@tangle-network/agent-eval 0.82.0 → 0.84.0

package/dist/chunk-T4SQEITX.js

@@ -0,0 +1,95 @@
+// src/runtime-trajectory.ts
+var DEFAULT_SPLIT_TAG = "search";
+function projectRuntimeTrajectoryEvidence(options) {
+  const diagnostics = [];
+  const runsById = /* @__PURE__ */ new Map();
+  const events = [];
+  let recordWithRuntimeEventsCount = 0;
+  let defaultedSplitCount = 0;
+  for (let recordIndex = 0; recordIndex < options.records.length; recordIndex += 1) {
+    const record = options.records[recordIndex];
+    const key = runtimeTrajectoryRecordKey(record, recordIndex, options.recordIdOf);
+    const splitTag = record.splitTag ?? options.defaultSplitTag ?? DEFAULT_SPLIT_TAG;
+    if (record.splitTag === void 0) defaultedSplitCount += 1;
+    const rawEvents = record.runtimeEvents;
+    if (!Array.isArray(rawEvents)) {
+      diagnostics.push(
+        `${key}: runtimeEvents is not an array; no runtime run join can be extracted`
+      );
+      continue;
+    }
+    if (rawEvents.length === 0) {
+      diagnostics.push(`${key}: no runtimeEvents; no runtime run join can be extracted`);
+      continue;
+    }
+    recordWithRuntimeEventsCount += 1;
+    for (let index = 0; index < rawEvents.length; index += 1) {
+      const event = parseRuntimeTrajectoryHookEvent(rawEvents[index]);
+      if (!event) {
+        diagnostics.push(`${key}: runtimeEvents[${index}] is not a RuntimeHookEvent`);
+        continue;
+      }
+      events.push(event);
+      const scenarioId = event.scenarioId ?? stringOrUndefined(options.scenarioIdOf?.(record, recordIndex)) ?? stringOrUndefined(record.scenarioId);
+      const prior = runsById.get(event.runId);
+      if (!prior) {
+        runsById.set(event.runId, { runId: event.runId, scenarioId, splitTag });
+        continue;
+      }
+      if (prior.scenarioId !== scenarioId || prior.splitTag !== splitTag) {
+        diagnostics.push(`${key}: runId ${event.runId} has conflicting scenario/split metadata`);
+      }
+    }
+  }
+  const runs = [...runsById.values()];
+  return {
+    runs,
+    events,
+    summary: {
+      recordCount: options.records.length,
+      recordWithRuntimeEventsCount,
+      runtimeRunCount: runs.length,
+      lifecycleEventCount: events.length,
+      defaultedSplitCount
+    },
+    diagnostics
+  };
+}
+function parseRuntimeTrajectoryHookEvent(input) {
+  if (!isRecord(input)) return null;
+  if (typeof input.id !== "string" || input.id.length === 0) return null;
+  if (typeof input.runId !== "string" || input.runId.length === 0) return null;
+  if (typeof input.target !== "string" || input.target.length === 0) return null;
+  if (typeof input.phase !== "string" || input.phase.length === 0) return null;
+  if (typeof input.timestamp !== "number" || !Number.isFinite(input.timestamp)) return null;
+  return {
+    id: input.id,
+    runId: input.runId,
+    scenarioId: stringOrUndefined(input.scenarioId),
+    target: input.target,
+    phase: input.phase,
+    timestamp: input.timestamp,
+    stepIndex: finiteNumberOrUndefined(input.stepIndex),
+    parentId: stringOrUndefined(input.parentId),
+    payload: input.payload,
+    metadata: isRecord(input.metadata) ? { ...input.metadata } : void 0
+  };
+}
+function runtimeTrajectoryRecordKey(record, index, recordIdOf) {
+  return stringOrUndefined(recordIdOf?.(record, index)) ?? stringOrUndefined(record.id) ?? `record[${index}]`;
+}
+function isRecord(value) {
+  return typeof value === "object" && value !== null && !Array.isArray(value);
+}
+function stringOrUndefined(value) {
+  return typeof value === "string" && value.length > 0 ? value : void 0;
+}
+function finiteNumberOrUndefined(value) {
+  return typeof value === "number" && Number.isFinite(value) ? value : void 0;
+}
+
+export {
+  projectRuntimeTrajectoryEvidence,
+  parseRuntimeTrajectoryHookEvent
+};
+//# sourceMappingURL=chunk-T4SQEITX.js.map

package/dist/chunk-T4SQEITX.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/runtime-trajectory.ts"],"sourcesContent":["import type { RunSplitTag } from './run-record'\n\nexport interface RuntimeTrajectoryHookEvent {\n  id: string\n  runId: string\n  scenarioId?: string\n  target: string\n  phase: string\n  timestamp: number\n  stepIndex?: number\n  parentId?: string\n  payload?: unknown\n  metadata?: Record<string, unknown>\n}\n\nexport interface RuntimeTrajectoryRecord {\n  id?: string\n  scenarioId?: string\n  splitTag?: RunSplitTag\n  runtimeEvents?: unknown\n  [key: string]: unknown\n}\n\nexport interface RuntimeTrajectoryRunRecord {\n  runId: string\n  scenarioId?: string\n  splitTag: RunSplitTag\n}\n\nexport interface RuntimeTrajectoryEvidenceSummary {\n  recordCount: number\n  recordWithRuntimeEventsCount: number\n  runtimeRunCount: number\n  lifecycleEventCount: number\n  defaultedSplitCount: number\n}\n\nexport interface RuntimeTrajectoryEvidenceProjection {\n  runs: RuntimeTrajectoryRunRecord[]\n  events: RuntimeTrajectoryHookEvent[]\n  summary: RuntimeTrajectoryEvidenceSummary\n  diagnostics: string[]\n}\n\nexport interface ProjectRuntimeTrajectoryEvidenceOptions<\n  TRecord extends RuntimeTrajectoryRecord = RuntimeTrajectoryRecord,\n> {\n  records: TRecord[]\n  defaultSplitTag?: RunSplitTag\n  recordIdOf?: (record: TRecord, index: number) => string | undefined\n  scenarioIdOf?: (record: TRecord, index: number) => string | undefined\n}\n\nconst DEFAULT_SPLIT_TAG: RunSplitTag = 'search'\n\nexport function projectRuntimeTrajectoryEvidence<TRecord extends RuntimeTrajectoryRecord>(\n  options: ProjectRuntimeTrajectoryEvidenceOptions<TRecord>,\n): RuntimeTrajectoryEvidenceProjection {\n  const diagnostics: string[] = []\n  const runsById = new Map<string, RuntimeTrajectoryRunRecord>()\n  const events: RuntimeTrajectoryHookEvent[] = []\n  let recordWithRuntimeEventsCount = 0\n  let defaultedSplitCount = 0\n\n  for (let recordIndex = 0; recordIndex < options.records.length; recordIndex += 1) {\n    const record = options.records[recordIndex]!\n    const key = runtimeTrajectoryRecordKey(record, recordIndex, options.recordIdOf)\n    const splitTag = record.splitTag ?? options.defaultSplitTag ?? DEFAULT_SPLIT_TAG\n    if (record.splitTag === undefined) defaultedSplitCount += 1\n\n    const rawEvents = record.runtimeEvents\n    if (!Array.isArray(rawEvents)) {\n      diagnostics.push(\n        `${key}: runtimeEvents is not an array; no runtime run join can be extracted`,\n      )\n      continue\n    }\n    if (rawEvents.length === 0) {\n      diagnostics.push(`${key}: no runtimeEvents; no runtime run join can be extracted`)\n      continue\n    }\n    recordWithRuntimeEventsCount += 1\n\n    for (let index = 0; index < rawEvents.length; index += 1) {\n      const event = parseRuntimeTrajectoryHookEvent(rawEvents[index])\n      if (!event) {\n        diagnostics.push(`${key}: runtimeEvents[${index}] is not a RuntimeHookEvent`)\n        continue\n      }\n      events.push(event)\n\n      const scenarioId =\n        event.scenarioId ??\n        stringOrUndefined(options.scenarioIdOf?.(record, recordIndex)) ??\n        stringOrUndefined(record.scenarioId)\n      const prior = runsById.get(event.runId)\n      if (!prior) {\n        runsById.set(event.runId, { runId: event.runId, scenarioId, splitTag })\n        continue\n      }\n      if (prior.scenarioId !== scenarioId || prior.splitTag !== splitTag) {\n        diagnostics.push(`${key}: runId ${event.runId} has conflicting scenario/split metadata`)\n      }\n    }\n  }\n\n  const runs = [...runsById.values()]\n  return {\n    runs,\n    events,\n    summary: {\n      recordCount: options.records.length,\n      recordWithRuntimeEventsCount,\n      runtimeRunCount: runs.length,\n      lifecycleEventCount: events.length,\n      defaultedSplitCount,\n    },\n    diagnostics,\n  }\n}\n\nexport function parseRuntimeTrajectoryHookEvent(input: unknown): RuntimeTrajectoryHookEvent | null {\n  if (!isRecord(input)) return null\n  if (typeof input.id !== 'string' || input.id.length === 0) return null\n  if (typeof input.runId !== 'string' || input.runId.length === 0) return null\n  if (typeof input.target !== 'string' || input.target.length === 0) return null\n  if (typeof input.phase !== 'string' || input.phase.length === 0) return null\n  if (typeof input.timestamp !== 'number' || !Number.isFinite(input.timestamp)) return null\n\n  return {\n    id: input.id,\n    runId: input.runId,\n    scenarioId: stringOrUndefined(input.scenarioId),\n    target: input.target,\n    phase: input.phase,\n    timestamp: input.timestamp,\n    stepIndex: finiteNumberOrUndefined(input.stepIndex),\n    parentId: stringOrUndefined(input.parentId),\n    payload: input.payload,\n    metadata: isRecord(input.metadata) ? { ...input.metadata } : undefined,\n  }\n}\n\nfunction runtimeTrajectoryRecordKey<TRecord extends RuntimeTrajectoryRecord>(\n  record: TRecord,\n  index: number,\n  recordIdOf?: (record: TRecord, index: number) => string | undefined,\n): string {\n  return (\n    stringOrUndefined(recordIdOf?.(record, index)) ??\n    stringOrUndefined(record.id) ??\n    `record[${index}]`\n  )\n}\n\nfunction isRecord(value: unknown): value is Record<string, unknown> {\n  return typeof value === 'object' && value !== null && !Array.isArray(value)\n}\n\nfunction stringOrUndefined(value: unknown): string | undefined {\n  return typeof value === 'string' && value.length > 0 ? value : undefined\n}\n\nfunction finiteNumberOrUndefined(value: unknown): number | undefined {\n  return typeof value === 'number' && Number.isFinite(value) ? value : undefined\n}\n"],"mappings":";AAqDA,IAAM,oBAAiC;AAEhC,SAAS,iCACd,SACqC;AACrC,QAAM,cAAwB,CAAC;AAC/B,QAAM,WAAW,oBAAI,IAAwC;AAC7D,QAAM,SAAuC,CAAC;AAC9C,MAAI,+BAA+B;AACnC,MAAI,sBAAsB;AAE1B,WAAS,cAAc,GAAG,cAAc,QAAQ,QAAQ,QAAQ,eAAe,GAAG;AAChF,UAAM,SAAS,QAAQ,QAAQ,WAAW;AAC1C,UAAM,MAAM,2BAA2B,QAAQ,aAAa,QAAQ,UAAU;AAC9E,UAAM,WAAW,OAAO,YAAY,QAAQ,mBAAmB;AAC/D,QAAI,OAAO,aAAa,OAAW,wBAAuB;AAE1D,UAAM,YAAY,OAAO;AACzB,QAAI,CAAC,MAAM,QAAQ,SAAS,GAAG;AAC7B,kBAAY;AAAA,QACV,GAAG,GAAG;AAAA,MACR;AACA;AAAA,IACF;AACA,QAAI,UAAU,WAAW,GAAG;AAC1B,kBAAY,KAAK,GAAG,GAAG,0DAA0D;AACjF;AAAA,IACF;AACA,oCAAgC;AAEhC,aAAS,QAAQ,GAAG,QAAQ,UAAU,QAAQ,SAAS,GAAG;AACxD,YAAM,QAAQ,gCAAgC,UAAU,KAAK,CAAC;AAC9D,UAAI,CAAC,OAAO;AACV,oBAAY,KAAK,GAAG,GAAG,mBAAmB,KAAK,6BAA6B;AAC5E;AAAA,MACF;AACA,aAAO,KAAK,KAAK;AAEjB,YAAM,aACJ,MAAM,cACN,kBAAkB,QAAQ,eAAe,QAAQ,WAAW,CAAC,KAC7D,kBAAkB,OAAO,UAAU;AACrC,YAAM,QAAQ,SAAS,IAAI,MAAM,KAAK;AACtC,UAAI,CAAC,OAAO;AACV,iBAAS,IAAI,MAAM,OAAO,EAAE,OAAO,MAAM,OAAO,YAAY,SAAS,CAAC;AACtE;AAAA,MACF;AACA,UAAI,MAAM,eAAe,cAAc,MAAM,aAAa,UAAU;AAClE,oBAAY,KAAK,GAAG,GAAG,WAAW,MAAM,KAAK,0CAA0C;AAAA,MACzF;AAAA,IACF;AAAA,EACF;AAEA,QAAM,OAAO,CAAC,GAAG,SAAS,OAAO,CAAC;AAClC,SAAO;AAAA,IACL;AAAA,IACA;AAAA,IACA,SAAS;AAAA,MACP,aAAa,QAAQ,QAAQ;AAAA,MAC7B;AAAA,MACA,iBAAiB,KAAK;AAAA,MACtB,qBAAqB,OAAO;AAAA,MAC5B;AAAA,IACF;AAAA,IACA;AAAA,EACF;AACF;AAEO,SAAS,gCAAgC,OAAmD;AACjG,MAAI,CAAC,SAAS,KAAK,EAAG,QAAO;AAC7B,MAAI,OAAO,MAAM,OAAO,YAAY,MAAM,GAAG,WAAW,EAAG,QAAO;AAClE,MAAI,OAAO,MAAM,UAAU,YAAY,MAAM,MAAM,WAAW,EAAG,QAAO;AACxE,MAAI,OAAO,MAAM,WAAW,YAAY,MAAM,OAAO,WAAW,EAAG,QAAO;AAC1E,MAAI,OAAO,MAAM,UAAU,YAAY,MAAM,MAAM,WAAW,EAAG,QAAO;AACxE,MAAI,OAAO,MAAM,cAAc,YAAY,CAAC,OAAO,SAAS,MAAM,SAAS,EAAG,QAAO;AAErF,SAAO;AAAA,IACL,IAAI,MAAM;AAAA,IACV,OAAO,MAAM;AAAA,IACb,YAAY,kBAAkB,MAAM,UAAU;AAAA,IAC9C,QAAQ,MAAM;AAAA,IACd,OAAO,MAAM;AAAA,IACb,WAAW,MAAM;AAAA,IACjB,WAAW,wBAAwB,MAAM,SAAS;AAAA,IAClD,UAAU,kBAAkB,MAAM,QAAQ;AAAA,IAC1C,SAAS,MAAM;AAAA,IACf,UAAU,SAAS,MAAM,QAAQ,IAAI,EAAE,GAAG,MAAM,SAAS,IAAI;AAAA,EAC/D;AACF;AAEA,SAAS,2BACP,QACA,OACA,YACQ;AACR,SACE,kBAAkB,aAAa,QAAQ,KAAK,CAAC,KAC7C,kBAAkB,OAAO,EAAE,KAC3B,UAAU,KAAK;AAEnB;AAEA,SAAS,SAAS,OAAkD;AAClE,SAAO,OAAO,UAAU,YAAY,UAAU,QAAQ,CAAC,MAAM,QAAQ,KAAK;AAC5E;AAEA,SAAS,kBAAkB,OAAoC;AAC7D,SAAO,OAAO,UAAU,YAAY,MAAM,SAAS,IAAI,QAAQ;AACjE;AAEA,SAAS,wBAAwB,OAAoC;AACnE,SAAO,OAAO,UAAU,YAAY,OAAO,SAAS,KAAK,IAAI,QAAQ;AACvE;","names":[]}

package/dist/{chunk-DFS3FEXO.js → chunk-ZZUXHH3R.js}

@@ -73,7 +73,7 @@ function createHostedClient(tenant) {
     }
   };
 }
-function hostedClientFromEnv(overrides = {}) {
+function hostedTenantFromEnv(overrides = {}) {
   const env = overrides.env ?? process.env;
   const endpoint = (overrides.endpoint ?? env.TANGLE_INGEST_URL ?? env.TANGLE_ORCHESTRATOR_URL)?.trim();
   const apiKey = (overrides.apiKey ?? env.TANGLE_INGEST_API_KEY ?? env.TANGLE_API_KEY)?.trim();
@@ -83,12 +83,17 @@ function hostedClientFromEnv(overrides = {}) {
   if (overrides.fetchImpl) tenant.fetchImpl = overrides.fetchImpl;
   if (overrides.timeoutMs !== void 0) tenant.timeoutMs = overrides.timeoutMs;
   if (overrides.retries !== void 0) tenant.retries = overrides.retries;
-  return createHostedClient(tenant);
+  return tenant;
+}
+function hostedClientFromEnv(overrides = {}) {
+  const tenant = hostedTenantFromEnv(overrides);
+  return tenant ? createHostedClient(tenant) : void 0;
 }
 
 export {
   HOSTED_WIRE_VERSION,
   createHostedClient,
+  hostedTenantFromEnv,
   hostedClientFromEnv
 };
-//# sourceMappingURL=chunk-DFS3FEXO.js.map
+//# sourceMappingURL=chunk-ZZUXHH3R.js.map

package/dist/{chunk-DFS3FEXO.js.map → chunk-ZZUXHH3R.js.map}

@@ -1 +1 @@
-{"version":3,"sources":["../src/hosted/types.ts","../src/hosted/client.ts"],"sourcesContent":["/**\n * # Hosted-tier wire format — the schema that EVERY orchestrator (ours,\n * a partner's self-hosted one, a future open implementation) must accept.\n *\n * **Stability:** every type in this file is committed under semver. New\n * minors only ADD optional fields. Breaking changes mean a major bump\n * (`HostedWireVersion` literal increment).\n *\n * The wire format is two event streams in one transport:\n *\n *   1. **Eval-run events** (`POST /v1/ingest/eval-runs`). Posted when a\n *      campaign / improvement-loop completes (or per-generation if\n *      streaming). Carries the structured result + per-cell scores +\n *      surface diffs the orchestrator stores for the dashboard.\n *\n *   2. **Trace spans** (`POST /v1/ingest/traces`). Standard OTLP-shaped\n *      spans with a few additional attributes so the orchestrator can\n *      pivot from eval-run → underlying execution. Compatible with any\n *      OTel collector.\n *\n * Both endpoints are authenticated with a bearer token + a tenant id\n * header. Tenants isolate everything downstream of ingest; no tenant\n * ever sees another tenant's data.\n */\n\nimport type { GateDecision, MutableSurface } from '../campaign/types'\nimport type { InsightReport } from '../contract/insight-report'\n\n// re-export so wire-format consumers can import the optional payload type\n// from `@tangle-network/agent-eval/hosted` without reaching into /contract.\nexport type { InsightReport } from '../contract/insight-report'\n\nexport const HOSTED_WIRE_VERSION = '2026-05-26.v1' as const\nexport type HostedWireVersion = typeof HOSTED_WIRE_VERSION\n\n// ── Transport headers ───────────────────────────────────────────────\n\n/** Every ingest request carries these. */\nexport interface HostedIngestHeaders {\n  /** Bearer token. The orchestrator validates against the tenant key. */\n  authorization: `Bearer ${string}`\n  /** Stable tenant id (the orchestrator-side primary key for the tenant). */\n  'x-tangle-tenant-id': string\n  /** Wire-version pin so the server can reject incompatible payloads. */\n  'x-tangle-wire-version': HostedWireVersion\n  /** Optional idempotency key for retry-safe ingest. */\n  'idempotency-key'?: string\n}\n\n// ── Eval-run event ──────────────────────────────────────────────────\n\n/** Lifecycle stages of an eval-run as the substrate reports them. */\nexport type EvalRunStatus =\n  | 'started'\n  | 'baseline-complete'\n  | 'generation-complete'\n  | 'gate-decided'\n  | 'finished'\n  | 'errored'\n\nexport interface EvalRunCellScore {\n  /** Stable scenario id from the consumer's scenario set. */\n  scenarioId: string\n  /** Repetition index when reps > 1; 0 for the default. */\n  rep: number\n  /** Composite score across all judges + dimensions for this cell. */\n  compositeMean: number\n  /** Per-judge → per-dimension scores; null where the judge did not run. */\n  dimensions: Record<string, Record<string, number>>\n  /** Per-cell error message if the dispatch threw. Null on success. */\n  errorMessage?: string\n}\n\nexport interface EvalRunGenerationSnapshot {\n  /** Generation index. 0 is baseline. */\n  index: number\n  /** Candidate surface fingerprint (stable hash) — pivot key into the\n   *  trace stream to fetch the underlying execution. */\n  surfaceHash: string\n  /** The candidate surface itself. May be omitted to avoid PII when the\n   *  consumer prefers not to ship verbatim prompts. */\n  surface?: MutableSurface\n  /** Per-cell scores for this generation. */\n  cells: EvalRunCellScore[]\n  /** Aggregate composite mean across all cells in this generation. */\n  compositeMean: number\n  /** Total $ spent across this generation. */\n  costUsd: number\n  /** Wall-clock duration of this generation. */\n  durationMs: number\n}\n\n/**\n * The top-level eval-run event. One ingest call per logical eval-run;\n * generations stream in incrementally via repeated calls with the same\n * `runId`. The orchestrator deduplicates by `(runId, generation.index)`.\n */\nexport interface EvalRunEvent {\n  /** Stable run id (the substrate's `runId`). UUID or substrate-generated. */\n  runId: string\n  /** Where this run was happening — derived from `RunCampaignOptions.runDir`. */\n  runDir: string\n  /** ISO-8601 timestamp the substrate recorded the event. */\n  timestamp: string\n  /** Lifecycle stage this event represents. */\n  status: EvalRunStatus\n  /** Free-form consumer tags (env, branch, model id, etc.). Searchable. */\n  labels: Record<string, string>\n  /** Baseline campaign snapshot. Present when status >= baseline-complete. */\n  baseline?: EvalRunGenerationSnapshot\n  /** Per-generation snapshots. Streams in; orchestrator appends. */\n  generations: EvalRunGenerationSnapshot[]\n  /** Final gate decision. Present when status >= gate-decided. */\n  gateDecision?: GateDecision\n  /** Held-out lift = winner-on-holdout - baseline-on-holdout. */\n  holdoutLift?: number\n  /** Total $ spent across baseline + every generation. */\n  totalCostUsd: number\n  /** Total wall-clock duration. */\n  totalDurationMs: number\n  /** Error message if status === 'errored'. */\n  errorMessage?: string\n  /** Rigor packet emitted alongside the run — distributional summary,\n   *  paired-bootstrap lift CI, judge stats, inter-rater agreement,\n   *  contamination check, failure clusters (when an analyst is wired),\n   *  outcome correlation (when downstream signal is supplied), and the\n   *  recommendations the dashboard surfaces verbatim. Additive; older\n   *  clients that don't know about this field continue to work. */\n  insightReport?: InsightReport\n}\n\n// ── Trace span event ────────────────────────────────────────────────\n\n/**\n * OTel-shape span with a few additional attributes for eval-run pivoting.\n * Compatible with any OTLP collector — `name`, `traceId`, `spanId`,\n * `startTimeUnixNano`, `endTimeUnixNano`, `attributes` are stock OTel.\n */\nexport interface TraceSpanEvent {\n  traceId: string\n  spanId: string\n  parentSpanId?: string\n  name: string\n  startTimeUnixNano: number\n  endTimeUnixNano: number\n  attributes: Record<string, string | number | boolean>\n  events?: Array<{\n    timeUnixNano: number\n    name: string\n    attributes?: Record<string, string | number | boolean>\n  }>\n  status?: { code: 'OK' | 'ERROR' | 'UNSET'; message?: string }\n  /** Pivot back into the eval-run stream. */\n  'tangle.runId'?: string\n  /** Pivot to the specific generation. */\n  'tangle.generation'?: number\n  /** Pivot to the specific cell. */\n  'tangle.cellId'?: string\n  /** Pivot to the specific scenario. */\n  'tangle.scenarioId'?: string\n}\n\n// ── Ingest request bodies ───────────────────────────────────────────\n\nexport interface IngestEvalRunsRequest {\n  wireVersion: HostedWireVersion\n  events: EvalRunEvent[]\n}\n\nexport interface IngestTracesRequest {\n  wireVersion: HostedWireVersion\n  spans: TraceSpanEvent[]\n}\n\nexport interface IngestResponse {\n  /** Accepted events / spans count. */\n  accepted: number\n  /** Rejected events with reasons (validation failures, dup idempotency key, etc.). */\n  rejected: Array<{ index: number; reason: string }>\n}\n","/**\n * # Hosted-tier ingest client.\n *\n * Ships eval-run events + trace spans to any orchestrator (ours, a\n * partner's self-hosted one, or a future open implementation) that\n * speaks the wire format in `./types.ts`.\n *\n * Three modes:\n *   - **Ours:** point at `https://orchestrator.tangle.tools` (the host root —\n *     the client appends the versioned `/v1/ingest/...` path itself; a trailing\n *     `/v1` on the endpoint is tolerated and normalized away). We handle ingest\n *     + storage + dashboard.\n *   - **Self-hosted:** point at whatever URL runs the reference receiver\n *     from `examples/hosted-ingest-server/`.\n *   - **Off (default):** when `hostedTenant` is unset, nothing is sent.\n *     Everything stays local.\n */\n\nimport {\n  type EvalRunEvent,\n  HOSTED_WIRE_VERSION,\n  type HostedWireVersion,\n  type IngestEvalRunsRequest,\n  type IngestResponse,\n  type IngestTracesRequest,\n  type TraceSpanEvent,\n} from './types'\n\nexport interface HostedTenant {\n  /** Orchestrator endpoint base URL (no trailing slash). Required. */\n  endpoint: string\n  /** Bearer token issued by the orchestrator. Required. */\n  apiKey: string\n  /** Tenant id — the orchestrator's primary key for this consumer. Required. */\n  tenantId: string\n  /** Optional `fetch` override (auth wrappers, custom agent, test mocks). */\n  fetchImpl?: typeof fetch\n  /** Per-call timeout in ms. Default 30s. */\n  timeoutMs?: number\n  /** Retries on 5xx / network errors. Default 2. */\n  retries?: number\n}\n\nexport interface HostedClient {\n  ingestEvalRun(event: EvalRunEvent, idempotencyKey?: string): Promise<IngestResponse>\n  ingestEvalRuns(events: EvalRunEvent[], idempotencyKey?: string): Promise<IngestResponse>\n  ingestTraces(spans: TraceSpanEvent[], idempotencyKey?: string): Promise<IngestResponse>\n  readonly tenant: HostedTenant\n  readonly wireVersion: HostedWireVersion\n}\n\ninterface RequestOptions {\n  idempotencyKey?: string\n  signal?: AbortSignal\n}\n\nfunction sleep(ms: number): Promise<void> {\n  return new Promise((resolve) => {\n    const t = setTimeout(resolve, ms)\n    if (typeof (t as { unref?: () => void }).unref === 'function')\n      (t as { unref: () => void }).unref()\n  })\n}\n\nasync function post<TReq, TRes>(\n  tenant: HostedTenant,\n  path: string,\n  body: TReq,\n  opts: RequestOptions = {},\n): Promise<TRes> {\n  const timeoutMs = tenant.timeoutMs ?? 30_000\n  const maxRetries = tenant.retries ?? 2\n  const f: typeof fetch = tenant.fetchImpl ?? ((...args) => fetch(...args))\n  // `path` already carries the `/v1` version prefix (e.g. `/v1/ingest/eval-runs`).\n  // Strip a trailing slash AND a trailing `/v1` from the endpoint so a base of\n  // either `https://host` or `https://host/v1` resolves to the same correct URL\n  // — callers routinely pass the versioned base and would otherwise hit\n  // `/v1/v1/ingest/...` (404).\n  const base = tenant.endpoint.replace(/\\/+$/, '').replace(/\\/v1$/, '')\n  const url = `${base}${path}`\n\n  let lastError: unknown\n  for (let attempt = 0; attempt <= maxRetries; attempt++) {\n    const ourTimeout = AbortSignal.timeout(timeoutMs)\n    const combinedSignal = opts.signal ? AbortSignal.any([opts.signal, ourTimeout]) : ourTimeout\n    try {\n      const headers: Record<string, string> = {\n        'content-type': 'application/json',\n        authorization: `Bearer ${tenant.apiKey}`,\n        'x-tangle-tenant-id': tenant.tenantId,\n        'x-tangle-wire-version': HOSTED_WIRE_VERSION,\n      }\n      if (opts.idempotencyKey) headers['idempotency-key'] = opts.idempotencyKey\n\n      const res = await f(url, {\n        method: 'POST',\n        headers,\n        body: JSON.stringify(body),\n        signal: combinedSignal,\n      })\n      if (!res.ok) {\n        const retryable = res.status >= 500 || res.status === 408 || res.status === 429\n        if (!retryable || attempt === maxRetries) {\n          const text = await res.text().catch(() => '')\n          throw new Error(`hosted ingest ${url} failed (${res.status}): ${text.slice(0, 500)}`)\n        }\n        await sleep(2 ** attempt * 200 + Math.random() * 200)\n        continue\n      }\n      return (await res.json()) as TRes\n    } catch (err) {\n      if (opts.signal?.aborted) throw err\n      lastError = err\n      if (attempt === maxRetries) throw err\n      await sleep(2 ** attempt * 200 + Math.random() * 200)\n    }\n  }\n  throw lastError ?? new Error('hosted ingest exhausted retries')\n}\n\nexport function createHostedClient(tenant: HostedTenant): HostedClient {\n  return {\n    tenant,\n    wireVersion: HOSTED_WIRE_VERSION,\n\n    async ingestEvalRun(event, idempotencyKey) {\n      return this.ingestEvalRuns([event], idempotencyKey)\n    },\n\n    async ingestEvalRuns(events, idempotencyKey) {\n      const body: IngestEvalRunsRequest = { wireVersion: HOSTED_WIRE_VERSION, events }\n      return post<IngestEvalRunsRequest, IngestResponse>(tenant, '/v1/ingest/eval-runs', body, {\n        idempotencyKey,\n      })\n    },\n\n    async ingestTraces(spans, idempotencyKey) {\n      const body: IngestTracesRequest = { wireVersion: HOSTED_WIRE_VERSION, spans }\n      return post<IngestTracesRequest, IngestResponse>(tenant, '/v1/ingest/traces', body, {\n        idempotencyKey,\n      })\n    },\n  }\n}\n\n/**\n * Build a `HostedClient` from environment, or `undefined` when ingest is not\n * configured — the canonical, fail-soft wiring every product uses so eval-run +\n * trace provenance lands in the Intelligence dashboard with ONE call:\n *\n *   const hosted = hostedClientFromEnv()\n *   // ...run the loop...\n *   await emitLoopProvenance({ ..., hostedClient: hosted })  // no-op if undefined\n *\n * Returns `undefined` (NOT an error) when any of endpoint / apiKey / tenantId is\n * missing — so a product wires the ship call unconditionally and it stays a\n * no-op until the env is set. Env precedence:\n *   - endpoint:  `TANGLE_INGEST_URL` → `TANGLE_ORCHESTRATOR_URL`\n *   - apiKey:    `TANGLE_INGEST_API_KEY` → `TANGLE_API_KEY`\n *   - tenantId:  `TANGLE_TENANT_ID`\n * A trailing slash on the endpoint is stripped. Pass `overrides` to supply any\n * field directly (e.g. a fixed `tenantId` per product) — overrides win over env.\n */\nexport function hostedClientFromEnv(\n  overrides: Partial<HostedTenant> & { env?: Record<string, string | undefined> } = {},\n): HostedClient | undefined {\n  const env = overrides.env ?? process.env\n  const endpoint = (\n    overrides.endpoint ??\n    env.TANGLE_INGEST_URL ??\n    env.TANGLE_ORCHESTRATOR_URL\n  )?.trim()\n  const apiKey = (overrides.apiKey ?? env.TANGLE_INGEST_API_KEY ?? env.TANGLE_API_KEY)?.trim()\n  const tenantId = (overrides.tenantId ?? env.TANGLE_TENANT_ID)?.trim()\n  if (!endpoint || !apiKey || !tenantId) return undefined\n  const tenant: HostedTenant = { endpoint: endpoint.replace(/\\/+$/, ''), apiKey, tenantId }\n  if (overrides.fetchImpl) tenant.fetchImpl = overrides.fetchImpl\n  if (overrides.timeoutMs !== undefined) tenant.timeoutMs = overrides.timeoutMs\n  if (overrides.retries !== undefined) tenant.retries = overrides.retries\n  return createHostedClient(tenant)\n}\n"],"mappings":";AAgCO,IAAM,sBAAsB;;;ACwBnC,SAAS,MAAM,IAA2B;AACxC,SAAO,IAAI,QAAQ,CAAC,YAAY;AAC9B,UAAM,IAAI,WAAW,SAAS,EAAE;AAChC,QAAI,OAAQ,EAA6B,UAAU;AACjD,MAAC,EAA4B,MAAM;AAAA,EACvC,CAAC;AACH;AAEA,eAAe,KACb,QACA,MACA,MACA,OAAuB,CAAC,GACT;AACf,QAAM,YAAY,OAAO,aAAa;AACtC,QAAM,aAAa,OAAO,WAAW;AACrC,QAAM,IAAkB,OAAO,cAAc,IAAI,SAAS,MAAM,GAAG,IAAI;AAMvE,QAAM,OAAO,OAAO,SAAS,QAAQ,QAAQ,EAAE,EAAE,QAAQ,SAAS,EAAE;AACpE,QAAM,MAAM,GAAG,IAAI,GAAG,IAAI;AAE1B,MAAI;AACJ,WAAS,UAAU,GAAG,WAAW,YAAY,WAAW;AACtD,UAAM,aAAa,YAAY,QAAQ,SAAS;AAChD,UAAM,iBAAiB,KAAK,SAAS,YAAY,IAAI,CAAC,KAAK,QAAQ,UAAU,CAAC,IAAI;AAClF,QAAI;AACF,YAAM,UAAkC;AAAA,QACtC,gBAAgB;AAAA,QAChB,eAAe,UAAU,OAAO,MAAM;AAAA,QACtC,sBAAsB,OAAO;AAAA,QAC7B,yBAAyB;AAAA,MAC3B;AACA,UAAI,KAAK,eAAgB,SAAQ,iBAAiB,IAAI,KAAK;AAE3D,YAAM,MAAM,MAAM,EAAE,KAAK;AAAA,QACvB,QAAQ;AAAA,QACR;AAAA,QACA,MAAM,KAAK,UAAU,IAAI;AAAA,QACzB,QAAQ;AAAA,MACV,CAAC;AACD,UAAI,CAAC,IAAI,IAAI;AACX,cAAM,YAAY,IAAI,UAAU,OAAO,IAAI,WAAW,OAAO,IAAI,WAAW;AAC5E,YAAI,CAAC,aAAa,YAAY,YAAY;AACxC,gBAAM,OAAO,MAAM,IAAI,KAAK,EAAE,MAAM,MAAM,EAAE;AAC5C,gBAAM,IAAI,MAAM,iBAAiB,GAAG,YAAY,IAAI,MAAM,MAAM,KAAK,MAAM,GAAG,GAAG,CAAC,EAAE;AAAA,QACtF;AACA,cAAM,MAAM,KAAK,UAAU,MAAM,KAAK,OAAO,IAAI,GAAG;AACpD;AAAA,MACF;AACA,aAAQ,MAAM,IAAI,KAAK;AAAA,IACzB,SAAS,KAAK;AACZ,UAAI,KAAK,QAAQ,QAAS,OAAM;AAChC,kBAAY;AACZ,UAAI,YAAY,WAAY,OAAM;AAClC,YAAM,MAAM,KAAK,UAAU,MAAM,KAAK,OAAO,IAAI,GAAG;AAAA,IACtD;AAAA,EACF;AACA,QAAM,aAAa,IAAI,MAAM,iCAAiC;AAChE;AAEO,SAAS,mBAAmB,QAAoC;AACrE,SAAO;AAAA,IACL;AAAA,IACA,aAAa;AAAA,IAEb,MAAM,cAAc,OAAO,gBAAgB;AACzC,aAAO,KAAK,eAAe,CAAC,KAAK,GAAG,cAAc;AAAA,IACpD;AAAA,IAEA,MAAM,eAAe,QAAQ,gBAAgB;AAC3C,YAAM,OAA8B,EAAE,aAAa,qBAAqB,OAAO;AAC/E,aAAO,KAA4C,QAAQ,wBAAwB,MAAM;AAAA,QACvF;AAAA,MACF,CAAC;AAAA,IACH;AAAA,IAEA,MAAM,aAAa,OAAO,gBAAgB;AACxC,YAAM,OAA4B,EAAE,aAAa,qBAAqB,MAAM;AAC5E,aAAO,KAA0C,QAAQ,qBAAqB,MAAM;AAAA,QAClF;AAAA,MACF,CAAC;AAAA,IACH;AAAA,EACF;AACF;AAoBO,SAAS,oBACd,YAAkF,CAAC,GACzD;AAC1B,QAAM,MAAM,UAAU,OAAO,QAAQ;AACrC,QAAM,YACJ,UAAU,YACV,IAAI,qBACJ,IAAI,0BACH,KAAK;AACR,QAAM,UAAU,UAAU,UAAU,IAAI,yBAAyB,IAAI,iBAAiB,KAAK;AAC3F,QAAM,YAAY,UAAU,YAAY,IAAI,mBAAmB,KAAK;AACpE,MAAI,CAAC,YAAY,CAAC,UAAU,CAAC,SAAU,QAAO;AAC9C,QAAM,SAAuB,EAAE,UAAU,SAAS,QAAQ,QAAQ,EAAE,GAAG,QAAQ,SAAS;AACxF,MAAI,UAAU,UAAW,QAAO,YAAY,UAAU;AACtD,MAAI,UAAU,cAAc,OAAW,QAAO,YAAY,UAAU;AACpE,MAAI,UAAU,YAAY,OAAW,QAAO,UAAU,UAAU;AAChE,SAAO,mBAAmB,MAAM;AAClC;","names":[]}
+{"version":3,"sources":["../src/hosted/types.ts","../src/hosted/client.ts"],"sourcesContent":["/**\n * # Hosted-tier wire format — the schema that EVERY orchestrator (ours,\n * a partner's self-hosted one, a future open implementation) must accept.\n *\n * **Stability:** every type in this file is committed under semver. New\n * minors only ADD optional fields. Breaking changes mean a major bump\n * (`HostedWireVersion` literal increment).\n *\n * The wire format is two event streams in one transport:\n *\n *   1. **Eval-run events** (`POST /v1/ingest/eval-runs`). Posted when a\n *      campaign / improvement-loop completes (or per-generation if\n *      streaming). Carries the structured result + per-cell scores +\n *      surface diffs the orchestrator stores for the dashboard.\n *\n *   2. **Trace spans** (`POST /v1/ingest/traces`). Standard OTLP-shaped\n *      spans with a few additional attributes so the orchestrator can\n *      pivot from eval-run → underlying execution. Compatible with any\n *      OTel collector.\n *\n * Both endpoints are authenticated with a bearer token + a tenant id\n * header. Tenants isolate everything downstream of ingest; no tenant\n * ever sees another tenant's data.\n */\n\nimport type { GateDecision, MutableSurface } from '../campaign/types'\nimport type { InsightReport } from '../contract/insight-report'\n\n// re-export so wire-format consumers can import the optional payload type\n// from `@tangle-network/agent-eval/hosted` without reaching into /contract.\nexport type { InsightReport } from '../contract/insight-report'\n\nexport const HOSTED_WIRE_VERSION = '2026-05-26.v1' as const\nexport type HostedWireVersion = typeof HOSTED_WIRE_VERSION\n\n// ── Transport headers ───────────────────────────────────────────────\n\n/** Every ingest request carries these. */\nexport interface HostedIngestHeaders {\n  /** Bearer token. The orchestrator validates against the tenant key. */\n  authorization: `Bearer ${string}`\n  /** Stable tenant id (the orchestrator-side primary key for the tenant). */\n  'x-tangle-tenant-id': string\n  /** Wire-version pin so the server can reject incompatible payloads. */\n  'x-tangle-wire-version': HostedWireVersion\n  /** Optional idempotency key for retry-safe ingest. */\n  'idempotency-key'?: string\n}\n\n// ── Eval-run event ──────────────────────────────────────────────────\n\n/** Lifecycle stages of an eval-run as the substrate reports them. */\nexport type EvalRunStatus =\n  | 'started'\n  | 'baseline-complete'\n  | 'generation-complete'\n  | 'gate-decided'\n  | 'finished'\n  | 'errored'\n\nexport interface EvalRunCellScore {\n  /** Stable scenario id from the consumer's scenario set. */\n  scenarioId: string\n  /** Repetition index when reps > 1; 0 for the default. */\n  rep: number\n  /** Composite score across all judges + dimensions for this cell. */\n  compositeMean: number\n  /** Per-judge → per-dimension scores; null where the judge did not run. */\n  dimensions: Record<string, Record<string, number>>\n  /** Per-cell error message if the dispatch threw. Null on success. */\n  errorMessage?: string\n}\n\nexport interface EvalRunGenerationSnapshot {\n  /** Generation index. 0 is baseline. */\n  index: number\n  /** Candidate surface fingerprint (stable hash) — pivot key into the\n   *  trace stream to fetch the underlying execution. */\n  surfaceHash: string\n  /** The candidate surface itself. May be omitted to avoid PII when the\n   *  consumer prefers not to ship verbatim prompts. */\n  surface?: MutableSurface\n  /** Per-cell scores for this generation. */\n  cells: EvalRunCellScore[]\n  /** Aggregate composite mean across all cells in this generation. */\n  compositeMean: number\n  /** Total $ spent across this generation. */\n  costUsd: number\n  /** Wall-clock duration of this generation. */\n  durationMs: number\n}\n\n/**\n * The top-level eval-run event. One ingest call per logical eval-run;\n * generations stream in incrementally via repeated calls with the same\n * `runId`. The orchestrator deduplicates by `(runId, generation.index)`.\n */\nexport interface EvalRunEvent {\n  /** Stable run id (the substrate's `runId`). UUID or substrate-generated. */\n  runId: string\n  /** Where this run was happening — derived from `RunCampaignOptions.runDir`. */\n  runDir: string\n  /** ISO-8601 timestamp the substrate recorded the event. */\n  timestamp: string\n  /** Lifecycle stage this event represents. */\n  status: EvalRunStatus\n  /** Free-form consumer tags (env, branch, model id, etc.). Searchable. */\n  labels: Record<string, string>\n  /** Baseline campaign snapshot. Present when status >= baseline-complete. */\n  baseline?: EvalRunGenerationSnapshot\n  /** Per-generation snapshots. Streams in; orchestrator appends. */\n  generations: EvalRunGenerationSnapshot[]\n  /** Final gate decision. Present when status >= gate-decided. */\n  gateDecision?: GateDecision\n  /** Held-out lift = winner-on-holdout - baseline-on-holdout. */\n  holdoutLift?: number\n  /** Total $ spent across baseline + every generation. */\n  totalCostUsd: number\n  /** Total wall-clock duration. */\n  totalDurationMs: number\n  /** Error message if status === 'errored'. */\n  errorMessage?: string\n  /** Rigor packet emitted alongside the run — distributional summary,\n   *  paired-bootstrap lift CI, judge stats, inter-rater agreement,\n   *  contamination check, failure clusters (when an analyst is wired),\n   *  outcome correlation (when downstream signal is supplied), and the\n   *  recommendations the dashboard surfaces verbatim. Additive; older\n   *  clients that don't know about this field continue to work. */\n  insightReport?: InsightReport\n}\n\n// ── Trace span event ────────────────────────────────────────────────\n\n/**\n * OTel-shape span with a few additional attributes for eval-run pivoting.\n * Compatible with any OTLP collector — `name`, `traceId`, `spanId`,\n * `startTimeUnixNano`, `endTimeUnixNano`, `attributes` are stock OTel.\n */\nexport interface TraceSpanEvent {\n  traceId: string\n  spanId: string\n  parentSpanId?: string\n  name: string\n  startTimeUnixNano: number\n  endTimeUnixNano: number\n  attributes: Record<string, string | number | boolean>\n  events?: Array<{\n    timeUnixNano: number\n    name: string\n    attributes?: Record<string, string | number | boolean>\n  }>\n  status?: { code: 'OK' | 'ERROR' | 'UNSET'; message?: string }\n  /** Pivot back into the eval-run stream. */\n  'tangle.runId'?: string\n  /** Pivot to the specific generation. */\n  'tangle.generation'?: number\n  /** Pivot to the specific cell. */\n  'tangle.cellId'?: string\n  /** Pivot to the specific scenario. */\n  'tangle.scenarioId'?: string\n}\n\n// ── Ingest request bodies ───────────────────────────────────────────\n\nexport interface IngestEvalRunsRequest {\n  wireVersion: HostedWireVersion\n  events: EvalRunEvent[]\n}\n\nexport interface IngestTracesRequest {\n  wireVersion: HostedWireVersion\n  spans: TraceSpanEvent[]\n}\n\nexport interface IngestResponse {\n  /** Accepted events / spans count. */\n  accepted: number\n  /** Rejected events with reasons (validation failures, dup idempotency key, etc.). */\n  rejected: Array<{ index: number; reason: string }>\n}\n","/**\n * # Hosted-tier ingest client.\n *\n * Ships eval-run events + trace spans to any orchestrator (ours, a\n * partner's self-hosted one, or a future open implementation) that\n * speaks the wire format in `./types.ts`.\n *\n * Three modes:\n *   - **Ours:** point at `https://orchestrator.tangle.tools` (the host root —\n *     the client appends the versioned `/v1/ingest/...` path itself; a trailing\n *     `/v1` on the endpoint is tolerated and normalized away). We handle ingest\n *     + storage + dashboard.\n *   - **Self-hosted:** point at whatever URL runs the reference receiver\n *     from `examples/hosted-ingest-server/`.\n *   - **Off (default):** when `hostedTenant` is unset, nothing is sent.\n *     Everything stays local.\n */\n\nimport {\n  type EvalRunEvent,\n  HOSTED_WIRE_VERSION,\n  type HostedWireVersion,\n  type IngestEvalRunsRequest,\n  type IngestResponse,\n  type IngestTracesRequest,\n  type TraceSpanEvent,\n} from './types'\n\nexport interface HostedTenant {\n  /** Orchestrator endpoint base URL (no trailing slash). Required. */\n  endpoint: string\n  /** Bearer token issued by the orchestrator. Required. */\n  apiKey: string\n  /** Tenant id — the orchestrator's primary key for this consumer. Required. */\n  tenantId: string\n  /** Optional `fetch` override (auth wrappers, custom agent, test mocks). */\n  fetchImpl?: typeof fetch\n  /** Per-call timeout in ms. Default 30s. */\n  timeoutMs?: number\n  /** Retries on 5xx / network errors. Default 2. */\n  retries?: number\n}\n\nexport interface HostedClient {\n  ingestEvalRun(event: EvalRunEvent, idempotencyKey?: string): Promise<IngestResponse>\n  ingestEvalRuns(events: EvalRunEvent[], idempotencyKey?: string): Promise<IngestResponse>\n  ingestTraces(spans: TraceSpanEvent[], idempotencyKey?: string): Promise<IngestResponse>\n  readonly tenant: HostedTenant\n  readonly wireVersion: HostedWireVersion\n}\n\ninterface RequestOptions {\n  idempotencyKey?: string\n  signal?: AbortSignal\n}\n\nfunction sleep(ms: number): Promise<void> {\n  return new Promise((resolve) => {\n    const t = setTimeout(resolve, ms)\n    if (typeof (t as { unref?: () => void }).unref === 'function')\n      (t as { unref: () => void }).unref()\n  })\n}\n\nasync function post<TReq, TRes>(\n  tenant: HostedTenant,\n  path: string,\n  body: TReq,\n  opts: RequestOptions = {},\n): Promise<TRes> {\n  const timeoutMs = tenant.timeoutMs ?? 30_000\n  const maxRetries = tenant.retries ?? 2\n  const f: typeof fetch = tenant.fetchImpl ?? ((...args) => fetch(...args))\n  // `path` already carries the `/v1` version prefix (e.g. `/v1/ingest/eval-runs`).\n  // Strip a trailing slash AND a trailing `/v1` from the endpoint so a base of\n  // either `https://host` or `https://host/v1` resolves to the same correct URL\n  // — callers routinely pass the versioned base and would otherwise hit\n  // `/v1/v1/ingest/...` (404).\n  const base = tenant.endpoint.replace(/\\/+$/, '').replace(/\\/v1$/, '')\n  const url = `${base}${path}`\n\n  let lastError: unknown\n  for (let attempt = 0; attempt <= maxRetries; attempt++) {\n    const ourTimeout = AbortSignal.timeout(timeoutMs)\n    const combinedSignal = opts.signal ? AbortSignal.any([opts.signal, ourTimeout]) : ourTimeout\n    try {\n      const headers: Record<string, string> = {\n        'content-type': 'application/json',\n        authorization: `Bearer ${tenant.apiKey}`,\n        'x-tangle-tenant-id': tenant.tenantId,\n        'x-tangle-wire-version': HOSTED_WIRE_VERSION,\n      }\n      if (opts.idempotencyKey) headers['idempotency-key'] = opts.idempotencyKey\n\n      const res = await f(url, {\n        method: 'POST',\n        headers,\n        body: JSON.stringify(body),\n        signal: combinedSignal,\n      })\n      if (!res.ok) {\n        const retryable = res.status >= 500 || res.status === 408 || res.status === 429\n        if (!retryable || attempt === maxRetries) {\n          const text = await res.text().catch(() => '')\n          throw new Error(`hosted ingest ${url} failed (${res.status}): ${text.slice(0, 500)}`)\n        }\n        await sleep(2 ** attempt * 200 + Math.random() * 200)\n        continue\n      }\n      return (await res.json()) as TRes\n    } catch (err) {\n      if (opts.signal?.aborted) throw err\n      lastError = err\n      if (attempt === maxRetries) throw err\n      await sleep(2 ** attempt * 200 + Math.random() * 200)\n    }\n  }\n  throw lastError ?? new Error('hosted ingest exhausted retries')\n}\n\nexport function createHostedClient(tenant: HostedTenant): HostedClient {\n  return {\n    tenant,\n    wireVersion: HOSTED_WIRE_VERSION,\n\n    async ingestEvalRun(event, idempotencyKey) {\n      return this.ingestEvalRuns([event], idempotencyKey)\n    },\n\n    async ingestEvalRuns(events, idempotencyKey) {\n      const body: IngestEvalRunsRequest = { wireVersion: HOSTED_WIRE_VERSION, events }\n      return post<IngestEvalRunsRequest, IngestResponse>(tenant, '/v1/ingest/eval-runs', body, {\n        idempotencyKey,\n      })\n    },\n\n    async ingestTraces(spans, idempotencyKey) {\n      const body: IngestTracesRequest = { wireVersion: HOSTED_WIRE_VERSION, spans }\n      return post<IngestTracesRequest, IngestResponse>(tenant, '/v1/ingest/traces', body, {\n        idempotencyKey,\n      })\n    },\n  }\n}\n\n/**\n * Build a `HostedClient` from environment, or `undefined` when ingest is not\n * configured — the canonical, fail-soft wiring every product uses so eval-run +\n * trace provenance lands in the Intelligence dashboard with ONE call:\n *\n *   const hosted = hostedClientFromEnv()\n *   // ...run the loop...\n *   await emitLoopProvenance({ ..., hostedClient: hosted })  // no-op if undefined\n *\n * Returns `undefined` (NOT an error) when any of endpoint / apiKey / tenantId is\n * missing — so a product wires the ship call unconditionally and it stays a\n * no-op until the env is set. Env precedence:\n *   - endpoint:  `TANGLE_INGEST_URL` → `TANGLE_ORCHESTRATOR_URL`\n *   - apiKey:    `TANGLE_INGEST_API_KEY` → `TANGLE_API_KEY`\n *   - tenantId:  `TANGLE_TENANT_ID`\n * A trailing slash on the endpoint is stripped. Pass `overrides` to supply any\n * field directly (e.g. a fixed `tenantId` per product) — overrides win over env.\n */\n/**\n * Build a {@link HostedTenant} config from env — the input `selfImprove`'s\n * `hostedTenant` and `emitLoopProvenance` take. Same env precedence + overrides\n * as {@link hostedClientFromEnv}; returns `undefined` (not an error) when any of\n * endpoint / apiKey / tenantId is missing, so a product wires\n * `hostedTenant: hostedTenantFromEnv({ tenantId: 'my-agent' })` unconditionally\n * and it stays off until the env is set.\n */\nexport function hostedTenantFromEnv(\n  overrides: Partial<HostedTenant> & { env?: Record<string, string | undefined> } = {},\n): HostedTenant | undefined {\n  const env = overrides.env ?? process.env\n  const endpoint = (\n    overrides.endpoint ??\n    env.TANGLE_INGEST_URL ??\n    env.TANGLE_ORCHESTRATOR_URL\n  )?.trim()\n  const apiKey = (overrides.apiKey ?? env.TANGLE_INGEST_API_KEY ?? env.TANGLE_API_KEY)?.trim()\n  const tenantId = (overrides.tenantId ?? env.TANGLE_TENANT_ID)?.trim()\n  if (!endpoint || !apiKey || !tenantId) return undefined\n  const tenant: HostedTenant = { endpoint: endpoint.replace(/\\/+$/, ''), apiKey, tenantId }\n  if (overrides.fetchImpl) tenant.fetchImpl = overrides.fetchImpl\n  if (overrides.timeoutMs !== undefined) tenant.timeoutMs = overrides.timeoutMs\n  if (overrides.retries !== undefined) tenant.retries = overrides.retries\n  return tenant\n}\n\nexport function hostedClientFromEnv(\n  overrides: Partial<HostedTenant> & { env?: Record<string, string | undefined> } = {},\n): HostedClient | undefined {\n  const tenant = hostedTenantFromEnv(overrides)\n  return tenant ? createHostedClient(tenant) : undefined\n}\n"],"mappings":";AAgCO,IAAM,sBAAsB;;;ACwBnC,SAAS,MAAM,IAA2B;AACxC,SAAO,IAAI,QAAQ,CAAC,YAAY;AAC9B,UAAM,IAAI,WAAW,SAAS,EAAE;AAChC,QAAI,OAAQ,EAA6B,UAAU;AACjD,MAAC,EAA4B,MAAM;AAAA,EACvC,CAAC;AACH;AAEA,eAAe,KACb,QACA,MACA,MACA,OAAuB,CAAC,GACT;AACf,QAAM,YAAY,OAAO,aAAa;AACtC,QAAM,aAAa,OAAO,WAAW;AACrC,QAAM,IAAkB,OAAO,cAAc,IAAI,SAAS,MAAM,GAAG,IAAI;AAMvE,QAAM,OAAO,OAAO,SAAS,QAAQ,QAAQ,EAAE,EAAE,QAAQ,SAAS,EAAE;AACpE,QAAM,MAAM,GAAG,IAAI,GAAG,IAAI;AAE1B,MAAI;AACJ,WAAS,UAAU,GAAG,WAAW,YAAY,WAAW;AACtD,UAAM,aAAa,YAAY,QAAQ,SAAS;AAChD,UAAM,iBAAiB,KAAK,SAAS,YAAY,IAAI,CAAC,KAAK,QAAQ,UAAU,CAAC,IAAI;AAClF,QAAI;AACF,YAAM,UAAkC;AAAA,QACtC,gBAAgB;AAAA,QAChB,eAAe,UAAU,OAAO,MAAM;AAAA,QACtC,sBAAsB,OAAO;AAAA,QAC7B,yBAAyB;AAAA,MAC3B;AACA,UAAI,KAAK,eAAgB,SAAQ,iBAAiB,IAAI,KAAK;AAE3D,YAAM,MAAM,MAAM,EAAE,KAAK;AAAA,QACvB,QAAQ;AAAA,QACR;AAAA,QACA,MAAM,KAAK,UAAU,IAAI;AAAA,QACzB,QAAQ;AAAA,MACV,CAAC;AACD,UAAI,CAAC,IAAI,IAAI;AACX,cAAM,YAAY,IAAI,UAAU,OAAO,IAAI,WAAW,OAAO,IAAI,WAAW;AAC5E,YAAI,CAAC,aAAa,YAAY,YAAY;AACxC,gBAAM,OAAO,MAAM,IAAI,KAAK,EAAE,MAAM,MAAM,EAAE;AAC5C,gBAAM,IAAI,MAAM,iBAAiB,GAAG,YAAY,IAAI,MAAM,MAAM,KAAK,MAAM,GAAG,GAAG,CAAC,EAAE;AAAA,QACtF;AACA,cAAM,MAAM,KAAK,UAAU,MAAM,KAAK,OAAO,IAAI,GAAG;AACpD;AAAA,MACF;AACA,aAAQ,MAAM,IAAI,KAAK;AAAA,IACzB,SAAS,KAAK;AACZ,UAAI,KAAK,QAAQ,QAAS,OAAM;AAChC,kBAAY;AACZ,UAAI,YAAY,WAAY,OAAM;AAClC,YAAM,MAAM,KAAK,UAAU,MAAM,KAAK,OAAO,IAAI,GAAG;AAAA,IACtD;AAAA,EACF;AACA,QAAM,aAAa,IAAI,MAAM,iCAAiC;AAChE;AAEO,SAAS,mBAAmB,QAAoC;AACrE,SAAO;AAAA,IACL;AAAA,IACA,aAAa;AAAA,IAEb,MAAM,cAAc,OAAO,gBAAgB;AACzC,aAAO,KAAK,eAAe,CAAC,KAAK,GAAG,cAAc;AAAA,IACpD;AAAA,IAEA,MAAM,eAAe,QAAQ,gBAAgB;AAC3C,YAAM,OAA8B,EAAE,aAAa,qBAAqB,OAAO;AAC/E,aAAO,KAA4C,QAAQ,wBAAwB,MAAM;AAAA,QACvF;AAAA,MACF,CAAC;AAAA,IACH;AAAA,IAEA,MAAM,aAAa,OAAO,gBAAgB;AACxC,YAAM,OAA4B,EAAE,aAAa,qBAAqB,MAAM;AAC5E,aAAO,KAA0C,QAAQ,qBAAqB,MAAM;AAAA,QAClF;AAAA,MACF,CAAC;AAAA,IACH;AAAA,EACF;AACF;AA4BO,SAAS,oBACd,YAAkF,CAAC,GACzD;AAC1B,QAAM,MAAM,UAAU,OAAO,QAAQ;AACrC,QAAM,YACJ,UAAU,YACV,IAAI,qBACJ,IAAI,0BACH,KAAK;AACR,QAAM,UAAU,UAAU,UAAU,IAAI,yBAAyB,IAAI,iBAAiB,KAAK;AAC3F,QAAM,YAAY,UAAU,YAAY,IAAI,mBAAmB,KAAK;AACpE,MAAI,CAAC,YAAY,CAAC,UAAU,CAAC,SAAU,QAAO;AAC9C,QAAM,SAAuB,EAAE,UAAU,SAAS,QAAQ,QAAQ,EAAE,GAAG,QAAQ,SAAS;AACxF,MAAI,UAAU,UAAW,QAAO,YAAY,UAAU;AACtD,MAAI,UAAU,cAAc,OAAW,QAAO,YAAY,UAAU;AACpE,MAAI,UAAU,YAAY,OAAW,QAAO,UAAU,UAAU;AAChE,SAAO;AACT;AAEO,SAAS,oBACd,YAAkF,CAAC,GACzD;AAC1B,QAAM,SAAS,oBAAoB,SAAS;AAC5C,SAAO,SAAS,mBAAmB,MAAM,IAAI;AAC/C;","names":[]}

package/dist/contract/index.js

@@ -1,6 +1,6 @@
 import {
   createHostedClient
-} from "../chunk-DFS3FEXO.js";
+} from "../chunk-ZZUXHH3R.js";
 import {
   buildEvidenceVector,
   composeGate,

package/dist/hosted/index.d.ts

@@ -223,8 +223,19 @@ declare function createHostedClient(tenant: HostedTenant): HostedClient;
  * A trailing slash on the endpoint is stripped. Pass `overrides` to supply any
  * field directly (e.g. a fixed `tenantId` per product) — overrides win over env.
  */
+/**
+ * Build a {@link HostedTenant} config from env — the input `selfImprove`'s
+ * `hostedTenant` and `emitLoopProvenance` take. Same env precedence + overrides
+ * as {@link hostedClientFromEnv}; returns `undefined` (not an error) when any of
+ * endpoint / apiKey / tenantId is missing, so a product wires
+ * `hostedTenant: hostedTenantFromEnv({ tenantId: 'my-agent' })` unconditionally
+ * and it stays off until the env is set.
+ */
+declare function hostedTenantFromEnv(overrides?: Partial<HostedTenant> & {
+    env?: Record<string, string | undefined>;
+}): HostedTenant | undefined;
 declare function hostedClientFromEnv(overrides?: Partial<HostedTenant> & {
     env?: Record<string, string | undefined>;
 }): HostedClient | undefined;
 
-export { type EvalRunCellScore, type EvalRunEvent, type EvalRunGenerationSnapshot, type EvalRunStatus, HOSTED_WIRE_VERSION, type HostedClient, type HostedIngestHeaders, type HostedTenant, type HostedWireVersion, type IngestEvalRunsRequest, type IngestResponse, type IngestTracesRequest, type TraceSpanEvent, createHostedClient, hostedClientFromEnv };
+export { type EvalRunCellScore, type EvalRunEvent, type EvalRunGenerationSnapshot, type EvalRunStatus, HOSTED_WIRE_VERSION, type HostedClient, type HostedIngestHeaders, type HostedTenant, type HostedWireVersion, type IngestEvalRunsRequest, type IngestResponse, type IngestTracesRequest, type TraceSpanEvent, createHostedClient, hostedClientFromEnv, hostedTenantFromEnv };

package/dist/hosted/index.js

@@ -1,12 +1,14 @@
 import {
   HOSTED_WIRE_VERSION,
   createHostedClient,
-  hostedClientFromEnv
-} from "../chunk-DFS3FEXO.js";
+  hostedClientFromEnv,
+  hostedTenantFromEnv
+} from "../chunk-ZZUXHH3R.js";
 import "../chunk-PZ5AY32C.js";
 export {
   HOSTED_WIRE_VERSION,
   createHostedClient,
-  hostedClientFromEnv
+  hostedClientFromEnv,
+  hostedTenantFromEnv
 };
 //# sourceMappingURL=index.js.map

package/dist/index.d.ts

@@ -42,6 +42,7 @@ export { A as Artifact, E as EventKind, i as FAILURE_CLASSES, F as FailureClass,
 import { T as TraceStore, R as RunFilter } from './store-CKUAgsJz.js';
 export { E as EventFilter, F as FileSystemTraceStore, a as FileSystemTraceStoreOptions, I as InMemoryTraceStore, S as SpanFilter } from './store-CKUAgsJz.js';
 export { D as DEFAULT_FAILURE_RULES, b as FailureClassification, c as FailureContext, d as FailureRule, e as classifyFailure } from './failure-cluster-CL7IVgkJ.js';
+export { P as ProjectRuntimeTrajectoryEvidenceOptions, a as RuntimeTrajectoryEvidenceProjection, b as RuntimeTrajectoryEvidenceSummary, c as RuntimeTrajectoryHookEvent, R as RuntimeTrajectoryRecord, d as RuntimeTrajectoryRunRecord, p as parseRuntimeTrajectoryHookEvent, e as projectRuntimeTrajectoryEvidence } from './runtime-trajectory-BLRiaifm.js';
 import { a as BaselineReport } from './baseline-DE36-Np7.js';
 export { B as BaselineOptions, M as MetricSamples, b as MetricVerdict, T as ToolStats, d as ToolUseMetrics, e as ToolUseOptions, f as compareToBaseline, c as computeToolUseMetrics, i as iqr, w as welchsTTest } from './baseline-DE36-Np7.js';
 import { T as Trajectory, a as TrajectoryStep } from './trajectory-GEdXJCL5.js';
@@ -3738,6 +3739,124 @@ declare class Mutex {
     get pending(): number;
 }
 
+/**
+ * DescriptionLengthGate — a Minimum-Description-Length promotion gate, the
+ * Builder/Breaker acceptance rule from Wang & Buehler, "Self-Revising Discovery
+ * Systems for Science" (arXiv:2606.01444, MIT LAMM), eq. 5:
+ *
+ *     L(M, D) = L_model(M) + L_data(D | M)
+ *     accept M' over M  iff  L(M', D∪E) < L(M, D∪E)
+ *
+ * Both candidate and baseline are scored on the SAME enlarged evidence set
+ * (every accumulated task — NOT just the held-out split), and the candidate is
+ * accepted only if it lowers the TOTAL bit cost. This is the gate's whole
+ * point and what distinguishes it from a monotone held-out delta:
+ *
+ *   - L_model(M)  — the candidate's own description length: the compressed size
+ *     of its model text (a prompt, skill, profile, or symbolic model). A bigger
+ *     model pays more bits.
+ *   - L_data(D|M) — the residual: bits of "surprise" that the model did not
+ *     simply succeed, −Σ_i log2(s_i) over the model's per-task score s_i.
+ *     Perfect scores cost 0 bits; failure costs a lot (capped, not infinite).
+ *
+ * A candidate that merely memorizes new counterexamples grows L_model faster
+ * than it shrinks L_data and LOSES — a principled, complexity-penalized
+ * alternative to HeldOutGate's held-out paired delta. Use this gate
+ * when the model text whose size you want to penalize is available; use
+ * HeldOutGate when promotion should turn on held-out generalization with an
+ * overfit-gap check instead.
+ *
+ * Scale / calibration: a gzip'd prose model is hundreds–thousands of bits;
+ * a single task contributes at most −log2(scoreFloor) data bits (≈10). So with
+ * little evidence the model term dominates and the gate is conservative about
+ * model GROWTH — it promotes a larger model only once accumulated evidence
+ * genuinely pays for the added bits (exactly the paper's regime, where D∪E
+ * grows). `lambda` is the lever: λ<1 discounts model bits (more permissive),
+ * λ>1 is stricter. A shrinking-or-equal model that does no worse always wins.
+ *
+ * Stateless: construct once with the description-length budget, call
+ * `evaluate` per (candidate, baseline) pair.
+ */
+type DescriptionLengthRejectionCode = 'few_tasks' | 'no_total_gain' | 'model_bloat';
+interface DescriptionLengthConfig {
+    /** Stable label of the baseline. Required — paper-grade evaluation never
+     *  compares two unlabelled candidates. */
+    baselineKey: string;
+    /** Weight on model bits relative to data bits (the description-length
+     *  budget λ). 1 = bits are bits. >1 = more complexity-averse. Default 1. */
+    lambda?: number;
+    /** The candidate must beat the baseline by at least this many bits to
+     *  promote — a robustness margin against measurement noise. Default 0
+     *  (strict `<`, as the paper). */
+    marginBits?: number;
+    /** Per-task score floor for the residual code: −log2(max(s, floor)). Caps a
+     *  total-failure task's surprise instead of letting it diverge. Default
+     *  2^-10 (a failed task costs 10 bits, not ∞). */
+    scoreFloor?: number;
+    /** Minimum number of shared (candidate, baseline) tasks before the gate will
+     *  consider promoting. Default 3. */
+    minTasks?: number;
+}
+interface DescriptionLengthEvidence {
+    /** Shared tasks scored on both sides (the enlarged evidence D∪E). */
+    tasks: number;
+    /** Compressed-model bits — L_model. */
+    modelBits: {
+        candidate: number;
+        baseline: number;
+    };
+    /** Residual surprise bits — L_data(D|M). */
+    dataBits: {
+        candidate: number;
+        baseline: number;
+    };
+    /** λ·L_model + L_data — the quantity the gate minimizes. */
+    totalBits: {
+        candidate: number;
+        baseline: number;
+    };
+    /** candidate − baseline total. Negative = candidate compresses better. */
+    deltaBits: number;
+    /** Per-component deltas, for audit: did the win come from a smaller model,
+     *  better outcomes, or both? */
+    modelBitsDelta: number;
+    dataBitsDelta: number;
+}
+interface DescriptionLengthDecision {
+    promote: boolean;
+    candidateId: string;
+    baselineId: string;
+    evidence: DescriptionLengthEvidence;
+    reason: string;
+    rejectionCode: DescriptionLengthRejectionCode | null;
+}
+interface DescriptionLengthCandidate {
+    /** The model text whose size is L_model (a prompt, skill, profile, or
+     *  symbolic model; concatenated if several files). */
+    content: string;
+    /** Runs whose per-task scores form L_data. */
+    runs: RunRecord[];
+}
+/** Compressed-model bits — the model's description length L_model. gzip is a
+ *  deterministic, dependency-free stand-in for Kolmogorov complexity; it
+ *  rewards genuine compactness and penalizes boilerplate padding. */
+declare function modelDescriptionBits(content: string): number;
+/** Residual surprise L_data(D|M) = −Σ_i log2(max(s_i, floor)) over the given
+ *  per-task scores. Lower = the model more reliably succeeds. */
+declare function dataDescriptionBits(scoreByTask: Map<string, number>, keys: Iterable<string>, scoreFloor: number): number;
+declare class DescriptionLengthGate {
+    private readonly baselineKey;
+    private readonly lambda;
+    private readonly marginBits;
+    private readonly scoreFloor;
+    private readonly minTasks;
+    constructor(config: DescriptionLengthConfig);
+    /** Decide whether `candidate` should replace `baseline`. Both are scored on
+     *  the shared task set (the enlarged evidence); the candidate promotes only
+     *  if λ·L_model + L_data is strictly lower by at least `marginBits`. */
+    evaluate(candidate: DescriptionLengthCandidate, baseline: DescriptionLengthCandidate): DescriptionLengthDecision;
+}
+
 /**
  * Walk a personas directory and return every file matching the convention
  * `NN-slug.{yaml,yml,json,md}`. Sorted by filename so the numeric prefix
@@ -4610,4 +4729,4 @@ declare namespace index {
   export { type index_AgentProfile as AgentProfile, type index_AgentProfileSection as AgentProfileSection, index_BASELINE_ROLES as BASELINE_ROLES, type index_BaselineRoleKey as BaselineRoleKey, type index_ProfileSkill as ProfileSkill, index_applyDomainPatch as applyDomainPatch, index_baselineProfile as baselineProfile, index_baselineProfileFromRole as baselineProfileFromRole, index_engineerRole as engineerRole, index_generalistRole as generalistRole, index_prodProfile as prodProfile, index_profileToSurface as profileToSurface, index_renderProfile as renderProfile, index_researcherRole as researcherRole, index_sectionHash as sectionHash };
 }
 
-export { type ActiveLearningOptions, type AdapterRun, AgentDriver, type AgentDriverConfig, AgentEvalError, AgentProfile$1 as AgentProfile, type AgreementResult, type AlignmentOp, AnalyzeTracesInput, AnalyzeTracesOptions, AnalyzeTracesResult, type AntiSlopConfig, type AntiSlopIssue, type AntiSlopReport, type AssertCrossFamilyOptions, type AssertSingleBackendOptions, type AutoPrClient, AxGepaSteeringOptimizer, type AxSteeringOptimizerConfig, type BackendDescriptor, BaselineReport, BehaviorAssertion, BenchmarkReport, BenchmarkRunner, BenchmarkRunnerConfig, type BisectOptions, type BisectResult, type BisectStep, BudgetBreachError, BudgetGuard, BudgetLedgerEntry, BudgetSpec, type BuildAgreementJudgeOptions, CallExpectation, type CanaryAlert, type CanaryKind, type CanaryLeak, type CanaryOptions, type CanaryReport, type CanarySeverity, type CandidateScenario, type CausalAttributionReport, type CellVerdict, ChatRequest, CheckResult, CollectedArtifacts, type CommandRunner, type CompareLabels, CompletionCriterion, type ContinuityCheck, type ContinuityCheckResult, type ContinuityReport, type ContinuitySnapshotPair, type ContractMetric, type ContractReport, ConvergenceTracker, type CostEntry, type CostSummary, CostTracker, type CounterfactualContext, type CounterfactualMutation, type CounterfactualResult, type CounterfactualRunner, CreateChatClientOpts, type CreateDefaultReviewerOptions, type CreateSandboxPoolOpts, CrossFamilyError, type CrossTraceDiff, type CrossTraceDiffOptions, DEFAULT_AGENT_SLOS, DEFAULT_FINDERS, DEFAULT_MUTATION_PRIMITIVES, DEFAULT_MUTATORS, DEFAULT_PR_REVIEW_SCORE_WEIGHTS, DEFAULT_SEVERITY_WEIGHTS, Dataset, DatasetScenario, type DecideNextUserTurnOpts, type DeployFamily, type DeployGateLayerInput, type DeployRunResult, type DeployRunner, type DiffScorecardOptions, type DirEntry, type DiscoverPersonasOptions, type DiscoveredPersona, DriverResult, DriverState, DualAgentBench, type DualAgentBenchConfig, type DualAgentReport, type DualAgentRound, type DualAgentScenario, type DualAgentScenarioResult, ERROR_COUNT_PATTERNS, type EnsembleAggregate, type ErrorCountPattern, type EvolutionRound, type ExecutorConfig, type Expectation, type ExportedRewardModel, type ExtractOptions, type ExtractResult, type FactorContribution, type FactorialCell, FeedbackLabel, FeedbackTrajectory, FeedbackTrajectoryStore, type FieldAgreementSpec, type FileChange, type FlowAction, type FlowLayerEnv, type FlowLayerFactoryInput, type FlowRunner, type FlowRunnerStepResult, type FlowSpec, type FlowStep, type GhCliClientOptions, type GoldScenario, type GoldSplit, type GoldenSeverity, type GoldenSpec, HarnessConfig, HoldoutAuditor, type HttpGithubClientOptions, type HypothesisManifest, type HypothesisResult, INTENT_MATCH_JUDGE_VERSION, type ImageData, InMemoryWorkspaceInspector, type InferenceScorer, type InspectorContext, type IntentMatchInput, type IntentMatchOptions, type IntentMatchResult, type InteractionContribution, type JudgeFamily, type JudgeFleetOptions, JudgeFn, type JudgeReplayResult, type JudgeRetryOutcome, type JudgeRetryPolicy, JudgeRunner, type JudgeVerdict, type KeywordConceptSpec, type KeywordCoverageFinding, type KeywordCoverageOptions, type KeywordCoverageResult, type LangfuseEnvelope, type LangfuseGeneration, type LangfuseScore, Layer, LayerResult, type LiveProofArtifact, type LiveProofConfig, type LiveProofContext, type LiveProofResult, LlmClientOptions, LlmSpan, LockedJsonlAppender, MODEL_PRICING, type MatchResult, type MatcherResult, type MergeOptions, MetricsCollector, type MuffledFinder, type MuffledFinding, type MultiToolchainLayerConfig, type Mutator, Mutex, type Oracle, type OracleObservation, type OracleReport, type OracleResult, type OrthogonalityInput, type OrthogonalityResult, OtelExportConfig, OtelExporter, type OtelPipelineHandle, type OtelPipelineOptions, PairwiseSteeringOptimizer, type ParaphraseRobustnessScenarioInput, type ParaphraseRobustnessScenarioResult, type ParseStudentLabel, PersonaConfig, type Playbook, type PlaybookEntry, type PoolSlot, type PrReviewAuditCase, type PrReviewBenchmarkSummary, type PrReviewComment, type PrReviewMatchedFinding, type PrReviewOutcome, type PrReviewReferenceFinding, type PrReviewScore, type PrReviewScoreWeights, type PrReviewSeverity, type PrReviewSource, ProductClient, ProductClientConfig, type PromptHandle, PromptRegistry, type ProposeAutomatedPullRequestInput, type ProposeAutomatedPullRequestResult, type RecordRunsOptions, type ReferenceMatchResult, type ReferenceReplayAdapter, type ReferenceReplayAdapterFn, type ReferenceReplayAdapterLike, type ReferenceReplayAggregate, type ReferenceReplayCandidate, type ReferenceReplayCase, type ReferenceReplayCaseRun, type ReferenceReplayExecutionScenario, type ReferenceReplayItem, type ReferenceReplayMatch, type ReferenceReplayMatchStrategy, type ReferenceReplayMatcher, type ReferenceReplayPromotionDecision, type ReferenceReplayPromotionPolicy, type ReferenceReplayRun, type ReferenceReplayRunContext, type ReferenceReplayRunOptions, type ReferenceReplayRunStore, type ReferenceReplayScenario, type ReferenceReplayScenarioScore, type ReferenceReplayScore, type ReferenceReplayScoreOptions, type ReferenceReplaySplit, type ReferenceReplaySplitComparison, type ReferenceReplaySteeringRowsOptions, type ReflectionContext, type ReflectionProposal, ReleaseConfidenceScorecard, ReleaseConfidenceThresholds, type RenderStudentPrompt, type RepoRef, type ReviewerMemoryEntry, type ReviewerOutput, type ReviewerPromptInput, type ReviewerSoftFailDefaults, type ReviewerVerificationSummary, type RobustnessResult, Run, type RunCommandInput, type RunCommandResult, type RunDistillationOptions, type RunDistillationResult, RunFilter, RunRecord, RunScore, RunScoreWeights, SandboxDriver, SandboxHarnessResult, type SandboxJudgeKind, type SandboxJudgeResult, type SandboxJudgeSpec, type SandboxPool, type ScanOptions, Scenario, type ScenarioCost, ScenarioFile, ScenarioRegistry, ScenarioResult, type Scorecard, type ScorecardCell, type ScorecardCellDiff, type ScorecardDiff, type ScorecardEntry, type ScorecardLogLine, type ScoredTarget, type SelfPlayOptions, type SelfPlayProposer, type SelfPlayScorer, type SeriesConvergenceOptions, type SeriesConvergenceResult, Severity, type SignedManifest, type SignedManifestAlgo, type SingleBackendDivergence, SingleBackendError, type SingleBackendField, type SingleBackendReport, type Slo, type SloCheckResult, type SloComparator, type SloReport, type SloSeverity, type SlopCategory, type SlotFactory, type SplitGoldOptions, SteeringBundle, type SteeringOptimizationResult, type SteeringOptimizationRow, type SteeringOptimizationSelector, type SteeringOptimizerBackend, type SteeringOptimizerConfig, type StepAttribution, type SynthesisReason, type SynthesisTarget, TestResult, type ThresholdContract, TokenCounter, type TokenSpec, TraceEmitter, TraceStore, type TracedAnalystOptions, type TracedJudgeOptions, Trajectory, TrajectoryStep, type TrialTrace, TurnMetrics, UI_FINDING_SEVERITIES, UI_LENSES, UNIVERSAL_FINDERS, type UiFinding, type UiFindingScreenshot, type UiFindingSeverity, type UiLens, VerifyContext, type VisualDiffOptions, type VisualDiffResult, type ViteDeployRunnerInput, type WorkspaceAssertion, type WorkspaceAssertionResult, type WorkspaceInspector, type WorkspaceSnapshot, type WranglerDeployRunnerInput, adversarialJudge, aggregateJudgeVerdicts, aggregatePrReviewScore, analyzeAntiSlop, analyzeSeries, appendScorecard, assertCrossFamily, assertSingleBackend, attributeCounterfactuals, bisect, buildAgreementJudge, buildDriverSystemPrompt, buildReflectionPrompt, buildReviewerPrompt, canaryLeakView, canonicalize, causalAttribution, checkBehavioralCanary, checkCanaries, checkSlos, codeExecutionJudge, coherenceJudge, collectionPreserved, commentsForSource, commitBisect, compareReferenceReplay, compilerJudge, createAntiSlopJudge, createCustomJudge, createDefaultReviewer, createDomainExpertJudge, createIntentMatchJudge, createSandboxPool, crossTraceDiff, decideNextUserTurn, decideReferenceReplayPromotion, decideReferenceReplayRunPromotion, defaultJudges, defaultParseStudentLabel, defaultReferenceReplayMatcher, defaultRenderStudentPrompt, deployGateLayer, diffScorecard, discoverPersonas, distillPlaybook, estimateCost, estimateTokens, evaluateContract, evaluateHypothesis, evaluateOracles, executeScenario, expectAgent, exportRewardModel, extractAssetUrls, extractErrorCount, fieldAgreement, fileContains, fileExists, findAutoMatchNoExpectation, findConstructorCwdDropped, findFallbackToPass, findLiteralTruePass, findSkipCountsAsPass, flowLayer, formatBenchmarkReport, formatDriverReport, formatFindings, formatScorecardDiff, ghCliClient, precision as goldenPrecision, hashContent, hashJson, htmlContainsElement, httpGithubClient, inMemoryReferenceReplayStore, isModelPriced, isOtelConfigured, jsonShape, jsonlReferenceReplayStore, judgeFamily, keyPreserved, linterJudge, loadGoldScenarios, loadScorecard, loadScorerFromGrader, localCommandRunner, lowercaseMutator, matchGoldens, mergeLayerResults, multiToolchainLayer, notBlocked, paraphraseRobustness, paraphraseRobustnessScenarios, parseGoldJsonl, parseReflectionResponse, passOrthogonality, pixelDeltaRatio, politenessPrefixMutator, printDriverSummary, index as profile, promptBisect, proposeAutomatedPullRequest, proposeSynthesisTargets, recordRuns, recordRunsToScorecard, referenceReplayRunsToSteeringRows, referenceReplayScenarioToRunScore, regexMatches, renderMarkdownReport, renderPlaybookMarkdown, replayScorerOverCorpus, replayTraceThroughJudge, resetLockedAppendersForTesting, resolveModelPricing, rowCount, rowWhere, runAssertions, runBehavioralCanaries, runCanaries, runCounterfactual, runDistillation, runE2EWorkflow, runExpectations, runIntentMatchJudge, runJudgeFleet, runKeywordCoverageJudge, runKeywordCoverageJudgeUrl, runLiveProof, runReferenceReplay, runSelfPlay, scanForMuffledGates, scoreContinuity, scorePrReviewComments, scorePrReviewSource, scoreReferenceReplay, securityJudge, sentenceReorderMutator, signManifest, splitGold, statusAdvanced, summarizePrReviewBenchmark, testJudge, textInSnapshot, toLangfuseEnvelope, toPrometheusText, traceJudge, traceJudgeEnsemble, tracedAnalyzeTraces, typoMutator, urlContains, verifyManifest, visualDiff, viteDeployRunner, weightedRecall, whitespaceCollapseMutator, withJudgeRetry, withOtelPipeline, wranglerDeployRunner };
+export { type ActiveLearningOptions, type AdapterRun, AgentDriver, type AgentDriverConfig, AgentEvalError, AgentProfile$1 as AgentProfile, type AgreementResult, type AlignmentOp, AnalyzeTracesInput, AnalyzeTracesOptions, AnalyzeTracesResult, type AntiSlopConfig, type AntiSlopIssue, type AntiSlopReport, type AssertCrossFamilyOptions, type AssertSingleBackendOptions, type AutoPrClient, AxGepaSteeringOptimizer, type AxSteeringOptimizerConfig, type BackendDescriptor, BaselineReport, BehaviorAssertion, BenchmarkReport, BenchmarkRunner, BenchmarkRunnerConfig, type BisectOptions, type BisectResult, type BisectStep, BudgetBreachError, BudgetGuard, BudgetLedgerEntry, BudgetSpec, type BuildAgreementJudgeOptions, CallExpectation, type CanaryAlert, type CanaryKind, type CanaryLeak, type CanaryOptions, type CanaryReport, type CanarySeverity, type CandidateScenario, type CausalAttributionReport, type CellVerdict, ChatRequest, CheckResult, CollectedArtifacts, type CommandRunner, type CompareLabels, CompletionCriterion, type ContinuityCheck, type ContinuityCheckResult, type ContinuityReport, type ContinuitySnapshotPair, type ContractMetric, type ContractReport, ConvergenceTracker, type CostEntry, type CostSummary, CostTracker, type CounterfactualContext, type CounterfactualMutation, type CounterfactualResult, type CounterfactualRunner, CreateChatClientOpts, type CreateDefaultReviewerOptions, type CreateSandboxPoolOpts, CrossFamilyError, type CrossTraceDiff, type CrossTraceDiffOptions, DEFAULT_AGENT_SLOS, DEFAULT_FINDERS, DEFAULT_MUTATION_PRIMITIVES, DEFAULT_MUTATORS, DEFAULT_PR_REVIEW_SCORE_WEIGHTS, DEFAULT_SEVERITY_WEIGHTS, Dataset, DatasetScenario, type DecideNextUserTurnOpts, type DeployFamily, type DeployGateLayerInput, type DeployRunResult, type DeployRunner, type DescriptionLengthCandidate, type DescriptionLengthConfig, type DescriptionLengthDecision, type DescriptionLengthEvidence, DescriptionLengthGate, type DescriptionLengthRejectionCode, type DiffScorecardOptions, type DirEntry, type DiscoverPersonasOptions, type DiscoveredPersona, DriverResult, DriverState, DualAgentBench, type DualAgentBenchConfig, type DualAgentReport, type DualAgentRound, type DualAgentScenario, type DualAgentScenarioResult, ERROR_COUNT_PATTERNS, type EnsembleAggregate, type ErrorCountPattern, type EvolutionRound, type ExecutorConfig, type Expectation, type ExportedRewardModel, type ExtractOptions, type ExtractResult, type FactorContribution, type FactorialCell, FeedbackLabel, FeedbackTrajectory, FeedbackTrajectoryStore, type FieldAgreementSpec, type FileChange, type FlowAction, type FlowLayerEnv, type FlowLayerFactoryInput, type FlowRunner, type FlowRunnerStepResult, type FlowSpec, type FlowStep, type GhCliClientOptions, type GoldScenario, type GoldSplit, type GoldenSeverity, type GoldenSpec, HarnessConfig, HoldoutAuditor, type HttpGithubClientOptions, type HypothesisManifest, type HypothesisResult, INTENT_MATCH_JUDGE_VERSION, type ImageData, InMemoryWorkspaceInspector, type InferenceScorer, type InspectorContext, type IntentMatchInput, type IntentMatchOptions, type IntentMatchResult, type InteractionContribution, type JudgeFamily, type JudgeFleetOptions, JudgeFn, type JudgeReplayResult, type JudgeRetryOutcome, type JudgeRetryPolicy, JudgeRunner, type JudgeVerdict, type KeywordConceptSpec, type KeywordCoverageFinding, type KeywordCoverageOptions, type KeywordCoverageResult, type LangfuseEnvelope, type LangfuseGeneration, type LangfuseScore, Layer, LayerResult, type LiveProofArtifact, type LiveProofConfig, type LiveProofContext, type LiveProofResult, LlmClientOptions, LlmSpan, LockedJsonlAppender, MODEL_PRICING, type MatchResult, type MatcherResult, type MergeOptions, MetricsCollector, type MuffledFinder, type MuffledFinding, type MultiToolchainLayerConfig, type Mutator, Mutex, type Oracle, type OracleObservation, type OracleReport, type OracleResult, type OrthogonalityInput, type OrthogonalityResult, OtelExportConfig, OtelExporter, type OtelPipelineHandle, type OtelPipelineOptions, PairwiseSteeringOptimizer, type ParaphraseRobustnessScenarioInput, type ParaphraseRobustnessScenarioResult, type ParseStudentLabel, PersonaConfig, type Playbook, type PlaybookEntry, type PoolSlot, type PrReviewAuditCase, type PrReviewBenchmarkSummary, type PrReviewComment, type PrReviewMatchedFinding, type PrReviewOutcome, type PrReviewReferenceFinding, type PrReviewScore, type PrReviewScoreWeights, type PrReviewSeverity, type PrReviewSource, ProductClient, ProductClientConfig, type PromptHandle, PromptRegistry, type ProposeAutomatedPullRequestInput, type ProposeAutomatedPullRequestResult, type RecordRunsOptions, type ReferenceMatchResult, type ReferenceReplayAdapter, type ReferenceReplayAdapterFn, type ReferenceReplayAdapterLike, type ReferenceReplayAggregate, type ReferenceReplayCandidate, type ReferenceReplayCase, type ReferenceReplayCaseRun, type ReferenceReplayExecutionScenario, type ReferenceReplayItem, type ReferenceReplayMatch, type ReferenceReplayMatchStrategy, type ReferenceReplayMatcher, type ReferenceReplayPromotionDecision, type ReferenceReplayPromotionPolicy, type ReferenceReplayRun, type ReferenceReplayRunContext, type ReferenceReplayRunOptions, type ReferenceReplayRunStore, type ReferenceReplayScenario, type ReferenceReplayScenarioScore, type ReferenceReplayScore, type ReferenceReplayScoreOptions, type ReferenceReplaySplit, type ReferenceReplaySplitComparison, type ReferenceReplaySteeringRowsOptions, type ReflectionContext, type ReflectionProposal, ReleaseConfidenceScorecard, ReleaseConfidenceThresholds, type RenderStudentPrompt, type RepoRef, type ReviewerMemoryEntry, type ReviewerOutput, type ReviewerPromptInput, type ReviewerSoftFailDefaults, type ReviewerVerificationSummary, type RobustnessResult, Run, type RunCommandInput, type RunCommandResult, type RunDistillationOptions, type RunDistillationResult, RunFilter, RunRecord, RunScore, RunScoreWeights, SandboxDriver, SandboxHarnessResult, type SandboxJudgeKind, type SandboxJudgeResult, type SandboxJudgeSpec, type SandboxPool, type ScanOptions, Scenario, type ScenarioCost, ScenarioFile, ScenarioRegistry, ScenarioResult, type Scorecard, type ScorecardCell, type ScorecardCellDiff, type ScorecardDiff, type ScorecardEntry, type ScorecardLogLine, type ScoredTarget, type SelfPlayOptions, type SelfPlayProposer, type SelfPlayScorer, type SeriesConvergenceOptions, type SeriesConvergenceResult, Severity, type SignedManifest, type SignedManifestAlgo, type SingleBackendDivergence, SingleBackendError, type SingleBackendField, type SingleBackendReport, type Slo, type SloCheckResult, type SloComparator, type SloReport, type SloSeverity, type SlopCategory, type SlotFactory, type SplitGoldOptions, SteeringBundle, type SteeringOptimizationResult, type SteeringOptimizationRow, type SteeringOptimizationSelector, type SteeringOptimizerBackend, type SteeringOptimizerConfig, type StepAttribution, type SynthesisReason, type SynthesisTarget, TestResult, type ThresholdContract, TokenCounter, type TokenSpec, TraceEmitter, TraceStore, type TracedAnalystOptions, type TracedJudgeOptions, Trajectory, TrajectoryStep, type TrialTrace, TurnMetrics, UI_FINDING_SEVERITIES, UI_LENSES, UNIVERSAL_FINDERS, type UiFinding, type UiFindingScreenshot, type UiFindingSeverity, type UiLens, VerifyContext, type VisualDiffOptions, type VisualDiffResult, type ViteDeployRunnerInput, type WorkspaceAssertion, type WorkspaceAssertionResult, type WorkspaceInspector, type WorkspaceSnapshot, type WranglerDeployRunnerInput, adversarialJudge, aggregateJudgeVerdicts, aggregatePrReviewScore, analyzeAntiSlop, analyzeSeries, appendScorecard, assertCrossFamily, assertSingleBackend, attributeCounterfactuals, bisect, buildAgreementJudge, buildDriverSystemPrompt, buildReflectionPrompt, buildReviewerPrompt, canaryLeakView, canonicalize, causalAttribution, checkBehavioralCanary, checkCanaries, checkSlos, codeExecutionJudge, coherenceJudge, collectionPreserved, commentsForSource, commitBisect, compareReferenceReplay, compilerJudge, createAntiSlopJudge, createCustomJudge, createDefaultReviewer, createDomainExpertJudge, createIntentMatchJudge, createSandboxPool, crossTraceDiff, dataDescriptionBits, decideNextUserTurn, decideReferenceReplayPromotion, decideReferenceReplayRunPromotion, defaultJudges, defaultParseStudentLabel, defaultReferenceReplayMatcher, defaultRenderStudentPrompt, deployGateLayer, diffScorecard, discoverPersonas, distillPlaybook, estimateCost, estimateTokens, evaluateContract, evaluateHypothesis, evaluateOracles, executeScenario, expectAgent, exportRewardModel, extractAssetUrls, extractErrorCount, fieldAgreement, fileContains, fileExists, findAutoMatchNoExpectation, findConstructorCwdDropped, findFallbackToPass, findLiteralTruePass, findSkipCountsAsPass, flowLayer, formatBenchmarkReport, formatDriverReport, formatFindings, formatScorecardDiff, ghCliClient, precision as goldenPrecision, hashContent, hashJson, htmlContainsElement, httpGithubClient, inMemoryReferenceReplayStore, isModelPriced, isOtelConfigured, jsonShape, jsonlReferenceReplayStore, judgeFamily, keyPreserved, linterJudge, loadGoldScenarios, loadScorecard, loadScorerFromGrader, localCommandRunner, lowercaseMutator, matchGoldens, mergeLayerResults, modelDescriptionBits, multiToolchainLayer, notBlocked, paraphraseRobustness, paraphraseRobustnessScenarios, parseGoldJsonl, parseReflectionResponse, passOrthogonality, pixelDeltaRatio, politenessPrefixMutator, printDriverSummary, index as profile, promptBisect, proposeAutomatedPullRequest, proposeSynthesisTargets, recordRuns, recordRunsToScorecard, referenceReplayRunsToSteeringRows, referenceReplayScenarioToRunScore, regexMatches, renderMarkdownReport, renderPlaybookMarkdown, replayScorerOverCorpus, replayTraceThroughJudge, resetLockedAppendersForTesting, resolveModelPricing, rowCount, rowWhere, runAssertions, runBehavioralCanaries, runCanaries, runCounterfactual, runDistillation, runE2EWorkflow, runExpectations, runIntentMatchJudge, runJudgeFleet, runKeywordCoverageJudge, runKeywordCoverageJudgeUrl, runLiveProof, runReferenceReplay, runSelfPlay, scanForMuffledGates, scoreContinuity, scorePrReviewComments, scorePrReviewSource, scoreReferenceReplay, securityJudge, sentenceReorderMutator, signManifest, splitGold, statusAdvanced, summarizePrReviewBenchmark, testJudge, textInSnapshot, toLangfuseEnvelope, toPrometheusText, traceJudge, traceJudgeEnsemble, tracedAnalyzeTraces, typoMutator, urlContains, verifyManifest, visualDiff, viteDeployRunner, weightedRecall, whitespaceCollapseMutator, withJudgeRetry, withOtelPipeline, wranglerDeployRunner };