@tangle-network/agent-eval 0.60.0 → 0.62.0

package/dist/campaign/index.js

@@ -20,19 +20,32 @@ import {
   runOptimization,
   surfaceContentHash,
   surfaceHash
-} from "../chunk-LBSXXH56.js";
+} from "../chunk-CV2BS2OV.js";
 import {
   fsCampaignStorage,
   inMemoryCampaignStorage,
   runCampaign
-} from "../chunk-NOPYCRNG.js";
-import "../chunk-GBHRUAOF.js";
+} from "../chunk-7TPYV2ER.js";
+import {
+  agentProfileHash
+} from "../chunk-PQV2TKC3.js";
+import "../chunk-SS2SOBBT.js";
+import {
+  assertRealBackend,
+  summarizeBackendIntegrity
+} from "../chunk-E22YUOAL.js";
 import "../chunk-YV7J7X5N.js";
-import "../chunk-S3SDD56V.js";
+import {
+  validateRunRecord
+} from "../chunk-F3SRAAZO.js";
+import "../chunk-ITBRCT73.js";
 import "../chunk-GGE4NNQT.js";
-import "../chunk-VXNVVBZO.js";
+import "../chunk-VSMTAMNK.js";
+import "../chunk-IHDHUN2X.js";
 import "../chunk-PC4UYEBM.js";
-import "../chunk-QYJT52YW.js";
+import {
+  AgentEvalError
+} from "../chunk-3BFEG2F6.js";
 import "../chunk-PZ5AY32C.js";
 
 // src/campaign/labeled-store/fs-adapter.ts
@@ -246,10 +259,197 @@ function appendLine(path, line) {
   }
 }
 
+// src/campaign/presets/run-profile-matrix.ts
+import { createHash as createHash2 } from "crypto";
+import { join as join2 } from "path";
+var ProfileMatrixError = class extends AgentEvalError {
+  constructor(message) {
+    super("profile_matrix", message);
+  }
+};
+function sanitize(id) {
+  return id.replace(/[^a-zA-Z0-9_-]/g, "_");
+}
+function sha(input) {
+  return createHash2("sha256").update(JSON.stringify(input)).digest("hex");
+}
+function mean(xs) {
+  return xs.length === 0 ? 0 : xs.reduce((a, b) => a + b, 0) / xs.length;
+}
+function cellComposite(cell) {
+  const composites = Object.values(cell.judgeScores).map((s) => s.composite);
+  return composites.length === 0 ? 0 : mean(composites);
+}
+function buildRunRecord(args) {
+  const { cell, profile, profileHash, configHash, experimentId, splitTag, commitSha, matrixId } = args;
+  const composite = cellComposite(cell);
+  const raw = { composite };
+  const perJudge = {};
+  const dimAccum = {};
+  const notes = [];
+  for (const [judgeName, js] of Object.entries(cell.judgeScores)) {
+    perJudge[judgeName] = { ...js.dimensions };
+    for (const [dim, value] of Object.entries(js.dimensions)) {
+      raw[`${judgeName}.${dim}`] = value;
+      (dimAccum[dim] ??= []).push(value);
+    }
+    if (js.notes) notes.push(`${judgeName}: ${js.notes}`);
+  }
+  const perDimMean = {};
+  for (const [dim, values] of Object.entries(dimAccum)) perDimMean[dim] = mean(values);
+  const outcome = splitTag === "holdout" ? { holdoutScore: composite, raw } : { searchScore: composite, raw };
+  if (Object.keys(perJudge).length > 0) {
+    outcome.judgeScores = {
+      perJudge,
+      perDimMean,
+      composite,
+      ...notes.length > 0 ? { notes: notes.join(" | ") } : {}
+    };
+  }
+  return {
+    runId: `${matrixId}:${profile.id}:${cell.cellId}`,
+    experimentId,
+    candidateId: profile.id,
+    seed: cell.seed,
+    model: profile.model,
+    promptHash: profileHash,
+    configHash,
+    commitSha,
+    wallMs: cell.durationMs,
+    costUsd: cell.costUsd,
+    tokenUsage: cell.tokenUsage,
+    outcome,
+    splitTag,
+    scenarioId: cell.scenarioId,
+    ...cell.error ? { failureMode: cell.error } : {}
+  };
+}
+async function runProfileMatrix(opts) {
+  if (opts.profiles.length === 0) throw new ProfileMatrixError("profiles must not be empty");
+  if (opts.scenarios.length === 0) throw new ProfileMatrixError("scenarios must not be empty");
+  const splitTag = opts.splitTag ?? "search";
+  const seed = opts.seed ?? 42;
+  const validate = opts.validate ?? true;
+  const integrityMode = opts.integrity ?? "assert";
+  const profileIds = opts.profiles.map((p) => p.id);
+  const experimentId = opts.experimentId ?? `pm_${sha({ profileIds, scenarios: opts.scenarios.map((s) => s.id) }).slice(0, 16)}`;
+  const matrixId = `mtx_${sha({ experimentId, profileIds, seed, splitTag }).slice(0, 16)}`;
+  for (const profile of opts.profiles) {
+    const profileHash = agentProfileHash(profile);
+    try {
+      validateRunRecord({
+        runId: `${matrixId}:${profile.id}:probe`,
+        experimentId,
+        candidateId: profile.id,
+        seed,
+        model: profile.model,
+        promptHash: profileHash,
+        configHash: profileHash,
+        commitSha: opts.commitSha,
+        wallMs: 0,
+        costUsd: 0,
+        tokenUsage: { input: 0, output: 0 },
+        outcome: splitTag === "holdout" ? { holdoutScore: 0, raw: {} } : { searchScore: 0, raw: {} },
+        splitTag
+      });
+    } catch (err) {
+      throw new ProfileMatrixError(
+        `profile '${profile.id}' is not recordable: ${err instanceof Error ? err.message : String(err)}`
+      );
+    }
+  }
+  const records = [];
+  const campaigns = {};
+  const byProfile = {};
+  for (const profile of opts.profiles) {
+    const profileHash = agentProfileHash(profile);
+    const configHash = sha({
+      profile: profileHash,
+      judges: (opts.judges ?? []).map((j) => j.name),
+      seed,
+      splitTag
+    });
+    const dispatch = (scenario, ctx) => opts.dispatch(profile, scenario, ctx);
+    Object.defineProperty(dispatch, "name", { value: `profile_${sanitize(profile.id)}` });
+    const campaign = await runCampaign({
+      scenarios: opts.scenarios,
+      dispatch,
+      judges: opts.judges,
+      seed,
+      reps: opts.reps,
+      maxConcurrency: opts.maxConcurrency,
+      costCeiling: opts.costCeiling,
+      labeledStore: opts.labeledStore,
+      captureSource: opts.captureSource,
+      storage: opts.storage,
+      now: opts.now,
+      runDir: join2(opts.runDir, sanitize(profile.id))
+    });
+    campaigns[profile.id] = campaign;
+    const profileRecords = [];
+    for (const cell of campaign.cells) {
+      const record = buildRunRecord({
+        cell,
+        profile,
+        profileHash,
+        configHash,
+        experimentId,
+        splitTag,
+        commitSha: opts.commitSha,
+        matrixId
+      });
+      if (validate) validateRunRecord(record);
+      profileRecords.push(record);
+      records.push(record);
+    }
+    byProfile[profile.id] = {
+      profileId: profile.id,
+      profileHash,
+      model: profile.model,
+      records: profileRecords.length,
+      meanComposite: mean(profileRecords.map(compositeOf)),
+      totalCostUsd: profileRecords.reduce((a, r) => a + r.costUsd, 0),
+      integrity: summarizeBackendIntegrity(profileRecords)
+    };
+  }
+  const integrity = summarizeBackendIntegrity(records);
+  if (integrityMode === "assert") {
+    assertRealBackend(records, { allowMixed: opts.allowMixed ?? true });
+  } else if (integrityMode === "warn" && integrity.verdict !== "real") {
+    console.warn(
+      `[runProfileMatrix] backend integrity: ${integrity.verdict} \u2014 ${integrity.diagnosis}`
+    );
+  }
+  const byScenario = rollup(records, (r) => r.scenarioId);
+  const byPersona = opts.personaOf ? rollupByPersona(records, opts.scenarios, opts.personaOf) : void 0;
+  return { matrixId, experimentId, records, byProfile, byScenario, byPersona, integrity, campaigns };
+}
+function compositeOf(r) {
+  return r.outcome.holdoutScore ?? r.outcome.searchScore ?? 0;
+}
+function rollup(records, keyOf) {
+  const groups = /* @__PURE__ */ new Map();
+  for (const r of records) {
+    const key = keyOf(r);
+    if (key === void 0) continue;
+    const arr = groups.get(key) ?? [];
+    arr.push(compositeOf(r));
+    groups.set(key, arr);
+  }
+  const out = {};
+  for (const [key, xs] of groups) out[key] = { meanComposite: mean(xs), n: xs.length };
+  return out;
+}
+function rollupByPersona(records, scenarios, personaOf) {
+  const personaByScenarioId = /* @__PURE__ */ new Map();
+  for (const s of scenarios) personaByScenarioId.set(s.id, personaOf(s));
+  return rollup(records, (r) => r.scenarioId ? personaByScenarioId.get(r.scenarioId) : void 0);
+}
+
 // src/campaign/worktree/index.ts
 import { execFileSync } from "child_process";
 import { existsSync as existsSync2 } from "fs";
-import { basename, isAbsolute, join as join2 } from "path";
+import { basename, isAbsolute, join as join3 } from "path";
 var WorktreeAdapterError = class extends Error {
   constructor(message, cause) {
     super(message);
@@ -271,13 +471,13 @@ function slug(label) {
 }
 function gitWorktreeAdapter(opts) {
   const git = opts.git ?? defaultGit;
-  const worktreeDir = opts.worktreeDir ?? join2(opts.repoRoot, ".worktrees");
+  const worktreeDir = opts.worktreeDir ?? join3(opts.repoRoot, ".worktrees");
   const branchPrefix = opts.branchPrefix ?? "improve";
   return {
     async create({ baseRef, label }) {
       const id = `${slug(label)}-${Date.now().toString(36)}-${Math.random().toString(36).slice(2, 6)}`;
       const branch = `${branchPrefix}/${id}`;
-      const path = join2(worktreeDir, id);
+      const path = join3(worktreeDir, id);
       git(["worktree", "add", "-b", branch, path, baseRef], opts.repoRoot);
       return { path, branch, baseRef };
     },
@@ -302,12 +502,13 @@ function gitWorktreeAdapter(opts) {
 }
 function resolveWorktreePath(surface, worktreeDir) {
   if (isAbsolute(surface.worktreeRef) && existsSync2(surface.worktreeRef)) return surface.worktreeRef;
-  if (worktreeDir) return join2(worktreeDir, basename(surface.worktreeRef));
+  if (worktreeDir) return join3(worktreeDir, basename(surface.worktreeRef));
   return surface.worktreeRef;
 }
 export {
   FsLabeledScenarioStore,
   LabeledScenarioStoreError,
+  ProfileMatrixError,
   WorktreeAdapterError,
   buildLoopProvenanceRecord,
   composeGate,
@@ -333,6 +534,7 @@ export {
   runEval,
   runImprovementLoop,
   runOptimization,
+  runProfileMatrix,
   surfaceContentHash,
   surfaceHash
 };

package/dist/campaign/index.js.map

@@ -1 +1 @@
-{"version":3,"sources":["../../src/campaign/labeled-store/fs-adapter.ts","../../src/campaign/worktree/index.ts"],"sourcesContent":["/**\n * @experimental\n *\n * Filesystem `LabeledScenarioStore` adapter. The default capture sink for\n * traces + eval artifacts. Production deployments typically swap for a\n * Turso/SQLite adapter (same interface).\n *\n * Records land as one JSONL file per source under `<root>/<source>.jsonl`.\n * Each line is a `LabeledScenarioRecord`. Append-only — no in-place edits.\n *\n * Safety properties enforced at write-time:\n *\n *   - **Provenance required**: writes without `source`, `sourceVersionHash`,\n *     `capturedAt`, `redactionStatus` are rejected. Closes the alignment\n *     reviewer's data-poisoning gap.\n *   - **Per-source rate limits**: optional `rateLimitBucket` + `maxWritesPerMinute`\n *     stops a single tenant/source from flooding the store.\n *\n * Safety properties enforced at sample-time:\n *\n *   - **Required split + capturedBefore**: substrate refuses to sample without\n *     an explicit `split` ('train' | 'test') AND a temporal cutoff. Eliminates\n *     accidental train/test contamination.\n *   - **Default training-source filter**: when the store is sampled with\n *     `split: 'train'`, production-trace records are EXCLUDED unless the\n *     caller passes `filter.source: 'production-trace'` explicitly. Closes\n *     the contamination-by-default gap flagged by the senior eval engineer.\n */\n\nimport { createHash } from 'node:crypto'\nimport { existsSync, mkdirSync, readFileSync, writeFileSync } from 'node:fs'\nimport { join } from 'node:path'\nimport type {\n  LabeledScenarioRecord,\n  LabeledScenarioSampleArgs,\n  LabeledScenarioSource,\n  LabeledScenarioStore,\n  LabeledScenarioWrite,\n  LabelTrust,\n} from '../types'\nimport { labelTrustRank } from '../types'\n\nexport interface FsLabeledScenarioStoreOptions {\n  /** Root directory for JSONL files. Created if missing. */\n  root: string\n  /** Per-source rate limit. When set, writes exceeding the cap are rejected\n   *  with a typed error. Default: no limit. */\n  maxWritesPerMinutePerBucket?: number\n  /** Test seam — override `Date.now()` for deterministic tests. */\n  now?: () => number\n}\n\nexport class LabeledScenarioStoreError extends Error {\n  constructor(\n    public readonly code: string,\n    message: string,\n  ) {\n    super(message)\n    this.name = 'LabeledScenarioStoreError'\n  }\n}\n\ninterface RateLimitState {\n  bucket: string\n  windowStartMs: number\n  count: number\n}\n\nexport class FsLabeledScenarioStore implements LabeledScenarioStore {\n  private readonly now: () => number\n  private readonly rateLimits = new Map<string, RateLimitState>()\n\n  constructor(private readonly options: FsLabeledScenarioStoreOptions) {\n    if (!existsSync(options.root)) mkdirSync(options.root, { recursive: true })\n    this.now = options.now ?? Date.now\n  }\n\n  async observe(write: LabeledScenarioWrite): Promise<void> {\n    this.assertProvenance(write)\n    this.assertRateLimit(write)\n    const record = this.toRecord(write)\n    const path = this.pathForSource(write.source)\n    const line = `${JSON.stringify(record)}\\n`\n    // Append atomically. For high-throughput a writev-friendly buffered\n    // implementation lands in the Turso adapter; FS adapter is for tests +\n    // local dev + small workloads.\n    appendLine(path, line)\n  }\n\n  async sample(args: LabeledScenarioSampleArgs): Promise<LabeledScenarioRecord[]> {\n    if (!args.split) {\n      throw new LabeledScenarioStoreError(\n        'split_required',\n        'sample() requires an explicit `split` (train | test) — substrate refuses ambiguous reads',\n      )\n    }\n    if (!args.capturedBefore) {\n      throw new LabeledScenarioStoreError(\n        'capturedBefore_required',\n        'sample() requires an explicit `capturedBefore` timestamp for temporal-split discipline',\n      )\n    }\n\n    const all: LabeledScenarioRecord[] = []\n    for (const source of ALL_SOURCES) {\n      // Default training-source filter: when sampling train, EXCLUDE\n      // production-trace records unless the caller asks for them.\n      if (args.split === 'train' && source === 'production-trace') {\n        const explicit = sourceFilterContains(args.filter?.source, 'production-trace')\n        if (!explicit) continue\n      }\n      const path = this.pathForSource(source)\n      if (!existsSync(path)) continue\n      const lines = readFileSync(path, 'utf8').split('\\n').filter(Boolean)\n      for (const line of lines) {\n        let record: LabeledScenarioRecord\n        try {\n          record = JSON.parse(line) as LabeledScenarioRecord\n        } catch {\n          continue\n        }\n        if (!matchesFilter(record, args, source)) continue\n        all.push(record)\n      }\n    }\n\n    // Deterministic order: by capturedAt ascending, then recordHash.\n    all.sort((a, b) => {\n      if (a.capturedAt !== b.capturedAt) return a.capturedAt.localeCompare(b.capturedAt)\n      return a.recordHash.localeCompare(b.recordHash)\n    })\n\n    return all.slice(0, args.count)\n  }\n\n  async size(): Promise<{\n    train: number\n    test: number\n    bySource: Record<string, number>\n    byTrust: Record<LabelTrust, number>\n  }> {\n    const bySource: Record<string, number> = {}\n    const byTrust: Record<LabelTrust, number> = {\n      unverified: 0,\n      'verified-signal': 0,\n      'human-rated': 0,\n    }\n    let total = 0\n    for (const source of ALL_SOURCES) {\n      const path = this.pathForSource(source)\n      if (!existsSync(path)) {\n        bySource[source] = 0\n        continue\n      }\n      const lines = readFileSync(path, 'utf8').split('\\n').filter(Boolean)\n      bySource[source] = lines.length\n      total += lines.length\n      for (const line of lines) {\n        let trust: LabelTrust = 'unverified'\n        try {\n          trust = (JSON.parse(line) as LabeledScenarioRecord).labelTrust ?? 'unverified'\n        } catch {\n          // A malformed line counts as unverified — never silently gold.\n        }\n        byTrust[trust] += 1\n      }\n    }\n    // FS adapter doesn't track split assignments per-record (split is\n    // computed at sample-time based on `capturedBefore`). For size(), we\n    // report `train`+`test` as the same total — split is a sampling concept.\n    return { train: total, test: total, bySource, byTrust }\n  }\n\n  private assertProvenance(write: LabeledScenarioWrite): void {\n    if (!write.source) {\n      throw new LabeledScenarioStoreError(\n        'missing_source',\n        'LabeledScenarioWrite requires `source`',\n      )\n    }\n    if (!write.sourceVersionHash || write.sourceVersionHash.length === 0) {\n      throw new LabeledScenarioStoreError(\n        'missing_source_version',\n        'LabeledScenarioWrite requires `sourceVersionHash` (git sha or substrate version)',\n      )\n    }\n    if (!write.capturedAt) {\n      throw new LabeledScenarioStoreError(\n        'missing_captured_at',\n        'LabeledScenarioWrite requires `capturedAt` ISO timestamp',\n      )\n    }\n    if (!write.redactionStatus) {\n      throw new LabeledScenarioStoreError(\n        'missing_redaction_status',\n        'LabeledScenarioWrite requires explicit `redactionStatus` — raw / redacted-pii / redacted-secrets / fully-redacted',\n      )\n    }\n    if (!ALL_SOURCES.includes(write.source)) {\n      throw new LabeledScenarioStoreError(\n        'unknown_source',\n        `LabeledScenarioWrite.source must be one of: ${ALL_SOURCES.join(', ')}`,\n      )\n    }\n  }\n\n  private assertRateLimit(write: LabeledScenarioWrite): void {\n    const cap = this.options.maxWritesPerMinutePerBucket\n    if (!cap || !write.rateLimitBucket) return\n    const now = this.now()\n    const windowMs = 60_000\n    let state = this.rateLimits.get(write.rateLimitBucket)\n    if (!state || now - state.windowStartMs >= windowMs) {\n      state = { bucket: write.rateLimitBucket, windowStartMs: now, count: 0 }\n      this.rateLimits.set(write.rateLimitBucket, state)\n    }\n    if (state.count >= cap) {\n      throw new LabeledScenarioStoreError(\n        'rate_limit_exceeded',\n        `LabeledScenarioStore: bucket ${write.rateLimitBucket} exceeded ${cap} writes/min`,\n      )\n    }\n    state.count += 1\n  }\n\n  private toRecord(write: LabeledScenarioWrite): LabeledScenarioRecord {\n    const recordHash = sha256(\n      JSON.stringify({\n        id: write.scenario.id,\n        src: write.source,\n        at: write.capturedAt,\n        ver: write.sourceVersionHash,\n      }),\n    )\n    // FS adapter assigns split at sample-time, but we cache a hint here\n    // based on capturedAt vs the world's \"now\" — sampler overrides this.\n    return {\n      ...write,\n      recordHash,\n      split: 'train',\n    }\n  }\n\n  private pathForSource(source: string): string {\n    return join(this.options.root, `${source}.jsonl`)\n  }\n}\n\nconst ALL_SOURCES: LabeledScenarioWrite['source'][] = [\n  'production-trace',\n  'eval-run',\n  'manual',\n  'red-team',\n  'synthetic',\n]\n\nfunction sourceFilterContains(\n  filter: LabeledScenarioSource | LabeledScenarioSource[] | undefined,\n  needle: LabeledScenarioSource,\n): boolean {\n  if (!filter) return false\n  if (Array.isArray(filter)) return filter.includes(needle)\n  return filter === needle\n}\n\nfunction matchesFilter(\n  record: LabeledScenarioRecord,\n  args: LabeledScenarioSampleArgs,\n  source: string,\n): boolean {\n  // Temporal cutoff — train must be capturedAt < capturedBefore.\n  if (args.split === 'train' && record.capturedAt >= args.capturedBefore) return false\n  if (args.split === 'test' && record.capturedAt < args.capturedBefore) return false\n\n  const f = args.filter\n  if (!f) return true\n  if (f.kind && record.scenario.kind !== f.kind) return false\n  if (f.source) {\n    const sources = Array.isArray(f.source) ? f.source : [f.source]\n    if (!sources.includes(source as never)) return false\n  }\n  if (f.minComposite !== undefined || f.maxComposite !== undefined) {\n    const composites = Object.values(record.judgeScores).map((s) => s.composite)\n    const max = composites.length === 0 ? 0 : Math.max(...composites)\n    if (f.minComposite !== undefined && max < f.minComposite) return false\n    if (f.maxComposite !== undefined && max > f.maxComposite) return false\n  }\n  if (f.minTrust !== undefined && labelTrustRank(record.labelTrust) < labelTrustRank(f.minTrust)) {\n    return false\n  }\n  return true\n}\n\nfunction sha256(input: string): string {\n  return createHash('sha256').update(input).digest('hex').slice(0, 16)\n}\n\nfunction appendLine(path: string, line: string): void {\n  if (existsSync(path)) {\n    const existing = readFileSync(path, 'utf8')\n    writeFileSync(path, existing + line)\n  } else {\n    writeFileSync(path, line)\n  }\n}\n","/**\n * @experimental\n *\n * VCS-pluggable worktree adapter. One improvement = one worktree, PR-like\n * (multiple commits allowed). A code-tier driver's `propose()` creates a\n * worktree, an agent commits the change into it, and `finalize()` returns a\n * `CodeSurface{ worktreeRef }` the measurement checks out to run the worker\n * against the changed code. On promotion the worktree becomes the PR branch.\n *\n * The interface is VCS-agnostic so a future `jj` ([jj-vcs](https://github.com/jj-vcs/jj))\n * adapter can slot in without touching driver code. Only the git adapter\n * ships today. See `docs/design/self-improvement-engine.md`.\n */\n\nimport { execFileSync } from 'node:child_process'\nimport { existsSync } from 'node:fs'\nimport { basename, isAbsolute, join } from 'node:path'\nimport type { CodeSurface } from '../types'\n\nexport interface Worktree {\n  /** Absolute path to the checked-out worktree directory. */\n  path: string\n  /** The branch the worktree is on (becomes the PR branch on promotion). */\n  branch: string\n  /** The ref the worktree was forked from. */\n  baseRef: string\n}\n\nexport interface WorktreeAdapter {\n  /** Create an isolated worktree on a fresh branch off `baseRef`. */\n  create(opts: { baseRef: string; label: string }): Promise<Worktree>\n  /** Commit any pending changes in the worktree, then return a CodeSurface\n   *  pointing at it. The agent has already written its change into\n   *  `worktree.path` by the time this is called. */\n  finalize(worktree: Worktree, summary: string): Promise<CodeSurface>\n  /** Remove the worktree (and its branch) — called for losing candidates. */\n  discard(worktree: Worktree): Promise<void>\n}\n\nexport class WorktreeAdapterError extends Error {\n  constructor(\n    message: string,\n    readonly cause?: unknown,\n  ) {\n    super(message)\n    this.name = 'WorktreeAdapterError'\n  }\n}\n\nexport interface GitWorktreeAdapterOptions {\n  /** Repo root the worktrees fork from. */\n  repoRoot: string\n  /** Directory worktrees are created under. Default: `<repoRoot>/.worktrees`. */\n  worktreeDir?: string\n  /** Branch-name prefix. Default: `improve`. */\n  branchPrefix?: string\n  /** Test seam — defaults to a real `git` runner. */\n  git?: (args: string[], cwd: string) => string\n}\n\nfunction defaultGit(args: string[], cwd: string): string {\n  try {\n    return execFileSync('git', args, { cwd, encoding: 'utf8' }).trim()\n  } catch (err) {\n    const stderr =\n      err && typeof err === 'object' && 'stderr' in err\n        ? String((err as { stderr: unknown }).stderr)\n        : ''\n    throw new WorktreeAdapterError(`git ${args.join(' ')} failed: ${stderr || String(err)}`, err)\n  }\n}\n\n/** Slugify a label into a branch-safe segment. */\nfunction slug(label: string): string {\n  return (\n    label\n      .toLowerCase()\n      .replace(/[^a-z0-9]+/g, '-')\n      .replace(/^-+|-+$/g, '')\n      .slice(0, 48) || 'candidate'\n  )\n}\n\nexport function gitWorktreeAdapter(opts: GitWorktreeAdapterOptions): WorktreeAdapter {\n  const git = opts.git ?? defaultGit\n  const worktreeDir = opts.worktreeDir ?? join(opts.repoRoot, '.worktrees')\n  const branchPrefix = opts.branchPrefix ?? 'improve'\n\n  return {\n    async create({ baseRef, label }) {\n      const id = `${slug(label)}-${Date.now().toString(36)}-${Math.random().toString(36).slice(2, 6)}`\n      const branch = `${branchPrefix}/${id}`\n      const path = join(worktreeDir, id)\n      git(['worktree', 'add', '-b', branch, path, baseRef], opts.repoRoot)\n      return { path, branch, baseRef }\n    },\n\n    async finalize(worktree, summary) {\n      // Stage + commit any pending changes the agent left in the worktree.\n      // A no-op commit is refused by git, so only commit when the tree is dirty.\n      const status = git(['status', '--porcelain'], worktree.path)\n      if (status.length > 0) {\n        git(['add', '-A'], worktree.path)\n        git(['commit', '-m', summary], worktree.path)\n      }\n      return {\n        kind: 'code',\n        worktreeRef: worktree.path,\n        baseRef: worktree.baseRef,\n        summary,\n      }\n    },\n\n    async discard(worktree) {\n      // Remove the worktree, then delete its branch. Force-remove because the\n      // worktree may hold uncommitted experiment state we're discarding.\n      git(['worktree', 'remove', '--force', worktree.path], opts.repoRoot)\n      git(['branch', '-D', worktree.branch], opts.repoRoot)\n    },\n  }\n}\n\n/** Resolve a `CodeSurface`'s worktreeRef to a directory the measurement can\n *  run the worker in. A path ref is returned as-is; anything else is treated\n *  as a ref under the adapter's worktree dir. */\nexport function resolveWorktreePath(surface: CodeSurface, worktreeDir?: string): string {\n  if (isAbsolute(surface.worktreeRef) && existsSync(surface.worktreeRef)) return surface.worktreeRef\n  if (worktreeDir) return join(worktreeDir, basename(surface.worktreeRef))\n  return surface.worktreeRef\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AA6BA,SAAS,kBAAkB;AAC3B,SAAS,YAAY,WAAW,cAAc,qBAAqB;AACnE,SAAS,YAAY;AAqBd,IAAM,4BAAN,cAAwC,MAAM;AAAA,EACnD,YACkB,MAChB,SACA;AACA,UAAM,OAAO;AAHG;AAIhB,SAAK,OAAO;AAAA,EACd;AAAA,EALkB;AAMpB;AAQO,IAAM,yBAAN,MAA6D;AAAA,EAIlE,YAA6B,SAAwC;AAAxC;AAC3B,QAAI,CAAC,WAAW,QAAQ,IAAI,EAAG,WAAU,QAAQ,MAAM,EAAE,WAAW,KAAK,CAAC;AAC1E,SAAK,MAAM,QAAQ,OAAO,KAAK;AAAA,EACjC;AAAA,EAH6B;AAAA,EAHZ;AAAA,EACA,aAAa,oBAAI,IAA4B;AAAA,EAO9D,MAAM,QAAQ,OAA4C;AACxD,SAAK,iBAAiB,KAAK;AAC3B,SAAK,gBAAgB,KAAK;AAC1B,UAAM,SAAS,KAAK,SAAS,KAAK;AAClC,UAAM,OAAO,KAAK,cAAc,MAAM,MAAM;AAC5C,UAAM,OAAO,GAAG,KAAK,UAAU,MAAM,CAAC;AAAA;AAItC,eAAW,MAAM,IAAI;AAAA,EACvB;AAAA,EAEA,MAAM,OAAO,MAAmE;AAC9E,QAAI,CAAC,KAAK,OAAO;AACf,YAAM,IAAI;AAAA,QACR;AAAA,QACA;AAAA,MACF;AAAA,IACF;AACA,QAAI,CAAC,KAAK,gBAAgB;AACxB,YAAM,IAAI;AAAA,QACR;AAAA,QACA;AAAA,MACF;AAAA,IACF;AAEA,UAAM,MAA+B,CAAC;AACtC,eAAW,UAAU,aAAa;AAGhC,UAAI,KAAK,UAAU,WAAW,WAAW,oBAAoB;AAC3D,cAAM,WAAW,qBAAqB,KAAK,QAAQ,QAAQ,kBAAkB;AAC7E,YAAI,CAAC,SAAU;AAAA,MACjB;AACA,YAAM,OAAO,KAAK,cAAc,MAAM;AACtC,UAAI,CAAC,WAAW,IAAI,EAAG;AACvB,YAAM,QAAQ,aAAa,MAAM,MAAM,EAAE,MAAM,IAAI,EAAE,OAAO,OAAO;AACnE,iBAAW,QAAQ,OAAO;AACxB,YAAI;AACJ,YAAI;AACF,mBAAS,KAAK,MAAM,IAAI;AAAA,QAC1B,QAAQ;AACN;AAAA,QACF;AACA,YAAI,CAAC,cAAc,QAAQ,MAAM,MAAM,EAAG;AAC1C,YAAI,KAAK,MAAM;AAAA,MACjB;AAAA,IACF;AAGA,QAAI,KAAK,CAAC,GAAG,MAAM;AACjB,UAAI,EAAE,eAAe,EAAE,WAAY,QAAO,EAAE,WAAW,cAAc,EAAE,UAAU;AACjF,aAAO,EAAE,WAAW,cAAc,EAAE,UAAU;AAAA,IAChD,CAAC;AAED,WAAO,IAAI,MAAM,GAAG,KAAK,KAAK;AAAA,EAChC;AAAA,EAEA,MAAM,OAKH;AACD,UAAM,WAAmC,CAAC;AAC1C,UAAM,UAAsC;AAAA,MAC1C,YAAY;AAAA,MACZ,mBAAmB;AAAA,MACnB,eAAe;AAAA,IACjB;AACA,QAAI,QAAQ;AACZ,eAAW,UAAU,aAAa;AAChC,YAAM,OAAO,KAAK,cAAc,MAAM;AACtC,UAAI,CAAC,WAAW,IAAI,GAAG;AACrB,iBAAS,MAAM,IAAI;AACnB;AAAA,MACF;AACA,YAAM,QAAQ,aAAa,MAAM,MAAM,EAAE,MAAM,IAAI,EAAE,OAAO,OAAO;AACnE,eAAS,MAAM,IAAI,MAAM;AACzB,eAAS,MAAM;AACf,iBAAW,QAAQ,OAAO;AACxB,YAAI,QAAoB;AACxB,YAAI;AACF,kBAAS,KAAK,MAAM,IAAI,EAA4B,cAAc;AAAA,QACpE,QAAQ;AAAA,QAER;AACA,gBAAQ,KAAK,KAAK;AAAA,MACpB;AAAA,IACF;AAIA,WAAO,EAAE,OAAO,OAAO,MAAM,OAAO,UAAU,QAAQ;AAAA,EACxD;AAAA,EAEQ,iBAAiB,OAAmC;AAC1D,QAAI,CAAC,MAAM,QAAQ;AACjB,YAAM,IAAI;AAAA,QACR;AAAA,QACA;AAAA,MACF;AAAA,IACF;AACA,QAAI,CAAC,MAAM,qBAAqB,MAAM,kBAAkB,WAAW,GAAG;AACpE,YAAM,IAAI;AAAA,QACR;AAAA,QACA;AAAA,MACF;AAAA,IACF;AACA,QAAI,CAAC,MAAM,YAAY;AACrB,YAAM,IAAI;AAAA,QACR;AAAA,QACA;AAAA,MACF;AAAA,IACF;AACA,QAAI,CAAC,MAAM,iBAAiB;AAC1B,YAAM,IAAI;AAAA,QACR;AAAA,QACA;AAAA,MACF;AAAA,IACF;AACA,QAAI,CAAC,YAAY,SAAS,MAAM,MAAM,GAAG;AACvC,YAAM,IAAI;AAAA,QACR;AAAA,QACA,+CAA+C,YAAY,KAAK,IAAI,CAAC;AAAA,MACvE;AAAA,IACF;AAAA,EACF;AAAA,EAEQ,gBAAgB,OAAmC;AACzD,UAAM,MAAM,KAAK,QAAQ;AACzB,QAAI,CAAC,OAAO,CAAC,MAAM,gBAAiB;AACpC,UAAM,MAAM,KAAK,IAAI;AACrB,UAAM,WAAW;AACjB,QAAI,QAAQ,KAAK,WAAW,IAAI,MAAM,eAAe;AACrD,QAAI,CAAC,SAAS,MAAM,MAAM,iBAAiB,UAAU;AACnD,cAAQ,EAAE,QAAQ,MAAM,iBAAiB,eAAe,KAAK,OAAO,EAAE;AACtE,WAAK,WAAW,IAAI,MAAM,iBAAiB,KAAK;AAAA,IAClD;AACA,QAAI,MAAM,SAAS,KAAK;AACtB,YAAM,IAAI;AAAA,QACR;AAAA,QACA,gCAAgC,MAAM,eAAe,aAAa,GAAG;AAAA,MACvE;AAAA,IACF;AACA,UAAM,SAAS;AAAA,EACjB;AAAA,EAEQ,SAAS,OAAoD;AACnE,UAAM,aAAa;AAAA,MACjB,KAAK,UAAU;AAAA,QACb,IAAI,MAAM,SAAS;AAAA,QACnB,KAAK,MAAM;AAAA,QACX,IAAI,MAAM;AAAA,QACV,KAAK,MAAM;AAAA,MACb,CAAC;AAAA,IACH;AAGA,WAAO;AAAA,MACL,GAAG;AAAA,MACH;AAAA,MACA,OAAO;AAAA,IACT;AAAA,EACF;AAAA,EAEQ,cAAc,QAAwB;AAC5C,WAAO,KAAK,KAAK,QAAQ,MAAM,GAAG,MAAM,QAAQ;AAAA,EAClD;AACF;AAEA,IAAM,cAAgD;AAAA,EACpD;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AACF;AAEA,SAAS,qBACP,QACA,QACS;AACT,MAAI,CAAC,OAAQ,QAAO;AACpB,MAAI,MAAM,QAAQ,MAAM,EAAG,QAAO,OAAO,SAAS,MAAM;AACxD,SAAO,WAAW;AACpB;AAEA,SAAS,cACP,QACA,MACA,QACS;AAET,MAAI,KAAK,UAAU,WAAW,OAAO,cAAc,KAAK,eAAgB,QAAO;AAC/E,MAAI,KAAK,UAAU,UAAU,OAAO,aAAa,KAAK,eAAgB,QAAO;AAE7E,QAAM,IAAI,KAAK;AACf,MAAI,CAAC,EAAG,QAAO;AACf,MAAI,EAAE,QAAQ,OAAO,SAAS,SAAS,EAAE,KAAM,QAAO;AACtD,MAAI,EAAE,QAAQ;AACZ,UAAM,UAAU,MAAM,QAAQ,EAAE,MAAM,IAAI,EAAE,SAAS,CAAC,EAAE,MAAM;AAC9D,QAAI,CAAC,QAAQ,SAAS,MAAe,EAAG,QAAO;AAAA,EACjD;AACA,MAAI,EAAE,iBAAiB,UAAa,EAAE,iBAAiB,QAAW;AAChE,UAAM,aAAa,OAAO,OAAO,OAAO,WAAW,EAAE,IAAI,CAAC,MAAM,EAAE,SAAS;AAC3E,UAAM,MAAM,WAAW,WAAW,IAAI,IAAI,KAAK,IAAI,GAAG,UAAU;AAChE,QAAI,EAAE,iBAAiB,UAAa,MAAM,EAAE,aAAc,QAAO;AACjE,QAAI,EAAE,iBAAiB,UAAa,MAAM,EAAE,aAAc,QAAO;AAAA,EACnE;AACA,MAAI,EAAE,aAAa,UAAa,eAAe,OAAO,UAAU,IAAI,eAAe,EAAE,QAAQ,GAAG;AAC9F,WAAO;AAAA,EACT;AACA,SAAO;AACT;AAEA,SAAS,OAAO,OAAuB;AACrC,SAAO,WAAW,QAAQ,EAAE,OAAO,KAAK,EAAE,OAAO,KAAK,EAAE,MAAM,GAAG,EAAE;AACrE;AAEA,SAAS,WAAW,MAAc,MAAoB;AACpD,MAAI,WAAW,IAAI,GAAG;AACpB,UAAM,WAAW,aAAa,MAAM,MAAM;AAC1C,kBAAc,MAAM,WAAW,IAAI;AAAA,EACrC,OAAO;AACL,kBAAc,MAAM,IAAI;AAAA,EAC1B;AACF;;;AClSA,SAAS,oBAAoB;AAC7B,SAAS,cAAAA,mBAAkB;AAC3B,SAAS,UAAU,YAAY,QAAAC,aAAY;AAuBpC,IAAM,uBAAN,cAAmC,MAAM;AAAA,EAC9C,YACE,SACS,OACT;AACA,UAAM,OAAO;AAFJ;AAGT,SAAK,OAAO;AAAA,EACd;AAAA,EAJW;AAKb;AAaA,SAAS,WAAW,MAAgB,KAAqB;AACvD,MAAI;AACF,WAAO,aAAa,OAAO,MAAM,EAAE,KAAK,UAAU,OAAO,CAAC,EAAE,KAAK;AAAA,EACnE,SAAS,KAAK;AACZ,UAAM,SACJ,OAAO,OAAO,QAAQ,YAAY,YAAY,MAC1C,OAAQ,IAA4B,MAAM,IAC1C;AACN,UAAM,IAAI,qBAAqB,OAAO,KAAK,KAAK,GAAG,CAAC,YAAY,UAAU,OAAO,GAAG,CAAC,IAAI,GAAG;AAAA,EAC9F;AACF;AAGA,SAAS,KAAK,OAAuB;AACnC,SACE,MACG,YAAY,EACZ,QAAQ,eAAe,GAAG,EAC1B,QAAQ,YAAY,EAAE,EACtB,MAAM,GAAG,EAAE,KAAK;AAEvB;AAEO,SAAS,mBAAmB,MAAkD;AACnF,QAAM,MAAM,KAAK,OAAO;AACxB,QAAM,cAAc,KAAK,eAAeA,MAAK,KAAK,UAAU,YAAY;AACxE,QAAM,eAAe,KAAK,gBAAgB;AAE1C,SAAO;AAAA,IACL,MAAM,OAAO,EAAE,SAAS,MAAM,GAAG;AAC/B,YAAM,KAAK,GAAG,KAAK,KAAK,CAAC,IAAI,KAAK,IAAI,EAAE,SAAS,EAAE,CAAC,IAAI,KAAK,OAAO,EAAE,SAAS,EAAE,EAAE,MAAM,GAAG,CAAC,CAAC;AAC9F,YAAM,SAAS,GAAG,YAAY,IAAI,EAAE;AACpC,YAAM,OAAOA,MAAK,aAAa,EAAE;AACjC,UAAI,CAAC,YAAY,OAAO,MAAM,QAAQ,MAAM,OAAO,GAAG,KAAK,QAAQ;AACnE,aAAO,EAAE,MAAM,QAAQ,QAAQ;AAAA,IACjC;AAAA,IAEA,MAAM,SAAS,UAAU,SAAS;AAGhC,YAAM,SAAS,IAAI,CAAC,UAAU,aAAa,GAAG,SAAS,IAAI;AAC3D,UAAI,OAAO,SAAS,GAAG;AACrB,YAAI,CAAC,OAAO,IAAI,GAAG,SAAS,IAAI;AAChC,YAAI,CAAC,UAAU,MAAM,OAAO,GAAG,SAAS,IAAI;AAAA,MAC9C;AACA,aAAO;AAAA,QACL,MAAM;AAAA,QACN,aAAa,SAAS;AAAA,QACtB,SAAS,SAAS;AAAA,QAClB;AAAA,MACF;AAAA,IACF;AAAA,IAEA,MAAM,QAAQ,UAAU;AAGtB,UAAI,CAAC,YAAY,UAAU,WAAW,SAAS,IAAI,GAAG,KAAK,QAAQ;AACnE,UAAI,CAAC,UAAU,MAAM,SAAS,MAAM,GAAG,KAAK,QAAQ;AAAA,IACtD;AAAA,EACF;AACF;AAKO,SAAS,oBAAoB,SAAsB,aAA8B;AACtF,MAAI,WAAW,QAAQ,WAAW,KAAKD,YAAW,QAAQ,WAAW,EAAG,QAAO,QAAQ;AACvF,MAAI,YAAa,QAAOC,MAAK,aAAa,SAAS,QAAQ,WAAW,CAAC;AACvE,SAAO,QAAQ;AACjB;","names":["existsSync","join"]}
+{"version":3,"sources":["../../src/campaign/labeled-store/fs-adapter.ts","../../src/campaign/presets/run-profile-matrix.ts","../../src/campaign/worktree/index.ts"],"sourcesContent":["/**\n * @experimental\n *\n * Filesystem `LabeledScenarioStore` adapter. The default capture sink for\n * traces + eval artifacts. Production deployments typically swap for a\n * Turso/SQLite adapter (same interface).\n *\n * Records land as one JSONL file per source under `<root>/<source>.jsonl`.\n * Each line is a `LabeledScenarioRecord`. Append-only — no in-place edits.\n *\n * Safety properties enforced at write-time:\n *\n *   - **Provenance required**: writes without `source`, `sourceVersionHash`,\n *     `capturedAt`, `redactionStatus` are rejected. Closes the alignment\n *     reviewer's data-poisoning gap.\n *   - **Per-source rate limits**: optional `rateLimitBucket` + `maxWritesPerMinute`\n *     stops a single tenant/source from flooding the store.\n *\n * Safety properties enforced at sample-time:\n *\n *   - **Required split + capturedBefore**: substrate refuses to sample without\n *     an explicit `split` ('train' | 'test') AND a temporal cutoff. Eliminates\n *     accidental train/test contamination.\n *   - **Default training-source filter**: when the store is sampled with\n *     `split: 'train'`, production-trace records are EXCLUDED unless the\n *     caller passes `filter.source: 'production-trace'` explicitly. Closes\n *     the contamination-by-default gap flagged by the senior eval engineer.\n */\n\nimport { createHash } from 'node:crypto'\nimport { existsSync, mkdirSync, readFileSync, writeFileSync } from 'node:fs'\nimport { join } from 'node:path'\nimport type {\n  LabeledScenarioRecord,\n  LabeledScenarioSampleArgs,\n  LabeledScenarioSource,\n  LabeledScenarioStore,\n  LabeledScenarioWrite,\n  LabelTrust,\n} from '../types'\nimport { labelTrustRank } from '../types'\n\nexport interface FsLabeledScenarioStoreOptions {\n  /** Root directory for JSONL files. Created if missing. */\n  root: string\n  /** Per-source rate limit. When set, writes exceeding the cap are rejected\n   *  with a typed error. Default: no limit. */\n  maxWritesPerMinutePerBucket?: number\n  /** Test seam — override `Date.now()` for deterministic tests. */\n  now?: () => number\n}\n\nexport class LabeledScenarioStoreError extends Error {\n  constructor(\n    public readonly code: string,\n    message: string,\n  ) {\n    super(message)\n    this.name = 'LabeledScenarioStoreError'\n  }\n}\n\ninterface RateLimitState {\n  bucket: string\n  windowStartMs: number\n  count: number\n}\n\nexport class FsLabeledScenarioStore implements LabeledScenarioStore {\n  private readonly now: () => number\n  private readonly rateLimits = new Map<string, RateLimitState>()\n\n  constructor(private readonly options: FsLabeledScenarioStoreOptions) {\n    if (!existsSync(options.root)) mkdirSync(options.root, { recursive: true })\n    this.now = options.now ?? Date.now\n  }\n\n  async observe(write: LabeledScenarioWrite): Promise<void> {\n    this.assertProvenance(write)\n    this.assertRateLimit(write)\n    const record = this.toRecord(write)\n    const path = this.pathForSource(write.source)\n    const line = `${JSON.stringify(record)}\\n`\n    // Append atomically. For high-throughput a writev-friendly buffered\n    // implementation lands in the Turso adapter; FS adapter is for tests +\n    // local dev + small workloads.\n    appendLine(path, line)\n  }\n\n  async sample(args: LabeledScenarioSampleArgs): Promise<LabeledScenarioRecord[]> {\n    if (!args.split) {\n      throw new LabeledScenarioStoreError(\n        'split_required',\n        'sample() requires an explicit `split` (train | test) — substrate refuses ambiguous reads',\n      )\n    }\n    if (!args.capturedBefore) {\n      throw new LabeledScenarioStoreError(\n        'capturedBefore_required',\n        'sample() requires an explicit `capturedBefore` timestamp for temporal-split discipline',\n      )\n    }\n\n    const all: LabeledScenarioRecord[] = []\n    for (const source of ALL_SOURCES) {\n      // Default training-source filter: when sampling train, EXCLUDE\n      // production-trace records unless the caller asks for them.\n      if (args.split === 'train' && source === 'production-trace') {\n        const explicit = sourceFilterContains(args.filter?.source, 'production-trace')\n        if (!explicit) continue\n      }\n      const path = this.pathForSource(source)\n      if (!existsSync(path)) continue\n      const lines = readFileSync(path, 'utf8').split('\\n').filter(Boolean)\n      for (const line of lines) {\n        let record: LabeledScenarioRecord\n        try {\n          record = JSON.parse(line) as LabeledScenarioRecord\n        } catch {\n          continue\n        }\n        if (!matchesFilter(record, args, source)) continue\n        all.push(record)\n      }\n    }\n\n    // Deterministic order: by capturedAt ascending, then recordHash.\n    all.sort((a, b) => {\n      if (a.capturedAt !== b.capturedAt) return a.capturedAt.localeCompare(b.capturedAt)\n      return a.recordHash.localeCompare(b.recordHash)\n    })\n\n    return all.slice(0, args.count)\n  }\n\n  async size(): Promise<{\n    train: number\n    test: number\n    bySource: Record<string, number>\n    byTrust: Record<LabelTrust, number>\n  }> {\n    const bySource: Record<string, number> = {}\n    const byTrust: Record<LabelTrust, number> = {\n      unverified: 0,\n      'verified-signal': 0,\n      'human-rated': 0,\n    }\n    let total = 0\n    for (const source of ALL_SOURCES) {\n      const path = this.pathForSource(source)\n      if (!existsSync(path)) {\n        bySource[source] = 0\n        continue\n      }\n      const lines = readFileSync(path, 'utf8').split('\\n').filter(Boolean)\n      bySource[source] = lines.length\n      total += lines.length\n      for (const line of lines) {\n        let trust: LabelTrust = 'unverified'\n        try {\n          trust = (JSON.parse(line) as LabeledScenarioRecord).labelTrust ?? 'unverified'\n        } catch {\n          // A malformed line counts as unverified — never silently gold.\n        }\n        byTrust[trust] += 1\n      }\n    }\n    // FS adapter doesn't track split assignments per-record (split is\n    // computed at sample-time based on `capturedBefore`). For size(), we\n    // report `train`+`test` as the same total — split is a sampling concept.\n    return { train: total, test: total, bySource, byTrust }\n  }\n\n  private assertProvenance(write: LabeledScenarioWrite): void {\n    if (!write.source) {\n      throw new LabeledScenarioStoreError(\n        'missing_source',\n        'LabeledScenarioWrite requires `source`',\n      )\n    }\n    if (!write.sourceVersionHash || write.sourceVersionHash.length === 0) {\n      throw new LabeledScenarioStoreError(\n        'missing_source_version',\n        'LabeledScenarioWrite requires `sourceVersionHash` (git sha or substrate version)',\n      )\n    }\n    if (!write.capturedAt) {\n      throw new LabeledScenarioStoreError(\n        'missing_captured_at',\n        'LabeledScenarioWrite requires `capturedAt` ISO timestamp',\n      )\n    }\n    if (!write.redactionStatus) {\n      throw new LabeledScenarioStoreError(\n        'missing_redaction_status',\n        'LabeledScenarioWrite requires explicit `redactionStatus` — raw / redacted-pii / redacted-secrets / fully-redacted',\n      )\n    }\n    if (!ALL_SOURCES.includes(write.source)) {\n      throw new LabeledScenarioStoreError(\n        'unknown_source',\n        `LabeledScenarioWrite.source must be one of: ${ALL_SOURCES.join(', ')}`,\n      )\n    }\n  }\n\n  private assertRateLimit(write: LabeledScenarioWrite): void {\n    const cap = this.options.maxWritesPerMinutePerBucket\n    if (!cap || !write.rateLimitBucket) return\n    const now = this.now()\n    const windowMs = 60_000\n    let state = this.rateLimits.get(write.rateLimitBucket)\n    if (!state || now - state.windowStartMs >= windowMs) {\n      state = { bucket: write.rateLimitBucket, windowStartMs: now, count: 0 }\n      this.rateLimits.set(write.rateLimitBucket, state)\n    }\n    if (state.count >= cap) {\n      throw new LabeledScenarioStoreError(\n        'rate_limit_exceeded',\n        `LabeledScenarioStore: bucket ${write.rateLimitBucket} exceeded ${cap} writes/min`,\n      )\n    }\n    state.count += 1\n  }\n\n  private toRecord(write: LabeledScenarioWrite): LabeledScenarioRecord {\n    const recordHash = sha256(\n      JSON.stringify({\n        id: write.scenario.id,\n        src: write.source,\n        at: write.capturedAt,\n        ver: write.sourceVersionHash,\n      }),\n    )\n    // FS adapter assigns split at sample-time, but we cache a hint here\n    // based on capturedAt vs the world's \"now\" — sampler overrides this.\n    return {\n      ...write,\n      recordHash,\n      split: 'train',\n    }\n  }\n\n  private pathForSource(source: string): string {\n    return join(this.options.root, `${source}.jsonl`)\n  }\n}\n\nconst ALL_SOURCES: LabeledScenarioWrite['source'][] = [\n  'production-trace',\n  'eval-run',\n  'manual',\n  'red-team',\n  'synthetic',\n]\n\nfunction sourceFilterContains(\n  filter: LabeledScenarioSource | LabeledScenarioSource[] | undefined,\n  needle: LabeledScenarioSource,\n): boolean {\n  if (!filter) return false\n  if (Array.isArray(filter)) return filter.includes(needle)\n  return filter === needle\n}\n\nfunction matchesFilter(\n  record: LabeledScenarioRecord,\n  args: LabeledScenarioSampleArgs,\n  source: string,\n): boolean {\n  // Temporal cutoff — train must be capturedAt < capturedBefore.\n  if (args.split === 'train' && record.capturedAt >= args.capturedBefore) return false\n  if (args.split === 'test' && record.capturedAt < args.capturedBefore) return false\n\n  const f = args.filter\n  if (!f) return true\n  if (f.kind && record.scenario.kind !== f.kind) return false\n  if (f.source) {\n    const sources = Array.isArray(f.source) ? f.source : [f.source]\n    if (!sources.includes(source as never)) return false\n  }\n  if (f.minComposite !== undefined || f.maxComposite !== undefined) {\n    const composites = Object.values(record.judgeScores).map((s) => s.composite)\n    const max = composites.length === 0 ? 0 : Math.max(...composites)\n    if (f.minComposite !== undefined && max < f.minComposite) return false\n    if (f.maxComposite !== undefined && max > f.maxComposite) return false\n  }\n  if (f.minTrust !== undefined && labelTrustRank(record.labelTrust) < labelTrustRank(f.minTrust)) {\n    return false\n  }\n  return true\n}\n\nfunction sha256(input: string): string {\n  return createHash('sha256').update(input).digest('hex').slice(0, 16)\n}\n\nfunction appendLine(path: string, line: string): void {\n  if (existsSync(path)) {\n    const existing = readFileSync(path, 'utf8')\n    writeFileSync(path, existing + line)\n  } else {\n    writeFileSync(path, line)\n  }\n}\n","/**\n * @experimental\n *\n * `runProfileMatrix` — the missing keystone between `runAgentMatrix` and the\n * backend-integrity guard.\n *\n * The gap it closes: `runAgentMatrix` is a topology-opaque scheduler whose\n * cells return a bare `{ output, verdict, costUsd }` — no `tokenUsage`, not a\n * `RunRecord`. `assertRealBackend` / `summarizeBackendIntegrity` key on\n * `RunRecord.tokenUsage`, so they cannot run on a raw matrix result. Every\n * consumer therefore hand-writes the same bridge: fan a profile × scenario\n * cartesian, call dispatch, fabricate a `RunRecord` with token usage, thread it\n * back, run the integrity guard. That hand-rolled bridge is exactly the pile of\n * bespoke `eval:*` scripts the adoption skills keep trying (and failing) to\n * forbid.\n *\n * `runProfileMatrix` IS that bridge, once:\n *\n *   - axis 3 (PROFILE) = `profiles: AgentProfile[]`\n *   - axis 1 (PERSONA/SCENARIO) = `scenarios: Scenario[]` (each scenario carries\n *     its persona; `personaOf` groups them for the `byPersona` pivot)\n *   - the scoring axis = `judges`\n *\n * It runs `runCampaign` once per profile (reusing its seeds, reps, bootstrap\n * CIs, resumability, and the `LabeledScenarioStore` capture flywheel), maps\n * every cell to a validated `RunRecord` carrying the real `tokenUsage` the\n * dispatch reported via `ctx.cost.observeTokens`, and runs `assertRealBackend`\n * BY CONSTRUCTION before returning — so a stub-backend run fails loudly instead\n * of reporting a clean 0/N leaderboard.\n *\n * Dispatch contract: a dispatch that calls an LLM MUST report usage via\n * `ctx.cost.observeTokens({ input, output })` (and cost via `ctx.cost.observe`).\n * A dispatch that reports zero tokens is indistinguishable from a stub and the\n * integrity guard treats it as one.\n */\n\nimport { createHash } from 'node:crypto'\nimport { join } from 'node:path'\nimport { type AgentProfile, agentProfileHash } from '../../agent-profile'\nimport { AgentEvalError } from '../../errors'\nimport {\n  assertRealBackend,\n  type BackendIntegrityReport,\n  summarizeBackendIntegrity,\n} from '../../integrity/backend-integrity'\nimport {\n  type RunOutcome,\n  type RunRecord,\n  type RunSplitTag,\n  validateRunRecord,\n} from '../../run-record'\nimport { runCampaign } from '../run-campaign'\nimport type { CampaignStorage } from '../storage'\nimport type {\n  CampaignCellResult,\n  CampaignResult,\n  DispatchContext,\n  JudgeConfig,\n  LabeledScenarioSource,\n  LabeledScenarioStore,\n  Scenario,\n} from '../types'\n\n/** Thrown when the matrix is misconfigured (no profiles, a profile whose model\n *  lacks a snapshot version, etc.). Distinct from `BackendIntegrityError`,\n *  which signals a stub backend at run time. */\nexport class ProfileMatrixError extends AgentEvalError {\n  constructor(message: string) {\n    super('profile_matrix', message)\n  }\n}\n\n/** Dispatch for one cell: render `profile` against `scenario`, returning the\n *  artifact the judges score. Report LLM usage via `ctx.cost.observeTokens`\n *  and `ctx.cost.observe` — the integrity guard depends on it. */\nexport type ProfileDispatchFn<TScenario extends Scenario, TArtifact> = (\n  profile: AgentProfile,\n  scenario: TScenario,\n  ctx: DispatchContext,\n) => Promise<TArtifact>\n\nexport interface RunProfileMatrixOptions<TScenario extends Scenario, TArtifact> {\n  /** Axis 3 — the agent-under-test configurations. Each is one column. */\n  profiles: AgentProfile[]\n  /** Axis 1 — the persona/scenario corpus, run against every profile. */\n  scenarios: TScenario[]\n  /** Renders one (profile, scenario) cell. */\n  dispatch: ProfileDispatchFn<TScenario, TArtifact>\n  /** The scoring axis. */\n  judges?: JudgeConfig<TArtifact, TScenario>[]\n  /** Where each profile's campaign writes artifacts/traces. One subdir per\n   *  profile. */\n  runDir: string\n  /** Git SHA the harness ran from — stamped onto every RunRecord (mandatory\n   *  for paper-grade records). */\n  commitSha: string\n  /** Logical experiment id shared across the whole matrix so the promotion\n   *  gate can pair profiles on matched scenarios. Default: a hash of the\n   *  profile + scenario ids. */\n  experimentId?: string\n  /** Which split these runs belong to. Default `'search'`. */\n  splitTag?: RunSplitTag\n  /** Replicates per (profile, scenario) cell for CI bands. Default 1. */\n  reps?: number\n  /** Campaign seed (per profile). Default 42. */\n  seed?: number\n  /**\n   * Backend-integrity posture, enforced AFTER the matrix completes:\n   *   - `'assert'` (default) — throw `BackendIntegrityError` if the run was a\n   *     stub (and, with `allowMixed:false`, if it was mixed).\n   *   - `'warn'` — log the verdict but never throw.\n   *   - `'off'` — skip the guard entirely (only for offline/replay analysis).\n   */\n  integrity?: 'assert' | 'warn' | 'off'\n  /** Forwarded to `assertRealBackend`. Default true (tolerate partial 429\n   *  cascades); set false for strict CI gates. */\n  allowMixed?: boolean\n  /** Max concurrent cells WITHIN each profile's campaign. Default 2.\n   *  Profiles run sequentially so the cost ceiling is honored deterministically. */\n  maxConcurrency?: number\n  /** Cumulative USD cap per profile campaign. */\n  costCeiling?: number\n  /** Capture flywheel — forwarded to each campaign. */\n  labeledStore?: LabeledScenarioStore | 'off'\n  captureSource?: LabeledScenarioSource\n  /** Storage backend. Default `fsCampaignStorage`. Pass\n   *  `inMemoryCampaignStorage()` for edge/CF-Worker/test runs. */\n  storage?: CampaignStorage\n  /** Test seam — override the wall clock. */\n  now?: () => Date\n  /** Optional persona key per scenario — drives the `byPersona` pivot. When\n   *  unset, `byPersona` is omitted. */\n  personaOf?: (scenario: TScenario) => string\n  /** Validate every produced RunRecord with `validateRunRecord` (fail-loud).\n   *  Default true — catches bad model snapshots and non-finite judge dims at\n   *  the boundary instead of letting them poison downstream analysis. */\n  validate?: boolean\n}\n\nexport interface ProfileSummary {\n  profileId: string\n  profileHash: string\n  model: string\n  /** RunRecords produced for this profile (= scenarios × reps). */\n  records: number\n  /** Mean composite across this profile's records. */\n  meanComposite: number\n  totalCostUsd: number\n  /** Per-profile integrity verdict — surfaces a single profile that ran stub\n   *  even when the matrix as a whole looks real. */\n  integrity: BackendIntegrityReport\n}\n\nexport interface ScenarioRollup {\n  meanComposite: number\n  n: number\n}\n\nexport interface RunProfileMatrixResult<TArtifact, TScenario extends Scenario> {\n  matrixId: string\n  experimentId: string\n  /** One RunRecord per (profile, scenario, rep) cell — the integrity-checked,\n   *  paper-grade output. Feed straight into `analyzeRuns`, `HeldOutGate`,\n   *  scorecards, the hosted wire format. */\n  records: RunRecord[]\n  byProfile: Record<string, ProfileSummary>\n  byScenario: Record<string, ScenarioRollup>\n  /** Present only when `personaOf` was supplied. */\n  byPersona?: Record<string, ScenarioRollup>\n  /** Whole-matrix integrity report (the one `integrity:'assert'` enforces). */\n  integrity: BackendIntegrityReport\n  /** The raw per-profile campaign results, keyed by profile id. */\n  campaigns: Record<string, CampaignResult<TArtifact, TScenario>>\n}\n\nfunction sanitize(id: string): string {\n  return id.replace(/[^a-zA-Z0-9_-]/g, '_')\n}\n\nfunction sha(input: unknown): string {\n  return createHash('sha256').update(JSON.stringify(input)).digest('hex')\n}\n\nfunction mean(xs: number[]): number {\n  return xs.length === 0 ? 0 : xs.reduce((a, b) => a + b, 0) / xs.length\n}\n\nfunction cellComposite(cell: CampaignCellResult<unknown>): number {\n  const composites = Object.values(cell.judgeScores).map((s) => s.composite)\n  return composites.length === 0 ? 0 : mean(composites)\n}\n\ninterface BuildRecordArgs<TArtifact> {\n  cell: CampaignCellResult<TArtifact>\n  profile: AgentProfile\n  profileHash: string\n  configHash: string\n  experimentId: string\n  splitTag: RunSplitTag\n  commitSha: string\n  matrixId: string\n}\n\nfunction buildRunRecord<TArtifact>(args: BuildRecordArgs<TArtifact>): RunRecord {\n  const { cell, profile, profileHash, configHash, experimentId, splitTag, commitSha, matrixId } =\n    args\n  const composite = cellComposite(cell)\n\n  // Flatten judge dimensions (judge-prefixed to avoid collisions) into raw.\n  const raw: Record<string, number> = { composite }\n  const perJudge: Record<string, Record<string, number>> = {}\n  const dimAccum: Record<string, number[]> = {}\n  const notes: string[] = []\n  for (const [judgeName, js] of Object.entries(cell.judgeScores)) {\n    perJudge[judgeName] = { ...js.dimensions }\n    for (const [dim, value] of Object.entries(js.dimensions)) {\n      raw[`${judgeName}.${dim}`] = value\n      ;(dimAccum[dim] ??= []).push(value)\n    }\n    if (js.notes) notes.push(`${judgeName}: ${js.notes}`)\n  }\n  const perDimMean: Record<string, number> = {}\n  for (const [dim, values] of Object.entries(dimAccum)) perDimMean[dim] = mean(values)\n\n  const outcome: RunOutcome =\n    splitTag === 'holdout' ? { holdoutScore: composite, raw } : { searchScore: composite, raw }\n  if (Object.keys(perJudge).length > 0) {\n    outcome.judgeScores = {\n      perJudge,\n      perDimMean,\n      composite,\n      ...(notes.length > 0 ? { notes: notes.join(' | ') } : {}),\n    }\n  }\n\n  return {\n    runId: `${matrixId}:${profile.id}:${cell.cellId}`,\n    experimentId,\n    candidateId: profile.id,\n    seed: cell.seed,\n    model: profile.model,\n    promptHash: profileHash,\n    configHash,\n    commitSha,\n    wallMs: cell.durationMs,\n    costUsd: cell.costUsd,\n    tokenUsage: cell.tokenUsage,\n    outcome,\n    splitTag,\n    scenarioId: cell.scenarioId,\n    ...(cell.error ? { failureMode: cell.error } : {}),\n  }\n}\n\nexport async function runProfileMatrix<TScenario extends Scenario, TArtifact>(\n  opts: RunProfileMatrixOptions<TScenario, TArtifact>,\n): Promise<RunProfileMatrixResult<TArtifact, TScenario>> {\n  if (opts.profiles.length === 0) throw new ProfileMatrixError('profiles must not be empty')\n  if (opts.scenarios.length === 0) throw new ProfileMatrixError('scenarios must not be empty')\n\n  const splitTag = opts.splitTag ?? 'search'\n  const seed = opts.seed ?? 42\n  const validate = opts.validate ?? true\n  const integrityMode = opts.integrity ?? 'assert'\n  const profileIds = opts.profiles.map((p) => p.id)\n  const experimentId =\n    opts.experimentId ??\n    `pm_${sha({ profileIds, scenarios: opts.scenarios.map((s) => s.id) }).slice(0, 16)}`\n  const matrixId = `mtx_${sha({ experimentId, profileIds, seed, splitTag }).slice(0, 16)}`\n\n  // Preflight: every profile must hash (non-empty model) AND its model must\n  // carry a snapshot version, BEFORE any LLM spend. A probe record run through\n  // validateRunRecord catches both in the exact place they'd otherwise surface\n  // far downstream.\n  for (const profile of opts.profiles) {\n    const profileHash = agentProfileHash(profile)\n    try {\n      validateRunRecord({\n        runId: `${matrixId}:${profile.id}:probe`,\n        experimentId,\n        candidateId: profile.id,\n        seed,\n        model: profile.model,\n        promptHash: profileHash,\n        configHash: profileHash,\n        commitSha: opts.commitSha,\n        wallMs: 0,\n        costUsd: 0,\n        tokenUsage: { input: 0, output: 0 },\n        outcome:\n          splitTag === 'holdout' ? { holdoutScore: 0, raw: {} } : { searchScore: 0, raw: {} },\n        splitTag,\n      })\n    } catch (err) {\n      throw new ProfileMatrixError(\n        `profile '${profile.id}' is not recordable: ${err instanceof Error ? err.message : String(err)}`,\n      )\n    }\n  }\n\n  const records: RunRecord[] = []\n  const campaigns: Record<string, CampaignResult<TArtifact, TScenario>> = {}\n  const byProfile: Record<string, ProfileSummary> = {}\n\n  for (const profile of opts.profiles) {\n    const profileHash = agentProfileHash(profile)\n    const configHash = sha({\n      profile: profileHash,\n      judges: (opts.judges ?? []).map((j) => j.name),\n      seed,\n      splitTag,\n    })\n\n    // Bind the profile into a campaign dispatch. Name it so the campaign's\n    // manifest hash is stable + distinct per profile.\n    const dispatch = (scenario: TScenario, ctx: DispatchContext): Promise<TArtifact> =>\n      opts.dispatch(profile, scenario, ctx)\n    Object.defineProperty(dispatch, 'name', { value: `profile_${sanitize(profile.id)}` })\n\n    const campaign = await runCampaign<TScenario, TArtifact>({\n      scenarios: opts.scenarios,\n      dispatch,\n      judges: opts.judges,\n      seed,\n      reps: opts.reps,\n      maxConcurrency: opts.maxConcurrency,\n      costCeiling: opts.costCeiling,\n      labeledStore: opts.labeledStore,\n      captureSource: opts.captureSource,\n      storage: opts.storage,\n      now: opts.now,\n      runDir: join(opts.runDir, sanitize(profile.id)),\n    })\n    campaigns[profile.id] = campaign\n\n    const profileRecords: RunRecord[] = []\n    for (const cell of campaign.cells) {\n      const record = buildRunRecord({\n        cell,\n        profile,\n        profileHash,\n        configHash,\n        experimentId,\n        splitTag,\n        commitSha: opts.commitSha,\n        matrixId,\n      })\n      if (validate) validateRunRecord(record)\n      profileRecords.push(record)\n      records.push(record)\n    }\n\n    byProfile[profile.id] = {\n      profileId: profile.id,\n      profileHash,\n      model: profile.model,\n      records: profileRecords.length,\n      meanComposite: mean(profileRecords.map(compositeOf)),\n      totalCostUsd: profileRecords.reduce((a, r) => a + r.costUsd, 0),\n      integrity: summarizeBackendIntegrity(profileRecords),\n    }\n  }\n\n  // Integrity by construction — the whole point of the primitive.\n  const integrity = summarizeBackendIntegrity(records)\n  if (integrityMode === 'assert') {\n    assertRealBackend(records, { allowMixed: opts.allowMixed ?? true })\n  } else if (integrityMode === 'warn' && integrity.verdict !== 'real') {\n    // eslint-disable-next-line no-console\n    console.warn(\n      `[runProfileMatrix] backend integrity: ${integrity.verdict} — ${integrity.diagnosis}`,\n    )\n  }\n\n  // Pivots.\n  const byScenario = rollup(records, (r) => r.scenarioId)\n  const byPersona = opts.personaOf\n    ? rollupByPersona(records, opts.scenarios, opts.personaOf)\n    : undefined\n\n  return { matrixId, experimentId, records, byProfile, byScenario, byPersona, integrity, campaigns }\n}\n\n/** Composite for a produced RunRecord (the split score it carries). */\nfunction compositeOf(r: RunRecord): number {\n  return r.outcome.holdoutScore ?? r.outcome.searchScore ?? 0\n}\n\nfunction rollup(\n  records: RunRecord[],\n  keyOf: (r: RunRecord) => string | undefined,\n): Record<string, ScenarioRollup> {\n  const groups = new Map<string, number[]>()\n  for (const r of records) {\n    const key = keyOf(r)\n    if (key === undefined) continue\n    const arr = groups.get(key) ?? []\n    arr.push(compositeOf(r))\n    groups.set(key, arr)\n  }\n  const out: Record<string, ScenarioRollup> = {}\n  for (const [key, xs] of groups) out[key] = { meanComposite: mean(xs), n: xs.length }\n  return out\n}\n\nfunction rollupByPersona<TScenario extends Scenario>(\n  records: RunRecord[],\n  scenarios: TScenario[],\n  personaOf: (s: TScenario) => string,\n): Record<string, ScenarioRollup> {\n  const personaByScenarioId = new Map<string, string>()\n  for (const s of scenarios) personaByScenarioId.set(s.id, personaOf(s))\n  return rollup(records, (r) => (r.scenarioId ? personaByScenarioId.get(r.scenarioId) : undefined))\n}\n","/**\n * @experimental\n *\n * VCS-pluggable worktree adapter. One improvement = one worktree, PR-like\n * (multiple commits allowed). A code-tier driver's `propose()` creates a\n * worktree, an agent commits the change into it, and `finalize()` returns a\n * `CodeSurface{ worktreeRef }` the measurement checks out to run the worker\n * against the changed code. On promotion the worktree becomes the PR branch.\n *\n * The interface is VCS-agnostic so a future `jj` ([jj-vcs](https://github.com/jj-vcs/jj))\n * adapter can slot in without touching driver code. Only the git adapter\n * ships today. See `docs/design/self-improvement-engine.md`.\n */\n\nimport { execFileSync } from 'node:child_process'\nimport { existsSync } from 'node:fs'\nimport { basename, isAbsolute, join } from 'node:path'\nimport type { CodeSurface } from '../types'\n\nexport interface Worktree {\n  /** Absolute path to the checked-out worktree directory. */\n  path: string\n  /** The branch the worktree is on (becomes the PR branch on promotion). */\n  branch: string\n  /** The ref the worktree was forked from. */\n  baseRef: string\n}\n\nexport interface WorktreeAdapter {\n  /** Create an isolated worktree on a fresh branch off `baseRef`. */\n  create(opts: { baseRef: string; label: string }): Promise<Worktree>\n  /** Commit any pending changes in the worktree, then return a CodeSurface\n   *  pointing at it. The agent has already written its change into\n   *  `worktree.path` by the time this is called. */\n  finalize(worktree: Worktree, summary: string): Promise<CodeSurface>\n  /** Remove the worktree (and its branch) — called for losing candidates. */\n  discard(worktree: Worktree): Promise<void>\n}\n\nexport class WorktreeAdapterError extends Error {\n  constructor(\n    message: string,\n    readonly cause?: unknown,\n  ) {\n    super(message)\n    this.name = 'WorktreeAdapterError'\n  }\n}\n\nexport interface GitWorktreeAdapterOptions {\n  /** Repo root the worktrees fork from. */\n  repoRoot: string\n  /** Directory worktrees are created under. Default: `<repoRoot>/.worktrees`. */\n  worktreeDir?: string\n  /** Branch-name prefix. Default: `improve`. */\n  branchPrefix?: string\n  /** Test seam — defaults to a real `git` runner. */\n  git?: (args: string[], cwd: string) => string\n}\n\nfunction defaultGit(args: string[], cwd: string): string {\n  try {\n    return execFileSync('git', args, { cwd, encoding: 'utf8' }).trim()\n  } catch (err) {\n    const stderr =\n      err && typeof err === 'object' && 'stderr' in err\n        ? String((err as { stderr: unknown }).stderr)\n        : ''\n    throw new WorktreeAdapterError(`git ${args.join(' ')} failed: ${stderr || String(err)}`, err)\n  }\n}\n\n/** Slugify a label into a branch-safe segment. */\nfunction slug(label: string): string {\n  return (\n    label\n      .toLowerCase()\n      .replace(/[^a-z0-9]+/g, '-')\n      .replace(/^-+|-+$/g, '')\n      .slice(0, 48) || 'candidate'\n  )\n}\n\nexport function gitWorktreeAdapter(opts: GitWorktreeAdapterOptions): WorktreeAdapter {\n  const git = opts.git ?? defaultGit\n  const worktreeDir = opts.worktreeDir ?? join(opts.repoRoot, '.worktrees')\n  const branchPrefix = opts.branchPrefix ?? 'improve'\n\n  return {\n    async create({ baseRef, label }) {\n      const id = `${slug(label)}-${Date.now().toString(36)}-${Math.random().toString(36).slice(2, 6)}`\n      const branch = `${branchPrefix}/${id}`\n      const path = join(worktreeDir, id)\n      git(['worktree', 'add', '-b', branch, path, baseRef], opts.repoRoot)\n      return { path, branch, baseRef }\n    },\n\n    async finalize(worktree, summary) {\n      // Stage + commit any pending changes the agent left in the worktree.\n      // A no-op commit is refused by git, so only commit when the tree is dirty.\n      const status = git(['status', '--porcelain'], worktree.path)\n      if (status.length > 0) {\n        git(['add', '-A'], worktree.path)\n        git(['commit', '-m', summary], worktree.path)\n      }\n      return {\n        kind: 'code',\n        worktreeRef: worktree.path,\n        baseRef: worktree.baseRef,\n        summary,\n      }\n    },\n\n    async discard(worktree) {\n      // Remove the worktree, then delete its branch. Force-remove because the\n      // worktree may hold uncommitted experiment state we're discarding.\n      git(['worktree', 'remove', '--force', worktree.path], opts.repoRoot)\n      git(['branch', '-D', worktree.branch], opts.repoRoot)\n    },\n  }\n}\n\n/** Resolve a `CodeSurface`'s worktreeRef to a directory the measurement can\n *  run the worker in. A path ref is returned as-is; anything else is treated\n *  as a ref under the adapter's worktree dir. */\nexport function resolveWorktreePath(surface: CodeSurface, worktreeDir?: string): string {\n  if (isAbsolute(surface.worktreeRef) && existsSync(surface.worktreeRef)) return surface.worktreeRef\n  if (worktreeDir) return join(worktreeDir, basename(surface.worktreeRef))\n  return surface.worktreeRef\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AA6BA,SAAS,kBAAkB;AAC3B,SAAS,YAAY,WAAW,cAAc,qBAAqB;AACnE,SAAS,YAAY;AAqBd,IAAM,4BAAN,cAAwC,MAAM;AAAA,EACnD,YACkB,MAChB,SACA;AACA,UAAM,OAAO;AAHG;AAIhB,SAAK,OAAO;AAAA,EACd;AAAA,EALkB;AAMpB;AAQO,IAAM,yBAAN,MAA6D;AAAA,EAIlE,YAA6B,SAAwC;AAAxC;AAC3B,QAAI,CAAC,WAAW,QAAQ,IAAI,EAAG,WAAU,QAAQ,MAAM,EAAE,WAAW,KAAK,CAAC;AAC1E,SAAK,MAAM,QAAQ,OAAO,KAAK;AAAA,EACjC;AAAA,EAH6B;AAAA,EAHZ;AAAA,EACA,aAAa,oBAAI,IAA4B;AAAA,EAO9D,MAAM,QAAQ,OAA4C;AACxD,SAAK,iBAAiB,KAAK;AAC3B,SAAK,gBAAgB,KAAK;AAC1B,UAAM,SAAS,KAAK,SAAS,KAAK;AAClC,UAAM,OAAO,KAAK,cAAc,MAAM,MAAM;AAC5C,UAAM,OAAO,GAAG,KAAK,UAAU,MAAM,CAAC;AAAA;AAItC,eAAW,MAAM,IAAI;AAAA,EACvB;AAAA,EAEA,MAAM,OAAO,MAAmE;AAC9E,QAAI,CAAC,KAAK,OAAO;AACf,YAAM,IAAI;AAAA,QACR;AAAA,QACA;AAAA,MACF;AAAA,IACF;AACA,QAAI,CAAC,KAAK,gBAAgB;AACxB,YAAM,IAAI;AAAA,QACR;AAAA,QACA;AAAA,MACF;AAAA,IACF;AAEA,UAAM,MAA+B,CAAC;AACtC,eAAW,UAAU,aAAa;AAGhC,UAAI,KAAK,UAAU,WAAW,WAAW,oBAAoB;AAC3D,cAAM,WAAW,qBAAqB,KAAK,QAAQ,QAAQ,kBAAkB;AAC7E,YAAI,CAAC,SAAU;AAAA,MACjB;AACA,YAAM,OAAO,KAAK,cAAc,MAAM;AACtC,UAAI,CAAC,WAAW,IAAI,EAAG;AACvB,YAAM,QAAQ,aAAa,MAAM,MAAM,EAAE,MAAM,IAAI,EAAE,OAAO,OAAO;AACnE,iBAAW,QAAQ,OAAO;AACxB,YAAI;AACJ,YAAI;AACF,mBAAS,KAAK,MAAM,IAAI;AAAA,QAC1B,QAAQ;AACN;AAAA,QACF;AACA,YAAI,CAAC,cAAc,QAAQ,MAAM,MAAM,EAAG;AAC1C,YAAI,KAAK,MAAM;AAAA,MACjB;AAAA,IACF;AAGA,QAAI,KAAK,CAAC,GAAG,MAAM;AACjB,UAAI,EAAE,eAAe,EAAE,WAAY,QAAO,EAAE,WAAW,cAAc,EAAE,UAAU;AACjF,aAAO,EAAE,WAAW,cAAc,EAAE,UAAU;AAAA,IAChD,CAAC;AAED,WAAO,IAAI,MAAM,GAAG,KAAK,KAAK;AAAA,EAChC;AAAA,EAEA,MAAM,OAKH;AACD,UAAM,WAAmC,CAAC;AAC1C,UAAM,UAAsC;AAAA,MAC1C,YAAY;AAAA,MACZ,mBAAmB;AAAA,MACnB,eAAe;AAAA,IACjB;AACA,QAAI,QAAQ;AACZ,eAAW,UAAU,aAAa;AAChC,YAAM,OAAO,KAAK,cAAc,MAAM;AACtC,UAAI,CAAC,WAAW,IAAI,GAAG;AACrB,iBAAS,MAAM,IAAI;AACnB;AAAA,MACF;AACA,YAAM,QAAQ,aAAa,MAAM,MAAM,EAAE,MAAM,IAAI,EAAE,OAAO,OAAO;AACnE,eAAS,MAAM,IAAI,MAAM;AACzB,eAAS,MAAM;AACf,iBAAW,QAAQ,OAAO;AACxB,YAAI,QAAoB;AACxB,YAAI;AACF,kBAAS,KAAK,MAAM,IAAI,EAA4B,cAAc;AAAA,QACpE,QAAQ;AAAA,QAER;AACA,gBAAQ,KAAK,KAAK;AAAA,MACpB;AAAA,IACF;AAIA,WAAO,EAAE,OAAO,OAAO,MAAM,OAAO,UAAU,QAAQ;AAAA,EACxD;AAAA,EAEQ,iBAAiB,OAAmC;AAC1D,QAAI,CAAC,MAAM,QAAQ;AACjB,YAAM,IAAI;AAAA,QACR;AAAA,QACA;AAAA,MACF;AAAA,IACF;AACA,QAAI,CAAC,MAAM,qBAAqB,MAAM,kBAAkB,WAAW,GAAG;AACpE,YAAM,IAAI;AAAA,QACR;AAAA,QACA;AAAA,MACF;AAAA,IACF;AACA,QAAI,CAAC,MAAM,YAAY;AACrB,YAAM,IAAI;AAAA,QACR;AAAA,QACA;AAAA,MACF;AAAA,IACF;AACA,QAAI,CAAC,MAAM,iBAAiB;AAC1B,YAAM,IAAI;AAAA,QACR;AAAA,QACA;AAAA,MACF;AAAA,IACF;AACA,QAAI,CAAC,YAAY,SAAS,MAAM,MAAM,GAAG;AACvC,YAAM,IAAI;AAAA,QACR;AAAA,QACA,+CAA+C,YAAY,KAAK,IAAI,CAAC;AAAA,MACvE;AAAA,IACF;AAAA,EACF;AAAA,EAEQ,gBAAgB,OAAmC;AACzD,UAAM,MAAM,KAAK,QAAQ;AACzB,QAAI,CAAC,OAAO,CAAC,MAAM,gBAAiB;AACpC,UAAM,MAAM,KAAK,IAAI;AACrB,UAAM,WAAW;AACjB,QAAI,QAAQ,KAAK,WAAW,IAAI,MAAM,eAAe;AACrD,QAAI,CAAC,SAAS,MAAM,MAAM,iBAAiB,UAAU;AACnD,cAAQ,EAAE,QAAQ,MAAM,iBAAiB,eAAe,KAAK,OAAO,EAAE;AACtE,WAAK,WAAW,IAAI,MAAM,iBAAiB,KAAK;AAAA,IAClD;AACA,QAAI,MAAM,SAAS,KAAK;AACtB,YAAM,IAAI;AAAA,QACR;AAAA,QACA,gCAAgC,MAAM,eAAe,aAAa,GAAG;AAAA,MACvE;AAAA,IACF;AACA,UAAM,SAAS;AAAA,EACjB;AAAA,EAEQ,SAAS,OAAoD;AACnE,UAAM,aAAa;AAAA,MACjB,KAAK,UAAU;AAAA,QACb,IAAI,MAAM,SAAS;AAAA,QACnB,KAAK,MAAM;AAAA,QACX,IAAI,MAAM;AAAA,QACV,KAAK,MAAM;AAAA,MACb,CAAC;AAAA,IACH;AAGA,WAAO;AAAA,MACL,GAAG;AAAA,MACH;AAAA,MACA,OAAO;AAAA,IACT;AAAA,EACF;AAAA,EAEQ,cAAc,QAAwB;AAC5C,WAAO,KAAK,KAAK,QAAQ,MAAM,GAAG,MAAM,QAAQ;AAAA,EAClD;AACF;AAEA,IAAM,cAAgD;AAAA,EACpD;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AACF;AAEA,SAAS,qBACP,QACA,QACS;AACT,MAAI,CAAC,OAAQ,QAAO;AACpB,MAAI,MAAM,QAAQ,MAAM,EAAG,QAAO,OAAO,SAAS,MAAM;AACxD,SAAO,WAAW;AACpB;AAEA,SAAS,cACP,QACA,MACA,QACS;AAET,MAAI,KAAK,UAAU,WAAW,OAAO,cAAc,KAAK,eAAgB,QAAO;AAC/E,MAAI,KAAK,UAAU,UAAU,OAAO,aAAa,KAAK,eAAgB,QAAO;AAE7E,QAAM,IAAI,KAAK;AACf,MAAI,CAAC,EAAG,QAAO;AACf,MAAI,EAAE,QAAQ,OAAO,SAAS,SAAS,EAAE,KAAM,QAAO;AACtD,MAAI,EAAE,QAAQ;AACZ,UAAM,UAAU,MAAM,QAAQ,EAAE,MAAM,IAAI,EAAE,SAAS,CAAC,EAAE,MAAM;AAC9D,QAAI,CAAC,QAAQ,SAAS,MAAe,EAAG,QAAO;AAAA,EACjD;AACA,MAAI,EAAE,iBAAiB,UAAa,EAAE,iBAAiB,QAAW;AAChE,UAAM,aAAa,OAAO,OAAO,OAAO,WAAW,EAAE,IAAI,CAAC,MAAM,EAAE,SAAS;AAC3E,UAAM,MAAM,WAAW,WAAW,IAAI,IAAI,KAAK,IAAI,GAAG,UAAU;AAChE,QAAI,EAAE,iBAAiB,UAAa,MAAM,EAAE,aAAc,QAAO;AACjE,QAAI,EAAE,iBAAiB,UAAa,MAAM,EAAE,aAAc,QAAO;AAAA,EACnE;AACA,MAAI,EAAE,aAAa,UAAa,eAAe,OAAO,UAAU,IAAI,eAAe,EAAE,QAAQ,GAAG;AAC9F,WAAO;AAAA,EACT;AACA,SAAO;AACT;AAEA,SAAS,OAAO,OAAuB;AACrC,SAAO,WAAW,QAAQ,EAAE,OAAO,KAAK,EAAE,OAAO,KAAK,EAAE,MAAM,GAAG,EAAE;AACrE;AAEA,SAAS,WAAW,MAAc,MAAoB;AACpD,MAAI,WAAW,IAAI,GAAG;AACpB,UAAM,WAAW,aAAa,MAAM,MAAM;AAC1C,kBAAc,MAAM,WAAW,IAAI;AAAA,EACrC,OAAO;AACL,kBAAc,MAAM,IAAI;AAAA,EAC1B;AACF;;;AC5QA,SAAS,cAAAA,mBAAkB;AAC3B,SAAS,QAAAC,aAAY;AA6Bd,IAAM,qBAAN,cAAiC,eAAe;AAAA,EACrD,YAAY,SAAiB;AAC3B,UAAM,kBAAkB,OAAO;AAAA,EACjC;AACF;AAyGA,SAAS,SAAS,IAAoB;AACpC,SAAO,GAAG,QAAQ,mBAAmB,GAAG;AAC1C;AAEA,SAAS,IAAI,OAAwB;AACnC,SAAOC,YAAW,QAAQ,EAAE,OAAO,KAAK,UAAU,KAAK,CAAC,EAAE,OAAO,KAAK;AACxE;AAEA,SAAS,KAAK,IAAsB;AAClC,SAAO,GAAG,WAAW,IAAI,IAAI,GAAG,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,GAAG;AAClE;AAEA,SAAS,cAAc,MAA2C;AAChE,QAAM,aAAa,OAAO,OAAO,KAAK,WAAW,EAAE,IAAI,CAAC,MAAM,EAAE,SAAS;AACzE,SAAO,WAAW,WAAW,IAAI,IAAI,KAAK,UAAU;AACtD;AAaA,SAAS,eAA0B,MAA6C;AAC9E,QAAM,EAAE,MAAM,SAAS,aAAa,YAAY,cAAc,UAAU,WAAW,SAAS,IAC1F;AACF,QAAM,YAAY,cAAc,IAAI;AAGpC,QAAM,MAA8B,EAAE,UAAU;AAChD,QAAM,WAAmD,CAAC;AAC1D,QAAM,WAAqC,CAAC;AAC5C,QAAM,QAAkB,CAAC;AACzB,aAAW,CAAC,WAAW,EAAE,KAAK,OAAO,QAAQ,KAAK,WAAW,GAAG;AAC9D,aAAS,SAAS,IAAI,EAAE,GAAG,GAAG,WAAW;AACzC,eAAW,CAAC,KAAK,KAAK,KAAK,OAAO,QAAQ,GAAG,UAAU,GAAG;AACxD,UAAI,GAAG,SAAS,IAAI,GAAG,EAAE,IAAI;AAC5B,OAAC,SAAS,GAAG,MAAM,CAAC,GAAG,KAAK,KAAK;AAAA,IACpC;AACA,QAAI,GAAG,MAAO,OAAM,KAAK,GAAG,SAAS,KAAK,GAAG,KAAK,EAAE;AAAA,EACtD;AACA,QAAM,aAAqC,CAAC;AAC5C,aAAW,CAAC,KAAK,MAAM,KAAK,OAAO,QAAQ,QAAQ,EAAG,YAAW,GAAG,IAAI,KAAK,MAAM;AAEnF,QAAM,UACJ,aAAa,YAAY,EAAE,cAAc,WAAW,IAAI,IAAI,EAAE,aAAa,WAAW,IAAI;AAC5F,MAAI,OAAO,KAAK,QAAQ,EAAE,SAAS,GAAG;AACpC,YAAQ,cAAc;AAAA,MACpB;AAAA,MACA;AAAA,MACA;AAAA,MACA,GAAI,MAAM,SAAS,IAAI,EAAE,OAAO,MAAM,KAAK,KAAK,EAAE,IAAI,CAAC;AAAA,IACzD;AAAA,EACF;AAEA,SAAO;AAAA,IACL,OAAO,GAAG,QAAQ,IAAI,QAAQ,EAAE,IAAI,KAAK,MAAM;AAAA,IAC/C;AAAA,IACA,aAAa,QAAQ;AAAA,IACrB,MAAM,KAAK;AAAA,IACX,OAAO,QAAQ;AAAA,IACf,YAAY;AAAA,IACZ;AAAA,IACA;AAAA,IACA,QAAQ,KAAK;AAAA,IACb,SAAS,KAAK;AAAA,IACd,YAAY,KAAK;AAAA,IACjB;AAAA,IACA;AAAA,IACA,YAAY,KAAK;AAAA,IACjB,GAAI,KAAK,QAAQ,EAAE,aAAa,KAAK,MAAM,IAAI,CAAC;AAAA,EAClD;AACF;AAEA,eAAsB,iBACpB,MACuD;AACvD,MAAI,KAAK,SAAS,WAAW,EAAG,OAAM,IAAI,mBAAmB,4BAA4B;AACzF,MAAI,KAAK,UAAU,WAAW,EAAG,OAAM,IAAI,mBAAmB,6BAA6B;AAE3F,QAAM,WAAW,KAAK,YAAY;AAClC,QAAM,OAAO,KAAK,QAAQ;AAC1B,QAAM,WAAW,KAAK,YAAY;AAClC,QAAM,gBAAgB,KAAK,aAAa;AACxC,QAAM,aAAa,KAAK,SAAS,IAAI,CAAC,MAAM,EAAE,EAAE;AAChD,QAAM,eACJ,KAAK,gBACL,MAAM,IAAI,EAAE,YAAY,WAAW,KAAK,UAAU,IAAI,CAAC,MAAM,EAAE,EAAE,EAAE,CAAC,EAAE,MAAM,GAAG,EAAE,CAAC;AACpF,QAAM,WAAW,OAAO,IAAI,EAAE,cAAc,YAAY,MAAM,SAAS,CAAC,EAAE,MAAM,GAAG,EAAE,CAAC;AAMtF,aAAW,WAAW,KAAK,UAAU;AACnC,UAAM,cAAc,iBAAiB,OAAO;AAC5C,QAAI;AACF,wBAAkB;AAAA,QAChB,OAAO,GAAG,QAAQ,IAAI,QAAQ,EAAE;AAAA,QAChC;AAAA,QACA,aAAa,QAAQ;AAAA,QACrB;AAAA,QACA,OAAO,QAAQ;AAAA,QACf,YAAY;AAAA,QACZ,YAAY;AAAA,QACZ,WAAW,KAAK;AAAA,QAChB,QAAQ;AAAA,QACR,SAAS;AAAA,QACT,YAAY,EAAE,OAAO,GAAG,QAAQ,EAAE;AAAA,QAClC,SACE,aAAa,YAAY,EAAE,cAAc,GAAG,KAAK,CAAC,EAAE,IAAI,EAAE,aAAa,GAAG,KAAK,CAAC,EAAE;AAAA,QACpF;AAAA,MACF,CAAC;AAAA,IACH,SAAS,KAAK;AACZ,YAAM,IAAI;AAAA,QACR,YAAY,QAAQ,EAAE,wBAAwB,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG,CAAC;AAAA,MAChG;AAAA,IACF;AAAA,EACF;AAEA,QAAM,UAAuB,CAAC;AAC9B,QAAM,YAAkE,CAAC;AACzE,QAAM,YAA4C,CAAC;AAEnD,aAAW,WAAW,KAAK,UAAU;AACnC,UAAM,cAAc,iBAAiB,OAAO;AAC5C,UAAM,aAAa,IAAI;AAAA,MACrB,SAAS;AAAA,MACT,SAAS,KAAK,UAAU,CAAC,GAAG,IAAI,CAAC,MAAM,EAAE,IAAI;AAAA,MAC7C;AAAA,MACA;AAAA,IACF,CAAC;AAID,UAAM,WAAW,CAAC,UAAqB,QACrC,KAAK,SAAS,SAAS,UAAU,GAAG;AACtC,WAAO,eAAe,UAAU,QAAQ,EAAE,OAAO,WAAW,SAAS,QAAQ,EAAE,CAAC,GAAG,CAAC;AAEpF,UAAM,WAAW,MAAM,YAAkC;AAAA,MACvD,WAAW,KAAK;AAAA,MAChB;AAAA,MACA,QAAQ,KAAK;AAAA,MACb;AAAA,MACA,MAAM,KAAK;AAAA,MACX,gBAAgB,KAAK;AAAA,MACrB,aAAa,KAAK;AAAA,MAClB,cAAc,KAAK;AAAA,MACnB,eAAe,KAAK;AAAA,MACpB,SAAS,KAAK;AAAA,MACd,KAAK,KAAK;AAAA,MACV,QAAQC,MAAK,KAAK,QAAQ,SAAS,QAAQ,EAAE,CAAC;AAAA,IAChD,CAAC;AACD,cAAU,QAAQ,EAAE,IAAI;AAExB,UAAM,iBAA8B,CAAC;AACrC,eAAW,QAAQ,SAAS,OAAO;AACjC,YAAM,SAAS,eAAe;AAAA,QAC5B;AAAA,QACA;AAAA,QACA;AAAA,QACA;AAAA,QACA;AAAA,QACA;AAAA,QACA,WAAW,KAAK;AAAA,QAChB;AAAA,MACF,CAAC;AACD,UAAI,SAAU,mBAAkB,MAAM;AACtC,qBAAe,KAAK,MAAM;AAC1B,cAAQ,KAAK,MAAM;AAAA,IACrB;AAEA,cAAU,QAAQ,EAAE,IAAI;AAAA,MACtB,WAAW,QAAQ;AAAA,MACnB;AAAA,MACA,OAAO,QAAQ;AAAA,MACf,SAAS,eAAe;AAAA,MACxB,eAAe,KAAK,eAAe,IAAI,WAAW,CAAC;AAAA,MACnD,cAAc,eAAe,OAAO,CAAC,GAAG,MAAM,IAAI,EAAE,SAAS,CAAC;AAAA,MAC9D,WAAW,0BAA0B,cAAc;AAAA,IACrD;AAAA,EACF;AAGA,QAAM,YAAY,0BAA0B,OAAO;AACnD,MAAI,kBAAkB,UAAU;AAC9B,sBAAkB,SAAS,EAAE,YAAY,KAAK,cAAc,KAAK,CAAC;AAAA,EACpE,WAAW,kBAAkB,UAAU,UAAU,YAAY,QAAQ;AAEnE,YAAQ;AAAA,MACN,yCAAyC,UAAU,OAAO,WAAM,UAAU,SAAS;AAAA,IACrF;AAAA,EACF;AAGA,QAAM,aAAa,OAAO,SAAS,CAAC,MAAM,EAAE,UAAU;AACtD,QAAM,YAAY,KAAK,YACnB,gBAAgB,SAAS,KAAK,WAAW,KAAK,SAAS,IACvD;AAEJ,SAAO,EAAE,UAAU,cAAc,SAAS,WAAW,YAAY,WAAW,WAAW,UAAU;AACnG;AAGA,SAAS,YAAY,GAAsB;AACzC,SAAO,EAAE,QAAQ,gBAAgB,EAAE,QAAQ,eAAe;AAC5D;AAEA,SAAS,OACP,SACA,OACgC;AAChC,QAAM,SAAS,oBAAI,IAAsB;AACzC,aAAW,KAAK,SAAS;AACvB,UAAM,MAAM,MAAM,CAAC;AACnB,QAAI,QAAQ,OAAW;AACvB,UAAM,MAAM,OAAO,IAAI,GAAG,KAAK,CAAC;AAChC,QAAI,KAAK,YAAY,CAAC,CAAC;AACvB,WAAO,IAAI,KAAK,GAAG;AAAA,EACrB;AACA,QAAM,MAAsC,CAAC;AAC7C,aAAW,CAAC,KAAK,EAAE,KAAK,OAAQ,KAAI,GAAG,IAAI,EAAE,eAAe,KAAK,EAAE,GAAG,GAAG,GAAG,OAAO;AACnF,SAAO;AACT;AAEA,SAAS,gBACP,SACA,WACA,WACgC;AAChC,QAAM,sBAAsB,oBAAI,IAAoB;AACpD,aAAW,KAAK,UAAW,qBAAoB,IAAI,EAAE,IAAI,UAAU,CAAC,CAAC;AACrE,SAAO,OAAO,SAAS,CAAC,MAAO,EAAE,aAAa,oBAAoB,IAAI,EAAE,UAAU,IAAI,MAAU;AAClG;;;AC/YA,SAAS,oBAAoB;AAC7B,SAAS,cAAAC,mBAAkB;AAC3B,SAAS,UAAU,YAAY,QAAAC,aAAY;AAuBpC,IAAM,uBAAN,cAAmC,MAAM;AAAA,EAC9C,YACE,SACS,OACT;AACA,UAAM,OAAO;AAFJ;AAGT,SAAK,OAAO;AAAA,EACd;AAAA,EAJW;AAKb;AAaA,SAAS,WAAW,MAAgB,KAAqB;AACvD,MAAI;AACF,WAAO,aAAa,OAAO,MAAM,EAAE,KAAK,UAAU,OAAO,CAAC,EAAE,KAAK;AAAA,EACnE,SAAS,KAAK;AACZ,UAAM,SACJ,OAAO,OAAO,QAAQ,YAAY,YAAY,MAC1C,OAAQ,IAA4B,MAAM,IAC1C;AACN,UAAM,IAAI,qBAAqB,OAAO,KAAK,KAAK,GAAG,CAAC,YAAY,UAAU,OAAO,GAAG,CAAC,IAAI,GAAG;AAAA,EAC9F;AACF;AAGA,SAAS,KAAK,OAAuB;AACnC,SACE,MACG,YAAY,EACZ,QAAQ,eAAe,GAAG,EAC1B,QAAQ,YAAY,EAAE,EACtB,MAAM,GAAG,EAAE,KAAK;AAEvB;AAEO,SAAS,mBAAmB,MAAkD;AACnF,QAAM,MAAM,KAAK,OAAO;AACxB,QAAM,cAAc,KAAK,eAAeA,MAAK,KAAK,UAAU,YAAY;AACxE,QAAM,eAAe,KAAK,gBAAgB;AAE1C,SAAO;AAAA,IACL,MAAM,OAAO,EAAE,SAAS,MAAM,GAAG;AAC/B,YAAM,KAAK,GAAG,KAAK,KAAK,CAAC,IAAI,KAAK,IAAI,EAAE,SAAS,EAAE,CAAC,IAAI,KAAK,OAAO,EAAE,SAAS,EAAE,EAAE,MAAM,GAAG,CAAC,CAAC;AAC9F,YAAM,SAAS,GAAG,YAAY,IAAI,EAAE;AACpC,YAAM,OAAOA,MAAK,aAAa,EAAE;AACjC,UAAI,CAAC,YAAY,OAAO,MAAM,QAAQ,MAAM,OAAO,GAAG,KAAK,QAAQ;AACnE,aAAO,EAAE,MAAM,QAAQ,QAAQ;AAAA,IACjC;AAAA,IAEA,MAAM,SAAS,UAAU,SAAS;AAGhC,YAAM,SAAS,IAAI,CAAC,UAAU,aAAa,GAAG,SAAS,IAAI;AAC3D,UAAI,OAAO,SAAS,GAAG;AACrB,YAAI,CAAC,OAAO,IAAI,GAAG,SAAS,IAAI;AAChC,YAAI,CAAC,UAAU,MAAM,OAAO,GAAG,SAAS,IAAI;AAAA,MAC9C;AACA,aAAO;AAAA,QACL,MAAM;AAAA,QACN,aAAa,SAAS;AAAA,QACtB,SAAS,SAAS;AAAA,QAClB;AAAA,MACF;AAAA,IACF;AAAA,IAEA,MAAM,QAAQ,UAAU;AAGtB,UAAI,CAAC,YAAY,UAAU,WAAW,SAAS,IAAI,GAAG,KAAK,QAAQ;AACnE,UAAI,CAAC,UAAU,MAAM,SAAS,MAAM,GAAG,KAAK,QAAQ;AAAA,IACtD;AAAA,EACF;AACF;AAKO,SAAS,oBAAoB,SAAsB,aAA8B;AACtF,MAAI,WAAW,QAAQ,WAAW,KAAKD,YAAW,QAAQ,WAAW,EAAG,QAAO,QAAQ;AACvF,MAAI,YAAa,QAAOC,MAAK,aAAa,SAAS,QAAQ,WAAW,CAAC;AACvE,SAAO,QAAQ;AACjB;","names":["createHash","join","createHash","join","existsSync","join"]}

package/dist/{chunk-QDOSODID.js → chunk-3B7Y5AUR.js}

@@ -1,6 +1,6 @@
 import {
   cohensD
-} from "./chunk-S3SDD56V.js";
+} from "./chunk-ITBRCT73.js";
 import {
   argHash,
   groupBy,
@@ -551,4 +551,4 @@ export {
   iqr,
   welchsTTest
 };
-//# sourceMappingURL=chunk-QDOSODID.js.map
+//# sourceMappingURL=chunk-3B7Y5AUR.js.map

package/dist/{chunk-QYJT52YW.js → chunk-3BFEG2F6.js}

@@ -54,4 +54,4 @@ export {
   VerificationError,
   ReplayError
 };
-//# sourceMappingURL=chunk-QYJT52YW.js.map
+//# sourceMappingURL=chunk-3BFEG2F6.js.map

package/dist/chunk-3BFEG2F6.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/errors.ts"],"sourcesContent":["/**\n * Error taxonomy for `@tangle-network/agent-eval`.\n *\n * Every error this package throws as part of its *public contract* extends\n * `AgentEvalError`. Consumers can pattern-match by `instanceof <Subclass>` or\n * by the stable string `code` carried on the base class.\n *\n * The codes are stable across minor versions; new codes can be added, but\n * existing codes never change meaning. New subclasses are non-breaking.\n *\n * Internal invariant guards (`throw new Error('this should never happen')`)\n * remain plain `Error`s on purpose — they're programmer-mistake assertions,\n * not consumer-catchable contract failures.\n */\n\nexport type AgentEvalErrorCode =\n  | 'validation'\n  | 'not_found'\n  | 'config'\n  | 'capture_integrity'\n  | 'judge'\n  | 'verification'\n  | 'replay'\n  | 'backend_integrity'\n  | 'profile_matrix'\n\nexport class AgentEvalError extends Error {\n  /** Stable string code. Survives minification; safe to switch on. */\n  readonly code: AgentEvalErrorCode\n\n  constructor(code: AgentEvalErrorCode, message: string, options?: { cause?: unknown }) {\n    super(message, options)\n    this.name = this.constructor.name\n    this.code = code\n  }\n}\n\n/** Caller passed invalid arguments (out of range, mutually-exclusive options, bad shape). */\nexport class ValidationError extends AgentEvalError {\n  constructor(message: string, options?: { cause?: unknown }) {\n    super('validation', message, options)\n  }\n}\n\n/** A named resource (run, span, rubric, scenario, dataset row, route) does not exist. */\nexport class NotFoundError extends AgentEvalError {\n  constructor(message: string, options?: { cause?: unknown }) {\n    super('not_found', message, options)\n  }\n}\n\n/** Configuration missing or malformed (`HOME` unset, required image not supplied, env var absent). */\nexport class ConfigError extends AgentEvalError {\n  constructor(message: string, options?: { cause?: unknown }) {\n    super('config', message, options)\n  }\n}\n\n/**\n * A run is missing the artifacts a launch-grade check requires:\n * raw HTTP capture absent, no LLM spans, route assertion failed, run-end\n * assertion tripped. Block ship on this; do not catch and move on.\n */\nexport class CaptureIntegrityError extends AgentEvalError {\n  constructor(message: string, options?: { cause?: unknown }) {\n    super('capture_integrity', message, options)\n  }\n}\n\n/** A judge call failed in a way that's not retryable: schema parse failure, bad rubric, conflicting dimensions. */\nexport class JudgeError extends AgentEvalError {\n  constructor(message: string, options?: { cause?: unknown }) {\n    super('judge', message, options)\n  }\n}\n\n/** A verifier signalled a hard failure (compile, test, schema) — distinct from a low judge score. */\nexport class VerificationError extends AgentEvalError {\n  constructor(message: string, options?: { cause?: unknown }) {\n    super('verification', message, options)\n  }\n}\n\n/** Replay cache cannot satisfy a request: miss with no fallback, sink lacks list(), unsupported URL. */\nexport class ReplayError extends AgentEvalError {\n  constructor(message: string, options?: { cause?: unknown }) {\n    super('replay', message, options)\n  }\n}\n"],"mappings":";AA0BO,IAAM,iBAAN,cAA6B,MAAM;AAAA;AAAA,EAE/B;AAAA,EAET,YAAY,MAA0B,SAAiB,SAA+B;AACpF,UAAM,SAAS,OAAO;AACtB,SAAK,OAAO,KAAK,YAAY;AAC7B,SAAK,OAAO;AAAA,EACd;AACF;AAGO,IAAM,kBAAN,cAA8B,eAAe;AAAA,EAClD,YAAY,SAAiB,SAA+B;AAC1D,UAAM,cAAc,SAAS,OAAO;AAAA,EACtC;AACF;AAGO,IAAM,gBAAN,cAA4B,eAAe;AAAA,EAChD,YAAY,SAAiB,SAA+B;AAC1D,UAAM,aAAa,SAAS,OAAO;AAAA,EACrC;AACF;AAGO,IAAM,cAAN,cAA0B,eAAe;AAAA,EAC9C,YAAY,SAAiB,SAA+B;AAC1D,UAAM,UAAU,SAAS,OAAO;AAAA,EAClC;AACF;AAOO,IAAM,wBAAN,cAAoC,eAAe;AAAA,EACxD,YAAY,SAAiB,SAA+B;AAC1D,UAAM,qBAAqB,SAAS,OAAO;AAAA,EAC7C;AACF;AAGO,IAAM,aAAN,cAAyB,eAAe;AAAA,EAC7C,YAAY,SAAiB,SAA+B;AAC1D,UAAM,SAAS,SAAS,OAAO;AAAA,EACjC;AACF;AAGO,IAAM,oBAAN,cAAgC,eAAe;AAAA,EACpD,YAAY,SAAiB,SAA+B;AAC1D,UAAM,gBAAgB,SAAS,OAAO;AAAA,EACxC;AACF;AAGO,IAAM,cAAN,cAA0B,eAAe;AAAA,EAC9C,YAAY,SAAiB,SAA+B;AAC1D,UAAM,UAAU,SAAS,OAAO;AAAA,EAClC;AACF;","names":[]}

package/dist/{chunk-J4DIMSRK.js → chunk-6EKXFFGQ.js}

@@ -4,7 +4,7 @@ import {
 } from "./chunk-NCRFYPS3.js";
 import {
   validateRunRecord
-} from "./chunk-NCK5QLGT.js";
+} from "./chunk-F3SRAAZO.js";
 import {
   TraceEmitter
 } from "./chunk-TVVP3ZZQ.js";
@@ -610,4 +610,4 @@ export {
   runProposeReviewAsControlLoop,
   controlFailureClassFromVerification
 };
-//# sourceMappingURL=chunk-J4DIMSRK.js.map
+//# sourceMappingURL=chunk-6EKXFFGQ.js.map

package/dist/{chunk-63EPZQUZ.js → chunk-6REHLN5J.js}

@@ -1,6 +1,6 @@
 import {
   callLlmJson
-} from "./chunk-VXNVVBZO.js";
+} from "./chunk-IHDHUN2X.js";
 
 // src/wire/schemas.ts
 import { extendZodWithOpenApi } from "@asteasolutions/zod-to-openapi";
@@ -1002,4 +1002,4 @@ export {
   startServer,
   startServerAsync
 };
-//# sourceMappingURL=chunk-63EPZQUZ.js.map
+//# sourceMappingURL=chunk-6REHLN5J.js.map

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

@@ -1,6 +1,9 @@
+import {
+  BackendIntegrityError
+} from "./chunk-E22YUOAL.js";
 import {
   confidenceInterval
-} from "./chunk-S3SDD56V.js";
+} from "./chunk-ITBRCT73.js";
 
 // src/campaign/run-campaign.ts
 import { createHash } from "crypto";
@@ -111,6 +114,7 @@ async function runCampaign(opts) {
             signal: abortController.signal
           });
           cellsRef.push(result.cell);
+          enforceCellUsage(result.cell, opts.expectUsage ?? "warn");
           totalCostUsd += result.cell.costUsd;
           Object.assign(artifactsByPath, result.artifactsByPath);
           if (opts.costCeiling !== void 0 && totalCostUsd >= opts.costCeiling) {
@@ -187,13 +191,22 @@ async function executeCell(args) {
     }
   };
   let costSoFar = 0;
+  const tokensSoFar = { input: 0, output: 0 };
   const cost = {
     observe(amount, source) {
       costSoFar += amount;
       trace.span(`cost.${source}`, { amountUsd: amount }).end();
     },
+    observeTokens(usage) {
+      tokensSoFar.input += usage.input;
+      tokensSoFar.output += usage.output;
+      if (usage.cached) tokensSoFar.cached = (tokensSoFar.cached ?? 0) + usage.cached;
+    },
     current() {
       return costSoFar;
+    },
+    tokens() {
+      return { ...tokensSoFar };
     }
   };
   const placement = args.opts.cellPlacement?.({
@@ -241,6 +254,7 @@ async function executeCell(args) {
     artifact: artifact ?? null,
     judgeScores,
     costUsd: costSoFar,
+    tokenUsage: { ...tokensSoFar },
     durationMs: Date.now() - startMs,
     seed: args.slot.cellSeed,
     cached: false,
@@ -251,6 +265,28 @@ async function executeCell(args) {
   }
   return { cell, artifactsByPath };
 }
+function enforceCellUsage(cell, mode) {
+  if (mode === "off" || cell.error) return;
+  if (cell.artifact === null || cell.artifact === void 0) return;
+  const zeroTokens = cell.tokenUsage.input === 0 && cell.tokenUsage.output === 0;
+  if (cell.costUsd !== 0 || !zeroTokens) return;
+  const msg = `cell '${cell.cellId}' produced an artifact but reported zero cost and zero tokens \u2014 the dispatch never reported LLM usage via ctx.cost.observe/observeTokens (a stub cell)`;
+  if (mode === "assert") {
+    const report = {
+      totalRecords: 1,
+      stubRecords: 1,
+      realRecords: 0,
+      uncostedRecords: 0,
+      totalInputTokens: 0,
+      totalOutputTokens: 0,
+      totalCostUsd: 0,
+      verdict: "stub",
+      diagnosis: msg
+    };
+    throw new BackendIntegrityError(`expectUsage: ${msg}`, report);
+  }
+  console.warn(`[runCampaign] expectUsage: ${msg}`);
+}
 async function runJudgeCell(judge, input) {
   return judge.score(input);
 }
@@ -287,6 +323,7 @@ function skippedCell(slot, reason) {
     artifact: null,
     judgeScores: {},
     costUsd: 0,
+    tokenUsage: { input: 0, output: 0 },
     durationMs: 0,
     seed: slot.cellSeed,
     cached: false,
@@ -363,4 +400,4 @@ export {
   inMemoryCampaignStorage,
   runCampaign
 };
-//# sourceMappingURL=chunk-NOPYCRNG.js.map
+//# sourceMappingURL=chunk-7TPYV2ER.js.map