE,OAAO,EAAE,IAAI,CAAC,EAAE,SAAS,mEAA8D;AAAA,EAC7F,aAAa,EAAE,OAAO,EAAE,IAAI,CAAC,EAAE,SAAS,mDAAmD;AAC7F,CAAC,EACA,QAAQ,aAAa;AAIjB,IAAM,eAAe,EACzB,OAAO;AAAA,EACN,MAAM,EACH,OAAO,EACP,IAAI,CAAC,EACL,SAAS,4EAAuE;AAAA,EACnF,aAAa,EACV,OAAO,EACP,IAAI,CAAC,EACL,SAAS,0DAA0D;AAAA,EACtE,cAAc,EACX,OAAO,EACP,IAAI,CAAC,EACL;AAAA,IACC;AAAA,EACF;AAAA,EACF,YAAY,EACT,MAAM,qBAAqB,EAC3B,IAAI,CAAC,EACL,SAAS,+DAA+D;AAAA,EAC3E,cAAc,EACX,MAAM,iBAAiB,EACvB,QAAQ,CAAC,CAAC,EACV,SAAS,iFAAiF;AAAA,EAC7F,MAAM,EACH,MAAM,iBAAiB,EACvB,QAAQ,CAAC,CAAC,EACV,SAAS,uEAAuE;AACrF,CAAC,EACA,QAAQ,QAAQ;AAIZ,IAAM,qBAAqB,EAC/B,OAAO;AAAA,EACN,YAAY,EACT,OAAO,EACP,SAAS,EACT,SAAS,kEAAkE;AAAA,EAC9E,QAAQ,aAAa,SAAS,EAAE;AAAA,IAC9B;AAAA,EACF;AAAA,EACA,SAAS,EACN,OAAO,EACP,IAAI,CAAC,EACL,SAAS,uFAAkF;AAAA,EAC9F,SAAS,EACN,OAAO,EAAE,OAAO,GAAG,EAAE,QAAQ,CAAC,EAC9B,SAAS,EACT;AAAA,IACC;AAAA,EACF;AAAA,EACF,OAAO,EACJ,OAAO,EACP,SAAS,EACT,SAAS,+EAA+E;AAC7F,CAAC,EACA,OAAO,CAAC,MAAM,QAAQ,EAAE,UAAU,MAAM,QAAQ,EAAE,MAAM,GAAG;AAAA,EAC1D,SAAS;AACX,CAAC,EACA,QAAQ,cAAc;AAElB,IAAM,oBAAoB,EAC9B,OAAO;AAAA,EACN,WAAW,EACR,OAAO,EACP,IAAI,CAAC,EACL,IAAI,CAAC,EACL,SAAS,iFAAiF;AAAA,EAC7F,YAAY,EACT,OAAO,EAAE,OAAO,GAAG,EAAE,OAAO,CAAC,EAC7B,SAAS,mDAAmD;AAAA,EAC/D,cAAc,EACX,MAAM,EAAE,OAAO,CAAC,EAChB,QAAQ,CAAC,CAAC,EACV,SAAS,+EAA+E;AAAA,EAC3F,MAAM,EACH,MAAM,EAAE,OAAO,CAAC,EAChB,QAAQ,CAAC,CAAC,EACV,SAAS,8DAA8D;AAAA,EAC1E,WAAW,EACR,OAAO,EACP,SAAS,yEAAyE;AAAA,EACrF,eAAe,EACZ,OAAO,EACP;AAAA,IACC;AAAA,EACF;AAAA,EACF,OAAO,EAAE,OAAO,EAAE,SAAS,yDAAyD;AAAA,EACpF,YAAY,EAAE,OAAO,EAAE,IAAI,EAAE,YAAY,EAAE,SAAS,qCAAqC;AAC3F,CAAC,EACA,QAAQ,aAAa;AAIjB,IAAM,mBAAmB,EAC7B,OAAO;AAAA,EACN,MAAM,EAAE,OAAO,EAAE,SAAS,yCAAyC;AAAA,EACnE,aAAa,EAAE,OAAO,EAAE,SAAS,4BAA4B;AAAA,EAC7D,YAAY,EACT,MAAM,EAAE,OAAO,EAAE,IAAI,EAAE,OAAO,GAAG,aAAa,EAAE,OAAO,GAAG,QAAQ,EAAE,OAAO,EAAE,CAAC,CAAC,EAC/E,SAAS,kDAAkD;AAAA,EAC9D,cAAc,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,QAAQ,CAAC,CAAC,EAAE,SAAS,uCAAuC;AAAA,EAC9F,eAAe,EAAE,OAAO,EAAE,SAAS,8DAAyD;AAC9F,CAAC,EACA,QAAQ,YAAY;AAEhB,IAAM,4BAA4B,EACtC,OAAO;AAAA,EACN,SAAS,EAAE,MAAM,gBAAgB;AACnC,CAAC,EACA,QAAQ,qBAAqB;AAIzB,IAAM,wBAAwB,EAClC,OAAO;AAAA,EACN,SAAS,EAAE,OAAO,EAAE,SAAS,qDAAqD;AAAA,EAClF,SAAS,EAAE,OAAO,EAAE,SAAS,0DAA0D;AAAA,EACvF,aAAa,EACV,OAAO,EACP;AAAA,IACC;AAAA,EACF;AAAA,EACF,YAAY,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,SAAS,iCAAiC;AAC5E,CAAC,EACA,QAAQ,iBAAiB;AAErB,IAAM,uBAAuB,EACjC,OAAO;AAAA,EACN,QAAQ,EAAE,QAAQ,IAAI;AAAA,EACtB,WAAW,EAAE,OAAO;AACtB,CAAC,EACA,QAAQ,gBAAgB;AAUpB,IAAM,mBAAmB,EAC7B,OAAO;AAAA,EACN,SAAS,EAAE,OAAO,EAAE,IAAI,CAAC,EAAE,SAAS,4CAA4C;AAAA,EAChF,OAAO,EAAE,OAAO,EAAE,IAAI,CAAC,EAAE,SAAS,4BAA4B;AAAA,EAC9D,QAAQ,EAAE,OAAO,EAAE,SAAS,EAAE,SAAS,sCAAsC;AAAA,EAC7E,MAAM,EACH,KAAK;AAAA,IACJ;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,EACF,CAAC,EACA,SAAS,yEAAoE;AAAA,EAChF,WAAW,EACR,OAAO,EACP,IAAI,EACJ,YAAY,EACZ,SAAS,kEAAkE;AAAA,EAC9E,SAAS,EACN,OAAO,EAAE,OAAO,GAAG,EAAE,QAAQ,CAAC,EAC9B,SAAS,sDAAiD;AAC/D,CAAC,EACA,QAAQ,YAAY;AAEhB,IAAM,4BAA4B,EACtC,OAAO;AAAA,EACN,QAAQ,EACL,MAAM,gBAAgB,EACtB,IAAI,CAAC,EACL,IAAI,GAAM,EACV,SAAS,4EAAuE;AACrF,CAAC,EACA,QAAQ,qBAAqB;AAEzB,IAAM,6BAA6B,EACvC,OAAO;AAAA,EACN,UAAU,EAAE,OAAO,EAAE,IAAI,EAAE,YAAY,EAAE,SAAS,6BAA6B;AAAA,EAC/E,UAAU,EACP,OAAO,EACP,IAAI,EACJ,YAAY,EACZ,SAAS,uEAAkE;AAAA,EAC9E,QAAQ,EACL;AAAA,IACC,EAAE,OAAO;AAAA,MACP,SAAS,EAAE,OAAO,EAAE,SAAS,iCAAiC;AAAA,MAC9D,SAAS,EAAE,OAAO,EAAE,SAAS,6BAA6B;AAAA,IAC5D,CAAC;AAAA,EACH,EACC,QAAQ,CAAC,CAAC;AACf,CAAC,EACA,QAAQ,sBAAsB;AAE1B,IAAM,sBAAsB,EAChC,OAAO;AAAA,EACN,IAAI,EAAE,OAAO,EAAE,SAAS;AAAA,EACxB,QAAQ,EAAE,KAAK,CAAC,QAAQ,SAAS,eAAe,UAAU,UAAU,QAAQ,CAAC;AAAA,EAC7E,MAAM,EAAE,KAAK;AAAA,IACX;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,EACF,CAAC;AAAA,EACD,OAAO,EAAE,QAAQ;AAAA,EACjB,QAAQ,EAAE,OAAO,EAAE,SAAS;AAAA,EAC5B,UAAU,EAAE,KAAK,CAAC,QAAQ,WAAW,SAAS,UAAU,CAAC,EAAE,SAAS;AAAA,EACpE,WAAW,EAAE,OAAO,EAAE,SAAS,eAAe;AAAA,EAC9C,UAAU,EAAE,OAAO,EAAE,OAAO,GAAG,EAAE,QAAQ,CAAC,EAAE,SAAS;AACvD,CAAC,EACA,QAAQ,eAAe;AAEnB,IAAM,wBAAwB,EAClC,OAAO;AAAA,EACN,IAAI,EAAE,OAAO,EAAE,IAAI,CAAC;AAAA,EACpB,WAAW,EAAE,OAAO,EAAE,IAAI,EAAE,YAAY;AAAA,EACxC,cAAc,EAAE,KAAK;AAAA,IACnB;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,EACF,CAAC;AAAA,EACD,UAAU,EAAE,QAAQ;AAAA,EACpB,SAAS,EAAE,MAAM,EAAE,QAAQ,CAAC,EAAE,SAAS;AAAA,EACvC,gBAAgB,EACb,OAAO;AAAA,IACN,MAAM,EAAE,OAAO;AAAA,IACf,MAAM,EAAE,KAAK,CAAC,OAAO,UAAU,MAAM,CAAC,EAAE,SAAS;AAAA,IACjD,SAAS,EAAE,OAAO,EAAE,SAAS;AAAA,IAC7B,oBAAoB,EAAE,QAAQ,EAAE,SAAS;AAAA,IACzC,kBAAkB,EAAE,QAAQ,EAAE,SAAS;AAAA,IACvC,UAAU,EAAE,OAAO,EAAE,OAAO,GAAG,EAAE,QAAQ,CAAC,EAAE,SAAS;AAAA,EACvD,CAAC,EACA,SAAS;AAAA,EACZ,UAAU,EAAE,MAAM,mBAAmB,EAAE,SAAS;AAAA,EAChD,WAAW,EAAE,OAAO;AAAA,EACpB,UAAU,EAAE,OAAO,EAAE,OAAO,GAAG,EAAE,QAAQ,CAAC,EAAE,SAAS;AACvD,CAAC,EACA,QAAQ,iBAAiB;AAErB,IAAM,2BAA2B,EACrC,OAAO;AAAA,EACN,IAAI,EAAE,OAAO,EAAE,IAAI,CAAC,EAAE,SAAS,gDAAgD;AAAA,EAC/E,WAAW,EAAE,OAAO,EAAE,SAAS;AAAA,EAC/B,YAAY,EAAE,OAAO,EAAE,SAAS;AAAA,EAChC,MAAM,EAAE,OAAO;AAAA,IACb,QAAQ,EAAE,OAAO,EAAE,IAAI,CAAC;AAAA,IACxB,SAAS,EAAE,QAAQ,EAAE,SAAS;AAAA,EAChC,CAAC;AAAA,EACD,UAAU,EAAE,MAAM,qBAAqB,EAAE,QAAQ,CAAC,CAAC;AAAA,EACnD,QAAQ,EAAE,MAAM,mBAAmB,EAAE,QAAQ,CAAC,CAAC;AAAA,EAC/C,SAAS,EACN,OAAO;AAAA,IACN,SAAS,EAAE,QAAQ,EAAE,SAAS;AAAA,IAC9B,OAAO,EAAE,OAAO,EAAE,SAAS;AAAA,IAC3B,SAAS,EAAE,OAAO,EAAE,OAAO,GAAG,EAAE,OAAO,CAAC,EAAE,SAAS;AAAA,IACnD,SAAS,EAAE,OAAO,EAAE,SAAS;AAAA,IAC7B,QAAQ,EAAE,OAAO,EAAE,SAAS;AAAA,IAC5B,YAAY,EAAE,OAAO,EAAE,SAAS;AAAA,IAChC,UAAU,EAAE,OAAO,EAAE,OAAO,GAAG,EAAE,QAAQ,CAAC,EAAE,SAAS;AAAA,EACvD,CAAC,EACA,SAAS;AAAA,EACZ,OAAO,EAAE,KAAK,CAAC,SAAS,OAAO,QAAQ,SAAS,CAAC,EAAE,SAAS;AAAA,EAC5D,MAAM,EAAE,OAAO,EAAE,OAAO,GAAG,EAAE,OAAO,CAAC,EAAE,SAAS;AAAA,EAChD,WAAW,EAAE,OAAO,EAAE,SAAS,eAAe;AAAA,EAC9C,WAAW,EAAE,OAAO,EAAE,SAAS;AAAA,EAC/B,UAAU,EAAE,OAAO,EAAE,OAAO,GAAG,EAAE,QAAQ,CAAC,EAAE,SAAS;AACvD,CAAC,EACA,QAAQ,oBAAoB;AAExB,IAAM,+BAA+B,EACzC,OAAO;AAAA,EACN,IAAI,EAAE,OAAO,EAAE,SAAS,mCAAmC;AAAA,EAC3D,WAAW,EAAE,QAAQ,EAAE,SAAS,wDAAwD;AAC1F,CAAC,EACA,QAAQ,wBAAwB;AAU5B,IAAM,sBAAsB,EAChC,OAAO;AAAA,EACN,OAAO,EACJ,OAAO;AAAA,IACN,MAAM,EACH,OAAO,EACP;AAAA,MACC;AAAA,IACF;AAAA,IACF,SAAS,EAAE,OAAO,EAAE,SAAS,yBAAyB;AAAA,IACtD,SAAS,EAAE,QAAQ,EAAE,SAAS,EAAE,SAAS,6BAA6B;AAAA,EACxE,CAAC,EACA,SAAS,+DAA+D;AAC7E,CAAC,EACA,QAAQ,eAAe;AAoBnB,IAAM,eAAe;AAMrB,SAAS,WAAW,QAAwB;AACjD,QAAM,SAAS,gBAAgB,MAAM;AACrC,MAAI,IAAI;AACR,WAAS,IAAI,GAAG,IAAI,OAAO,QAAQ,KAAK;AACtC,QAAK,IAAI,KAAM,OAAO,WAAW,CAAC;AAAA,EACpC;AAEA,SAAO,GAAG,OAAO,IAAI,KAAK,MAAM,GAAG,SAAS,EAAE,EAAE,SAAS,GAAG,GAAG,CAAC;AAClE;AAEA,SAAS,gBAAgB,OAAwB;AAC/C,MAAI,MAAM,QAAQ,KAAK,EAAG,QAAO,IAAI,MAAM,IAAI,CAAC,SAAS,gBAAgB,IAAI,CAAC,EAAE,KAAK,GAAG,CAAC;AACzF,MAAI,SAAS,OAAO,UAAU,UAAU;AACtC,UAAM,UAAU,OAAO,QAAQ,KAAgC,EAC5D,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,cAAc,CAAC,CAAC,EACrC,IAAI,CAAC,CAAC,KAAK,IAAI,MAAM,GAAG,KAAK,UAAU,GAAG,CAAC,IAAI,gBAAgB,IAAI,CAAC,EAAE;AACzE,WAAO,IAAI,QAAQ,KAAK,GAAG,CAAC;AAAA,EAC9B;AACA,SAAO,KAAK,UAAU,KAAK;AAC7B;;;ACjYA,IAAM,YAAoB;AAAA,EACxB,MAAM;AAAA,EACN,aACE;AAAA,EACF,cAAc;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAuBd,YAAY;AAAA,IACV;AAAA,MACE,IAAI;AAAA,MACJ,aAAa;AAAA,MACb,QAAQ;AAAA,MACR,KAAK;AAAA,MACL,KAAK;AAAA,IACP;AAAA,IACA;AAAA,MACE,IAAI;AAAA,MACJ,aAAa;AAAA,MACb,QAAQ;AAAA,MACR,KAAK;AAAA,MACL,KAAK;AAAA,IACP;AAAA,IACA;AAAA,MACE,IAAI;AAAA,MACJ,aAAa;AAAA,MACb,QAAQ;AAAA,MACR,KAAK;AAAA,MACL,KAAK;AAAA,IACP;AAAA,EACF;AAAA,EACA,cAAc;AAAA,IACZ,EAAE,IAAI,cAAc,aAAa,qCAAqC;AAAA,IACtE,EAAE,IAAI,kBAAkB,aAAa,sCAAsC;AAAA,IAC3E,EAAE,IAAI,eAAe,aAAa,oCAAoC;AAAA,IACtE,EAAE,IAAI,WAAW,aAAa,kCAAkC;AAAA,IAChE,EAAE,IAAI,mBAAmB,aAAa,qCAAqC;AAAA,IAC3E,EAAE,IAAI,WAAW,aAAa,oCAAoC;AAAA,IAClE,EAAE,IAAI,eAAe,aAAa,uCAAuC;AAAA,EAC3E;AAAA,EACA,MAAM;AAAA,IACJ,EAAE,IAAI,sBAAsB,aAAa,qCAAqC;AAAA,IAC9E,EAAE,IAAI,iBAAiB,aAAa,4BAA4B;AAAA,IAChE,EAAE,IAAI,0BAA0B,aAAa,wBAAwB;AAAA,IACrE,EAAE,IAAI,kBAAkB,aAAa,oCAAoC;AAAA,EAC3E;AACF;AAIO,IAAM,kBAA0C;AAAA,EACrD,aAAa;AACf;AAGO,SAAS,iBAAiB,MAAkC;AACjE,SAAO,gBAAgB,IAAI;AAC7B;AAGO,SAAS,qBAAqB;AACnC,SAAO,OAAO,OAAO,eAAe,EAAE,IAAI,CAAC,OAAO;AAAA,IAChD,MAAM,EAAE;AAAA,IACR,aAAa,EAAE;AAAA,IACf,YAAY,EAAE,WAAW,IAAI,CAAC,OAAO;AAAA,MACnC,IAAI,EAAE;AAAA,MACN,aAAa,EAAE;AAAA,MACf,QAAQ,EAAE;AAAA,IACZ,EAAE;AAAA,IACF,cAAc,EAAE,aAAa,IAAI,CAAC,MAAM,EAAE,EAAE;AAAA,IAC5C,eAAe,WAAW,CAAC;AAAA,EAC7B,EAAE;AACJ;;;ACwHA,SAAS,oBAAoB;AAC7B,SAAS,SAAS,eAAe;AACjC,SAAS,qBAAqB;AA9MvB,IAAM,YAAN,cAAwB,MAAM;AAAA,EACnC,YACkB,MAChB,SACgB,SAAiB,KACjB,SAChB;AACA,UAAM,OAAO;AALG;AAEA;AACA;AAGhB,SAAK,OAAO;AAAA,EACd;AAAA,EAPkB;AAAA,EAEA;AAAA,EACA;AAKpB;AAKA,SAAS,kBAAkB,QAAgB;AACzC,SAAO;AAAA,IACL,MAAM;AAAA,IACN,QAAQ;AAAA,MACN,MAAM;AAAA,MACN,sBAAsB;AAAA,MACtB,YAAY;AAAA,QACV,YAAY;AAAA,UACV,MAAM;AAAA,UACN,sBAAsB;AAAA,UACtB,YAAY,OAAO;AAAA,YACjB,OAAO,WAAW,IAAI,CAAC,MAAM;AAAA,cAC3B,EAAE;AAAA,cACF,EAAE,MAAM,UAAU,SAAS,EAAE,KAAK,SAAS,EAAE,IAAI;AAAA,YACnD,CAAC;AAAA,UACH;AAAA,UACA,UAAU,OAAO,WAAW,IAAI,CAAC,MAAM,EAAE,EAAE;AAAA,QAC7C;AAAA,QACA,cAAc;AAAA,UACZ,MAAM;AAAA,UACN,OAAO,EAAE,MAAM,UAAU,MAAM,OAAO,aAAa,IAAI,CAAC,MAAM,EAAE,EAAE,EAAE;AAAA,QACtE;AAAA,QACA,MAAM;AAAA,UACJ,MAAM;AAAA,UACN,OAAO,EAAE,MAAM,UAAU,MAAM,OAAO,KAAK,IAAI,CAAC,MAAM,EAAE,EAAE,EAAE;AAAA,QAC9D;AAAA,QACA,WAAW,EAAE,MAAM,SAAS;AAAA,MAC9B;AAAA,MACA,UAAU,CAAC,cAAc,WAAW;AAAA,IACtC;AAAA,EACF;AACF;AASA,SAAS,oBAAoB,OAAgB,QAA6B;AACxE,MAAI,CAAC,SAAS,OAAO,UAAU,UAAU;AACvC,UAAM,IAAI,UAAU,eAAe,oCAAoC,KAAK,KAAK;AAAA,EACnF;AACA,QAAM,MAAM;AACZ,QAAM,gBAAgB,IAAI;AAC1B,MAAI,CAAC,iBAAiB,OAAO,kBAAkB,YAAY,MAAM,QAAQ,aAAa,GAAG;AACvF,UAAM,IAAI,UAAU,eAAe,wCAAwC,KAAK,KAAK;AAAA,EACvF;AAEA,QAAM,aAAqC,CAAC;AAC5C,QAAM,kBAAkB;AACxB,aAAW,OAAO,OAAO,YAAY;AACnC,UAAM,QAAQ,gBAAgB,IAAI,EAAE;AACpC,QACE,OAAO,UAAU,YACjB,CAAC,OAAO,SAAS,KAAK,KACtB,QAAQ,IAAI,OACZ,QAAQ,IAAI,KACZ;AACA,YAAM,IAAI;AAAA,QACR;AAAA,QACA,+CAA+C,IAAI,EAAE;AAAA,QACrD;AAAA,QACA;AAAA,MACF;AAAA,IACF;AACA,eAAW,IAAI,EAAE,IAAI;AAAA,EACvB;AAEA,QAAM,kBAAkB,IAAI,IAAI,OAAO,aAAa,IAAI,CAAC,SAAS,KAAK,EAAE,CAAC;AAC1E,QAAM,cAAc,IAAI,IAAI,OAAO,KAAK,IAAI,CAAC,QAAQ,IAAI,EAAE,CAAC;AAC5D,QAAM,eAAe,gBAAgB,IAAI,cAAc,iBAAiB,gBAAgB,KAAK;AAC7F,QAAM,OAAO,gBAAgB,IAAI,MAAM,aAAa,QAAQ,KAAK;AACjE,MAAI,OAAO,IAAI,cAAc,YAAY,IAAI,UAAU,KAAK,EAAE,WAAW,GAAG;AAC1E,UAAM,IAAI,UAAU,eAAe,qCAAqC,KAAK,KAAK;AAAA,EACpF;AAEA,SAAO,EAAE,YAAY,cAAc,MAAM,WAAW,IAAI,UAAU;AACpE;AAEA,SAAS,gBACP,KACA,SACA,OACA,UACU;AACV,MAAI,QAAQ,OAAW,QAAO,CAAC;AAC/B,MAAI,CAAC,MAAM,QAAQ,GAAG,GAAG;AACvB,UAAM,IAAI,UAAU,eAAe,4BAA4B,KAAK,KAAK,KAAK,QAAQ;AAAA,EACxF;AACA,QAAM,MAAgB,CAAC;AACvB,aAAW,QAAQ,KAAK;AACtB,QAAI,OAAO,SAAS,YAAY,CAAC,QAAQ,IAAI,IAAI,GAAG;AAClD,YAAM,IAAI;AAAA,QACR;AAAA,QACA,0BAA0B,KAAK,QAAQ,OAAO,IAAI,CAAC;AAAA,QACnD;AAAA,QACA;AAAA,MACF;AAAA,IACF;AACA,QAAI,KAAK,IAAI;AAAA,EACf;AACA,SAAO;AACT;AAEA,SAAS,eAAe,YAAoC,QAAwB;AAClF,MAAI,WAAW;AACf,MAAI,cAAc;AAClB,aAAW,OAAO,OAAO,YAAY;AACnC,UAAM,MAAM,WAAW,IAAI,EAAE,KAAK;AAClC,UAAM,QAAQ,IAAI,MAAM,IAAI,OAAO;AACnC,UAAM,aAAa,KAAK,IAAI,GAAG,KAAK,IAAI,IAAI,MAAM,IAAI,OAAO,KAAK,CAAC;AACnE,gBAAY,aAAa,IAAI;AAC7B,mBAAe,IAAI;AAAA,EACrB;AACA,SAAO,cAAc,IAAI,WAAW,cAAc;AACpD;AAEA,SAAS,iBAAiB,SAAiB,SAA0B;AACnE,QAAM,MAAM,WAAW,OAAO,KAAK,OAAiB,EAAE,SAAS,KAAK,UAAU,OAAO,IAAI;AACzF,SAAO;AAAA,IACL;AAAA,IACA;AAAA,IACA;AAAA,IACA,MAAM,yCAAyC;AAAA,IAC/C,MAAM,MAAM;AAAA,EACd,EACG,OAAO,OAAO,EACd,KAAK,IAAI;AACd;AAEA,IAAM,sBAAsB;AAE5B,eAAsB,YAAY,KAAyC;AAEzE,MAAI;AACJ,MAAI,IAAI,YAAY;AAClB,UAAM,QAAQ,iBAAiB,IAAI,UAAU;AAC7C,QAAI,CAAC,OAAO;AACV,YAAM,IAAI,UAAU,oBAAoB,6BAA6B,IAAI,UAAU,MAAM,GAAG;AAAA,IAC9F;AACA,aAAS;AAAA,EACX,WAAW,IAAI,QAAQ;AACrB,aAAS,IAAI;AAAA,EACf,OAAO;AAEL,UAAM,IAAI,UAAU,oBAAoB,4CAA4C,GAAG;AAAA,EACzF;AAEA,QAAM,YAAY,KAAK,IAAI;AAC3B,QAAM,QAAQ,IAAI,SAAS;AAE3B,QAAM,EAAE,OAAO,OAAO,IAAI,MAAM,YAAyB;AAAA,IACvD;AAAA,IACA,UAAU;AAAA,MACR,EAAE,MAAM,UAAU,SAAS,OAAO,aAAa;AAAA,MAC/C,EAAE,MAAM,QAAQ,SAAS,iBAAiB,IAAI,SAAS,IAAI,OAAO,EAAE;AAAA,IACtE;AAAA,IACA,YAAY,kBAAkB,MAAM;AAAA,IACpC,aAAa;AAAA,IACb,WAAW;AAAA,EACb,CAAC;AAED,QAAM,SAAS,oBAAoB,OAAO,MAAM;AAEhD,QAAM,YAAY,eAAe,OAAO,YAAY,MAAM;AAC1D,QAAM,aAAa,KAAK,IAAI,IAAI;AAEhC,SAAO;AAAA,IACL;AAAA,IACA,YAAY,OAAO;AAAA,IACnB,cAAc,OAAO,gBAAgB,CAAC;AAAA,IACtC,MAAM,OAAO,QAAQ,CAAC;AAAA,IACtB,WAAW,OAAO;AAAA,IAClB,eAAe,WAAW,MAAM;AAAA,IAChC,OAAO,OAAO;AAAA,IACd;AAAA,EACF;AACF;AAIO,SAAS,oBAAyC;AACvD,SAAO,EAAE,SAAS,mBAAmB,EAAE;AACzC;AAQA,IAAI;AAEJ,SAAS,qBAA6B;AACpC,MAAI,eAAgB,QAAO;AAG3B,QAAM,OAAO,QAAQ,cAAc,YAAY,GAAG,CAAC;AACnD,QAAM,aAAa;AAAA,IACjB,QAAQ,MAAM,MAAM,MAAM,cAAc;AAAA;AAAA,IACxC,QAAQ,MAAM,MAAM,cAAc;AAAA;AAAA,EACpC;AACA,aAAW,QAAQ,YAAY;AAC7B,QAAI;AACF,YAAM,MAAM,KAAK,MAAM,aAAa,MAAM,OAAO,CAAC;AAClD,UAAI,IAAI,SAAS;AACf,yBAAiB,IAAI;AACrB,eAAO,IAAI;AAAA,MACb;AAAA,IACF,QAAQ;AAAA,IAER;AAAA,EACF;AACA,SAAO;AACT;AAEO,SAAS,gBAAiC;AAC/C,SAAO;AAAA,IACL,SAAS;AAAA,IACT,SAAS,mBAAmB;AAAA,IAC5B,aAAa;AAAA,IACb,YAAY,CAAC,SAAS,eAAe,WAAW,mBAAmB,eAAe;AAAA,EACpF;AACF;AA2BA,eAAsB,mBACpB,KACA,QAC+B;AAC/B,MAAI,CAAC,OAAO,YAAY;AACtB,UAAM,IAAI;AAAA,MACR;AAAA,MACA;AAAA,MACA;AAAA,IACF;AAAA,EACF;AACA,QAAM,SAAsD,CAAC;AAC7D,MAAI,WAAW;AACf,aAAW,SAAS,IAAI,QAAQ;AAC9B,QAAI;AAEF,YAAM,OAAO,WAAW,YAAY,KAA2B;AAC/D;AAAA,IACF,SAAS,KAAK;AACZ,aAAO,KAAK;AAAA,QACV,SAAS,MAAM;AAAA,QACf,SAAS,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;AAAA,MAC1D,CAAC;AAAA,IACH;AAAA,EACF;AACA,SAAO,EAAE,UAAU,UAAU,OAAO,QAAQ,OAAO;AACrD;AAOA,eAAsB,qBACpB,KACA,QACiC;AACjC,MAAI,CAAC,OAAO,eAAe;AACzB,UAAM,IAAI;AAAA,MACR;AAAA,MACA;AAAA,MACA;AAAA,IACF;AAAA,EACF;AAIA,QAAM,OAAO,cAAc,KAAK,GAAgE;AAChG,SAAO,EAAE,IAAI,IAAI,IAAI,WAAW,KAAK;AACvC;;;AChVA,SAAS,iBAAiB,2BAA2B;AAiB9C,SAAS,aAAa,gBAAuC;AAClE,QAAM,WAAW,IAAI,gBAAgB;AAGrC,WAAS,SAAS,gBAAgB,kBAAkB;AACpD,WAAS,SAAS,eAAe,iBAAiB;AAClD,WAAS,SAAS,uBAAuB,yBAAyB;AAClE,WAAS,SAAS,mBAAmB,qBAAqB;AAC1D,WAAS,SAAS,kBAAkB,oBAAoB;AACxD,WAAS,SAAS,iBAAiB,mBAAmB;AACtD,WAAS,SAAS,uBAAuB,yBAAyB;AAClE,WAAS,SAAS,wBAAwB,0BAA0B;AACpE,WAAS,SAAS,sBAAsB,wBAAwB;AAChE,WAAS,SAAS,0BAA0B,4BAA4B;AAGxE,WAAS,aAAa;AAAA,IACpB,QAAQ;AAAA,IACR,MAAM;AAAA,IACN,SAAS;AAAA,IACT,aACE;AAAA,IACF,SAAS;AAAA,MACP,MAAM;AAAA,QACJ,SAAS;AAAA,UACP,oBAAoB,EAAE,QAAQ,mBAAmB;AAAA,QACnD;AAAA,MACF;AAAA,IACF;AAAA,IACA,WAAW;AAAA,MACT,KAAK;AAAA,QACH,aAAa;AAAA,QACb,SAAS,EAAE,oBAAoB,EAAE,QAAQ,kBAAkB,EAAE;AAAA,MAC/D;AAAA,MACA,KAAK;AAAA,QACH,aAAa;AAAA,QACb,SAAS,EAAE,oBAAoB,EAAE,QAAQ,oBAAoB,EAAE;AAAA,MACjE;AAAA,MACA,KAAK;AAAA,QACH,aAAa;AAAA,QACb,SAAS,EAAE,oBAAoB,EAAE,QAAQ,oBAAoB,EAAE;AAAA,MACjE;AAAA,MACA,KAAK;AAAA,QACH,aAAa;AAAA,QACb,SAAS,EAAE,oBAAoB,EAAE,QAAQ,oBAAoB,EAAE;AAAA,MACjE;AAAA,IACF;AAAA,EACF,CAAC;AAED,WAAS,aAAa;AAAA,IACpB,QAAQ;AAAA,IACR,MAAM;AAAA,IACN,SAAS;AAAA,IACT,aACE;AAAA,IACF,WAAW;AAAA,MACT,KAAK;AAAA,QACH,aAAa;AAAA,QACb,SAAS,EAAE,oBAAoB,EAAE,QAAQ,0BAA0B,EAAE;AAAA,MACvE;AAAA,IACF;AAAA,EACF,CAAC;AAED,WAAS,aAAa;AAAA,IACpB,QAAQ;AAAA,IACR,MAAM;AAAA,IACN,SAAS;AAAA,IACT,aAAa;AAAA,IACb,WAAW;AAAA,MACT,KAAK;AAAA,QACH,aAAa;AAAA,QACb,SAAS,EAAE,oBAAoB,EAAE,QAAQ,sBAAsB,EAAE;AAAA,MACnE;AAAA,IACF;AAAA,EACF,CAAC;AAED,WAAS,aAAa;AAAA,IACpB,QAAQ;AAAA,IACR,MAAM;AAAA,IACN,SAAS;AAAA,IACT,WAAW;AAAA,MACT,KAAK;AAAA,QACH,aAAa;AAAA,QACb,SAAS,EAAE,oBAAoB,EAAE,QAAQ,qBAAqB,EAAE;AAAA,MAClE;AAAA,IACF;AAAA,EACF,CAAC;AAED,WAAS,aAAa;AAAA,IACpB,QAAQ;AAAA,IACR,MAAM;AAAA,IACN,SAAS;AAAA,IACT,aACE;AAAA,IACF,SAAS;AAAA,MACP,MAAM;AAAA,QACJ,SAAS;AAAA,UACP,oBAAoB,EAAE,QAAQ,0BAA0B;AAAA,UACxD,wBAAwB,EAAE,QAAQ,0BAA0B;AAAA,QAC9D;AAAA,MACF;AAAA,IACF;AAAA,IACA,WAAW;AAAA,MACT,KAAK;AAAA,QACH,aAAa;AAAA,QACb,SAAS,EAAE,oBAAoB,EAAE,QAAQ,2BAA2B,EAAE;AAAA,MACxE;AAAA,MACA,KAAK;AAAA,QACH,aAAa;AAAA,QACb,SAAS,EAAE,oBAAoB,EAAE,QAAQ,oBAAoB,EAAE;AAAA,MACjE;AAAA,MACA,KAAK;AAAA,QACH,aAAa;AAAA,QACb,SAAS,EAAE,oBAAoB,EAAE,QAAQ,oBAAoB,EAAE;AAAA,MACjE;AAAA,MACA,KAAK;AAAA,QACH,aAAa;AAAA,QACb,SAAS,EAAE,oBAAoB,EAAE,QAAQ,oBAAoB,EAAE;AAAA,MACjE;AAAA,IACF;AAAA,EACF,CAAC;AAED,WAAS,aAAa;AAAA,IACpB,QAAQ;AAAA,IACR,MAAM;AAAA,IACN,SAAS;AAAA,IACT,aACE;AAAA,IACF,SAAS;AAAA,MACP,MAAM;AAAA,QACJ,SAAS;AAAA,UACP,oBAAoB,EAAE,QAAQ,yBAAyB;AAAA,QACzD;AAAA,MACF;AAAA,IACF;AAAA,IACA,WAAW;AAAA,MACT,KAAK;AAAA,QACH,aAAa;AAAA,QACb,SAAS,EAAE,oBAAoB,EAAE,QAAQ,6BAA6B,EAAE;AAAA,MAC1E;AAAA,MACA,KAAK;AAAA,QACH,aAAa;AAAA,QACb,SAAS,EAAE,oBAAoB,EAAE,QAAQ,oBAAoB,EAAE;AAAA,MACjE;AAAA,MACA,KAAK;AAAA,QACH,aAAa;AAAA,QACb,SAAS,EAAE,oBAAoB,EAAE,QAAQ,oBAAoB,EAAE;AAAA,MACjE;AAAA,MACA,KAAK;AAAA,QACH,aAAa;AAAA,QACb,SAAS,EAAE,oBAAoB,EAAE,QAAQ,oBAAoB,EAAE;AAAA,MACjE;AAAA,IACF;AAAA,EACF,CAAC;AAED,QAAM,YAAY,IAAI,oBAAoB,SAAS,WAAW;AAC9D,QAAM,MAAM,UAAU,iBAAiB;AAAA,IACrC,SAAS;AAAA,IACT,MAAM;AAAA,MACJ,OAAO;AAAA,MACP,SAAS;AAAA,MACT,aAAa;AAAA;AAAA,yBAEM,YAAY;AAAA,MAC/B,SAAS,EAAE,MAAM,kBAAkB,KAAK,+CAA+C;AAAA,MACvF,SAAS,EAAE,MAAM,MAAM;AAAA,IACzB;AAAA,IACA,SAAS,CAAC,EAAE,KAAK,yBAAyB,aAAa,yBAAyB,CAAC;AAAA,EACnF,CAAC;AACD,QAAM,YAAY,EAAE,MAAM,8BAA8B;AACxD,QAAM,oBAAoB;AAAA,IACxB,SAAS,EAAE,MAAM,UAAU,WAAW,EAAE;AAAA,IACxC,SAAS,EAAE,MAAM,UAAU,sBAAsB,KAAK;AAAA,IACtD,OAAO,EAAE,MAAM,SAAS;AAAA,EAC1B;AACA,MAAI,eAAe,CAAC;AACpB,MAAI,WAAW,YAAY,CAAC;AAC5B,MAAI,WAAW,QAAQ,eAAe;AAAA,IACpC,OAAO;AAAA,MACL;AAAA,QACE,MAAM;AAAA,QACN,sBAAsB;AAAA,QACtB,UAAU,CAAC,cAAc,SAAS;AAAA,QAClC,YAAY;AAAA,UACV,YAAY,EAAE,MAAM,UAAU,WAAW,EAAE;AAAA,UAC3C,GAAG;AAAA,QACL;AAAA,MACF;AAAA,MACA;AAAA,QACE,MAAM;AAAA,QACN,sBAAsB;AAAA,QACtB,UAAU,CAAC,UAAU,SAAS;AAAA,QAC9B,YAAY;AAAA,UACV,QAAQ;AAAA,UACR,GAAG;AAAA,QACL;AAAA,MACF;AAAA,IACF;AAAA,IACA,aAAa;AAAA,EACf;AACA,SAAO;AACT;;;ACxMA,eAAsB,YAAY,KAAiD;AACjF,MAAI;AACF,YAAQ,IAAI,QAAQ;AAAA,MAClB,KAAK,SAAS;AACZ,cAAM,SAAS,mBAAmB,UAAU,IAAI,MAAM;AACtD,YAAI,CAAC,OAAO,SAAS;AACnB,iBAAO;AAAA,YACL,OAAO;AAAA,cACL,MAAM;AAAA,cACN,SAAS;AAAA,cACT,SAAS,OAAO,MAAM;AAAA,YACxB;AAAA,UACF;AAAA,QACF;AACA,eAAO,EAAE,QAAQ,MAAM,YAAY,OAAO,IAAI,EAAE;AAAA,MAClD;AAAA,MACA,KAAK;AACH,eAAO,EAAE,QAAQ,kBAAkB,EAAE;AAAA,MACvC,KAAK;AACH,eAAO,EAAE,QAAQ,cAAc,EAAE;AAAA,MACnC;AACE,eAAO;AAAA,UACL,OAAO;AAAA,YACL,MAAM;AAAA,YACN,SAAS,mBAAoB,IAA2B,MAAM;AAAA,UAChE;AAAA,QACF;AAAA,IACJ;AAAA,EACF,SAAS,KAAK;AACZ,QAAI,eAAe,WAAW;AAC5B,aAAO,EAAE,OAAO,EAAE,MAAM,IAAI,MAAM,SAAS,IAAI,SAAS,SAAS,IAAI,QAAQ,EAAE;AAAA,IACjF;AACA,UAAM,UAAU,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;AAC/D,WAAO,EAAE,OAAO,EAAE,MAAM,kBAAkB,QAAQ,EAAE;AAAA,EACtD;AACF;AAIA,eAAe,QAAQ,QAAgD;AACrE,QAAM,SAAmB,CAAC;AAC1B,mBAAiB,SAAS,QAAQ;AAChC,WAAO,KAAK,OAAO,SAAS,KAAK,IAAI,QAAQ,OAAO,KAAK,KAAe,CAAC;AAAA,EAC3E;AACA,SAAO,OAAO,OAAO,MAAM,EAAE,SAAS,OAAO;AAC/C;AAGA,eAAsB,WAAW,QAAkC;AACjE,QAAM,MAAM,MAAM,QAAQ,QAAQ,KAAK;AACvC,MAAI;AACJ,MAAI;AACF,UAAM,OAAO,KAAK,MAAM,GAAG;AAC3B,UAAM,SAAS,EAAE,QAAwC,QAAQ,KAAK,IAAK;AAAA,EAC7E,SAAS,KAAK;AACZ,YAAQ,OAAO;AAAA,MACb,GAAG,KAAK,UAAU;AAAA,QAChB,OAAO;AAAA,UACL,MAAM;AAAA,UACN,SAAS,6BAA6B,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG,CAAC;AAAA,QACxF;AAAA,MACF,CAAC,CAAC;AAAA;AAAA,IACJ;AACA,WAAO;AAAA,EACT;AACA,QAAM,MAAM,MAAM,YAAY,GAAG;AACjC,UAAQ,OAAO,MAAM,GAAG,KAAK,UAAU,GAAG,CAAC;AAAA,CAAI;AAC/C,SAAO,WAAW,MAAM,IAAI;AAC9B;AAGA,eAAsB,YAAY,QAAkC;AAClE,QAAM,MAAM,MAAM,QAAQ,QAAQ,KAAK;AACvC,QAAM,QAAQ,IAAI,MAAM,IAAI,EAAE,OAAO,CAAC,MAAM,EAAE,KAAK,EAAE,SAAS,CAAC;AAC/D,MAAI,WAAW;AACf,aAAW,QAAQ,OAAO;AACxB,QAAI;AACJ,QAAI;AACF,YAAM,OAAO,KAAK,MAAM,IAAI;AAC5B,YAAM,SAAS,EAAE,QAAwC,QAAQ,KAAK,IAAK;AAAA,IAC7E,SAAS,KAAK;AACZ,cAAQ,OAAO;AAAA,QACb,GAAG,KAAK,UAAU;AAAA,UAChB,OAAO;AAAA,YACL,MAAM;AAAA,YACN,SAAS,4BAA4B,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG,CAAC;AAAA,UACvF;AAAA,QACF,CAAC,CAAC;AAAA;AAAA,MACJ;AACA,iBAAW;AACX;AAAA,IACF;AACA,UAAM,MAAM,MAAM,YAAY,GAAG;AACjC,YAAQ,OAAO,MAAM,GAAG,KAAK,UAAU,GAAG,CAAC;AAAA,CAAI;AAC/C,QAAI,WAAW,IAAK,YAAW;AAAA,EACjC;AACA,SAAO;AACT;;;AC/GA,SAA0B,aAAa;AACvC,SAAS,YAAY;AACrB,SAAS,YAAY;AAcrB,IAAM,aAAa,KAAK,IAAI;AAmB5B,IAAM,oBAAoB,oBAAI,IAAI,CAAC,YAAY,eAAe,eAAe,CAAC;AAEvE,SAAS,UAAU,OAAyB,CAAC,GAAG;AACrD,QAAM,MAAM,IAAI,KAAK;AAErB,MAAI,IAAI,KAAK,KAAK,CAAC;AAGnB,MAAI,KAAK,MAAM;AACb,UAAM,SAAS,KAAK,KAAK;AACzB,QAAI,IAAI,KAAK,OAAO,GAAG,SAAS;AAC9B,YAAM,OAAO,IAAI,IAAI,EAAE,IAAI,GAAG,EAAE;AAChC,UAAI,kBAAkB,IAAI,IAAI,EAAG,QAAO,KAAK;AAC7C,YAAM,MAAM,EAAE,IAAI,OAAO,eAAe,KAAK;AAC7C,YAAM,QAAQ,IAAI,MAAM,kBAAkB;AAC1C,UAAI,CAAC,OAAO;AACV,cAAM,IAAI,UAAU,gBAAgB,8CAA8C,GAAG;AAAA,MACvF;AACA,YAAM,QAAQ,MAAM,CAAC;AACrB,YAAM,KAAK,OAAO,WAAW,WAAW,UAAU,SAAS,MAAM,OAAO,KAAK;AAC7E,UAAI,CAAC,IAAI;AACP,cAAM,IAAI,UAAU,gBAAgB,yBAAyB,GAAG;AAAA,MAClE;AACA,aAAO,KAAK;AAAA,IACd,CAAC;AAAA,EACH;AAEA,MAAI,QAAQ,CAAC,KAAK,MAAM;AACtB,QAAI,eAAe,WAAW;AAC5B,YAAM,SAAS,IAAI;AACnB,aAAO,EAAE;AAAA,QACP,EAAE,OAAO,EAAE,MAAM,IAAI,MAAM,SAAS,IAAI,SAAS,SAAS,IAAI,QAAQ,EAAE;AAAA,QACxE;AAAA,MACF;AAAA,IACF;AAEA,YAAQ,MAAM,iCAAiC,GAAG;AAClD,WAAO,EAAE,KAAK,EAAE,OAAO,EAAE,MAAM,kBAAkB,SAAS,yBAAyB,EAAE,GAAG,GAAG;AAAA,EAC7F,CAAC;AAGD,MAAI;AAAA,IAAI;AAAA,IAAY,CAAC,MACnB,EAAE,KAAK,EAAE,QAAQ,MAAe,YAAY,KAAK,IAAI,IAAI,cAAc,IAAK,CAAC;AAAA,EAC/E;AAGA,MAAI,IAAI,eAAe,CAAC,MAAM,EAAE,KAAK,cAAc,CAAC,CAAC;AAGrD,MAAI,IAAI,eAAe,CAAC,MAAM,EAAE,KAAK,kBAAkB,CAAC,CAAC;AAGzD,MAAI,KAAK,aAAa,OAAO,MAAM;AACjC,UAAM,MAAM,MAAM,EAAE,IAAI,KAAK,EAAE,MAAM,MAAM,IAAI;AAC/C,QAAI,OAAO,MAAM;AACf,YAAM,IAAI,UAAU,oBAAoB,8BAA8B,GAAG;AAAA,IAC3E;AACA,UAAM,SAAS,mBAAmB,UAAU,GAAG;AAC/C,QAAI,CAAC,OAAO,SAAS;AACnB,YAAM,IAAI;AAAA,QACR;AAAA,QACA;AAAA,QACA;AAAA,QACA,OAAO,MAAM;AAAA,MACf;AAAA,IACF;AACA,UAAM,SAAS,MAAM,YAAY,OAAO,IAAI;AAC5C,WAAO,EAAE,KAAK,MAAM;AAAA,EACtB,CAAC;AAGD,MAAI,KAAK,qBAAqB,OAAO,MAAM;AACzC,UAAM,cAAc,EAAE,IAAI,OAAO,cAAc,KAAK;AACpD,QAAI;AACJ,QAAI,YAAY,SAAS,sBAAsB,GAAG;AAChD,YAAM,OAAO,MAAM,EAAE,IAAI,KAAK;AAC9B,YAAM,SAAS,KACZ,MAAM,IAAI,EACV,IAAI,CAAC,SAAS,KAAK,KAAK,CAAC,EACzB,OAAO,CAAC,SAAS,KAAK,SAAS,CAAC,EAChC,IAAI,CAAC,SAAS;AACb,YAAI;AACF,iBAAO,KAAK,MAAM,IAAI;AAAA,QACxB,QAAQ;AACN,gBAAM,IAAI;AAAA,YACR;AAAA,YACA;AAAA,YACA;AAAA,YACA,KAAK,MAAM,GAAG,GAAG;AAAA,UACnB;AAAA,QACF;AAAA,MACF,CAAC;AACH,gBAAU,EAAE,OAAO;AAAA,IACrB,OAAO;AACL,gBAAU,MAAM,EAAE,IAAI,KAAK,EAAE,MAAM,MAAM,IAAI;AAAA,IAC/C;AACA,QAAI,WAAW,MAAM;AACnB,YAAM,IAAI,UAAU,oBAAoB,wCAAwC,GAAG;AAAA,IACrF;AACA,UAAM,SAAS,0BAA0B,UAAU,OAAO;AAC1D,QAAI,CAAC,OAAO,SAAS;AACnB,YAAM,IAAI;AAAA,QACR;AAAA,QACA;AAAA,QACA;AAAA,QACA,OAAO,MAAM;AAAA,MACf;AAAA,IACF;AACA,UAAM,SAAS,MAAM,mBAAmB,OAAO,MAAM,KAAK,UAAU,CAAC,CAAC;AACtE,WAAO,EAAE,KAAK,MAAM;AAAA,EACtB,CAAC;AAGD,MAAI,KAAK,gBAAgB,OAAO,MAAM;AACpC,UAAM,MAAM,MAAM,EAAE,IAAI,KAAK,EAAE,MAAM,MAAM,IAAI;AAC/C,QAAI,OAAO,MAAM;AACf,YAAM,IAAI,UAAU,oBAAoB,8BAA8B,GAAG;AAAA,IAC3E;AACA,UAAM,SAAS,yBAAyB,UAAU,GAAG;AACrD,QAAI,CAAC,OAAO,SAAS;AACnB,YAAM,IAAI;AAAA,QACR;AAAA,QACA;AAAA,QACA;AAAA,QACA,OAAO,MAAM;AAAA,MACf;AAAA,IACF;AACA,UAAM,SAAS,MAAM,qBAAqB,OAAO,MAAM,KAAK,UAAU,CAAC,CAAC;AACxE,WAAO,EAAE,KAAK,MAAM;AAAA,EACtB,CAAC;AAGD,MAAI,IAAI,iBAAiB,CAAC,MAAM,EAAE,KAAK,aAAa,cAAc,EAAE,OAAO,CAAC,CAAC;AAE7E,SAAO;AACT;AASO,SAAS,YAAY,OAAqB,CAAC,GAAe;AAC/D,QAAM,MAAM,UAAU,IAAI;AAC1B,QAAM,OAAO,KAAK,QAAQ;AAC1B,QAAM,OAAO,KAAK,QAAQ;AAC1B,SAAO,MAAM,EAAE,OAAO,IAAI,OAAO,MAAM,UAAU,KAAK,GAAG,CAAC,EAAE,SAAS,MAAM,WAAW,MAAM;AAE1F,YAAQ,IAAI,kCAAkC,OAAO,IAAI,UAAU,EAAE;AAAA,EACvE,CAAC;AACH;","names":[]}
1
+ {"version":3,"sources":["../src/wire/schemas.ts","../src/wire/rubrics.ts","../src/wire/handlers.ts","../src/wire/openapi.ts","../src/wire/rpc.ts","../src/wire/server.ts"],"sourcesContent":["/**\n * Wire-protocol schemas.\n *\n * These Zod schemas are the contract between the agent-eval runtime and\n * any non-TypeScript client (Python, Rust, Go, …). They get rendered to\n * OpenAPI by `wire/openapi.ts` and code-generators consume that spec to\n * produce typed clients in other languages.\n *\n * Rule: if it's not in this file, it isn't on the wire. Keep names and\n * shapes self-explanatory — every field has a `.describe()` so the\n * generated docs are useful without reading the source.\n */\nimport { extendZodWithOpenApi } from '@asteasolutions/zod-to-openapi'\nimport { z } from 'zod'\n\nextendZodWithOpenApi(z)\n\n// ── Building blocks ─────────────────────────────────────────────────\n\nexport const RubricDimensionSchema = z\n .object({\n id: z\n .string()\n .min(1)\n .describe('Short stable id like \"buyer_quality\" — used as the key in scoring output.'),\n description: z\n .string()\n .min(1)\n .describe('One-line plain-English meaning. Read by humans reviewing low scores.'),\n weight: z\n .number()\n .min(0)\n .default(1)\n .describe('Relative weight in the composite score. Default 1; 0 disables.'),\n min: z.number().default(0).describe('Lower bound of valid score for this dimension.'),\n max: z.number().default(1).describe('Upper bound of valid score for this dimension.'),\n })\n .openapi('RubricDimension')\n\nexport const FailureModeSchema = z\n .object({\n id: z.string().min(1).describe('Short stable id like \"ai-cadence\" — used in detection lists.'),\n description: z.string().min(1).describe('Plain-English description of the failure pattern.'),\n })\n .openapi('FailureMode')\n\n// ── Rubric ──────────────────────────────────────────────────────────\n\nexport const RubricSchema = z\n .object({\n name: z\n .string()\n .min(1)\n .describe('Stable name like \"anti-slop\" — used by clients to invoke this rubric.'),\n description: z\n .string()\n .min(1)\n .describe('What this rubric measures. Shown in /v1/rubrics listing.'),\n systemPrompt: z\n .string()\n .min(1)\n .describe(\n 'Instructs the judging LLM. Should explain the persona (e.g. \"senior engineer reviewing voice\"), what to score on, and what to return.',\n ),\n dimensions: z\n .array(RubricDimensionSchema)\n .min(1)\n .describe('Scoring axes. The composite score is a weighted sum of these.'),\n failureModes: z\n .array(FailureModeSchema)\n .default([])\n .describe('Patterns to detect; each detected mode appears in the result.failureModes list.'),\n wins: z\n .array(FailureModeSchema)\n .default([])\n .describe('Positive patterns; each detected one appears in the result.wins list.'),\n })\n .openapi('Rubric')\n\n// ── Judge call ──────────────────────────────────────────────────────\n\nexport const JudgeRequestSchema = z\n .object({\n rubricName: z\n .string()\n .optional()\n .describe('Use a built-in rubric by name. Mutually exclusive with `rubric`.'),\n rubric: RubricSchema.optional().describe(\n 'Inline rubric definition. Mutually exclusive with `rubricName`.',\n ),\n content: z\n .string()\n .min(1)\n .describe('The text being judged — a tweet, a blog post, a code snippet, anything stringly.'),\n context: z\n .record(z.string(), z.unknown())\n .optional()\n .describe(\n 'Free-form metadata for the rubric to use — analytics, source URL, author, etc. Surfaced to the LLM.',\n ),\n model: z\n .string()\n .optional()\n .describe('Override the judge model (default routes via tcloud). e.g. \"claude-opus-4-7\".'),\n })\n .refine((v) => Boolean(v.rubricName) !== Boolean(v.rubric), {\n message: 'Provide exactly one of `rubricName` or `rubric`.',\n })\n .openapi('JudgeRequest')\n\nexport const JudgeResultSchema = z\n .object({\n composite: z\n .number()\n .min(0)\n .max(1)\n .describe('Weighted combination of dimension scores in 0..1. The single number to gate on.'),\n dimensions: z\n .record(z.string(), z.number())\n .describe('Per-dimension score, keyed by RubricDimension.id.'),\n failureModes: z\n .array(z.string())\n .default([])\n .describe('Failure-mode ids detected in the content (subset of rubric.failureModes ids).'),\n wins: z\n .array(z.string())\n .default([])\n .describe('Win ids detected in the content (subset of rubric.wins ids).'),\n rationale: z\n .string()\n .describe('Plain-English explanation of the score. Surfaced to the human reviewer.'),\n rubricVersion: z\n .string()\n .describe(\n 'Stable hash of the rubric used. Scores are only comparable across runs when this matches.',\n ),\n model: z.string().describe('Model that produced the judgement, for reproducibility.'),\n durationMs: z.number().int().nonnegative().describe('End-to-end wall time for this call.'),\n })\n .openapi('JudgeResult')\n\n// ── Rubric listing ──────────────────────────────────────────────────\n\nexport const RubricInfoSchema = z\n .object({\n name: z.string().describe('Pass this to /v1/judge as `rubricName`.'),\n description: z.string().describe('What this rubric measures.'),\n dimensions: z\n .array(z.object({ id: z.string(), description: z.string(), weight: z.number() }))\n .describe('The scoring axes this rubric uses, with weights.'),\n failureModes: z.array(z.string()).default([]).describe('Failure-mode ids this rubric detects.'),\n rubricVersion: z.string().describe('Stable hash — match this to compare scores across runs.'),\n })\n .openapi('RubricInfo')\n\nexport const ListRubricsResponseSchema = z\n .object({\n rubrics: z.array(RubricInfoSchema),\n })\n .openapi('ListRubricsResponse')\n\n// ── Version / health ────────────────────────────────────────────────\n\nexport const VersionResponseSchema = z\n .object({\n package: z.string().describe('Package name (always \"@tangle-network/agent-eval\").'),\n version: z.string().describe('Semver of the running server. Match your client to this.'),\n wireVersion: z\n .string()\n .describe(\n 'Wire-protocol semver. Bumps separately from package version when the schema changes.',\n ),\n apiSurface: z.array(z.string()).describe('List of supported method names.'),\n })\n .openapi('VersionResponse')\n\nexport const HealthResponseSchema = z\n .object({\n status: z.literal('ok'),\n uptimeSec: z.number(),\n })\n .openapi('HealthResponse')\n\n// ── Ingestion: production traces + user feedback ────────────────────\n\n/**\n * Minimal `TraceEvent` shape that the production runtime emits.\n * Matches `trace/schema.ts` `TraceEvent` but is duplicated here as a\n * wire schema so non-TypeScript clients can validate without depending\n * on internal types.\n */\nexport const TraceEventSchema = z\n .object({\n eventId: z.string().min(1).describe('Stable id for the event. Use ULID or UUID.'),\n runId: z.string().min(1).describe('Run this event belongs to.'),\n spanId: z.string().optional().describe('Span that emitted the event, if any.'),\n kind: z\n .enum([\n 'log',\n 'error',\n 'budget_decrement',\n 'budget_breach',\n 'state_mutation',\n 'policy_violation',\n 'redaction_applied',\n 'custom',\n ])\n .describe('Coarse event category — matches the TraceSchema v1 EventKind enum.'),\n timestamp: z\n .number()\n .int()\n .nonnegative()\n .describe('Unix millis. Must be monotonically non-decreasing within a span.'),\n payload: z\n .record(z.string(), z.unknown())\n .describe('Free-form payload — the runtime owns the shape.'),\n })\n .openapi('TraceEvent')\n\nexport const TracesIngestRequestSchema = z\n .object({\n events: z\n .array(TraceEventSchema)\n .min(1)\n .max(10_000)\n .describe('Batch of events. Max 10k per call — bigger streams should be chunked.'),\n })\n .openapi('TracesIngestRequest')\n\nexport const TracesIngestResponseSchema = z\n .object({\n accepted: z.number().int().nonnegative().describe('Number of events persisted.'),\n rejected: z\n .number()\n .int()\n .nonnegative()\n .describe('Number of events the store refused — see `errors[]` for reasons.'),\n errors: z\n .array(\n z.object({\n eventId: z.string().describe('Event id this error applies to.'),\n message: z.string().describe('Why the event was rejected.'),\n }),\n )\n .default([]),\n })\n .openapi('TracesIngestResponse')\n\nexport const FeedbackLabelSchema = z\n .object({\n id: z.string().optional(),\n source: z.enum(['user', 'judge', 'environment', 'metric', 'policy', 'system']),\n kind: z.enum([\n 'approve',\n 'reject',\n 'select',\n 'edit',\n 'rank',\n 'rate',\n 'comment',\n 'metric_outcome',\n 'policy_block',\n 'revision_request',\n ]),\n value: z.unknown(),\n reason: z.string().optional(),\n severity: z.enum(['info', 'warning', 'error', 'critical']).optional(),\n createdAt: z.string().describe('ISO-8601 UTC.'),\n metadata: z.record(z.string(), z.unknown()).optional(),\n })\n .openapi('FeedbackLabel')\n\nexport const FeedbackAttemptSchema = z\n .object({\n id: z.string().min(1),\n stepIndex: z.number().int().nonnegative(),\n artifactType: z.enum([\n 'text',\n 'code',\n 'plan',\n 'research',\n 'action',\n 'ui',\n 'decision',\n 'data',\n 'other',\n ]),\n artifact: z.unknown(),\n options: z.array(z.unknown()).optional(),\n proposedAction: z\n .object({\n type: z.string(),\n risk: z.enum(['low', 'medium', 'high']).optional(),\n costUsd: z.number().optional(),\n externalSideEffect: z.boolean().optional(),\n requiresApproval: z.boolean().optional(),\n metadata: z.record(z.string(), z.unknown()).optional(),\n })\n .optional(),\n feedback: z.array(FeedbackLabelSchema).optional(),\n createdAt: z.string(),\n metadata: z.record(z.string(), z.unknown()).optional(),\n })\n .openapi('FeedbackAttempt')\n\nexport const FeedbackTrajectorySchema = z\n .object({\n id: z.string().min(1).describe('Stable id; idempotency key for the trajectory.'),\n projectId: z.string().optional(),\n scenarioId: z.string().optional(),\n task: z.object({\n intent: z.string().min(1),\n context: z.unknown().optional(),\n }),\n attempts: z.array(FeedbackAttemptSchema).default([]),\n labels: z.array(FeedbackLabelSchema).default([]),\n outcome: z\n .object({\n success: z.boolean().optional(),\n score: z.number().optional(),\n metrics: z.record(z.string(), z.number()).optional(),\n costUsd: z.number().optional(),\n detail: z.string().optional(),\n observedAt: z.string().optional(),\n metadata: z.record(z.string(), z.unknown()).optional(),\n })\n .optional(),\n split: z.enum(['train', 'dev', 'test', 'holdout']).optional(),\n tags: z.record(z.string(), z.string()).optional(),\n createdAt: z.string().describe('ISO-8601 UTC.'),\n updatedAt: z.string().optional(),\n metadata: z.record(z.string(), z.unknown()).optional(),\n })\n .openapi('FeedbackTrajectory')\n\nexport const FeedbackIngestResponseSchema = z\n .object({\n id: z.string().describe('Trajectory id that was persisted.'),\n persisted: z.boolean().describe('True when the trajectory was saved (idempotent on id).'),\n })\n .openapi('FeedbackIngestResponse')\n\nexport type TraceEvent = z.infer<typeof TraceEventSchema>\nexport type TracesIngestRequest = z.infer<typeof TracesIngestRequestSchema>\nexport type TracesIngestResponse = z.infer<typeof TracesIngestResponseSchema>\nexport type FeedbackTrajectory = z.infer<typeof FeedbackTrajectorySchema>\nexport type FeedbackIngestResponse = z.infer<typeof FeedbackIngestResponseSchema>\n\n// ── Errors ──────────────────────────────────────────────────────────\n\nexport const ErrorResponseSchema = z\n .object({\n error: z\n .object({\n code: z\n .string()\n .describe(\n 'Machine-readable code: \"validation_error\", \"rubric_not_found\", \"judge_error\".',\n ),\n message: z.string().describe('Human-readable message.'),\n details: z.unknown().optional().describe('Optional structured detail.'),\n })\n .describe('Errors are always wrapped in this shape across all endpoints.'),\n })\n .openapi('ErrorResponse')\n\n// ── Type exports for callers in the same package ────────────────────\n\nexport type RubricDimension = z.infer<typeof RubricDimensionSchema>\nexport type FailureMode = z.infer<typeof FailureModeSchema>\nexport type Rubric = z.infer<typeof RubricSchema>\nexport type JudgeRequest = z.infer<typeof JudgeRequestSchema>\nexport type JudgeResult = z.infer<typeof JudgeResultSchema>\nexport type RubricInfo = z.infer<typeof RubricInfoSchema>\nexport type ListRubricsResponse = z.infer<typeof ListRubricsResponseSchema>\nexport type VersionResponse = z.infer<typeof VersionResponseSchema>\nexport type ErrorResponse = z.infer<typeof ErrorResponseSchema>\n\n// ── Wire-protocol version ───────────────────────────────────────────\n\n/**\n * Bump on any breaking change to a request/response schema.\n * Non-breaking (additive) changes don't require a bump.\n */\nexport const WIRE_VERSION = '1.0.0'\n\n/**\n * Stable hash of a rubric. Used to make scores comparable across runs:\n * if the rubricVersion matches, the rubric was identical.\n */\nexport function hashRubric(rubric: Rubric): string {\n const stable = stableStringify(rubric)\n let h = 5381\n for (let i = 0; i < stable.length; i++) {\n h = (h * 33) ^ stable.charCodeAt(i)\n }\n // Unsigned 32-bit hex, prefixed with rubric name + version slot\n return `${rubric.name}@${(h >>> 0).toString(16).padStart(8, '0')}`\n}\n\nfunction stableStringify(value: unknown): string {\n if (Array.isArray(value)) return `[${value.map((item) => stableStringify(item)).join(',')}]`\n if (value && typeof value === 'object') {\n const entries = Object.entries(value as Record<string, unknown>)\n .sort(([a], [b]) => a.localeCompare(b))\n .map(([key, item]) => `${JSON.stringify(key)}:${stableStringify(item)}`)\n return `{${entries.join(',')}}`\n }\n return JSON.stringify(value)\n}\n","/**\n * Built-in rubrics shipped with agent-eval.\n *\n * A rubric is a set of scoring axes plus a system prompt that tells the\n * judging LLM how to grade against those axes. Built-in rubrics are\n * curated for use cases that recur across Tangle projects — call them\n * by name from any client.\n *\n * Adding a rubric:\n * 1. Define the Rubric object below with a clear `description` and\n * named `dimensions`.\n * 2. Register it in `BUILTIN_RUBRICS` at the bottom.\n * 3. Add a test in `tests/wire/rubrics.test.ts`.\n *\n * Custom rubrics: callers pass `rubric` inline to /v1/judge instead of\n * `rubricName` — see schemas.ts.\n */\nimport type { Rubric } from './schemas'\nimport { hashRubric } from './schemas'\n\n// ── anti-slop ───────────────────────────────────────────────────────\n// Voice/style judge tuned for technical-buyer audiences. Used by the\n// Postiz autoresearch loop and any content-quality gate.\n\nconst ANTI_SLOP: Rubric = {\n name: 'anti-slop',\n description:\n 'Voice and signal quality for content aimed at senior engineers. Catches AI cadence, marketing tone, and engagement-bait shapes.',\n systemPrompt: `You are evaluating a piece of content written for senior engineers and technical founders.\n\nYou score three things:\n- buyer_quality (0..1): would a senior engineer in the target ICP find this worth their attention? High = specific, earned, technically interesting. Low = generic, hyped, off-target.\n- voice (0..1): does it read like a person who built the thing, or like AI/marketing copy?\n- signal (0..1): does it contain a non-obvious detail, constraint, or claim a reader couldn't get from the public docs?\n\nDetect failure modes (return ids matching):\n- ai-cadence: rule-of-three openings, em-dash flourish, \"Let me explain\", \"Here's the thing\", AI rhythm\n- marketing-tone: \"We're excited to announce\", \"thrilled\", \"delighted\", \"game-changer\", buzzword stack\n- vague-claim: technical claim without a specific component, file, or measurement\n- no-hook: opening doesn't earn attention from the target reader\n- engagement-bait: \"agree?\", \"thoughts?\", listicles, controversy-fishing, hook-detail-pitch\n- off-icp: content shape would attract motivational/grift/hype audiences instead of buyers\n- stale-claim: repeats a positioning line we've used many times this month\n\nDetect wins (return ids matching):\n- specific-component: names a real file, component, or measurement\n- earned-detail: shares a non-obvious detail not derivable from public docs\n- constraint-articulated: names a real tradeoff and the side chosen\n- honest-failure: describes a real failure mode and what was done about it\n\nReturn ONLY JSON matching the response schema. Be conservative — most content has 0-1 wins and 1-2 failure modes, not many of each.`,\n dimensions: [\n {\n id: 'buyer_quality',\n description: 'Would the target buyer find this worth attention?',\n weight: 0.5,\n min: 0,\n max: 1,\n },\n {\n id: 'voice',\n description: 'Does it sound like a builder, not AI or marketing?',\n weight: 0.3,\n min: 0,\n max: 1,\n },\n {\n id: 'signal',\n description: 'Non-obvious detail, constraint, or claim?',\n weight: 0.2,\n min: 0,\n max: 1,\n },\n ],\n failureModes: [\n { id: 'ai-cadence', description: 'AI-rhythm openings and transitions' },\n { id: 'marketing-tone', description: 'Buzzwords, hype, corporate-PR voice' },\n { id: 'vague-claim', description: 'Technical claim without specifics' },\n { id: 'no-hook', description: 'Opening fails to earn attention' },\n { id: 'engagement-bait', description: 'Listicle/controversy/agree-pattern' },\n { id: 'off-icp', description: 'Voice attracts the wrong audience' },\n { id: 'stale-claim', description: 'Reuses an over-used positioning line' },\n ],\n wins: [\n { id: 'specific-component', description: 'Names a real file/component/number' },\n { id: 'earned-detail', description: 'Detail not in public docs' },\n { id: 'constraint-articulated', description: 'Names a real tradeoff' },\n { id: 'honest-failure', description: 'Describes a real failure honestly' },\n ],\n}\n\n// ── Registry ────────────────────────────────────────────────────────\n\nexport const BUILTIN_RUBRICS: Record<string, Rubric> = {\n 'anti-slop': ANTI_SLOP,\n}\n\n/** Get a built-in rubric by name, or undefined. */\nexport function getBuiltinRubric(name: string): Rubric | undefined {\n return BUILTIN_RUBRICS[name]\n}\n\n/** List built-in rubrics with their stable versions. */\nexport function listBuiltinRubrics() {\n return Object.values(BUILTIN_RUBRICS).map((r) => ({\n name: r.name,\n description: r.description,\n dimensions: r.dimensions.map((d) => ({\n id: d.id,\n description: d.description,\n weight: d.weight,\n })),\n failureModes: r.failureModes.map((f) => f.id),\n rubricVersion: hashRubric(r),\n }))\n}\n","/**\n * Pure handler functions — the \"business logic\" behind every wire-protocol\n * method. The HTTP server (`server.ts`) and the stdio RPC (`rpc.ts`) both\n * call these. Tests call these directly without spinning a server.\n *\n * Each handler:\n * - Takes a parsed request (already Zod-validated by the transport).\n * - Returns a result that matches the response schema.\n * - Throws `WireError` for caller-fixable errors (404, 400, 422).\n * - Lets unexpected errors bubble — the transport maps them to 500.\n */\nimport type { FeedbackTrajectoryStore } from '../feedback-trajectory'\nimport { callLlmJson } from '../llm-client'\nimport type { TraceEvent as InternalTraceEvent } from '../trace/schema'\nimport type { TraceStore } from '../trace/store'\nimport { getBuiltinRubric, listBuiltinRubrics } from './rubrics'\nimport {\n type FeedbackIngestResponse,\n hashRubric,\n type JudgeRequest,\n type JudgeResult,\n type ListRubricsResponse,\n type Rubric,\n type TracesIngestRequest,\n type TracesIngestResponse,\n type VersionResponse,\n WIRE_VERSION,\n type FeedbackTrajectory as WireFeedbackTrajectory,\n} from './schemas'\n\n/** Caller-fixable error. The transport renders this to 4xx + ErrorResponse. */\nexport class WireError extends Error {\n constructor(\n public readonly code: string,\n message: string,\n public readonly status: number = 400,\n public readonly details?: unknown,\n ) {\n super(message)\n this.name = 'WireError'\n }\n}\n\n// ── judge ───────────────────────────────────────────────────────────\n\n/** The JSON schema we ask the judging LLM to fill in. */\nfunction judgeOutputSchema(rubric: Rubric) {\n return {\n name: 'JudgeOutput',\n schema: {\n type: 'object',\n additionalProperties: false,\n properties: {\n dimensions: {\n type: 'object',\n additionalProperties: false,\n properties: Object.fromEntries(\n rubric.dimensions.map((d) => [\n d.id,\n { type: 'number', minimum: d.min, maximum: d.max },\n ]),\n ),\n required: rubric.dimensions.map((d) => d.id),\n },\n failureModes: {\n type: 'array',\n items: { type: 'string', enum: rubric.failureModes.map((f) => f.id) },\n },\n wins: {\n type: 'array',\n items: { type: 'string', enum: rubric.wins.map((w) => w.id) },\n },\n rationale: { type: 'string' },\n },\n required: ['dimensions', 'rationale'],\n } as Record<string, unknown>,\n }\n}\n\ninterface JudgeOutput {\n dimensions: Record<string, number>\n failureModes?: string[]\n wins?: string[]\n rationale: string\n}\n\nfunction validateJudgeOutput(value: unknown, rubric: Rubric): JudgeOutput {\n if (!value || typeof value !== 'object') {\n throw new WireError('judge_error', 'Judge returned malformed output.', 500, value)\n }\n const raw = value as Record<string, unknown>\n const rawDimensions = raw.dimensions\n if (!rawDimensions || typeof rawDimensions !== 'object' || Array.isArray(rawDimensions)) {\n throw new WireError('judge_error', 'Judge returned malformed dimensions.', 500, value)\n }\n\n const dimensions: Record<string, number> = {}\n const dimensionRecord = rawDimensions as Record<string, unknown>\n for (const dim of rubric.dimensions) {\n const score = dimensionRecord[dim.id]\n if (\n typeof score !== 'number' ||\n !Number.isFinite(score) ||\n score < dim.min ||\n score > dim.max\n ) {\n throw new WireError(\n 'judge_error',\n `Judge returned invalid score for dimension \"${dim.id}\".`,\n 500,\n value,\n )\n }\n dimensions[dim.id] = score\n }\n\n const allowedFailures = new Set(rubric.failureModes.map((mode) => mode.id))\n const allowedWins = new Set(rubric.wins.map((win) => win.id))\n const failureModes = validateIdArray(raw.failureModes, allowedFailures, 'failureModes', value)\n const wins = validateIdArray(raw.wins, allowedWins, 'wins', value)\n if (typeof raw.rationale !== 'string' || raw.rationale.trim().length === 0) {\n throw new WireError('judge_error', 'Judge returned missing rationale.', 500, value)\n }\n\n return { dimensions, failureModes, wins, rationale: raw.rationale }\n}\n\nfunction validateIdArray(\n raw: unknown,\n allowed: Set<string>,\n field: 'failureModes' | 'wins',\n original: unknown,\n): string[] {\n if (raw === undefined) return []\n if (!Array.isArray(raw)) {\n throw new WireError('judge_error', `Judge returned non-array ${field}.`, 500, original)\n }\n const out: string[] = []\n for (const item of raw) {\n if (typeof item !== 'string' || !allowed.has(item)) {\n throw new WireError(\n 'judge_error',\n `Judge returned unknown ${field} id \"${String(item)}\".`,\n 500,\n original,\n )\n }\n out.push(item)\n }\n return out\n}\n\nfunction compositeScore(dimensions: Record<string, number>, rubric: Rubric): number {\n let weighted = 0\n let totalWeight = 0\n for (const dim of rubric.dimensions) {\n const raw = dimensions[dim.id] ?? 0\n const range = dim.max - dim.min || 1\n const normalized = Math.max(0, Math.min(1, (raw - dim.min) / range))\n weighted += normalized * dim.weight\n totalWeight += dim.weight\n }\n return totalWeight > 0 ? weighted / totalWeight : 0\n}\n\nfunction buildJudgePrompt(content: string, context: unknown): string {\n const ctx = context && Object.keys(context as object).length ? JSON.stringify(context) : ''\n return [\n `CONTENT TO JUDGE:`,\n content,\n '',\n ctx ? `CONTEXT (metadata, analytics, etc.):` : '',\n ctx ? ctx : '',\n ]\n .filter(Boolean)\n .join('\\n')\n}\n\nconst DEFAULT_JUDGE_MODEL = 'claude-sonnet-4-6'\n\nexport async function handleJudge(req: JudgeRequest): Promise<JudgeResult> {\n // Resolve rubric\n let rubric: Rubric\n if (req.rubricName) {\n const found = getBuiltinRubric(req.rubricName)\n if (!found) {\n throw new WireError('rubric_not_found', `No built-in rubric named \"${req.rubricName}\".`, 404)\n }\n rubric = found\n } else if (req.rubric) {\n rubric = req.rubric\n } else {\n // refine() in the schema should already have caught this — defense in depth\n throw new WireError('validation_error', 'Provide either `rubricName` or `rubric`.', 422)\n }\n\n const startedAt = Date.now()\n const model = req.model ?? DEFAULT_JUDGE_MODEL\n\n const { value, result } = await callLlmJson<JudgeOutput>({\n model,\n messages: [\n { role: 'system', content: rubric.systemPrompt },\n { role: 'user', content: buildJudgePrompt(req.content, req.context) },\n ],\n jsonSchema: judgeOutputSchema(rubric),\n temperature: 0.0,\n timeoutMs: 60_000,\n })\n\n const output = validateJudgeOutput(value, rubric)\n\n const composite = compositeScore(output.dimensions, rubric)\n const durationMs = Date.now() - startedAt\n\n return {\n composite,\n dimensions: output.dimensions,\n failureModes: output.failureModes ?? [],\n wins: output.wins ?? [],\n rationale: output.rationale,\n rubricVersion: hashRubric(rubric),\n model: result.model,\n durationMs,\n }\n}\n\n// ── listRubrics ─────────────────────────────────────────────────────\n\nexport function handleListRubrics(): ListRubricsResponse {\n return { rubrics: listBuiltinRubrics() }\n}\n\n// ── version ─────────────────────────────────────────────────────────\n\nimport { readFileSync } from 'node:fs'\nimport { dirname, resolve } from 'node:path'\nimport { fileURLToPath } from 'node:url'\n\nlet CACHED_VERSION: string | undefined\n\nfunction readPackageVersion(): string {\n if (CACHED_VERSION) return CACHED_VERSION\n // Walk up from this file looking for the nearest package.json.\n // In dist/ this is dist/.., in src/wire/ this is ../../package.json.\n const here = dirname(fileURLToPath(import.meta.url))\n const candidates = [\n resolve(here, '..', '..', 'package.json'), // src/wire → repo root\n resolve(here, '..', 'package.json'), // dist → repo root\n ]\n for (const path of candidates) {\n try {\n const pkg = JSON.parse(readFileSync(path, 'utf-8')) as { version?: string }\n if (pkg.version) {\n CACHED_VERSION = pkg.version\n return pkg.version\n }\n } catch {\n // try next\n }\n }\n return '0.0.0-unknown'\n}\n\nexport function handleVersion(): VersionResponse {\n return {\n package: '@tangle-network/agent-eval',\n version: readPackageVersion(),\n wireVersion: WIRE_VERSION,\n apiSurface: ['judge', 'listRubrics', 'version', 'feedback.ingest', 'traces.ingest'],\n }\n}\n\n// ── Ingestion handlers ───────────────────────────────────────────────\n\n/**\n * Pluggable stores the wire layer routes ingestion writes into. Both\n * are optional — when omitted, the corresponding endpoint returns 503.\n *\n * Production deployments wire a `FileSystemTraceStore` and\n * `FileSystemFeedbackTrajectoryStore` here. Tests substitute in-memory\n * stores.\n */\nexport interface IngestionStores {\n traceStore?: TraceStore\n feedbackStore?: FeedbackTrajectoryStore\n}\n\n/**\n * `POST /v1/traces/ingest` — accept a batch of `TraceEvent`s from the\n * production runtime. Best-effort: each event is appended independently;\n * one bad event does not poison the batch.\n *\n * Idempotency: the underlying store is append-only; consumers retrying\n * the same payload will get duplicate events. Consumers should\n * de-duplicate by `eventId` downstream — production traces frequently\n * land via at-least-once buses (Kafka, SQS) where dedup is unavoidable.\n */\nexport async function handleTracesIngest(\n req: TracesIngestRequest,\n stores: IngestionStores,\n): Promise<TracesIngestResponse> {\n if (!stores.traceStore) {\n throw new WireError(\n 'service_unavailable',\n 'No trace store configured on this server. Pass `traceStore` to `createApp`.',\n 503,\n )\n }\n const errors: Array<{ eventId: string; message: string }> = []\n let accepted = 0\n for (const event of req.events) {\n try {\n // The wire `TraceEvent` is structurally identical to the internal one.\n await stores.traceStore.appendEvent(event as InternalTraceEvent)\n accepted++\n } catch (err) {\n errors.push({\n eventId: event.eventId,\n message: err instanceof Error ? err.message : String(err),\n })\n }\n }\n return { accepted, rejected: errors.length, errors }\n}\n\n/**\n * `POST /v1/feedback` — accept a single `FeedbackTrajectory` from the\n * production runtime. Idempotent on `id`: re-posting the same trajectory\n * replaces the prior record.\n */\nexport async function handleFeedbackIngest(\n req: WireFeedbackTrajectory,\n stores: IngestionStores,\n): Promise<FeedbackIngestResponse> {\n if (!stores.feedbackStore) {\n throw new WireError(\n 'service_unavailable',\n 'No feedback store configured on this server. Pass `feedbackStore` to `createApp`.',\n 503,\n )\n }\n // The wire `FeedbackTrajectory` aligns 1:1 with the internal type;\n // cast through `unknown` since the wire schema is a Zod-inferred\n // structural type with optional fields the internal store consumes.\n await stores.feedbackStore.save(req as unknown as Parameters<FeedbackTrajectoryStore['save']>[0])\n return { id: req.id, persisted: true }\n}\n","/**\n * Build an OpenAPI spec from the wire schemas.\n *\n * The spec is the contract that other-language clients (Python, Rust,\n * Go, …) generate from. There is no hand-written client — clients are\n * derived artifacts of this file plus `schemas.ts`.\n *\n * Run `pnpm openapi` (defined in package.json) to write the spec to\n * `dist/openapi.json`. CI uses that file to regenerate the Python\n * client and gate the dual-publish workflow.\n */\nimport { OpenAPIRegistry, OpenApiGeneratorV31 } from '@asteasolutions/zod-to-openapi'\nimport type { OpenAPIObject } from 'openapi3-ts/oas31'\n\nimport {\n ErrorResponseSchema,\n FeedbackIngestResponseSchema,\n FeedbackTrajectorySchema,\n HealthResponseSchema,\n JudgeRequestSchema,\n JudgeResultSchema,\n ListRubricsResponseSchema,\n TracesIngestRequestSchema,\n TracesIngestResponseSchema,\n VersionResponseSchema,\n WIRE_VERSION,\n} from './schemas'\n\nexport function buildOpenApi(packageVersion: string): OpenAPIObject {\n const registry = new OpenAPIRegistry()\n\n // Components — each schema becomes a $ref-able component\n registry.register('JudgeRequest', JudgeRequestSchema)\n registry.register('JudgeResult', JudgeResultSchema)\n registry.register('ListRubricsResponse', ListRubricsResponseSchema)\n registry.register('VersionResponse', VersionResponseSchema)\n registry.register('HealthResponse', HealthResponseSchema)\n registry.register('ErrorResponse', ErrorResponseSchema)\n registry.register('TracesIngestRequest', TracesIngestRequestSchema)\n registry.register('TracesIngestResponse', TracesIngestResponseSchema)\n registry.register('FeedbackTrajectory', FeedbackTrajectorySchema)\n registry.register('FeedbackIngestResponse', FeedbackIngestResponseSchema)\n\n // Routes\n registry.registerPath({\n method: 'post',\n path: '/v1/judge',\n summary: 'Score a piece of content against a rubric',\n description:\n 'Runs the judging LLM with the named (or inline) rubric and returns dimension scores, detected failure modes, wins, and a composite score in 0..1.',\n request: {\n body: {\n content: {\n 'application/json': { schema: JudgeRequestSchema },\n },\n },\n },\n responses: {\n 200: {\n description: 'Successful judgement',\n content: { 'application/json': { schema: JudgeResultSchema } },\n },\n 400: {\n description: 'Validation error',\n content: { 'application/json': { schema: ErrorResponseSchema } },\n },\n 404: {\n description: 'Rubric not found',\n content: { 'application/json': { schema: ErrorResponseSchema } },\n },\n 500: {\n description: 'Judge error',\n content: { 'application/json': { schema: ErrorResponseSchema } },\n },\n },\n })\n\n registry.registerPath({\n method: 'get',\n path: '/v1/rubrics',\n summary: 'List built-in rubrics',\n description:\n 'Returns every rubric registered server-side, with their dimensions and stable rubricVersion hash.',\n responses: {\n 200: {\n description: 'Listing',\n content: { 'application/json': { schema: ListRubricsResponseSchema } },\n },\n },\n })\n\n registry.registerPath({\n method: 'get',\n path: '/v1/version',\n summary: 'Server and wire-protocol version',\n description: 'Match your client version to `version`; check `wireVersion` for compatibility.',\n responses: {\n 200: {\n description: 'Version info',\n content: { 'application/json': { schema: VersionResponseSchema } },\n },\n },\n })\n\n registry.registerPath({\n method: 'get',\n path: '/healthz',\n summary: 'Liveness check',\n responses: {\n 200: {\n description: 'OK',\n content: { 'application/json': { schema: HealthResponseSchema } },\n },\n },\n })\n\n registry.registerPath({\n method: 'post',\n path: '/v1/traces/ingest',\n summary: 'Ingest a batch of production TraceEvents',\n description:\n 'Append a batch of TraceEvents to the configured TraceStore. Accepts application/json ({events:[...]}) or application/x-ndjson (one event per line). Returns counts of accepted + rejected events.',\n request: {\n body: {\n content: {\n 'application/json': { schema: TracesIngestRequestSchema },\n 'application/x-ndjson': { schema: TracesIngestRequestSchema },\n },\n },\n },\n responses: {\n 200: {\n description: 'Ingestion summary',\n content: { 'application/json': { schema: TracesIngestResponseSchema } },\n },\n 400: {\n description: 'Validation error',\n content: { 'application/json': { schema: ErrorResponseSchema } },\n },\n 401: {\n description: 'Unauthorized (when bearer auth is configured)',\n content: { 'application/json': { schema: ErrorResponseSchema } },\n },\n 503: {\n description: 'No trace store configured',\n content: { 'application/json': { schema: ErrorResponseSchema } },\n },\n },\n })\n\n registry.registerPath({\n method: 'post',\n path: '/v1/feedback',\n summary: 'Ingest a FeedbackTrajectory from production',\n description:\n 'Persist a single FeedbackTrajectory. Idempotent on trajectory.id — re-posting replaces the prior record. Used by production runtimes to forward user 👍/👎/edits into the eval substrate.',\n request: {\n body: {\n content: {\n 'application/json': { schema: FeedbackTrajectorySchema },\n },\n },\n },\n responses: {\n 200: {\n description: 'Persisted',\n content: { 'application/json': { schema: FeedbackIngestResponseSchema } },\n },\n 400: {\n description: 'Validation error',\n content: { 'application/json': { schema: ErrorResponseSchema } },\n },\n 401: {\n description: 'Unauthorized (when bearer auth is configured)',\n content: { 'application/json': { schema: ErrorResponseSchema } },\n },\n 503: {\n description: 'No feedback store configured',\n content: { 'application/json': { schema: ErrorResponseSchema } },\n },\n },\n })\n\n const generator = new OpenApiGeneratorV31(registry.definitions)\n const doc = generator.generateDocument({\n openapi: '3.1.0',\n info: {\n title: '@tangle-network/agent-eval — wire protocol',\n version: packageVersion,\n description: `HTTP and stdio RPC interface to agent-eval. The TypeScript runtime is the source of truth; this spec is the contract that cross-language clients (Python, Rust, Go) generate from.\n\nWire-protocol version: ${WIRE_VERSION}. Bumps on breaking changes to request/response schemas.`,\n contact: { name: 'Tangle Network', url: 'https://github.com/tangle-network/agent-eval' },\n license: { name: 'MIT' },\n },\n servers: [{ url: 'http://localhost:5005', description: 'Local agent-eval serve' }],\n })\n const rubricRef = { $ref: '#/components/schemas/Rubric' } as const\n const commonJudgeFields = {\n content: { type: 'string', minLength: 1 },\n context: { type: 'object', additionalProperties: true },\n model: { type: 'string' },\n } as const\n doc.components ??= {}\n doc.components.schemas ??= {}\n doc.components.schemas.JudgeRequest = {\n oneOf: [\n {\n type: 'object',\n additionalProperties: false,\n required: ['rubricName', 'content'],\n properties: {\n rubricName: { type: 'string', minLength: 1 },\n ...commonJudgeFields,\n },\n },\n {\n type: 'object',\n additionalProperties: false,\n required: ['rubric', 'content'],\n properties: {\n rubric: rubricRef,\n ...commonJudgeFields,\n },\n },\n ],\n description: 'Judge request. Provide exactly one of rubricName or rubric.',\n }\n return doc\n}\n","/**\n * stdio RPC transport.\n *\n * For batch / cron use without a running server. The Python client falls\n * back to this when no server is reachable.\n *\n * Protocol (line-delimited JSON over stdin/stdout):\n * IN: one JSON object on stdin: {\"method\":\"judge\",\"params\":{...}}\n * OUT: one JSON object on stdout: {\"result\":{...}} or {\"error\":{...}}\n *\n * One request per process invocation. To pipeline many calls, the client\n * writes JSONL to stdin and reads JSONL from stdout — see batch mode below.\n */\nimport { handleJudge, handleListRubrics, handleVersion, WireError } from './handlers'\nimport { JudgeRequestSchema } from './schemas'\n\ninterface RpcRequest {\n method: 'judge' | 'listRubrics' | 'version'\n params?: unknown\n}\n\ninterface RpcSuccess {\n result: unknown\n}\n\ninterface RpcError {\n error: { code: string; message: string; details?: unknown }\n}\n\nexport async function dispatchRpc(req: RpcRequest): Promise<RpcSuccess | RpcError> {\n try {\n switch (req.method) {\n case 'judge': {\n const parsed = JudgeRequestSchema.safeParse(req.params)\n if (!parsed.success) {\n return {\n error: {\n code: 'validation_error',\n message: 'params did not match JudgeRequest schema.',\n details: parsed.error.issues,\n },\n }\n }\n return { result: await handleJudge(parsed.data) }\n }\n case 'listRubrics':\n return { result: handleListRubrics() }\n case 'version':\n return { result: handleVersion() }\n default:\n return {\n error: {\n code: 'unknown_method',\n message: `No such method: ${(req as { method: string }).method}`,\n },\n }\n }\n } catch (err) {\n if (err instanceof WireError) {\n return { error: { code: err.code, message: err.message, details: err.details } }\n }\n const message = err instanceof Error ? err.message : String(err)\n return { error: { code: 'internal_error', message } }\n }\n}\n\n// ── stdin/stdout driver ─────────────────────────────────────────────\n\nasync function readAll(stream: NodeJS.ReadableStream): Promise<string> {\n const chunks: Buffer[] = []\n for await (const chunk of stream) {\n chunks.push(Buffer.isBuffer(chunk) ? chunk : Buffer.from(chunk as string))\n }\n return Buffer.concat(chunks).toString('utf-8')\n}\n\n/** Read one JSON request from stdin, write one JSON response to stdout. */\nexport async function runRpcOnce(method?: string): Promise<number> {\n const raw = await readAll(process.stdin)\n let req: RpcRequest\n try {\n const body = JSON.parse(raw)\n req = method ? { method: method as RpcRequest['method'], params: body } : (body as RpcRequest)\n } catch (err) {\n process.stdout.write(\n `${JSON.stringify({\n error: {\n code: 'parse_error',\n message: `stdin was not valid JSON: ${err instanceof Error ? err.message : String(err)}`,\n },\n })}\\n`,\n )\n return 1\n }\n const out = await dispatchRpc(req)\n process.stdout.write(`${JSON.stringify(out)}\\n`)\n return 'error' in out ? 1 : 0\n}\n\n/** Read JSONL requests from stdin, write JSONL responses to stdout. */\nexport async function runRpcBatch(method?: string): Promise<number> {\n const raw = await readAll(process.stdin)\n const lines = raw.split('\\n').filter((l) => l.trim().length > 0)\n let exitCode = 0\n for (const line of lines) {\n let req: RpcRequest\n try {\n const body = JSON.parse(line)\n req = method ? { method: method as RpcRequest['method'], params: body } : (body as RpcRequest)\n } catch (err) {\n process.stdout.write(\n `${JSON.stringify({\n error: {\n code: 'parse_error',\n message: `line was not valid JSON: ${err instanceof Error ? err.message : String(err)}`,\n },\n })}\\n`,\n )\n exitCode = 1\n continue\n }\n const out = await dispatchRpc(req)\n process.stdout.write(`${JSON.stringify(out)}\\n`)\n if ('error' in out) exitCode = 1\n }\n return exitCode\n}\n","/**\n * HTTP transport for the wire protocol.\n *\n * Hono + @hono/node-server. Every endpoint:\n * 1. Validates the request against its Zod schema.\n * 2. Calls the matching handler in `handlers.ts`.\n * 3. Renders 4xx for `WireError` with structured body, 500 for unexpected.\n *\n * The server holds optional `IngestionStores` (passed to `createApp`)\n * to receive production traces and user feedback. With no stores wired,\n * the ingestion endpoints return 503 — read endpoints (`/v1/judge`,\n * `/v1/rubrics`, `/v1/version`) remain fully functional.\n *\n * Run via `agent-eval serve --port 5005`.\n */\nimport { type ServerType, serve } from '@hono/node-server'\nimport { Hono } from 'hono'\nimport { cors } from 'hono/cors'\n\nimport {\n handleFeedbackIngest,\n handleJudge,\n handleListRubrics,\n handleTracesIngest,\n handleVersion,\n type IngestionStores,\n WireError,\n} from './handlers'\nimport { buildOpenApi } from './openapi'\nimport { FeedbackTrajectorySchema, JudgeRequestSchema, TracesIngestRequestSchema } from './schemas'\n\nconst STARTED_AT = Date.now()\n\nexport interface CreateAppOptions {\n /** Stores wired to the ingestion endpoints. */\n stores?: IngestionStores\n /**\n * Bearer-token auth. When provided, every endpoint EXCEPT `/healthz`\n * and `/v1/version` requires `Authorization: Bearer <token>`. The\n * token may be a static string OR a function for time-bounded /\n * rotating tokens.\n *\n * Recommended for any server that accepts ingestion writes from the\n * public internet. Read-only deployments may omit it.\n */\n auth?: {\n bearer: string | ((token: string) => boolean | Promise<boolean>)\n }\n}\n\nconst AUTH_EXEMPT_PATHS = new Set(['/healthz', '/v1/version', '/openapi.json'])\n\nexport function createApp(opts: CreateAppOptions = {}) {\n const app = new Hono()\n\n app.use('*', cors())\n\n // Bearer-token middleware (only attached when configured).\n if (opts.auth) {\n const verify = opts.auth.bearer\n app.use('*', async (c, next) => {\n const path = new URL(c.req.url).pathname\n if (AUTH_EXEMPT_PATHS.has(path)) return next()\n const raw = c.req.header('authorization') ?? ''\n const match = raw.match(/^Bearer\\s+(.+)$/i)\n if (!match) {\n throw new WireError('unauthorized', 'Missing or malformed Authorization header.', 401)\n }\n const token = match[1] as string\n const ok = typeof verify === 'string' ? token === verify : await verify(token)\n if (!ok) {\n throw new WireError('unauthorized', 'Invalid bearer token.', 401)\n }\n return next()\n })\n }\n\n app.onError((err, c) => {\n if (err instanceof WireError) {\n const status = err.status as 400 | 401 | 404 | 422 | 500 | 503\n return c.json(\n { error: { code: err.code, message: err.message, details: err.details } },\n status,\n )\n }\n // Unexpected — log and return generic 500 without leaking internals.\n console.error('[agent-eval] unhandled error:', err)\n return c.json({ error: { code: 'internal_error', message: 'Internal server error.' } }, 500)\n })\n\n // ── Health ──\n app.get('/healthz', (c) =>\n c.json({ status: 'ok' as const, uptimeSec: (Date.now() - STARTED_AT) / 1000 }),\n )\n\n // ── Version ──\n app.get('/v1/version', (c) => c.json(handleVersion()))\n\n // ── Rubrics ──\n app.get('/v1/rubrics', (c) => c.json(handleListRubrics()))\n\n // ── Judge ──\n app.post('/v1/judge', async (c) => {\n const raw = await c.req.json().catch(() => null)\n if (raw == null) {\n throw new WireError('validation_error', 'Request body must be JSON.', 400)\n }\n const parsed = JudgeRequestSchema.safeParse(raw)\n if (!parsed.success) {\n throw new WireError(\n 'validation_error',\n 'Request did not match JudgeRequest schema.',\n 400,\n parsed.error.issues,\n )\n }\n const result = await handleJudge(parsed.data)\n return c.json(result)\n })\n\n // ── Traces ingest (NDJSON-friendly: accepts either {events:[...]} or NDJSON) ──\n app.post('/v1/traces/ingest', async (c) => {\n const contentType = c.req.header('content-type') ?? ''\n let payload: unknown\n if (contentType.includes('application/x-ndjson')) {\n const text = await c.req.text()\n const events = text\n .split('\\n')\n .map((line) => line.trim())\n .filter((line) => line.length > 0)\n .map((line) => {\n try {\n return JSON.parse(line)\n } catch {\n throw new WireError(\n 'validation_error',\n 'NDJSON line did not parse as JSON.',\n 400,\n line.slice(0, 200),\n )\n }\n })\n payload = { events }\n } else {\n payload = await c.req.json().catch(() => null)\n }\n if (payload == null) {\n throw new WireError('validation_error', 'Request body must be JSON or NDJSON.', 400)\n }\n const parsed = TracesIngestRequestSchema.safeParse(payload)\n if (!parsed.success) {\n throw new WireError(\n 'validation_error',\n 'Request did not match TracesIngestRequest schema.',\n 400,\n parsed.error.issues,\n )\n }\n const result = await handleTracesIngest(parsed.data, opts.stores ?? {})\n return c.json(result)\n })\n\n // ── Feedback ingest ──\n app.post('/v1/feedback', async (c) => {\n const raw = await c.req.json().catch(() => null)\n if (raw == null) {\n throw new WireError('validation_error', 'Request body must be JSON.', 400)\n }\n const parsed = FeedbackTrajectorySchema.safeParse(raw)\n if (!parsed.success) {\n throw new WireError(\n 'validation_error',\n 'Request did not match FeedbackTrajectory schema.',\n 400,\n parsed.error.issues,\n )\n }\n const result = await handleFeedbackIngest(parsed.data, opts.stores ?? {})\n return c.json(result)\n })\n\n // ── OpenAPI spec ──\n app.get('/openapi.json', (c) => c.json(buildOpenApi(handleVersion().version)))\n\n return app\n}\n\nexport interface ServeOptions extends CreateAppOptions {\n /** Default 5005. */\n port?: number\n /** Default '127.0.0.1'. Set to '0.0.0.0' to listen on all interfaces. */\n host?: string\n}\n\nexport function startServer(opts: ServeOptions = {}): ServerType {\n const app = createApp(opts)\n const port = opts.port ?? 5005\n const host = opts.host ?? '127.0.0.1'\n return serve({ fetch: app.fetch, port, hostname: host }, ({ address, port: actualPort }) => {\n // eslint-disable-next-line no-console\n console.log(`[agent-eval] serving on http://${address}:${actualPort}`)\n })\n}\n\nexport interface StartedServer {\n server: ServerType\n /** The OS-assigned port. When opts.port was 0, this is the actual port the\n * kernel bound — callers that need to dial back (smoke tests, sidecars\n * registering with a parent) read this rather than guessing a free port. */\n port: number\n /** Resolved host the server bound to (defaults to 127.0.0.1). */\n host: string\n /** Close the server. Resolves once active connections have drained. */\n close(): Promise<void>\n}\n\n/**\n * Promise-returning variant of `startServer` that resolves once the server is\n * listening and surfaces the resolved bound port. Use this from smoke tests\n * (`startServerAsync({ port: 0 })`) and any caller that needs to dial back.\n */\nexport function startServerAsync(opts: ServeOptions = {}): Promise<StartedServer> {\n const app = createApp(opts)\n const port = opts.port ?? 5005\n const host = opts.host ?? '127.0.0.1'\n return new Promise((resolve, reject) => {\n let settled = false\n let server: ServerType | undefined\n server = serve({ fetch: app.fetch, port, hostname: host }, ({ address, port: actualPort }) => {\n if (settled) return\n settled = true\n // eslint-disable-next-line no-console\n console.log(`[agent-eval] serving on http://${address}:${actualPort}`)\n resolve({\n server: server!,\n port: actualPort,\n host: address,\n close: () =>\n new Promise<void>((res, rej) => {\n server!.close((err) => (err ? rej(err) : res()))\n }),\n })\n })\n server.on('error', (err) => {\n if (settled) return\n settled = true\n reject(err)\n })\n })\n}\n"],"mappings":";;;;;AAYA,SAAS,4BAA4B;AACrC,SAAS,SAAS;AAElB,qBAAqB,CAAC;AAIf,IAAM,wBAAwB,EAClC,OAAO;AAAA,EACN,IAAI,EACD,OAAO,EACP,IAAI,CAAC,EACL,SAAS,gFAA2E;AAAA,EACvF,aAAa,EACV,OAAO,EACP,IAAI,CAAC,EACL,SAAS,sEAAsE;AAAA,EAClF,QAAQ,EACL,OAAO,EACP,IAAI,CAAC,EACL,QAAQ,CAAC,EACT,SAAS,gEAAgE;AAAA,EAC5E,KAAK,EAAE,OAAO,EAAE,QAAQ,CAAC,EAAE,SAAS,gDAAgD;AAAA,EACpF,KAAK,EAAE,OAAO,EAAE,QAAQ,CAAC,EAAE,SAAS,gDAAgD;AACtF,CAAC,EACA,QAAQ,iBAAiB;AAErB,IAAM,oBAAoB,EAC9B,OAAO;AAAA,EACN,IAAI,EAAE,OAAO,EAAE,IAAI,CAAC,EAAE,SAAS,mEAA8D;AAAA,EAC7F,aAAa,EAAE,OAAO,EAAE,IAAI,CAAC,EAAE,SAAS,mDAAmD;AAC7F,CAAC,EACA,QAAQ,aAAa;AAIjB,IAAM,eAAe,EACzB,OAAO;AAAA,EACN,MAAM,EACH,OAAO,EACP,IAAI,CAAC,EACL,SAAS,4EAAuE;AAAA,EACnF,aAAa,EACV,OAAO,EACP,IAAI,CAAC,EACL,SAAS,0DAA0D;AAAA,EACtE,cAAc,EACX,OAAO,EACP,IAAI,CAAC,EACL;AAAA,IACC;AAAA,EACF;AAAA,EACF,YAAY,EACT,MAAM,qBAAqB,EAC3B,IAAI,CAAC,EACL,SAAS,+DAA+D;AAAA,EAC3E,cAAc,EACX,MAAM,iBAAiB,EACvB,QAAQ,CAAC,CAAC,EACV,SAAS,iFAAiF;AAAA,EAC7F,MAAM,EACH,MAAM,iBAAiB,EACvB,QAAQ,CAAC,CAAC,EACV,SAAS,uEAAuE;AACrF,CAAC,EACA,QAAQ,QAAQ;AAIZ,IAAM,qBAAqB,EAC/B,OAAO;AAAA,EACN,YAAY,EACT,OAAO,EACP,SAAS,EACT,SAAS,kEAAkE;AAAA,EAC9E,QAAQ,aAAa,SAAS,EAAE;AAAA,IAC9B;AAAA,EACF;AAAA,EACA,SAAS,EACN,OAAO,EACP,IAAI,CAAC,EACL,SAAS,uFAAkF;AAAA,EAC9F,SAAS,EACN,OAAO,EAAE,OAAO,GAAG,EAAE,QAAQ,CAAC,EAC9B,SAAS,EACT;AAAA,IACC;AAAA,EACF;AAAA,EACF,OAAO,EACJ,OAAO,EACP,SAAS,EACT,SAAS,+EAA+E;AAC7F,CAAC,EACA,OAAO,CAAC,MAAM,QAAQ,EAAE,UAAU,MAAM,QAAQ,EAAE,MAAM,GAAG;AAAA,EAC1D,SAAS;AACX,CAAC,EACA,QAAQ,cAAc;AAElB,IAAM,oBAAoB,EAC9B,OAAO;AAAA,EACN,WAAW,EACR,OAAO,EACP,IAAI,CAAC,EACL,IAAI,CAAC,EACL,SAAS,iFAAiF;AAAA,EAC7F,YAAY,EACT,OAAO,EAAE,OAAO,GAAG,EAAE,OAAO,CAAC,EAC7B,SAAS,mDAAmD;AAAA,EAC/D,cAAc,EACX,MAAM,EAAE,OAAO,CAAC,EAChB,QAAQ,CAAC,CAAC,EACV,SAAS,+EAA+E;AAAA,EAC3F,MAAM,EACH,MAAM,EAAE,OAAO,CAAC,EAChB,QAAQ,CAAC,CAAC,EACV,SAAS,8DAA8D;AAAA,EAC1E,WAAW,EACR,OAAO,EACP,SAAS,yEAAyE;AAAA,EACrF,eAAe,EACZ,OAAO,EACP;AAAA,IACC;AAAA,EACF;AAAA,EACF,OAAO,EAAE,OAAO,EAAE,SAAS,yDAAyD;AAAA,EACpF,YAAY,EAAE,OAAO,EAAE,IAAI,EAAE,YAAY,EAAE,SAAS,qCAAqC;AAC3F,CAAC,EACA,QAAQ,aAAa;AAIjB,IAAM,mBAAmB,EAC7B,OAAO;AAAA,EACN,MAAM,EAAE,OAAO,EAAE,SAAS,yCAAyC;AAAA,EACnE,aAAa,EAAE,OAAO,EAAE,SAAS,4BAA4B;AAAA,EAC7D,YAAY,EACT,MAAM,EAAE,OAAO,EAAE,IAAI,EAAE,OAAO,GAAG,aAAa,EAAE,OAAO,GAAG,QAAQ,EAAE,OAAO,EAAE,CAAC,CAAC,EAC/E,SAAS,kDAAkD;AAAA,EAC9D,cAAc,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,QAAQ,CAAC,CAAC,EAAE,SAAS,uCAAuC;AAAA,EAC9F,eAAe,EAAE,OAAO,EAAE,SAAS,8DAAyD;AAC9F,CAAC,EACA,QAAQ,YAAY;AAEhB,IAAM,4BAA4B,EACtC,OAAO;AAAA,EACN,SAAS,EAAE,MAAM,gBAAgB;AACnC,CAAC,EACA,QAAQ,qBAAqB;AAIzB,IAAM,wBAAwB,EAClC,OAAO;AAAA,EACN,SAAS,EAAE,OAAO,EAAE,SAAS,qDAAqD;AAAA,EAClF,SAAS,EAAE,OAAO,EAAE,SAAS,0DAA0D;AAAA,EACvF,aAAa,EACV,OAAO,EACP;AAAA,IACC;AAAA,EACF;AAAA,EACF,YAAY,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,SAAS,iCAAiC;AAC5E,CAAC,EACA,QAAQ,iBAAiB;AAErB,IAAM,uBAAuB,EACjC,OAAO;AAAA,EACN,QAAQ,EAAE,QAAQ,IAAI;AAAA,EACtB,WAAW,EAAE,OAAO;AACtB,CAAC,EACA,QAAQ,gBAAgB;AAUpB,IAAM,mBAAmB,EAC7B,OAAO;AAAA,EACN,SAAS,EAAE,OAAO,EAAE,IAAI,CAAC,EAAE,SAAS,4CAA4C;AAAA,EAChF,OAAO,EAAE,OAAO,EAAE,IAAI,CAAC,EAAE,SAAS,4BAA4B;AAAA,EAC9D,QAAQ,EAAE,OAAO,EAAE,SAAS,EAAE,SAAS,sCAAsC;AAAA,EAC7E,MAAM,EACH,KAAK;AAAA,IACJ;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,EACF,CAAC,EACA,SAAS,yEAAoE;AAAA,EAChF,WAAW,EACR,OAAO,EACP,IAAI,EACJ,YAAY,EACZ,SAAS,kEAAkE;AAAA,EAC9E,SAAS,EACN,OAAO,EAAE,OAAO,GAAG,EAAE,QAAQ,CAAC,EAC9B,SAAS,sDAAiD;AAC/D,CAAC,EACA,QAAQ,YAAY;AAEhB,IAAM,4BAA4B,EACtC,OAAO;AAAA,EACN,QAAQ,EACL,MAAM,gBAAgB,EACtB,IAAI,CAAC,EACL,IAAI,GAAM,EACV,SAAS,4EAAuE;AACrF,CAAC,EACA,QAAQ,qBAAqB;AAEzB,IAAM,6BAA6B,EACvC,OAAO;AAAA,EACN,UAAU,EAAE,OAAO,EAAE,IAAI,EAAE,YAAY,EAAE,SAAS,6BAA6B;AAAA,EAC/E,UAAU,EACP,OAAO,EACP,IAAI,EACJ,YAAY,EACZ,SAAS,uEAAkE;AAAA,EAC9E,QAAQ,EACL;AAAA,IACC,EAAE,OAAO;AAAA,MACP,SAAS,EAAE,OAAO,EAAE,SAAS,iCAAiC;AAAA,MAC9D,SAAS,EAAE,OAAO,EAAE,SAAS,6BAA6B;AAAA,IAC5D,CAAC;AAAA,EACH,EACC,QAAQ,CAAC,CAAC;AACf,CAAC,EACA,QAAQ,sBAAsB;AAE1B,IAAM,sBAAsB,EAChC,OAAO;AAAA,EACN,IAAI,EAAE,OAAO,EAAE,SAAS;AAAA,EACxB,QAAQ,EAAE,KAAK,CAAC,QAAQ,SAAS,eAAe,UAAU,UAAU,QAAQ,CAAC;AAAA,EAC7E,MAAM,EAAE,KAAK;AAAA,IACX;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,EACF,CAAC;AAAA,EACD,OAAO,EAAE,QAAQ;AAAA,EACjB,QAAQ,EAAE,OAAO,EAAE,SAAS;AAAA,EAC5B,UAAU,EAAE,KAAK,CAAC,QAAQ,WAAW,SAAS,UAAU,CAAC,EAAE,SAAS;AAAA,EACpE,WAAW,EAAE,OAAO,EAAE,SAAS,eAAe;AAAA,EAC9C,UAAU,EAAE,OAAO,EAAE,OAAO,GAAG,EAAE,QAAQ,CAAC,EAAE,SAAS;AACvD,CAAC,EACA,QAAQ,eAAe;AAEnB,IAAM,wBAAwB,EAClC,OAAO;AAAA,EACN,IAAI,EAAE,OAAO,EAAE,IAAI,CAAC;AAAA,EACpB,WAAW,EAAE,OAAO,EAAE,IAAI,EAAE,YAAY;AAAA,EACxC,cAAc,EAAE,KAAK;AAAA,IACnB;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,EACF,CAAC;AAAA,EACD,UAAU,EAAE,QAAQ;AAAA,EACpB,SAAS,EAAE,MAAM,EAAE,QAAQ,CAAC,EAAE,SAAS;AAAA,EACvC,gBAAgB,EACb,OAAO;AAAA,IACN,MAAM,EAAE,OAAO;AAAA,IACf,MAAM,EAAE,KAAK,CAAC,OAAO,UAAU,MAAM,CAAC,EAAE,SAAS;AAAA,IACjD,SAAS,EAAE,OAAO,EAAE,SAAS;AAAA,IAC7B,oBAAoB,EAAE,QAAQ,EAAE,SAAS;AAAA,IACzC,kBAAkB,EAAE,QAAQ,EAAE,SAAS;AAAA,IACvC,UAAU,EAAE,OAAO,EAAE,OAAO,GAAG,EAAE,QAAQ,CAAC,EAAE,SAAS;AAAA,EACvD,CAAC,EACA,SAAS;AAAA,EACZ,UAAU,EAAE,MAAM,mBAAmB,EAAE,SAAS;AAAA,EAChD,WAAW,EAAE,OAAO;AAAA,EACpB,UAAU,EAAE,OAAO,EAAE,OAAO,GAAG,EAAE,QAAQ,CAAC,EAAE,SAAS;AACvD,CAAC,EACA,QAAQ,iBAAiB;AAErB,IAAM,2BAA2B,EACrC,OAAO;AAAA,EACN,IAAI,EAAE,OAAO,EAAE,IAAI,CAAC,EAAE,SAAS,gDAAgD;AAAA,EAC/E,WAAW,EAAE,OAAO,EAAE,SAAS;AAAA,EAC/B,YAAY,EAAE,OAAO,EAAE,SAAS;AAAA,EAChC,MAAM,EAAE,OAAO;AAAA,IACb,QAAQ,EAAE,OAAO,EAAE,IAAI,CAAC;AAAA,IACxB,SAAS,EAAE,QAAQ,EAAE,SAAS;AAAA,EAChC,CAAC;AAAA,EACD,UAAU,EAAE,MAAM,qBAAqB,EAAE,QAAQ,CAAC,CAAC;AAAA,EACnD,QAAQ,EAAE,MAAM,mBAAmB,EAAE,QAAQ,CAAC,CAAC;AAAA,EAC/C,SAAS,EACN,OAAO;AAAA,IACN,SAAS,EAAE,QAAQ,EAAE,SAAS;AAAA,IAC9B,OAAO,EAAE,OAAO,EAAE,SAAS;AAAA,IAC3B,SAAS,EAAE,OAAO,EAAE,OAAO,GAAG,EAAE,OAAO,CAAC,EAAE,SAAS;AAAA,IACnD,SAAS,EAAE,OAAO,EAAE,SAAS;AAAA,IAC7B,QAAQ,EAAE,OAAO,EAAE,SAAS;AAAA,IAC5B,YAAY,EAAE,OAAO,EAAE,SAAS;AAAA,IAChC,UAAU,EAAE,OAAO,EAAE,OAAO,GAAG,EAAE,QAAQ,CAAC,EAAE,SAAS;AAAA,EACvD,CAAC,EACA,SAAS;AAAA,EACZ,OAAO,EAAE,KAAK,CAAC,SAAS,OAAO,QAAQ,SAAS,CAAC,EAAE,SAAS;AAAA,EAC5D,MAAM,EAAE,OAAO,EAAE,OAAO,GAAG,EAAE,OAAO,CAAC,EAAE,SAAS;AAAA,EAChD,WAAW,EAAE,OAAO,EAAE,SAAS,eAAe;AAAA,EAC9C,WAAW,EAAE,OAAO,EAAE,SAAS;AAAA,EAC/B,UAAU,EAAE,OAAO,EAAE,OAAO,GAAG,EAAE,QAAQ,CAAC,EAAE,SAAS;AACvD,CAAC,EACA,QAAQ,oBAAoB;AAExB,IAAM,+BAA+B,EACzC,OAAO;AAAA,EACN,IAAI,EAAE,OAAO,EAAE,SAAS,mCAAmC;AAAA,EAC3D,WAAW,EAAE,QAAQ,EAAE,SAAS,wDAAwD;AAC1F,CAAC,EACA,QAAQ,wBAAwB;AAU5B,IAAM,sBAAsB,EAChC,OAAO;AAAA,EACN,OAAO,EACJ,OAAO;AAAA,IACN,MAAM,EACH,OAAO,EACP;AAAA,MACC;AAAA,IACF;AAAA,IACF,SAAS,EAAE,OAAO,EAAE,SAAS,yBAAyB;AAAA,IACtD,SAAS,EAAE,QAAQ,EAAE,SAAS,EAAE,SAAS,6BAA6B;AAAA,EACxE,CAAC,EACA,SAAS,+DAA+D;AAC7E,CAAC,EACA,QAAQ,eAAe;AAoBnB,IAAM,eAAe;AAMrB,SAAS,WAAW,QAAwB;AACjD,QAAM,SAAS,gBAAgB,MAAM;AACrC,MAAI,IAAI;AACR,WAAS,IAAI,GAAG,IAAI,OAAO,QAAQ,KAAK;AACtC,QAAK,IAAI,KAAM,OAAO,WAAW,CAAC;AAAA,EACpC;AAEA,SAAO,GAAG,OAAO,IAAI,KAAK,MAAM,GAAG,SAAS,EAAE,EAAE,SAAS,GAAG,GAAG,CAAC;AAClE;AAEA,SAAS,gBAAgB,OAAwB;AAC/C,MAAI,MAAM,QAAQ,KAAK,EAAG,QAAO,IAAI,MAAM,IAAI,CAAC,SAAS,gBAAgB,IAAI,CAAC,EAAE,KAAK,GAAG,CAAC;AACzF,MAAI,SAAS,OAAO,UAAU,UAAU;AACtC,UAAM,UAAU,OAAO,QAAQ,KAAgC,EAC5D,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,cAAc,CAAC,CAAC,EACrC,IAAI,CAAC,CAAC,KAAK,IAAI,MAAM,GAAG,KAAK,UAAU,GAAG,CAAC,IAAI,gBAAgB,IAAI,CAAC,EAAE;AACzE,WAAO,IAAI,QAAQ,KAAK,GAAG,CAAC;AAAA,EAC9B;AACA,SAAO,KAAK,UAAU,KAAK;AAC7B;;;ACjYA,IAAM,YAAoB;AAAA,EACxB,MAAM;AAAA,EACN,aACE;AAAA,EACF,cAAc;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAuBd,YAAY;AAAA,IACV;AAAA,MACE,IAAI;AAAA,MACJ,aAAa;AAAA,MACb,QAAQ;AAAA,MACR,KAAK;AAAA,MACL,KAAK;AAAA,IACP;AAAA,IACA;AAAA,MACE,IAAI;AAAA,MACJ,aAAa;AAAA,MACb,QAAQ;AAAA,MACR,KAAK;AAAA,MACL,KAAK;AAAA,IACP;AAAA,IACA;AAAA,MACE,IAAI;AAAA,MACJ,aAAa;AAAA,MACb,QAAQ;AAAA,MACR,KAAK;AAAA,MACL,KAAK;AAAA,IACP;AAAA,EACF;AAAA,EACA,cAAc;AAAA,IACZ,EAAE,IAAI,cAAc,aAAa,qCAAqC;AAAA,IACtE,EAAE,IAAI,kBAAkB,aAAa,sCAAsC;AAAA,IAC3E,EAAE,IAAI,eAAe,aAAa,oCAAoC;AAAA,IACtE,EAAE,IAAI,WAAW,aAAa,kCAAkC;AAAA,IAChE,EAAE,IAAI,mBAAmB,aAAa,qCAAqC;AAAA,IAC3E,EAAE,IAAI,WAAW,aAAa,oCAAoC;AAAA,IAClE,EAAE,IAAI,eAAe,aAAa,uCAAuC;AAAA,EAC3E;AAAA,EACA,MAAM;AAAA,IACJ,EAAE,IAAI,sBAAsB,aAAa,qCAAqC;AAAA,IAC9E,EAAE,IAAI,iBAAiB,aAAa,4BAA4B;AAAA,IAChE,EAAE,IAAI,0BAA0B,aAAa,wBAAwB;AAAA,IACrE,EAAE,IAAI,kBAAkB,aAAa,oCAAoC;AAAA,EAC3E;AACF;AAIO,IAAM,kBAA0C;AAAA,EACrD,aAAa;AACf;AAGO,SAAS,iBAAiB,MAAkC;AACjE,SAAO,gBAAgB,IAAI;AAC7B;AAGO,SAAS,qBAAqB;AACnC,SAAO,OAAO,OAAO,eAAe,EAAE,IAAI,CAAC,OAAO;AAAA,IAChD,MAAM,EAAE;AAAA,IACR,aAAa,EAAE;AAAA,IACf,YAAY,EAAE,WAAW,IAAI,CAAC,OAAO;AAAA,MACnC,IAAI,EAAE;AAAA,MACN,aAAa,EAAE;AAAA,MACf,QAAQ,EAAE;AAAA,IACZ,EAAE;AAAA,IACF,cAAc,EAAE,aAAa,IAAI,CAAC,MAAM,EAAE,EAAE;AAAA,IAC5C,eAAe,WAAW,CAAC;AAAA,EAC7B,EAAE;AACJ;;;ACwHA,SAAS,oBAAoB;AAC7B,SAAS,SAAS,eAAe;AACjC,SAAS,qBAAqB;AA9MvB,IAAM,YAAN,cAAwB,MAAM;AAAA,EACnC,YACkB,MAChB,SACgB,SAAiB,KACjB,SAChB;AACA,UAAM,OAAO;AALG;AAEA;AACA;AAGhB,SAAK,OAAO;AAAA,EACd;AAAA,EAPkB;AAAA,EAEA;AAAA,EACA;AAKpB;AAKA,SAAS,kBAAkB,QAAgB;AACzC,SAAO;AAAA,IACL,MAAM;AAAA,IACN,QAAQ;AAAA,MACN,MAAM;AAAA,MACN,sBAAsB;AAAA,MACtB,YAAY;AAAA,QACV,YAAY;AAAA,UACV,MAAM;AAAA,UACN,sBAAsB;AAAA,UACtB,YAAY,OAAO;AAAA,YACjB,OAAO,WAAW,IAAI,CAAC,MAAM;AAAA,cAC3B,EAAE;AAAA,cACF,EAAE,MAAM,UAAU,SAAS,EAAE,KAAK,SAAS,EAAE,IAAI;AAAA,YACnD,CAAC;AAAA,UACH;AAAA,UACA,UAAU,OAAO,WAAW,IAAI,CAAC,MAAM,EAAE,EAAE;AAAA,QAC7C;AAAA,QACA,cAAc;AAAA,UACZ,MAAM;AAAA,UACN,OAAO,EAAE,MAAM,UAAU,MAAM,OAAO,aAAa,IAAI,CAAC,MAAM,EAAE,EAAE,EAAE;AAAA,QACtE;AAAA,QACA,MAAM;AAAA,UACJ,MAAM;AAAA,UACN,OAAO,EAAE,MAAM,UAAU,MAAM,OAAO,KAAK,IAAI,CAAC,MAAM,EAAE,EAAE,EAAE;AAAA,QAC9D;AAAA,QACA,WAAW,EAAE,MAAM,SAAS;AAAA,MAC9B;AAAA,MACA,UAAU,CAAC,cAAc,WAAW;AAAA,IACtC;AAAA,EACF;AACF;AASA,SAAS,oBAAoB,OAAgB,QAA6B;AACxE,MAAI,CAAC,SAAS,OAAO,UAAU,UAAU;AACvC,UAAM,IAAI,UAAU,eAAe,oCAAoC,KAAK,KAAK;AAAA,EACnF;AACA,QAAM,MAAM;AACZ,QAAM,gBAAgB,IAAI;AAC1B,MAAI,CAAC,iBAAiB,OAAO,kBAAkB,YAAY,MAAM,QAAQ,aAAa,GAAG;AACvF,UAAM,IAAI,UAAU,eAAe,wCAAwC,KAAK,KAAK;AAAA,EACvF;AAEA,QAAM,aAAqC,CAAC;AAC5C,QAAM,kBAAkB;AACxB,aAAW,OAAO,OAAO,YAAY;AACnC,UAAM,QAAQ,gBAAgB,IAAI,EAAE;AACpC,QACE,OAAO,UAAU,YACjB,CAAC,OAAO,SAAS,KAAK,KACtB,QAAQ,IAAI,OACZ,QAAQ,IAAI,KACZ;AACA,YAAM,IAAI;AAAA,QACR;AAAA,QACA,+CAA+C,IAAI,EAAE;AAAA,QACrD;AAAA,QACA;AAAA,MACF;AAAA,IACF;AACA,eAAW,IAAI,EAAE,IAAI;AAAA,EACvB;AAEA,QAAM,kBAAkB,IAAI,IAAI,OAAO,aAAa,IAAI,CAAC,SAAS,KAAK,EAAE,CAAC;AAC1E,QAAM,cAAc,IAAI,IAAI,OAAO,KAAK,IAAI,CAAC,QAAQ,IAAI,EAAE,CAAC;AAC5D,QAAM,eAAe,gBAAgB,IAAI,cAAc,iBAAiB,gBAAgB,KAAK;AAC7F,QAAM,OAAO,gBAAgB,IAAI,MAAM,aAAa,QAAQ,KAAK;AACjE,MAAI,OAAO,IAAI,cAAc,YAAY,IAAI,UAAU,KAAK,EAAE,WAAW,GAAG;AAC1E,UAAM,IAAI,UAAU,eAAe,qCAAqC,KAAK,KAAK;AAAA,EACpF;AAEA,SAAO,EAAE,YAAY,cAAc,MAAM,WAAW,IAAI,UAAU;AACpE;AAEA,SAAS,gBACP,KACA,SACA,OACA,UACU;AACV,MAAI,QAAQ,OAAW,QAAO,CAAC;AAC/B,MAAI,CAAC,MAAM,QAAQ,GAAG,GAAG;AACvB,UAAM,IAAI,UAAU,eAAe,4BAA4B,KAAK,KAAK,KAAK,QAAQ;AAAA,EACxF;AACA,QAAM,MAAgB,CAAC;AACvB,aAAW,QAAQ,KAAK;AACtB,QAAI,OAAO,SAAS,YAAY,CAAC,QAAQ,IAAI,IAAI,GAAG;AAClD,YAAM,IAAI;AAAA,QACR;AAAA,QACA,0BAA0B,KAAK,QAAQ,OAAO,IAAI,CAAC;AAAA,QACnD;AAAA,QACA;AAAA,MACF;AAAA,IACF;AACA,QAAI,KAAK,IAAI;AAAA,EACf;AACA,SAAO;AACT;AAEA,SAAS,eAAe,YAAoC,QAAwB;AAClF,MAAI,WAAW;AACf,MAAI,cAAc;AAClB,aAAW,OAAO,OAAO,YAAY;AACnC,UAAM,MAAM,WAAW,IAAI,EAAE,KAAK;AAClC,UAAM,QAAQ,IAAI,MAAM,IAAI,OAAO;AACnC,UAAM,aAAa,KAAK,IAAI,GAAG,KAAK,IAAI,IAAI,MAAM,IAAI,OAAO,KAAK,CAAC;AACnE,gBAAY,aAAa,IAAI;AAC7B,mBAAe,IAAI;AAAA,EACrB;AACA,SAAO,cAAc,IAAI,WAAW,cAAc;AACpD;AAEA,SAAS,iBAAiB,SAAiB,SAA0B;AACnE,QAAM,MAAM,WAAW,OAAO,KAAK,OAAiB,EAAE,SAAS,KAAK,UAAU,OAAO,IAAI;AACzF,SAAO;AAAA,IACL;AAAA,IACA;AAAA,IACA;AAAA,IACA,MAAM,yCAAyC;AAAA,IAC/C,MAAM,MAAM;AAAA,EACd,EACG,OAAO,OAAO,EACd,KAAK,IAAI;AACd;AAEA,IAAM,sBAAsB;AAE5B,eAAsB,YAAY,KAAyC;AAEzE,MAAI;AACJ,MAAI,IAAI,YAAY;AAClB,UAAM,QAAQ,iBAAiB,IAAI,UAAU;AAC7C,QAAI,CAAC,OAAO;AACV,YAAM,IAAI,UAAU,oBAAoB,6BAA6B,IAAI,UAAU,MAAM,GAAG;AAAA,IAC9F;AACA,aAAS;AAAA,EACX,WAAW,IAAI,QAAQ;AACrB,aAAS,IAAI;AAAA,EACf,OAAO;AAEL,UAAM,IAAI,UAAU,oBAAoB,4CAA4C,GAAG;AAAA,EACzF;AAEA,QAAM,YAAY,KAAK,IAAI;AAC3B,QAAM,QAAQ,IAAI,SAAS;AAE3B,QAAM,EAAE,OAAO,OAAO,IAAI,MAAM,YAAyB;AAAA,IACvD;AAAA,IACA,UAAU;AAAA,MACR,EAAE,MAAM,UAAU,SAAS,OAAO,aAAa;AAAA,MAC/C,EAAE,MAAM,QAAQ,SAAS,iBAAiB,IAAI,SAAS,IAAI,OAAO,EAAE;AAAA,IACtE;AAAA,IACA,YAAY,kBAAkB,MAAM;AAAA,IACpC,aAAa;AAAA,IACb,WAAW;AAAA,EACb,CAAC;AAED,QAAM,SAAS,oBAAoB,OAAO,MAAM;AAEhD,QAAM,YAAY,eAAe,OAAO,YAAY,MAAM;AAC1D,QAAM,aAAa,KAAK,IAAI,IAAI;AAEhC,SAAO;AAAA,IACL;AAAA,IACA,YAAY,OAAO;AAAA,IACnB,cAAc,OAAO,gBAAgB,CAAC;AAAA,IACtC,MAAM,OAAO,QAAQ,CAAC;AAAA,IACtB,WAAW,OAAO;AAAA,IAClB,eAAe,WAAW,MAAM;AAAA,IAChC,OAAO,OAAO;AAAA,IACd;AAAA,EACF;AACF;AAIO,SAAS,oBAAyC;AACvD,SAAO,EAAE,SAAS,mBAAmB,EAAE;AACzC;AAQA,IAAI;AAEJ,SAAS,qBAA6B;AACpC,MAAI,eAAgB,QAAO;AAG3B,QAAM,OAAO,QAAQ,cAAc,YAAY,GAAG,CAAC;AACnD,QAAM,aAAa;AAAA,IACjB,QAAQ,MAAM,MAAM,MAAM,cAAc;AAAA;AAAA,IACxC,QAAQ,MAAM,MAAM,cAAc;AAAA;AAAA,EACpC;AACA,aAAW,QAAQ,YAAY;AAC7B,QAAI;AACF,YAAM,MAAM,KAAK,MAAM,aAAa,MAAM,OAAO,CAAC;AAClD,UAAI,IAAI,SAAS;AACf,yBAAiB,IAAI;AACrB,eAAO,IAAI;AAAA,MACb;AAAA,IACF,QAAQ;AAAA,IAER;AAAA,EACF;AACA,SAAO;AACT;AAEO,SAAS,gBAAiC;AAC/C,SAAO;AAAA,IACL,SAAS;AAAA,IACT,SAAS,mBAAmB;AAAA,IAC5B,aAAa;AAAA,IACb,YAAY,CAAC,SAAS,eAAe,WAAW,mBAAmB,eAAe;AAAA,EACpF;AACF;AA2BA,eAAsB,mBACpB,KACA,QAC+B;AAC/B,MAAI,CAAC,OAAO,YAAY;AACtB,UAAM,IAAI;AAAA,MACR;AAAA,MACA;AAAA,MACA;AAAA,IACF;AAAA,EACF;AACA,QAAM,SAAsD,CAAC;AAC7D,MAAI,WAAW;AACf,aAAW,SAAS,IAAI,QAAQ;AAC9B,QAAI;AAEF,YAAM,OAAO,WAAW,YAAY,KAA2B;AAC/D;AAAA,IACF,SAAS,KAAK;AACZ,aAAO,KAAK;AAAA,QACV,SAAS,MAAM;AAAA,QACf,SAAS,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;AAAA,MAC1D,CAAC;AAAA,IACH;AAAA,EACF;AACA,SAAO,EAAE,UAAU,UAAU,OAAO,QAAQ,OAAO;AACrD;AAOA,eAAsB,qBACpB,KACA,QACiC;AACjC,MAAI,CAAC,OAAO,eAAe;AACzB,UAAM,IAAI;AAAA,MACR;AAAA,MACA;AAAA,MACA;AAAA,IACF;AAAA,EACF;AAIA,QAAM,OAAO,cAAc,KAAK,GAAgE;AAChG,SAAO,EAAE,IAAI,IAAI,IAAI,WAAW,KAAK;AACvC;;;AChVA,SAAS,iBAAiB,2BAA2B;AAiB9C,SAAS,aAAa,gBAAuC;AAClE,QAAM,WAAW,IAAI,gBAAgB;AAGrC,WAAS,SAAS,gBAAgB,kBAAkB;AACpD,WAAS,SAAS,eAAe,iBAAiB;AAClD,WAAS,SAAS,uBAAuB,yBAAyB;AAClE,WAAS,SAAS,mBAAmB,qBAAqB;AAC1D,WAAS,SAAS,kBAAkB,oBAAoB;AACxD,WAAS,SAAS,iBAAiB,mBAAmB;AACtD,WAAS,SAAS,uBAAuB,yBAAyB;AAClE,WAAS,SAAS,wBAAwB,0BAA0B;AACpE,WAAS,SAAS,sBAAsB,wBAAwB;AAChE,WAAS,SAAS,0BAA0B,4BAA4B;AAGxE,WAAS,aAAa;AAAA,IACpB,QAAQ;AAAA,IACR,MAAM;AAAA,IACN,SAAS;AAAA,IACT,aACE;AAAA,IACF,SAAS;AAAA,MACP,MAAM;AAAA,QACJ,SAAS;AAAA,UACP,oBAAoB,EAAE,QAAQ,mBAAmB;AAAA,QACnD;AAAA,MACF;AAAA,IACF;AAAA,IACA,WAAW;AAAA,MACT,KAAK;AAAA,QACH,aAAa;AAAA,QACb,SAAS,EAAE,oBAAoB,EAAE,QAAQ,kBAAkB,EAAE;AAAA,MAC/D;AAAA,MACA,KAAK;AAAA,QACH,aAAa;AAAA,QACb,SAAS,EAAE,oBAAoB,EAAE,QAAQ,oBAAoB,EAAE;AAAA,MACjE;AAAA,MACA,KAAK;AAAA,QACH,aAAa;AAAA,QACb,SAAS,EAAE,oBAAoB,EAAE,QAAQ,oBAAoB,EAAE;AAAA,MACjE;AAAA,MACA,KAAK;AAAA,QACH,aAAa;AAAA,QACb,SAAS,EAAE,oBAAoB,EAAE,QAAQ,oBAAoB,EAAE;AAAA,MACjE;AAAA,IACF;AAAA,EACF,CAAC;AAED,WAAS,aAAa;AAAA,IACpB,QAAQ;AAAA,IACR,MAAM;AAAA,IACN,SAAS;AAAA,IACT,aACE;AAAA,IACF,WAAW;AAAA,MACT,KAAK;AAAA,QACH,aAAa;AAAA,QACb,SAAS,EAAE,oBAAoB,EAAE,QAAQ,0BAA0B,EAAE;AAAA,MACvE;AAAA,IACF;AAAA,EACF,CAAC;AAED,WAAS,aAAa;AAAA,IACpB,QAAQ;AAAA,IACR,MAAM;AAAA,IACN,SAAS;AAAA,IACT,aAAa;AAAA,IACb,WAAW;AAAA,MACT,KAAK;AAAA,QACH,aAAa;AAAA,QACb,SAAS,EAAE,oBAAoB,EAAE,QAAQ,sBAAsB,EAAE;AAAA,MACnE;AAAA,IACF;AAAA,EACF,CAAC;AAED,WAAS,aAAa;AAAA,IACpB,QAAQ;AAAA,IACR,MAAM;AAAA,IACN,SAAS;AAAA,IACT,WAAW;AAAA,MACT,KAAK;AAAA,QACH,aAAa;AAAA,QACb,SAAS,EAAE,oBAAoB,EAAE,QAAQ,qBAAqB,EAAE;AAAA,MAClE;AAAA,IACF;AAAA,EACF,CAAC;AAED,WAAS,aAAa;AAAA,IACpB,QAAQ;AAAA,IACR,MAAM;AAAA,IACN,SAAS;AAAA,IACT,aACE;AAAA,IACF,SAAS;AAAA,MACP,MAAM;AAAA,QACJ,SAAS;AAAA,UACP,oBAAoB,EAAE,QAAQ,0BAA0B;AAAA,UACxD,wBAAwB,EAAE,QAAQ,0BAA0B;AAAA,QAC9D;AAAA,MACF;AAAA,IACF;AAAA,IACA,WAAW;AAAA,MACT,KAAK;AAAA,QACH,aAAa;AAAA,QACb,SAAS,EAAE,oBAAoB,EAAE,QAAQ,2BAA2B,EAAE;AAAA,MACxE;AAAA,MACA,KAAK;AAAA,QACH,aAAa;AAAA,QACb,SAAS,EAAE,oBAAoB,EAAE,QAAQ,oBAAoB,EAAE;AAAA,MACjE;AAAA,MACA,KAAK;AAAA,QACH,aAAa;AAAA,QACb,SAAS,EAAE,oBAAoB,EAAE,QAAQ,oBAAoB,EAAE;AAAA,MACjE;AAAA,MACA,KAAK;AAAA,QACH,aAAa;AAAA,QACb,SAAS,EAAE,oBAAoB,EAAE,QAAQ,oBAAoB,EAAE;AAAA,MACjE;AAAA,IACF;AAAA,EACF,CAAC;AAED,WAAS,aAAa;AAAA,IACpB,QAAQ;AAAA,IACR,MAAM;AAAA,IACN,SAAS;AAAA,IACT,aACE;AAAA,IACF,SAAS;AAAA,MACP,MAAM;AAAA,QACJ,SAAS;AAAA,UACP,oBAAoB,EAAE,QAAQ,yBAAyB;AAAA,QACzD;AAAA,MACF;AAAA,IACF;AAAA,IACA,WAAW;AAAA,MACT,KAAK;AAAA,QACH,aAAa;AAAA,QACb,SAAS,EAAE,oBAAoB,EAAE,QAAQ,6BAA6B,EAAE;AAAA,MAC1E;AAAA,MACA,KAAK;AAAA,QACH,aAAa;AAAA,QACb,SAAS,EAAE,oBAAoB,EAAE,QAAQ,oBAAoB,EAAE;AAAA,MACjE;AAAA,MACA,KAAK;AAAA,QACH,aAAa;AAAA,QACb,SAAS,EAAE,oBAAoB,EAAE,QAAQ,oBAAoB,EAAE;AAAA,MACjE;AAAA,MACA,KAAK;AAAA,QACH,aAAa;AAAA,QACb,SAAS,EAAE,oBAAoB,EAAE,QAAQ,oBAAoB,EAAE;AAAA,MACjE;AAAA,IACF;AAAA,EACF,CAAC;AAED,QAAM,YAAY,IAAI,oBAAoB,SAAS,WAAW;AAC9D,QAAM,MAAM,UAAU,iBAAiB;AAAA,IACrC,SAAS;AAAA,IACT,MAAM;AAAA,MACJ,OAAO;AAAA,MACP,SAAS;AAAA,MACT,aAAa;AAAA;AAAA,yBAEM,YAAY;AAAA,MAC/B,SAAS,EAAE,MAAM,kBAAkB,KAAK,+CAA+C;AAAA,MACvF,SAAS,EAAE,MAAM,MAAM;AAAA,IACzB;AAAA,IACA,SAAS,CAAC,EAAE,KAAK,yBAAyB,aAAa,yBAAyB,CAAC;AAAA,EACnF,CAAC;AACD,QAAM,YAAY,EAAE,MAAM,8BAA8B;AACxD,QAAM,oBAAoB;AAAA,IACxB,SAAS,EAAE,MAAM,UAAU,WAAW,EAAE;AAAA,IACxC,SAAS,EAAE,MAAM,UAAU,sBAAsB,KAAK;AAAA,IACtD,OAAO,EAAE,MAAM,SAAS;AAAA,EAC1B;AACA,MAAI,eAAe,CAAC;AACpB,MAAI,WAAW,YAAY,CAAC;AAC5B,MAAI,WAAW,QAAQ,eAAe;AAAA,IACpC,OAAO;AAAA,MACL;AAAA,QACE,MAAM;AAAA,QACN,sBAAsB;AAAA,QACtB,UAAU,CAAC,cAAc,SAAS;AAAA,QAClC,YAAY;AAAA,UACV,YAAY,EAAE,MAAM,UAAU,WAAW,EAAE;AAAA,UAC3C,GAAG;AAAA,QACL;AAAA,MACF;AAAA,MACA;AAAA,QACE,MAAM;AAAA,QACN,sBAAsB;AAAA,QACtB,UAAU,CAAC,UAAU,SAAS;AAAA,QAC9B,YAAY;AAAA,UACV,QAAQ;AAAA,UACR,GAAG;AAAA,QACL;AAAA,MACF;AAAA,IACF;AAAA,IACA,aAAa;AAAA,EACf;AACA,SAAO;AACT;;;ACxMA,eAAsB,YAAY,KAAiD;AACjF,MAAI;AACF,YAAQ,IAAI,QAAQ;AAAA,MAClB,KAAK,SAAS;AACZ,cAAM,SAAS,mBAAmB,UAAU,IAAI,MAAM;AACtD,YAAI,CAAC,OAAO,SAAS;AACnB,iBAAO;AAAA,YACL,OAAO;AAAA,cACL,MAAM;AAAA,cACN,SAAS;AAAA,cACT,SAAS,OAAO,MAAM;AAAA,YACxB;AAAA,UACF;AAAA,QACF;AACA,eAAO,EAAE,QAAQ,MAAM,YAAY,OAAO,IAAI,EAAE;AAAA,MAClD;AAAA,MACA,KAAK;AACH,eAAO,EAAE,QAAQ,kBAAkB,EAAE;AAAA,MACvC,KAAK;AACH,eAAO,EAAE,QAAQ,cAAc,EAAE;AAAA,MACnC;AACE,eAAO;AAAA,UACL,OAAO;AAAA,YACL,MAAM;AAAA,YACN,SAAS,mBAAoB,IAA2B,MAAM;AAAA,UAChE;AAAA,QACF;AAAA,IACJ;AAAA,EACF,SAAS,KAAK;AACZ,QAAI,eAAe,WAAW;AAC5B,aAAO,EAAE,OAAO,EAAE,MAAM,IAAI,MAAM,SAAS,IAAI,SAAS,SAAS,IAAI,QAAQ,EAAE;AAAA,IACjF;AACA,UAAM,UAAU,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;AAC/D,WAAO,EAAE,OAAO,EAAE,MAAM,kBAAkB,QAAQ,EAAE;AAAA,EACtD;AACF;AAIA,eAAe,QAAQ,QAAgD;AACrE,QAAM,SAAmB,CAAC;AAC1B,mBAAiB,SAAS,QAAQ;AAChC,WAAO,KAAK,OAAO,SAAS,KAAK,IAAI,QAAQ,OAAO,KAAK,KAAe,CAAC;AAAA,EAC3E;AACA,SAAO,OAAO,OAAO,MAAM,EAAE,SAAS,OAAO;AAC/C;AAGA,eAAsB,WAAW,QAAkC;AACjE,QAAM,MAAM,MAAM,QAAQ,QAAQ,KAAK;AACvC,MAAI;AACJ,MAAI;AACF,UAAM,OAAO,KAAK,MAAM,GAAG;AAC3B,UAAM,SAAS,EAAE,QAAwC,QAAQ,KAAK,IAAK;AAAA,EAC7E,SAAS,KAAK;AACZ,YAAQ,OAAO;AAAA,MACb,GAAG,KAAK,UAAU;AAAA,QAChB,OAAO;AAAA,UACL,MAAM;AAAA,UACN,SAAS,6BAA6B,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG,CAAC;AAAA,QACxF;AAAA,MACF,CAAC,CAAC;AAAA;AAAA,IACJ;AACA,WAAO;AAAA,EACT;AACA,QAAM,MAAM,MAAM,YAAY,GAAG;AACjC,UAAQ,OAAO,MAAM,GAAG,KAAK,UAAU,GAAG,CAAC;AAAA,CAAI;AAC/C,SAAO,WAAW,MAAM,IAAI;AAC9B;AAGA,eAAsB,YAAY,QAAkC;AAClE,QAAM,MAAM,MAAM,QAAQ,QAAQ,KAAK;AACvC,QAAM,QAAQ,IAAI,MAAM,IAAI,EAAE,OAAO,CAAC,MAAM,EAAE,KAAK,EAAE,SAAS,CAAC;AAC/D,MAAI,WAAW;AACf,aAAW,QAAQ,OAAO;AACxB,QAAI;AACJ,QAAI;AACF,YAAM,OAAO,KAAK,MAAM,IAAI;AAC5B,YAAM,SAAS,EAAE,QAAwC,QAAQ,KAAK,IAAK;AAAA,IAC7E,SAAS,KAAK;AACZ,cAAQ,OAAO;AAAA,QACb,GAAG,KAAK,UAAU;AAAA,UAChB,OAAO;AAAA,YACL,MAAM;AAAA,YACN,SAAS,4BAA4B,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG,CAAC;AAAA,UACvF;AAAA,QACF,CAAC,CAAC;AAAA;AAAA,MACJ;AACA,iBAAW;AACX;AAAA,IACF;AACA,UAAM,MAAM,MAAM,YAAY,GAAG;AACjC,YAAQ,OAAO,MAAM,GAAG,KAAK,UAAU,GAAG,CAAC;AAAA,CAAI;AAC/C,QAAI,WAAW,IAAK,YAAW;AAAA,EACjC;AACA,SAAO;AACT;;;AC/GA,SAA0B,aAAa;AACvC,SAAS,YAAY;AACrB,SAAS,YAAY;AAcrB,IAAM,aAAa,KAAK,IAAI;AAmB5B,IAAM,oBAAoB,oBAAI,IAAI,CAAC,YAAY,eAAe,eAAe,CAAC;AAEvE,SAAS,UAAU,OAAyB,CAAC,GAAG;AACrD,QAAM,MAAM,IAAI,KAAK;AAErB,MAAI,IAAI,KAAK,KAAK,CAAC;AAGnB,MAAI,KAAK,MAAM;AACb,UAAM,SAAS,KAAK,KAAK;AACzB,QAAI,IAAI,KAAK,OAAO,GAAG,SAAS;AAC9B,YAAM,OAAO,IAAI,IAAI,EAAE,IAAI,GAAG,EAAE;AAChC,UAAI,kBAAkB,IAAI,IAAI,EAAG,QAAO,KAAK;AAC7C,YAAM,MAAM,EAAE,IAAI,OAAO,eAAe,KAAK;AAC7C,YAAM,QAAQ,IAAI,MAAM,kBAAkB;AAC1C,UAAI,CAAC,OAAO;AACV,cAAM,IAAI,UAAU,gBAAgB,8CAA8C,GAAG;AAAA,MACvF;AACA,YAAM,QAAQ,MAAM,CAAC;AACrB,YAAM,KAAK,OAAO,WAAW,WAAW,UAAU,SAAS,MAAM,OAAO,KAAK;AAC7E,UAAI,CAAC,IAAI;AACP,cAAM,IAAI,UAAU,gBAAgB,yBAAyB,GAAG;AAAA,MAClE;AACA,aAAO,KAAK;AAAA,IACd,CAAC;AAAA,EACH;AAEA,MAAI,QAAQ,CAAC,KAAK,MAAM;AACtB,QAAI,eAAe,WAAW;AAC5B,YAAM,SAAS,IAAI;AACnB,aAAO,EAAE;AAAA,QACP,EAAE,OAAO,EAAE,MAAM,IAAI,MAAM,SAAS,IAAI,SAAS,SAAS,IAAI,QAAQ,EAAE;AAAA,QACxE;AAAA,MACF;AAAA,IACF;AAEA,YAAQ,MAAM,iCAAiC,GAAG;AAClD,WAAO,EAAE,KAAK,EAAE,OAAO,EAAE,MAAM,kBAAkB,SAAS,yBAAyB,EAAE,GAAG,GAAG;AAAA,EAC7F,CAAC;AAGD,MAAI;AAAA,IAAI;AAAA,IAAY,CAAC,MACnB,EAAE,KAAK,EAAE,QAAQ,MAAe,YAAY,KAAK,IAAI,IAAI,cAAc,IAAK,CAAC;AAAA,EAC/E;AAGA,MAAI,IAAI,eAAe,CAAC,MAAM,EAAE,KAAK,cAAc,CAAC,CAAC;AAGrD,MAAI,IAAI,eAAe,CAAC,MAAM,EAAE,KAAK,kBAAkB,CAAC,CAAC;AAGzD,MAAI,KAAK,aAAa,OAAO,MAAM;AACjC,UAAM,MAAM,MAAM,EAAE,IAAI,KAAK,EAAE,MAAM,MAAM,IAAI;AAC/C,QAAI,OAAO,MAAM;AACf,YAAM,IAAI,UAAU,oBAAoB,8BAA8B,GAAG;AAAA,IAC3E;AACA,UAAM,SAAS,mBAAmB,UAAU,GAAG;AAC/C,QAAI,CAAC,OAAO,SAAS;AACnB,YAAM,IAAI;AAAA,QACR;AAAA,QACA;AAAA,QACA;AAAA,QACA,OAAO,MAAM;AAAA,MACf;AAAA,IACF;AACA,UAAM,SAAS,MAAM,YAAY,OAAO,IAAI;AAC5C,WAAO,EAAE,KAAK,MAAM;AAAA,EACtB,CAAC;AAGD,MAAI,KAAK,qBAAqB,OAAO,MAAM;AACzC,UAAM,cAAc,EAAE,IAAI,OAAO,cAAc,KAAK;AACpD,QAAI;AACJ,QAAI,YAAY,SAAS,sBAAsB,GAAG;AAChD,YAAM,OAAO,MAAM,EAAE,IAAI,KAAK;AAC9B,YAAM,SAAS,KACZ,MAAM,IAAI,EACV,IAAI,CAAC,SAAS,KAAK,KAAK,CAAC,EACzB,OAAO,CAAC,SAAS,KAAK,SAAS,CAAC,EAChC,IAAI,CAAC,SAAS;AACb,YAAI;AACF,iBAAO,KAAK,MAAM,IAAI;AAAA,QACxB,QAAQ;AACN,gBAAM,IAAI;AAAA,YACR;AAAA,YACA;AAAA,YACA;AAAA,YACA,KAAK,MAAM,GAAG,GAAG;AAAA,UACnB;AAAA,QACF;AAAA,MACF,CAAC;AACH,gBAAU,EAAE,OAAO;AAAA,IACrB,OAAO;AACL,gBAAU,MAAM,EAAE,IAAI,KAAK,EAAE,MAAM,MAAM,IAAI;AAAA,IAC/C;AACA,QAAI,WAAW,MAAM;AACnB,YAAM,IAAI,UAAU,oBAAoB,wCAAwC,GAAG;AAAA,IACrF;AACA,UAAM,SAAS,0BAA0B,UAAU,OAAO;AAC1D,QAAI,CAAC,OAAO,SAAS;AACnB,YAAM,IAAI;AAAA,QACR;AAAA,QACA;AAAA,QACA;AAAA,QACA,OAAO,MAAM;AAAA,MACf;AAAA,IACF;AACA,UAAM,SAAS,MAAM,mBAAmB,OAAO,MAAM,KAAK,UAAU,CAAC,CAAC;AACtE,WAAO,EAAE,KAAK,MAAM;AAAA,EACtB,CAAC;AAGD,MAAI,KAAK,gBAAgB,OAAO,MAAM;AACpC,UAAM,MAAM,MAAM,EAAE,IAAI,KAAK,EAAE,MAAM,MAAM,IAAI;AAC/C,QAAI,OAAO,MAAM;AACf,YAAM,IAAI,UAAU,oBAAoB,8BAA8B,GAAG;AAAA,IAC3E;AACA,UAAM,SAAS,yBAAyB,UAAU,GAAG;AACrD,QAAI,CAAC,OAAO,SAAS;AACnB,YAAM,IAAI;AAAA,QACR;AAAA,QACA;AAAA,QACA;AAAA,QACA,OAAO,MAAM;AAAA,MACf;AAAA,IACF;AACA,UAAM,SAAS,MAAM,qBAAqB,OAAO,MAAM,KAAK,UAAU,CAAC,CAAC;AACxE,WAAO,EAAE,KAAK,MAAM;AAAA,EACtB,CAAC;AAGD,MAAI,IAAI,iBAAiB,CAAC,MAAM,EAAE,KAAK,aAAa,cAAc,EAAE,OAAO,CAAC,CAAC;AAE7E,SAAO;AACT;AASO,SAAS,YAAY,OAAqB,CAAC,GAAe;AAC/D,QAAM,MAAM,UAAU,IAAI;AAC1B,QAAM,OAAO,KAAK,QAAQ;AAC1B,QAAM,OAAO,KAAK,QAAQ;AAC1B,SAAO,MAAM,EAAE,OAAO,IAAI,OAAO,MAAM,UAAU,KAAK,GAAG,CAAC,EAAE,SAAS,MAAM,WAAW,MAAM;AAE1F,YAAQ,IAAI,kCAAkC,OAAO,IAAI,UAAU,EAAE;AAAA,EACvE,CAAC;AACH;AAmBO,SAAS,iBAAiB,OAAqB,CAAC,GAA2B;AAChF,QAAM,MAAM,UAAU,IAAI;AAC1B,QAAM,OAAO,KAAK,QAAQ;AAC1B,QAAM,OAAO,KAAK,QAAQ;AAC1B,SAAO,IAAI,QAAQ,CAACA,UAAS,WAAW;AACtC,QAAI,UAAU;AACd,QAAI;AACJ,aAAS,MAAM,EAAE,OAAO,IAAI,OAAO,MAAM,UAAU,KAAK,GAAG,CAAC,EAAE,SAAS,MAAM,WAAW,MAAM;AAC5F,UAAI,QAAS;AACb,gBAAU;AAEV,cAAQ,IAAI,kCAAkC,OAAO,IAAI,UAAU,EAAE;AACrE,MAAAA,SAAQ;AAAA,QACN;AAAA,QACA,MAAM;AAAA,QACN,MAAM;AAAA,QACN,OAAO,MACL,IAAI,QAAc,CAAC,KAAK,QAAQ;AAC9B,iBAAQ,MAAM,CAAC,QAAS,MAAM,IAAI,GAAG,IAAI,IAAI,CAAE;AAAA,QACjD,CAAC;AAAA,MACL,CAAC;AAAA,IACH,CAAC;AACD,WAAO,GAAG,SAAS,CAAC,QAAQ;AAC1B,UAAI,QAAS;AACb,gBAAU;AACV,aAAO,GAAG;AAAA,IACZ,CAAC;AAAA,EACH,CAAC;AACH;","names":["resolve"]}
package/dist/cli.js CHANGED
@@ -5,7 +5,7 @@ import {
5
5
  runRpcBatch,
6
6
  runRpcOnce,
7
7
  startServer
8
- } from "./chunk-YU3G6I7F.js";
8
+ } from "./chunk-63EPZQUZ.js";
9
9
  import "./chunk-VXNVVBZO.js";
10
10
  import "./chunk-PC4UYEBM.js";
11
11
  import "./chunk-QYJT52YW.js";
@@ -0,0 +1,141 @@
1
+ import { DefaultVerdict } from '@tangle-network/agent-runtime/loops';
2
+ export { DefaultVerdict } from '@tangle-network/agent-runtime/loops';
3
+
4
+ /**
5
+ * @experimental
6
+ *
7
+ * N-axis cartesian matrix over substrate types — types module.
8
+ *
9
+ * The matrix is a runner + aggregator. It iterates the cartesian product of
10
+ * caller-provided axes (any value type — `AgentProfile` from sandbox, `Driver`
11
+ * / `Validator` from agent-runtime, rubric records, thinking levels, anything)
12
+ * and aggregates per-axis pass/score/cost summaries. Substrate types are
13
+ * imported at the boundary by JSDoc only; the matrix never wraps them.
14
+ */
15
+
16
+ /** One axis = one dimension to iterate. `V` is the value type — pass any
17
+ * substrate type (AgentProfile, Driver, Validator, rubric record). */
18
+ interface MatrixAxis<V> {
19
+ /** Axis name. Becomes the key in `MatrixResult.byAxis`. */
20
+ name: string;
21
+ /** Stable id per value. Used as the bucket key in aggregation. */
22
+ values: Array<{
23
+ id: string;
24
+ value: V;
25
+ }>;
26
+ /** Optional bucket label override. Receives the same `(value, id)` the
27
+ * runner stored on the cell; default label is `id`. */
28
+ label?: (value: V, id: string) => string;
29
+ }
30
+ /** A cell carries one picked value from each axis, keyed by axis name. */
31
+ interface MatrixCell {
32
+ axes: Record<string, {
33
+ id: string;
34
+ value: unknown;
35
+ }>;
36
+ /** 0-based replicate index within the same axis combination. */
37
+ rep: number;
38
+ /** Stable sort key — preserves cartesian order across concurrent execution. */
39
+ ordinal: number;
40
+ }
41
+ interface CellResult<Output> {
42
+ output: Output;
43
+ verdict: DefaultVerdict;
44
+ costUsd: number;
45
+ durationMs: number;
46
+ runId?: string;
47
+ /** Populated when `runCell` threw. The cell contributes 0 to passRate AND
48
+ * meanScore regardless of `verdict`. */
49
+ error?: {
50
+ message: string;
51
+ kind: string;
52
+ };
53
+ }
54
+ interface AxisSummary {
55
+ axisName: string;
56
+ axisValue: string;
57
+ cells: number;
58
+ passRate: number;
59
+ meanScore: number;
60
+ p50Score: number;
61
+ p90Score: number;
62
+ totalCostUsd: number;
63
+ meanDurationMs: number;
64
+ }
65
+ interface MatrixResult<Output> {
66
+ cells: Array<{
67
+ cell: MatrixCell;
68
+ runs: CellResult<Output>[];
69
+ }>;
70
+ /** `byAxis[axisName][axisValueId] = summary`. Populated only for axes
71
+ * named in `aggregateBy` (default = every axis in `axes`). */
72
+ byAxis: Record<string, Record<string, AxisSummary>>;
73
+ summary: {
74
+ totalCells: number;
75
+ runsExecuted: number;
76
+ /** Cells removed by `filter` plus cells unscheduled after the cost
77
+ * ceiling or abort signal tripped. */
78
+ cellsSkipped: number;
79
+ overallPassRate: number;
80
+ overallMeanScore: number;
81
+ totalCostUsd: number;
82
+ durationMs: number;
83
+ };
84
+ /** Stable id-like string generated at the end of the run. */
85
+ matrixId: string;
86
+ }
87
+ interface RunAgentMatrixOptions<Output> {
88
+ axes: MatrixAxis<unknown>[];
89
+ /** User-supplied cell executor. May throw; the matrix captures throws as
90
+ * `CellResult.error` and continues. */
91
+ runCell: (cell: MatrixCell) => Promise<CellResult<Output>>;
92
+ /** Replicates per cell. Default 1. */
93
+ reps?: number;
94
+ /** Prune cells from the cartesian BEFORE rep expansion. */
95
+ filter?: (cell: Omit<MatrixCell, 'rep' | 'ordinal'>) => boolean;
96
+ /** Axes to aggregate into `byAxis`. Default: every axis in `axes`. */
97
+ aggregateBy?: string[];
98
+ /** Max concurrent in-flight `runCell` invocations. Default 4. */
99
+ maxConcurrency?: number;
100
+ /** Cumulative-cost abort threshold (USD). When the running sum of
101
+ * `result.costUsd` crosses this value, no new cells are scheduled.
102
+ * In-flight cells finish. Default `Infinity`. */
103
+ costCeiling?: number;
104
+ /** Fires once per executed cell, after its promise settles. */
105
+ onCellComplete?: (cell: MatrixCell, result: CellResult<Output>) => void;
106
+ /** External cancellation. Aborts in-flight cells via a forwarded signal
107
+ * and suppresses scheduling of new ones. */
108
+ signal?: AbortSignal;
109
+ }
110
+
111
+ /**
112
+ * Per-axis aggregation of cell runs into `AxisSummary` rows.
113
+ *
114
+ * Pure: consumes the final `cells: [{cell, runs}]` array and returns the
115
+ * `byAxis` table. Error runs contribute 0 to passRate and meanScore. Cost
116
+ * and duration always count — the budget was spent regardless.
117
+ */
118
+
119
+ interface Row<Output> {
120
+ cell: MatrixCell;
121
+ result: CellResult<Output>;
122
+ }
123
+ declare function summariseRows<Output>(rows: Row<Output>[], axisName: string, axisValue: string): AxisSummary;
124
+ declare function buildByAxis<Output>(cells: MatrixResult<Output>['cells'], axes: MatrixAxis<unknown>[], aggregateBy: string[]): Record<string, Record<string, AxisSummary>>;
125
+
126
+ /**
127
+ * N-axis cartesian runner.
128
+ *
129
+ * Expansion order: cartesian over `axes` in declared order, then `reps` as the
130
+ * inner-most dim → `ordinal = (cartIdx * reps) + rep`. The returned
131
+ * `cells[]` is sorted by `ordinal` so concurrent execution does not reorder
132
+ * the output.
133
+ *
134
+ * Scheduling is a sliding window of in-flight promises capped at
135
+ * `maxConcurrency`. The window stops admitting new cells when the cost
136
+ * ceiling trips or the abort signal fires; in-flight cells finish.
137
+ */
138
+
139
+ declare function runAgentMatrix<Output>(opts: RunAgentMatrixOptions<Output>): Promise<MatrixResult<Output>>;
140
+
141
+ export { type AxisSummary, type CellResult, type MatrixAxis, type MatrixCell, type MatrixResult, type RunAgentMatrixOptions, buildByAxis, runAgentMatrix, summariseRows };
@@ -0,0 +1,277 @@
1
+ import "../chunk-PZ5AY32C.js";
2
+
3
+ // src/matrix/aggregation.ts
4
+ function flattenRuns(cells) {
5
+ const rows = [];
6
+ for (const { cell, runs } of cells) {
7
+ for (const result of runs) rows.push({ cell, result });
8
+ }
9
+ return rows;
10
+ }
11
+ function quantile(sorted, q) {
12
+ if (sorted.length === 0) return 0;
13
+ if (sorted.length === 1) return sorted[0];
14
+ const pos = (sorted.length - 1) * q;
15
+ const lo = Math.floor(pos);
16
+ const hi = Math.ceil(pos);
17
+ if (lo === hi) return sorted[lo];
18
+ const frac = pos - lo;
19
+ return sorted[lo] * (1 - frac) + sorted[hi] * frac;
20
+ }
21
+ function summariseRows(rows, axisName, axisValue) {
22
+ if (rows.length === 0) {
23
+ return {
24
+ axisName,
25
+ axisValue,
26
+ cells: 0,
27
+ passRate: 0,
28
+ meanScore: 0,
29
+ p50Score: 0,
30
+ p90Score: 0,
31
+ totalCostUsd: 0,
32
+ meanDurationMs: 0
33
+ };
34
+ }
35
+ let pass = 0;
36
+ let scoreSum = 0;
37
+ let costSum = 0;
38
+ let durSum = 0;
39
+ const scores = [];
40
+ for (const { result } of rows) {
41
+ const errored = result.error !== void 0;
42
+ const score = errored ? 0 : result.verdict.score;
43
+ const valid = !errored && result.verdict.valid;
44
+ if (valid) pass++;
45
+ scoreSum += score;
46
+ scores.push(score);
47
+ costSum += result.costUsd;
48
+ durSum += result.durationMs;
49
+ }
50
+ scores.sort((a, b) => a - b);
51
+ return {
52
+ axisName,
53
+ axisValue,
54
+ cells: rows.length,
55
+ passRate: pass / rows.length,
56
+ meanScore: scoreSum / rows.length,
57
+ p50Score: quantile(scores, 0.5),
58
+ p90Score: quantile(scores, 0.9),
59
+ totalCostUsd: costSum,
60
+ meanDurationMs: durSum / rows.length
61
+ };
62
+ }
63
+ function bucketBy(rows, axisName, labelFor) {
64
+ const buckets = /* @__PURE__ */ new Map();
65
+ for (const row of rows) {
66
+ const slot = row.cell.axes[axisName];
67
+ if (!slot) continue;
68
+ const id = slot.id;
69
+ let arr = buckets.get(id);
70
+ if (!arr) {
71
+ arr = [];
72
+ buckets.set(id, arr);
73
+ }
74
+ arr.push(row);
75
+ }
76
+ const out = {};
77
+ for (const id of [...buckets.keys()].sort()) {
78
+ out[id] = summariseRows(buckets.get(id), axisName, labelFor(id));
79
+ }
80
+ return out;
81
+ }
82
+ function buildByAxis(cells, axes, aggregateBy) {
83
+ const rows = flattenRuns(cells);
84
+ const byName = new Map(axes.map((a) => [a.name, a]));
85
+ const byAxis = {};
86
+ for (const name of aggregateBy) {
87
+ const axis = byName.get(name);
88
+ const labelFor = (id) => {
89
+ if (!axis?.label) return id;
90
+ const found = axis.values.find((v) => v.id === id);
91
+ if (!found) return id;
92
+ return axis.label(found.value, id);
93
+ };
94
+ byAxis[name] = bucketBy(rows, name, labelFor);
95
+ }
96
+ return byAxis;
97
+ }
98
+
99
+ // src/matrix/runner.ts
100
+ function cartesian(axes) {
101
+ if (axes.length === 0) return [{ axes: {} }];
102
+ for (const a of axes) if (a.values.length === 0) return [];
103
+ const out = [];
104
+ const idx = new Array(axes.length).fill(0);
105
+ while (true) {
106
+ const slot = {};
107
+ for (let i2 = 0; i2 < axes.length; i2++) {
108
+ const axis = axes[i2];
109
+ const v = axis.values[idx[i2]];
110
+ slot[axis.name] = { id: v.id, value: v.value };
111
+ }
112
+ out.push({ axes: slot });
113
+ let i = 0;
114
+ while (i < axes.length) {
115
+ const next = idx[i] + 1;
116
+ const axis = axes[i];
117
+ if (next < axis.values.length) {
118
+ idx[i] = next;
119
+ break;
120
+ }
121
+ idx[i] = 0;
122
+ i++;
123
+ }
124
+ if (i === axes.length) break;
125
+ }
126
+ return out;
127
+ }
128
+ function makeMatrixId() {
129
+ const t = Date.now().toString(36);
130
+ let r = "";
131
+ for (let i = 0; i < 8; i++) r += Math.floor(Math.random() * 16).toString(16);
132
+ return `mtx_${t}_${r}`;
133
+ }
134
+ function makeErrorResult(err) {
135
+ const e = err;
136
+ return {
137
+ output: void 0,
138
+ verdict: { valid: false, score: 0 },
139
+ costUsd: 0,
140
+ durationMs: 0,
141
+ error: {
142
+ message: typeof e?.message === "string" ? e.message : String(err),
143
+ kind: typeof e?.name === "string" ? e.name : "Error"
144
+ }
145
+ };
146
+ }
147
+ async function runAgentMatrix(opts) {
148
+ const startedAt = Date.now();
149
+ const reps = Math.max(1, opts.reps ?? 1);
150
+ const maxConcurrency = Math.max(1, opts.maxConcurrency ?? 4);
151
+ const costCeiling = opts.costCeiling ?? Number.POSITIVE_INFINITY;
152
+ const aggregateBy = opts.aggregateBy ?? opts.axes.map((a) => a.name);
153
+ const base = cartesian(opts.axes);
154
+ const filtered = opts.filter ? base.filter((c) => opts.filter(c)) : base;
155
+ const filteredOut = base.length - filtered.length;
156
+ const planned = [];
157
+ for (let i = 0; i < filtered.length; i++) {
158
+ for (let r = 0; r < reps; r++) {
159
+ planned.push({
160
+ axes: filtered[i].axes,
161
+ rep: r,
162
+ ordinal: i * reps + r
163
+ });
164
+ }
165
+ }
166
+ const cellRecords = [];
167
+ let cumulativeCost = 0;
168
+ let costCeilingReached = false;
169
+ let runsExecuted = 0;
170
+ let cellsUnscheduled = 0;
171
+ const aborted = () => opts.signal?.aborted === true;
172
+ let inFlight = 0;
173
+ let cursor = 0;
174
+ let resolveAll;
175
+ const done = new Promise((res) => {
176
+ resolveAll = res;
177
+ });
178
+ const pump = () => {
179
+ while (inFlight < maxConcurrency && cursor < planned.length) {
180
+ if (aborted() || costCeilingReached) {
181
+ const left = planned.length - cursor;
182
+ cellsUnscheduled += left;
183
+ cursor = planned.length;
184
+ break;
185
+ }
186
+ const cell = planned[cursor++];
187
+ inFlight++;
188
+ const record = { cell, runs: [] };
189
+ cellRecords.push(record);
190
+ const promise = (async () => {
191
+ try {
192
+ return await opts.runCell(cell);
193
+ } catch (err) {
194
+ return makeErrorResult(err);
195
+ }
196
+ })();
197
+ promise.then((result) => {
198
+ record.runs.push(result);
199
+ runsExecuted++;
200
+ cumulativeCost += result.costUsd;
201
+ if (cumulativeCost >= costCeiling && !costCeilingReached) {
202
+ costCeilingReached = true;
203
+ console.warn("[matrix] cost ceiling reached");
204
+ }
205
+ try {
206
+ opts.onCellComplete?.(cell, result);
207
+ } catch {
208
+ }
209
+ inFlight--;
210
+ if (cursor < planned.length) {
211
+ pump();
212
+ } else if (inFlight === 0) {
213
+ resolveAll?.();
214
+ }
215
+ });
216
+ }
217
+ if (cursor >= planned.length && inFlight === 0) resolveAll?.();
218
+ };
219
+ const onAbort = () => {
220
+ if (cursor < planned.length) {
221
+ cellsUnscheduled += planned.length - cursor;
222
+ cursor = planned.length;
223
+ }
224
+ if (inFlight === 0) resolveAll?.();
225
+ };
226
+ if (opts.signal) {
227
+ if (opts.signal.aborted) {
228
+ cellsUnscheduled = planned.length;
229
+ cursor = planned.length;
230
+ resolveAll?.();
231
+ } else {
232
+ opts.signal.addEventListener("abort", onAbort, { once: true });
233
+ }
234
+ }
235
+ if (planned.length === 0) {
236
+ resolveAll?.();
237
+ } else {
238
+ pump();
239
+ }
240
+ await done;
241
+ if (opts.signal) opts.signal.removeEventListener("abort", onAbort);
242
+ cellRecords.sort((a, b) => a.cell.ordinal - b.cell.ordinal);
243
+ let pass = 0;
244
+ let scoreSum = 0;
245
+ let totalCost = 0;
246
+ let runCount = 0;
247
+ for (const { runs } of cellRecords) {
248
+ for (const r of runs) {
249
+ runCount++;
250
+ const errored = r.error !== void 0;
251
+ if (!errored && r.verdict.valid) pass++;
252
+ scoreSum += errored ? 0 : r.verdict.score;
253
+ totalCost += r.costUsd;
254
+ }
255
+ }
256
+ const byAxis = buildByAxis(cellRecords, opts.axes, aggregateBy);
257
+ return {
258
+ cells: cellRecords,
259
+ byAxis,
260
+ summary: {
261
+ totalCells: planned.length,
262
+ runsExecuted,
263
+ cellsSkipped: cellsUnscheduled + filteredOut * reps,
264
+ overallPassRate: runCount === 0 ? 0 : pass / runCount,
265
+ overallMeanScore: runCount === 0 ? 0 : scoreSum / runCount,
266
+ totalCostUsd: totalCost,
267
+ durationMs: Date.now() - startedAt
268
+ },
269
+ matrixId: makeMatrixId()
270
+ };
271
+ }
272
+ export {
273
+ buildByAxis,
274
+ runAgentMatrix,
275
+ summariseRows
276
+ };
277
+ //# sourceMappingURL=index.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"sources":["../../src/matrix/aggregation.ts","../../src/matrix/runner.ts"],"sourcesContent":["/**\n * Per-axis aggregation of cell runs into `AxisSummary` rows.\n *\n * Pure: consumes the final `cells: [{cell, runs}]` array and returns the\n * `byAxis` table. Error runs contribute 0 to passRate and meanScore. Cost\n * and duration always count — the budget was spent regardless.\n */\n\nimport type { AxisSummary, CellResult, MatrixAxis, MatrixCell, MatrixResult } from './types'\n\ninterface Row<Output> {\n cell: MatrixCell\n result: CellResult<Output>\n}\n\nfunction flattenRuns<Output>(cells: MatrixResult<Output>['cells']): Row<Output>[] {\n const rows: Row<Output>[] = []\n for (const { cell, runs } of cells) {\n for (const result of runs) rows.push({ cell, result })\n }\n return rows\n}\n\nfunction quantile(sorted: number[], q: number): number {\n if (sorted.length === 0) return 0\n if (sorted.length === 1) return sorted[0] as number\n const pos = (sorted.length - 1) * q\n const lo = Math.floor(pos)\n const hi = Math.ceil(pos)\n if (lo === hi) return sorted[lo] as number\n const frac = pos - lo\n return (sorted[lo] as number) * (1 - frac) + (sorted[hi] as number) * frac\n}\n\nexport function summariseRows<Output>(\n rows: Row<Output>[],\n axisName: string,\n axisValue: string,\n): AxisSummary {\n if (rows.length === 0) {\n return {\n axisName,\n axisValue,\n cells: 0,\n passRate: 0,\n meanScore: 0,\n p50Score: 0,\n p90Score: 0,\n totalCostUsd: 0,\n meanDurationMs: 0,\n }\n }\n let pass = 0\n let scoreSum = 0\n let costSum = 0\n let durSum = 0\n const scores: number[] = []\n for (const { result } of rows) {\n const errored = result.error !== undefined\n const score = errored ? 0 : result.verdict.score\n const valid = !errored && result.verdict.valid\n if (valid) pass++\n scoreSum += score\n scores.push(score)\n costSum += result.costUsd\n durSum += result.durationMs\n }\n scores.sort((a, b) => a - b)\n return {\n axisName,\n axisValue,\n cells: rows.length,\n passRate: pass / rows.length,\n meanScore: scoreSum / rows.length,\n p50Score: quantile(scores, 0.5),\n p90Score: quantile(scores, 0.9),\n totalCostUsd: costSum,\n meanDurationMs: durSum / rows.length,\n }\n}\n\nfunction bucketBy<Output>(\n rows: Row<Output>[],\n axisName: string,\n labelFor: (id: string) => string,\n): Record<string, AxisSummary> {\n const buckets = new Map<string, Row<Output>[]>()\n for (const row of rows) {\n const slot = row.cell.axes[axisName]\n if (!slot) continue\n const id = slot.id\n let arr = buckets.get(id)\n if (!arr) {\n arr = []\n buckets.set(id, arr)\n }\n arr.push(row)\n }\n const out: Record<string, AxisSummary> = {}\n // Sorted keys for deterministic JSON serialisation.\n for (const id of [...buckets.keys()].sort()) {\n out[id] = summariseRows(buckets.get(id) as Row<Output>[], axisName, labelFor(id))\n }\n return out\n}\n\nexport function buildByAxis<Output>(\n cells: MatrixResult<Output>['cells'],\n axes: MatrixAxis<unknown>[],\n aggregateBy: string[],\n): Record<string, Record<string, AxisSummary>> {\n const rows = flattenRuns(cells)\n const byName = new Map(axes.map((a) => [a.name, a]))\n const byAxis: Record<string, Record<string, AxisSummary>> = {}\n for (const name of aggregateBy) {\n const axis = byName.get(name)\n const labelFor = (id: string): string => {\n if (!axis?.label) return id\n const found = axis.values.find((v) => v.id === id)\n if (!found) return id\n return axis.label(found.value, id)\n }\n byAxis[name] = bucketBy(rows, name, labelFor)\n }\n return byAxis\n}\n","/**\n * N-axis cartesian runner.\n *\n * Expansion order: cartesian over `axes` in declared order, then `reps` as the\n * inner-most dim → `ordinal = (cartIdx * reps) + rep`. The returned\n * `cells[]` is sorted by `ordinal` so concurrent execution does not reorder\n * the output.\n *\n * Scheduling is a sliding window of in-flight promises capped at\n * `maxConcurrency`. The window stops admitting new cells when the cost\n * ceiling trips or the abort signal fires; in-flight cells finish.\n */\n\nimport { buildByAxis } from './aggregation'\nimport type {\n CellResult,\n MatrixAxis,\n MatrixCell,\n MatrixResult,\n RunAgentMatrixOptions,\n} from './types'\n\ninterface BaseCell {\n axes: Record<string, { id: string; value: unknown }>\n}\n\nfunction cartesian(axes: MatrixAxis<unknown>[]): BaseCell[] {\n // Empty axes (`values=[]`) collapse the whole product to zero cells. An\n // empty `axes` array yields a single empty-axes cell — degenerate but\n // valid (caller is iterating only reps).\n if (axes.length === 0) return [{ axes: {} }]\n for (const a of axes) if (a.values.length === 0) return []\n const out: BaseCell[] = []\n const idx = new Array(axes.length).fill(0)\n while (true) {\n const slot: Record<string, { id: string; value: unknown }> = {}\n for (let i = 0; i < axes.length; i++) {\n const axis = axes[i] as MatrixAxis<unknown>\n const v = axis.values[idx[i] as number] as { id: string; value: unknown }\n slot[axis.name] = { id: v.id, value: v.value }\n }\n out.push({ axes: slot })\n // Increment like an odometer, left-most axis is fastest.\n let i = 0\n while (i < axes.length) {\n const next = (idx[i] as number) + 1\n const axis = axes[i] as MatrixAxis<unknown>\n if (next < axis.values.length) {\n idx[i] = next\n break\n }\n idx[i] = 0\n i++\n }\n if (i === axes.length) break\n }\n return out\n}\n\nfunction makeMatrixId(): string {\n // Stable id-like string: time + 8 random hex chars. Avoids node:crypto\n // import to keep the matrix dep-free.\n const t = Date.now().toString(36)\n let r = ''\n for (let i = 0; i < 8; i++) r += Math.floor(Math.random() * 16).toString(16)\n return `mtx_${t}_${r}`\n}\n\nfunction makeErrorResult<Output>(err: unknown): CellResult<Output> {\n const e = err as { message?: string; name?: string }\n return {\n output: undefined as unknown as Output,\n verdict: { valid: false, score: 0 },\n costUsd: 0,\n durationMs: 0,\n error: {\n message: typeof e?.message === 'string' ? e.message : String(err),\n kind: typeof e?.name === 'string' ? e.name : 'Error',\n },\n }\n}\n\nexport async function runAgentMatrix<Output>(\n opts: RunAgentMatrixOptions<Output>,\n): Promise<MatrixResult<Output>> {\n const startedAt = Date.now()\n const reps = Math.max(1, opts.reps ?? 1)\n const maxConcurrency = Math.max(1, opts.maxConcurrency ?? 4)\n const costCeiling = opts.costCeiling ?? Number.POSITIVE_INFINITY\n const aggregateBy = opts.aggregateBy ?? opts.axes.map((a) => a.name)\n\n const base = cartesian(opts.axes)\n const filtered = opts.filter\n ? base.filter((c) => (opts.filter as (b: BaseCell) => boolean)(c))\n : base\n const filteredOut = base.length - filtered.length\n\n const planned: MatrixCell[] = []\n for (let i = 0; i < filtered.length; i++) {\n for (let r = 0; r < reps; r++) {\n planned.push({\n axes: (filtered[i] as BaseCell).axes,\n rep: r,\n ordinal: i * reps + r,\n })\n }\n }\n\n const cellRecords: Array<{ cell: MatrixCell; runs: CellResult<Output>[] }> = []\n let cumulativeCost = 0\n let costCeilingReached = false\n let runsExecuted = 0\n let cellsUnscheduled = 0\n\n const aborted = (): boolean => opts.signal?.aborted === true\n\n // Per-run abort controller forwards the external signal so cell executors\n // see cancellation. We don't expose it on `MatrixCell` — the signature on\n // `runCell` per the public API is `(cell) => Promise<...>`. Executors that\n // need cancellation use the external signal directly via closure.\n\n let inFlight = 0\n let cursor = 0\n let resolveAll: (() => void) | undefined\n const done = new Promise<void>((res) => {\n resolveAll = res\n })\n\n const pump = (): void => {\n while (inFlight < maxConcurrency && cursor < planned.length) {\n if (aborted() || costCeilingReached) {\n // Drain remaining as unscheduled.\n const left = planned.length - cursor\n cellsUnscheduled += left\n cursor = planned.length\n break\n }\n const cell = planned[cursor++] as MatrixCell\n inFlight++\n // Lazily allocate the record so cells appear in `cells[]` in any\n // arrival order; we sort by ordinal at the end.\n const record = { cell, runs: [] as CellResult<Output>[] }\n cellRecords.push(record)\n const promise: Promise<CellResult<Output>> = (async () => {\n try {\n return await opts.runCell(cell)\n } catch (err) {\n return makeErrorResult<Output>(err)\n }\n })()\n promise.then((result) => {\n record.runs.push(result)\n runsExecuted++\n cumulativeCost += result.costUsd\n if (cumulativeCost >= costCeiling && !costCeilingReached) {\n costCeilingReached = true\n // eslint-disable-next-line no-console\n console.warn('[matrix] cost ceiling reached')\n }\n try {\n opts.onCellComplete?.(cell, result)\n } catch {\n // onCellComplete is observational — swallow throws so a noisy\n // callback can't tank the run.\n }\n inFlight--\n if (cursor < planned.length) {\n pump()\n } else if (inFlight === 0) {\n resolveAll?.()\n }\n })\n }\n if (cursor >= planned.length && inFlight === 0) resolveAll?.()\n }\n\n const onAbort = (): void => {\n // External abort: stop scheduling. In-flight cells finish; their\n // executors observe `opts.signal.aborted` directly via closure.\n if (cursor < planned.length) {\n cellsUnscheduled += planned.length - cursor\n cursor = planned.length\n }\n if (inFlight === 0) resolveAll?.()\n }\n if (opts.signal) {\n if (opts.signal.aborted) {\n cellsUnscheduled = planned.length\n cursor = planned.length\n resolveAll?.()\n } else {\n opts.signal.addEventListener('abort', onAbort, { once: true })\n }\n }\n\n if (planned.length === 0) {\n resolveAll?.()\n } else {\n pump()\n }\n\n await done\n if (opts.signal) opts.signal.removeEventListener('abort', onAbort)\n\n cellRecords.sort((a, b) => a.cell.ordinal - b.cell.ordinal)\n\n let pass = 0\n let scoreSum = 0\n let totalCost = 0\n let runCount = 0\n for (const { runs } of cellRecords) {\n for (const r of runs) {\n runCount++\n const errored = r.error !== undefined\n if (!errored && r.verdict.valid) pass++\n scoreSum += errored ? 0 : r.verdict.score\n totalCost += r.costUsd\n }\n }\n\n const byAxis = buildByAxis(cellRecords, opts.axes, aggregateBy)\n\n return {\n cells: cellRecords,\n byAxis,\n summary: {\n totalCells: planned.length,\n runsExecuted,\n cellsSkipped: cellsUnscheduled + filteredOut * reps,\n overallPassRate: runCount === 0 ? 0 : pass / runCount,\n overallMeanScore: runCount === 0 ? 0 : scoreSum / runCount,\n totalCostUsd: totalCost,\n durationMs: Date.now() - startedAt,\n },\n matrixId: makeMatrixId(),\n }\n}\n"],"mappings":";;;AAeA,SAAS,YAAoB,OAAqD;AAChF,QAAM,OAAsB,CAAC;AAC7B,aAAW,EAAE,MAAM,KAAK,KAAK,OAAO;AAClC,eAAW,UAAU,KAAM,MAAK,KAAK,EAAE,MAAM,OAAO,CAAC;AAAA,EACvD;AACA,SAAO;AACT;AAEA,SAAS,SAAS,QAAkB,GAAmB;AACrD,MAAI,OAAO,WAAW,EAAG,QAAO;AAChC,MAAI,OAAO,WAAW,EAAG,QAAO,OAAO,CAAC;AACxC,QAAM,OAAO,OAAO,SAAS,KAAK;AAClC,QAAM,KAAK,KAAK,MAAM,GAAG;AACzB,QAAM,KAAK,KAAK,KAAK,GAAG;AACxB,MAAI,OAAO,GAAI,QAAO,OAAO,EAAE;AAC/B,QAAM,OAAO,MAAM;AACnB,SAAQ,OAAO,EAAE,KAAgB,IAAI,QAAS,OAAO,EAAE,IAAe;AACxE;AAEO,SAAS,cACd,MACA,UACA,WACa;AACb,MAAI,KAAK,WAAW,GAAG;AACrB,WAAO;AAAA,MACL;AAAA,MACA;AAAA,MACA,OAAO;AAAA,MACP,UAAU;AAAA,MACV,WAAW;AAAA,MACX,UAAU;AAAA,MACV,UAAU;AAAA,MACV,cAAc;AAAA,MACd,gBAAgB;AAAA,IAClB;AAAA,EACF;AACA,MAAI,OAAO;AACX,MAAI,WAAW;AACf,MAAI,UAAU;AACd,MAAI,SAAS;AACb,QAAM,SAAmB,CAAC;AAC1B,aAAW,EAAE,OAAO,KAAK,MAAM;AAC7B,UAAM,UAAU,OAAO,UAAU;AACjC,UAAM,QAAQ,UAAU,IAAI,OAAO,QAAQ;AAC3C,UAAM,QAAQ,CAAC,WAAW,OAAO,QAAQ;AACzC,QAAI,MAAO;AACX,gBAAY;AACZ,WAAO,KAAK,KAAK;AACjB,eAAW,OAAO;AAClB,cAAU,OAAO;AAAA,EACnB;AACA,SAAO,KAAK,CAAC,GAAG,MAAM,IAAI,CAAC;AAC3B,SAAO;AAAA,IACL;AAAA,IACA;AAAA,IACA,OAAO,KAAK;AAAA,IACZ,UAAU,OAAO,KAAK;AAAA,IACtB,WAAW,WAAW,KAAK;AAAA,IAC3B,UAAU,SAAS,QAAQ,GAAG;AAAA,IAC9B,UAAU,SAAS,QAAQ,GAAG;AAAA,IAC9B,cAAc;AAAA,IACd,gBAAgB,SAAS,KAAK;AAAA,EAChC;AACF;AAEA,SAAS,SACP,MACA,UACA,UAC6B;AAC7B,QAAM,UAAU,oBAAI,IAA2B;AAC/C,aAAW,OAAO,MAAM;AACtB,UAAM,OAAO,IAAI,KAAK,KAAK,QAAQ;AACnC,QAAI,CAAC,KAAM;AACX,UAAM,KAAK,KAAK;AAChB,QAAI,MAAM,QAAQ,IAAI,EAAE;AACxB,QAAI,CAAC,KAAK;AACR,YAAM,CAAC;AACP,cAAQ,IAAI,IAAI,GAAG;AAAA,IACrB;AACA,QAAI,KAAK,GAAG;AAAA,EACd;AACA,QAAM,MAAmC,CAAC;AAE1C,aAAW,MAAM,CAAC,GAAG,QAAQ,KAAK,CAAC,EAAE,KAAK,GAAG;AAC3C,QAAI,EAAE,IAAI,cAAc,QAAQ,IAAI,EAAE,GAAoB,UAAU,SAAS,EAAE,CAAC;AAAA,EAClF;AACA,SAAO;AACT;AAEO,SAAS,YACd,OACA,MACA,aAC6C;AAC7C,QAAM,OAAO,YAAY,KAAK;AAC9B,QAAM,SAAS,IAAI,IAAI,KAAK,IAAI,CAAC,MAAM,CAAC,EAAE,MAAM,CAAC,CAAC,CAAC;AACnD,QAAM,SAAsD,CAAC;AAC7D,aAAW,QAAQ,aAAa;AAC9B,UAAM,OAAO,OAAO,IAAI,IAAI;AAC5B,UAAM,WAAW,CAAC,OAAuB;AACvC,UAAI,CAAC,MAAM,MAAO,QAAO;AACzB,YAAM,QAAQ,KAAK,OAAO,KAAK,CAAC,MAAM,EAAE,OAAO,EAAE;AACjD,UAAI,CAAC,MAAO,QAAO;AACnB,aAAO,KAAK,MAAM,MAAM,OAAO,EAAE;AAAA,IACnC;AACA,WAAO,IAAI,IAAI,SAAS,MAAM,MAAM,QAAQ;AAAA,EAC9C;AACA,SAAO;AACT;;;ACnGA,SAAS,UAAU,MAAyC;AAI1D,MAAI,KAAK,WAAW,EAAG,QAAO,CAAC,EAAE,MAAM,CAAC,EAAE,CAAC;AAC3C,aAAW,KAAK,KAAM,KAAI,EAAE,OAAO,WAAW,EAAG,QAAO,CAAC;AACzD,QAAM,MAAkB,CAAC;AACzB,QAAM,MAAM,IAAI,MAAM,KAAK,MAAM,EAAE,KAAK,CAAC;AACzC,SAAO,MAAM;AACX,UAAM,OAAuD,CAAC;AAC9D,aAASA,KAAI,GAAGA,KAAI,KAAK,QAAQA,MAAK;AACpC,YAAM,OAAO,KAAKA,EAAC;AACnB,YAAM,IAAI,KAAK,OAAO,IAAIA,EAAC,CAAW;AACtC,WAAK,KAAK,IAAI,IAAI,EAAE,IAAI,EAAE,IAAI,OAAO,EAAE,MAAM;AAAA,IAC/C;AACA,QAAI,KAAK,EAAE,MAAM,KAAK,CAAC;AAEvB,QAAI,IAAI;AACR,WAAO,IAAI,KAAK,QAAQ;AACtB,YAAM,OAAQ,IAAI,CAAC,IAAe;AAClC,YAAM,OAAO,KAAK,CAAC;AACnB,UAAI,OAAO,KAAK,OAAO,QAAQ;AAC7B,YAAI,CAAC,IAAI;AACT;AAAA,MACF;AACA,UAAI,CAAC,IAAI;AACT;AAAA,IACF;AACA,QAAI,MAAM,KAAK,OAAQ;AAAA,EACzB;AACA,SAAO;AACT;AAEA,SAAS,eAAuB;AAG9B,QAAM,IAAI,KAAK,IAAI,EAAE,SAAS,EAAE;AAChC,MAAI,IAAI;AACR,WAAS,IAAI,GAAG,IAAI,GAAG,IAAK,MAAK,KAAK,MAAM,KAAK,OAAO,IAAI,EAAE,EAAE,SAAS,EAAE;AAC3E,SAAO,OAAO,CAAC,IAAI,CAAC;AACtB;AAEA,SAAS,gBAAwB,KAAkC;AACjE,QAAM,IAAI;AACV,SAAO;AAAA,IACL,QAAQ;AAAA,IACR,SAAS,EAAE,OAAO,OAAO,OAAO,EAAE;AAAA,IAClC,SAAS;AAAA,IACT,YAAY;AAAA,IACZ,OAAO;AAAA,MACL,SAAS,OAAO,GAAG,YAAY,WAAW,EAAE,UAAU,OAAO,GAAG;AAAA,MAChE,MAAM,OAAO,GAAG,SAAS,WAAW,EAAE,OAAO;AAAA,IAC/C;AAAA,EACF;AACF;AAEA,eAAsB,eACpB,MAC+B;AAC/B,QAAM,YAAY,KAAK,IAAI;AAC3B,QAAM,OAAO,KAAK,IAAI,GAAG,KAAK,QAAQ,CAAC;AACvC,QAAM,iBAAiB,KAAK,IAAI,GAAG,KAAK,kBAAkB,CAAC;AAC3D,QAAM,cAAc,KAAK,eAAe,OAAO;AAC/C,QAAM,cAAc,KAAK,eAAe,KAAK,KAAK,IAAI,CAAC,MAAM,EAAE,IAAI;AAEnE,QAAM,OAAO,UAAU,KAAK,IAAI;AAChC,QAAM,WAAW,KAAK,SAClB,KAAK,OAAO,CAAC,MAAO,KAAK,OAAoC,CAAC,CAAC,IAC/D;AACJ,QAAM,cAAc,KAAK,SAAS,SAAS;AAE3C,QAAM,UAAwB,CAAC;AAC/B,WAAS,IAAI,GAAG,IAAI,SAAS,QAAQ,KAAK;AACxC,aAAS,IAAI,GAAG,IAAI,MAAM,KAAK;AAC7B,cAAQ,KAAK;AAAA,QACX,MAAO,SAAS,CAAC,EAAe;AAAA,QAChC,KAAK;AAAA,QACL,SAAS,IAAI,OAAO;AAAA,MACtB,CAAC;AAAA,IACH;AAAA,EACF;AAEA,QAAM,cAAuE,CAAC;AAC9E,MAAI,iBAAiB;AACrB,MAAI,qBAAqB;AACzB,MAAI,eAAe;AACnB,MAAI,mBAAmB;AAEvB,QAAM,UAAU,MAAe,KAAK,QAAQ,YAAY;AAOxD,MAAI,WAAW;AACf,MAAI,SAAS;AACb,MAAI;AACJ,QAAM,OAAO,IAAI,QAAc,CAAC,QAAQ;AACtC,iBAAa;AAAA,EACf,CAAC;AAED,QAAM,OAAO,MAAY;AACvB,WAAO,WAAW,kBAAkB,SAAS,QAAQ,QAAQ;AAC3D,UAAI,QAAQ,KAAK,oBAAoB;AAEnC,cAAM,OAAO,QAAQ,SAAS;AAC9B,4BAAoB;AACpB,iBAAS,QAAQ;AACjB;AAAA,MACF;AACA,YAAM,OAAO,QAAQ,QAAQ;AAC7B;AAGA,YAAM,SAAS,EAAE,MAAM,MAAM,CAAC,EAA0B;AACxD,kBAAY,KAAK,MAAM;AACvB,YAAM,WAAwC,YAAY;AACxD,YAAI;AACF,iBAAO,MAAM,KAAK,QAAQ,IAAI;AAAA,QAChC,SAAS,KAAK;AACZ,iBAAO,gBAAwB,GAAG;AAAA,QACpC;AAAA,MACF,GAAG;AACH,cAAQ,KAAK,CAAC,WAAW;AACvB,eAAO,KAAK,KAAK,MAAM;AACvB;AACA,0BAAkB,OAAO;AACzB,YAAI,kBAAkB,eAAe,CAAC,oBAAoB;AACxD,+BAAqB;AAErB,kBAAQ,KAAK,+BAA+B;AAAA,QAC9C;AACA,YAAI;AACF,eAAK,iBAAiB,MAAM,MAAM;AAAA,QACpC,QAAQ;AAAA,QAGR;AACA;AACA,YAAI,SAAS,QAAQ,QAAQ;AAC3B,eAAK;AAAA,QACP,WAAW,aAAa,GAAG;AACzB,uBAAa;AAAA,QACf;AAAA,MACF,CAAC;AAAA,IACH;AACA,QAAI,UAAU,QAAQ,UAAU,aAAa,EAAG,cAAa;AAAA,EAC/D;AAEA,QAAM,UAAU,MAAY;AAG1B,QAAI,SAAS,QAAQ,QAAQ;AAC3B,0BAAoB,QAAQ,SAAS;AACrC,eAAS,QAAQ;AAAA,IACnB;AACA,QAAI,aAAa,EAAG,cAAa;AAAA,EACnC;AACA,MAAI,KAAK,QAAQ;AACf,QAAI,KAAK,OAAO,SAAS;AACvB,yBAAmB,QAAQ;AAC3B,eAAS,QAAQ;AACjB,mBAAa;AAAA,IACf,OAAO;AACL,WAAK,OAAO,iBAAiB,SAAS,SAAS,EAAE,MAAM,KAAK,CAAC;AAAA,IAC/D;AAAA,EACF;AAEA,MAAI,QAAQ,WAAW,GAAG;AACxB,iBAAa;AAAA,EACf,OAAO;AACL,SAAK;AAAA,EACP;AAEA,QAAM;AACN,MAAI,KAAK,OAAQ,MAAK,OAAO,oBAAoB,SAAS,OAAO;AAEjE,cAAY,KAAK,CAAC,GAAG,MAAM,EAAE,KAAK,UAAU,EAAE,KAAK,OAAO;AAE1D,MAAI,OAAO;AACX,MAAI,WAAW;AACf,MAAI,YAAY;AAChB,MAAI,WAAW;AACf,aAAW,EAAE,KAAK,KAAK,aAAa;AAClC,eAAW,KAAK,MAAM;AACpB;AACA,YAAM,UAAU,EAAE,UAAU;AAC5B,UAAI,CAAC,WAAW,EAAE,QAAQ,MAAO;AACjC,kBAAY,UAAU,IAAI,EAAE,QAAQ;AACpC,mBAAa,EAAE;AAAA,IACjB;AAAA,EACF;AAEA,QAAM,SAAS,YAAY,aAAa,KAAK,MAAM,WAAW;AAE9D,SAAO;AAAA,IACL,OAAO;AAAA,IACP;AAAA,IACA,SAAS;AAAA,MACP,YAAY,QAAQ;AAAA,MACpB;AAAA,MACA,cAAc,mBAAmB,cAAc;AAAA,MAC/C,iBAAiB,aAAa,IAAI,IAAI,OAAO;AAAA,MAC7C,kBAAkB,aAAa,IAAI,IAAI,WAAW;AAAA,MAClD,cAAc;AAAA,MACd,YAAY,KAAK,IAAI,IAAI;AAAA,IAC3B;AAAA,IACA,UAAU,aAAa;AAAA,EACzB;AACF;","names":["i"]}
package/dist/openapi.json CHANGED
@@ -2,7 +2,7 @@
2
2
  "openapi": "3.1.0",
3
3
  "info": {
4
4
  "title": "@tangle-network/agent-eval — wire protocol",
5
- "version": "0.34.1",
5
+ "version": "0.36.0",
6
6
  "description": "HTTP and stdio RPC interface to agent-eval. The TypeScript runtime is the source of truth; this spec is the contract that cross-language clients (Python, Rust, Go) generate from.\n\nWire-protocol version: 1.0.0. Bumps on breaking changes to request/response schemas.",
7
7
  "contact": {
8
8
  "name": "Tangle Network",
@@ -551,5 +551,22 @@ interface ServeOptions extends CreateAppOptions {
551
551
  host?: string;
552
552
  }
553
553
  declare function startServer(opts?: ServeOptions): ServerType;
554
+ interface StartedServer {
555
+ server: ServerType;
556
+ /** The OS-assigned port. When opts.port was 0, this is the actual port the
557
+ * kernel bound — callers that need to dial back (smoke tests, sidecars
558
+ * registering with a parent) read this rather than guessing a free port. */
559
+ port: number;
560
+ /** Resolved host the server bound to (defaults to 127.0.0.1). */
561
+ host: string;
562
+ /** Close the server. Resolves once active connections have drained. */
563
+ close(): Promise<void>;
564
+ }
565
+ /**
566
+ * Promise-returning variant of `startServer` that resolves once the server is
567
+ * listening and surfaces the resolved bound port. Use this from smoke tests
568
+ * (`startServerAsync({ port: 0 })`) and any caller that needs to dial back.
569
+ */
570
+ declare function startServerAsync(opts?: ServeOptions): Promise<StartedServer>;
554
571
 
555
- export { BUILTIN_RUBRICS, type ErrorResponse, ErrorResponseSchema, type FailureMode, FailureModeSchema, FeedbackAttemptSchema, type FeedbackIngestResponse, FeedbackIngestResponseSchema, FeedbackLabelSchema, type FeedbackTrajectory, FeedbackTrajectorySchema, HealthResponseSchema, type IngestionStores, type JudgeRequest, JudgeRequestSchema, type JudgeResult, JudgeResultSchema, type ListRubricsResponse, ListRubricsResponseSchema, type Rubric, type RubricDimension, RubricDimensionSchema, type RubricInfo, RubricInfoSchema, RubricSchema, type ServeOptions, type TraceEvent, TraceEventSchema, type TracesIngestRequest, TracesIngestRequestSchema, type TracesIngestResponse, TracesIngestResponseSchema, type VersionResponse, VersionResponseSchema, WIRE_VERSION, WireError, buildOpenApi, createApp, dispatchRpc, getBuiltinRubric, handleFeedbackIngest, handleJudge, handleListRubrics, handleTracesIngest, handleVersion, hashRubric, listBuiltinRubrics, runRpcBatch, runRpcOnce, startServer };
572
+ export { BUILTIN_RUBRICS, type ErrorResponse, ErrorResponseSchema, type FailureMode, FailureModeSchema, FeedbackAttemptSchema, type FeedbackIngestResponse, FeedbackIngestResponseSchema, FeedbackLabelSchema, type FeedbackTrajectory, FeedbackTrajectorySchema, HealthResponseSchema, type IngestionStores, type JudgeRequest, JudgeRequestSchema, type JudgeResult, JudgeResultSchema, type ListRubricsResponse, ListRubricsResponseSchema, type Rubric, type RubricDimension, RubricDimensionSchema, type RubricInfo, RubricInfoSchema, RubricSchema, type ServeOptions, type StartedServer, type TraceEvent, TraceEventSchema, type TracesIngestRequest, TracesIngestRequestSchema, type TracesIngestResponse, TracesIngestResponseSchema, type VersionResponse, VersionResponseSchema, WIRE_VERSION, WireError, buildOpenApi, createApp, dispatchRpc, getBuiltinRubric, handleFeedbackIngest, handleJudge, handleListRubrics, handleTracesIngest, handleVersion, hashRubric, listBuiltinRubrics, runRpcBatch, runRpcOnce, startServer, startServerAsync };
@@ -32,8 +32,9 @@ import {
32
32
  listBuiltinRubrics,
33
33
  runRpcBatch,
34
34
  runRpcOnce,
35
- startServer
36
- } from "../chunk-YU3G6I7F.js";
35
+ startServer,
36
+ startServerAsync
37
+ } from "../chunk-63EPZQUZ.js";
37
38
  import "../chunk-VXNVVBZO.js";
38
39
  import "../chunk-PC4UYEBM.js";
39
40
  import "../chunk-QYJT52YW.js";
@@ -72,6 +73,7 @@ export {
72
73
  listBuiltinRubrics,
73
74
  runRpcBatch,
74
75
  runRpcOnce,
75
- startServer
76
+ startServer,
77
+ startServerAsync
76
78
  };
77
79
  //# sourceMappingURL=index.js.map
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@tangle-network/agent-eval",
3
- "version": "0.34.1",
3
+ "version": "0.36.0",
4
4
  "description": "Substrate for self-improving agents: traces, verifiable rewards, preferences, GEPA / reflective mutation, auto-research, replay, sequential anytime-valid stats, and release gates.",
5
5
  "homepage": "https://github.com/tangle-network/agent-eval#readme",
6
6
  "repository": {
@@ -94,6 +94,11 @@
94
94
  "import": "./dist/knowledge/index.js",
95
95
  "default": "./dist/knowledge/index.js"
96
96
  },
97
+ "./matrix": {
98
+ "types": "./dist/matrix/index.d.ts",
99
+ "import": "./dist/matrix/index.js",
100
+ "default": "./dist/matrix/index.js"
101
+ },
97
102
  "./openapi.json": {
98
103
  "default": "./dist/openapi.json"
99
104
  }
@@ -109,6 +114,17 @@
109
114
  "publishConfig": {
110
115
  "access": "public"
111
116
  },
117
+ "scripts": {
118
+ "build": "tsup && pnpm openapi",
119
+ "dev": "tsup --watch",
120
+ "prepare": "pnpm build",
121
+ "test": "vitest run",
122
+ "test:watch": "vitest",
123
+ "typecheck": "tsc --noEmit",
124
+ "lint": "biome check src",
125
+ "format": "biome format --write src",
126
+ "openapi": "node dist/cli.js openapi --out dist/openapi.json"
127
+ },
112
128
  "dependencies": {
113
129
  "@asteasolutions/zod-to-openapi": "^8.5.0",
114
130
  "@ax-llm/ax": "^19.0.25",
@@ -117,26 +133,35 @@
117
133
  "hono": "^4.12.16",
118
134
  "zod": "^4.3.6"
119
135
  },
136
+ "peerDependencies": {
137
+ "@tangle-network/agent-runtime": "^0.21.0",
138
+ "@tangle-network/sandbox": "^0.2.1"
139
+ },
140
+ "peerDependenciesMeta": {
141
+ "@tangle-network/agent-runtime": { "optional": true },
142
+ "@tangle-network/sandbox": { "optional": true }
143
+ },
120
144
  "devDependencies": {
121
145
  "@biomejs/biome": "^2.4.15",
146
+ "@tangle-network/agent-runtime": "^0.21.0",
147
+ "@tangle-network/sandbox": "^0.2.1",
122
148
  "@types/node": "^25.6.0",
123
149
  "openapi3-ts": "^4.5.0",
124
150
  "tsup": "^8.0.0",
125
151
  "typescript": "^5.7.0",
126
152
  "vitest": "^3.0.0"
127
153
  },
154
+ "pnpm": {
155
+ "minimumReleaseAge": 4320,
156
+ "minimumReleaseAgeExclude": ["@tangle-network/sandbox", "@tangle-network/agent-runtime"],
157
+ "overrides": {
158
+ "postcss@<8.5.10": "^8.5.10",
159
+ "ws@>=8.0.0 <8.20.1": "^8.20.1"
160
+ }
161
+ },
128
162
  "engines": {
129
163
  "node": ">=20"
130
164
  },
131
165
  "license": "MIT",
132
- "scripts": {
133
- "build": "tsup && pnpm openapi",
134
- "dev": "tsup --watch",
135
- "test": "vitest run",
136
- "test:watch": "vitest",
137
- "typecheck": "tsc --noEmit",
138
- "lint": "biome check src",
139
- "format": "biome format --write src",
140
- "openapi": "node dist/cli.js openapi --out dist/openapi.json"
141
- }
142
- }
166
+ "packageManager": "pnpm@10.22.0"
167
+ }