@tangle-network/agent-eval 0.82.0 → 0.83.0

package/CHANGELOG.md

@@ -4,6 +4,14 @@ All notable changes to `@tangle-network/agent-eval` and its sibling `agent-eval-
 
 ---
 
+## [0.83.0] — 2026-06-05 — hostedTenantFromEnv
+
+### Added
+
+- **`hostedTenantFromEnv` (`/hosted`).** Builds a `HostedTenant` config from env (the input `selfImprove({ hostedTenant })` and `emitLoopProvenance` take), with the same env precedence + overrides as `hostedClientFromEnv` — which now composes it. Returns `undefined` (not an error) when unconfigured, so a product wires `hostedTenant: hostedTenantFromEnv({ tenantId: 'my-agent' })` unconditionally and hosted ingest stays off until the env is set. Removes the env→tenant mapping every product would otherwise hand-roll when collapsing onto `selfImprove`.
+
+---
+
 ## [0.82.0] — 2026-06-05 — selfImprove forwards the full loop surface
 
 ### Changed
@@ -21,7 +29,7 @@ A deterministic offline test that drives `selfImprove` with a mock agent must no
 
 ### Added
 
-- **`aggregateJudgeVerdicts<D>` (root).** Generic judge-ensemble reducer: fan out N uncorrelated judges, mean each rubric dimension over the SURVIVORS, report the inter-rater disagreement spread, sum cost. Replaces the same reduction hand-rolled in legal (`aggregateEnsemble`), creative (`production-loop/judges.ts`), and tax (`judge-ensemble.ts`). Fail-loud: a failed judge (`perDimension: null`) is recorded in `failedJudges`, never folded into a zero; all-failed throws; a failed judge's cost is still summed. Composite reuses `weightedComposite`.
+- **`aggregateJudgeVerdicts<D>` (root).** Generic judge-ensemble reducer: fan out N uncorrelated judges, mean each rubric dimension over the SURVIVORS, report the inter-rater disagreement spread, sum cost. Replaces the same reduction hand-rolled across multiple product agents. Fail-loud: a failed judge (`perDimension: null`) is recorded in `failedJudges`, never folded into a zero; all-failed throws; a failed judge's cost is still summed. Composite reuses `weightedComposite`.
 - **`createTokenRecallChecker` (root).** The deterministic, no-LLM `CorrectnessChecker` — sibling of `createLlmCorrectnessChecker`. A produced item fulfils a requirement when its content is substantive and recalls ≥ `minRecall` of the requirement title's significant tokens. The default completion gate for apps/tests without an LLM judge.
 - **`ErrorCluster` (root + `/analyst`).** The failure-cluster element type is now a named export, so consumers import it instead of deriving `DatasetOverview['error_clusters'][number]`.
 

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/openapi.json

@@ -2,7 +2,7 @@
   "openapi": "3.1.0",
   "info": {
     "title": "@tangle-network/agent-eval — wire protocol",
-    "version": "0.82.0",
+    "version": "0.83.0",
     "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.",
     "contact": {
       "name": "Tangle Network",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tangle-network/agent-eval",
-  "version": "0.82.0",
+  "version": "0.83.0",
   "description": "Evaluate and improve AI agents from runs, traces, judges, and feedback. Compare candidates, cluster failures, measure lift, and gate releases.",
   "homepage": "https://github.com/tangle-network/agent-eval#readme",
   "repository": {