@tangle-network/agent-eval 0.65.0 → 0.67.0

package/dist/{chunk-7TPYV2ER.js → chunk-6XQIEUQ2.js}

@@ -1,14 +1,115 @@
-import {
-  BackendIntegrityError
-} from "./chunk-E22YUOAL.js";
 import {
   confidenceInterval
 } from "./chunk-ITBRCT73.js";
+import {
+  AgentEvalError
+} from "./chunk-3BFEG2F6.js";
 
 // src/campaign/run-campaign.ts
 import { createHash } from "crypto";
 import { join } from "path";
 
+// src/integrity/backend-integrity.ts
+var BackendIntegrityError = class extends AgentEvalError {
+  constructor(message, report) {
+    super("backend_integrity", message);
+    this.report = report;
+  }
+  report;
+};
+function isStubRecord(rec) {
+  return rec.tokenUsage.input === 0 && rec.tokenUsage.output === 0;
+}
+function isUncostedRecord(rec) {
+  return rec.tokenUsage.output > 0 && rec.costUsd === 0;
+}
+function summarizeBackendIntegrity(records) {
+  const totalRecords = records.length;
+  let stubRecords = 0;
+  let realRecords = 0;
+  let uncostedRecords = 0;
+  let totalInputTokens = 0;
+  let totalOutputTokens = 0;
+  let totalCostUsd = 0;
+  for (const rec of records) {
+    totalInputTokens += rec.tokenUsage.input;
+    totalOutputTokens += rec.tokenUsage.output;
+    totalCostUsd += rec.costUsd;
+    if (isStubRecord(rec)) stubRecords++;
+    else realRecords++;
+    if (isUncostedRecord(rec)) uncostedRecords++;
+  }
+  const verdict = totalRecords === 0 ? "stub" : stubRecords === totalRecords ? "stub" : stubRecords === 0 ? "real" : "mixed";
+  const diagnosis = buildDiagnosis({
+    totalRecords,
+    stubRecords,
+    realRecords,
+    uncostedRecords,
+    totalInputTokens,
+    totalOutputTokens,
+    totalCostUsd,
+    verdict
+  });
+  return {
+    totalRecords,
+    stubRecords,
+    realRecords,
+    uncostedRecords,
+    totalInputTokens,
+    totalOutputTokens,
+    totalCostUsd,
+    verdict,
+    diagnosis
+  };
+}
+function buildDiagnosis(r) {
+  if (r.totalRecords === 0) {
+    return "no records \u2014 eval produced zero runs; backend likely failed before first turn";
+  }
+  if (r.verdict === "stub") {
+    return [
+      `all ${r.totalRecords} records have zero token usage \u2014 the LLM backend was never called.`,
+      "common causes: --backend sandbox without a sandbox bridge running; stub model returning hard-coded strings;",
+      "auth misconfigured so requests were silently dropped before the LLM. Re-run with --backend tcloud and TANGLE_API_KEY set,",
+      "or boot the cli-bridge / sandbox before invoking the eval."
+    ].join(" ");
+  }
+  if (r.verdict === "mixed") {
+    const pct = (r.stubRecords / r.totalRecords * 100).toFixed(0);
+    return [
+      `${r.stubRecords}/${r.totalRecords} records (${pct}%) have zero token usage \u2014 the backend partially failed.`,
+      "common causes: rate-limit cascade (429s after the first N personas);",
+      "transient auth expiry mid-run; provider outage. Treat the affected records as missing data, not agent failures."
+    ].join(" ");
+  }
+  if (r.uncostedRecords > 0) {
+    const pct = (r.uncostedRecords / r.totalRecords * 100).toFixed(0);
+    return [
+      `${r.totalRecords} records with real LLM activity (in=${r.totalInputTokens}, out=${r.totalOutputTokens} tokens).`,
+      `${r.uncostedRecords} (${pct}%) have output tokens but costUsd=0 \u2014 cost ledger is mis-wired (no input-token`,
+      "propagation from the runtime stream into RunRecord)."
+    ].join(" ");
+  }
+  return `${r.totalRecords} records with real LLM activity (in=${r.totalInputTokens}, out=${r.totalOutputTokens} tokens, $${r.totalCostUsd.toFixed(4)}).`;
+}
+function assertRealBackend(records, opts = {}) {
+  const report = summarizeBackendIntegrity(records);
+  const allowMixed = opts.allowMixed ?? true;
+  if (report.verdict === "stub") {
+    throw new BackendIntegrityError(
+      `backend-integrity: ran against a stub or unconfigured backend \u2014 ${report.diagnosis}`,
+      report
+    );
+  }
+  if (!allowMixed && report.verdict === "mixed") {
+    throw new BackendIntegrityError(
+      `backend-integrity: partial backend failure rejected \u2014 ${report.diagnosis}`,
+      report
+    );
+  }
+  return report;
+}
+
 // src/campaign/storage.ts
 import { createRequire } from "module";
 function fsCampaignStorage() {
@@ -111,7 +212,8 @@ async function runCampaign(opts) {
             now,
             storage,
             buildTraceWriter: opts.buildTraceWriter ?? defaultBuildTraceWriter(storage),
-            signal: abortController.signal
+            signal: abortController.signal,
+            dispatchTimeoutMs: opts.dispatchTimeoutMs
           });
           cellsRef.push(result.cell);
           enforceCellUsage(result.cell, opts.expectUsage ?? "warn");
@@ -213,11 +315,15 @@ async function executeCell(args) {
     scenario: args.slot.scenario,
     rep: args.slot.rep
   });
+  const cellAbort = new AbortController();
+  const onCampaignAbort = () => cellAbort.abort(args.signal.reason);
+  if (args.signal.aborted) cellAbort.abort(args.signal.reason);
+  else args.signal.addEventListener("abort", onCampaignAbort, { once: true });
   const ctx = {
     cellId: args.slot.cellId,
     rep: args.slot.rep,
     seed: args.slot.cellSeed,
-    signal: args.signal,
+    signal: cellAbort.signal,
     trace,
     artifacts,
     cost,
@@ -225,10 +331,34 @@ async function executeCell(args) {
   };
   let artifact;
   let errorMessage;
+  const timeoutMs = args.dispatchTimeoutMs;
+  let timeoutTimer;
   try {
-    artifact = await args.opts.dispatch(args.slot.scenario, ctx);
+    const dispatched = args.opts.dispatch(args.slot.scenario, ctx);
+    if (timeoutMs !== void 0 && timeoutMs > 0) {
+      artifact = await Promise.race([
+        dispatched,
+        new Promise((_, reject) => {
+          timeoutTimer = setTimeout(() => {
+            cellAbort.abort(new Error("dispatch timeout"));
+            reject(
+              new Error(
+                `dispatch exceeded ${timeoutMs}ms for cell '${args.slot.cellId}' \u2014 aborted and failed loud (no silent hang)`
+              )
+            );
+          }, timeoutMs);
+          if (typeof timeoutTimer.unref === "function")
+            timeoutTimer.unref();
+        })
+      ]);
+    } else {
+      artifact = await dispatched;
+    }
   } catch (err) {
     errorMessage = err instanceof Error ? err.message : String(err);
+  } finally {
+    if (timeoutTimer) clearTimeout(timeoutTimer);
+    args.signal.removeEventListener("abort", onCampaignAbort);
   }
   const judgeScores = {};
   if (artifact !== void 0) {
@@ -396,8 +526,11 @@ function aggregate(samples, seed) {
 }
 
 export {
+  BackendIntegrityError,
+  summarizeBackendIntegrity,
+  assertRealBackend,
   fsCampaignStorage,
   inMemoryCampaignStorage,
   runCampaign
 };
-//# sourceMappingURL=chunk-7TPYV2ER.js.map
+//# sourceMappingURL=chunk-6XQIEUQ2.js.map

package/dist/chunk-6XQIEUQ2.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/campaign/run-campaign.ts","../src/integrity/backend-integrity.ts","../src/campaign/storage.ts"],"sourcesContent":["/**\n * @experimental\n *\n * `runCampaign` — Pass A substrate primitive. ONE function that orchestrates\n * scenarios → dispatch → artifacts → judges → aggregates, with full\n * reproducibility (seed + manifest hash), cell-level resumability, bootstrap\n * CIs, and the `LabeledScenarioStore` capture flywheel.\n *\n * Improvement loops (optimizer / gate / autoOnPromote) ride on top of this\n * primitive but live in `presets/run-improvement-loop.ts`. This file keeps\n * the core orchestrator minimal — Phase 1 of the Pass A track.\n */\n\nimport { createHash } from 'node:crypto'\nimport { join } from 'node:path'\nimport { BackendIntegrityError, type BackendIntegrityReport } from '../integrity/backend-integrity'\nimport { confidenceInterval } from '../statistics'\nimport { type CampaignStorage, fsCampaignStorage } from './storage'\nimport type {\n  CampaignAggregates,\n  CampaignArtifactWriter,\n  CampaignCellResult,\n  CampaignCostMeter,\n  CampaignResult,\n  CampaignTokenUsage,\n  CampaignTraceWriter,\n  DispatchContext,\n  DispatchFn,\n  JudgeAggregate,\n  JudgeConfig,\n  JudgeScore,\n  LabeledScenarioStore,\n  Scenario,\n  ScenarioAggregate,\n  TraceSpan,\n} from './types'\n\nexport interface RunCampaignOptions<TScenario extends Scenario, TArtifact> {\n  scenarios: TScenario[]\n  dispatch: DispatchFn<TScenario, TArtifact>\n  judges?: JudgeConfig<TArtifact, TScenario>[]\n  /** Required for reproducibility. Default 42. */\n  seed?: number\n  /** Per-scenario replicates for CI bands. Default 1; raise to 5+ for\n   *  bootstrap-tight intervals on critical eval. */\n  reps?: number\n  /** When true (default), completed cells are cached by\n   *  (manifestHash, scenarioId, rep, generation). Re-runs skip cached cells. */\n  resumable?: boolean\n  /** Optional store — when present, every artifact + judge score is captured\n   *  with the configured `captureSource`. Capture is default ON; pass `'off'`\n   *  to disable. */\n  labeledStore?: LabeledScenarioStore | 'off'\n  captureSource?: 'production-trace' | 'eval-run' | 'manual' | 'red-team' | 'synthetic'\n  captureSourceVersionHash?: string\n  /** Wall-clock cost cap across all cells. Cells beyond ceiling are skipped. */\n  costCeiling?: number\n  /** Max concurrent cells. Default 2. */\n  maxConcurrency?: number\n  /**\n   * Per-cell dispatch deadline in ms. A `dispatch` that neither resolves nor\n   * rejects within this window is a hang (a stalled model request, an\n   * exhausted runtime resource, a backend that never closes its stream). When\n   * set, the cell's `ctx.signal` is aborted and the cell is recorded as a LOUD\n   * error (`dispatch exceeded <N>ms`) so the campaign proceeds and the failure\n   * is visible — instead of one wedged cell silently hanging the whole run (and\n   * every loop/CI job above it) forever. `undefined`/`0` = unbounded (legacy).\n   */\n  dispatchTimeoutMs?: number\n  /** Required: where artifacts + traces land. */\n  runDir: string\n  /** Tracing posture. Default is the substrate's `FileSystemTraceStore` rooted\n   *  at `<runDir>/traces/`. `'off'` disables capture entirely — substrate\n   *  refuses this when the caller wires `autoOnPromote !== 'none'`. */\n  tracing?: 'on' | 'off'\n  /**\n   * Per-cell usage expectation — the early, fine-grained sibling of the\n   * batch `assertRealBackend` guard. A cell that produced an artifact (no\n   * error) but reported `costUsd === 0` AND zero tokens is a stub: the\n   * dispatch never reported LLM activity via `ctx.cost`. Modes:\n   *   - `'warn'` (default) — log the offending cell loudly, keep going.\n   *   - `'assert'` — throw `BackendIntegrityError` on the first such cell\n   *     (fail-fast; recommended for CI campaigns expecting real LLM calls).\n   *   - `'off'` — no check (replay / deterministic-only / offline analysis).\n   */\n  expectUsage?: 'assert' | 'warn' | 'off'\n  /** Test seam — override the wall clock for deterministic tests. */\n  now?: () => Date\n  /** Test seam — override per-cell trace writer factory. */\n  buildTraceWriter?: (cellId: string, dir: string) => CampaignTraceWriter\n  /** Storage backend for run/cell dirs, the resumability cache, artifacts,\n   *  and trace spans. Default: the Node filesystem (`fsCampaignStorage`).\n   *  Pass `inMemoryCampaignStorage()` to run in a filesystem-less runtime\n   *  (Cloudflare Workers, Deno, edge) — the `CampaignResult` is still\n   *  produced; artifacts/traces just aren't persisted to disk. */\n  storage?: CampaignStorage\n  /**\n   * Optional per-cell placement strategy. Returns an opaque string the\n   * substrate forwards as `ctx.placement` to the Dispatch — placement-aware\n   * Dispatches (e.g. `httpDispatch` from `/adapters/http`) use it to route\n   * each cell to the right worker, region, or sandbox. When unset, every\n   * cell receives `ctx.placement = undefined` and behaves identically to\n   * the in-process case.\n   *\n   * @example\n   *   cellPlacement: ({ scenario }) => scenario.tags?.includes('eu') ? 'eu-west' : 'us-east'\n   */\n  cellPlacement?: (input: {\n    scenario: TScenario\n    rep: number\n    generation?: number\n  }) => string | undefined\n}\n\nexport async function runCampaign<TScenario extends Scenario, TArtifact>(\n  opts: RunCampaignOptions<TScenario, TArtifact>,\n): Promise<CampaignResult<TArtifact, TScenario>> {\n  const seed = opts.seed ?? 42\n  const reps = opts.reps ?? 1\n  const resumable = opts.resumable ?? true\n  const maxConcurrency = opts.maxConcurrency ?? 2\n  const now = opts.now ?? (() => new Date())\n  const judges = opts.judges ?? []\n  const storage = opts.storage ?? fsCampaignStorage()\n\n  storage.ensureDir(opts.runDir)\n\n  const manifestHash = computeManifestHash({\n    scenarios: opts.scenarios,\n    judges: judges as unknown as JudgeConfig<unknown>[],\n    dispatchRef: opts.dispatch.name || 'anonymous',\n    seed,\n    reps,\n  })\n\n  const startedAt = now()\n  const cells: CampaignCellResult<TArtifact>[] = []\n  const artifactsByPath: Record<string, string> = {}\n\n  // Build the cell schedule (scenario × rep).\n  const schedule: Array<{ scenario: TScenario; rep: number; cellId: string; cellSeed: number }> = []\n  let cellIndex = 0\n  for (const scenario of opts.scenarios) {\n    for (let rep = 0; rep < reps; rep++) {\n      const cellId = `${scenario.id}:${rep}`\n      const cellSeed = seed + cellIndex\n      schedule.push({ scenario, rep, cellId, cellSeed })\n      cellIndex += 1\n    }\n  }\n\n  // Concurrency-limited execution.\n  let totalCostUsd = 0\n  let costCeilingReached = false\n  const abortController = new AbortController()\n  // Concurrency lanes that drain the cell schedule. Named \"lanes\" — not\n  // \"workers\" — to avoid clashing with the taxonomy's worker (= the agent\n  // harness in a sandbox, invoked behind `dispatch`). See loop-taxonomy.md.\n  const lanes: Promise<void>[] = []\n  let nextIdx = 0\n  const cellsRef = cells\n\n  for (let i = 0; i < maxConcurrency; i++) {\n    lanes.push(\n      (async () => {\n        while (true) {\n          const myIdx = nextIdx++\n          if (myIdx >= schedule.length) return\n          const slot = schedule[myIdx]!\n          if (costCeilingReached) {\n            cellsRef.push(skippedCell(slot, 'cost_ceiling_reached'))\n            continue\n          }\n          const result = await executeCell({\n            slot,\n            opts,\n            manifestHash,\n            resumable,\n            now,\n            storage,\n            buildTraceWriter: opts.buildTraceWriter ?? defaultBuildTraceWriter(storage),\n            signal: abortController.signal,\n            dispatchTimeoutMs: opts.dispatchTimeoutMs,\n          })\n          cellsRef.push(result.cell)\n          enforceCellUsage(result.cell, opts.expectUsage ?? 'warn')\n          totalCostUsd += result.cell.costUsd\n          Object.assign(artifactsByPath, result.artifactsByPath)\n          if (opts.costCeiling !== undefined && totalCostUsd >= opts.costCeiling) {\n            costCeilingReached = true\n          }\n          // Capture into LabeledScenarioStore unless explicitly disabled.\n          if (opts.labeledStore && opts.labeledStore !== 'off' && !result.cell.error) {\n            await captureToStore({\n              store: opts.labeledStore,\n              cell: result.cell,\n              scenario: slot.scenario,\n              opts,\n              now,\n            }).catch((err) => {\n              // Capture failures are non-fatal — log but don't crash the campaign.\n              // (Trace would normally land here.)\n              console.warn(\n                `[runCampaign] capture failed for ${result.cell.cellId}: ${err instanceof Error ? err.message : String(err)}`,\n              )\n            })\n          }\n        }\n      })(),\n    )\n  }\n  await Promise.all(lanes)\n\n  const endedAt = now()\n  cellsRef.sort((a, b) => a.cellId.localeCompare(b.cellId))\n\n  const aggregates = computeAggregates(\n    cellsRef,\n    judges as unknown as JudgeConfig<TArtifact>[],\n    seed,\n  )\n\n  return {\n    manifestHash,\n    seed,\n    startedAt: startedAt.toISOString(),\n    endedAt: endedAt.toISOString(),\n    durationMs: endedAt.getTime() - startedAt.getTime(),\n    cells: cellsRef,\n    aggregates,\n    runDir: opts.runDir,\n    artifactsByPath,\n    scenarios: opts.scenarios.map((s) => ({ id: s.id, kind: s.kind })),\n  }\n}\n\n// ── Internals ─────────────────────────────────────────────────────────\n\ninterface ExecuteCellArgs<TScenario extends Scenario, TArtifact> {\n  slot: { scenario: TScenario; rep: number; cellId: string; cellSeed: number }\n  opts: RunCampaignOptions<TScenario, TArtifact>\n  manifestHash: string\n  resumable: boolean\n  now: () => Date\n  storage: CampaignStorage\n  buildTraceWriter: (cellId: string, dir: string) => CampaignTraceWriter\n  signal: AbortSignal\n  dispatchTimeoutMs?: number\n}\n\nasync function executeCell<TScenario extends Scenario, TArtifact>(\n  args: ExecuteCellArgs<TScenario, TArtifact>,\n): Promise<{ cell: CampaignCellResult<TArtifact>; artifactsByPath: Record<string, string> }> {\n  const storage = args.storage\n  const cellDir = join(args.opts.runDir, args.slot.cellId.replace(/[^a-zA-Z0-9_-]/g, '_'))\n  storage.ensureDir(cellDir)\n\n  // Resumability: cache key = (manifestHash, scenarioId, rep)\n  const cachePath = join(cellDir, 'cached-result.json')\n  if (args.resumable) {\n    const raw = storage.read(cachePath)\n    if (raw !== undefined) {\n      try {\n        const cached = JSON.parse(raw) as CampaignCellResult<TArtifact>\n        if (cached.cellId === args.slot.cellId) {\n          return { cell: { ...cached, cached: true }, artifactsByPath: {} }\n        }\n      } catch {\n        // Corrupt cache — fall through to re-run.\n      }\n    }\n  }\n\n  const startMs = Date.now()\n  const trace = args.buildTraceWriter(args.slot.cellId, cellDir)\n  const artifactsByPath: Record<string, string> = {}\n  const artifacts: CampaignArtifactWriter = {\n    async write(path, content) {\n      const fullPath = join(cellDir, path)\n      storage.ensureDir(join(fullPath, '..'))\n      storage.write(fullPath, content)\n      artifactsByPath[`${args.slot.cellId}/${path}`] = fullPath\n      return fullPath\n    },\n    async writeJson(path, value) {\n      return artifacts.write(path, JSON.stringify(value, null, 2))\n    },\n  }\n  let costSoFar = 0\n  const tokensSoFar: CampaignTokenUsage = { input: 0, output: 0 }\n  const cost: CampaignCostMeter = {\n    observe(amount, source) {\n      costSoFar += amount\n      trace.span(`cost.${source}`, { amountUsd: amount }).end()\n    },\n    observeTokens(usage) {\n      tokensSoFar.input += usage.input\n      tokensSoFar.output += usage.output\n      if (usage.cached) tokensSoFar.cached = (tokensSoFar.cached ?? 0) + usage.cached\n    },\n    current() {\n      return costSoFar\n    },\n    tokens() {\n      return { ...tokensSoFar }\n    },\n  }\n\n  const placement = args.opts.cellPlacement?.({\n    scenario: args.slot.scenario,\n    rep: args.slot.rep,\n  })\n\n  // Per-cell abort signal, chained to the campaign signal. The dispatch sees\n  // THIS signal so a timeout (below) can abort just this cell's in-flight work\n  // without tearing down sibling cells — and a signal-honoring dispatch\n  // releases its open request instead of leaking it past the deadline.\n  const cellAbort = new AbortController()\n  const onCampaignAbort = () => cellAbort.abort((args.signal as { reason?: unknown }).reason)\n  if (args.signal.aborted) cellAbort.abort((args.signal as { reason?: unknown }).reason)\n  else args.signal.addEventListener('abort', onCampaignAbort, { once: true })\n\n  const ctx: DispatchContext = {\n    cellId: args.slot.cellId,\n    rep: args.slot.rep,\n    seed: args.slot.cellSeed,\n    signal: cellAbort.signal,\n    trace,\n    artifacts,\n    cost,\n    placement,\n  }\n\n  let artifact: TArtifact | undefined\n  let errorMessage: string | undefined\n  const timeoutMs = args.dispatchTimeoutMs\n  let timeoutTimer: ReturnType<typeof setTimeout> | undefined\n  try {\n    const dispatched = args.opts.dispatch(args.slot.scenario, ctx)\n    if (timeoutMs !== undefined && timeoutMs > 0) {\n      // A dispatch that never settles (stalled model request, exhausted runtime\n      // resource, a stream that never closes) must NOT hang the cell — and with\n      // it the lane, the campaign, the loop, the CI job — forever. Race it\n      // against the deadline; on timeout, abort the cell and fail it LOUD.\n      artifact = await Promise.race([\n        dispatched,\n        new Promise<never>((_, reject) => {\n          timeoutTimer = setTimeout(() => {\n            cellAbort.abort(new Error('dispatch timeout'))\n            reject(\n              new Error(\n                `dispatch exceeded ${timeoutMs}ms for cell '${args.slot.cellId}' — aborted and failed loud (no silent hang)`,\n              ),\n            )\n          }, timeoutMs)\n          if (typeof (timeoutTimer as { unref?: () => void }).unref === 'function')\n            (timeoutTimer as { unref: () => void }).unref()\n        }),\n      ])\n    } else {\n      artifact = await dispatched\n    }\n  } catch (err) {\n    errorMessage = err instanceof Error ? err.message : String(err)\n  } finally {\n    if (timeoutTimer) clearTimeout(timeoutTimer)\n    args.signal.removeEventListener('abort', onCampaignAbort)\n  }\n\n  // Run judges (only if we have an artifact). A judge that throws invalidates\n  // the cell — recorded as `error`, NOT folded into a fake composite:0 (a fake\n  // zero is indistinguishable from a real zero and poisons every aggregate).\n  const judgeScores: Record<string, JudgeScore> = {}\n  if (artifact !== undefined) {\n    for (const judge of args.opts.judges ?? []) {\n      if (judge.appliesTo && !judge.appliesTo(args.slot.scenario)) continue\n      try {\n        judgeScores[judge.name] = await runJudgeCell(judge, {\n          artifact,\n          scenario: args.slot.scenario,\n          signal: args.signal,\n        })\n      } catch (err) {\n        errorMessage = `judge '${judge.name}' failed: ${err instanceof Error ? err.message : String(err)}`\n        break\n      }\n    }\n  }\n\n  await trace.flush()\n\n  const cell: CampaignCellResult<TArtifact> = {\n    cellId: args.slot.cellId,\n    scenarioId: args.slot.scenario.id,\n    rep: args.slot.rep,\n    artifact: (artifact ?? null) as TArtifact,\n    judgeScores,\n    costUsd: costSoFar,\n    tokenUsage: { ...tokensSoFar },\n    durationMs: Date.now() - startMs,\n    seed: args.slot.cellSeed,\n    cached: false,\n    error: errorMessage,\n  }\n\n  if (!errorMessage && args.resumable) {\n    storage.write(cachePath, JSON.stringify(cell))\n  }\n\n  return { cell, artifactsByPath }\n}\n\n/**\n * Per-cell stub guard. A cell that produced an artifact (no error) but reported\n * `costUsd === 0` AND zero tokens means the dispatch never called `ctx.cost` —\n * i.e. it ran against a stub or silently dropped its usage. `'warn'` logs it,\n * `'assert'` throws (fail-fast), `'off'` skips. An errored/skipped cell or a\n * deterministic judge-only run that genuinely made no LLM call is not flagged.\n */\nfunction enforceCellUsage<TArtifact>(\n  cell: CampaignCellResult<TArtifact>,\n  mode: 'assert' | 'warn' | 'off',\n): void {\n  if (mode === 'off' || cell.error) return\n  if (cell.artifact === null || cell.artifact === undefined) return\n  const zeroTokens = cell.tokenUsage.input === 0 && cell.tokenUsage.output === 0\n  if (cell.costUsd !== 0 || !zeroTokens) return\n  const msg = `cell '${cell.cellId}' produced an artifact but reported zero cost and zero tokens — the dispatch never reported LLM usage via ctx.cost.observe/observeTokens (a stub cell)`\n  if (mode === 'assert') {\n    const report: BackendIntegrityReport = {\n      totalRecords: 1,\n      stubRecords: 1,\n      realRecords: 0,\n      uncostedRecords: 0,\n      totalInputTokens: 0,\n      totalOutputTokens: 0,\n      totalCostUsd: 0,\n      verdict: 'stub',\n      diagnosis: msg,\n    }\n    throw new BackendIntegrityError(`expectUsage: ${msg}`, report)\n  }\n  // eslint-disable-next-line no-console\n  console.warn(`[runCampaign] expectUsage: ${msg}`)\n}\n\nasync function runJudgeCell<TArtifact, TScenario extends Scenario>(\n  judge: JudgeConfig<TArtifact, TScenario>,\n  input: { artifact: TArtifact; scenario: TScenario; signal: AbortSignal },\n): Promise<JudgeScore> {\n  return judge.score(input)\n}\n\nfunction defaultBuildTraceWriter(\n  storage: CampaignStorage,\n): (cellId: string, dir: string) => CampaignTraceWriter {\n  return (cellId, dir) => {\n    const spans: Array<Record<string, unknown>> = []\n    return {\n      span(name, attributes) {\n        const startMs = Date.now()\n        const record: Record<string, unknown> = { name, cellId, startMs, ...(attributes ?? {}) }\n        const finish: TraceSpan = {\n          end(endAttrs) {\n            record.durationMs = Date.now() - startMs\n            if (endAttrs) Object.assign(record, endAttrs)\n            spans.push(record)\n          },\n          setAttribute(key, value) {\n            record[key] = value\n          },\n        }\n        return finish\n      },\n      async flush() {\n        storage.write(join(dir, 'spans.jsonl'), spans.map((s) => JSON.stringify(s)).join('\\n'))\n      },\n    }\n  }\n}\n\nfunction skippedCell<TScenario extends Scenario, TArtifact>(\n  slot: { scenario: TScenario; rep: number; cellId: string; cellSeed: number },\n  reason: string,\n): CampaignCellResult<TArtifact> {\n  return {\n    cellId: slot.cellId,\n    scenarioId: slot.scenario.id,\n    rep: slot.rep,\n    artifact: null as unknown as TArtifact,\n    judgeScores: {},\n    costUsd: 0,\n    tokenUsage: { input: 0, output: 0 },\n    durationMs: 0,\n    seed: slot.cellSeed,\n    cached: false,\n    error: `skipped: ${reason}`,\n  }\n}\n\ninterface CaptureArgs<TScenario extends Scenario, TArtifact> {\n  store: LabeledScenarioStore\n  cell: CampaignCellResult<TArtifact>\n  scenario: TScenario\n  opts: RunCampaignOptions<TScenario, TArtifact>\n  now: () => Date\n}\n\nasync function captureToStore<TScenario extends Scenario, TArtifact>(\n  args: CaptureArgs<TScenario, TArtifact>,\n): Promise<void> {\n  await args.store.observe({\n    scenario: args.scenario,\n    artifact: args.cell.artifact,\n    judgeScores: args.cell.judgeScores,\n    source: args.opts.captureSource ?? 'eval-run',\n    sourceVersionHash: args.opts.captureSourceVersionHash ?? 'unknown',\n    capturedAt: args.now().toISOString(),\n    redactionStatus: 'raw',\n  })\n}\n\n// ── Aggregates + manifest hash ────────────────────────────────────────\n\nfunction computeManifestHash(input: {\n  scenarios: Scenario[]\n  judges: JudgeConfig<unknown>[]\n  dispatchRef: string\n  seed: number\n  reps: number\n}): string {\n  const canonical = {\n    scenarios: input.scenarios.map((s) => ({ id: s.id, kind: s.kind })),\n    judges: input.judges.map((j) => ({ name: j.name, dims: j.dimensions.map((d) => d.key) })),\n    dispatch: input.dispatchRef,\n    seed: input.seed,\n    reps: input.reps,\n  }\n  return createHash('sha256').update(JSON.stringify(canonical)).digest('hex')\n}\n\nfunction computeAggregates<TArtifact>(\n  cells: CampaignCellResult<TArtifact>[],\n  judges: JudgeConfig<TArtifact>[],\n  seed: number,\n): CampaignAggregates {\n  const byJudge: Record<string, JudgeAggregate> = {}\n  for (const judge of judges) {\n    const scores: number[] = []\n    for (const cell of cells) {\n      const s = cell.judgeScores[judge.name]\n      if (s !== undefined) scores.push(s.composite)\n    }\n    byJudge[judge.name] = aggregate(scores, seed)\n  }\n  const byScenario: Record<string, ScenarioAggregate> = {}\n  const scenarioGroups = new Map<string, number[]>()\n  for (const cell of cells) {\n    const composites = Object.values(cell.judgeScores).map((s) => s.composite)\n    if (composites.length === 0) continue\n    const mean = composites.reduce((a, b) => a + b, 0) / composites.length\n    const arr = scenarioGroups.get(cell.scenarioId) ?? []\n    arr.push(mean)\n    scenarioGroups.set(cell.scenarioId, arr)\n  }\n  for (const [scenarioId, samples] of scenarioGroups) {\n    const ag = aggregate(samples, seed)\n    byScenario[scenarioId] = { meanComposite: ag.mean, ci95: ag.ci95, n: ag.n }\n  }\n  return {\n    byJudge,\n    byScenario,\n    totalCostUsd: cells.reduce((a, c) => a + c.costUsd, 0),\n    cellsExecuted: cells.filter((c) => !c.error).length,\n    cellsSkipped: cells.filter((c) => c.error?.startsWith('skipped:')).length,\n    cellsCached: cells.filter((c) => c.cached).length,\n    cellsFailed: cells.filter((c) => c.error && !c.error.startsWith('skipped:')).length,\n  }\n}\n\n// Percentile bootstrap CI95 via seeded resampling. Deterministic for a given\n// seed — same campaign re-run produces identical CI bands. Falls back to\n// degenerate intervals at n<=1 (the bootstrap is undefined there).\nfunction aggregate(samples: number[], seed: number): JudgeAggregate {\n  const n = samples.length\n  if (n === 0) return { mean: 0, stdev: 0, ci95: [0, 0], n: 0 }\n  const mean = samples.reduce((a, b) => a + b, 0) / n\n  const variance = samples.reduce((a, b) => a + (b - mean) ** 2, 0) / Math.max(1, n - 1)\n  const stdev = Math.sqrt(variance)\n  const ci = confidenceInterval(samples, 0.95, { seed, resamples: 1000 })\n  return { mean, stdev, ci95: [ci.lower, ci.upper], n }\n}\n","/**\n * Backend-integrity guard: distinguish \"agent failed\" from \"eval ran against\n * a stub / unconfigured backend.\" Without this guard a canonical eval can\n * silently report `0/N passed` and look like an agent-quality problem when\n * the LLM was never actually called — the failure mode we just hit running\n * the 4-vertical parallel eval (legal-sandbox-stub returned hard-coded 33-104\n * char strings; gtm/creative defaulted to a cli-bridge that wasn't running).\n *\n * The shape:\n *\n *   const report = summarizeBackendIntegrity(records)\n *   assertRealBackend(records)   // throws BackendIntegrityError if 100% stub\n *\n * A record is \"stub-mode\" if its `tokenUsage.input === 0 && tokenUsage.output === 0`.\n * (`costUsd` alone is unreliable — some backends successfully call LLMs but\n *  don't propagate pricing, producing real tokens with $0 cost.)\n *\n * Verdicts:\n *   - `real`   — at least one record has nonzero token usage\n *   - `stub`   — every record is stub-mode (eval ran blind)\n *   - `mixed`  — some records real, some stub (partial backend failure;\n *                often the 429-cascade or auth-half-failed case)\n */\n\nimport { AgentEvalError } from '../errors'\nimport type { RunRecord } from '../run-record'\n\nexport interface BackendIntegrityReport {\n  /** Total records inspected. */\n  totalRecords: number\n  /** Records with input=0 AND output=0 (a stub fingerprint). */\n  stubRecords: number\n  /** Records with nonzero token usage (real LLM activity). */\n  realRecords: number\n  /** Records where output>0 but costUsd=0 (real LLM, broken cost ledger). */\n  uncostedRecords: number\n  /** Sum of input tokens across all records. */\n  totalInputTokens: number\n  /** Sum of output tokens across all records. */\n  totalOutputTokens: number\n  /** Sum of costUsd across all records. */\n  totalCostUsd: number\n  /** Worst-case integrity verdict. */\n  verdict: 'real' | 'mixed' | 'stub'\n  /** Human-readable diagnosis suitable for terminal output. */\n  diagnosis: string\n}\n\n/**\n * Error thrown when an integrity assertion fails. Caller can pattern-match\n * by `code === 'AGENT_EVAL_BACKEND_STUB'` to differentiate from other\n * errors.\n */\nexport class BackendIntegrityError extends AgentEvalError {\n  constructor(\n    message: string,\n    public readonly report: BackendIntegrityReport,\n  ) {\n    super('backend_integrity', message)\n  }\n}\n\nfunction isStubRecord(rec: RunRecord): boolean {\n  return rec.tokenUsage.input === 0 && rec.tokenUsage.output === 0\n}\n\nfunction isUncostedRecord(rec: RunRecord): boolean {\n  return rec.tokenUsage.output > 0 && rec.costUsd === 0\n}\n\n/**\n * Inspect a batch of RunRecords and return an integrity report. Pure\n * function — no I/O, no logging. The caller decides what to do with the\n * verdict (print warning, throw, gate CI, etc.).\n */\nexport function summarizeBackendIntegrity(\n  records: ReadonlyArray<RunRecord>,\n): BackendIntegrityReport {\n  const totalRecords = records.length\n  let stubRecords = 0\n  let realRecords = 0\n  let uncostedRecords = 0\n  let totalInputTokens = 0\n  let totalOutputTokens = 0\n  let totalCostUsd = 0\n  for (const rec of records) {\n    totalInputTokens += rec.tokenUsage.input\n    totalOutputTokens += rec.tokenUsage.output\n    totalCostUsd += rec.costUsd\n    if (isStubRecord(rec)) stubRecords++\n    else realRecords++\n    if (isUncostedRecord(rec)) uncostedRecords++\n  }\n  const verdict: BackendIntegrityReport['verdict'] =\n    totalRecords === 0\n      ? 'stub'\n      : stubRecords === totalRecords\n        ? 'stub'\n        : stubRecords === 0\n          ? 'real'\n          : 'mixed'\n  const diagnosis = buildDiagnosis({\n    totalRecords,\n    stubRecords,\n    realRecords,\n    uncostedRecords,\n    totalInputTokens,\n    totalOutputTokens,\n    totalCostUsd,\n    verdict,\n  })\n  return {\n    totalRecords,\n    stubRecords,\n    realRecords,\n    uncostedRecords,\n    totalInputTokens,\n    totalOutputTokens,\n    totalCostUsd,\n    verdict,\n    diagnosis,\n  }\n}\n\nfunction buildDiagnosis(r: Omit<BackendIntegrityReport, 'diagnosis'>): string {\n  if (r.totalRecords === 0) {\n    return 'no records — eval produced zero runs; backend likely failed before first turn'\n  }\n  if (r.verdict === 'stub') {\n    return [\n      `all ${r.totalRecords} records have zero token usage — the LLM backend was never called.`,\n      'common causes: --backend sandbox without a sandbox bridge running; stub model returning hard-coded strings;',\n      'auth misconfigured so requests were silently dropped before the LLM. Re-run with --backend tcloud and TANGLE_API_KEY set,',\n      'or boot the cli-bridge / sandbox before invoking the eval.',\n    ].join(' ')\n  }\n  if (r.verdict === 'mixed') {\n    const pct = ((r.stubRecords / r.totalRecords) * 100).toFixed(0)\n    return [\n      `${r.stubRecords}/${r.totalRecords} records (${pct}%) have zero token usage — the backend partially failed.`,\n      'common causes: rate-limit cascade (429s after the first N personas);',\n      'transient auth expiry mid-run; provider outage. Treat the affected records as missing data, not agent failures.',\n    ].join(' ')\n  }\n  // verdict === 'real'\n  if (r.uncostedRecords > 0) {\n    const pct = ((r.uncostedRecords / r.totalRecords) * 100).toFixed(0)\n    return [\n      `${r.totalRecords} records with real LLM activity (in=${r.totalInputTokens}, out=${r.totalOutputTokens} tokens).`,\n      `${r.uncostedRecords} (${pct}%) have output tokens but costUsd=0 — cost ledger is mis-wired (no input-token`,\n      'propagation from the runtime stream into RunRecord).',\n    ].join(' ')\n  }\n  return `${r.totalRecords} records with real LLM activity (in=${r.totalInputTokens}, out=${r.totalOutputTokens} tokens, $${r.totalCostUsd.toFixed(4)}).`\n}\n\n/**\n * Throw BackendIntegrityError if the verdict is 'stub' — i.e. every record\n * shows zero LLM activity. Non-strict callers can pass `{ allowMixed: false }`\n * to also reject mixed verdicts (recommended for CI gates).\n *\n * Real backends pass through silently.\n */\nexport function assertRealBackend(\n  records: ReadonlyArray<RunRecord>,\n  opts: { allowMixed?: boolean } = {},\n): BackendIntegrityReport {\n  const report = summarizeBackendIntegrity(records)\n  const allowMixed = opts.allowMixed ?? true\n  if (report.verdict === 'stub') {\n    throw new BackendIntegrityError(\n      `backend-integrity: ran against a stub or unconfigured backend — ${report.diagnosis}`,\n      report,\n    )\n  }\n  if (!allowMixed && report.verdict === 'mixed') {\n    throw new BackendIntegrityError(\n      `backend-integrity: partial backend failure rejected — ${report.diagnosis}`,\n      report,\n    )\n  }\n  return report\n}\n","import { createRequire } from 'node:module'\n\n/**\n * @experimental\n *\n * `CampaignStorage` — the filesystem seam `runCampaign` writes through\n * (run/cell dirs, the resumability cache, per-cell artifacts, trace spans).\n *\n * The default (`fsCampaignStorage`) is the Node filesystem — identical\n * behavior to the inline `node:fs` calls it replaces, so existing CLI\n * consumers are unaffected. `inMemoryCampaignStorage` keeps everything in a\n * `Map`, so the substrate runs in environments WITHOUT a filesystem\n * (Cloudflare Workers, Deno Deploy, other edge runtimes) — the campaign\n * still produces its `CampaignResult` (cells + aggregates) in memory;\n * artifacts/traces simply aren't persisted to disk.\n *\n * Paths are opaque keys to the in-memory adapter — it does not parse them,\n * so the same `join(...)`-built paths work unchanged across both adapters.\n */\nexport interface CampaignStorage {\n  /** Ensure a directory exists (recursive). No-op for in-memory. */\n  ensureDir(dir: string): void\n  /** Does this path exist (as a written file or an ensured dir)? */\n  exists(path: string): boolean\n  /** Read a UTF-8 file; `undefined` when missing or unreadable. */\n  read(path: string): string | undefined\n  /** Write a file (string or bytes). Parent dir is assumed ensured. */\n  write(path: string, content: string | Uint8Array): void\n}\n\n/** Node-filesystem storage — the default. Lazily requires `node:fs` so the\n *  module imports cleanly in non-Node runtimes (where the caller passes\n *  `inMemoryCampaignStorage` instead and never constructs this).\n *\n *  `createRequire(import.meta.url)` is the ESM-native lazy require — a bare\n *  `require` is a ReferenceError under `\"type\": \"module\"`, which is exactly\n *  the shape this package publishes. */\nexport function fsCampaignStorage(): CampaignStorage {\n  const nodeRequire = createRequire(import.meta.url)\n  const { existsSync, mkdirSync, readFileSync, writeFileSync } = nodeRequire(\n    'node:fs',\n  ) as typeof import('node:fs')\n  return {\n    ensureDir(dir) {\n      if (!existsSync(dir)) mkdirSync(dir, { recursive: true })\n    },\n    exists(path) {\n      return existsSync(path)\n    },\n    read(path) {\n      try {\n        return readFileSync(path, 'utf8')\n      } catch {\n        return undefined\n      }\n    },\n    write(path, content) {\n      writeFileSync(path, content as Uint8Array)\n    },\n  }\n}\n\n/** In-memory storage for filesystem-less runtimes. Artifacts + trace spans\n *  live in a `Map` for the duration of the run; the `CampaignResult` is\n *  fully populated, but nothing is persisted to disk. */\nexport function inMemoryCampaignStorage(): CampaignStorage {\n  const files = new Map<string, string | Uint8Array>()\n  const dirs = new Set<string>()\n  return {\n    ensureDir(dir) {\n      dirs.add(dir)\n    },\n    exists(path) {\n      return files.has(path) || dirs.has(path)\n    },\n    read(path) {\n      const value = files.get(path)\n      if (value === undefined) return undefined\n      return typeof value === 'string' ? value : new TextDecoder().decode(value)\n    },\n    write(path, content) {\n      files.set(path, content)\n    },\n  }\n}\n"],"mappings":";;;;;;;;AAaA,SAAS,kBAAkB;AAC3B,SAAS,YAAY;;;ACuCd,IAAM,wBAAN,cAAoC,eAAe;AAAA,EACxD,YACE,SACgB,QAChB;AACA,UAAM,qBAAqB,OAAO;AAFlB;AAAA,EAGlB;AAAA,EAHkB;AAIpB;AAEA,SAAS,aAAa,KAAyB;AAC7C,SAAO,IAAI,WAAW,UAAU,KAAK,IAAI,WAAW,WAAW;AACjE;AAEA,SAAS,iBAAiB,KAAyB;AACjD,SAAO,IAAI,WAAW,SAAS,KAAK,IAAI,YAAY;AACtD;AAOO,SAAS,0BACd,SACwB;AACxB,QAAM,eAAe,QAAQ;AAC7B,MAAI,cAAc;AAClB,MAAI,cAAc;AAClB,MAAI,kBAAkB;AACtB,MAAI,mBAAmB;AACvB,MAAI,oBAAoB;AACxB,MAAI,eAAe;AACnB,aAAW,OAAO,SAAS;AACzB,wBAAoB,IAAI,WAAW;AACnC,yBAAqB,IAAI,WAAW;AACpC,oBAAgB,IAAI;AACpB,QAAI,aAAa,GAAG,EAAG;AAAA,QAClB;AACL,QAAI,iBAAiB,GAAG,EAAG;AAAA,EAC7B;AACA,QAAM,UACJ,iBAAiB,IACb,SACA,gBAAgB,eACd,SACA,gBAAgB,IACd,SACA;AACV,QAAM,YAAY,eAAe;AAAA,IAC/B;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,EACF,CAAC;AACD,SAAO;AAAA,IACL;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,EACF;AACF;AAEA,SAAS,eAAe,GAAsD;AAC5E,MAAI,EAAE,iBAAiB,GAAG;AACxB,WAAO;AAAA,EACT;AACA,MAAI,EAAE,YAAY,QAAQ;AACxB,WAAO;AAAA,MACL,OAAO,EAAE,YAAY;AAAA,MACrB;AAAA,MACA;AAAA,MACA;AAAA,IACF,EAAE,KAAK,GAAG;AAAA,EACZ;AACA,MAAI,EAAE,YAAY,SAAS;AACzB,UAAM,OAAQ,EAAE,cAAc,EAAE,eAAgB,KAAK,QAAQ,CAAC;AAC9D,WAAO;AAAA,MACL,GAAG,EAAE,WAAW,IAAI,EAAE,YAAY,aAAa,GAAG;AAAA,MAClD;AAAA,MACA;AAAA,IACF,EAAE,KAAK,GAAG;AAAA,EACZ;AAEA,MAAI,EAAE,kBAAkB,GAAG;AACzB,UAAM,OAAQ,EAAE,kBAAkB,EAAE,eAAgB,KAAK,QAAQ,CAAC;AAClE,WAAO;AAAA,MACL,GAAG,EAAE,YAAY,uCAAuC,EAAE,gBAAgB,SAAS,EAAE,iBAAiB;AAAA,MACtG,GAAG,EAAE,eAAe,KAAK,GAAG;AAAA,MAC5B;AAAA,IACF,EAAE,KAAK,GAAG;AAAA,EACZ;AACA,SAAO,GAAG,EAAE,YAAY,uCAAuC,EAAE,gBAAgB,SAAS,EAAE,iBAAiB,aAAa,EAAE,aAAa,QAAQ,CAAC,CAAC;AACrJ;AASO,SAAS,kBACd,SACA,OAAiC,CAAC,GACV;AACxB,QAAM,SAAS,0BAA0B,OAAO;AAChD,QAAM,aAAa,KAAK,cAAc;AACtC,MAAI,OAAO,YAAY,QAAQ;AAC7B,UAAM,IAAI;AAAA,MACR,wEAAmE,OAAO,SAAS;AAAA,MACnF;AAAA,IACF;AAAA,EACF;AACA,MAAI,CAAC,cAAc,OAAO,YAAY,SAAS;AAC7C,UAAM,IAAI;AAAA,MACR,8DAAyD,OAAO,SAAS;AAAA,MACzE;AAAA,IACF;AAAA,EACF;AACA,SAAO;AACT;;;ACtLA,SAAS,qBAAqB;AAqCvB,SAAS,oBAAqC;AACnD,QAAM,cAAc,cAAc,YAAY,GAAG;AACjD,QAAM,EAAE,YAAY,WAAW,cAAc,cAAc,IAAI;AAAA,IAC7D;AAAA,EACF;AACA,SAAO;AAAA,IACL,UAAU,KAAK;AACb,UAAI,CAAC,WAAW,GAAG,EAAG,WAAU,KAAK,EAAE,WAAW,KAAK,CAAC;AAAA,IAC1D;AAAA,IACA,OAAO,MAAM;AACX,aAAO,WAAW,IAAI;AAAA,IACxB;AAAA,IACA,KAAK,MAAM;AACT,UAAI;AACF,eAAO,aAAa,MAAM,MAAM;AAAA,MAClC,QAAQ;AACN,eAAO;AAAA,MACT;AAAA,IACF;AAAA,IACA,MAAM,MAAM,SAAS;AACnB,oBAAc,MAAM,OAAqB;AAAA,IAC3C;AAAA,EACF;AACF;AAKO,SAAS,0BAA2C;AACzD,QAAM,QAAQ,oBAAI,IAAiC;AACnD,QAAM,OAAO,oBAAI,IAAY;AAC7B,SAAO;AAAA,IACL,UAAU,KAAK;AACb,WAAK,IAAI,GAAG;AAAA,IACd;AAAA,IACA,OAAO,MAAM;AACX,aAAO,MAAM,IAAI,IAAI,KAAK,KAAK,IAAI,IAAI;AAAA,IACzC;AAAA,IACA,KAAK,MAAM;AACT,YAAM,QAAQ,MAAM,IAAI,IAAI;AAC5B,UAAI,UAAU,OAAW,QAAO;AAChC,aAAO,OAAO,UAAU,WAAW,QAAQ,IAAI,YAAY,EAAE,OAAO,KAAK;AAAA,IAC3E;AAAA,IACA,MAAM,MAAM,SAAS;AACnB,YAAM,IAAI,MAAM,OAAO;AAAA,IACzB;AAAA,EACF;AACF;;;AF8BA,eAAsB,YACpB,MAC+C;AAC/C,QAAM,OAAO,KAAK,QAAQ;AAC1B,QAAM,OAAO,KAAK,QAAQ;AAC1B,QAAM,YAAY,KAAK,aAAa;AACpC,QAAM,iBAAiB,KAAK,kBAAkB;AAC9C,QAAM,MAAM,KAAK,QAAQ,MAAM,oBAAI,KAAK;AACxC,QAAM,SAAS,KAAK,UAAU,CAAC;AAC/B,QAAM,UAAU,KAAK,WAAW,kBAAkB;AAElD,UAAQ,UAAU,KAAK,MAAM;AAE7B,QAAM,eAAe,oBAAoB;AAAA,IACvC,WAAW,KAAK;AAAA,IAChB;AAAA,IACA,aAAa,KAAK,SAAS,QAAQ;AAAA,IACnC;AAAA,IACA;AAAA,EACF,CAAC;AAED,QAAM,YAAY,IAAI;AACtB,QAAM,QAAyC,CAAC;AAChD,QAAM,kBAA0C,CAAC;AAGjD,QAAM,WAA0F,CAAC;AACjG,MAAI,YAAY;AAChB,aAAW,YAAY,KAAK,WAAW;AACrC,aAAS,MAAM,GAAG,MAAM,MAAM,OAAO;AACnC,YAAM,SAAS,GAAG,SAAS,EAAE,IAAI,GAAG;AACpC,YAAM,WAAW,OAAO;AACxB,eAAS,KAAK,EAAE,UAAU,KAAK,QAAQ,SAAS,CAAC;AACjD,mBAAa;AAAA,IACf;AAAA,EACF;AAGA,MAAI,eAAe;AACnB,MAAI,qBAAqB;AACzB,QAAM,kBAAkB,IAAI,gBAAgB;AAI5C,QAAM,QAAyB,CAAC;AAChC,MAAI,UAAU;AACd,QAAM,WAAW;AAEjB,WAAS,IAAI,GAAG,IAAI,gBAAgB,KAAK;AACvC,UAAM;AAAA,OACH,YAAY;AACX,eAAO,MAAM;AACX,gBAAM,QAAQ;AACd,cAAI,SAAS,SAAS,OAAQ;AAC9B,gBAAM,OAAO,SAAS,KAAK;AAC3B,cAAI,oBAAoB;AACtB,qBAAS,KAAK,YAAY,MAAM,sBAAsB,CAAC;AACvD;AAAA,UACF;AACA,gBAAM,SAAS,MAAM,YAAY;AAAA,YAC/B;AAAA,YACA;AAAA,YACA;AAAA,YACA;AAAA,YACA;AAAA,YACA;AAAA,YACA,kBAAkB,KAAK,oBAAoB,wBAAwB,OAAO;AAAA,YAC1E,QAAQ,gBAAgB;AAAA,YACxB,mBAAmB,KAAK;AAAA,UAC1B,CAAC;AACD,mBAAS,KAAK,OAAO,IAAI;AACzB,2BAAiB,OAAO,MAAM,KAAK,eAAe,MAAM;AACxD,0BAAgB,OAAO,KAAK;AAC5B,iBAAO,OAAO,iBAAiB,OAAO,eAAe;AACrD,cAAI,KAAK,gBAAgB,UAAa,gBAAgB,KAAK,aAAa;AACtE,iCAAqB;AAAA,UACvB;AAEA,cAAI,KAAK,gBAAgB,KAAK,iBAAiB,SAAS,CAAC,OAAO,KAAK,OAAO;AAC1E,kBAAM,eAAe;AAAA,cACnB,OAAO,KAAK;AAAA,cACZ,MAAM,OAAO;AAAA,cACb,UAAU,KAAK;AAAA,cACf;AAAA,cACA;AAAA,YACF,CAAC,EAAE,MAAM,CAAC,QAAQ;AAGhB,sBAAQ;AAAA,gBACN,oCAAoC,OAAO,KAAK,MAAM,KAAK,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG,CAAC;AAAA,cAC7G;AAAA,YACF,CAAC;AAAA,UACH;AAAA,QACF;AAAA,MACF,GAAG;AAAA,IACL;AAAA,EACF;AACA,QAAM,QAAQ,IAAI,KAAK;AAEvB,QAAM,UAAU,IAAI;AACpB,WAAS,KAAK,CAAC,GAAG,MAAM,EAAE,OAAO,cAAc,EAAE,MAAM,CAAC;AAExD,QAAM,aAAa;AAAA,IACjB;AAAA,IACA;AAAA,IACA;AAAA,EACF;AAEA,SAAO;AAAA,IACL;AAAA,IACA;AAAA,IACA,WAAW,UAAU,YAAY;AAAA,IACjC,SAAS,QAAQ,YAAY;AAAA,IAC7B,YAAY,QAAQ,QAAQ,IAAI,UAAU,QAAQ;AAAA,IAClD,OAAO;AAAA,IACP;AAAA,IACA,QAAQ,KAAK;AAAA,IACb;AAAA,IACA,WAAW,KAAK,UAAU,IAAI,CAAC,OAAO,EAAE,IAAI,EAAE,IAAI,MAAM,EAAE,KAAK,EAAE;AAAA,EACnE;AACF;AAgBA,eAAe,YACb,MAC2F;AAC3F,QAAM,UAAU,KAAK;AACrB,QAAM,UAAU,KAAK,KAAK,KAAK,QAAQ,KAAK,KAAK,OAAO,QAAQ,mBAAmB,GAAG,CAAC;AACvF,UAAQ,UAAU,OAAO;AAGzB,QAAM,YAAY,KAAK,SAAS,oBAAoB;AACpD,MAAI,KAAK,WAAW;AAClB,UAAM,MAAM,QAAQ,KAAK,SAAS;AAClC,QAAI,QAAQ,QAAW;AACrB,UAAI;AACF,cAAM,SAAS,KAAK,MAAM,GAAG;AAC7B,YAAI,OAAO,WAAW,KAAK,KAAK,QAAQ;AACtC,iBAAO,EAAE,MAAM,EAAE,GAAG,QAAQ,QAAQ,KAAK,GAAG,iBAAiB,CAAC,EAAE;AAAA,QAClE;AAAA,MACF,QAAQ;AAAA,MAER;AAAA,IACF;AAAA,EACF;AAEA,QAAM,UAAU,KAAK,IAAI;AACzB,QAAM,QAAQ,KAAK,iBAAiB,KAAK,KAAK,QAAQ,OAAO;AAC7D,QAAM,kBAA0C,CAAC;AACjD,QAAM,YAAoC;AAAA,IACxC,MAAM,MAAM,MAAM,SAAS;AACzB,YAAM,WAAW,KAAK,SAAS,IAAI;AACnC,cAAQ,UAAU,KAAK,UAAU,IAAI,CAAC;AACtC,cAAQ,MAAM,UAAU,OAAO;AAC/B,sBAAgB,GAAG,KAAK,KAAK,MAAM,IAAI,IAAI,EAAE,IAAI;AACjD,aAAO;AAAA,IACT;AAAA,IACA,MAAM,UAAU,MAAM,OAAO;AAC3B,aAAO,UAAU,MAAM,MAAM,KAAK,UAAU,OAAO,MAAM,CAAC,CAAC;AAAA,IAC7D;AAAA,EACF;AACA,MAAI,YAAY;AAChB,QAAM,cAAkC,EAAE,OAAO,GAAG,QAAQ,EAAE;AAC9D,QAAM,OAA0B;AAAA,IAC9B,QAAQ,QAAQ,QAAQ;AACtB,mBAAa;AACb,YAAM,KAAK,QAAQ,MAAM,IAAI,EAAE,WAAW,OAAO,CAAC,EAAE,IAAI;AAAA,IAC1D;AAAA,IACA,cAAc,OAAO;AACnB,kBAAY,SAAS,MAAM;AAC3B,kBAAY,UAAU,MAAM;AAC5B,UAAI,MAAM,OAAQ,aAAY,UAAU,YAAY,UAAU,KAAK,MAAM;AAAA,IAC3E;AAAA,IACA,UAAU;AACR,aAAO;AAAA,IACT;AAAA,IACA,SAAS;AACP,aAAO,EAAE,GAAG,YAAY;AAAA,IAC1B;AAAA,EACF;AAEA,QAAM,YAAY,KAAK,KAAK,gBAAgB;AAAA,IAC1C,UAAU,KAAK,KAAK;AAAA,IACpB,KAAK,KAAK,KAAK;AAAA,EACjB,CAAC;AAMD,QAAM,YAAY,IAAI,gBAAgB;AACtC,QAAM,kBAAkB,MAAM,UAAU,MAAO,KAAK,OAAgC,MAAM;AAC1F,MAAI,KAAK,OAAO,QAAS,WAAU,MAAO,KAAK,OAAgC,MAAM;AAAA,MAChF,MAAK,OAAO,iBAAiB,SAAS,iBAAiB,EAAE,MAAM,KAAK,CAAC;AAE1E,QAAM,MAAuB;AAAA,IAC3B,QAAQ,KAAK,KAAK;AAAA,IAClB,KAAK,KAAK,KAAK;AAAA,IACf,MAAM,KAAK,KAAK;AAAA,IAChB,QAAQ,UAAU;AAAA,IAClB;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,EACF;AAEA,MAAI;AACJ,MAAI;AACJ,QAAM,YAAY,KAAK;AACvB,MAAI;AACJ,MAAI;AACF,UAAM,aAAa,KAAK,KAAK,SAAS,KAAK,KAAK,UAAU,GAAG;AAC7D,QAAI,cAAc,UAAa,YAAY,GAAG;AAK5C,iBAAW,MAAM,QAAQ,KAAK;AAAA,QAC5B;AAAA,QACA,IAAI,QAAe,CAAC,GAAG,WAAW;AAChC,yBAAe,WAAW,MAAM;AAC9B,sBAAU,MAAM,IAAI,MAAM,kBAAkB,CAAC;AAC7C;AAAA,cACE,IAAI;AAAA,gBACF,qBAAqB,SAAS,gBAAgB,KAAK,KAAK,MAAM;AAAA,cAChE;AAAA,YACF;AAAA,UACF,GAAG,SAAS;AACZ,cAAI,OAAQ,aAAwC,UAAU;AAC5D,YAAC,aAAuC,MAAM;AAAA,QAClD,CAAC;AAAA,MACH,CAAC;AAAA,IACH,OAAO;AACL,iBAAW,MAAM;AAAA,IACnB;AAAA,EACF,SAAS,KAAK;AACZ,mBAAe,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;AAAA,EAChE,UAAE;AACA,QAAI,aAAc,cAAa,YAAY;AAC3C,SAAK,OAAO,oBAAoB,SAAS,eAAe;AAAA,EAC1D;AAKA,QAAM,cAA0C,CAAC;AACjD,MAAI,aAAa,QAAW;AAC1B,eAAW,SAAS,KAAK,KAAK,UAAU,CAAC,GAAG;AAC1C,UAAI,MAAM,aAAa,CAAC,MAAM,UAAU,KAAK,KAAK,QAAQ,EAAG;AAC7D,UAAI;AACF,oBAAY,MAAM,IAAI,IAAI,MAAM,aAAa,OAAO;AAAA,UAClD;AAAA,UACA,UAAU,KAAK,KAAK;AAAA,UACpB,QAAQ,KAAK;AAAA,QACf,CAAC;AAAA,MACH,SAAS,KAAK;AACZ,uBAAe,UAAU,MAAM,IAAI,aAAa,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG,CAAC;AAChG;AAAA,MACF;AAAA,IACF;AAAA,EACF;AAEA,QAAM,MAAM,MAAM;AAElB,QAAM,OAAsC;AAAA,IAC1C,QAAQ,KAAK,KAAK;AAAA,IAClB,YAAY,KAAK,KAAK,SAAS;AAAA,IAC/B,KAAK,KAAK,KAAK;AAAA,IACf,UAAW,YAAY;AAAA,IACvB;AAAA,IACA,SAAS;AAAA,IACT,YAAY,EAAE,GAAG,YAAY;AAAA,IAC7B,YAAY,KAAK,IAAI,IAAI;AAAA,IACzB,MAAM,KAAK,KAAK;AAAA,IAChB,QAAQ;AAAA,IACR,OAAO;AAAA,EACT;AAEA,MAAI,CAAC,gBAAgB,KAAK,WAAW;AACnC,YAAQ,MAAM,WAAW,KAAK,UAAU,IAAI,CAAC;AAAA,EAC/C;AAEA,SAAO,EAAE,MAAM,gBAAgB;AACjC;AASA,SAAS,iBACP,MACA,MACM;AACN,MAAI,SAAS,SAAS,KAAK,MAAO;AAClC,MAAI,KAAK,aAAa,QAAQ,KAAK,aAAa,OAAW;AAC3D,QAAM,aAAa,KAAK,WAAW,UAAU,KAAK,KAAK,WAAW,WAAW;AAC7E,MAAI,KAAK,YAAY,KAAK,CAAC,WAAY;AACvC,QAAM,MAAM,SAAS,KAAK,MAAM;AAChC,MAAI,SAAS,UAAU;AACrB,UAAM,SAAiC;AAAA,MACrC,cAAc;AAAA,MACd,aAAa;AAAA,MACb,aAAa;AAAA,MACb,iBAAiB;AAAA,MACjB,kBAAkB;AAAA,MAClB,mBAAmB;AAAA,MACnB,cAAc;AAAA,MACd,SAAS;AAAA,MACT,WAAW;AAAA,IACb;AACA,UAAM,IAAI,sBAAsB,gBAAgB,GAAG,IAAI,MAAM;AAAA,EAC/D;AAEA,UAAQ,KAAK,8BAA8B,GAAG,EAAE;AAClD;AAEA,eAAe,aACb,OACA,OACqB;AACrB,SAAO,MAAM,MAAM,KAAK;AAC1B;AAEA,SAAS,wBACP,SACsD;AACtD,SAAO,CAAC,QAAQ,QAAQ;AACtB,UAAM,QAAwC,CAAC;AAC/C,WAAO;AAAA,MACL,KAAK,MAAM,YAAY;AACrB,cAAM,UAAU,KAAK,IAAI;AACzB,cAAM,SAAkC,EAAE,MAAM,QAAQ,SAAS,GAAI,cAAc,CAAC,EAAG;AACvF,cAAM,SAAoB;AAAA,UACxB,IAAI,UAAU;AACZ,mBAAO,aAAa,KAAK,IAAI,IAAI;AACjC,gBAAI,SAAU,QAAO,OAAO,QAAQ,QAAQ;AAC5C,kBAAM,KAAK,MAAM;AAAA,UACnB;AAAA,UACA,aAAa,KAAK,OAAO;AACvB,mBAAO,GAAG,IAAI;AAAA,UAChB;AAAA,QACF;AACA,eAAO;AAAA,MACT;AAAA,MACA,MAAM,QAAQ;AACZ,gBAAQ,MAAM,KAAK,KAAK,aAAa,GAAG,MAAM,IAAI,CAAC,MAAM,KAAK,UAAU,CAAC,CAAC,EAAE,KAAK,IAAI,CAAC;AAAA,MACxF;AAAA,IACF;AAAA,EACF;AACF;AAEA,SAAS,YACP,MACA,QAC+B;AAC/B,SAAO;AAAA,IACL,QAAQ,KAAK;AAAA,IACb,YAAY,KAAK,SAAS;AAAA,IAC1B,KAAK,KAAK;AAAA,IACV,UAAU;AAAA,IACV,aAAa,CAAC;AAAA,IACd,SAAS;AAAA,IACT,YAAY,EAAE,OAAO,GAAG,QAAQ,EAAE;AAAA,IAClC,YAAY;AAAA,IACZ,MAAM,KAAK;AAAA,IACX,QAAQ;AAAA,IACR,OAAO,YAAY,MAAM;AAAA,EAC3B;AACF;AAUA,eAAe,eACb,MACe;AACf,QAAM,KAAK,MAAM,QAAQ;AAAA,IACvB,UAAU,KAAK;AAAA,IACf,UAAU,KAAK,KAAK;AAAA,IACpB,aAAa,KAAK,KAAK;AAAA,IACvB,QAAQ,KAAK,KAAK,iBAAiB;AAAA,IACnC,mBAAmB,KAAK,KAAK,4BAA4B;AAAA,IACzD,YAAY,KAAK,IAAI,EAAE,YAAY;AAAA,IACnC,iBAAiB;AAAA,EACnB,CAAC;AACH;AAIA,SAAS,oBAAoB,OAMlB;AACT,QAAM,YAAY;AAAA,IAChB,WAAW,MAAM,UAAU,IAAI,CAAC,OAAO,EAAE,IAAI,EAAE,IAAI,MAAM,EAAE,KAAK,EAAE;AAAA,IAClE,QAAQ,MAAM,OAAO,IAAI,CAAC,OAAO,EAAE,MAAM,EAAE,MAAM,MAAM,EAAE,WAAW,IAAI,CAAC,MAAM,EAAE,GAAG,EAAE,EAAE;AAAA,IACxF,UAAU,MAAM;AAAA,IAChB,MAAM,MAAM;AAAA,IACZ,MAAM,MAAM;AAAA,EACd;AACA,SAAO,WAAW,QAAQ,EAAE,OAAO,KAAK,UAAU,SAAS,CAAC,EAAE,OAAO,KAAK;AAC5E;AAEA,SAAS,kBACP,OACA,QACA,MACoB;AACpB,QAAM,UAA0C,CAAC;AACjD,aAAW,SAAS,QAAQ;AAC1B,UAAM,SAAmB,CAAC;AAC1B,eAAW,QAAQ,OAAO;AACxB,YAAM,IAAI,KAAK,YAAY,MAAM,IAAI;AACrC,UAAI,MAAM,OAAW,QAAO,KAAK,EAAE,SAAS;AAAA,IAC9C;AACA,YAAQ,MAAM,IAAI,IAAI,UAAU,QAAQ,IAAI;AAAA,EAC9C;AACA,QAAM,aAAgD,CAAC;AACvD,QAAM,iBAAiB,oBAAI,IAAsB;AACjD,aAAW,QAAQ,OAAO;AACxB,UAAM,aAAa,OAAO,OAAO,KAAK,WAAW,EAAE,IAAI,CAAC,MAAM,EAAE,SAAS;AACzE,QAAI,WAAW,WAAW,EAAG;AAC7B,UAAM,OAAO,WAAW,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,WAAW;AAChE,UAAM,MAAM,eAAe,IAAI,KAAK,UAAU,KAAK,CAAC;AACpD,QAAI,KAAK,IAAI;AACb,mBAAe,IAAI,KAAK,YAAY,GAAG;AAAA,EACzC;AACA,aAAW,CAAC,YAAY,OAAO,KAAK,gBAAgB;AAClD,UAAM,KAAK,UAAU,SAAS,IAAI;AAClC,eAAW,UAAU,IAAI,EAAE,eAAe,GAAG,MAAM,MAAM,GAAG,MAAM,GAAG,GAAG,EAAE;AAAA,EAC5E;AACA,SAAO;AAAA,IACL;AAAA,IACA;AAAA,IACA,cAAc,MAAM,OAAO,CAAC,GAAG,MAAM,IAAI,EAAE,SAAS,CAAC;AAAA,IACrD,eAAe,MAAM,OAAO,CAAC,MAAM,CAAC,EAAE,KAAK,EAAE;AAAA,IAC7C,cAAc,MAAM,OAAO,CAAC,MAAM,EAAE,OAAO,WAAW,UAAU,CAAC,EAAE;AAAA,IACnE,aAAa,MAAM,OAAO,CAAC,MAAM,EAAE,MAAM,EAAE;AAAA,IAC3C,aAAa,MAAM,OAAO,CAAC,MAAM,EAAE,SAAS,CAAC,EAAE,MAAM,WAAW,UAAU,CAAC,EAAE;AAAA,EAC/E;AACF;AAKA,SAAS,UAAU,SAAmB,MAA8B;AAClE,QAAM,IAAI,QAAQ;AAClB,MAAI,MAAM,EAAG,QAAO,EAAE,MAAM,GAAG,OAAO,GAAG,MAAM,CAAC,GAAG,CAAC,GAAG,GAAG,EAAE;AAC5D,QAAM,OAAO,QAAQ,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI;AAClD,QAAM,WAAW,QAAQ,OAAO,CAAC,GAAG,MAAM,KAAK,IAAI,SAAS,GAAG,CAAC,IAAI,KAAK,IAAI,GAAG,IAAI,CAAC;AACrF,QAAM,QAAQ,KAAK,KAAK,QAAQ;AAChC,QAAM,KAAK,mBAAmB,SAAS,MAAM,EAAE,MAAM,WAAW,IAAK,CAAC;AACtE,SAAO,EAAE,MAAM,OAAO,MAAM,CAAC,GAAG,OAAO,GAAG,KAAK,GAAG,EAAE;AACtD;","names":[]}

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

@@ -13,7 +13,8 @@ async function post(tenant, path, body, opts = {}) {
   const timeoutMs = tenant.timeoutMs ?? 3e4;
   const maxRetries = tenant.retries ?? 2;
   const f = tenant.fetchImpl ?? ((...args) => fetch(...args));
-  const url = `${tenant.endpoint.replace(/\/$/, "")}${path}`;
+  const base = tenant.endpoint.replace(/\/+$/, "").replace(/\/v1$/, "");
+  const url = `${base}${path}`;
   let lastError;
   for (let attempt = 0; attempt <= maxRetries; attempt++) {
     const ourTimeout = AbortSignal.timeout(timeoutMs);
@@ -90,4 +91,4 @@ export {
   createHostedClient,
   hostedClientFromEnv
 };
-//# sourceMappingURL=chunk-HKINEDRZ.js.map
+//# sourceMappingURL=chunk-DFS3FEXO.js.map

package/dist/chunk-DFS3FEXO.js.map

@@ -0,0 +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":[]}