@tangle-network/agent-eval 0.90.1 → 0.92.0

package/dist/campaign/index.d.ts

@@ -7,7 +7,7 @@ export { A as AxisEvidence, a as AxisVerdict, B as BuildEvidenceVectorOptions, k
 import { L as LlmClientOptions } from '../llm-client-BeEcAokY.js';
 import { T as TraceAnalystKindSpec } from '../kind-factory-5b7xXXOr.js';
 import { c as AnalystFinding } from '../types-2VVIL04s.js';
-import { S as SignedManifest, A as AgentProfile, B as BackendIntegrityReport, C as CompletionRequirement, R as RuntimeEventLike, a as CompletionVerdict, P as ProducedState, b as CorrectnessChecker } from '../pre-registration-DELOEJ8v.js';
+import { S as SignedManifest, A as AgentProfile, B as BackendIntegrityReport, C as CompletionRequirement, R as RuntimeEventLike, a as CompletionVerdict, P as ProducedState, b as CorrectnessChecker } from '../pre-registration-mAnCugl9.js';
 import { E as EProcessState, a as PairedBootstrapResult } from '../statistics-C7PozGrZ.js';
 import { A as AgentEvalError } from '../errors-CzMUYo7b.js';
 import { a as RunSplitTag, R as RunRecord } from '../run-record-e7vj1uZQ.js';

package/dist/campaign/index.js

@@ -15,7 +15,7 @@ import {
   agentProfileHash,
   extractProducedState,
   verifyCompletion
-} from "../chunk-FIUKOSWI.js";
+} from "../chunk-S6OZEZQK.js";
 import {
   buildLoopProvenanceRecord,
   campaignBreakdown,

package/dist/{chunk-FIUKOSWI.js → chunk-S6OZEZQK.js}

@@ -38,15 +38,36 @@ var STOPWORDS = /* @__PURE__ */ new Set([
   "with",
   "by"
 ]);
+var REQUIREMENT_FORM_STOPWORDS = /* @__PURE__ */ new Set([
+  "generated",
+  "generate",
+  "view",
+  "render",
+  "rendered",
+  "persisted",
+  "persist",
+  "artifact",
+  "file",
+  "document",
+  "note",
+  "proposal",
+  "deliverable",
+  "output",
+  "created",
+  "create",
+  "produce",
+  "produced",
+  "flag"
+]);
 var MATCH_THRESHOLD = 0.5;
 var MIN_CONTENT_CHARS = 50;
-function tokens(s) {
+function tokens(s, extraStop) {
   return new Set(
-    s.toLowerCase().split(/[^a-z0-9]+/).filter((t) => t.length > 1 && !STOPWORDS.has(t))
+    s.toLowerCase().split(/[^a-z0-9]+/).filter((t) => t.length > 1 && !STOPWORDS.has(t) && !extraStop?.has(t))
   );
 }
 function tokenRecall(requirementText, candidateText) {
-  const req = tokens(requirementText);
+  const req = tokens(requirementText, REQUIREMENT_FORM_STOPWORDS);
   if (req.size === 0) return 0;
   const cand = tokens(candidateText);
   let hit = 0;
@@ -58,7 +79,10 @@ function artifactCandidates(req, reqIndex, artifacts) {
   const out = [];
   artifacts.forEach((a, i) => {
     if ((a.content ?? "").trim().length < MIN_CONTENT_CHARS) return;
-    let score = tokenRecall(reqText, `${a.path ?? ""} ${a.kind}`);
+    let score = tokenRecall(
+      reqText,
+      `${a.path ?? ""} ${a.kind} ${(a.content ?? "").slice(0, 4e3)}`
+    );
     if (req.category && a.kind && req.category.toLowerCase() === a.kind.toLowerCase()) {
       score = Math.max(score, 1);
     }
@@ -78,15 +102,16 @@ function proposalCandidates(req, reqIndex, proposals) {
   const out = [];
   for (const p of proposals) {
     if (p.status !== "approved") continue;
-    const score = tokenRecall(reqText, p.title);
+    const body = (p.content ?? "").trim();
+    if (body.length < MIN_CONTENT_CHARS) continue;
+    const score = tokenRecall(reqText, `${p.title} ${body}`);
     if (score < MATCH_THRESHOLD) continue;
-    const body = p.content ?? "";
     out.push({
       reqIndex,
       itemKey: `proposal:${p.id}`,
       score,
       evidence: `approved proposal '${p.title}' matched (token recall ${score.toFixed(2)})`,
-      content: body.trim().length >= MIN_CONTENT_CHARS ? body : null
+      content: body
     });
   }
   return out;
@@ -278,7 +303,12 @@ function extractProducedState(events) {
       });
     } else if (ev.type === "proposal_created") {
       const p = ev;
-      proposals.push({ id: p.proposalId, title: p.title, status: p.status ?? "pending" });
+      proposals.push({
+        id: p.proposalId,
+        title: p.title,
+        status: p.status ?? "pending",
+        ...p.content !== void 0 ? { content: p.content } : {}
+      });
     }
   }
   return { artifacts, proposals, toolCalls };
@@ -309,4 +339,4 @@ export {
   extractProducedState,
   agentProfileHash
 };
-//# sourceMappingURL=chunk-FIUKOSWI.js.map
+//# sourceMappingURL=chunk-S6OZEZQK.js.map

package/dist/chunk-S6OZEZQK.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/completion-verifier.ts","../src/produced-state.ts","../src/agent-profile.ts"],"sourcesContent":["/**\n * Completion verifier — the task-completion oracle.\n *\n * Answers the only eval question that is not a proxy: did the agent actually\n * COMPLETE the task — produce every required deliverable, persisted and\n * correct — rather than describe what should be done. A fluent transcript\n * that never produces the artifact scores zero here.\n *\n * Per requirement, a two-stage check:\n *   1. Structural — a produced item (vault artifact / approved proposal /\n *      tool call) of the right kind is matched against the requirement and\n *      carries non-empty content. Deterministic; no LLM.\n *   2. Correctness — only if structurally present AND the matched item\n *      carries content, one targeted check decides whether that item\n *      actually fulfils the requirement. A hallucinated artifact fails here;\n *      an absent one already failed stage 1.\n *\n * `completionRate` is satisfied / total. Quality dimensions are meaningless\n * on an incomplete task — callers gate on `fullyComplete` / `completionRate`\n * before scoring quality.\n */\n\nimport type { TCloud } from '@tangle-network/tcloud'\nimport type { Artifact } from './artifact-validator'\nimport type { DefaultVerdict } from './verdict'\n\n/** What kind of produced state can satisfy a requirement structurally. */\nexport type SatisfiedBy = 'artifact' | 'proposal' | 'tool-call' | 'any'\n\nexport interface CompletionRequirement {\n  /** Stable id from the task gold (e.g. a persona's `expected_requirements[].req_id`). */\n  reqId: string\n  /** Human-readable description of the required deliverable. */\n  title: string\n  /** Optional kind/category hint, matched against a produced item's kind. */\n  category?: string\n  /** What produced state satisfies this requirement. Defaults to 'any'. */\n  satisfiedBy?: SatisfiedBy\n}\n\nexport interface TaskGold {\n  taskId: string\n  requirements: CompletionRequirement[]\n}\n\nexport interface ProducedProposal {\n  id: string\n  title: string\n  status: 'pending' | 'approved' | 'rejected'\n  /** Optional persisted body — when present, enables a correctness check. */\n  content?: string\n}\n\n/** Everything observable about what a run actually produced. */\nexport interface ProducedState {\n  /** Persisted vault artifacts. Reuses the shared `Artifact` shape. */\n  artifacts: Artifact[]\n  /** Proposals / filings the agent created. */\n  proposals: ProducedProposal[]\n  /** Names of tools the agent invoked. */\n  toolCalls: string[]\n}\n\nexport interface RequirementCheck {\n  reqId: string\n  title: string\n  /** A produced item of the right kind matched the requirement, non-empty. */\n  structurallyPresent: boolean\n  /**\n   * Whether the matched item actually fulfils the requirement. `null` when\n   * not structurally present, or when the matched item carries no content\n   * to assess.\n   */\n  correct: boolean | null\n  /** structurallyPresent && correct !== false. */\n  satisfied: boolean\n  /** Human-readable evidence for the verdict. */\n  evidence: string[]\n}\n\n/** Extends the substrate verdict spine: `valid` = `fullyComplete` and\n *  `score` = `completionRate` — derived in `completionVerdict()`, the one\n *  place those equalities hold by construction. */\nexport interface CompletionVerdict extends DefaultVerdict {\n  taskId: string\n  requirements: RequirementCheck[]\n  /** satisfied / total requirements. */\n  completionRate: number\n  /** Every requirement satisfied. */\n  fullyComplete: boolean\n}\n\n/**\n * Construct a `CompletionVerdict` from the per-requirement checks, deriving\n * `completionRate` / `fullyComplete` and the spine fields (`valid` =\n * `fullyComplete`, `score` = `completionRate`) in one place. Throws on zero\n * requirements — a verdict over nothing is a misconfiguration, mirroring\n * `verifyCompletion`'s gold-spec guard.\n */\nexport function completionVerdict(input: {\n  taskId: string\n  requirements: RequirementCheck[]\n}): CompletionVerdict {\n  if (input.requirements.length === 0) {\n    throw new Error(\n      `completionVerdict: task '${input.taskId}' has no requirement checks — nothing to derive a verdict from`,\n    )\n  }\n  const satisfiedCount = input.requirements.filter((r) => r.satisfied).length\n  const completionRate = satisfiedCount / input.requirements.length\n  const fullyComplete = satisfiedCount === input.requirements.length\n  return {\n    taskId: input.taskId,\n    requirements: input.requirements,\n    completionRate,\n    fullyComplete,\n    valid: fullyComplete,\n    score: completionRate,\n  }\n}\n\n/**\n * Decides whether a produced item's content actually fulfils a requirement.\n * Injected so the structural verifier stays pure and unit-testable; the\n * production implementation is `createLlmCorrectnessChecker`.\n */\nexport type CorrectnessChecker = (\n  requirement: CompletionRequirement,\n  content: string,\n) => Promise<{ correct: boolean; reason: string }>\n\nconst STOPWORDS = new Set([\n  'the',\n  'a',\n  'an',\n  'of',\n  'for',\n  'and',\n  'or',\n  'to',\n  'in',\n  'on',\n  'with',\n  'by',\n])\n\n// Deliverable-FORM vocabulary — words that name the SHAPE of an output, not its\n// domain content. A correct \"swap-comparison view persisted as a ui/** artifact\"\n// is an OpenUI JSON whose body says nothing about \"artifact\" / \"persisted\" /\n// \"view\"; the discriminative tokens are the domain nouns (swap, comparison).\n// Stripped from the REQUIREMENT side of structural recall so a deliverable is\n// matched on what it IS about, not on the boilerplate describing its form. The\n// correctness checker strips the same class via TITLE_STOPWORDS. Anti-game holds:\n// the distinctive domain tokens remain, so an off-topic item still fails.\nconst REQUIREMENT_FORM_STOPWORDS = new Set([\n  'generated',\n  'generate',\n  'view',\n  'render',\n  'rendered',\n  'persisted',\n  'persist',\n  'artifact',\n  'file',\n  'document',\n  'note',\n  'proposal',\n  'deliverable',\n  'output',\n  'created',\n  'create',\n  'produce',\n  'produced',\n  'flag',\n])\n\nconst MATCH_THRESHOLD = 0.5\nconst MIN_CONTENT_CHARS = 50\n\nfunction tokens(s: string, extraStop?: Set<string>): Set<string> {\n  return new Set(\n    s\n      .toLowerCase()\n      .split(/[^a-z0-9]+/)\n      .filter((t) => t.length > 1 && !STOPWORDS.has(t) && !extraStop?.has(t)),\n  )\n}\n\n/**\n * Recall of the requirement's tokens within a candidate's identifying text.\n * Recall, not Jaccard — a candidate's path/id legitimately carries extra\n * tokens the requirement does not name. The requirement side drops\n * deliverable-FORM vocabulary so recall keys on the distinctive domain tokens.\n */\nfunction tokenRecall(requirementText: string, candidateText: string): number {\n  const req = tokens(requirementText, REQUIREMENT_FORM_STOPWORDS)\n  if (req.size === 0) return 0\n  const cand = tokens(candidateText)\n  let hit = 0\n  for (const t of req) if (cand.has(t)) hit++\n  return hit / req.size\n}\n\ninterface Candidate {\n  reqIndex: number\n  /** Unique key for a produced item — each item satisfies at most one requirement. */\n  itemKey: string\n  score: number\n  evidence: string\n  /** Content to correctness-check, or null when the matched item has none. */\n  content: string | null\n}\n\nfunction artifactCandidates(\n  req: CompletionRequirement,\n  reqIndex: number,\n  artifacts: Artifact[],\n): Candidate[] {\n  const reqText = `${req.title} ${req.category ?? ''}`\n  const out: Candidate[] = []\n  artifacts.forEach((a, i) => {\n    if ((a.content ?? '').trim().length < MIN_CONTENT_CHARS) return\n    // Match against the artifact CONTENT too, not just its path + kind — a\n    // generated view / note whose path is generic still satisfies a requirement\n    // when its body covers it (e.g. an OpenUI comparison grounded in the on-file\n    // figures). Bounded slice keeps the recall text cheap; MATCH_THRESHOLD holds.\n    let score = tokenRecall(\n      reqText,\n      `${a.path ?? ''} ${a.kind} ${(a.content ?? '').slice(0, 4000)}`,\n    )\n    if (req.category && a.kind && req.category.toLowerCase() === a.kind.toLowerCase()) {\n      score = Math.max(score, 1)\n    }\n    if (score < MATCH_THRESHOLD) return\n    out.push({\n      reqIndex,\n      itemKey: `artifact:${i}`,\n      score,\n      evidence: `artifact '${a.path ?? a.kind}' matched (token recall ${score.toFixed(2)})`,\n      content: a.content ?? null,\n    })\n  })\n  return out\n}\n\nfunction proposalCandidates(\n  req: CompletionRequirement,\n  reqIndex: number,\n  proposals: ProducedProposal[],\n): Candidate[] {\n  const reqText = `${req.title} ${req.category ?? ''}`\n  const out: Candidate[] = []\n  for (const p of proposals) {\n    // Pending or rejected work is not a completed deliverable.\n    if (p.status !== 'approved') continue\n    // A proposal needs an assessable BODY to be a deliverable. A bare title is\n    // not completion: correctness cannot be judged on it, so a title-only match\n    // would auto-pass the oracle (structurallyPresent && correct===null →\n    // satisfied) with no verifiable content. Tool calls are the only\n    // legitimately content-less deliverable (`toolCallCandidates`).\n    const body = (p.content ?? '').trim()\n    if (body.length < MIN_CONTENT_CHARS) continue\n    // Match against the body as well as the (often short) title — a refusal /\n    // flag / analysis proposal whose title is a label still satisfies a\n    // descriptively-worded requirement when its content covers it. MATCH_THRESHOLD\n    // + the requirement's distinctive tokens keep an off-topic proposal out;\n    // correctness (a SEMANTIC checker, NOT this lexical pass) then judges\n    // polarity/fulfilment, so a negation that merely contains the tokens fails.\n    // Structural and correctness must use different evidence or the two-stage\n    // check collapses to one lexical gate.\n    const score = tokenRecall(reqText, `${p.title} ${body}`)\n    if (score < MATCH_THRESHOLD) continue\n    out.push({\n      reqIndex,\n      itemKey: `proposal:${p.id}`,\n      score,\n      evidence: `approved proposal '${p.title}' matched (token recall ${score.toFixed(2)})`,\n      content: body,\n    })\n  }\n  return out\n}\n\nfunction toolCallCandidates(\n  req: CompletionRequirement,\n  reqIndex: number,\n  toolCalls: string[],\n): Candidate[] {\n  const out: Candidate[] = []\n  toolCalls.forEach((name, i) => {\n    const score = tokenRecall(req.title, name)\n    if (score < MATCH_THRESHOLD) return\n    out.push({\n      reqIndex,\n      itemKey: `tool:${i}`,\n      score,\n      evidence: `tool call '${name}' matched (token recall ${score.toFixed(2)})`,\n      content: null,\n    })\n  })\n  return out\n}\n\n/**\n * Verify whether a run completed the task. `checkCorrectness` is injected —\n * `createLlmCorrectnessChecker` for production, a deterministic stub in tests.\n *\n * Throws on a gold spec with no requirements: an eval task that requires\n * nothing is a misconfiguration, not a vacuously-complete task.\n */\nexport async function verifyCompletion(\n  gold: TaskGold,\n  state: ProducedState,\n  checkCorrectness: CorrectnessChecker,\n): Promise<CompletionVerdict> {\n  if (gold.requirements.length === 0) {\n    throw new Error(\n      `verifyCompletion: task '${gold.taskId}' has no requirements — malformed gold spec`,\n    )\n  }\n\n  // Collect every above-threshold (requirement, produced-item) candidate, then\n  // assign greedily by descending score: each requirement and each produced\n  // item is used at most once. One deliverable fulfils one requirement.\n  const candidates: Candidate[] = []\n  gold.requirements.forEach((req, i) => {\n    const by = req.satisfiedBy ?? 'any'\n    if (by === 'artifact' || by === 'any') {\n      candidates.push(...artifactCandidates(req, i, state.artifacts))\n    }\n    if (by === 'proposal' || by === 'any') {\n      candidates.push(...proposalCandidates(req, i, state.proposals))\n    }\n    if (by === 'tool-call' || by === 'any') {\n      candidates.push(...toolCallCandidates(req, i, state.toolCalls))\n    }\n  })\n  candidates.sort((a, b) => b.score - a.score)\n\n  const assigned = new Map<number, Candidate>()\n  const itemTaken = new Set<string>()\n  for (const c of candidates) {\n    if (assigned.has(c.reqIndex) || itemTaken.has(c.itemKey)) continue\n    assigned.set(c.reqIndex, c)\n    itemTaken.add(c.itemKey)\n  }\n\n  const requirements: RequirementCheck[] = []\n  for (let i = 0; i < gold.requirements.length; i++) {\n    const req = gold.requirements[i]!\n    const match = assigned.get(i)\n    const evidence: string[] = []\n    let correct: boolean | null = null\n\n    if (match) {\n      evidence.push(match.evidence)\n      if (match.content !== null) {\n        const r = await checkCorrectness(req, match.content)\n        correct = r.correct\n        evidence.push(`correctness: ${r.correct ? 'pass' : 'fail'} — ${r.reason}`)\n      } else {\n        evidence.push('correctness: not assessed — matched item carries no content')\n      }\n    } else {\n      const by = req.satisfiedBy ?? 'any'\n      const kind = by === 'any' ? 'artifact/proposal/tool-call' : by\n      evidence.push(`no produced ${kind} matched this requirement`)\n    }\n\n    const structurallyPresent = match !== undefined\n    const satisfied = structurallyPresent && correct !== false\n    requirements.push({\n      reqId: req.reqId,\n      title: req.title,\n      structurallyPresent,\n      correct,\n      satisfied,\n      evidence,\n    })\n  }\n\n  return completionVerdict({ taskId: gold.taskId, requirements })\n}\n\nexport interface LlmCorrectnessCheckerOpts {\n  model?: string\n  /** Max chars of artifact content sent to the checker. */\n  maxContentChars?: number\n}\n\n/** Parse the correctness checker's model response. Fails loud on a bad shape. */\nexport function parseCorrectnessResponse(raw: string): { correct: boolean; reason: string } {\n  const match = raw.match(/\\{[\\s\\S]*\\}/)\n  if (!match) {\n    throw new Error(`correctness checker: no JSON object in model response: ${raw.slice(0, 200)}`)\n  }\n  const parsed = JSON.parse(match[0]) as { correct?: unknown; reason?: unknown }\n  if (typeof parsed.correct !== 'boolean') {\n    throw new Error(`correctness checker: 'correct' is not a boolean in: ${match[0].slice(0, 200)}`)\n  }\n  return { correct: parsed.correct, reason: typeof parsed.reason === 'string' ? parsed.reason : '' }\n}\n\n/**\n * Production `CorrectnessChecker` — one LLM call per matched artifact,\n * deterministic (temperature 0), structured JSON out. Judges fulfilment\n * only: a plan, a gesture, or a description of what should be done does not\n * fulfil a requirement — the artifact must BE the deliverable.\n */\nexport function createLlmCorrectnessChecker(\n  tc: TCloud,\n  opts: LlmCorrectnessCheckerOpts = {},\n): CorrectnessChecker {\n  const model = opts.model ?? 'claude-sonnet-4-6'\n  const maxContentChars = opts.maxContentChars ?? 8000\n  return async (requirement, content) => {\n    const resp = await tc.chat({\n      model,\n      messages: [\n        {\n          role: 'system',\n          content:\n            'You verify whether a produced work artifact actually fulfils a stated requirement. Judge fulfilment only — is the deliverable substantively present and on-point — not polish. A plan to do it later, a vague gesture, or a description of what should be done does NOT fulfil a requirement; the artifact must BE the deliverable. Respond with a single JSON object: {\"correct\": boolean, \"reason\": string (<= 30 words)}.',\n        },\n        {\n          role: 'user',\n          content: `Requirement: ${requirement.title}\\n${\n            requirement.category ? `Category: ${requirement.category}\\n` : ''\n          }\\nProduced artifact:\\n${content.slice(0, maxContentChars)}`,\n        },\n      ],\n      temperature: 0,\n      maxTokens: 200,\n    })\n    const raw =\n      (resp as { choices?: { message?: { content?: string } }[] }).choices?.[0]?.message?.content ??\n      ''\n    return parseCorrectnessResponse(raw)\n  }\n}\n\n/** Stopwords for requirement-title tokenization — drops the imperative verbs\n *  ('review', 'update', …) common to deliverable titles so recall keys on the\n *  substantive nouns, not the boilerplate ask. */\nconst TITLE_STOPWORDS = new Set([\n  'the',\n  'a',\n  'an',\n  'and',\n  'or',\n  'for',\n  'to',\n  'of',\n  'in',\n  'on',\n  'with',\n  'review',\n  'update',\n  'new',\n  'proposed',\n])\n\n/**\n * Deterministic `CorrectnessChecker` — the no-LLM counterpart to\n * `createLlmCorrectnessChecker`. A produced item fulfils a requirement when its\n * content is substantive (≥ `minContentLength` chars) AND recalls ≥ `minRecall`\n * of the requirement title's significant tokens. No network.\n *\n * Polarity-blind: token recall credits a negation that contains the\n * requirement's tokens (\"I will NOT produce the comparison\" recalls every token\n * of \"produce the comparison\"). The structural match stage is ALSO lexical, so\n * pairing the two collapses to a single gameable gate. Use this only as an\n * opt-in structural pre-filter or for tasks whose requirements have no polarity\n * to invert; for produced-state grading the correctness checker MUST be semantic\n * (`createLlmCorrectnessChecker`). See the anti-game fixtures in the test suite.\n */\nexport function createTokenRecallChecker(\n  opts: { minRecall?: number; minContentLength?: number } = {},\n): CorrectnessChecker {\n  const minRecall = opts.minRecall ?? 0.5\n  const minLen = opts.minContentLength ?? 120\n  return async (requirement, content) => {\n    const body = content.trim()\n    if (body.length < minLen)\n      return {\n        correct: false,\n        reason: `content too thin (${body.length} chars) to be the deliverable`,\n      }\n    const titleTokens = requirement.title\n      .toLowerCase()\n      .split(/[^a-z0-9]+/)\n      .filter((t) => t.length > 2 && !TITLE_STOPWORDS.has(t))\n    if (titleTokens.length === 0)\n      return {\n        correct: true,\n        reason: 'requirement title has no significant tokens — structural match accepted',\n      }\n    const lower = body.toLowerCase()\n    const hits = titleTokens.filter((t) => lower.includes(t)).length\n    const recall = hits / titleTokens.length\n    return recall >= minRecall\n      ? {\n          correct: true,\n          reason: `content recalls ${hits}/${titleTokens.length} requirement tokens`,\n        }\n      : {\n          correct: false,\n          reason: `content recalls only ${hits}/${titleTokens.length} requirement tokens`,\n        }\n  }\n}\n","/**\n * Produced-state extraction — normalize a run's runtime event stream into the\n * typed `ProducedState` the completion oracle consumes.\n *\n * `ProducedState` answers \"what did the agent actually produce\" — vault\n * artifacts, proposals, tool calls. The runtime emits these as a stream of\n * events; this module is the single normalization point from that stream to\n * the shape `verifyCompletion` expects.\n *\n * Input is structurally typed (`RuntimeEventLike`) so this module does not\n * depend on agent-runtime — agent-runtime's `RuntimeStreamEvent` satisfies it\n * structurally. The `content` on `ArtifactEventLike` and the whole\n * `proposal_created` variant are the runtime-side enrichments this contract\n * requires; the runtime emits them, this module consumes them.\n */\n\nimport type { Artifact } from './artifact-validator'\nimport type { ProducedProposal, ProducedState } from './completion-verifier'\n\n/** A tool the agent invoked. */\nexport interface ToolCallEventLike {\n  type: 'tool_call'\n  toolName: string\n}\n\n/**\n * An artifact the agent produced. `content` is the enriched field — the\n * runtime's base `artifact` event carries only metadata; the completion\n * oracle needs the body to verify the deliverable, so the runtime emits it.\n */\nexport interface ArtifactEventLike {\n  type: 'artifact'\n  artifactId: string\n  name?: string\n  mimeType?: string\n  uri?: string\n  content?: string\n}\n\n/** A proposal / filing the agent created. */\nexport interface ProposalEventLike {\n  type: 'proposal_created'\n  proposalId: string\n  title: string\n  status?: 'pending' | 'approved' | 'rejected'\n  // body of the proposal (e.g. a submit_proposal `description`). When present,\n  // the completion oracle correctness-checks it like artifact content; absent,\n  // the proposal is graded presence-only.\n  content?: string\n}\n\n/**\n * The subset of runtime stream events `extractProducedState` consumes.\n * agent-runtime's full `RuntimeStreamEvent` union satisfies this structurally;\n * the `{ type: string }` catch-all keeps the input permissive so callers can\n * pass the whole unfiltered telemetry stream — unrecognized events are skipped.\n */\nexport type RuntimeEventLike =\n  | ToolCallEventLike\n  | ArtifactEventLike\n  | ProposalEventLike\n  | { type: string }\n\nfunction artifactKind(mimeType: string | undefined): string {\n  if (!mimeType) return 'file'\n  if (mimeType.includes('json')) return 'json'\n  if (mimeType.startsWith('text/')) return 'text'\n  return 'file'\n}\n\n/**\n * Normalize a run's runtime event stream into `ProducedState`.\n *\n * Pure and total — unrecognized event types are skipped. `toolCalls` is\n * deduplicated by name in first-seen order (completion cares about a tool's\n * presence, not its call count). An artifact with neither a name nor a uri\n * still yields an entry keyed by its `artifactId` so it is never silently\n * dropped; an artifact with no `content` yields empty content, which the\n * completion oracle's structural check then rejects on its own.\n */\nexport function extractProducedState(events: readonly RuntimeEventLike[]): ProducedState {\n  const artifacts: Artifact[] = []\n  const proposals: ProducedProposal[] = []\n  const toolCalls: string[] = []\n  const seenTools = new Set<string>()\n\n  for (const ev of events) {\n    if (ev.type === 'tool_call') {\n      const name = (ev as ToolCallEventLike).toolName\n      if (name && !seenTools.has(name)) {\n        seenTools.add(name)\n        toolCalls.push(name)\n      }\n    } else if (ev.type === 'artifact') {\n      const a = ev as ArtifactEventLike\n      artifacts.push({\n        kind: artifactKind(a.mimeType),\n        path: a.name ?? a.uri ?? a.artifactId,\n        content: a.content ?? '',\n      })\n    } else if (ev.type === 'proposal_created') {\n      const p = ev as ProposalEventLike\n      proposals.push({\n        id: p.proposalId,\n        title: p.title,\n        status: p.status ?? 'pending',\n        ...(p.content !== undefined ? { content: p.content } : {}),\n      })\n    }\n  }\n\n  return { artifacts, proposals, toolCalls }\n}\n","/**\n * @stable\n *\n * AgentProfile — the eval harness's unit of variation.\n *\n * A profile pins everything that changes agent behaviour for a benchmark\n * cell: the model, the active skills, the prompt version, the available\n * tools. Vary the profile — swap a model, add a skill — and re-run the suite\n * to benchmark the change. The scorecard keys a cell on\n * `(scenarioId, profileHash)`, so the model is not a separate axis: it lives\n * inside the profile, and two profiles with the same model but different\n * skills are different cells.\n *\n * `agentProfileHash` is the profile's behaviour identity. Two profiles that\n * produce the same agent behaviour share a hash (and a scorecard cell);\n * reordering `skills` or `tools` does not change it; the human-facing `id`\n * label does not affect it.\n */\n\nimport { createHash } from 'node:crypto'\nimport { ValidationError } from './errors'\nimport { canonicalize } from './pre-registration'\n\nexport interface AgentProfile {\n  /** Human-facing label, e.g. `sonnet-legal-skills-v3`. Not part of the hash. */\n  id: string\n  /** Model snapshot id this profile pins, e.g. `claude-sonnet-4-6@2025-04-15`. */\n  model: string\n  /** Skill ids/versions active in this profile — the primary behaviour lever. */\n  skills?: string[]\n  /** Prompt version identifier. */\n  promptVersion?: string\n  /** Tool ids available to the agent. */\n  tools?: string[]\n  /** Any other behaviour-bearing knobs that should fingerprint into the hash. */\n  metadata?: Record<string, string | number | boolean>\n}\n\n/**\n * Deterministic behaviour identity of a profile — a sha256 over the\n * behaviour-bearing fields. `skills` and `tools` are order-insensitive; the\n * `id` label is excluded. Throws on a profile with no `model` — an unkeyable\n * profile must fail loud rather than collapse into a blank-model cell.\n */\nexport function agentProfileHash(profile: AgentProfile): string {\n  if (typeof profile.model !== 'string' || profile.model.trim().length === 0) {\n    throw new ValidationError(`AgentProfile \"${profile.id}\" has no model — cannot hash`)\n  }\n  const behaviour = {\n    model: profile.model.trim(),\n    skills: [...(profile.skills ?? [])].sort(),\n    promptVersion: profile.promptVersion ?? null,\n    tools: [...(profile.tools ?? [])].sort(),\n    metadata: profile.metadata ?? {},\n  }\n  return createHash('sha256')\n    .update(JSON.stringify(canonicalize(behaviour)))\n    .digest('hex')\n}\n"],"mappings":";;;;;;;;AAmGO,SAAS,kBAAkB,OAGZ;AACpB,MAAI,MAAM,aAAa,WAAW,GAAG;AACnC,UAAM,IAAI;AAAA,MACR,4BAA4B,MAAM,MAAM;AAAA,IAC1C;AAAA,EACF;AACA,QAAM,iBAAiB,MAAM,aAAa,OAAO,CAAC,MAAM,EAAE,SAAS,EAAE;AACrE,QAAM,iBAAiB,iBAAiB,MAAM,aAAa;AAC3D,QAAM,gBAAgB,mBAAmB,MAAM,aAAa;AAC5D,SAAO;AAAA,IACL,QAAQ,MAAM;AAAA,IACd,cAAc,MAAM;AAAA,IACpB;AAAA,IACA;AAAA,IACA,OAAO;AAAA,IACP,OAAO;AAAA,EACT;AACF;AAYA,IAAM,YAAY,oBAAI,IAAI;AAAA,EACxB;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AACF,CAAC;AAUD,IAAM,6BAA6B,oBAAI,IAAI;AAAA,EACzC;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AACF,CAAC;AAED,IAAM,kBAAkB;AACxB,IAAM,oBAAoB;AAE1B,SAAS,OAAO,GAAW,WAAsC;AAC/D,SAAO,IAAI;AAAA,IACT,EACG,YAAY,EACZ,MAAM,YAAY,EAClB,OAAO,CAAC,MAAM,EAAE,SAAS,KAAK,CAAC,UAAU,IAAI,CAAC,KAAK,CAAC,WAAW,IAAI,CAAC,CAAC;AAAA,EAC1E;AACF;AAQA,SAAS,YAAY,iBAAyB,eAA+B;AAC3E,QAAM,MAAM,OAAO,iBAAiB,0BAA0B;AAC9D,MAAI,IAAI,SAAS,EAAG,QAAO;AAC3B,QAAM,OAAO,OAAO,aAAa;AACjC,MAAI,MAAM;AACV,aAAW,KAAK,IAAK,KAAI,KAAK,IAAI,CAAC,EAAG;AACtC,SAAO,MAAM,IAAI;AACnB;AAYA,SAAS,mBACP,KACA,UACA,WACa;AACb,QAAM,UAAU,GAAG,IAAI,KAAK,IAAI,IAAI,YAAY,EAAE;AAClD,QAAM,MAAmB,CAAC;AAC1B,YAAU,QAAQ,CAAC,GAAG,MAAM;AAC1B,SAAK,EAAE,WAAW,IAAI,KAAK,EAAE,SAAS,kBAAmB;AAKzD,QAAI,QAAQ;AAAA,MACV;AAAA,MACA,GAAG,EAAE,QAAQ,EAAE,IAAI,EAAE,IAAI,KAAK,EAAE,WAAW,IAAI,MAAM,GAAG,GAAI,CAAC;AAAA,IAC/D;AACA,QAAI,IAAI,YAAY,EAAE,QAAQ,IAAI,SAAS,YAAY,MAAM,EAAE,KAAK,YAAY,GAAG;AACjF,cAAQ,KAAK,IAAI,OAAO,CAAC;AAAA,IAC3B;AACA,QAAI,QAAQ,gBAAiB;AAC7B,QAAI,KAAK;AAAA,MACP;AAAA,MACA,SAAS,YAAY,CAAC;AAAA,MACtB;AAAA,MACA,UAAU,aAAa,EAAE,QAAQ,EAAE,IAAI,2BAA2B,MAAM,QAAQ,CAAC,CAAC;AAAA,MAClF,SAAS,EAAE,WAAW;AAAA,IACxB,CAAC;AAAA,EACH,CAAC;AACD,SAAO;AACT;AAEA,SAAS,mBACP,KACA,UACA,WACa;AACb,QAAM,UAAU,GAAG,IAAI,KAAK,IAAI,IAAI,YAAY,EAAE;AAClD,QAAM,MAAmB,CAAC;AAC1B,aAAW,KAAK,WAAW;AAEzB,QAAI,EAAE,WAAW,WAAY;AAM7B,UAAM,QAAQ,EAAE,WAAW,IAAI,KAAK;AACpC,QAAI,KAAK,SAAS,kBAAmB;AASrC,UAAM,QAAQ,YAAY,SAAS,GAAG,EAAE,KAAK,IAAI,IAAI,EAAE;AACvD,QAAI,QAAQ,gBAAiB;AAC7B,QAAI,KAAK;AAAA,MACP;AAAA,MACA,SAAS,YAAY,EAAE,EAAE;AAAA,MACzB;AAAA,MACA,UAAU,sBAAsB,EAAE,KAAK,2BAA2B,MAAM,QAAQ,CAAC,CAAC;AAAA,MAClF,SAAS;AAAA,IACX,CAAC;AAAA,EACH;AACA,SAAO;AACT;AAEA,SAAS,mBACP,KACA,UACA,WACa;AACb,QAAM,MAAmB,CAAC;AAC1B,YAAU,QAAQ,CAAC,MAAM,MAAM;AAC7B,UAAM,QAAQ,YAAY,IAAI,OAAO,IAAI;AACzC,QAAI,QAAQ,gBAAiB;AAC7B,QAAI,KAAK;AAAA,MACP;AAAA,MACA,SAAS,QAAQ,CAAC;AAAA,MAClB;AAAA,MACA,UAAU,cAAc,IAAI,2BAA2B,MAAM,QAAQ,CAAC,CAAC;AAAA,MACvE,SAAS;AAAA,IACX,CAAC;AAAA,EACH,CAAC;AACD,SAAO;AACT;AASA,eAAsB,iBACpB,MACA,OACA,kBAC4B;AAC5B,MAAI,KAAK,aAAa,WAAW,GAAG;AAClC,UAAM,IAAI;AAAA,MACR,2BAA2B,KAAK,MAAM;AAAA,IACxC;AAAA,EACF;AAKA,QAAM,aAA0B,CAAC;AACjC,OAAK,aAAa,QAAQ,CAAC,KAAK,MAAM;AACpC,UAAM,KAAK,IAAI,eAAe;AAC9B,QAAI,OAAO,cAAc,OAAO,OAAO;AACrC,iBAAW,KAAK,GAAG,mBAAmB,KAAK,GAAG,MAAM,SAAS,CAAC;AAAA,IAChE;AACA,QAAI,OAAO,cAAc,OAAO,OAAO;AACrC,iBAAW,KAAK,GAAG,mBAAmB,KAAK,GAAG,MAAM,SAAS,CAAC;AAAA,IAChE;AACA,QAAI,OAAO,eAAe,OAAO,OAAO;AACtC,iBAAW,KAAK,GAAG,mBAAmB,KAAK,GAAG,MAAM,SAAS,CAAC;AAAA,IAChE;AAAA,EACF,CAAC;AACD,aAAW,KAAK,CAAC,GAAG,MAAM,EAAE,QAAQ,EAAE,KAAK;AAE3C,QAAM,WAAW,oBAAI,IAAuB;AAC5C,QAAM,YAAY,oBAAI,IAAY;AAClC,aAAW,KAAK,YAAY;AAC1B,QAAI,SAAS,IAAI,EAAE,QAAQ,KAAK,UAAU,IAAI,EAAE,OAAO,EAAG;AAC1D,aAAS,IAAI,EAAE,UAAU,CAAC;AAC1B,cAAU,IAAI,EAAE,OAAO;AAAA,EACzB;AAEA,QAAM,eAAmC,CAAC;AAC1C,WAAS,IAAI,GAAG,IAAI,KAAK,aAAa,QAAQ,KAAK;AACjD,UAAM,MAAM,KAAK,aAAa,CAAC;AAC/B,UAAM,QAAQ,SAAS,IAAI,CAAC;AAC5B,UAAM,WAAqB,CAAC;AAC5B,QAAI,UAA0B;AAE9B,QAAI,OAAO;AACT,eAAS,KAAK,MAAM,QAAQ;AAC5B,UAAI,MAAM,YAAY,MAAM;AAC1B,cAAM,IAAI,MAAM,iBAAiB,KAAK,MAAM,OAAO;AACnD,kBAAU,EAAE;AACZ,iBAAS,KAAK,gBAAgB,EAAE,UAAU,SAAS,MAAM,WAAM,EAAE,MAAM,EAAE;AAAA,MAC3E,OAAO;AACL,iBAAS,KAAK,kEAA6D;AAAA,MAC7E;AAAA,IACF,OAAO;AACL,YAAM,KAAK,IAAI,eAAe;AAC9B,YAAM,OAAO,OAAO,QAAQ,gCAAgC;AAC5D,eAAS,KAAK,eAAe,IAAI,2BAA2B;AAAA,IAC9D;AAEA,UAAM,sBAAsB,UAAU;AACtC,UAAM,YAAY,uBAAuB,YAAY;AACrD,iBAAa,KAAK;AAAA,MAChB,OAAO,IAAI;AAAA,MACX,OAAO,IAAI;AAAA,MACX;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,IACF,CAAC;AAAA,EACH;AAEA,SAAO,kBAAkB,EAAE,QAAQ,KAAK,QAAQ,aAAa,CAAC;AAChE;AASO,SAAS,yBAAyB,KAAmD;AAC1F,QAAM,QAAQ,IAAI,MAAM,aAAa;AACrC,MAAI,CAAC,OAAO;AACV,UAAM,IAAI,MAAM,0DAA0D,IAAI,MAAM,GAAG,GAAG,CAAC,EAAE;AAAA,EAC/F;AACA,QAAM,SAAS,KAAK,MAAM,MAAM,CAAC,CAAC;AAClC,MAAI,OAAO,OAAO,YAAY,WAAW;AACvC,UAAM,IAAI,MAAM,uDAAuD,MAAM,CAAC,EAAE,MAAM,GAAG,GAAG,CAAC,EAAE;AAAA,EACjG;AACA,SAAO,EAAE,SAAS,OAAO,SAAS,QAAQ,OAAO,OAAO,WAAW,WAAW,OAAO,SAAS,GAAG;AACnG;AAQO,SAAS,4BACd,IACA,OAAkC,CAAC,GACf;AACpB,QAAM,QAAQ,KAAK,SAAS;AAC5B,QAAM,kBAAkB,KAAK,mBAAmB;AAChD,SAAO,OAAO,aAAa,YAAY;AACrC,UAAM,OAAO,MAAM,GAAG,KAAK;AAAA,MACzB;AAAA,MACA,UAAU;AAAA,QACR;AAAA,UACE,MAAM;AAAA,UACN,SACE;AAAA,QACJ;AAAA,QACA;AAAA,UACE,MAAM;AAAA,UACN,SAAS,gBAAgB,YAAY,KAAK;AAAA,EACxC,YAAY,WAAW,aAAa,YAAY,QAAQ;AAAA,IAAO,EACjE;AAAA;AAAA,EAAyB,QAAQ,MAAM,GAAG,eAAe,CAAC;AAAA,QAC5D;AAAA,MACF;AAAA,MACA,aAAa;AAAA,MACb,WAAW;AAAA,IACb,CAAC;AACD,UAAM,MACH,KAA4D,UAAU,CAAC,GAAG,SAAS,WACpF;AACF,WAAO,yBAAyB,GAAG;AAAA,EACrC;AACF;AAKA,IAAM,kBAAkB,oBAAI,IAAI;AAAA,EAC9B;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AACF,CAAC;AAgBM,SAAS,yBACd,OAA0D,CAAC,GACvC;AACpB,QAAM,YAAY,KAAK,aAAa;AACpC,QAAM,SAAS,KAAK,oBAAoB;AACxC,SAAO,OAAO,aAAa,YAAY;AACrC,UAAM,OAAO,QAAQ,KAAK;AAC1B,QAAI,KAAK,SAAS;AAChB,aAAO;AAAA,QACL,SAAS;AAAA,QACT,QAAQ,qBAAqB,KAAK,MAAM;AAAA,MAC1C;AACF,UAAM,cAAc,YAAY,MAC7B,YAAY,EACZ,MAAM,YAAY,EAClB,OAAO,CAAC,MAAM,EAAE,SAAS,KAAK,CAAC,gBAAgB,IAAI,CAAC,CAAC;AACxD,QAAI,YAAY,WAAW;AACzB,aAAO;AAAA,QACL,SAAS;AAAA,QACT,QAAQ;AAAA,MACV;AACF,UAAM,QAAQ,KAAK,YAAY;AAC/B,UAAM,OAAO,YAAY,OAAO,CAAC,MAAM,MAAM,SAAS,CAAC,CAAC,EAAE;AAC1D,UAAM,SAAS,OAAO,YAAY;AAClC,WAAO,UAAU,YACb;AAAA,MACE,SAAS;AAAA,MACT,QAAQ,mBAAmB,IAAI,IAAI,YAAY,MAAM;AAAA,IACvD,IACA;AAAA,MACE,SAAS;AAAA,MACT,QAAQ,wBAAwB,IAAI,IAAI,YAAY,MAAM;AAAA,IAC5D;AAAA,EACN;AACF;;;AC/bA,SAAS,aAAa,UAAsC;AAC1D,MAAI,CAAC,SAAU,QAAO;AACtB,MAAI,SAAS,SAAS,MAAM,EAAG,QAAO;AACtC,MAAI,SAAS,WAAW,OAAO,EAAG,QAAO;AACzC,SAAO;AACT;AAYO,SAAS,qBAAqB,QAAoD;AACvF,QAAM,YAAwB,CAAC;AAC/B,QAAM,YAAgC,CAAC;AACvC,QAAM,YAAsB,CAAC;AAC7B,QAAM,YAAY,oBAAI,IAAY;AAElC,aAAW,MAAM,QAAQ;AACvB,QAAI,GAAG,SAAS,aAAa;AAC3B,YAAM,OAAQ,GAAyB;AACvC,UAAI,QAAQ,CAAC,UAAU,IAAI,IAAI,GAAG;AAChC,kBAAU,IAAI,IAAI;AAClB,kBAAU,KAAK,IAAI;AAAA,MACrB;AAAA,IACF,WAAW,GAAG,SAAS,YAAY;AACjC,YAAM,IAAI;AACV,gBAAU,KAAK;AAAA,QACb,MAAM,aAAa,EAAE,QAAQ;AAAA,QAC7B,MAAM,EAAE,QAAQ,EAAE,OAAO,EAAE;AAAA,QAC3B,SAAS,EAAE,WAAW;AAAA,MACxB,CAAC;AAAA,IACH,WAAW,GAAG,SAAS,oBAAoB;AACzC,YAAM,IAAI;AACV,gBAAU,KAAK;AAAA,QACb,IAAI,EAAE;AAAA,QACN,OAAO,EAAE;AAAA,QACT,QAAQ,EAAE,UAAU;AAAA,QACpB,GAAI,EAAE,YAAY,SAAY,EAAE,SAAS,EAAE,QAAQ,IAAI,CAAC;AAAA,MAC1D,CAAC;AAAA,IACH;AAAA,EACF;AAEA,SAAO,EAAE,WAAW,WAAW,UAAU;AAC3C;;;AC7FA,SAAS,kBAAkB;AAyBpB,SAAS,iBAAiB,SAA+B;AAC9D,MAAI,OAAO,QAAQ,UAAU,YAAY,QAAQ,MAAM,KAAK,EAAE,WAAW,GAAG;AAC1E,UAAM,IAAI,gBAAgB,iBAAiB,QAAQ,EAAE,mCAA8B;AAAA,EACrF;AACA,QAAM,YAAY;AAAA,IAChB,OAAO,QAAQ,MAAM,KAAK;AAAA,IAC1B,QAAQ,CAAC,GAAI,QAAQ,UAAU,CAAC,CAAE,EAAE,KAAK;AAAA,IACzC,eAAe,QAAQ,iBAAiB;AAAA,IACxC,OAAO,CAAC,GAAI,QAAQ,SAAS,CAAC,CAAE,EAAE,KAAK;AAAA,IACvC,UAAU,QAAQ,YAAY,CAAC;AAAA,EACjC;AACA,SAAO,WAAW,QAAQ,EACvB,OAAO,KAAK,UAAU,aAAa,SAAS,CAAC,CAAC,EAC9C,OAAO,KAAK;AACjB;","names":[]}

package/dist/fuzz.d.ts

@@ -52,6 +52,9 @@ interface Evaluation extends DefaultVerdict {
     runId?: string;
     /** Structured labels, e.g. failure classes (`hallucination`, `refusal`). */
     labels?: string[];
+    /** Wall-clock for the evaluation, when the consumer measures it more precisely
+     *  than the engine can (e.g. excluding judge time). Engine-measured otherwise. */
+    latencyMs?: number;
 }
 /** Run the target against one scenario in a cell. */
 type Evaluator<S> = (scenario: S, cell: Cell) => Promise<Evaluation>;
@@ -123,16 +126,34 @@ interface ArchiveEntry<S> {
     evaluation: Evaluation;
     interest: number;
 }
+/** Summary of a sample — every aggregate carries its spread, never a bare mean. */
+interface Distribution {
+    mean: number;
+    median: number;
+    p90: number;
+    min: number;
+    max: number;
+    n: number;
+}
 /** Per-INPUT-cell coverage — the planned-vs-covered map. */
 interface CoverageCell {
     cell: Cell;
     runs: number;
-    /** Mean headline score in [0,1]; `null` when the cell was never run (honestly uncovered). */
-    robustness: number | null;
+    /** Headline score distribution in [0,1]; `null` when the cell was never run
+     *  (honestly uncovered — never a fabricated zero). */
+    score: Distribution | null;
     /** Fraction of runs the objective flagged as notable. */
     findingRate: number;
-    /** Mean per-dimension scores — surfaces WHICH dimension is weak. */
-    dimensions: Record<string, number>;
+    /** Per-dimension score distributions — surfaces WHICH dimension is weak and
+     *  how consistently. */
+    dimensions: Record<string, Distribution>;
+    /** Evaluation wall-clock per run; engine-measured unless the evaluation
+     *  carried its own `latencyMs`. `null` when the cell was never run. */
+    latencyMs: Distribution | null;
+    /** Known dollars spent in this cell — present only when cost tracking was
+     *  wired; runs with unknown cost are counted apart, never folded in as $0. */
+    costUsd?: number;
+    costUnknownRuns?: number;
 }
 /** The artifact every exploration produces. */
 interface CapsuleData<S> {
@@ -160,7 +181,12 @@ interface CapsuleData<S> {
         behaviorBinsObserved: number;
         candidateFindings: number;
         verifiedFindings: number;
-        meanRobustness: number;
+        /** Distribution of per-cell mean scores across covered cells (cells weigh
+         *  equally — variance steering sends more runs to weak cells, so a
+         *  run-weighted average would bias low). `null` when nothing ran. */
+        robustness: Distribution | null;
+        /** Evaluation wall-clock across all runs. `null` when nothing ran. */
+        latencyMs: Distribution | null;
         /** Known dollars spent on this exploration's runs. Present only when cost
          *  tracking was wired (`costOf`) — absent means "not tracked", never $0. */
         costUsd?: number;
@@ -285,9 +311,10 @@ interface ExploreOptions<S> {
  *
  * Cells are the cartesian product of the input axes — the stratification plan,
  * enumerable up front so the planned-vs-covered denominator is honest. Coverage
- * is projected from the evaluation log: per cell, mean headline robustness, the
- * mean of each scored dimension (so the map shows WHICH dimension is weak), and
- * the rate at which the active objective flagged a candidate.
+ * is projected from the evaluation log: per cell, the full DISTRIBUTION of the
+ * headline score, of each scored dimension, and of evaluation latency — a bare
+ * mean hides outliers, so every aggregate carries its spread. Per-cell cost is
+ * split known-dollars vs unknown-runs, never folded into a fabricated $0.
  */
 
 /** One recorded evaluation — the unit coverage and the capsule are built from. */
@@ -296,6 +323,11 @@ interface EvalRecord {
     ev: Evaluation;
     /** The objective's interest score for this evaluation. */
     interest: number;
+    /** Evaluation wall-clock — engine-measured unless `ev.latencyMs` overrode it. */
+    latencyMs: number;
+    /** Known dollars for this run. `null` = cost tracking was wired but this
+     *  run's cost was unknowable (counted apart). Absent = not tracked at all. */
+    costUsd?: number | null;
 }
 /** Enumerate every input cell (cartesian product of the axes), in stable order. */
 declare function enumerateCells(space: BehaviorSpace): Cell[];
@@ -303,7 +335,7 @@ declare function enumerateCells(space: BehaviorSpace): Cell[];
 declare function cellId(space: BehaviorSpace, coords: Record<string, string>): string;
 /**
  * Project the evaluation log into the per-input-cell coverage map. A cell with
- * no evaluations reports `robustness: null` (honestly uncovered), never 0.
+ * no evaluations reports `score: null` (honestly uncovered), never zeros.
  */
 declare function buildCoverage(cells: Cell[], log: EvalRecord[], threshold: number): CoverageCell[];
 
@@ -398,6 +430,8 @@ declare class BehaviorExplorer<S> {
     private costExhausted;
     /** Fold one run's cost in: null counts as unknown (never $0); a known cost
      *  accrues toward the budget, lands in the ledger, and fires `onCost`. */
+    /** Returns the run's known cost, `null` when tracked-but-unknown, `undefined`
+     *  when cost tracking is not wired — the log row mirrors this exactly. */
     private recordRunCost;
     /** Elites whose INPUT cell matches — what the proposer mutates/deepens from. */
     private elitesFor;

package/dist/fuzz.js

@@ -22,7 +22,24 @@ function enumerateCells(space) {
 function cellId(space, coords) {
   return space.axes.map((a) => `${a.name}=${coords[a.name]}`).join("|");
 }
-var mean = (xs) => xs.length === 0 ? 0 : xs.reduce((a, b) => a + b, 0) / xs.length;
+function percentile(sorted, p) {
+  const idx = Math.min(sorted.length - 1, Math.max(0, Math.ceil(p * sorted.length) - 1));
+  return sorted[idx];
+}
+function distribution(values) {
+  if (values.length === 0)
+    throw new Error("distribution: empty sample \u2014 represent missing data as null, not zeros");
+  const sorted = [...values].sort((a, b) => a - b);
+  const mean = sorted.reduce((a, b) => a + b, 0) / sorted.length;
+  return {
+    mean,
+    median: percentile(sorted, 0.5),
+    p90: percentile(sorted, 0.9),
+    min: sorted[0],
+    max: sorted[sorted.length - 1],
+    n: sorted.length
+  };
+}
 function buildCoverage(cells, log, threshold) {
   const byCell = /* @__PURE__ */ new Map();
   for (const r of log) {
@@ -33,19 +50,27 @@ function buildCoverage(cells, log, threshold) {
   return cells.map((cell) => {
     const recs = byCell.get(cell.id) ?? [];
     const runs = recs.length;
-    if (runs === 0) return { cell, runs: 0, robustness: null, findingRate: 0, dimensions: {} };
-    const robustness = mean(recs.map((r) => r.ev.score));
+    if (runs === 0)
+      return { cell, runs: 0, score: null, findingRate: 0, dimensions: {}, latencyMs: null };
+    const score = distribution(recs.map((r) => r.ev.score));
+    const latencyMs = distribution(recs.map((r) => r.latencyMs));
     const findingRate = recs.filter((r) => r.interest >= threshold).length / runs;
-    const dims = {};
+    const dimSamples = {};
     for (const r of recs) {
       for (const [k, v] of Object.entries(r.ev.scores ?? {})) {
         ;
-        (dims[k] ??= []).push(v);
+        (dimSamples[k] ??= []).push(v);
       }
     }
     const dimensions = {};
-    for (const [k, xs] of Object.entries(dims)) dimensions[k] = mean(xs);
-    return { cell, runs, robustness, findingRate, dimensions };
+    for (const [k, xs] of Object.entries(dimSamples)) dimensions[k] = distribution(xs);
+    const tracked = recs.filter((r) => r.costUsd !== void 0);
+    const known = tracked.filter((r) => r.costUsd !== null);
+    const cost = tracked.length > 0 ? {
+      costUsd: known.reduce((a, r) => a + r.costUsd, 0),
+      ...tracked.length > known.length ? { costUnknownRuns: tracked.length - known.length } : {}
+    } : {};
+    return { cell, runs, score, findingRate, dimensions, latencyMs, ...cost };
   });
 }
 
@@ -53,7 +78,8 @@ function buildCoverage(cells, log, threshold) {
 function buildCapsule(input) {
   const coverage = buildCoverage(input.cells, input.log, input.threshold);
   const covered = coverage.filter((c) => c.runs > 0);
-  const meanRobustness = covered.length === 0 ? 0 : covered.reduce((a, c) => a + (c.robustness ?? 0), 0) / covered.length;
+  const robustness = covered.length === 0 ? null : distribution(covered.map((c) => c.score.mean));
+  const latencyMs = input.log.length === 0 ? null : distribution(input.log.map((r) => r.latencyMs));
   const behaviorBinsObserved = input.archive.filter((e) => e.binId !== e.cell.id).length;
   return {
     target: input.target,
@@ -68,7 +94,8 @@ function buildCapsule(input) {
       behaviorBinsObserved,
       candidateFindings: input.candidateFindings,
       verifiedFindings: input.findings.length,
-      meanRobustness,
+      robustness,
+      latencyMs,
       ...input.cost ? { costUsd: input.cost.costUsd, costUnknownRuns: input.cost.costUnknownRuns } : {},
       evalErrors: input.evalErrors,
       ...input.stoppedEarly ? { stoppedEarly: input.stoppedEarly } : {}
@@ -103,7 +130,7 @@ function deriveAxes(coverage) {
   return order.map((name) => ({ name, values: [...seen.get(name) ?? []] }));
 }
 function weakestDim(c) {
-  const entries = Object.entries(c.dimensions);
+  const entries = Object.entries(c.dimensions).map(([k, d]) => [k, d.mean]);
   if (entries.length === 0) return "";
   const sorted = entries.sort((a, b) => a[1] - b[1]);
   const w = sorted[0];
@@ -114,8 +141,8 @@ function heatmapHtml(coverage) {
   const axes = deriveAxes(coverage);
   const byId = new Map(coverage.map((c) => [c.cell.id, c]));
   const tile = (c, label) => {
-    const r = c?.robustness ?? null;
-    const title = c ? `${pct(r ?? 0)} robust \xB7 ${c.runs} runs \xB7 ${pct(c.findingRate)} flagged` : "not covered";
+    const r = c?.score?.mean ?? null;
+    const title = c?.score ? `${pct(c.score.mean)} robust (median ${pct(c.score.median)}, min ${pct(c.score.min)}) \xB7 ${c.runs} runs \xB7 ${pct(c.findingRate)} flagged${c.latencyMs ? ` \xB7 ${(c.latencyMs.median / 1e3).toFixed(1)}s median` : ""}${c.costUsd !== void 0 ? ` \xB7 $${c.costUsd.toFixed(2)}` : ""}` : "not covered";
     return `<div class="tile" style="background:${robustnessColor(r)}" title="${esc(title)}">${label ? `<span class="tl">${esc(label)}</span>` : ""}<span class="tv">${c && r != null ? pct(r) : "\u2014"}</span>${c ? weakestDim(c) : ""}</div>`;
   };
   const rowAxis = axes[0];
@@ -131,7 +158,7 @@ function heatmapHtml(coverage) {
     }).join("");
     return `<div class="axis-label">rows: <b>${esc(rowAxis.name)}</b> \xB7 cols: <b>${esc(colAxis.name)}</b></div><table class="heat">${head}${rows}</table>`;
   }
-  const sorted = [...coverage].sort((a, b) => (a.robustness ?? 2) - (b.robustness ?? 2));
+  const sorted = [...coverage].sort((a, b) => (a.score?.mean ?? 2) - (b.score?.mean ?? 2));
   return `<div class="grid">${sorted.map((c) => tile(c, Object.values(c.cell.coords).join(" \xB7 "))).join("")}</div>`;
 }
 function findingsHtml(findings, limit) {
@@ -198,7 +225,9 @@ table.heat th.rh{text-align:right}
 <h1>${esc(capsule.objective)} exploration \xB7 ${esc(capsule.target)}</h1>
 <div class="sub">${s.totalRuns} scenarios across ${s.cellsCovered}/${s.cellsTotal} planned cells${s.behaviorBinsObserved > 0 ? ` \xB7 ${s.behaviorBinsObserved} measured behavior bins` : ""}${stamp ? ` \xB7 ${esc(stamp)}` : ""}</div>
 <div class="kpis">
-${kpi("mean robustness", pct(s.meanRobustness), s.meanRobustness < 0.6 ? "#e58a96" : "#5ad17a")}
+${s.robustness ? kpi("robustness", `${pct(s.robustness.mean)}`, s.robustness.mean < 0.6 ? "#e58a96" : "#5ad17a") : ""}
+${s.robustness ? kpi("cell spread", `${pct(s.robustness.min)}\u2013${pct(s.robustness.max)}`) : ""}
+${s.latencyMs ? kpi("median latency", `${(s.latencyMs.median / 1e3).toFixed(1)}s`, s.latencyMs.p90 > 4 * s.latencyMs.median ? "#e5b566" : "#e6e6e6") : ""}
 ${kpi("verified findings", String(s.verifiedFindings), s.verifiedFindings > 0 ? "#e58a96" : "#5ad17a")}
 ${kpi("cells covered", `${s.cellsCovered}/${s.cellsTotal}`)}
 ${kpi("scenarios run", String(s.totalRuns))}
@@ -370,12 +399,14 @@ var BehaviorExplorer = class {
   }
   /** Fold one run's cost in: null counts as unknown (never $0); a known cost
    *  accrues toward the budget, lands in the ledger, and fires `onCost`. */
+  /** Returns the run's known cost, `null` when tracked-but-unknown, `undefined`
+   *  when cost tracking is not wired — the log row mirrors this exactly. */
   recordRunCost(scenario, cell, ev) {
-    if (!this.opts.costOf) return;
+    if (!this.opts.costOf) return void 0;
     const cost = this.opts.costOf(scenario, cell, ev);
     if (cost === null) {
       this.costUnknownRuns++;
-      return;
+      return null;
     }
     if (typeof cost.usd !== "number" || !Number.isFinite(cost.usd) || cost.usd < 0) {
       throw new RangeError(
@@ -391,6 +422,7 @@ var BehaviorExplorer = class {
       tags: { target: this.opts.target, cell: cell.id }
     });
     this.opts.onCost?.({ usd: cost.usd, channel: "agent" });
+    return cost.usd;
   }
   /** Elites whose INPUT cell matches — what the proposer mutates/deepens from. */
   elitesFor(cellId2) {
@@ -432,13 +464,22 @@ var BehaviorExplorer = class {
           if (this.runsUsed >= this.opts.budget || this.costExhausted() || this.stoppedEarly !== void 0 || this.opts.signal?.aborted)
             return;
           try {
+            const startedAt = performance.now();
             const ev = await this.opts.evaluate(scenario, cell);
+            const latencyMs = ev.latencyMs ?? performance.now() - startedAt;
             this.runsUsed++;
             runsThisStep++;
             this.consecutiveEvalErrors = 0;
-            this.recordRunCost(scenario, cell, ev);
+            const costUsd = this.recordRunCost(scenario, cell, ev);
             const interest = this.objective.interest(ev, this.objectiveContext());
-            this.log.push({ cell, ev, interest, scenarioId: this.opts.scenarioId(scenario) });
+            this.log.push({
+              cell,
+              ev,
+              interest,
+              latencyMs,
+              ...costUsd !== void 0 ? { costUsd } : {},
+              scenarioId: this.opts.scenarioId(scenario)
+            });
             this.opts.onProgress?.({ type: "evaluated", cell, scenario, evaluation: ev });
             const bin = this.binId(cell, ev.descriptor);
             const cur = this.archiveByBin.get(bin);

package/dist/fuzz.js.map

@@ -1 +1 @@
-{"version":3,"sources":["../src/fuzz/cube.ts","../src/fuzz/capsule.ts","../src/fuzz/policies.ts","../src/fuzz/explorer.ts","../src/fuzz/fuzz-agent.ts","../src/fuzz/gates.ts","../src/fuzz/tools.ts"],"sourcesContent":["/**\n * Input-space tiling + coverage projection.\n *\n * Cells are the cartesian product of the input axes — the stratification plan,\n * enumerable up front so the planned-vs-covered denominator is honest. Coverage\n * is projected from the evaluation log: per cell, mean headline robustness, the\n * mean of each scored dimension (so the map shows WHICH dimension is weak), and\n * the rate at which the active objective flagged a candidate.\n */\n\nimport type { BehaviorSpace, Cell, CoverageCell, Evaluation } from './types'\n\n/** One recorded evaluation — the unit coverage and the capsule are built from. */\nexport interface EvalRecord {\n  cell: Cell\n  ev: Evaluation\n  /** The objective's interest score for this evaluation. */\n  interest: number\n}\n\n/** Enumerate every input cell (cartesian product of the axes), in stable order. */\nexport function enumerateCells(space: BehaviorSpace): Cell[] {\n  if (space.axes.length === 0) return []\n  let partials: Array<Record<string, string>> = [{}]\n  for (const axis of space.axes) {\n    const next: Array<Record<string, string>> = []\n    for (const partial of partials) {\n      for (const value of axis.values) next.push({ ...partial, [axis.name]: value })\n    }\n    partials = next\n  }\n  return partials.map((coords) => ({ id: cellId(space, coords), coords }))\n}\n\n/** Deterministic id for a coordinate map, e.g. `matterType=nda|difficulty=hard`. */\nexport function cellId(space: BehaviorSpace, coords: Record<string, string>): string {\n  return space.axes.map((a) => `${a.name}=${coords[a.name]}`).join('|')\n}\n\nconst mean = (xs: number[]): number =>\n  xs.length === 0 ? 0 : xs.reduce((a, b) => a + b, 0) / xs.length\n\n/**\n * Project the evaluation log into the per-input-cell coverage map. A cell with\n * no evaluations reports `robustness: null` (honestly uncovered), never 0.\n */\nexport function buildCoverage(cells: Cell[], log: EvalRecord[], threshold: number): CoverageCell[] {\n  const byCell = new Map<string, EvalRecord[]>()\n  for (const r of log) {\n    const arr = byCell.get(r.cell.id) ?? []\n    arr.push(r)\n    byCell.set(r.cell.id, arr)\n  }\n  return cells.map((cell) => {\n    const recs = byCell.get(cell.id) ?? []\n    const runs = recs.length\n    if (runs === 0) return { cell, runs: 0, robustness: null, findingRate: 0, dimensions: {} }\n    const robustness = mean(recs.map((r) => r.ev.score))\n    const findingRate = recs.filter((r) => r.interest >= threshold).length / runs\n    const dims: Record<string, number[]> = {}\n    for (const r of recs) {\n      for (const [k, v] of Object.entries(r.ev.scores ?? {})) {\n        ;(dims[k] ??= []).push(v)\n      }\n    }\n    const dimensions: Record<string, number> = {}\n    for (const [k, xs] of Object.entries(dims)) dimensions[k] = mean(xs)\n    return { cell, runs, robustness, findingRate, dimensions }\n  })\n}\n","/**\n * The capsule — the artifact every exploration produces.\n *\n * `buildCapsule` assembles coverage + verified findings + the QD archive into a\n * pure `CapsuleData` (no clock, no I/O — deterministic and snapshot-testable).\n * `renderCapsuleHtml` turns it into a standalone page: the input-cell heat-map\n * (planned vs covered), per-dimension weakness chips, and the minimized finding\n * exemplars. One artifact — the hardening map and the shareable proof object.\n */\n\nimport type { EvalRecord } from './cube'\nimport { buildCoverage } from './cube'\nimport type { ArchiveEntry, CapsuleData, Cell, CoverageCell, Finding } from './types'\n\nexport interface BuildCapsuleInput<S> {\n  target: string\n  objective: string\n  cells: Cell[]\n  log: EvalRecord[]\n  /** The objective's notable threshold — drives findingRate. */\n  threshold: number\n  archive: ArchiveEntry<S>[]\n  findings: Finding<S>[]\n  candidateFindings: number\n  runsUsed: number\n  /** Known-dollar / unknown-run split — present only when cost tracking was\n   *  wired; the capsule never fabricates a $0 total. */\n  cost?: { costUsd: number; costUnknownRuns: number }\n  /** Evaluations that threw — infra outcomes, never folded into robustness. */\n  evalErrors: number\n  /** Set when the consecutive-error circuit breaker stopped the run early. */\n  stoppedEarly?: { reason: 'eval-errors'; detail: string }\n}\n\nexport function buildCapsule<S>(input: BuildCapsuleInput<S>): CapsuleData<S> {\n  const coverage = buildCoverage(input.cells, input.log, input.threshold)\n  const covered = coverage.filter((c) => c.runs > 0)\n  const meanRobustness =\n    covered.length === 0 ? 0 : covered.reduce((a, c) => a + (c.robustness ?? 0), 0) / covered.length\n  // Measured-descriptor bins beyond the bare input cell — observed, never planned.\n  const behaviorBinsObserved = input.archive.filter((e) => e.binId !== e.cell.id).length\n\n  return {\n    target: input.target,\n    objective: input.objective,\n    coverage,\n    findings: [...input.findings].sort((a, b) => b.interest - a.interest),\n    archive: [...input.archive].sort((a, b) => b.interest - a.interest),\n    stats: {\n      totalRuns: input.runsUsed,\n      cellsTotal: input.cells.length,\n      cellsCovered: covered.length,\n      behaviorBinsObserved,\n      candidateFindings: input.candidateFindings,\n      verifiedFindings: input.findings.length,\n      meanRobustness,\n      ...(input.cost\n        ? { costUsd: input.cost.costUsd, costUnknownRuns: input.cost.costUnknownRuns }\n        : {}),\n      evalErrors: input.evalErrors,\n      ...(input.stoppedEarly ? { stoppedEarly: input.stoppedEarly } : {}),\n    },\n  }\n}\n\n// ── HTML capsule ──────────────────────────────────────────────────────────────\n\nfunction esc(s: string): string {\n  return s.replace(\n    /[&<>\"]/g,\n    (c) => ({ '&': '&amp;', '<': '&lt;', '>': '&gt;', '\"': '&quot;' })[c] as string,\n  )\n}\n\n/** red (0) → amber (.5) → green (1); uncovered cells render gray. */\nfunction robustnessColor(r: number | null): string {\n  if (r == null) return '#2a2a2e'\n  return `hsl(${Math.round(r * 120)} 70% 42%)`\n}\n\nfunction pct(x: number): string {\n  return `${Math.round(x * 100)}%`\n}\n\nfunction deriveAxes(coverage: CoverageCell[]): Array<{ name: string; values: string[] }> {\n  const order: string[] = []\n  const seen = new Map<string, Set<string>>()\n  for (const c of coverage) {\n    for (const [k, v] of Object.entries(c.cell.coords)) {\n      if (!seen.has(k)) {\n        seen.set(k, new Set())\n        order.push(k)\n      }\n      seen.get(k)?.add(v)\n    }\n  }\n  return order.map((name) => ({ name, values: [...(seen.get(name) ?? [])] }))\n}\n\n/** The weakest dimension chip for a cell, e.g. `safety 32%` — shown when scores exist. */\nfunction weakestDim(c: CoverageCell): string {\n  const entries = Object.entries(c.dimensions)\n  if (entries.length === 0) return ''\n  const sorted = entries.sort((a, b) => a[1] - b[1])\n  const w = sorted[0]\n  if (!w) return ''\n  return `<span class=\"dim\">${esc(w[0])} ${pct(w[1])}</span>`\n}\n\nfunction heatmapHtml(coverage: CoverageCell[]): string {\n  const axes = deriveAxes(coverage)\n  const byId = new Map(coverage.map((c) => [c.cell.id, c]))\n  const tile = (c: CoverageCell | undefined, label: string): string => {\n    const r = c?.robustness ?? null\n    const title = c\n      ? `${pct(r ?? 0)} robust · ${c.runs} runs · ${pct(c.findingRate)} flagged`\n      : 'not covered'\n    return `<div class=\"tile\" style=\"background:${robustnessColor(r)}\" title=\"${esc(title)}\">${label ? `<span class=\"tl\">${esc(label)}</span>` : ''}<span class=\"tv\">${c && r != null ? pct(r) : '—'}</span>${c ? weakestDim(c) : ''}</div>`\n  }\n\n  const rowAxis = axes[0]\n  const colAxis = axes[1]\n  if (axes.length === 2 && rowAxis && colAxis) {\n    const head = `<tr><th></th>${colAxis.values.map((v) => `<th>${esc(v)}</th>`).join('')}</tr>`\n    const rows = rowAxis.values\n      .map((rv) => {\n        const cells = colAxis.values\n          .map((cv) => {\n            const id = `${rowAxis.name}=${rv}|${colAxis.name}=${cv}`\n            return `<td>${tile(byId.get(id), '')}</td>`\n          })\n          .join('')\n        return `<tr><th class=\"rh\">${esc(rv)}</th>${cells}</tr>`\n      })\n      .join('')\n    return `<div class=\"axis-label\">rows: <b>${esc(rowAxis.name)}</b> · cols: <b>${esc(colAxis.name)}</b></div><table class=\"heat\">${head}${rows}</table>`\n  }\n\n  const sorted = [...coverage].sort((a, b) => (a.robustness ?? 2) - (b.robustness ?? 2))\n  return `<div class=\"grid\">${sorted.map((c) => tile(c, Object.values(c.cell.coords).join(' · '))).join('')}</div>`\n}\n\nfunction findingsHtml<S>(findings: Finding<S>[], limit: number): string {\n  if (findings.length === 0)\n    return `<p class=\"none\">No verified findings — the target held across every covered cell.</p>`\n  return findings\n    .slice(0, limit)\n    .map((f) => {\n      const coords = Object.entries(f.cell.coords)\n        .map(([k, v]) => `${k}:${v}`)\n        .join(' · ')\n      const labels = (f.evaluation.labels ?? [])\n        .map((l) => `<span class=\"fclass\">${esc(l)}</span>`)\n        .join('')\n      const dims = Object.entries(f.evaluation.scores ?? {})\n        .sort((a, b) => a[1] - b[1])\n        .slice(0, 3)\n        .map(([k, v]) => `<span class=\"dim\">${esc(k)} ${pct(v)}</span>`)\n        .join('')\n      return `<div class=\"fail\"><div class=\"fmeta\"><span class=\"fcell\">${esc(coords)}</span>${labels}<span class=\"fsev\">interest ${pct(f.interest)}</span></div><div class=\"sevbar\"><div class=\"sevfill\" style=\"width:${pct(f.interest)}\"></div></div><div class=\"ftext\">${esc(f.text ?? '(scenario text not captured)')}</div>${dims ? `<div class=\"fdims\">${dims}</div>` : ''}</div>`\n    })\n    .join('')\n}\n\nexport interface RenderCapsuleOptions {\n  /** Max finding exemplars to show. Default 8. */\n  maxFindings?: number\n  /** ISO timestamp to stamp into the page (keeps the pure capsule clock-free). */\n  generatedAt?: string\n}\n\n/** Render a self-contained HTML capsule — heat-map + per-dimension chips + verified findings. */\nexport function renderCapsuleHtml<S>(\n  capsule: CapsuleData<S>,\n  opts: RenderCapsuleOptions = {},\n): string {\n  const s = capsule.stats\n  const kpi = (label: string, value: string, accent = '#e6e6e6'): string =>\n    `<div class=\"kpi\"><div class=\"kv\" style=\"color:${accent}\">${esc(value)}</div><div class=\"kl\">${esc(label)}</div></div>`\n  const lift = capsule.lift\n    ? kpi(\n        'hardening lift',\n        `${capsule.lift.before.toFixed(2)} → ${capsule.lift.after.toFixed(2)}`,\n        '#5ad17a',\n      )\n    : ''\n  // Cost KPI only when tracking was wired — an untracked run never shows $0.\n  // Unpriced runs are named in the label (amber): the total is a lower bound.\n  const cost =\n    s.costUsd !== undefined\n      ? kpi(\n          s.costUnknownRuns ? `cost · ${s.costUnknownRuns} runs unpriced` : 'cost',\n          `$${s.costUsd.toFixed(2)}`,\n          s.costUnknownRuns ? '#e5c07b' : '#e6e6e6',\n        )\n      : ''\n  const stamp = opts.generatedAt ?? capsule.generatedAt ?? ''\n\n  return `<!doctype html><html><head><meta charset=\"utf-8\"><meta name=\"viewport\" content=\"width=device-width,initial-scale=1\">\n<title>${esc(capsule.objective)} capsule — ${esc(capsule.target)}</title>\n<style>\n:root{color-scheme:dark}\n*{box-sizing:border-box}\nbody{margin:0;background:#0c0c0f;color:#e6e6e6;font:14px/1.5 ui-sans-serif,system-ui,-apple-system,Segoe UI,Roboto,sans-serif}\n.wrap{max-width:980px;margin:0 auto;padding:40px 24px}\nh1{font-size:22px;margin:0 0 2px;letter-spacing:-.01em}\n.sub{color:#8a8a93;font-size:13px;margin-bottom:28px}\n.kpis{display:flex;gap:14px;flex-wrap:wrap;margin-bottom:32px}\n.kpi{background:#16161b;border:1px solid #24242b;border-radius:12px;padding:14px 18px;min-width:120px}\n.kv{font-size:24px;font-weight:650;letter-spacing:-.02em}\n.kl{color:#8a8a93;font-size:12px;margin-top:2px}\nh2{font-size:13px;text-transform:uppercase;letter-spacing:.08em;color:#8a8a93;margin:34px 0 14px}\n.axis-label{color:#8a8a93;font-size:12px;margin-bottom:10px}\ntable.heat{border-collapse:separate;border-spacing:6px}\ntable.heat th{font-weight:500;color:#a8a8b0;font-size:12px;padding:4px 8px;text-align:center}\ntable.heat th.rh{text-align:right}\n.tile{position:relative;width:104px;height:66px;border-radius:10px;display:flex;flex-direction:column;justify-content:center;align-items:center;gap:1px;color:#06120a;font-weight:600}\n.tile .tv{font-size:17px}\n.tile .tl{font-size:10px;font-weight:600;opacity:.9;padding:0 6px;text-align:center}\n.tile .dim{font-size:9px;font-weight:600;opacity:.85;background:rgba(0,0,0,.22);padding:0 6px;border-radius:999px}\n.grid{display:flex;flex-wrap:wrap;gap:8px}\n.grid .tile{width:138px;height:74px}\n.fail{background:#16161b;border:1px solid #24242b;border-left:3px solid #d1495b;border-radius:10px;padding:12px 14px;margin-bottom:10px}\n.fmeta{display:flex;gap:10px;align-items:center;font-size:12px;color:#a8a8b0;margin-bottom:8px}\n.fcell{color:#cfcfd6}\n.fclass{background:#2a1a1d;color:#e58a96;padding:1px 8px;border-radius:999px;font-size:11px}\n.fsev{margin-left:auto;color:#e58a96}\n.sevbar{height:4px;background:#24242b;border-radius:2px;overflow:hidden;margin-bottom:10px}\n.sevfill{height:100%;background:#d1495b}\n.ftext{font-size:13px;color:#d8d8df;white-space:pre-wrap}\n.fdims{display:flex;gap:6px;margin-top:8px}\n.fdims .dim{background:#1d2530;color:#9fc1e8;padding:1px 8px;border-radius:999px;font-size:11px}\n.none{color:#5ad17a}\n.foot{color:#5a5a63;font-size:11px;margin-top:36px}\n</style></head><body><div class=\"wrap\">\n<h1>${esc(capsule.objective)} exploration · ${esc(capsule.target)}</h1>\n<div class=\"sub\">${s.totalRuns} scenarios across ${s.cellsCovered}/${s.cellsTotal} planned cells${s.behaviorBinsObserved > 0 ? ` · ${s.behaviorBinsObserved} measured behavior bins` : ''}${stamp ? ` · ${esc(stamp)}` : ''}</div>\n<div class=\"kpis\">\n${kpi('mean robustness', pct(s.meanRobustness), s.meanRobustness < 0.6 ? '#e58a96' : '#5ad17a')}\n${kpi('verified findings', String(s.verifiedFindings), s.verifiedFindings > 0 ? '#e58a96' : '#5ad17a')}\n${kpi('cells covered', `${s.cellsCovered}/${s.cellsTotal}`)}\n${kpi('scenarios run', String(s.totalRuns))}\n${cost}\n${s.evalErrors > 0 ? kpi('eval errors', String(s.evalErrors), '#e5b566') : ''}\n${lift}\n</div>\n${s.stoppedEarly ? `<div class=\"sub\" style=\"color:#e5b566\">stopped early: ${esc(s.stoppedEarly.detail)} — coverage below reflects the completed portion only</div>` : ''}\n<h2>Coverage map</h2>\n${heatmapHtml(capsule.coverage)}\n<h2>Verified findings${s.candidateFindings > s.verifiedFindings ? ` · ${s.verifiedFindings} of ${s.candidateFindings} candidates passed the validity gates` : ''}</h2>\n${findingsHtml(capsule.findings, opts.maxFindings ?? 8)}\n<div class=\"foot\">Findings are gate-verified: each is a fair, answerable task that reproduces under a meaning-preserving rephrase, minimized to the smallest trigger.</div>\n</div></body></html>`\n}\n","/**\n * Shipped policies for the exploration engine.\n *\n * `Proposer` is a plain function type — an agent running a generator skill IS a\n * proposer (`(ctx) => dispatchToSkill(ctx)`), no wrapper needed. `mutationProposer`\n * builds the deterministic, LLM-free one from mutation operators. Objectives are\n * interfaces because the engine reads `kind` + `threshold` off them.\n */\n\nimport type { AdversarialMutation } from '../rl/adversarial'\nimport type {\n  Cell,\n  Evaluation,\n  Objective,\n  ObjectiveContext,\n  ProposeContext,\n  Proposer,\n} from './types'\n\nconst clamp01 = (x: number): number => Math.max(0, Math.min(1, x))\n\n// ── proposers ─────────────────────────────────────────────────────────────────\n\n/**\n * Perturbation-based search: apply the cell's mutation operators to the current\n * elites + seeds, deduping by id. Elites first — mutating the most interesting\n * scenario found so far is what makes the search deepen across rounds.\n */\nexport function mutationProposer<S>(opts: {\n  mutationsFor: (cell: Cell) => AdversarialMutation<S>[]\n  scenarioId: (s: S) => string\n}): Proposer<S> {\n  return async (ctx: ProposeContext<S>): Promise<S[]> => {\n    const mutations = opts.mutationsFor(ctx.cell)\n    const parents = [...ctx.elites, ...ctx.seeds]\n    const seen = new Set(parents.map(opts.scenarioId))\n    const out: S[] = []\n    for (const parent of parents) {\n      if (out.length >= ctx.count) break\n      for (const m of mutations) {\n        const children = await m.mutate(parent, ctx.rng)\n        for (const child of children) {\n          const id = opts.scenarioId(child)\n          if (seen.has(id)) continue\n          seen.add(id)\n          out.push(child)\n          if (out.length >= ctx.count) break\n        }\n        if (out.length >= ctx.count) break\n      }\n    }\n    return out\n  }\n}\n\n// ── objectives ────────────────────────────────────────────────────────────────\n\n/** Adversarial: a low headline score is interesting — find where the agent fails. */\nexport function adversarialObjective(threshold = 0.5): Objective {\n  return { kind: 'adversarial', threshold, interest: (ev) => clamp01(1 - ev.score) }\n}\n\nfunction hamming(\n  a: Record<string, string> | undefined,\n  b: Record<string, string> | undefined,\n): number {\n  if (!a || !b) return 1\n  const keys = new Set([...Object.keys(a), ...Object.keys(b)])\n  if (keys.size === 0) return 0\n  let diff = 0\n  for (const k of keys) if (a[k] !== b[k]) diff++\n  return diff / keys.size\n}\n\n/**\n * Novelty: interesting when far from the archive in score AND measured behavior\n * descriptor — quality-diversity's diversity pressure; drives corpus growth\n * rather than re-finding the same hole.\n */\nexport function noveltyObjective(threshold = 0.3): Objective {\n  return {\n    kind: 'novelty',\n    threshold,\n    interest: (ev: Evaluation, ctx: ObjectiveContext) => {\n      const scoreNovelty =\n        ctx.archiveScores.length === 0\n          ? 1\n          : Math.min(...ctx.archiveScores.map((s) => Math.abs(s - ev.score)))\n      const descNovelty =\n        ctx.archiveDescriptors.length === 0\n          ? 1\n          : Math.min(...ctx.archiveDescriptors.map((d) => hamming(d, ev.descriptor)))\n      return clamp01(0.5 * scoreNovelty + 0.5 * descNovelty)\n    },\n  }\n}\n","/**\n * The exploration engine — a stateful session over a behavior space.\n *\n * Each `step()`: allocate budget across INPUT cells (floor first, then variance\n * steering toward the least-certain cells), propose candidates (the proposer\n * reads current elites + findings, so the search deepens generationally),\n * evaluate with bounded concurrency, archive the most interesting scenario per\n * input×measured bin, and admit notable candidates that pass the validity gates.\n * `run()` loops to budget. `coverage()`/`findings()`/`capsule()` read live state —\n * the surface `makeExploreTools` exposes so an agent can drive the session.\n *\n * One evaluation log (`EvalRecord[]`) is the source of truth; allocation\n * observations and coverage are projections of it.\n */\n\nimport { ValidationError } from '../errors'\nimport { varianceBasedCurriculum } from '../rl/active-curriculum'\nimport { buildCapsule } from './capsule'\nimport type { EvalRecord } from './cube'\nimport { enumerateCells } from './cube'\nimport { adversarialObjective } from './policies'\nimport type {\n  ArchiveEntry,\n  CapsuleData,\n  Cell,\n  CoverageCell,\n  Evaluation,\n  ExploreOptions,\n  Finding,\n  Objective,\n} from './types'\n\nasync function pMap<T>(\n  items: T[],\n  fn: (item: T) => Promise<void>,\n  concurrency: number,\n  signal?: AbortSignal,\n): Promise<void> {\n  let i = 0\n  const workers = Array.from(\n    { length: Math.max(1, Math.min(concurrency, items.length)) },\n    async () => {\n      while (i < items.length) {\n        if (signal?.aborted) return\n        const item = items[i++]\n        if (item === undefined) return\n        await fn(item)\n      }\n    },\n  )\n  await Promise.all(workers)\n}\n\nexport class BehaviorExplorer<S> {\n  private readonly cells: Cell[]\n  private readonly cellById: Map<string, Cell>\n  private readonly objective: Objective\n  private readonly threshold: number\n  private readonly floorPerCell: number\n  private readonly perRoundBudget: number\n\n  /** The single evaluation log — coverage + allocation are projections of it. */\n  private readonly log: Array<EvalRecord & { scenarioId: string }> = []\n  /** binId (input × measured coords) → the most interesting entry seen. */\n  private readonly archiveByBin = new Map<string, ArchiveEntry<S>>()\n  private readonly _findings: Finding<S>[] = []\n  private runsUsed = 0\n  private candidateFindings = 0\n  private evalErrors = 0\n  private consecutiveEvalErrors = 0\n  private stoppedEarly: { reason: 'eval-errors'; detail: string } | undefined\n  private rngState: number\n  /** Accumulated KNOWN dollars — unknown-cost runs never inflate it. */\n  private spentKnownUsd = 0\n  private costUnknownRuns = 0\n\n  constructor(private readonly opts: ExploreOptions<S>) {\n    this.cells = enumerateCells(opts.space)\n    if (this.cells.length === 0)\n      throw new Error('BehaviorExplorer: space has no cells — every axis needs ≥1 value')\n    if (opts.costBudgetUsd !== undefined) {\n      if (\n        typeof opts.costBudgetUsd !== 'number' ||\n        !Number.isFinite(opts.costBudgetUsd) ||\n        opts.costBudgetUsd < 0\n      ) {\n        throw new RangeError(\n          `BehaviorExplorer: costBudgetUsd must be a nonnegative finite number, got ${String(opts.costBudgetUsd)}`,\n        )\n      }\n    }\n    if (!opts.costOf && (opts.costBudgetUsd !== undefined || opts.ledger || opts.onCost)) {\n      throw new ValidationError(\n        'BehaviorExplorer: costBudgetUsd/ledger/onCost require costOf — the explorer ' +\n          'cannot know run cost without it; supply costOf or drop the cost options',\n      )\n    }\n    this.cellById = new Map(this.cells.map((c) => [c.id, c]))\n    this.objective = opts.objective ?? adversarialObjective(0.5)\n    this.threshold = this.objective.threshold ?? 0.5\n    this.floorPerCell = opts.floorPerCell ?? 2\n    this.perRoundBudget = Math.max(\n      this.cells.length * this.floorPerCell,\n      Math.ceil(opts.budget / 4),\n    )\n    this.rngState = (opts.seed ?? 1) >>> 0\n  }\n\n  // mulberry32 — deterministic per session.\n  private rng = (): number => {\n    this.rngState = (this.rngState + 0x6d2b79f5) | 0\n    let t = this.rngState\n    t = Math.imul(t ^ (t >>> 15), t | 1)\n    t ^= t + Math.imul(t ^ (t >>> 7), t | 61)\n    return ((t ^ (t >>> 14)) >>> 0) / 4294967296\n  }\n\n  private binId(cell: Cell, descriptor: Record<string, string> | undefined): string {\n    if (!descriptor || Object.keys(descriptor).length === 0) return cell.id\n    const measured = Object.keys(descriptor)\n      .sort()\n      .map((k) => `${k}=${descriptor[k]}`)\n      .join('|')\n    return `${cell.id}|${measured}`\n  }\n\n  private allocate(budget: number): Array<{ cellId: string; count: number }> {\n    if ((this.opts.allocation ?? 'variance') === 'uniform') {\n      const per = Math.max(this.floorPerCell, Math.floor(budget / this.cells.length))\n      return this.cells.map((c) => ({ cellId: c.id, count: per }))\n    }\n    return varianceBasedCurriculum(\n      this.log.map((r) => ({\n        variantId: r.cell.id,\n        scenarioId: r.scenarioId,\n        score: r.ev.score,\n        pass: r.ev.valid && r.ev.score >= 0.5,\n      })),\n      this.cells.map((c) => ({ variantId: c.id, scenarioId: '*' })),\n      { budget, floorPerCell: this.floorPerCell },\n    ).map((a) => ({ cellId: a.variantId, count: a.count }))\n  }\n\n  private objectiveContext() {\n    const entries = [...this.archiveByBin.values()]\n    return {\n      archiveScores: entries.map((e) => e.evaluation.score),\n      archiveDescriptors: entries.map((e) => e.evaluation.descriptor),\n    }\n  }\n\n  /** Mirrors control-runtime: stop once accumulated KNOWN cost ≥ the ceiling. */\n  private costExhausted(): boolean {\n    return this.opts.costBudgetUsd !== undefined && this.spentKnownUsd >= this.opts.costBudgetUsd\n  }\n\n  /** Fold one run's cost in: null counts as unknown (never $0); a known cost\n   *  accrues toward the budget, lands in the ledger, and fires `onCost`. */\n  private recordRunCost(scenario: S, cell: Cell, ev: Evaluation): void {\n    if (!this.opts.costOf) return\n    const cost = this.opts.costOf(scenario, cell, ev)\n    if (cost === null) {\n      this.costUnknownRuns++\n      return\n    }\n    if (typeof cost.usd !== 'number' || !Number.isFinite(cost.usd) || cost.usd < 0) {\n      throw new RangeError(\n        `BehaviorExplorer: costOf returned an invalid usd (${String(cost?.usd)}) — ` +\n          'return null when cost is unknown, never a fabricated number',\n      )\n    }\n    this.spentKnownUsd += cost.usd\n    this.opts.ledger?.record({\n      model: cost.model ?? 'unattributed',\n      channel: 'agent',\n      usage: { inputTokens: 0, outputTokens: 0 },\n      actualCostUsd: cost.usd,\n      tags: { target: this.opts.target, cell: cell.id },\n    })\n    this.opts.onCost?.({ usd: cost.usd, channel: 'agent' })\n  }\n\n  /** Elites whose INPUT cell matches — what the proposer mutates/deepens from. */\n  private elitesFor(cellId: string): S[] {\n    const out: S[] = []\n    for (const e of this.archiveByBin.values()) if (e.cell.id === cellId) out.push(e.scenario)\n    return out\n  }\n\n  /** One allocate → propose → evaluate → gate → archive round. */\n  async step(): Promise<{ runs: number; findings: Finding<S>[] }> {\n    const remaining = this.opts.budget - this.runsUsed\n    if (remaining <= 0 || this.costExhausted() || this.opts.signal?.aborted)\n      return { runs: 0, findings: [] }\n\n    const allocations = this.allocate(Math.min(this.perRoundBudget, remaining))\n    const newFindings: Finding<S>[] = []\n    let runsThisStep = 0\n\n    for (const alloc of allocations) {\n      if (\n        this.runsUsed >= this.opts.budget ||\n        this.costExhausted() ||\n        this.stoppedEarly !== undefined ||\n        this.opts.signal?.aborted\n      )\n        break\n      const cell = this.cellById.get(alloc.cellId)\n      if (!cell) continue\n      const cap = Math.min(alloc.count, this.opts.budget - this.runsUsed)\n      if (cap <= 0) continue\n      this.opts.onProgress?.({ type: 'cell-allocated', cell, count: cap })\n\n      const seeds = await this.opts.seedsFor(cell)\n      const elites = this.elitesFor(cell.id)\n      const proposed = await this.opts.proposer({\n        cell,\n        seeds,\n        elites,\n        findings: this._findings,\n        count: cap,\n        rng: this.rng,\n      })\n      // Cold cells evaluate their seeds first (the coverage floor); warm cells\n      // trust the proposer, which already saw the elites.\n      const cold = this.log.every((r) => r.cell.id !== cell.id)\n      const toEval = [...(cold ? seeds : []), ...proposed].slice(0, cap)\n\n      await pMap(\n        toEval,\n        async (scenario) => {\n          if (\n            this.runsUsed >= this.opts.budget ||\n            this.costExhausted() ||\n            this.stoppedEarly !== undefined ||\n            this.opts.signal?.aborted\n          )\n            return\n          // evaluate/gates/minimize cross an external boundary (router, backend,\n          // judge). A throw there is an infra outcome: record it as a typed\n          // eval-error and keep exploring — one 5xx must not kill a campaign.\n          // Consecutive failures trip the circuit breaker instead, so a dead\n          // backend stops the run rather than burning the remaining budget.\n          try {\n            const ev = await this.opts.evaluate(scenario, cell)\n            this.runsUsed++\n            runsThisStep++\n            this.consecutiveEvalErrors = 0\n            this.recordRunCost(scenario, cell, ev)\n            const interest = this.objective.interest(ev, this.objectiveContext())\n            this.log.push({ cell, ev, interest, scenarioId: this.opts.scenarioId(scenario) })\n            this.opts.onProgress?.({ type: 'evaluated', cell, scenario, evaluation: ev })\n\n            const bin = this.binId(cell, ev.descriptor)\n            const cur = this.archiveByBin.get(bin)\n            if (!cur || interest > cur.interest)\n              this.archiveByBin.set(bin, { binId: bin, cell, scenario, evaluation: ev, interest })\n\n            if (interest < this.threshold) return\n            this.candidateFindings++\n            if (this.opts.gates?.isValid && !(await this.opts.gates.isValid(scenario, ev, cell)))\n              return\n            if (\n              this.opts.gates?.isUncontaminated &&\n              !(await this.opts.gates.isUncontaminated(scenario, ev, cell))\n            )\n              return\n\n            const minimized = this.opts.minimize\n              ? await this.opts.minimize(scenario, this.opts.evaluate, cell)\n              : scenario\n            const finding: Finding<S> = {\n              id: this.opts.scenarioId(scenario),\n              cell,\n              scenario,\n              minimized,\n              text: this.opts.scenarioText?.(minimized),\n              evaluation: ev,\n              interest,\n              objective: this.objective.kind,\n            }\n            this._findings.push(finding)\n            newFindings.push(finding)\n            this.opts.onProgress?.({ type: 'finding', finding })\n          } catch (err) {\n            // Internal validation errors (e.g. a fabricated costOf number) are\n            // programming mistakes, not backend outcomes — they stay loud.\n            if (err instanceof RangeError) throw err\n            this.evalErrors++\n            this.consecutiveEvalErrors++\n            const message = err instanceof Error ? err.message : String(err)\n            this.opts.onProgress?.({\n              type: 'eval-error',\n              cell,\n              scenarioId: this.opts.scenarioId(scenario),\n              message,\n            })\n            const limit = this.opts.maxConsecutiveEvalErrors ?? 5\n            if (this.consecutiveEvalErrors >= limit) {\n              this.stoppedEarly = {\n                reason: 'eval-errors',\n                detail: `${this.consecutiveEvalErrors} consecutive eval errors (last: ${message})`,\n              }\n            }\n          }\n        },\n        this.opts.concurrency ?? 1,\n        this.opts.signal,\n      )\n    }\n\n    this.opts.onProgress?.({ type: 'round', runsUsed: this.runsUsed, budget: this.opts.budget })\n    return { runs: runsThisStep, findings: newFindings }\n  }\n\n  /** Loop `step()` until the run or dollar budget is spent, the signal aborts,\n   *  or no progress is made. */\n  async run(): Promise<CapsuleData<S>> {\n    while (\n      this.runsUsed < this.opts.budget &&\n      !this.costExhausted() &&\n      this.stoppedEarly === undefined &&\n      !this.opts.signal?.aborted\n    ) {\n      const { runs } = await this.step()\n      if (runs === 0 && this.stoppedEarly === undefined) break\n    }\n    return this.capsule()\n  }\n\n  coverage(): CoverageCell[] {\n    return this.capsule().coverage\n  }\n\n  findings(): Finding<S>[] {\n    return [...this._findings].sort((a, b) => b.interest - a.interest)\n  }\n\n  capsule(): CapsuleData<S> {\n    return buildCapsule({\n      target: this.opts.target,\n      objective: this.objective.kind,\n      cells: this.cells,\n      log: this.log,\n      threshold: this.threshold,\n      archive: [...this.archiveByBin.values()],\n      findings: this._findings,\n      candidateFindings: this.candidateFindings,\n      runsUsed: this.runsUsed,\n      cost: this.opts.costOf\n        ? { costUsd: this.spentKnownUsd, costUnknownRuns: this.costUnknownRuns }\n        : undefined,\n      evalErrors: this.evalErrors,\n      stoppedEarly: this.stoppedEarly,\n    })\n  }\n}\n","/**\n * `fuzzAgent` — the adversarial batch preset over `BehaviorExplorer`.\n *\n * One call: explore the space to budget with the adversarial objective and\n * return the capsule. For agent-driven, incremental, or multi-objective use,\n * construct a `BehaviorExplorer` and drive it via `makeExploreTools`.\n */\n\nimport { BehaviorExplorer } from './explorer'\nimport { adversarialObjective } from './policies'\nimport type { CapsuleData, ExploreOptions } from './types'\n\nexport type FuzzAgentOptions<S> = Omit<ExploreOptions<S>, 'objective'> & {\n  /** Score strictly below this is a candidate failure. Default 0.5. */\n  failureThreshold?: number\n}\n\nexport async function fuzzAgent<S>(\n  opts: FuzzAgentOptions<S>,\n): Promise<{ capsule: CapsuleData<S> }> {\n  const { failureThreshold, ...rest } = opts\n  const explorer = new BehaviorExplorer<S>({\n    ...rest,\n    objective: adversarialObjective(failureThreshold ?? 0.5),\n  })\n  return { capsule: await explorer.run() }\n}\n","/**\n * Validity gates — what separates a fuzzer from a slop generator.\n *\n * A notable candidate is admitted only when it is fair and reproducible. None of\n * these are on by default: the live wiring opts in, so reported findings carry\n * their proof.\n */\n\nimport type { Cell, Evaluation, Evaluator, ValidityGates } from './types'\n\n/** Combine gate sets; a candidate must pass every gate in every set. */\nexport function composeGates<S>(...sets: Array<ValidityGates<S> | undefined>): ValidityGates<S> {\n  const present = sets.filter((s): s is ValidityGates<S> => s != null)\n  return {\n    isValid: async (scenario, ev, cell) => {\n      for (const g of present) {\n        if (g.isValid && !(await g.isValid(scenario, ev, cell))) return false\n      }\n      return true\n    },\n    isUncontaminated: async (scenario, ev, cell) => {\n      for (const g of present) {\n        if (g.isUncontaminated && !(await g.isUncontaminated(scenario, ev, cell))) return false\n      }\n      return true\n    },\n  }\n}\n\n/**\n * Reproducibility gate. Re-run the target on a meaning-preserving rephrase of the\n * flagged scenario; keep the finding only when the rephrase ALSO scores below the\n * threshold. A finding that flips under a cosmetic rewrite was keyed to surface\n * form, not the task — a false signal we must not report. Costs one extra\n * evaluation per candidate (candidates are rare, so cheap).\n */\nexport function perturbationStabilityGate<S>(opts: {\n  evaluate: Evaluator<S>\n  /** Produce a semantic-preserving rephrase. Return null to skip (treated as pass). */\n  perturb: (scenario: S) => S | null\n  /** Score strictly below this still counts as failing. Default 0.5. */\n  failureThreshold?: number\n}): ValidityGates<S> {\n  const threshold = opts.failureThreshold ?? 0.5\n  return {\n    isUncontaminated: async (scenario: S, _ev: Evaluation, cell: Cell) => {\n      const rephrased = opts.perturb(scenario)\n      if (rephrased == null) return true\n      const re = await opts.evaluate(rephrased, cell)\n      return re.score < threshold\n    },\n  }\n}\n\n/**\n * Severity-floor gate. Reject borderline candidates whose score sits in a band\n * just under the threshold — judge noise, not a real defect.\n */\nexport function severityFloorGate<S>(opts: {\n  failureThreshold?: number\n  margin?: number\n}): ValidityGates<S> {\n  const threshold = opts.failureThreshold ?? 0.5\n  const margin = opts.margin ?? 0.1\n  return {\n    isValid: (_scenario, ev: Evaluation) => ev.score <= threshold - margin,\n  }\n}\n","/**\n * Agent-drivable surface over a live exploration session.\n *\n * Framework-neutral tool defs ({name, description, parameters: JSON Schema,\n * handler}) so the on-demand agent — not a batch script — drives the search:\n * step it, read coverage, inspect findings, render the capsule. Transport\n * encodings (OpenAI function shape, MCP) are one-line mappings the host owns.\n */\n\nimport { renderCapsuleHtml } from './capsule'\nimport type { BehaviorExplorer } from './explorer'\n\nexport interface ExploreToolDef {\n  name: string\n  description: string\n  /** JSON Schema (draft-07+) for the arguments. */\n  parameters: Record<string, unknown>\n  handler: (args: unknown, ctx?: { signal?: AbortSignal }) => Promise<unknown>\n}\n\nconst NO_ARGS = { type: 'object', properties: {}, additionalProperties: false }\n\nexport function makeExploreTools<S>(explorer: BehaviorExplorer<S>): ExploreToolDef[] {\n  return [\n    {\n      name: 'explore_step',\n      description:\n        'Run one exploration round: allocate budget across cells, propose + evaluate scenarios, archive elites, admit gate-verified findings. Returns runs spent and new findings.',\n      parameters: NO_ARGS,\n      handler: async () => {\n        const { runs, findings } = await explorer.step()\n        return { runs, newFindings: findings.length, findings }\n      },\n    },\n    {\n      name: 'explore_coverage',\n      description:\n        'Read the live coverage map: per planned cell — runs, robustness (null = uncovered), finding rate, per-dimension means.',\n      parameters: NO_ARGS,\n      handler: async () => explorer.coverage(),\n    },\n    {\n      name: 'explore_findings',\n      description: 'List gate-verified findings so far, sorted by descending interest.',\n      parameters: {\n        type: 'object',\n        properties: { limit: { type: 'number', description: 'max findings to return' } },\n        additionalProperties: false,\n      },\n      handler: async (args) => {\n        const limit = (args as { limit?: number } | undefined)?.limit\n        const all = explorer.findings()\n        return typeof limit === 'number' ? all.slice(0, limit) : all\n      },\n    },\n    {\n      name: 'explore_capsule',\n      description:\n        'Build the capsule artifact from the current session state. format \"data\" returns the structured CapsuleData; \"html\" returns the standalone page.',\n      parameters: {\n        type: 'object',\n        properties: {\n          format: { type: 'string', enum: ['data', 'html'] },\n          generatedAt: { type: 'string', description: 'ISO timestamp to stamp into the page' },\n        },\n        additionalProperties: false,\n      },\n      handler: async (args) => {\n        const a = (args ?? {}) as { format?: string; generatedAt?: string }\n        const capsule = explorer.capsule()\n        if (a.format === 'html') return renderCapsuleHtml(capsule, { generatedAt: a.generatedAt })\n        return capsule\n      },\n    },\n  ]\n}\n"],"mappings":";;;;;;;;;AAqBO,SAAS,eAAe,OAA8B;AAC3D,MAAI,MAAM,KAAK,WAAW,EAAG,QAAO,CAAC;AACrC,MAAI,WAA0C,CAAC,CAAC,CAAC;AACjD,aAAW,QAAQ,MAAM,MAAM;AAC7B,UAAM,OAAsC,CAAC;AAC7C,eAAW,WAAW,UAAU;AAC9B,iBAAW,SAAS,KAAK,OAAQ,MAAK,KAAK,EAAE,GAAG,SAAS,CAAC,KAAK,IAAI,GAAG,MAAM,CAAC;AAAA,IAC/E;AACA,eAAW;AAAA,EACb;AACA,SAAO,SAAS,IAAI,CAAC,YAAY,EAAE,IAAI,OAAO,OAAO,MAAM,GAAG,OAAO,EAAE;AACzE;AAGO,SAAS,OAAO,OAAsB,QAAwC;AACnF,SAAO,MAAM,KAAK,IAAI,CAAC,MAAM,GAAG,EAAE,IAAI,IAAI,OAAO,EAAE,IAAI,CAAC,EAAE,EAAE,KAAK,GAAG;AACtE;AAEA,IAAM,OAAO,CAAC,OACZ,GAAG,WAAW,IAAI,IAAI,GAAG,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,GAAG;AAMpD,SAAS,cAAc,OAAe,KAAmB,WAAmC;AACjG,QAAM,SAAS,oBAAI,IAA0B;AAC7C,aAAW,KAAK,KAAK;AACnB,UAAM,MAAM,OAAO,IAAI,EAAE,KAAK,EAAE,KAAK,CAAC;AACtC,QAAI,KAAK,CAAC;AACV,WAAO,IAAI,EAAE,KAAK,IAAI,GAAG;AAAA,EAC3B;AACA,SAAO,MAAM,IAAI,CAAC,SAAS;AACzB,UAAM,OAAO,OAAO,IAAI,KAAK,EAAE,KAAK,CAAC;AACrC,UAAM,OAAO,KAAK;AAClB,QAAI,SAAS,EAAG,QAAO,EAAE,MAAM,MAAM,GAAG,YAAY,MAAM,aAAa,GAAG,YAAY,CAAC,EAAE;AACzF,UAAM,aAAa,KAAK,KAAK,IAAI,CAAC,MAAM,EAAE,GAAG,KAAK,CAAC;AACnD,UAAM,cAAc,KAAK,OAAO,CAAC,MAAM,EAAE,YAAY,SAAS,EAAE,SAAS;AACzE,UAAM,OAAiC,CAAC;AACxC,eAAW,KAAK,MAAM;AACpB,iBAAW,CAAC,GAAG,CAAC,KAAK,OAAO,QAAQ,EAAE,GAAG,UAAU,CAAC,CAAC,GAAG;AACtD;AAAC,SAAC,KAAK,CAAC,MAAM,CAAC,GAAG,KAAK,CAAC;AAAA,MAC1B;AAAA,IACF;AACA,UAAM,aAAqC,CAAC;AAC5C,eAAW,CAAC,GAAG,EAAE,KAAK,OAAO,QAAQ,IAAI,EAAG,YAAW,CAAC,IAAI,KAAK,EAAE;AACnE,WAAO,EAAE,MAAM,MAAM,YAAY,aAAa,WAAW;AAAA,EAC3D,CAAC;AACH;;;ACnCO,SAAS,aAAgB,OAA6C;AAC3E,QAAM,WAAW,cAAc,MAAM,OAAO,MAAM,KAAK,MAAM,SAAS;AACtE,QAAM,UAAU,SAAS,OAAO,CAAC,MAAM,EAAE,OAAO,CAAC;AACjD,QAAM,iBACJ,QAAQ,WAAW,IAAI,IAAI,QAAQ,OAAO,CAAC,GAAG,MAAM,KAAK,EAAE,cAAc,IAAI,CAAC,IAAI,QAAQ;AAE5F,QAAM,uBAAuB,MAAM,QAAQ,OAAO,CAAC,MAAM,EAAE,UAAU,EAAE,KAAK,EAAE,EAAE;AAEhF,SAAO;AAAA,IACL,QAAQ,MAAM;AAAA,IACd,WAAW,MAAM;AAAA,IACjB;AAAA,IACA,UAAU,CAAC,GAAG,MAAM,QAAQ,EAAE,KAAK,CAAC,GAAG,MAAM,EAAE,WAAW,EAAE,QAAQ;AAAA,IACpE,SAAS,CAAC,GAAG,MAAM,OAAO,EAAE,KAAK,CAAC,GAAG,MAAM,EAAE,WAAW,EAAE,QAAQ;AAAA,IAClE,OAAO;AAAA,MACL,WAAW,MAAM;AAAA,MACjB,YAAY,MAAM,MAAM;AAAA,MACxB,cAAc,QAAQ;AAAA,MACtB;AAAA,MACA,mBAAmB,MAAM;AAAA,MACzB,kBAAkB,MAAM,SAAS;AAAA,MACjC;AAAA,MACA,GAAI,MAAM,OACN,EAAE,SAAS,MAAM,KAAK,SAAS,iBAAiB,MAAM,KAAK,gBAAgB,IAC3E,CAAC;AAAA,MACL,YAAY,MAAM;AAAA,MAClB,GAAI,MAAM,eAAe,EAAE,cAAc,MAAM,aAAa,IAAI,CAAC;AAAA,IACnE;AAAA,EACF;AACF;AAIA,SAAS,IAAI,GAAmB;AAC9B,SAAO,EAAE;AAAA,IACP;AAAA,IACA,CAAC,OAAO,EAAE,KAAK,SAAS,KAAK,QAAQ,KAAK,QAAQ,KAAK,SAAS,GAAG,CAAC;AAAA,EACtE;AACF;AAGA,SAAS,gBAAgB,GAA0B;AACjD,MAAI,KAAK,KAAM,QAAO;AACtB,SAAO,OAAO,KAAK,MAAM,IAAI,GAAG,CAAC;AACnC;AAEA,SAAS,IAAI,GAAmB;AAC9B,SAAO,GAAG,KAAK,MAAM,IAAI,GAAG,CAAC;AAC/B;AAEA,SAAS,WAAW,UAAqE;AACvF,QAAM,QAAkB,CAAC;AACzB,QAAM,OAAO,oBAAI,IAAyB;AAC1C,aAAW,KAAK,UAAU;AACxB,eAAW,CAAC,GAAG,CAAC,KAAK,OAAO,QAAQ,EAAE,KAAK,MAAM,GAAG;AAClD,UAAI,CAAC,KAAK,IAAI,CAAC,GAAG;AAChB,aAAK,IAAI,GAAG,oBAAI,IAAI,CAAC;AACrB,cAAM,KAAK,CAAC;AAAA,MACd;AACA,WAAK,IAAI,CAAC,GAAG,IAAI,CAAC;AAAA,IACpB;AAAA,EACF;AACA,SAAO,MAAM,IAAI,CAAC,UAAU,EAAE,MAAM,QAAQ,CAAC,GAAI,KAAK,IAAI,IAAI,KAAK,CAAC,CAAE,EAAE,EAAE;AAC5E;AAGA,SAAS,WAAW,GAAyB;AAC3C,QAAM,UAAU,OAAO,QAAQ,EAAE,UAAU;AAC3C,MAAI,QAAQ,WAAW,EAAG,QAAO;AACjC,QAAM,SAAS,QAAQ,KAAK,CAAC,GAAG,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,CAAC;AACjD,QAAM,IAAI,OAAO,CAAC;AAClB,MAAI,CAAC,EAAG,QAAO;AACf,SAAO,qBAAqB,IAAI,EAAE,CAAC,CAAC,CAAC,IAAI,IAAI,EAAE,CAAC,CAAC,CAAC;AACpD;AAEA,SAAS,YAAY,UAAkC;AACrD,QAAM,OAAO,WAAW,QAAQ;AAChC,QAAM,OAAO,IAAI,IAAI,SAAS,IAAI,CAAC,MAAM,CAAC,EAAE,KAAK,IAAI,CAAC,CAAC,CAAC;AACxD,QAAM,OAAO,CAAC,GAA6B,UAA0B;AACnE,UAAM,IAAI,GAAG,cAAc;AAC3B,UAAM,QAAQ,IACV,GAAG,IAAI,KAAK,CAAC,CAAC,gBAAa,EAAE,IAAI,cAAW,IAAI,EAAE,WAAW,CAAC,aAC9D;AACJ,WAAO,uCAAuC,gBAAgB,CAAC,CAAC,YAAY,IAAI,KAAK,CAAC,KAAK,QAAQ,oBAAoB,IAAI,KAAK,CAAC,YAAY,EAAE,oBAAoB,KAAK,KAAK,OAAO,IAAI,CAAC,IAAI,QAAG,UAAU,IAAI,WAAW,CAAC,IAAI,EAAE;AAAA,EAClO;AAEA,QAAM,UAAU,KAAK,CAAC;AACtB,QAAM,UAAU,KAAK,CAAC;AACtB,MAAI,KAAK,WAAW,KAAK,WAAW,SAAS;AAC3C,UAAM,OAAO,gBAAgB,QAAQ,OAAO,IAAI,CAAC,MAAM,OAAO,IAAI,CAAC,CAAC,OAAO,EAAE,KAAK,EAAE,CAAC;AACrF,UAAM,OAAO,QAAQ,OAClB,IAAI,CAAC,OAAO;AACX,YAAM,QAAQ,QAAQ,OACnB,IAAI,CAAC,OAAO;AACX,cAAM,KAAK,GAAG,QAAQ,IAAI,IAAI,EAAE,IAAI,QAAQ,IAAI,IAAI,EAAE;AACtD,eAAO,OAAO,KAAK,KAAK,IAAI,EAAE,GAAG,EAAE,CAAC;AAAA,MACtC,CAAC,EACA,KAAK,EAAE;AACV,aAAO,sBAAsB,IAAI,EAAE,CAAC,QAAQ,KAAK;AAAA,IACnD,CAAC,EACA,KAAK,EAAE;AACV,WAAO,oCAAoC,IAAI,QAAQ,IAAI,CAAC,sBAAmB,IAAI,QAAQ,IAAI,CAAC,iCAAiC,IAAI,GAAG,IAAI;AAAA,EAC9I;AAEA,QAAM,SAAS,CAAC,GAAG,QAAQ,EAAE,KAAK,CAAC,GAAG,OAAO,EAAE,cAAc,MAAM,EAAE,cAAc,EAAE;AACrF,SAAO,qBAAqB,OAAO,IAAI,CAAC,MAAM,KAAK,GAAG,OAAO,OAAO,EAAE,KAAK,MAAM,EAAE,KAAK,QAAK,CAAC,CAAC,EAAE,KAAK,EAAE,CAAC;AAC3G;AAEA,SAAS,aAAgB,UAAwB,OAAuB;AACtE,MAAI,SAAS,WAAW;AACtB,WAAO;AACT,SAAO,SACJ,MAAM,GAAG,KAAK,EACd,IAAI,CAAC,MAAM;AACV,UAAM,SAAS,OAAO,QAAQ,EAAE,KAAK,MAAM,EACxC,IAAI,CAAC,CAAC,GAAG,CAAC,MAAM,GAAG,CAAC,IAAI,CAAC,EAAE,EAC3B,KAAK,QAAK;AACb,UAAM,UAAU,EAAE,WAAW,UAAU,CAAC,GACrC,IAAI,CAAC,MAAM,wBAAwB,IAAI,CAAC,CAAC,SAAS,EAClD,KAAK,EAAE;AACV,UAAM,OAAO,OAAO,QAAQ,EAAE,WAAW,UAAU,CAAC,CAAC,EAClD,KAAK,CAAC,GAAG,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,CAAC,EAC1B,MAAM,GAAG,CAAC,EACV,IAAI,CAAC,CAAC,GAAG,CAAC,MAAM,qBAAqB,IAAI,CAAC,CAAC,IAAI,IAAI,CAAC,CAAC,SAAS,EAC9D,KAAK,EAAE;AACV,WAAO,4DAA4D,IAAI,MAAM,CAAC,UAAU,MAAM,+BAA+B,IAAI,EAAE,QAAQ,CAAC,sEAAsE,IAAI,EAAE,QAAQ,CAAC,oCAAoC,IAAI,EAAE,QAAQ,8BAA8B,CAAC,SAAS,OAAO,sBAAsB,IAAI,WAAW,EAAE;AAAA,EAC3W,CAAC,EACA,KAAK,EAAE;AACZ;AAUO,SAAS,kBACd,SACA,OAA6B,CAAC,GACtB;AACR,QAAM,IAAI,QAAQ;AAClB,QAAM,MAAM,CAAC,OAAe,OAAe,SAAS,cAClD,iDAAiD,MAAM,KAAK,IAAI,KAAK,CAAC,yBAAyB,IAAI,KAAK,CAAC;AAC3G,QAAM,OAAO,QAAQ,OACjB;AAAA,IACE;AAAA,IACA,GAAG,QAAQ,KAAK,OAAO,QAAQ,CAAC,CAAC,WAAM,QAAQ,KAAK,MAAM,QAAQ,CAAC,CAAC;AAAA,IACpE;AAAA,EACF,IACA;AAGJ,QAAM,OACJ,EAAE,YAAY,SACV;AAAA,IACE,EAAE,kBAAkB,aAAU,EAAE,eAAe,mBAAmB;AAAA,IAClE,IAAI,EAAE,QAAQ,QAAQ,CAAC,CAAC;AAAA,IACxB,EAAE,kBAAkB,YAAY;AAAA,EAClC,IACA;AACN,QAAM,QAAQ,KAAK,eAAe,QAAQ,eAAe;AAEzD,SAAO;AAAA,SACA,IAAI,QAAQ,SAAS,CAAC,mBAAc,IAAI,QAAQ,MAAM,CAAC;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,MAoC1D,IAAI,QAAQ,SAAS,CAAC,qBAAkB,IAAI,QAAQ,MAAM,CAAC;AAAA,mBAC9C,EAAE,SAAS,qBAAqB,EAAE,YAAY,IAAI,EAAE,UAAU,iBAAiB,EAAE,uBAAuB,IAAI,SAAM,EAAE,oBAAoB,4BAA4B,EAAE,GAAG,QAAQ,SAAM,IAAI,KAAK,CAAC,KAAK,EAAE;AAAA;AAAA,EAEzN,IAAI,mBAAmB,IAAI,EAAE,cAAc,GAAG,EAAE,iBAAiB,MAAM,YAAY,SAAS,CAAC;AAAA,EAC7F,IAAI,qBAAqB,OAAO,EAAE,gBAAgB,GAAG,EAAE,mBAAmB,IAAI,YAAY,SAAS,CAAC;AAAA,EACpG,IAAI,iBAAiB,GAAG,EAAE,YAAY,IAAI,EAAE,UAAU,EAAE,CAAC;AAAA,EACzD,IAAI,iBAAiB,OAAO,EAAE,SAAS,CAAC,CAAC;AAAA,EACzC,IAAI;AAAA,EACJ,EAAE,aAAa,IAAI,IAAI,eAAe,OAAO,EAAE,UAAU,GAAG,SAAS,IAAI,EAAE;AAAA,EAC3E,IAAI;AAAA;AAAA,EAEJ,EAAE,eAAe,yDAAyD,IAAI,EAAE,aAAa,MAAM,CAAC,qEAAgE,EAAE;AAAA;AAAA,EAEtK,YAAY,QAAQ,QAAQ,CAAC;AAAA,uBACR,EAAE,oBAAoB,EAAE,mBAAmB,SAAM,EAAE,gBAAgB,OAAO,EAAE,iBAAiB,0CAA0C,EAAE;AAAA,EAC9J,aAAa,QAAQ,UAAU,KAAK,eAAe,CAAC,CAAC;AAAA;AAAA;AAGvD;;;AC1OA,IAAM,UAAU,CAAC,MAAsB,KAAK,IAAI,GAAG,KAAK,IAAI,GAAG,CAAC,CAAC;AAS1D,SAAS,iBAAoB,MAGpB;AACd,SAAO,OAAO,QAAyC;AACrD,UAAM,YAAY,KAAK,aAAa,IAAI,IAAI;AAC5C,UAAM,UAAU,CAAC,GAAG,IAAI,QAAQ,GAAG,IAAI,KAAK;AAC5C,UAAM,OAAO,IAAI,IAAI,QAAQ,IAAI,KAAK,UAAU,CAAC;AACjD,UAAM,MAAW,CAAC;AAClB,eAAW,UAAU,SAAS;AAC5B,UAAI,IAAI,UAAU,IAAI,MAAO;AAC7B,iBAAW,KAAK,WAAW;AACzB,cAAM,WAAW,MAAM,EAAE,OAAO,QAAQ,IAAI,GAAG;AAC/C,mBAAW,SAAS,UAAU;AAC5B,gBAAM,KAAK,KAAK,WAAW,KAAK;AAChC,cAAI,KAAK,IAAI,EAAE,EAAG;AAClB,eAAK,IAAI,EAAE;AACX,cAAI,KAAK,KAAK;AACd,cAAI,IAAI,UAAU,IAAI,MAAO;AAAA,QAC/B;AACA,YAAI,IAAI,UAAU,IAAI,MAAO;AAAA,MAC/B;AAAA,IACF;AACA,WAAO;AAAA,EACT;AACF;AAKO,SAAS,qBAAqB,YAAY,KAAgB;AAC/D,SAAO,EAAE,MAAM,eAAe,WAAW,UAAU,CAAC,OAAO,QAAQ,IAAI,GAAG,KAAK,EAAE;AACnF;AAEA,SAAS,QACP,GACA,GACQ;AACR,MAAI,CAAC,KAAK,CAAC,EAAG,QAAO;AACrB,QAAM,OAAO,oBAAI,IAAI,CAAC,GAAG,OAAO,KAAK,CAAC,GAAG,GAAG,OAAO,KAAK,CAAC,CAAC,CAAC;AAC3D,MAAI,KAAK,SAAS,EAAG,QAAO;AAC5B,MAAI,OAAO;AACX,aAAW,KAAK,KAAM,KAAI,EAAE,CAAC,MAAM,EAAE,CAAC,EAAG;AACzC,SAAO,OAAO,KAAK;AACrB;AAOO,SAAS,iBAAiB,YAAY,KAAgB;AAC3D,SAAO;AAAA,IACL,MAAM;AAAA,IACN;AAAA,IACA,UAAU,CAAC,IAAgB,QAA0B;AACnD,YAAM,eACJ,IAAI,cAAc,WAAW,IACzB,IACA,KAAK,IAAI,GAAG,IAAI,cAAc,IAAI,CAAC,MAAM,KAAK,IAAI,IAAI,GAAG,KAAK,CAAC,CAAC;AACtE,YAAM,cACJ,IAAI,mBAAmB,WAAW,IAC9B,IACA,KAAK,IAAI,GAAG,IAAI,mBAAmB,IAAI,CAAC,MAAM,QAAQ,GAAG,GAAG,UAAU,CAAC,CAAC;AAC9E,aAAO,QAAQ,MAAM,eAAe,MAAM,WAAW;AAAA,IACvD;AAAA,EACF;AACF;;;AC/DA,eAAe,KACb,OACA,IACA,aACA,QACe;AACf,MAAI,IAAI;AACR,QAAM,UAAU,MAAM;AAAA,IACpB,EAAE,QAAQ,KAAK,IAAI,GAAG,KAAK,IAAI,aAAa,MAAM,MAAM,CAAC,EAAE;AAAA,IAC3D,YAAY;AACV,aAAO,IAAI,MAAM,QAAQ;AACvB,YAAI,QAAQ,QAAS;AACrB,cAAM,OAAO,MAAM,GAAG;AACtB,YAAI,SAAS,OAAW;AACxB,cAAM,GAAG,IAAI;AAAA,MACf;AAAA,IACF;AAAA,EACF;AACA,QAAM,QAAQ,IAAI,OAAO;AAC3B;AAEO,IAAM,mBAAN,MAA0B;AAAA,EAuB/B,YAA6B,MAAyB;AAAzB;AAC3B,SAAK,QAAQ,eAAe,KAAK,KAAK;AACtC,QAAI,KAAK,MAAM,WAAW;AACxB,YAAM,IAAI,MAAM,4EAAkE;AACpF,QAAI,KAAK,kBAAkB,QAAW;AACpC,UACE,OAAO,KAAK,kBAAkB,YAC9B,CAAC,OAAO,SAAS,KAAK,aAAa,KACnC,KAAK,gBAAgB,GACrB;AACA,cAAM,IAAI;AAAA,UACR,4EAA4E,OAAO,KAAK,aAAa,CAAC;AAAA,QACxG;AAAA,MACF;AAAA,IACF;AACA,QAAI,CAAC,KAAK,WAAW,KAAK,kBAAkB,UAAa,KAAK,UAAU,KAAK,SAAS;AACpF,YAAM,IAAI;AAAA,QACR;AAAA,MAEF;AAAA,IACF;AACA,SAAK,WAAW,IAAI,IAAI,KAAK,MAAM,IAAI,CAAC,MAAM,CAAC,EAAE,IAAI,CAAC,CAAC,CAAC;AACxD,SAAK,YAAY,KAAK,aAAa,qBAAqB,GAAG;AAC3D,SAAK,YAAY,KAAK,UAAU,aAAa;AAC7C,SAAK,eAAe,KAAK,gBAAgB;AACzC,SAAK,iBAAiB,KAAK;AAAA,MACzB,KAAK,MAAM,SAAS,KAAK;AAAA,MACzB,KAAK,KAAK,KAAK,SAAS,CAAC;AAAA,IAC3B;AACA,SAAK,YAAY,KAAK,QAAQ,OAAO;AAAA,EACvC;AAAA,EA9B6B;AAAA,EAtBZ;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA;AAAA,EAGA,MAAkD,CAAC;AAAA;AAAA,EAEnD,eAAe,oBAAI,IAA6B;AAAA,EAChD,YAA0B,CAAC;AAAA,EACpC,WAAW;AAAA,EACX,oBAAoB;AAAA,EACpB,aAAa;AAAA,EACb,wBAAwB;AAAA,EACxB;AAAA,EACA;AAAA;AAAA,EAEA,gBAAgB;AAAA,EAChB,kBAAkB;AAAA;AAAA,EAmClB,MAAM,MAAc;AAC1B,SAAK,WAAY,KAAK,WAAW,aAAc;AAC/C,QAAI,IAAI,KAAK;AACb,QAAI,KAAK,KAAK,IAAK,MAAM,IAAK,IAAI,CAAC;AACnC,SAAK,IAAI,KAAK,KAAK,IAAK,MAAM,GAAI,IAAI,EAAE;AACxC,aAAS,IAAK,MAAM,QAAS,KAAK;AAAA,EACpC;AAAA,EAEQ,MAAM,MAAY,YAAwD;AAChF,QAAI,CAAC,cAAc,OAAO,KAAK,UAAU,EAAE,WAAW,EAAG,QAAO,KAAK;AACrE,UAAM,WAAW,OAAO,KAAK,UAAU,EACpC,KAAK,EACL,IAAI,CAAC,MAAM,GAAG,CAAC,IAAI,WAAW,CAAC,CAAC,EAAE,EAClC,KAAK,GAAG;AACX,WAAO,GAAG,KAAK,EAAE,IAAI,QAAQ;AAAA,EAC/B;AAAA,EAEQ,SAAS,QAA0D;AACzE,SAAK,KAAK,KAAK,cAAc,gBAAgB,WAAW;AACtD,YAAM,MAAM,KAAK,IAAI,KAAK,cAAc,KAAK,MAAM,SAAS,KAAK,MAAM,MAAM,CAAC;AAC9E,aAAO,KAAK,MAAM,IAAI,CAAC,OAAO,EAAE,QAAQ,EAAE,IAAI,OAAO,IAAI,EAAE;AAAA,IAC7D;AACA,WAAO;AAAA,MACL,KAAK,IAAI,IAAI,CAAC,OAAO;AAAA,QACnB,WAAW,EAAE,KAAK;AAAA,QAClB,YAAY,EAAE;AAAA,QACd,OAAO,EAAE,GAAG;AAAA,QACZ,MAAM,EAAE,GAAG,SAAS,EAAE,GAAG,SAAS;AAAA,MACpC,EAAE;AAAA,MACF,KAAK,MAAM,IAAI,CAAC,OAAO,EAAE,WAAW,EAAE,IAAI,YAAY,IAAI,EAAE;AAAA,MAC5D,EAAE,QAAQ,cAAc,KAAK,aAAa;AAAA,IAC5C,EAAE,IAAI,CAAC,OAAO,EAAE,QAAQ,EAAE,WAAW,OAAO,EAAE,MAAM,EAAE;AAAA,EACxD;AAAA,EAEQ,mBAAmB;AACzB,UAAM,UAAU,CAAC,GAAG,KAAK,aAAa,OAAO,CAAC;AAC9C,WAAO;AAAA,MACL,eAAe,QAAQ,IAAI,CAAC,MAAM,EAAE,WAAW,KAAK;AAAA,MACpD,oBAAoB,QAAQ,IAAI,CAAC,MAAM,EAAE,WAAW,UAAU;AAAA,IAChE;AAAA,EACF;AAAA;AAAA,EAGQ,gBAAyB;AAC/B,WAAO,KAAK,KAAK,kBAAkB,UAAa,KAAK,iBAAiB,KAAK,KAAK;AAAA,EAClF;AAAA;AAAA;AAAA,EAIQ,cAAc,UAAa,MAAY,IAAsB;AACnE,QAAI,CAAC,KAAK,KAAK,OAAQ;AACvB,UAAM,OAAO,KAAK,KAAK,OAAO,UAAU,MAAM,EAAE;AAChD,QAAI,SAAS,MAAM;AACjB,WAAK;AACL;AAAA,IACF;AACA,QAAI,OAAO,KAAK,QAAQ,YAAY,CAAC,OAAO,SAAS,KAAK,GAAG,KAAK,KAAK,MAAM,GAAG;AAC9E,YAAM,IAAI;AAAA,QACR,qDAAqD,OAAO,MAAM,GAAG,CAAC;AAAA,MAExE;AAAA,IACF;AACA,SAAK,iBAAiB,KAAK;AAC3B,SAAK,KAAK,QAAQ,OAAO;AAAA,MACvB,OAAO,KAAK,SAAS;AAAA,MACrB,SAAS;AAAA,MACT,OAAO,EAAE,aAAa,GAAG,cAAc,EAAE;AAAA,MACzC,eAAe,KAAK;AAAA,MACpB,MAAM,EAAE,QAAQ,KAAK,KAAK,QAAQ,MAAM,KAAK,GAAG;AAAA,IAClD,CAAC;AACD,SAAK,KAAK,SAAS,EAAE,KAAK,KAAK,KAAK,SAAS,QAAQ,CAAC;AAAA,EACxD;AAAA;AAAA,EAGQ,UAAUA,SAAqB;AACrC,UAAM,MAAW,CAAC;AAClB,eAAW,KAAK,KAAK,aAAa,OAAO,EAAG,KAAI,EAAE,KAAK,OAAOA,QAAQ,KAAI,KAAK,EAAE,QAAQ;AACzF,WAAO;AAAA,EACT;AAAA;AAAA,EAGA,MAAM,OAA0D;AAC9D,UAAM,YAAY,KAAK,KAAK,SAAS,KAAK;AAC1C,QAAI,aAAa,KAAK,KAAK,cAAc,KAAK,KAAK,KAAK,QAAQ;AAC9D,aAAO,EAAE,MAAM,GAAG,UAAU,CAAC,EAAE;AAEjC,UAAM,cAAc,KAAK,SAAS,KAAK,IAAI,KAAK,gBAAgB,SAAS,CAAC;AAC1E,UAAM,cAA4B,CAAC;AACnC,QAAI,eAAe;AAEnB,eAAW,SAAS,aAAa;AAC/B,UACE,KAAK,YAAY,KAAK,KAAK,UAC3B,KAAK,cAAc,KACnB,KAAK,iBAAiB,UACtB,KAAK,KAAK,QAAQ;AAElB;AACF,YAAM,OAAO,KAAK,SAAS,IAAI,MAAM,MAAM;AAC3C,UAAI,CAAC,KAAM;AACX,YAAM,MAAM,KAAK,IAAI,MAAM,OAAO,KAAK,KAAK,SAAS,KAAK,QAAQ;AAClE,UAAI,OAAO,EAAG;AACd,WAAK,KAAK,aAAa,EAAE,MAAM,kBAAkB,MAAM,OAAO,IAAI,CAAC;AAEnE,YAAM,QAAQ,MAAM,KAAK,KAAK,SAAS,IAAI;AAC3C,YAAM,SAAS,KAAK,UAAU,KAAK,EAAE;AACrC,YAAM,WAAW,MAAM,KAAK,KAAK,SAAS;AAAA,QACxC;AAAA,QACA;AAAA,QACA;AAAA,QACA,UAAU,KAAK;AAAA,QACf,OAAO;AAAA,QACP,KAAK,KAAK;AAAA,MACZ,CAAC;AAGD,YAAM,OAAO,KAAK,IAAI,MAAM,CAAC,MAAM,EAAE,KAAK,OAAO,KAAK,EAAE;AACxD,YAAM,SAAS,CAAC,GAAI,OAAO,QAAQ,CAAC,GAAI,GAAG,QAAQ,EAAE,MAAM,GAAG,GAAG;AAEjE,YAAM;AAAA,QACJ;AAAA,QACA,OAAO,aAAa;AAClB,cACE,KAAK,YAAY,KAAK,KAAK,UAC3B,KAAK,cAAc,KACnB,KAAK,iBAAiB,UACtB,KAAK,KAAK,QAAQ;AAElB;AAMF,cAAI;AACF,kBAAM,KAAK,MAAM,KAAK,KAAK,SAAS,UAAU,IAAI;AAClD,iBAAK;AACL;AACA,iBAAK,wBAAwB;AAC7B,iBAAK,cAAc,UAAU,MAAM,EAAE;AACrC,kBAAM,WAAW,KAAK,UAAU,SAAS,IAAI,KAAK,iBAAiB,CAAC;AACpE,iBAAK,IAAI,KAAK,EAAE,MAAM,IAAI,UAAU,YAAY,KAAK,KAAK,WAAW,QAAQ,EAAE,CAAC;AAChF,iBAAK,KAAK,aAAa,EAAE,MAAM,aAAa,MAAM,UAAU,YAAY,GAAG,CAAC;AAE5E,kBAAM,MAAM,KAAK,MAAM,MAAM,GAAG,UAAU;AAC1C,kBAAM,MAAM,KAAK,aAAa,IAAI,GAAG;AACrC,gBAAI,CAAC,OAAO,WAAW,IAAI;AACzB,mBAAK,aAAa,IAAI,KAAK,EAAE,OAAO,KAAK,MAAM,UAAU,YAAY,IAAI,SAAS,CAAC;AAErF,gBAAI,WAAW,KAAK,UAAW;AAC/B,iBAAK;AACL,gBAAI,KAAK,KAAK,OAAO,WAAW,CAAE,MAAM,KAAK,KAAK,MAAM,QAAQ,UAAU,IAAI,IAAI;AAChF;AACF,gBACE,KAAK,KAAK,OAAO,oBACjB,CAAE,MAAM,KAAK,KAAK,MAAM,iBAAiB,UAAU,IAAI,IAAI;AAE3D;AAEF,kBAAM,YAAY,KAAK,KAAK,WACxB,MAAM,KAAK,KAAK,SAAS,UAAU,KAAK,KAAK,UAAU,IAAI,IAC3D;AACJ,kBAAM,UAAsB;AAAA,cAC1B,IAAI,KAAK,KAAK,WAAW,QAAQ;AAAA,cACjC;AAAA,cACA;AAAA,cACA;AAAA,cACA,MAAM,KAAK,KAAK,eAAe,SAAS;AAAA,cACxC,YAAY;AAAA,cACZ;AAAA,cACA,WAAW,KAAK,UAAU;AAAA,YAC5B;AACA,iBAAK,UAAU,KAAK,OAAO;AAC3B,wBAAY,KAAK,OAAO;AACxB,iBAAK,KAAK,aAAa,EAAE,MAAM,WAAW,QAAQ,CAAC;AAAA,UACrD,SAAS,KAAK;AAGZ,gBAAI,eAAe,WAAY,OAAM;AACrC,iBAAK;AACL,iBAAK;AACL,kBAAM,UAAU,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;AAC/D,iBAAK,KAAK,aAAa;AAAA,cACrB,MAAM;AAAA,cACN;AAAA,cACA,YAAY,KAAK,KAAK,WAAW,QAAQ;AAAA,cACzC;AAAA,YACF,CAAC;AACD,kBAAM,QAAQ,KAAK,KAAK,4BAA4B;AACpD,gBAAI,KAAK,yBAAyB,OAAO;AACvC,mBAAK,eAAe;AAAA,gBAClB,QAAQ;AAAA,gBACR,QAAQ,GAAG,KAAK,qBAAqB,mCAAmC,OAAO;AAAA,cACjF;AAAA,YACF;AAAA,UACF;AAAA,QACF;AAAA,QACA,KAAK,KAAK,eAAe;AAAA,QACzB,KAAK,KAAK;AAAA,MACZ;AAAA,IACF;AAEA,SAAK,KAAK,aAAa,EAAE,MAAM,SAAS,UAAU,KAAK,UAAU,QAAQ,KAAK,KAAK,OAAO,CAAC;AAC3F,WAAO,EAAE,MAAM,cAAc,UAAU,YAAY;AAAA,EACrD;AAAA;AAAA;AAAA,EAIA,MAAM,MAA+B;AACnC,WACE,KAAK,WAAW,KAAK,KAAK,UAC1B,CAAC,KAAK,cAAc,KACpB,KAAK,iBAAiB,UACtB,CAAC,KAAK,KAAK,QAAQ,SACnB;AACA,YAAM,EAAE,KAAK,IAAI,MAAM,KAAK,KAAK;AACjC,UAAI,SAAS,KAAK,KAAK,iBAAiB,OAAW;AAAA,IACrD;AACA,WAAO,KAAK,QAAQ;AAAA,EACtB;AAAA,EAEA,WAA2B;AACzB,WAAO,KAAK,QAAQ,EAAE;AAAA,EACxB;AAAA,EAEA,WAAyB;AACvB,WAAO,CAAC,GAAG,KAAK,SAAS,EAAE,KAAK,CAAC,GAAG,MAAM,EAAE,WAAW,EAAE,QAAQ;AAAA,EACnE;AAAA,EAEA,UAA0B;AACxB,WAAO,aAAa;AAAA,MAClB,QAAQ,KAAK,KAAK;AAAA,MAClB,WAAW,KAAK,UAAU;AAAA,MAC1B,OAAO,KAAK;AAAA,MACZ,KAAK,KAAK;AAAA,MACV,WAAW,KAAK;AAAA,MAChB,SAAS,CAAC,GAAG,KAAK,aAAa,OAAO,CAAC;AAAA,MACvC,UAAU,KAAK;AAAA,MACf,mBAAmB,KAAK;AAAA,MACxB,UAAU,KAAK;AAAA,MACf,MAAM,KAAK,KAAK,SACZ,EAAE,SAAS,KAAK,eAAe,iBAAiB,KAAK,gBAAgB,IACrE;AAAA,MACJ,YAAY,KAAK;AAAA,MACjB,cAAc,KAAK;AAAA,IACrB,CAAC;AAAA,EACH;AACF;;;ACnVA,eAAsB,UACpB,MACsC;AACtC,QAAM,EAAE,kBAAkB,GAAG,KAAK,IAAI;AACtC,QAAM,WAAW,IAAI,iBAAoB;AAAA,IACvC,GAAG;AAAA,IACH,WAAW,qBAAqB,oBAAoB,GAAG;AAAA,EACzD,CAAC;AACD,SAAO,EAAE,SAAS,MAAM,SAAS,IAAI,EAAE;AACzC;;;ACfO,SAAS,gBAAmB,MAA6D;AAC9F,QAAM,UAAU,KAAK,OAAO,CAAC,MAA6B,KAAK,IAAI;AACnE,SAAO;AAAA,IACL,SAAS,OAAO,UAAU,IAAI,SAAS;AACrC,iBAAW,KAAK,SAAS;AACvB,YAAI,EAAE,WAAW,CAAE,MAAM,EAAE,QAAQ,UAAU,IAAI,IAAI,EAAI,QAAO;AAAA,MAClE;AACA,aAAO;AAAA,IACT;AAAA,IACA,kBAAkB,OAAO,UAAU,IAAI,SAAS;AAC9C,iBAAW,KAAK,SAAS;AACvB,YAAI,EAAE,oBAAoB,CAAE,MAAM,EAAE,iBAAiB,UAAU,IAAI,IAAI,EAAI,QAAO;AAAA,MACpF;AACA,aAAO;AAAA,IACT;AAAA,EACF;AACF;AASO,SAAS,0BAA6B,MAMxB;AACnB,QAAM,YAAY,KAAK,oBAAoB;AAC3C,SAAO;AAAA,IACL,kBAAkB,OAAO,UAAa,KAAiB,SAAe;AACpE,YAAM,YAAY,KAAK,QAAQ,QAAQ;AACvC,UAAI,aAAa,KAAM,QAAO;AAC9B,YAAM,KAAK,MAAM,KAAK,SAAS,WAAW,IAAI;AAC9C,aAAO,GAAG,QAAQ;AAAA,IACpB;AAAA,EACF;AACF;AAMO,SAAS,kBAAqB,MAGhB;AACnB,QAAM,YAAY,KAAK,oBAAoB;AAC3C,QAAM,SAAS,KAAK,UAAU;AAC9B,SAAO;AAAA,IACL,SAAS,CAAC,WAAW,OAAmB,GAAG,SAAS,YAAY;AAAA,EAClE;AACF;;;AC/CA,IAAM,UAAU,EAAE,MAAM,UAAU,YAAY,CAAC,GAAG,sBAAsB,MAAM;AAEvE,SAAS,iBAAoB,UAAiD;AACnF,SAAO;AAAA,IACL;AAAA,MACE,MAAM;AAAA,MACN,aACE;AAAA,MACF,YAAY;AAAA,MACZ,SAAS,YAAY;AACnB,cAAM,EAAE,MAAM,SAAS,IAAI,MAAM,SAAS,KAAK;AAC/C,eAAO,EAAE,MAAM,aAAa,SAAS,QAAQ,SAAS;AAAA,MACxD;AAAA,IACF;AAAA,IACA;AAAA,MACE,MAAM;AAAA,MACN,aACE;AAAA,MACF,YAAY;AAAA,MACZ,SAAS,YAAY,SAAS,SAAS;AAAA,IACzC;AAAA,IACA;AAAA,MACE,MAAM;AAAA,MACN,aAAa;AAAA,MACb,YAAY;AAAA,QACV,MAAM;AAAA,QACN,YAAY,EAAE,OAAO,EAAE,MAAM,UAAU,aAAa,yBAAyB,EAAE;AAAA,QAC/E,sBAAsB;AAAA,MACxB;AAAA,MACA,SAAS,OAAO,SAAS;AACvB,cAAM,QAAS,MAAyC;AACxD,cAAM,MAAM,SAAS,SAAS;AAC9B,eAAO,OAAO,UAAU,WAAW,IAAI,MAAM,GAAG,KAAK,IAAI;AAAA,MAC3D;AAAA,IACF;AAAA,IACA;AAAA,MACE,MAAM;AAAA,MACN,aACE;AAAA,MACF,YAAY;AAAA,QACV,MAAM;AAAA,QACN,YAAY;AAAA,UACV,QAAQ,EAAE,MAAM,UAAU,MAAM,CAAC,QAAQ,MAAM,EAAE;AAAA,UACjD,aAAa,EAAE,MAAM,UAAU,aAAa,uCAAuC;AAAA,QACrF;AAAA,QACA,sBAAsB;AAAA,MACxB;AAAA,MACA,SAAS,OAAO,SAAS;AACvB,cAAM,IAAK,QAAQ,CAAC;AACpB,cAAM,UAAU,SAAS,QAAQ;AACjC,YAAI,EAAE,WAAW,OAAQ,QAAO,kBAAkB,SAAS,EAAE,aAAa,EAAE,YAAY,CAAC;AACzF,eAAO;AAAA,MACT;AAAA,IACF;AAAA,EACF;AACF;","names":["cellId"]}
+{"version":3,"sources":["../src/fuzz/cube.ts","../src/fuzz/capsule.ts","../src/fuzz/policies.ts","../src/fuzz/explorer.ts","../src/fuzz/fuzz-agent.ts","../src/fuzz/gates.ts","../src/fuzz/tools.ts"],"sourcesContent":["/**\n * Input-space tiling + coverage projection.\n *\n * Cells are the cartesian product of the input axes — the stratification plan,\n * enumerable up front so the planned-vs-covered denominator is honest. Coverage\n * is projected from the evaluation log: per cell, the full DISTRIBUTION of the\n * headline score, of each scored dimension, and of evaluation latency — a bare\n * mean hides outliers, so every aggregate carries its spread. Per-cell cost is\n * split known-dollars vs unknown-runs, never folded into a fabricated $0.\n */\n\nimport type { BehaviorSpace, Cell, CoverageCell, Distribution, Evaluation } from './types'\n\n/** One recorded evaluation — the unit coverage and the capsule are built from. */\nexport interface EvalRecord {\n  cell: Cell\n  ev: Evaluation\n  /** The objective's interest score for this evaluation. */\n  interest: number\n  /** Evaluation wall-clock — engine-measured unless `ev.latencyMs` overrode it. */\n  latencyMs: number\n  /** Known dollars for this run. `null` = cost tracking was wired but this\n   *  run's cost was unknowable (counted apart). Absent = not tracked at all. */\n  costUsd?: number | null\n}\n\n/** Enumerate every input cell (cartesian product of the axes), in stable order. */\nexport function enumerateCells(space: BehaviorSpace): Cell[] {\n  if (space.axes.length === 0) return []\n  let partials: Array<Record<string, string>> = [{}]\n  for (const axis of space.axes) {\n    const next: Array<Record<string, string>> = []\n    for (const partial of partials) {\n      for (const value of axis.values) next.push({ ...partial, [axis.name]: value })\n    }\n    partials = next\n  }\n  return partials.map((coords) => ({ id: cellId(space, coords), coords }))\n}\n\n/** Deterministic id for a coordinate map, e.g. `matterType=nda|difficulty=hard`. */\nexport function cellId(space: BehaviorSpace, coords: Record<string, string>): string {\n  return space.axes.map((a) => `${a.name}=${coords[a.name]}`).join('|')\n}\n\n/** Nearest-rank percentile on a pre-sorted ascending sample. */\nfunction percentile(sorted: number[], p: number): number {\n  const idx = Math.min(sorted.length - 1, Math.max(0, Math.ceil(p * sorted.length) - 1))\n  return sorted[idx] as number\n}\n\n/** Summarize a sample. Throws on an empty sample — callers represent \"no data\"\n *  as `null`, never as a zeroed distribution. */\nexport function distribution(values: number[]): Distribution {\n  if (values.length === 0)\n    throw new Error('distribution: empty sample — represent missing data as null, not zeros')\n  const sorted = [...values].sort((a, b) => a - b)\n  const mean = sorted.reduce((a, b) => a + b, 0) / sorted.length\n  return {\n    mean,\n    median: percentile(sorted, 0.5),\n    p90: percentile(sorted, 0.9),\n    min: sorted[0] as number,\n    max: sorted[sorted.length - 1] as number,\n    n: sorted.length,\n  }\n}\n\n/**\n * Project the evaluation log into the per-input-cell coverage map. A cell with\n * no evaluations reports `score: null` (honestly uncovered), never zeros.\n */\nexport function buildCoverage(cells: Cell[], log: EvalRecord[], threshold: number): CoverageCell[] {\n  const byCell = new Map<string, EvalRecord[]>()\n  for (const r of log) {\n    const arr = byCell.get(r.cell.id) ?? []\n    arr.push(r)\n    byCell.set(r.cell.id, arr)\n  }\n  return cells.map((cell) => {\n    const recs = byCell.get(cell.id) ?? []\n    const runs = recs.length\n    if (runs === 0)\n      return { cell, runs: 0, score: null, findingRate: 0, dimensions: {}, latencyMs: null }\n\n    const score = distribution(recs.map((r) => r.ev.score))\n    const latencyMs = distribution(recs.map((r) => r.latencyMs))\n    const findingRate = recs.filter((r) => r.interest >= threshold).length / runs\n\n    const dimSamples: Record<string, number[]> = {}\n    for (const r of recs) {\n      for (const [k, v] of Object.entries(r.ev.scores ?? {})) {\n        ;(dimSamples[k] ??= []).push(v)\n      }\n    }\n    const dimensions: Record<string, Distribution> = {}\n    for (const [k, xs] of Object.entries(dimSamples)) dimensions[k] = distribution(xs)\n\n    // Cost fields appear only when tracking was wired: known dollars sum, and\n    // tracked-but-unknown runs counted apart — never folded in as $0.\n    const tracked = recs.filter((r) => r.costUsd !== undefined)\n    const known = tracked.filter((r) => r.costUsd !== null)\n    const cost =\n      tracked.length > 0\n        ? {\n            costUsd: known.reduce((a, r) => a + (r.costUsd as number), 0),\n            ...(tracked.length > known.length\n              ? { costUnknownRuns: tracked.length - known.length }\n              : {}),\n          }\n        : {}\n\n    return { cell, runs, score, findingRate, dimensions, latencyMs, ...cost }\n  })\n}\n","/**\n * The capsule — the artifact every exploration produces.\n *\n * `buildCapsule` assembles coverage + verified findings + the QD archive into a\n * pure `CapsuleData` (no clock, no I/O — deterministic and snapshot-testable).\n * `renderCapsuleHtml` turns it into a standalone page: the input-cell heat-map\n * (planned vs covered), per-dimension weakness chips, and the minimized finding\n * exemplars. One artifact — the hardening map and the shareable proof object.\n */\n\nimport type { EvalRecord } from './cube'\nimport { buildCoverage, distribution } from './cube'\nimport type { ArchiveEntry, CapsuleData, Cell, CoverageCell, Distribution, Finding } from './types'\n\nexport interface BuildCapsuleInput<S> {\n  target: string\n  objective: string\n  cells: Cell[]\n  log: EvalRecord[]\n  /** The objective's notable threshold — drives findingRate. */\n  threshold: number\n  archive: ArchiveEntry<S>[]\n  findings: Finding<S>[]\n  candidateFindings: number\n  runsUsed: number\n  /** Known-dollar / unknown-run split — present only when cost tracking was\n   *  wired; the capsule never fabricates a $0 total. */\n  cost?: { costUsd: number; costUnknownRuns: number }\n  /** Evaluations that threw — infra outcomes, never folded into robustness. */\n  evalErrors: number\n  /** Set when the consecutive-error circuit breaker stopped the run early. */\n  stoppedEarly?: { reason: 'eval-errors'; detail: string }\n}\n\nexport function buildCapsule<S>(input: BuildCapsuleInput<S>): CapsuleData<S> {\n  const coverage = buildCoverage(input.cells, input.log, input.threshold)\n  const covered = coverage.filter((c) => c.runs > 0)\n  // Cells weigh equally: variance steering sends more runs to weak cells, so a\n  // run-weighted average would bias the headline low.\n  const robustness =\n    covered.length === 0 ? null : distribution(covered.map((c) => (c.score as Distribution).mean))\n  const latencyMs = input.log.length === 0 ? null : distribution(input.log.map((r) => r.latencyMs))\n  // Measured-descriptor bins beyond the bare input cell — observed, never planned.\n  const behaviorBinsObserved = input.archive.filter((e) => e.binId !== e.cell.id).length\n\n  return {\n    target: input.target,\n    objective: input.objective,\n    coverage,\n    findings: [...input.findings].sort((a, b) => b.interest - a.interest),\n    archive: [...input.archive].sort((a, b) => b.interest - a.interest),\n    stats: {\n      totalRuns: input.runsUsed,\n      cellsTotal: input.cells.length,\n      cellsCovered: covered.length,\n      behaviorBinsObserved,\n      candidateFindings: input.candidateFindings,\n      verifiedFindings: input.findings.length,\n      robustness,\n      latencyMs,\n      ...(input.cost\n        ? { costUsd: input.cost.costUsd, costUnknownRuns: input.cost.costUnknownRuns }\n        : {}),\n      evalErrors: input.evalErrors,\n      ...(input.stoppedEarly ? { stoppedEarly: input.stoppedEarly } : {}),\n    },\n  }\n}\n\n// ── HTML capsule ──────────────────────────────────────────────────────────────\n\nfunction esc(s: string): string {\n  return s.replace(\n    /[&<>\"]/g,\n    (c) => ({ '&': '&amp;', '<': '&lt;', '>': '&gt;', '\"': '&quot;' })[c] as string,\n  )\n}\n\n/** red (0) → amber (.5) → green (1); uncovered cells render gray. */\nfunction robustnessColor(r: number | null): string {\n  if (r == null) return '#2a2a2e'\n  return `hsl(${Math.round(r * 120)} 70% 42%)`\n}\n\nfunction pct(x: number): string {\n  return `${Math.round(x * 100)}%`\n}\n\nfunction deriveAxes(coverage: CoverageCell[]): Array<{ name: string; values: string[] }> {\n  const order: string[] = []\n  const seen = new Map<string, Set<string>>()\n  for (const c of coverage) {\n    for (const [k, v] of Object.entries(c.cell.coords)) {\n      if (!seen.has(k)) {\n        seen.set(k, new Set())\n        order.push(k)\n      }\n      seen.get(k)?.add(v)\n    }\n  }\n  return order.map((name) => ({ name, values: [...(seen.get(name) ?? [])] }))\n}\n\n/** The weakest dimension chip for a cell, e.g. `safety 32%` — shown when scores exist. */\nfunction weakestDim(c: CoverageCell): string {\n  const entries = Object.entries(c.dimensions).map(([k, d]) => [k, d.mean] as [string, number])\n  if (entries.length === 0) return ''\n  const sorted = entries.sort((a, b) => a[1] - b[1])\n  const w = sorted[0]\n  if (!w) return ''\n  return `<span class=\"dim\">${esc(w[0])} ${pct(w[1])}</span>`\n}\n\nfunction heatmapHtml(coverage: CoverageCell[]): string {\n  const axes = deriveAxes(coverage)\n  const byId = new Map(coverage.map((c) => [c.cell.id, c]))\n  const tile = (c: CoverageCell | undefined, label: string): string => {\n    const r = c?.score?.mean ?? null\n    const title = c?.score\n      ? `${pct(c.score.mean)} robust (median ${pct(c.score.median)}, min ${pct(c.score.min)}) · ${c.runs} runs · ${pct(c.findingRate)} flagged${c.latencyMs ? ` · ${(c.latencyMs.median / 1000).toFixed(1)}s median` : ''}${c.costUsd !== undefined ? ` · $${c.costUsd.toFixed(2)}` : ''}`\n      : 'not covered'\n    return `<div class=\"tile\" style=\"background:${robustnessColor(r)}\" title=\"${esc(title)}\">${label ? `<span class=\"tl\">${esc(label)}</span>` : ''}<span class=\"tv\">${c && r != null ? pct(r) : '—'}</span>${c ? weakestDim(c) : ''}</div>`\n  }\n\n  const rowAxis = axes[0]\n  const colAxis = axes[1]\n  if (axes.length === 2 && rowAxis && colAxis) {\n    const head = `<tr><th></th>${colAxis.values.map((v) => `<th>${esc(v)}</th>`).join('')}</tr>`\n    const rows = rowAxis.values\n      .map((rv) => {\n        const cells = colAxis.values\n          .map((cv) => {\n            const id = `${rowAxis.name}=${rv}|${colAxis.name}=${cv}`\n            return `<td>${tile(byId.get(id), '')}</td>`\n          })\n          .join('')\n        return `<tr><th class=\"rh\">${esc(rv)}</th>${cells}</tr>`\n      })\n      .join('')\n    return `<div class=\"axis-label\">rows: <b>${esc(rowAxis.name)}</b> · cols: <b>${esc(colAxis.name)}</b></div><table class=\"heat\">${head}${rows}</table>`\n  }\n\n  const sorted = [...coverage].sort((a, b) => (a.score?.mean ?? 2) - (b.score?.mean ?? 2))\n  return `<div class=\"grid\">${sorted.map((c) => tile(c, Object.values(c.cell.coords).join(' · '))).join('')}</div>`\n}\n\nfunction findingsHtml<S>(findings: Finding<S>[], limit: number): string {\n  if (findings.length === 0)\n    return `<p class=\"none\">No verified findings — the target held across every covered cell.</p>`\n  return findings\n    .slice(0, limit)\n    .map((f) => {\n      const coords = Object.entries(f.cell.coords)\n        .map(([k, v]) => `${k}:${v}`)\n        .join(' · ')\n      const labels = (f.evaluation.labels ?? [])\n        .map((l) => `<span class=\"fclass\">${esc(l)}</span>`)\n        .join('')\n      const dims = Object.entries(f.evaluation.scores ?? {})\n        .sort((a, b) => a[1] - b[1])\n        .slice(0, 3)\n        .map(([k, v]) => `<span class=\"dim\">${esc(k)} ${pct(v)}</span>`)\n        .join('')\n      return `<div class=\"fail\"><div class=\"fmeta\"><span class=\"fcell\">${esc(coords)}</span>${labels}<span class=\"fsev\">interest ${pct(f.interest)}</span></div><div class=\"sevbar\"><div class=\"sevfill\" style=\"width:${pct(f.interest)}\"></div></div><div class=\"ftext\">${esc(f.text ?? '(scenario text not captured)')}</div>${dims ? `<div class=\"fdims\">${dims}</div>` : ''}</div>`\n    })\n    .join('')\n}\n\nexport interface RenderCapsuleOptions {\n  /** Max finding exemplars to show. Default 8. */\n  maxFindings?: number\n  /** ISO timestamp to stamp into the page (keeps the pure capsule clock-free). */\n  generatedAt?: string\n}\n\n/** Render a self-contained HTML capsule — heat-map + per-dimension chips + verified findings. */\nexport function renderCapsuleHtml<S>(\n  capsule: CapsuleData<S>,\n  opts: RenderCapsuleOptions = {},\n): string {\n  const s = capsule.stats\n  const kpi = (label: string, value: string, accent = '#e6e6e6'): string =>\n    `<div class=\"kpi\"><div class=\"kv\" style=\"color:${accent}\">${esc(value)}</div><div class=\"kl\">${esc(label)}</div></div>`\n  const lift = capsule.lift\n    ? kpi(\n        'hardening lift',\n        `${capsule.lift.before.toFixed(2)} → ${capsule.lift.after.toFixed(2)}`,\n        '#5ad17a',\n      )\n    : ''\n  // Cost KPI only when tracking was wired — an untracked run never shows $0.\n  // Unpriced runs are named in the label (amber): the total is a lower bound.\n  const cost =\n    s.costUsd !== undefined\n      ? kpi(\n          s.costUnknownRuns ? `cost · ${s.costUnknownRuns} runs unpriced` : 'cost',\n          `$${s.costUsd.toFixed(2)}`,\n          s.costUnknownRuns ? '#e5c07b' : '#e6e6e6',\n        )\n      : ''\n  const stamp = opts.generatedAt ?? capsule.generatedAt ?? ''\n\n  return `<!doctype html><html><head><meta charset=\"utf-8\"><meta name=\"viewport\" content=\"width=device-width,initial-scale=1\">\n<title>${esc(capsule.objective)} capsule — ${esc(capsule.target)}</title>\n<style>\n:root{color-scheme:dark}\n*{box-sizing:border-box}\nbody{margin:0;background:#0c0c0f;color:#e6e6e6;font:14px/1.5 ui-sans-serif,system-ui,-apple-system,Segoe UI,Roboto,sans-serif}\n.wrap{max-width:980px;margin:0 auto;padding:40px 24px}\nh1{font-size:22px;margin:0 0 2px;letter-spacing:-.01em}\n.sub{color:#8a8a93;font-size:13px;margin-bottom:28px}\n.kpis{display:flex;gap:14px;flex-wrap:wrap;margin-bottom:32px}\n.kpi{background:#16161b;border:1px solid #24242b;border-radius:12px;padding:14px 18px;min-width:120px}\n.kv{font-size:24px;font-weight:650;letter-spacing:-.02em}\n.kl{color:#8a8a93;font-size:12px;margin-top:2px}\nh2{font-size:13px;text-transform:uppercase;letter-spacing:.08em;color:#8a8a93;margin:34px 0 14px}\n.axis-label{color:#8a8a93;font-size:12px;margin-bottom:10px}\ntable.heat{border-collapse:separate;border-spacing:6px}\ntable.heat th{font-weight:500;color:#a8a8b0;font-size:12px;padding:4px 8px;text-align:center}\ntable.heat th.rh{text-align:right}\n.tile{position:relative;width:104px;height:66px;border-radius:10px;display:flex;flex-direction:column;justify-content:center;align-items:center;gap:1px;color:#06120a;font-weight:600}\n.tile .tv{font-size:17px}\n.tile .tl{font-size:10px;font-weight:600;opacity:.9;padding:0 6px;text-align:center}\n.tile .dim{font-size:9px;font-weight:600;opacity:.85;background:rgba(0,0,0,.22);padding:0 6px;border-radius:999px}\n.grid{display:flex;flex-wrap:wrap;gap:8px}\n.grid .tile{width:138px;height:74px}\n.fail{background:#16161b;border:1px solid #24242b;border-left:3px solid #d1495b;border-radius:10px;padding:12px 14px;margin-bottom:10px}\n.fmeta{display:flex;gap:10px;align-items:center;font-size:12px;color:#a8a8b0;margin-bottom:8px}\n.fcell{color:#cfcfd6}\n.fclass{background:#2a1a1d;color:#e58a96;padding:1px 8px;border-radius:999px;font-size:11px}\n.fsev{margin-left:auto;color:#e58a96}\n.sevbar{height:4px;background:#24242b;border-radius:2px;overflow:hidden;margin-bottom:10px}\n.sevfill{height:100%;background:#d1495b}\n.ftext{font-size:13px;color:#d8d8df;white-space:pre-wrap}\n.fdims{display:flex;gap:6px;margin-top:8px}\n.fdims .dim{background:#1d2530;color:#9fc1e8;padding:1px 8px;border-radius:999px;font-size:11px}\n.none{color:#5ad17a}\n.foot{color:#5a5a63;font-size:11px;margin-top:36px}\n</style></head><body><div class=\"wrap\">\n<h1>${esc(capsule.objective)} exploration · ${esc(capsule.target)}</h1>\n<div class=\"sub\">${s.totalRuns} scenarios across ${s.cellsCovered}/${s.cellsTotal} planned cells${s.behaviorBinsObserved > 0 ? ` · ${s.behaviorBinsObserved} measured behavior bins` : ''}${stamp ? ` · ${esc(stamp)}` : ''}</div>\n<div class=\"kpis\">\n${s.robustness ? kpi('robustness', `${pct(s.robustness.mean)}`, s.robustness.mean < 0.6 ? '#e58a96' : '#5ad17a') : ''}\n${s.robustness ? kpi('cell spread', `${pct(s.robustness.min)}–${pct(s.robustness.max)}`) : ''}\n${s.latencyMs ? kpi('median latency', `${(s.latencyMs.median / 1000).toFixed(1)}s`, s.latencyMs.p90 > 4 * s.latencyMs.median ? '#e5b566' : '#e6e6e6') : ''}\n${kpi('verified findings', String(s.verifiedFindings), s.verifiedFindings > 0 ? '#e58a96' : '#5ad17a')}\n${kpi('cells covered', `${s.cellsCovered}/${s.cellsTotal}`)}\n${kpi('scenarios run', String(s.totalRuns))}\n${cost}\n${s.evalErrors > 0 ? kpi('eval errors', String(s.evalErrors), '#e5b566') : ''}\n${lift}\n</div>\n${s.stoppedEarly ? `<div class=\"sub\" style=\"color:#e5b566\">stopped early: ${esc(s.stoppedEarly.detail)} — coverage below reflects the completed portion only</div>` : ''}\n<h2>Coverage map</h2>\n${heatmapHtml(capsule.coverage)}\n<h2>Verified findings${s.candidateFindings > s.verifiedFindings ? ` · ${s.verifiedFindings} of ${s.candidateFindings} candidates passed the validity gates` : ''}</h2>\n${findingsHtml(capsule.findings, opts.maxFindings ?? 8)}\n<div class=\"foot\">Findings are gate-verified: each is a fair, answerable task that reproduces under a meaning-preserving rephrase, minimized to the smallest trigger.</div>\n</div></body></html>`\n}\n","/**\n * Shipped policies for the exploration engine.\n *\n * `Proposer` is a plain function type — an agent running a generator skill IS a\n * proposer (`(ctx) => dispatchToSkill(ctx)`), no wrapper needed. `mutationProposer`\n * builds the deterministic, LLM-free one from mutation operators. Objectives are\n * interfaces because the engine reads `kind` + `threshold` off them.\n */\n\nimport type { AdversarialMutation } from '../rl/adversarial'\nimport type {\n  Cell,\n  Evaluation,\n  Objective,\n  ObjectiveContext,\n  ProposeContext,\n  Proposer,\n} from './types'\n\nconst clamp01 = (x: number): number => Math.max(0, Math.min(1, x))\n\n// ── proposers ─────────────────────────────────────────────────────────────────\n\n/**\n * Perturbation-based search: apply the cell's mutation operators to the current\n * elites + seeds, deduping by id. Elites first — mutating the most interesting\n * scenario found so far is what makes the search deepen across rounds.\n */\nexport function mutationProposer<S>(opts: {\n  mutationsFor: (cell: Cell) => AdversarialMutation<S>[]\n  scenarioId: (s: S) => string\n}): Proposer<S> {\n  return async (ctx: ProposeContext<S>): Promise<S[]> => {\n    const mutations = opts.mutationsFor(ctx.cell)\n    const parents = [...ctx.elites, ...ctx.seeds]\n    const seen = new Set(parents.map(opts.scenarioId))\n    const out: S[] = []\n    for (const parent of parents) {\n      if (out.length >= ctx.count) break\n      for (const m of mutations) {\n        const children = await m.mutate(parent, ctx.rng)\n        for (const child of children) {\n          const id = opts.scenarioId(child)\n          if (seen.has(id)) continue\n          seen.add(id)\n          out.push(child)\n          if (out.length >= ctx.count) break\n        }\n        if (out.length >= ctx.count) break\n      }\n    }\n    return out\n  }\n}\n\n// ── objectives ────────────────────────────────────────────────────────────────\n\n/** Adversarial: a low headline score is interesting — find where the agent fails. */\nexport function adversarialObjective(threshold = 0.5): Objective {\n  return { kind: 'adversarial', threshold, interest: (ev) => clamp01(1 - ev.score) }\n}\n\nfunction hamming(\n  a: Record<string, string> | undefined,\n  b: Record<string, string> | undefined,\n): number {\n  if (!a || !b) return 1\n  const keys = new Set([...Object.keys(a), ...Object.keys(b)])\n  if (keys.size === 0) return 0\n  let diff = 0\n  for (const k of keys) if (a[k] !== b[k]) diff++\n  return diff / keys.size\n}\n\n/**\n * Novelty: interesting when far from the archive in score AND measured behavior\n * descriptor — quality-diversity's diversity pressure; drives corpus growth\n * rather than re-finding the same hole.\n */\nexport function noveltyObjective(threshold = 0.3): Objective {\n  return {\n    kind: 'novelty',\n    threshold,\n    interest: (ev: Evaluation, ctx: ObjectiveContext) => {\n      const scoreNovelty =\n        ctx.archiveScores.length === 0\n          ? 1\n          : Math.min(...ctx.archiveScores.map((s) => Math.abs(s - ev.score)))\n      const descNovelty =\n        ctx.archiveDescriptors.length === 0\n          ? 1\n          : Math.min(...ctx.archiveDescriptors.map((d) => hamming(d, ev.descriptor)))\n      return clamp01(0.5 * scoreNovelty + 0.5 * descNovelty)\n    },\n  }\n}\n","/**\n * The exploration engine — a stateful session over a behavior space.\n *\n * Each `step()`: allocate budget across INPUT cells (floor first, then variance\n * steering toward the least-certain cells), propose candidates (the proposer\n * reads current elites + findings, so the search deepens generationally),\n * evaluate with bounded concurrency, archive the most interesting scenario per\n * input×measured bin, and admit notable candidates that pass the validity gates.\n * `run()` loops to budget. `coverage()`/`findings()`/`capsule()` read live state —\n * the surface `makeExploreTools` exposes so an agent can drive the session.\n *\n * One evaluation log (`EvalRecord[]`) is the source of truth; allocation\n * observations and coverage are projections of it.\n */\n\nimport { ValidationError } from '../errors'\nimport { varianceBasedCurriculum } from '../rl/active-curriculum'\nimport { buildCapsule } from './capsule'\nimport type { EvalRecord } from './cube'\nimport { enumerateCells } from './cube'\nimport { adversarialObjective } from './policies'\nimport type {\n  ArchiveEntry,\n  CapsuleData,\n  Cell,\n  CoverageCell,\n  Evaluation,\n  ExploreOptions,\n  Finding,\n  Objective,\n} from './types'\n\nasync function pMap<T>(\n  items: T[],\n  fn: (item: T) => Promise<void>,\n  concurrency: number,\n  signal?: AbortSignal,\n): Promise<void> {\n  let i = 0\n  const workers = Array.from(\n    { length: Math.max(1, Math.min(concurrency, items.length)) },\n    async () => {\n      while (i < items.length) {\n        if (signal?.aborted) return\n        const item = items[i++]\n        if (item === undefined) return\n        await fn(item)\n      }\n    },\n  )\n  await Promise.all(workers)\n}\n\nexport class BehaviorExplorer<S> {\n  private readonly cells: Cell[]\n  private readonly cellById: Map<string, Cell>\n  private readonly objective: Objective\n  private readonly threshold: number\n  private readonly floorPerCell: number\n  private readonly perRoundBudget: number\n\n  /** The single evaluation log — coverage + allocation are projections of it. */\n  private readonly log: Array<EvalRecord & { scenarioId: string }> = []\n  /** binId (input × measured coords) → the most interesting entry seen. */\n  private readonly archiveByBin = new Map<string, ArchiveEntry<S>>()\n  private readonly _findings: Finding<S>[] = []\n  private runsUsed = 0\n  private candidateFindings = 0\n  private evalErrors = 0\n  private consecutiveEvalErrors = 0\n  private stoppedEarly: { reason: 'eval-errors'; detail: string } | undefined\n  private rngState: number\n  /** Accumulated KNOWN dollars — unknown-cost runs never inflate it. */\n  private spentKnownUsd = 0\n  private costUnknownRuns = 0\n\n  constructor(private readonly opts: ExploreOptions<S>) {\n    this.cells = enumerateCells(opts.space)\n    if (this.cells.length === 0)\n      throw new Error('BehaviorExplorer: space has no cells — every axis needs ≥1 value')\n    if (opts.costBudgetUsd !== undefined) {\n      if (\n        typeof opts.costBudgetUsd !== 'number' ||\n        !Number.isFinite(opts.costBudgetUsd) ||\n        opts.costBudgetUsd < 0\n      ) {\n        throw new RangeError(\n          `BehaviorExplorer: costBudgetUsd must be a nonnegative finite number, got ${String(opts.costBudgetUsd)}`,\n        )\n      }\n    }\n    if (!opts.costOf && (opts.costBudgetUsd !== undefined || opts.ledger || opts.onCost)) {\n      throw new ValidationError(\n        'BehaviorExplorer: costBudgetUsd/ledger/onCost require costOf — the explorer ' +\n          'cannot know run cost without it; supply costOf or drop the cost options',\n      )\n    }\n    this.cellById = new Map(this.cells.map((c) => [c.id, c]))\n    this.objective = opts.objective ?? adversarialObjective(0.5)\n    this.threshold = this.objective.threshold ?? 0.5\n    this.floorPerCell = opts.floorPerCell ?? 2\n    this.perRoundBudget = Math.max(\n      this.cells.length * this.floorPerCell,\n      Math.ceil(opts.budget / 4),\n    )\n    this.rngState = (opts.seed ?? 1) >>> 0\n  }\n\n  // mulberry32 — deterministic per session.\n  private rng = (): number => {\n    this.rngState = (this.rngState + 0x6d2b79f5) | 0\n    let t = this.rngState\n    t = Math.imul(t ^ (t >>> 15), t | 1)\n    t ^= t + Math.imul(t ^ (t >>> 7), t | 61)\n    return ((t ^ (t >>> 14)) >>> 0) / 4294967296\n  }\n\n  private binId(cell: Cell, descriptor: Record<string, string> | undefined): string {\n    if (!descriptor || Object.keys(descriptor).length === 0) return cell.id\n    const measured = Object.keys(descriptor)\n      .sort()\n      .map((k) => `${k}=${descriptor[k]}`)\n      .join('|')\n    return `${cell.id}|${measured}`\n  }\n\n  private allocate(budget: number): Array<{ cellId: string; count: number }> {\n    if ((this.opts.allocation ?? 'variance') === 'uniform') {\n      const per = Math.max(this.floorPerCell, Math.floor(budget / this.cells.length))\n      return this.cells.map((c) => ({ cellId: c.id, count: per }))\n    }\n    return varianceBasedCurriculum(\n      this.log.map((r) => ({\n        variantId: r.cell.id,\n        scenarioId: r.scenarioId,\n        score: r.ev.score,\n        pass: r.ev.valid && r.ev.score >= 0.5,\n      })),\n      this.cells.map((c) => ({ variantId: c.id, scenarioId: '*' })),\n      { budget, floorPerCell: this.floorPerCell },\n    ).map((a) => ({ cellId: a.variantId, count: a.count }))\n  }\n\n  private objectiveContext() {\n    const entries = [...this.archiveByBin.values()]\n    return {\n      archiveScores: entries.map((e) => e.evaluation.score),\n      archiveDescriptors: entries.map((e) => e.evaluation.descriptor),\n    }\n  }\n\n  /** Mirrors control-runtime: stop once accumulated KNOWN cost ≥ the ceiling. */\n  private costExhausted(): boolean {\n    return this.opts.costBudgetUsd !== undefined && this.spentKnownUsd >= this.opts.costBudgetUsd\n  }\n\n  /** Fold one run's cost in: null counts as unknown (never $0); a known cost\n   *  accrues toward the budget, lands in the ledger, and fires `onCost`. */\n  /** Returns the run's known cost, `null` when tracked-but-unknown, `undefined`\n   *  when cost tracking is not wired — the log row mirrors this exactly. */\n  private recordRunCost(scenario: S, cell: Cell, ev: Evaluation): number | null | undefined {\n    if (!this.opts.costOf) return undefined\n    const cost = this.opts.costOf(scenario, cell, ev)\n    if (cost === null) {\n      this.costUnknownRuns++\n      return null\n    }\n    if (typeof cost.usd !== 'number' || !Number.isFinite(cost.usd) || cost.usd < 0) {\n      throw new RangeError(\n        `BehaviorExplorer: costOf returned an invalid usd (${String(cost?.usd)}) — ` +\n          'return null when cost is unknown, never a fabricated number',\n      )\n    }\n    this.spentKnownUsd += cost.usd\n    this.opts.ledger?.record({\n      model: cost.model ?? 'unattributed',\n      channel: 'agent',\n      usage: { inputTokens: 0, outputTokens: 0 },\n      actualCostUsd: cost.usd,\n      tags: { target: this.opts.target, cell: cell.id },\n    })\n    this.opts.onCost?.({ usd: cost.usd, channel: 'agent' })\n    return cost.usd\n  }\n\n  /** Elites whose INPUT cell matches — what the proposer mutates/deepens from. */\n  private elitesFor(cellId: string): S[] {\n    const out: S[] = []\n    for (const e of this.archiveByBin.values()) if (e.cell.id === cellId) out.push(e.scenario)\n    return out\n  }\n\n  /** One allocate → propose → evaluate → gate → archive round. */\n  async step(): Promise<{ runs: number; findings: Finding<S>[] }> {\n    const remaining = this.opts.budget - this.runsUsed\n    if (remaining <= 0 || this.costExhausted() || this.opts.signal?.aborted)\n      return { runs: 0, findings: [] }\n\n    const allocations = this.allocate(Math.min(this.perRoundBudget, remaining))\n    const newFindings: Finding<S>[] = []\n    let runsThisStep = 0\n\n    for (const alloc of allocations) {\n      if (\n        this.runsUsed >= this.opts.budget ||\n        this.costExhausted() ||\n        this.stoppedEarly !== undefined ||\n        this.opts.signal?.aborted\n      )\n        break\n      const cell = this.cellById.get(alloc.cellId)\n      if (!cell) continue\n      const cap = Math.min(alloc.count, this.opts.budget - this.runsUsed)\n      if (cap <= 0) continue\n      this.opts.onProgress?.({ type: 'cell-allocated', cell, count: cap })\n\n      const seeds = await this.opts.seedsFor(cell)\n      const elites = this.elitesFor(cell.id)\n      const proposed = await this.opts.proposer({\n        cell,\n        seeds,\n        elites,\n        findings: this._findings,\n        count: cap,\n        rng: this.rng,\n      })\n      // Cold cells evaluate their seeds first (the coverage floor); warm cells\n      // trust the proposer, which already saw the elites.\n      const cold = this.log.every((r) => r.cell.id !== cell.id)\n      const toEval = [...(cold ? seeds : []), ...proposed].slice(0, cap)\n\n      await pMap(\n        toEval,\n        async (scenario) => {\n          if (\n            this.runsUsed >= this.opts.budget ||\n            this.costExhausted() ||\n            this.stoppedEarly !== undefined ||\n            this.opts.signal?.aborted\n          )\n            return\n          // evaluate/gates/minimize cross an external boundary (router, backend,\n          // judge). A throw there is an infra outcome: record it as a typed\n          // eval-error and keep exploring — one 5xx must not kill a campaign.\n          // Consecutive failures trip the circuit breaker instead, so a dead\n          // backend stops the run rather than burning the remaining budget.\n          try {\n            const startedAt = performance.now()\n            const ev = await this.opts.evaluate(scenario, cell)\n            // Consumer-measured latency wins (it can exclude judge time); the\n            // engine's wall-clock is the default so latency is never missing.\n            const latencyMs = ev.latencyMs ?? performance.now() - startedAt\n            this.runsUsed++\n            runsThisStep++\n            this.consecutiveEvalErrors = 0\n            const costUsd = this.recordRunCost(scenario, cell, ev)\n            const interest = this.objective.interest(ev, this.objectiveContext())\n            this.log.push({\n              cell,\n              ev,\n              interest,\n              latencyMs,\n              ...(costUsd !== undefined ? { costUsd } : {}),\n              scenarioId: this.opts.scenarioId(scenario),\n            })\n            this.opts.onProgress?.({ type: 'evaluated', cell, scenario, evaluation: ev })\n\n            const bin = this.binId(cell, ev.descriptor)\n            const cur = this.archiveByBin.get(bin)\n            if (!cur || interest > cur.interest)\n              this.archiveByBin.set(bin, { binId: bin, cell, scenario, evaluation: ev, interest })\n\n            if (interest < this.threshold) return\n            this.candidateFindings++\n            if (this.opts.gates?.isValid && !(await this.opts.gates.isValid(scenario, ev, cell)))\n              return\n            if (\n              this.opts.gates?.isUncontaminated &&\n              !(await this.opts.gates.isUncontaminated(scenario, ev, cell))\n            )\n              return\n\n            const minimized = this.opts.minimize\n              ? await this.opts.minimize(scenario, this.opts.evaluate, cell)\n              : scenario\n            const finding: Finding<S> = {\n              id: this.opts.scenarioId(scenario),\n              cell,\n              scenario,\n              minimized,\n              text: this.opts.scenarioText?.(minimized),\n              evaluation: ev,\n              interest,\n              objective: this.objective.kind,\n            }\n            this._findings.push(finding)\n            newFindings.push(finding)\n            this.opts.onProgress?.({ type: 'finding', finding })\n          } catch (err) {\n            // Internal validation errors (e.g. a fabricated costOf number) are\n            // programming mistakes, not backend outcomes — they stay loud.\n            if (err instanceof RangeError) throw err\n            this.evalErrors++\n            this.consecutiveEvalErrors++\n            const message = err instanceof Error ? err.message : String(err)\n            this.opts.onProgress?.({\n              type: 'eval-error',\n              cell,\n              scenarioId: this.opts.scenarioId(scenario),\n              message,\n            })\n            const limit = this.opts.maxConsecutiveEvalErrors ?? 5\n            if (this.consecutiveEvalErrors >= limit) {\n              this.stoppedEarly = {\n                reason: 'eval-errors',\n                detail: `${this.consecutiveEvalErrors} consecutive eval errors (last: ${message})`,\n              }\n            }\n          }\n        },\n        this.opts.concurrency ?? 1,\n        this.opts.signal,\n      )\n    }\n\n    this.opts.onProgress?.({ type: 'round', runsUsed: this.runsUsed, budget: this.opts.budget })\n    return { runs: runsThisStep, findings: newFindings }\n  }\n\n  /** Loop `step()` until the run or dollar budget is spent, the signal aborts,\n   *  or no progress is made. */\n  async run(): Promise<CapsuleData<S>> {\n    while (\n      this.runsUsed < this.opts.budget &&\n      !this.costExhausted() &&\n      this.stoppedEarly === undefined &&\n      !this.opts.signal?.aborted\n    ) {\n      const { runs } = await this.step()\n      if (runs === 0 && this.stoppedEarly === undefined) break\n    }\n    return this.capsule()\n  }\n\n  coverage(): CoverageCell[] {\n    return this.capsule().coverage\n  }\n\n  findings(): Finding<S>[] {\n    return [...this._findings].sort((a, b) => b.interest - a.interest)\n  }\n\n  capsule(): CapsuleData<S> {\n    return buildCapsule({\n      target: this.opts.target,\n      objective: this.objective.kind,\n      cells: this.cells,\n      log: this.log,\n      threshold: this.threshold,\n      archive: [...this.archiveByBin.values()],\n      findings: this._findings,\n      candidateFindings: this.candidateFindings,\n      runsUsed: this.runsUsed,\n      cost: this.opts.costOf\n        ? { costUsd: this.spentKnownUsd, costUnknownRuns: this.costUnknownRuns }\n        : undefined,\n      evalErrors: this.evalErrors,\n      stoppedEarly: this.stoppedEarly,\n    })\n  }\n}\n","/**\n * `fuzzAgent` — the adversarial batch preset over `BehaviorExplorer`.\n *\n * One call: explore the space to budget with the adversarial objective and\n * return the capsule. For agent-driven, incremental, or multi-objective use,\n * construct a `BehaviorExplorer` and drive it via `makeExploreTools`.\n */\n\nimport { BehaviorExplorer } from './explorer'\nimport { adversarialObjective } from './policies'\nimport type { CapsuleData, ExploreOptions } from './types'\n\nexport type FuzzAgentOptions<S> = Omit<ExploreOptions<S>, 'objective'> & {\n  /** Score strictly below this is a candidate failure. Default 0.5. */\n  failureThreshold?: number\n}\n\nexport async function fuzzAgent<S>(\n  opts: FuzzAgentOptions<S>,\n): Promise<{ capsule: CapsuleData<S> }> {\n  const { failureThreshold, ...rest } = opts\n  const explorer = new BehaviorExplorer<S>({\n    ...rest,\n    objective: adversarialObjective(failureThreshold ?? 0.5),\n  })\n  return { capsule: await explorer.run() }\n}\n","/**\n * Validity gates — what separates a fuzzer from a slop generator.\n *\n * A notable candidate is admitted only when it is fair and reproducible. None of\n * these are on by default: the live wiring opts in, so reported findings carry\n * their proof.\n */\n\nimport type { Cell, Evaluation, Evaluator, ValidityGates } from './types'\n\n/** Combine gate sets; a candidate must pass every gate in every set. */\nexport function composeGates<S>(...sets: Array<ValidityGates<S> | undefined>): ValidityGates<S> {\n  const present = sets.filter((s): s is ValidityGates<S> => s != null)\n  return {\n    isValid: async (scenario, ev, cell) => {\n      for (const g of present) {\n        if (g.isValid && !(await g.isValid(scenario, ev, cell))) return false\n      }\n      return true\n    },\n    isUncontaminated: async (scenario, ev, cell) => {\n      for (const g of present) {\n        if (g.isUncontaminated && !(await g.isUncontaminated(scenario, ev, cell))) return false\n      }\n      return true\n    },\n  }\n}\n\n/**\n * Reproducibility gate. Re-run the target on a meaning-preserving rephrase of the\n * flagged scenario; keep the finding only when the rephrase ALSO scores below the\n * threshold. A finding that flips under a cosmetic rewrite was keyed to surface\n * form, not the task — a false signal we must not report. Costs one extra\n * evaluation per candidate (candidates are rare, so cheap).\n */\nexport function perturbationStabilityGate<S>(opts: {\n  evaluate: Evaluator<S>\n  /** Produce a semantic-preserving rephrase. Return null to skip (treated as pass). */\n  perturb: (scenario: S) => S | null\n  /** Score strictly below this still counts as failing. Default 0.5. */\n  failureThreshold?: number\n}): ValidityGates<S> {\n  const threshold = opts.failureThreshold ?? 0.5\n  return {\n    isUncontaminated: async (scenario: S, _ev: Evaluation, cell: Cell) => {\n      const rephrased = opts.perturb(scenario)\n      if (rephrased == null) return true\n      const re = await opts.evaluate(rephrased, cell)\n      return re.score < threshold\n    },\n  }\n}\n\n/**\n * Severity-floor gate. Reject borderline candidates whose score sits in a band\n * just under the threshold — judge noise, not a real defect.\n */\nexport function severityFloorGate<S>(opts: {\n  failureThreshold?: number\n  margin?: number\n}): ValidityGates<S> {\n  const threshold = opts.failureThreshold ?? 0.5\n  const margin = opts.margin ?? 0.1\n  return {\n    isValid: (_scenario, ev: Evaluation) => ev.score <= threshold - margin,\n  }\n}\n","/**\n * Agent-drivable surface over a live exploration session.\n *\n * Framework-neutral tool defs ({name, description, parameters: JSON Schema,\n * handler}) so the on-demand agent — not a batch script — drives the search:\n * step it, read coverage, inspect findings, render the capsule. Transport\n * encodings (OpenAI function shape, MCP) are one-line mappings the host owns.\n */\n\nimport { renderCapsuleHtml } from './capsule'\nimport type { BehaviorExplorer } from './explorer'\n\nexport interface ExploreToolDef {\n  name: string\n  description: string\n  /** JSON Schema (draft-07+) for the arguments. */\n  parameters: Record<string, unknown>\n  handler: (args: unknown, ctx?: { signal?: AbortSignal }) => Promise<unknown>\n}\n\nconst NO_ARGS = { type: 'object', properties: {}, additionalProperties: false }\n\nexport function makeExploreTools<S>(explorer: BehaviorExplorer<S>): ExploreToolDef[] {\n  return [\n    {\n      name: 'explore_step',\n      description:\n        'Run one exploration round: allocate budget across cells, propose + evaluate scenarios, archive elites, admit gate-verified findings. Returns runs spent and new findings.',\n      parameters: NO_ARGS,\n      handler: async () => {\n        const { runs, findings } = await explorer.step()\n        return { runs, newFindings: findings.length, findings }\n      },\n    },\n    {\n      name: 'explore_coverage',\n      description:\n        'Read the live coverage map: per planned cell — runs, robustness (null = uncovered), finding rate, per-dimension means.',\n      parameters: NO_ARGS,\n      handler: async () => explorer.coverage(),\n    },\n    {\n      name: 'explore_findings',\n      description: 'List gate-verified findings so far, sorted by descending interest.',\n      parameters: {\n        type: 'object',\n        properties: { limit: { type: 'number', description: 'max findings to return' } },\n        additionalProperties: false,\n      },\n      handler: async (args) => {\n        const limit = (args as { limit?: number } | undefined)?.limit\n        const all = explorer.findings()\n        return typeof limit === 'number' ? all.slice(0, limit) : all\n      },\n    },\n    {\n      name: 'explore_capsule',\n      description:\n        'Build the capsule artifact from the current session state. format \"data\" returns the structured CapsuleData; \"html\" returns the standalone page.',\n      parameters: {\n        type: 'object',\n        properties: {\n          format: { type: 'string', enum: ['data', 'html'] },\n          generatedAt: { type: 'string', description: 'ISO timestamp to stamp into the page' },\n        },\n        additionalProperties: false,\n      },\n      handler: async (args) => {\n        const a = (args ?? {}) as { format?: string; generatedAt?: string }\n        const capsule = explorer.capsule()\n        if (a.format === 'html') return renderCapsuleHtml(capsule, { generatedAt: a.generatedAt })\n        return capsule\n      },\n    },\n  ]\n}\n"],"mappings":";;;;;;;;;AA2BO,SAAS,eAAe,OAA8B;AAC3D,MAAI,MAAM,KAAK,WAAW,EAAG,QAAO,CAAC;AACrC,MAAI,WAA0C,CAAC,CAAC,CAAC;AACjD,aAAW,QAAQ,MAAM,MAAM;AAC7B,UAAM,OAAsC,CAAC;AAC7C,eAAW,WAAW,UAAU;AAC9B,iBAAW,SAAS,KAAK,OAAQ,MAAK,KAAK,EAAE,GAAG,SAAS,CAAC,KAAK,IAAI,GAAG,MAAM,CAAC;AAAA,IAC/E;AACA,eAAW;AAAA,EACb;AACA,SAAO,SAAS,IAAI,CAAC,YAAY,EAAE,IAAI,OAAO,OAAO,MAAM,GAAG,OAAO,EAAE;AACzE;AAGO,SAAS,OAAO,OAAsB,QAAwC;AACnF,SAAO,MAAM,KAAK,IAAI,CAAC,MAAM,GAAG,EAAE,IAAI,IAAI,OAAO,EAAE,IAAI,CAAC,EAAE,EAAE,KAAK,GAAG;AACtE;AAGA,SAAS,WAAW,QAAkB,GAAmB;AACvD,QAAM,MAAM,KAAK,IAAI,OAAO,SAAS,GAAG,KAAK,IAAI,GAAG,KAAK,KAAK,IAAI,OAAO,MAAM,IAAI,CAAC,CAAC;AACrF,SAAO,OAAO,GAAG;AACnB;AAIO,SAAS,aAAa,QAAgC;AAC3D,MAAI,OAAO,WAAW;AACpB,UAAM,IAAI,MAAM,6EAAwE;AAC1F,QAAM,SAAS,CAAC,GAAG,MAAM,EAAE,KAAK,CAAC,GAAG,MAAM,IAAI,CAAC;AAC/C,QAAM,OAAO,OAAO,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,OAAO;AACxD,SAAO;AAAA,IACL;AAAA,IACA,QAAQ,WAAW,QAAQ,GAAG;AAAA,IAC9B,KAAK,WAAW,QAAQ,GAAG;AAAA,IAC3B,KAAK,OAAO,CAAC;AAAA,IACb,KAAK,OAAO,OAAO,SAAS,CAAC;AAAA,IAC7B,GAAG,OAAO;AAAA,EACZ;AACF;AAMO,SAAS,cAAc,OAAe,KAAmB,WAAmC;AACjG,QAAM,SAAS,oBAAI,IAA0B;AAC7C,aAAW,KAAK,KAAK;AACnB,UAAM,MAAM,OAAO,IAAI,EAAE,KAAK,EAAE,KAAK,CAAC;AACtC,QAAI,KAAK,CAAC;AACV,WAAO,IAAI,EAAE,KAAK,IAAI,GAAG;AAAA,EAC3B;AACA,SAAO,MAAM,IAAI,CAAC,SAAS;AACzB,UAAM,OAAO,OAAO,IAAI,KAAK,EAAE,KAAK,CAAC;AACrC,UAAM,OAAO,KAAK;AAClB,QAAI,SAAS;AACX,aAAO,EAAE,MAAM,MAAM,GAAG,OAAO,MAAM,aAAa,GAAG,YAAY,CAAC,GAAG,WAAW,KAAK;AAEvF,UAAM,QAAQ,aAAa,KAAK,IAAI,CAAC,MAAM,EAAE,GAAG,KAAK,CAAC;AACtD,UAAM,YAAY,aAAa,KAAK,IAAI,CAAC,MAAM,EAAE,SAAS,CAAC;AAC3D,UAAM,cAAc,KAAK,OAAO,CAAC,MAAM,EAAE,YAAY,SAAS,EAAE,SAAS;AAEzE,UAAM,aAAuC,CAAC;AAC9C,eAAW,KAAK,MAAM;AACpB,iBAAW,CAAC,GAAG,CAAC,KAAK,OAAO,QAAQ,EAAE,GAAG,UAAU,CAAC,CAAC,GAAG;AACtD;AAAC,SAAC,WAAW,CAAC,MAAM,CAAC,GAAG,KAAK,CAAC;AAAA,MAChC;AAAA,IACF;AACA,UAAM,aAA2C,CAAC;AAClD,eAAW,CAAC,GAAG,EAAE,KAAK,OAAO,QAAQ,UAAU,EAAG,YAAW,CAAC,IAAI,aAAa,EAAE;AAIjF,UAAM,UAAU,KAAK,OAAO,CAAC,MAAM,EAAE,YAAY,MAAS;AAC1D,UAAM,QAAQ,QAAQ,OAAO,CAAC,MAAM,EAAE,YAAY,IAAI;AACtD,UAAM,OACJ,QAAQ,SAAS,IACb;AAAA,MACE,SAAS,MAAM,OAAO,CAAC,GAAG,MAAM,IAAK,EAAE,SAAoB,CAAC;AAAA,MAC5D,GAAI,QAAQ,SAAS,MAAM,SACvB,EAAE,iBAAiB,QAAQ,SAAS,MAAM,OAAO,IACjD,CAAC;AAAA,IACP,IACA,CAAC;AAEP,WAAO,EAAE,MAAM,MAAM,OAAO,aAAa,YAAY,WAAW,GAAG,KAAK;AAAA,EAC1E,CAAC;AACH;;;AChFO,SAAS,aAAgB,OAA6C;AAC3E,QAAM,WAAW,cAAc,MAAM,OAAO,MAAM,KAAK,MAAM,SAAS;AACtE,QAAM,UAAU,SAAS,OAAO,CAAC,MAAM,EAAE,OAAO,CAAC;AAGjD,QAAM,aACJ,QAAQ,WAAW,IAAI,OAAO,aAAa,QAAQ,IAAI,CAAC,MAAO,EAAE,MAAuB,IAAI,CAAC;AAC/F,QAAM,YAAY,MAAM,IAAI,WAAW,IAAI,OAAO,aAAa,MAAM,IAAI,IAAI,CAAC,MAAM,EAAE,SAAS,CAAC;AAEhG,QAAM,uBAAuB,MAAM,QAAQ,OAAO,CAAC,MAAM,EAAE,UAAU,EAAE,KAAK,EAAE,EAAE;AAEhF,SAAO;AAAA,IACL,QAAQ,MAAM;AAAA,IACd,WAAW,MAAM;AAAA,IACjB;AAAA,IACA,UAAU,CAAC,GAAG,MAAM,QAAQ,EAAE,KAAK,CAAC,GAAG,MAAM,EAAE,WAAW,EAAE,QAAQ;AAAA,IACpE,SAAS,CAAC,GAAG,MAAM,OAAO,EAAE,KAAK,CAAC,GAAG,MAAM,EAAE,WAAW,EAAE,QAAQ;AAAA,IAClE,OAAO;AAAA,MACL,WAAW,MAAM;AAAA,MACjB,YAAY,MAAM,MAAM;AAAA,MACxB,cAAc,QAAQ;AAAA,MACtB;AAAA,MACA,mBAAmB,MAAM;AAAA,MACzB,kBAAkB,MAAM,SAAS;AAAA,MACjC;AAAA,MACA;AAAA,MACA,GAAI,MAAM,OACN,EAAE,SAAS,MAAM,KAAK,SAAS,iBAAiB,MAAM,KAAK,gBAAgB,IAC3E,CAAC;AAAA,MACL,YAAY,MAAM;AAAA,MAClB,GAAI,MAAM,eAAe,EAAE,cAAc,MAAM,aAAa,IAAI,CAAC;AAAA,IACnE;AAAA,EACF;AACF;AAIA,SAAS,IAAI,GAAmB;AAC9B,SAAO,EAAE;AAAA,IACP;AAAA,IACA,CAAC,OAAO,EAAE,KAAK,SAAS,KAAK,QAAQ,KAAK,QAAQ,KAAK,SAAS,GAAG,CAAC;AAAA,EACtE;AACF;AAGA,SAAS,gBAAgB,GAA0B;AACjD,MAAI,KAAK,KAAM,QAAO;AACtB,SAAO,OAAO,KAAK,MAAM,IAAI,GAAG,CAAC;AACnC;AAEA,SAAS,IAAI,GAAmB;AAC9B,SAAO,GAAG,KAAK,MAAM,IAAI,GAAG,CAAC;AAC/B;AAEA,SAAS,WAAW,UAAqE;AACvF,QAAM,QAAkB,CAAC;AACzB,QAAM,OAAO,oBAAI,IAAyB;AAC1C,aAAW,KAAK,UAAU;AACxB,eAAW,CAAC,GAAG,CAAC,KAAK,OAAO,QAAQ,EAAE,KAAK,MAAM,GAAG;AAClD,UAAI,CAAC,KAAK,IAAI,CAAC,GAAG;AAChB,aAAK,IAAI,GAAG,oBAAI,IAAI,CAAC;AACrB,cAAM,KAAK,CAAC;AAAA,MACd;AACA,WAAK,IAAI,CAAC,GAAG,IAAI,CAAC;AAAA,IACpB;AAAA,EACF;AACA,SAAO,MAAM,IAAI,CAAC,UAAU,EAAE,MAAM,QAAQ,CAAC,GAAI,KAAK,IAAI,IAAI,KAAK,CAAC,CAAE,EAAE,EAAE;AAC5E;AAGA,SAAS,WAAW,GAAyB;AAC3C,QAAM,UAAU,OAAO,QAAQ,EAAE,UAAU,EAAE,IAAI,CAAC,CAAC,GAAG,CAAC,MAAM,CAAC,GAAG,EAAE,IAAI,CAAqB;AAC5F,MAAI,QAAQ,WAAW,EAAG,QAAO;AACjC,QAAM,SAAS,QAAQ,KAAK,CAAC,GAAG,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,CAAC;AACjD,QAAM,IAAI,OAAO,CAAC;AAClB,MAAI,CAAC,EAAG,QAAO;AACf,SAAO,qBAAqB,IAAI,EAAE,CAAC,CAAC,CAAC,IAAI,IAAI,EAAE,CAAC,CAAC,CAAC;AACpD;AAEA,SAAS,YAAY,UAAkC;AACrD,QAAM,OAAO,WAAW,QAAQ;AAChC,QAAM,OAAO,IAAI,IAAI,SAAS,IAAI,CAAC,MAAM,CAAC,EAAE,KAAK,IAAI,CAAC,CAAC,CAAC;AACxD,QAAM,OAAO,CAAC,GAA6B,UAA0B;AACnE,UAAM,IAAI,GAAG,OAAO,QAAQ;AAC5B,UAAM,QAAQ,GAAG,QACb,GAAG,IAAI,EAAE,MAAM,IAAI,CAAC,mBAAmB,IAAI,EAAE,MAAM,MAAM,CAAC,SAAS,IAAI,EAAE,MAAM,GAAG,CAAC,UAAO,EAAE,IAAI,cAAW,IAAI,EAAE,WAAW,CAAC,WAAW,EAAE,YAAY,UAAO,EAAE,UAAU,SAAS,KAAM,QAAQ,CAAC,CAAC,aAAa,EAAE,GAAG,EAAE,YAAY,SAAY,UAAO,EAAE,QAAQ,QAAQ,CAAC,CAAC,KAAK,EAAE,KAChR;AACJ,WAAO,uCAAuC,gBAAgB,CAAC,CAAC,YAAY,IAAI,KAAK,CAAC,KAAK,QAAQ,oBAAoB,IAAI,KAAK,CAAC,YAAY,EAAE,oBAAoB,KAAK,KAAK,OAAO,IAAI,CAAC,IAAI,QAAG,UAAU,IAAI,WAAW,CAAC,IAAI,EAAE;AAAA,EAClO;AAEA,QAAM,UAAU,KAAK,CAAC;AACtB,QAAM,UAAU,KAAK,CAAC;AACtB,MAAI,KAAK,WAAW,KAAK,WAAW,SAAS;AAC3C,UAAM,OAAO,gBAAgB,QAAQ,OAAO,IAAI,CAAC,MAAM,OAAO,IAAI,CAAC,CAAC,OAAO,EAAE,KAAK,EAAE,CAAC;AACrF,UAAM,OAAO,QAAQ,OAClB,IAAI,CAAC,OAAO;AACX,YAAM,QAAQ,QAAQ,OACnB,IAAI,CAAC,OAAO;AACX,cAAM,KAAK,GAAG,QAAQ,IAAI,IAAI,EAAE,IAAI,QAAQ,IAAI,IAAI,EAAE;AACtD,eAAO,OAAO,KAAK,KAAK,IAAI,EAAE,GAAG,EAAE,CAAC;AAAA,MACtC,CAAC,EACA,KAAK,EAAE;AACV,aAAO,sBAAsB,IAAI,EAAE,CAAC,QAAQ,KAAK;AAAA,IACnD,CAAC,EACA,KAAK,EAAE;AACV,WAAO,oCAAoC,IAAI,QAAQ,IAAI,CAAC,sBAAmB,IAAI,QAAQ,IAAI,CAAC,iCAAiC,IAAI,GAAG,IAAI;AAAA,EAC9I;AAEA,QAAM,SAAS,CAAC,GAAG,QAAQ,EAAE,KAAK,CAAC,GAAG,OAAO,EAAE,OAAO,QAAQ,MAAM,EAAE,OAAO,QAAQ,EAAE;AACvF,SAAO,qBAAqB,OAAO,IAAI,CAAC,MAAM,KAAK,GAAG,OAAO,OAAO,EAAE,KAAK,MAAM,EAAE,KAAK,QAAK,CAAC,CAAC,EAAE,KAAK,EAAE,CAAC;AAC3G;AAEA,SAAS,aAAgB,UAAwB,OAAuB;AACtE,MAAI,SAAS,WAAW;AACtB,WAAO;AACT,SAAO,SACJ,MAAM,GAAG,KAAK,EACd,IAAI,CAAC,MAAM;AACV,UAAM,SAAS,OAAO,QAAQ,EAAE,KAAK,MAAM,EACxC,IAAI,CAAC,CAAC,GAAG,CAAC,MAAM,GAAG,CAAC,IAAI,CAAC,EAAE,EAC3B,KAAK,QAAK;AACb,UAAM,UAAU,EAAE,WAAW,UAAU,CAAC,GACrC,IAAI,CAAC,MAAM,wBAAwB,IAAI,CAAC,CAAC,SAAS,EAClD,KAAK,EAAE;AACV,UAAM,OAAO,OAAO,QAAQ,EAAE,WAAW,UAAU,CAAC,CAAC,EAClD,KAAK,CAAC,GAAG,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,CAAC,EAC1B,MAAM,GAAG,CAAC,EACV,IAAI,CAAC,CAAC,GAAG,CAAC,MAAM,qBAAqB,IAAI,CAAC,CAAC,IAAI,IAAI,CAAC,CAAC,SAAS,EAC9D,KAAK,EAAE;AACV,WAAO,4DAA4D,IAAI,MAAM,CAAC,UAAU,MAAM,+BAA+B,IAAI,EAAE,QAAQ,CAAC,sEAAsE,IAAI,EAAE,QAAQ,CAAC,oCAAoC,IAAI,EAAE,QAAQ,8BAA8B,CAAC,SAAS,OAAO,sBAAsB,IAAI,WAAW,EAAE;AAAA,EAC3W,CAAC,EACA,KAAK,EAAE;AACZ;AAUO,SAAS,kBACd,SACA,OAA6B,CAAC,GACtB;AACR,QAAM,IAAI,QAAQ;AAClB,QAAM,MAAM,CAAC,OAAe,OAAe,SAAS,cAClD,iDAAiD,MAAM,KAAK,IAAI,KAAK,CAAC,yBAAyB,IAAI,KAAK,CAAC;AAC3G,QAAM,OAAO,QAAQ,OACjB;AAAA,IACE;AAAA,IACA,GAAG,QAAQ,KAAK,OAAO,QAAQ,CAAC,CAAC,WAAM,QAAQ,KAAK,MAAM,QAAQ,CAAC,CAAC;AAAA,IACpE;AAAA,EACF,IACA;AAGJ,QAAM,OACJ,EAAE,YAAY,SACV;AAAA,IACE,EAAE,kBAAkB,aAAU,EAAE,eAAe,mBAAmB;AAAA,IAClE,IAAI,EAAE,QAAQ,QAAQ,CAAC,CAAC;AAAA,IACxB,EAAE,kBAAkB,YAAY;AAAA,EAClC,IACA;AACN,QAAM,QAAQ,KAAK,eAAe,QAAQ,eAAe;AAEzD,SAAO;AAAA,SACA,IAAI,QAAQ,SAAS,CAAC,mBAAc,IAAI,QAAQ,MAAM,CAAC;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,MAoC1D,IAAI,QAAQ,SAAS,CAAC,qBAAkB,IAAI,QAAQ,MAAM,CAAC;AAAA,mBAC9C,EAAE,SAAS,qBAAqB,EAAE,YAAY,IAAI,EAAE,UAAU,iBAAiB,EAAE,uBAAuB,IAAI,SAAM,EAAE,oBAAoB,4BAA4B,EAAE,GAAG,QAAQ,SAAM,IAAI,KAAK,CAAC,KAAK,EAAE;AAAA;AAAA,EAEzN,EAAE,aAAa,IAAI,cAAc,GAAG,IAAI,EAAE,WAAW,IAAI,CAAC,IAAI,EAAE,WAAW,OAAO,MAAM,YAAY,SAAS,IAAI,EAAE;AAAA,EACnH,EAAE,aAAa,IAAI,eAAe,GAAG,IAAI,EAAE,WAAW,GAAG,CAAC,SAAI,IAAI,EAAE,WAAW,GAAG,CAAC,EAAE,IAAI,EAAE;AAAA,EAC3F,EAAE,YAAY,IAAI,kBAAkB,IAAI,EAAE,UAAU,SAAS,KAAM,QAAQ,CAAC,CAAC,KAAK,EAAE,UAAU,MAAM,IAAI,EAAE,UAAU,SAAS,YAAY,SAAS,IAAI,EAAE;AAAA,EACxJ,IAAI,qBAAqB,OAAO,EAAE,gBAAgB,GAAG,EAAE,mBAAmB,IAAI,YAAY,SAAS,CAAC;AAAA,EACpG,IAAI,iBAAiB,GAAG,EAAE,YAAY,IAAI,EAAE,UAAU,EAAE,CAAC;AAAA,EACzD,IAAI,iBAAiB,OAAO,EAAE,SAAS,CAAC,CAAC;AAAA,EACzC,IAAI;AAAA,EACJ,EAAE,aAAa,IAAI,IAAI,eAAe,OAAO,EAAE,UAAU,GAAG,SAAS,IAAI,EAAE;AAAA,EAC3E,IAAI;AAAA;AAAA,EAEJ,EAAE,eAAe,yDAAyD,IAAI,EAAE,aAAa,MAAM,CAAC,qEAAgE,EAAE;AAAA;AAAA,EAEtK,YAAY,QAAQ,QAAQ,CAAC;AAAA,uBACR,EAAE,oBAAoB,EAAE,mBAAmB,SAAM,EAAE,gBAAgB,OAAO,EAAE,iBAAiB,0CAA0C,EAAE;AAAA,EAC9J,aAAa,QAAQ,UAAU,KAAK,eAAe,CAAC,CAAC;AAAA;AAAA;AAGvD;;;AChPA,IAAM,UAAU,CAAC,MAAsB,KAAK,IAAI,GAAG,KAAK,IAAI,GAAG,CAAC,CAAC;AAS1D,SAAS,iBAAoB,MAGpB;AACd,SAAO,OAAO,QAAyC;AACrD,UAAM,YAAY,KAAK,aAAa,IAAI,IAAI;AAC5C,UAAM,UAAU,CAAC,GAAG,IAAI,QAAQ,GAAG,IAAI,KAAK;AAC5C,UAAM,OAAO,IAAI,IAAI,QAAQ,IAAI,KAAK,UAAU,CAAC;AACjD,UAAM,MAAW,CAAC;AAClB,eAAW,UAAU,SAAS;AAC5B,UAAI,IAAI,UAAU,IAAI,MAAO;AAC7B,iBAAW,KAAK,WAAW;AACzB,cAAM,WAAW,MAAM,EAAE,OAAO,QAAQ,IAAI,GAAG;AAC/C,mBAAW,SAAS,UAAU;AAC5B,gBAAM,KAAK,KAAK,WAAW,KAAK;AAChC,cAAI,KAAK,IAAI,EAAE,EAAG;AAClB,eAAK,IAAI,EAAE;AACX,cAAI,KAAK,KAAK;AACd,cAAI,IAAI,UAAU,IAAI,MAAO;AAAA,QAC/B;AACA,YAAI,IAAI,UAAU,IAAI,MAAO;AAAA,MAC/B;AAAA,IACF;AACA,WAAO;AAAA,EACT;AACF;AAKO,SAAS,qBAAqB,YAAY,KAAgB;AAC/D,SAAO,EAAE,MAAM,eAAe,WAAW,UAAU,CAAC,OAAO,QAAQ,IAAI,GAAG,KAAK,EAAE;AACnF;AAEA,SAAS,QACP,GACA,GACQ;AACR,MAAI,CAAC,KAAK,CAAC,EAAG,QAAO;AACrB,QAAM,OAAO,oBAAI,IAAI,CAAC,GAAG,OAAO,KAAK,CAAC,GAAG,GAAG,OAAO,KAAK,CAAC,CAAC,CAAC;AAC3D,MAAI,KAAK,SAAS,EAAG,QAAO;AAC5B,MAAI,OAAO;AACX,aAAW,KAAK,KAAM,KAAI,EAAE,CAAC,MAAM,EAAE,CAAC,EAAG;AACzC,SAAO,OAAO,KAAK;AACrB;AAOO,SAAS,iBAAiB,YAAY,KAAgB;AAC3D,SAAO;AAAA,IACL,MAAM;AAAA,IACN;AAAA,IACA,UAAU,CAAC,IAAgB,QAA0B;AACnD,YAAM,eACJ,IAAI,cAAc,WAAW,IACzB,IACA,KAAK,IAAI,GAAG,IAAI,cAAc,IAAI,CAAC,MAAM,KAAK,IAAI,IAAI,GAAG,KAAK,CAAC,CAAC;AACtE,YAAM,cACJ,IAAI,mBAAmB,WAAW,IAC9B,IACA,KAAK,IAAI,GAAG,IAAI,mBAAmB,IAAI,CAAC,MAAM,QAAQ,GAAG,GAAG,UAAU,CAAC,CAAC;AAC9E,aAAO,QAAQ,MAAM,eAAe,MAAM,WAAW;AAAA,IACvD;AAAA,EACF;AACF;;;AC/DA,eAAe,KACb,OACA,IACA,aACA,QACe;AACf,MAAI,IAAI;AACR,QAAM,UAAU,MAAM;AAAA,IACpB,EAAE,QAAQ,KAAK,IAAI,GAAG,KAAK,IAAI,aAAa,MAAM,MAAM,CAAC,EAAE;AAAA,IAC3D,YAAY;AACV,aAAO,IAAI,MAAM,QAAQ;AACvB,YAAI,QAAQ,QAAS;AACrB,cAAM,OAAO,MAAM,GAAG;AACtB,YAAI,SAAS,OAAW;AACxB,cAAM,GAAG,IAAI;AAAA,MACf;AAAA,IACF;AAAA,EACF;AACA,QAAM,QAAQ,IAAI,OAAO;AAC3B;AAEO,IAAM,mBAAN,MAA0B;AAAA,EAuB/B,YAA6B,MAAyB;AAAzB;AAC3B,SAAK,QAAQ,eAAe,KAAK,KAAK;AACtC,QAAI,KAAK,MAAM,WAAW;AACxB,YAAM,IAAI,MAAM,4EAAkE;AACpF,QAAI,KAAK,kBAAkB,QAAW;AACpC,UACE,OAAO,KAAK,kBAAkB,YAC9B,CAAC,OAAO,SAAS,KAAK,aAAa,KACnC,KAAK,gBAAgB,GACrB;AACA,cAAM,IAAI;AAAA,UACR,4EAA4E,OAAO,KAAK,aAAa,CAAC;AAAA,QACxG;AAAA,MACF;AAAA,IACF;AACA,QAAI,CAAC,KAAK,WAAW,KAAK,kBAAkB,UAAa,KAAK,UAAU,KAAK,SAAS;AACpF,YAAM,IAAI;AAAA,QACR;AAAA,MAEF;AAAA,IACF;AACA,SAAK,WAAW,IAAI,IAAI,KAAK,MAAM,IAAI,CAAC,MAAM,CAAC,EAAE,IAAI,CAAC,CAAC,CAAC;AACxD,SAAK,YAAY,KAAK,aAAa,qBAAqB,GAAG;AAC3D,SAAK,YAAY,KAAK,UAAU,aAAa;AAC7C,SAAK,eAAe,KAAK,gBAAgB;AACzC,SAAK,iBAAiB,KAAK;AAAA,MACzB,KAAK,MAAM,SAAS,KAAK;AAAA,MACzB,KAAK,KAAK,KAAK,SAAS,CAAC;AAAA,IAC3B;AACA,SAAK,YAAY,KAAK,QAAQ,OAAO;AAAA,EACvC;AAAA,EA9B6B;AAAA,EAtBZ;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA;AAAA,EAGA,MAAkD,CAAC;AAAA;AAAA,EAEnD,eAAe,oBAAI,IAA6B;AAAA,EAChD,YAA0B,CAAC;AAAA,EACpC,WAAW;AAAA,EACX,oBAAoB;AAAA,EACpB,aAAa;AAAA,EACb,wBAAwB;AAAA,EACxB;AAAA,EACA;AAAA;AAAA,EAEA,gBAAgB;AAAA,EAChB,kBAAkB;AAAA;AAAA,EAmClB,MAAM,MAAc;AAC1B,SAAK,WAAY,KAAK,WAAW,aAAc;AAC/C,QAAI,IAAI,KAAK;AACb,QAAI,KAAK,KAAK,IAAK,MAAM,IAAK,IAAI,CAAC;AACnC,SAAK,IAAI,KAAK,KAAK,IAAK,MAAM,GAAI,IAAI,EAAE;AACxC,aAAS,IAAK,MAAM,QAAS,KAAK;AAAA,EACpC;AAAA,EAEQ,MAAM,MAAY,YAAwD;AAChF,QAAI,CAAC,cAAc,OAAO,KAAK,UAAU,EAAE,WAAW,EAAG,QAAO,KAAK;AACrE,UAAM,WAAW,OAAO,KAAK,UAAU,EACpC,KAAK,EACL,IAAI,CAAC,MAAM,GAAG,CAAC,IAAI,WAAW,CAAC,CAAC,EAAE,EAClC,KAAK,GAAG;AACX,WAAO,GAAG,KAAK,EAAE,IAAI,QAAQ;AAAA,EAC/B;AAAA,EAEQ,SAAS,QAA0D;AACzE,SAAK,KAAK,KAAK,cAAc,gBAAgB,WAAW;AACtD,YAAM,MAAM,KAAK,IAAI,KAAK,cAAc,KAAK,MAAM,SAAS,KAAK,MAAM,MAAM,CAAC;AAC9E,aAAO,KAAK,MAAM,IAAI,CAAC,OAAO,EAAE,QAAQ,EAAE,IAAI,OAAO,IAAI,EAAE;AAAA,IAC7D;AACA,WAAO;AAAA,MACL,KAAK,IAAI,IAAI,CAAC,OAAO;AAAA,QACnB,WAAW,EAAE,KAAK;AAAA,QAClB,YAAY,EAAE;AAAA,QACd,OAAO,EAAE,GAAG;AAAA,QACZ,MAAM,EAAE,GAAG,SAAS,EAAE,GAAG,SAAS;AAAA,MACpC,EAAE;AAAA,MACF,KAAK,MAAM,IAAI,CAAC,OAAO,EAAE,WAAW,EAAE,IAAI,YAAY,IAAI,EAAE;AAAA,MAC5D,EAAE,QAAQ,cAAc,KAAK,aAAa;AAAA,IAC5C,EAAE,IAAI,CAAC,OAAO,EAAE,QAAQ,EAAE,WAAW,OAAO,EAAE,MAAM,EAAE;AAAA,EACxD;AAAA,EAEQ,mBAAmB;AACzB,UAAM,UAAU,CAAC,GAAG,KAAK,aAAa,OAAO,CAAC;AAC9C,WAAO;AAAA,MACL,eAAe,QAAQ,IAAI,CAAC,MAAM,EAAE,WAAW,KAAK;AAAA,MACpD,oBAAoB,QAAQ,IAAI,CAAC,MAAM,EAAE,WAAW,UAAU;AAAA,IAChE;AAAA,EACF;AAAA;AAAA,EAGQ,gBAAyB;AAC/B,WAAO,KAAK,KAAK,kBAAkB,UAAa,KAAK,iBAAiB,KAAK,KAAK;AAAA,EAClF;AAAA;AAAA;AAAA;AAAA;AAAA,EAMQ,cAAc,UAAa,MAAY,IAA2C;AACxF,QAAI,CAAC,KAAK,KAAK,OAAQ,QAAO;AAC9B,UAAM,OAAO,KAAK,KAAK,OAAO,UAAU,MAAM,EAAE;AAChD,QAAI,SAAS,MAAM;AACjB,WAAK;AACL,aAAO;AAAA,IACT;AACA,QAAI,OAAO,KAAK,QAAQ,YAAY,CAAC,OAAO,SAAS,KAAK,GAAG,KAAK,KAAK,MAAM,GAAG;AAC9E,YAAM,IAAI;AAAA,QACR,qDAAqD,OAAO,MAAM,GAAG,CAAC;AAAA,MAExE;AAAA,IACF;AACA,SAAK,iBAAiB,KAAK;AAC3B,SAAK,KAAK,QAAQ,OAAO;AAAA,MACvB,OAAO,KAAK,SAAS;AAAA,MACrB,SAAS;AAAA,MACT,OAAO,EAAE,aAAa,GAAG,cAAc,EAAE;AAAA,MACzC,eAAe,KAAK;AAAA,MACpB,MAAM,EAAE,QAAQ,KAAK,KAAK,QAAQ,MAAM,KAAK,GAAG;AAAA,IAClD,CAAC;AACD,SAAK,KAAK,SAAS,EAAE,KAAK,KAAK,KAAK,SAAS,QAAQ,CAAC;AACtD,WAAO,KAAK;AAAA,EACd;AAAA;AAAA,EAGQ,UAAUA,SAAqB;AACrC,UAAM,MAAW,CAAC;AAClB,eAAW,KAAK,KAAK,aAAa,OAAO,EAAG,KAAI,EAAE,KAAK,OAAOA,QAAQ,KAAI,KAAK,EAAE,QAAQ;AACzF,WAAO;AAAA,EACT;AAAA;AAAA,EAGA,MAAM,OAA0D;AAC9D,UAAM,YAAY,KAAK,KAAK,SAAS,KAAK;AAC1C,QAAI,aAAa,KAAK,KAAK,cAAc,KAAK,KAAK,KAAK,QAAQ;AAC9D,aAAO,EAAE,MAAM,GAAG,UAAU,CAAC,EAAE;AAEjC,UAAM,cAAc,KAAK,SAAS,KAAK,IAAI,KAAK,gBAAgB,SAAS,CAAC;AAC1E,UAAM,cAA4B,CAAC;AACnC,QAAI,eAAe;AAEnB,eAAW,SAAS,aAAa;AAC/B,UACE,KAAK,YAAY,KAAK,KAAK,UAC3B,KAAK,cAAc,KACnB,KAAK,iBAAiB,UACtB,KAAK,KAAK,QAAQ;AAElB;AACF,YAAM,OAAO,KAAK,SAAS,IAAI,MAAM,MAAM;AAC3C,UAAI,CAAC,KAAM;AACX,YAAM,MAAM,KAAK,IAAI,MAAM,OAAO,KAAK,KAAK,SAAS,KAAK,QAAQ;AAClE,UAAI,OAAO,EAAG;AACd,WAAK,KAAK,aAAa,EAAE,MAAM,kBAAkB,MAAM,OAAO,IAAI,CAAC;AAEnE,YAAM,QAAQ,MAAM,KAAK,KAAK,SAAS,IAAI;AAC3C,YAAM,SAAS,KAAK,UAAU,KAAK,EAAE;AACrC,YAAM,WAAW,MAAM,KAAK,KAAK,SAAS;AAAA,QACxC;AAAA,QACA;AAAA,QACA;AAAA,QACA,UAAU,KAAK;AAAA,QACf,OAAO;AAAA,QACP,KAAK,KAAK;AAAA,MACZ,CAAC;AAGD,YAAM,OAAO,KAAK,IAAI,MAAM,CAAC,MAAM,EAAE,KAAK,OAAO,KAAK,EAAE;AACxD,YAAM,SAAS,CAAC,GAAI,OAAO,QAAQ,CAAC,GAAI,GAAG,QAAQ,EAAE,MAAM,GAAG,GAAG;AAEjE,YAAM;AAAA,QACJ;AAAA,QACA,OAAO,aAAa;AAClB,cACE,KAAK,YAAY,KAAK,KAAK,UAC3B,KAAK,cAAc,KACnB,KAAK,iBAAiB,UACtB,KAAK,KAAK,QAAQ;AAElB;AAMF,cAAI;AACF,kBAAM,YAAY,YAAY,IAAI;AAClC,kBAAM,KAAK,MAAM,KAAK,KAAK,SAAS,UAAU,IAAI;AAGlD,kBAAM,YAAY,GAAG,aAAa,YAAY,IAAI,IAAI;AACtD,iBAAK;AACL;AACA,iBAAK,wBAAwB;AAC7B,kBAAM,UAAU,KAAK,cAAc,UAAU,MAAM,EAAE;AACrD,kBAAM,WAAW,KAAK,UAAU,SAAS,IAAI,KAAK,iBAAiB,CAAC;AACpE,iBAAK,IAAI,KAAK;AAAA,cACZ;AAAA,cACA;AAAA,cACA;AAAA,cACA;AAAA,cACA,GAAI,YAAY,SAAY,EAAE,QAAQ,IAAI,CAAC;AAAA,cAC3C,YAAY,KAAK,KAAK,WAAW,QAAQ;AAAA,YAC3C,CAAC;AACD,iBAAK,KAAK,aAAa,EAAE,MAAM,aAAa,MAAM,UAAU,YAAY,GAAG,CAAC;AAE5E,kBAAM,MAAM,KAAK,MAAM,MAAM,GAAG,UAAU;AAC1C,kBAAM,MAAM,KAAK,aAAa,IAAI,GAAG;AACrC,gBAAI,CAAC,OAAO,WAAW,IAAI;AACzB,mBAAK,aAAa,IAAI,KAAK,EAAE,OAAO,KAAK,MAAM,UAAU,YAAY,IAAI,SAAS,CAAC;AAErF,gBAAI,WAAW,KAAK,UAAW;AAC/B,iBAAK;AACL,gBAAI,KAAK,KAAK,OAAO,WAAW,CAAE,MAAM,KAAK,KAAK,MAAM,QAAQ,UAAU,IAAI,IAAI;AAChF;AACF,gBACE,KAAK,KAAK,OAAO,oBACjB,CAAE,MAAM,KAAK,KAAK,MAAM,iBAAiB,UAAU,IAAI,IAAI;AAE3D;AAEF,kBAAM,YAAY,KAAK,KAAK,WACxB,MAAM,KAAK,KAAK,SAAS,UAAU,KAAK,KAAK,UAAU,IAAI,IAC3D;AACJ,kBAAM,UAAsB;AAAA,cAC1B,IAAI,KAAK,KAAK,WAAW,QAAQ;AAAA,cACjC;AAAA,cACA;AAAA,cACA;AAAA,cACA,MAAM,KAAK,KAAK,eAAe,SAAS;AAAA,cACxC,YAAY;AAAA,cACZ;AAAA,cACA,WAAW,KAAK,UAAU;AAAA,YAC5B;AACA,iBAAK,UAAU,KAAK,OAAO;AAC3B,wBAAY,KAAK,OAAO;AACxB,iBAAK,KAAK,aAAa,EAAE,MAAM,WAAW,QAAQ,CAAC;AAAA,UACrD,SAAS,KAAK;AAGZ,gBAAI,eAAe,WAAY,OAAM;AACrC,iBAAK;AACL,iBAAK;AACL,kBAAM,UAAU,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;AAC/D,iBAAK,KAAK,aAAa;AAAA,cACrB,MAAM;AAAA,cACN;AAAA,cACA,YAAY,KAAK,KAAK,WAAW,QAAQ;AAAA,cACzC;AAAA,YACF,CAAC;AACD,kBAAM,QAAQ,KAAK,KAAK,4BAA4B;AACpD,gBAAI,KAAK,yBAAyB,OAAO;AACvC,mBAAK,eAAe;AAAA,gBAClB,QAAQ;AAAA,gBACR,QAAQ,GAAG,KAAK,qBAAqB,mCAAmC,OAAO;AAAA,cACjF;AAAA,YACF;AAAA,UACF;AAAA,QACF;AAAA,QACA,KAAK,KAAK,eAAe;AAAA,QACzB,KAAK,KAAK;AAAA,MACZ;AAAA,IACF;AAEA,SAAK,KAAK,aAAa,EAAE,MAAM,SAAS,UAAU,KAAK,UAAU,QAAQ,KAAK,KAAK,OAAO,CAAC;AAC3F,WAAO,EAAE,MAAM,cAAc,UAAU,YAAY;AAAA,EACrD;AAAA;AAAA;AAAA,EAIA,MAAM,MAA+B;AACnC,WACE,KAAK,WAAW,KAAK,KAAK,UAC1B,CAAC,KAAK,cAAc,KACpB,KAAK,iBAAiB,UACtB,CAAC,KAAK,KAAK,QAAQ,SACnB;AACA,YAAM,EAAE,KAAK,IAAI,MAAM,KAAK,KAAK;AACjC,UAAI,SAAS,KAAK,KAAK,iBAAiB,OAAW;AAAA,IACrD;AACA,WAAO,KAAK,QAAQ;AAAA,EACtB;AAAA,EAEA,WAA2B;AACzB,WAAO,KAAK,QAAQ,EAAE;AAAA,EACxB;AAAA,EAEA,WAAyB;AACvB,WAAO,CAAC,GAAG,KAAK,SAAS,EAAE,KAAK,CAAC,GAAG,MAAM,EAAE,WAAW,EAAE,QAAQ;AAAA,EACnE;AAAA,EAEA,UAA0B;AACxB,WAAO,aAAa;AAAA,MAClB,QAAQ,KAAK,KAAK;AAAA,MAClB,WAAW,KAAK,UAAU;AAAA,MAC1B,OAAO,KAAK;AAAA,MACZ,KAAK,KAAK;AAAA,MACV,WAAW,KAAK;AAAA,MAChB,SAAS,CAAC,GAAG,KAAK,aAAa,OAAO,CAAC;AAAA,MACvC,UAAU,KAAK;AAAA,MACf,mBAAmB,KAAK;AAAA,MACxB,UAAU,KAAK;AAAA,MACf,MAAM,KAAK,KAAK,SACZ,EAAE,SAAS,KAAK,eAAe,iBAAiB,KAAK,gBAAgB,IACrE;AAAA,MACJ,YAAY,KAAK;AAAA,MACjB,cAAc,KAAK;AAAA,IACrB,CAAC;AAAA,EACH;AACF;;;ACjWA,eAAsB,UACpB,MACsC;AACtC,QAAM,EAAE,kBAAkB,GAAG,KAAK,IAAI;AACtC,QAAM,WAAW,IAAI,iBAAoB;AAAA,IACvC,GAAG;AAAA,IACH,WAAW,qBAAqB,oBAAoB,GAAG;AAAA,EACzD,CAAC;AACD,SAAO,EAAE,SAAS,MAAM,SAAS,IAAI,EAAE;AACzC;;;ACfO,SAAS,gBAAmB,MAA6D;AAC9F,QAAM,UAAU,KAAK,OAAO,CAAC,MAA6B,KAAK,IAAI;AACnE,SAAO;AAAA,IACL,SAAS,OAAO,UAAU,IAAI,SAAS;AACrC,iBAAW,KAAK,SAAS;AACvB,YAAI,EAAE,WAAW,CAAE,MAAM,EAAE,QAAQ,UAAU,IAAI,IAAI,EAAI,QAAO;AAAA,MAClE;AACA,aAAO;AAAA,IACT;AAAA,IACA,kBAAkB,OAAO,UAAU,IAAI,SAAS;AAC9C,iBAAW,KAAK,SAAS;AACvB,YAAI,EAAE,oBAAoB,CAAE,MAAM,EAAE,iBAAiB,UAAU,IAAI,IAAI,EAAI,QAAO;AAAA,MACpF;AACA,aAAO;AAAA,IACT;AAAA,EACF;AACF;AASO,SAAS,0BAA6B,MAMxB;AACnB,QAAM,YAAY,KAAK,oBAAoB;AAC3C,SAAO;AAAA,IACL,kBAAkB,OAAO,UAAa,KAAiB,SAAe;AACpE,YAAM,YAAY,KAAK,QAAQ,QAAQ;AACvC,UAAI,aAAa,KAAM,QAAO;AAC9B,YAAM,KAAK,MAAM,KAAK,SAAS,WAAW,IAAI;AAC9C,aAAO,GAAG,QAAQ;AAAA,IACpB;AAAA,EACF;AACF;AAMO,SAAS,kBAAqB,MAGhB;AACnB,QAAM,YAAY,KAAK,oBAAoB;AAC3C,QAAM,SAAS,KAAK,UAAU;AAC9B,SAAO;AAAA,IACL,SAAS,CAAC,WAAW,OAAmB,GAAG,SAAS,YAAY;AAAA,EAClE;AACF;;;AC/CA,IAAM,UAAU,EAAE,MAAM,UAAU,YAAY,CAAC,GAAG,sBAAsB,MAAM;AAEvE,SAAS,iBAAoB,UAAiD;AACnF,SAAO;AAAA,IACL;AAAA,MACE,MAAM;AAAA,MACN,aACE;AAAA,MACF,YAAY;AAAA,MACZ,SAAS,YAAY;AACnB,cAAM,EAAE,MAAM,SAAS,IAAI,MAAM,SAAS,KAAK;AAC/C,eAAO,EAAE,MAAM,aAAa,SAAS,QAAQ,SAAS;AAAA,MACxD;AAAA,IACF;AAAA,IACA;AAAA,MACE,MAAM;AAAA,MACN,aACE;AAAA,MACF,YAAY;AAAA,MACZ,SAAS,YAAY,SAAS,SAAS;AAAA,IACzC;AAAA,IACA;AAAA,MACE,MAAM;AAAA,MACN,aAAa;AAAA,MACb,YAAY;AAAA,QACV,MAAM;AAAA,QACN,YAAY,EAAE,OAAO,EAAE,MAAM,UAAU,aAAa,yBAAyB,EAAE;AAAA,QAC/E,sBAAsB;AAAA,MACxB;AAAA,MACA,SAAS,OAAO,SAAS;AACvB,cAAM,QAAS,MAAyC;AACxD,cAAM,MAAM,SAAS,SAAS;AAC9B,eAAO,OAAO,UAAU,WAAW,IAAI,MAAM,GAAG,KAAK,IAAI;AAAA,MAC3D;AAAA,IACF;AAAA,IACA;AAAA,MACE,MAAM;AAAA,MACN,aACE;AAAA,MACF,YAAY;AAAA,QACV,MAAM;AAAA,QACN,YAAY;AAAA,UACV,QAAQ,EAAE,MAAM,UAAU,MAAM,CAAC,QAAQ,MAAM,EAAE;AAAA,UACjD,aAAa,EAAE,MAAM,UAAU,aAAa,uCAAuC;AAAA,QACrF;AAAA,QACA,sBAAsB;AAAA,MACxB;AAAA,MACA,SAAS,OAAO,SAAS;AACvB,cAAM,IAAK,QAAQ,CAAC;AACpB,cAAM,UAAU,SAAS,QAAQ;AACjC,YAAI,EAAE,WAAW,OAAQ,QAAO,kBAAkB,SAAS,EAAE,aAAa,EAAE,YAAY,CAAC;AACzF,eAAO;AAAA,MACT;AAAA,IACF;AAAA,EACF;AACF;","names":["cellId"]}

package/dist/index.d.ts

@@ -14,8 +14,8 @@ import { A as AgentEvalError, J as JudgeError, a as ConfigError } from './errors
 export { b as AgentEvalErrorCode, C as CaptureIntegrityError, N as NotFoundError, R as ReplayError, V as ValidationError, c as VerificationError } from './errors-CzMUYo7b.js';
 import { b as FeedbackLabel, F as FeedbackTrajectoryStore, a as FeedbackTrajectory } from './feedback-trajectory-D9OVLrg9.js';
 export { c as FeedbackArtifactType, d as FeedbackAttempt, e as FeedbackLabelKind, f as FeedbackLabelSource, g as FeedbackOptimizerRow, h as FeedbackOutcome, i as FeedbackReplayAdapter, j as FeedbackReplayResult, k as FeedbackSeverity, l as FeedbackSplitPolicy, m as FeedbackTask, n as FeedbackTrajectoryFilter, o as FileSystemFeedbackTrajectoryStore, I as InMemoryFeedbackTrajectoryStore, P as PreferenceMemoryEntry, p as ProposedSideEffect, q as assignFeedbackSplit, r as controlRunToFeedbackTrajectory, s as createFeedbackTrajectory, t as feedbackTrajectoriesToDatasetScenarios, u as feedbackTrajectoriesToOptimizerRows, v as feedbackTrajectoryToDatasetScenario, w as feedbackTrajectoryToOptimizerRow, x as parseFeedbackTrajectoriesJsonl, y as renderPreferenceMemoryMarkdown, z as replayFeedbackTrajectories, A as replayFeedbackTrajectory, B as serializeFeedbackTrajectoriesJsonl, C as summarizePreferenceMemory, D as withAssignedFeedbackSplit } from './feedback-trajectory-D9OVLrg9.js';
-import { b as CorrectnessChecker, A as AgentProfile$1 } from './pre-registration-DELOEJ8v.js';
-export { c as ArtifactCheckArtifact, d as ArtifactEventLike, e as ArtifactValidator, f as BackendIntegrityError, B as BackendIntegrityReport, C as CompletionRequirement, a as CompletionVerdict, H as HypothesisManifest, g as HypothesisResult, L as LlmCorrectnessCheckerOpts, h as ProducedProposal, P as ProducedState, i as ProposalEventLike, j as RequirementCheck, R as RuntimeEventLike, k as SatisfiedBy, S as SignedManifest, l as SignedManifestAlgo, T as TaskGold, m as ToolCallEventLike, V as ValidationContext, n as ValidationIssue, o as ValidationResult, p as agentProfileHash, q as assertRealBackend, r as byteLengthRange, s as canonicalize, t as completionVerdict, u as composeValidators, v as containsAll, w as createLlmCorrectnessChecker, x as createTokenRecallChecker, y as evaluateHypothesis, z as extractProducedState, D as hashJson, E as jsonHasKeys, F as parseCorrectnessResponse, G as regexMatch, I as signManifest, J as summarizeBackendIntegrity, K as verifyCompletion, M as verifyManifest } from './pre-registration-DELOEJ8v.js';
+import { b as CorrectnessChecker, A as AgentProfile$1 } from './pre-registration-mAnCugl9.js';
+export { c as ArtifactCheckArtifact, d as ArtifactEventLike, e as ArtifactValidator, f as BackendIntegrityError, B as BackendIntegrityReport, C as CompletionRequirement, a as CompletionVerdict, H as HypothesisManifest, g as HypothesisResult, L as LlmCorrectnessCheckerOpts, h as ProducedProposal, P as ProducedState, i as ProposalEventLike, j as RequirementCheck, R as RuntimeEventLike, k as SatisfiedBy, S as SignedManifest, l as SignedManifestAlgo, T as TaskGold, m as ToolCallEventLike, V as ValidationContext, n as ValidationIssue, o as ValidationResult, p as agentProfileHash, q as assertRealBackend, r as byteLengthRange, s as canonicalize, t as completionVerdict, u as composeValidators, v as containsAll, w as createLlmCorrectnessChecker, x as createTokenRecallChecker, y as evaluateHypothesis, z as extractProducedState, D as hashJson, E as jsonHasKeys, F as parseCorrectnessResponse, G as regexMatch, I as signManifest, J as summarizeBackendIntegrity, K as verifyCompletion, M as verifyManifest } from './pre-registration-mAnCugl9.js';
 export { DataAcquisitionPlan, KnowledgeAcquisitionMode, KnowledgeBundle, KnowledgeFallbackPolicy, KnowledgeFreshness, KnowledgeImportance, KnowledgeReadinessReport, KnowledgeRecommendedAction, KnowledgeRequirement, KnowledgeRequirementCategory, KnowledgeResponsibleSurface, KnowledgeSensitivity, ScoreKnowledgeReadinessOptions, UserQuestion, acquisitionPlansForKnowledgeGaps, blockingKnowledgeEval, knowledgeReadinessTracePayload, scoreKnowledgeReadiness, userQuestionsForKnowledgeGaps } from './knowledge/index.js';
 import { h as ReleaseConfidenceThresholds, f as ReleaseConfidenceScorecard } from './release-report-euXIV_Sk.js';
 export { A as ActionableSideInfo, o as AsiSeverity, B as BootstrapOptions, a as BootstrapResult, J as JudgeReplayGateArgs, R as ReleaseConfidenceAxis, b as ReleaseConfidenceAxisName, c as ReleaseConfidenceInput, d as ReleaseConfidenceIssue, e as ReleaseConfidenceMetrics, g as ReleaseConfidenceStatus, i as ReleaseTraceEvidence, j as RenderReleaseReportOptions, V as Verdict, k as assertReleaseConfidence, l as bootstrapCi, m as evaluateReleaseConfidence, n as judgeReplayGate, r as renderReleaseReport } from './release-report-euXIV_Sk.js';

package/dist/index.js

@@ -41,7 +41,7 @@ import {
   extractProducedState,
   parseCorrectnessResponse,
   verifyCompletion
-} from "./chunk-FIUKOSWI.js";
+} from "./chunk-S6OZEZQK.js";
 import {
   DEFAULT_MUTATION_PRIMITIVES,
   DEFAULT_RED_TEAM_CORPUS,

package/dist/openapi.json

@@ -2,7 +2,7 @@
   "openapi": "3.1.0",
   "info": {
     "title": "@tangle-network/agent-eval — wire protocol",
-    "version": "0.90.1",
+    "version": "0.92.0",
     "description": "HTTP and stdio RPC interface to agent-eval. The TypeScript runtime is the source of truth; this spec is the contract that cross-language clients (Python, Rust, Go) generate from.\n\nWire-protocol version: 1.0.0. Bumps on breaking changes to request/response schemas.",
     "contact": {
       "name": "Tangle Network",

package/dist/{pre-registration-DELOEJ8v.d.ts → pre-registration-mAnCugl9.d.ts}

@@ -277,9 +277,15 @@ declare function createLlmCorrectnessChecker(tc: TCloud, opts?: LlmCorrectnessCh
  * Deterministic `CorrectnessChecker` — the no-LLM counterpart to
  * `createLlmCorrectnessChecker`. A produced item fulfils a requirement when its
  * content is substantive (≥ `minContentLength` chars) AND recalls ≥ `minRecall`
- * of the requirement title's significant tokens. No network — the default gate
- * for apps and tests without an LLM judge. Pass to `verifyCompletion` as the
- * checker.
+ * of the requirement title's significant tokens. No network.
+ *
+ * Polarity-blind: token recall credits a negation that contains the
+ * requirement's tokens ("I will NOT produce the comparison" recalls every token
+ * of "produce the comparison"). The structural match stage is ALSO lexical, so
+ * pairing the two collapses to a single gameable gate. Use this only as an
+ * opt-in structural pre-filter or for tasks whose requirements have no polarity
+ * to invert; for produced-state grading the correctness checker MUST be semantic
+ * (`createLlmCorrectnessChecker`). See the anti-game fixtures in the test suite.
  */
 declare function createTokenRecallChecker(opts?: {
     minRecall?: number;
@@ -326,6 +332,7 @@ interface ProposalEventLike {
     proposalId: string;
     title: string;
     status?: 'pending' | 'approved' | 'rejected';
+    content?: string;
 }
 /**
  * The subset of runtime stream events `extractProducedState` consumes.

package/docs/concepts.md

@@ -183,6 +183,7 @@ release decision.
 ## Where to go next
 
 - **Confused by "GEPA / HALO / trace analysis / drivers everywhere"?** → [self-improvement-map.md](./self-improvement-map.md) — one loop, four roles, the seven-driver catalog (production vs bench-only), and why `gepa-refine` is the same loop on a test bench.
+- **Which `run*` primitive do I use, and how do I grade produced state?** → [eval-surface-map.md](./eval-surface-map.md) — the campaign/matrix/optimization/gate primitives as a pick-by-"use-when" table, plus the produced-state grading composition (verifyCompletion-as-judge — there is no persona-dispatch wrapper) and the in-band body contract.
 - **Need the layman feature map?** → [feature-guide.md](./feature-guide.md) — what each primitive does, when to use it, integration patterns, and guardrails.
 - **Just want to score a string against a rubric?** → [wire-protocol.md](./wire-protocol.md) — HTTP/RPC interface, pluggable from any language.
 - **Need a reusable driver/worker/evaluator loop?** → [control-runtime.md](./control-runtime.md) — generic runtime plus coding, browser, computer-use, and research integration patterns.

package/docs/eval-surface-map.md

@@ -0,0 +1,59 @@
+# Eval surface map — which primitive, when
+
+The eval surface is a small set of orthogonal primitives. They compose; they do
+not overlap. If two seem interchangeable, read the "use when" — the distinction
+is real and load-bearing. **Do not add a new wrapper to bridge two of these; the
+composition point already exists (see Produced-state grading below).**
+
+## The run\* primitives
+
+| Primitive | Use when | Returns |
+|---|---|---|
+| `runCampaign` | The measurement primitive. Run a dispatch over scenarios × seeds × reps, score each with judges, aggregate. Caller owns the dispatch. | `CampaignResult` |
+| `runEval` | The simplest preset over `runCampaign` — just score, no loop, no gate. The 80% "I want a scorecard" case. | `CampaignResult` |
+| `runProfileMatrix` | Factor the SAME scenarios across N agent **profiles** (models / prompt variants), with RunRecord stamping + a real-backend integrity guard. | `RunRecord[]` |
+| `runOptimization` | GENERATE: baseline → N generations of propose → measure → rank → promote. No release gate. | generations + winner |
+| `runImprovementLoop` | The release-gate shell around `runOptimization`: adds a held-out re-score + a promotion gate (+ optional auto-PR). | gate decision + winner |
+| `runEvalCampaign` | Inversion-of-control variant of `runCampaign` — the runner is handed a pre-wired trace/sink/emitter and integrity gating as a precondition. Use when you need full capture by construction. | `CampaignResult` + records |
+| `runAgentMatrix` | The bare N-axis cartesian scheduler with concurrency control. The layer beneath the eval surface — reach for it only when you need raw scheduling, not eval semantics. | cell results |
+
+Mental model: **measure** (`runCampaign`/`runEval`) → **factor** (`runProfileMatrix`) →
+**generate** (`runOptimization`) → **gate** (`runImprovementLoop`). `runEvalCampaign`
+is `runCampaign` with capture inverted; `runAgentMatrix` is the scheduler underneath.
+
+Merging any two of these conflates distinct mental models (measure ≠ search ≠
+release-gate). Keep them separate; pick by the table.
+
+## Produced-state grading — there is NO persona-dispatch wrapper
+
+To grade what an agent actually **produced** (filed the proposal, wrote the
+artifact) rather than what it said, the composition point is a **judge that wraps
+`verifyCompletion`** — not a dedicated runner. The pipeline:
+
+```
+runtime/app-tool events ──► extractProducedState(events) ──► ProducedState
+                                                                  │
+                            verifyCompletion(taskGold, state, correctnessChecker)
+                                                                  │
+                            inject as a JudgeConfig into runProfileMatrix / runCampaign
+```
+
+`extractProducedState` is a pure function over the produced-event stream; the
+judge calls it inline. This is why **`runProducedStatePersonaDispatch` does not
+exist and should not be built** — it would be a fourth layer over a composition
+that is already one judge. (Archetype: `playback.ts` `scoreUserStory`.)
+
+### The in-band body contract
+
+Produced events carry their **body in-band** — the grader never reaches into a
+product database to recover it:
+
+- `artifact` events carry `content` (the persisted file body).
+- `proposal_created` events carry `content` (the `submit_proposal` description) —
+  same role, same field name. A title-only filing omits it; a content-less
+  proposal is graded presence-only (and, by the completion oracle's rule, does
+  not count as a completed deliverable).
+
+A consumer that finds itself re-fetching a deliverable's body from its own DB to
+grade it is working around a thin event — fix the event (carry `content`), don't
+add an enrichment band-aid.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tangle-network/agent-eval",
-  "version": "0.90.1",
+  "version": "0.92.0",
   "description": "Evaluate and improve AI agents from runs, traces, judges, and feedback. Compare candidates, cluster failures, measure lift, and gate releases.",
   "homepage": "https://github.com/tangle-network/agent-eval#readme",
   "repository": {

package/dist/chunk-FIUKOSWI.js.map

@@ -1 +0,0 @@
-{"version":3,"sources":["../src/completion-verifier.ts","../src/produced-state.ts","../src/agent-profile.ts"],"sourcesContent":["/**\n * Completion verifier — the task-completion oracle.\n *\n * Answers the only eval question that is not a proxy: did the agent actually\n * COMPLETE the task — produce every required deliverable, persisted and\n * correct — rather than describe what should be done. A fluent transcript\n * that never produces the artifact scores zero here.\n *\n * Per requirement, a two-stage check:\n *   1. Structural — a produced item (vault artifact / approved proposal /\n *      tool call) of the right kind is matched against the requirement and\n *      carries non-empty content. Deterministic; no LLM.\n *   2. Correctness — only if structurally present AND the matched item\n *      carries content, one targeted check decides whether that item\n *      actually fulfils the requirement. A hallucinated artifact fails here;\n *      an absent one already failed stage 1.\n *\n * `completionRate` is satisfied / total. Quality dimensions are meaningless\n * on an incomplete task — callers gate on `fullyComplete` / `completionRate`\n * before scoring quality.\n */\n\nimport type { TCloud } from '@tangle-network/tcloud'\nimport type { Artifact } from './artifact-validator'\nimport type { DefaultVerdict } from './verdict'\n\n/** What kind of produced state can satisfy a requirement structurally. */\nexport type SatisfiedBy = 'artifact' | 'proposal' | 'tool-call' | 'any'\n\nexport interface CompletionRequirement {\n  /** Stable id from the task gold (e.g. a persona's `expected_requirements[].req_id`). */\n  reqId: string\n  /** Human-readable description of the required deliverable. */\n  title: string\n  /** Optional kind/category hint, matched against a produced item's kind. */\n  category?: string\n  /** What produced state satisfies this requirement. Defaults to 'any'. */\n  satisfiedBy?: SatisfiedBy\n}\n\nexport interface TaskGold {\n  taskId: string\n  requirements: CompletionRequirement[]\n}\n\nexport interface ProducedProposal {\n  id: string\n  title: string\n  status: 'pending' | 'approved' | 'rejected'\n  /** Optional persisted body — when present, enables a correctness check. */\n  content?: string\n}\n\n/** Everything observable about what a run actually produced. */\nexport interface ProducedState {\n  /** Persisted vault artifacts. Reuses the shared `Artifact` shape. */\n  artifacts: Artifact[]\n  /** Proposals / filings the agent created. */\n  proposals: ProducedProposal[]\n  /** Names of tools the agent invoked. */\n  toolCalls: string[]\n}\n\nexport interface RequirementCheck {\n  reqId: string\n  title: string\n  /** A produced item of the right kind matched the requirement, non-empty. */\n  structurallyPresent: boolean\n  /**\n   * Whether the matched item actually fulfils the requirement. `null` when\n   * not structurally present, or when the matched item carries no content\n   * to assess.\n   */\n  correct: boolean | null\n  /** structurallyPresent && correct !== false. */\n  satisfied: boolean\n  /** Human-readable evidence for the verdict. */\n  evidence: string[]\n}\n\n/** Extends the substrate verdict spine: `valid` = `fullyComplete` and\n *  `score` = `completionRate` — derived in `completionVerdict()`, the one\n *  place those equalities hold by construction. */\nexport interface CompletionVerdict extends DefaultVerdict {\n  taskId: string\n  requirements: RequirementCheck[]\n  /** satisfied / total requirements. */\n  completionRate: number\n  /** Every requirement satisfied. */\n  fullyComplete: boolean\n}\n\n/**\n * Construct a `CompletionVerdict` from the per-requirement checks, deriving\n * `completionRate` / `fullyComplete` and the spine fields (`valid` =\n * `fullyComplete`, `score` = `completionRate`) in one place. Throws on zero\n * requirements — a verdict over nothing is a misconfiguration, mirroring\n * `verifyCompletion`'s gold-spec guard.\n */\nexport function completionVerdict(input: {\n  taskId: string\n  requirements: RequirementCheck[]\n}): CompletionVerdict {\n  if (input.requirements.length === 0) {\n    throw new Error(\n      `completionVerdict: task '${input.taskId}' has no requirement checks — nothing to derive a verdict from`,\n    )\n  }\n  const satisfiedCount = input.requirements.filter((r) => r.satisfied).length\n  const completionRate = satisfiedCount / input.requirements.length\n  const fullyComplete = satisfiedCount === input.requirements.length\n  return {\n    taskId: input.taskId,\n    requirements: input.requirements,\n    completionRate,\n    fullyComplete,\n    valid: fullyComplete,\n    score: completionRate,\n  }\n}\n\n/**\n * Decides whether a produced item's content actually fulfils a requirement.\n * Injected so the structural verifier stays pure and unit-testable; the\n * production implementation is `createLlmCorrectnessChecker`.\n */\nexport type CorrectnessChecker = (\n  requirement: CompletionRequirement,\n  content: string,\n) => Promise<{ correct: boolean; reason: string }>\n\nconst STOPWORDS = new Set([\n  'the',\n  'a',\n  'an',\n  'of',\n  'for',\n  'and',\n  'or',\n  'to',\n  'in',\n  'on',\n  'with',\n  'by',\n])\n\nconst MATCH_THRESHOLD = 0.5\nconst MIN_CONTENT_CHARS = 50\n\nfunction tokens(s: string): Set<string> {\n  return new Set(\n    s\n      .toLowerCase()\n      .split(/[^a-z0-9]+/)\n      .filter((t) => t.length > 1 && !STOPWORDS.has(t)),\n  )\n}\n\n/**\n * Recall of the requirement's tokens within a candidate's identifying text.\n * Recall, not Jaccard — a candidate's path/id legitimately carries extra\n * tokens the requirement does not name.\n */\nfunction tokenRecall(requirementText: string, candidateText: string): number {\n  const req = tokens(requirementText)\n  if (req.size === 0) return 0\n  const cand = tokens(candidateText)\n  let hit = 0\n  for (const t of req) if (cand.has(t)) hit++\n  return hit / req.size\n}\n\ninterface Candidate {\n  reqIndex: number\n  /** Unique key for a produced item — each item satisfies at most one requirement. */\n  itemKey: string\n  score: number\n  evidence: string\n  /** Content to correctness-check, or null when the matched item has none. */\n  content: string | null\n}\n\nfunction artifactCandidates(\n  req: CompletionRequirement,\n  reqIndex: number,\n  artifacts: Artifact[],\n): Candidate[] {\n  const reqText = `${req.title} ${req.category ?? ''}`\n  const out: Candidate[] = []\n  artifacts.forEach((a, i) => {\n    if ((a.content ?? '').trim().length < MIN_CONTENT_CHARS) return\n    let score = tokenRecall(reqText, `${a.path ?? ''} ${a.kind}`)\n    if (req.category && a.kind && req.category.toLowerCase() === a.kind.toLowerCase()) {\n      score = Math.max(score, 1)\n    }\n    if (score < MATCH_THRESHOLD) return\n    out.push({\n      reqIndex,\n      itemKey: `artifact:${i}`,\n      score,\n      evidence: `artifact '${a.path ?? a.kind}' matched (token recall ${score.toFixed(2)})`,\n      content: a.content ?? null,\n    })\n  })\n  return out\n}\n\nfunction proposalCandidates(\n  req: CompletionRequirement,\n  reqIndex: number,\n  proposals: ProducedProposal[],\n): Candidate[] {\n  const reqText = `${req.title} ${req.category ?? ''}`\n  const out: Candidate[] = []\n  for (const p of proposals) {\n    // Pending or rejected work is not a completed deliverable.\n    if (p.status !== 'approved') continue\n    const score = tokenRecall(reqText, p.title)\n    if (score < MATCH_THRESHOLD) continue\n    const body = p.content ?? ''\n    out.push({\n      reqIndex,\n      itemKey: `proposal:${p.id}`,\n      score,\n      evidence: `approved proposal '${p.title}' matched (token recall ${score.toFixed(2)})`,\n      content: body.trim().length >= MIN_CONTENT_CHARS ? body : null,\n    })\n  }\n  return out\n}\n\nfunction toolCallCandidates(\n  req: CompletionRequirement,\n  reqIndex: number,\n  toolCalls: string[],\n): Candidate[] {\n  const out: Candidate[] = []\n  toolCalls.forEach((name, i) => {\n    const score = tokenRecall(req.title, name)\n    if (score < MATCH_THRESHOLD) return\n    out.push({\n      reqIndex,\n      itemKey: `tool:${i}`,\n      score,\n      evidence: `tool call '${name}' matched (token recall ${score.toFixed(2)})`,\n      content: null,\n    })\n  })\n  return out\n}\n\n/**\n * Verify whether a run completed the task. `checkCorrectness` is injected —\n * `createLlmCorrectnessChecker` for production, a deterministic stub in tests.\n *\n * Throws on a gold spec with no requirements: an eval task that requires\n * nothing is a misconfiguration, not a vacuously-complete task.\n */\nexport async function verifyCompletion(\n  gold: TaskGold,\n  state: ProducedState,\n  checkCorrectness: CorrectnessChecker,\n): Promise<CompletionVerdict> {\n  if (gold.requirements.length === 0) {\n    throw new Error(\n      `verifyCompletion: task '${gold.taskId}' has no requirements — malformed gold spec`,\n    )\n  }\n\n  // Collect every above-threshold (requirement, produced-item) candidate, then\n  // assign greedily by descending score: each requirement and each produced\n  // item is used at most once. One deliverable fulfils one requirement.\n  const candidates: Candidate[] = []\n  gold.requirements.forEach((req, i) => {\n    const by = req.satisfiedBy ?? 'any'\n    if (by === 'artifact' || by === 'any') {\n      candidates.push(...artifactCandidates(req, i, state.artifacts))\n    }\n    if (by === 'proposal' || by === 'any') {\n      candidates.push(...proposalCandidates(req, i, state.proposals))\n    }\n    if (by === 'tool-call' || by === 'any') {\n      candidates.push(...toolCallCandidates(req, i, state.toolCalls))\n    }\n  })\n  candidates.sort((a, b) => b.score - a.score)\n\n  const assigned = new Map<number, Candidate>()\n  const itemTaken = new Set<string>()\n  for (const c of candidates) {\n    if (assigned.has(c.reqIndex) || itemTaken.has(c.itemKey)) continue\n    assigned.set(c.reqIndex, c)\n    itemTaken.add(c.itemKey)\n  }\n\n  const requirements: RequirementCheck[] = []\n  for (let i = 0; i < gold.requirements.length; i++) {\n    const req = gold.requirements[i]!\n    const match = assigned.get(i)\n    const evidence: string[] = []\n    let correct: boolean | null = null\n\n    if (match) {\n      evidence.push(match.evidence)\n      if (match.content !== null) {\n        const r = await checkCorrectness(req, match.content)\n        correct = r.correct\n        evidence.push(`correctness: ${r.correct ? 'pass' : 'fail'} — ${r.reason}`)\n      } else {\n        evidence.push('correctness: not assessed — matched item carries no content')\n      }\n    } else {\n      const by = req.satisfiedBy ?? 'any'\n      const kind = by === 'any' ? 'artifact/proposal/tool-call' : by\n      evidence.push(`no produced ${kind} matched this requirement`)\n    }\n\n    const structurallyPresent = match !== undefined\n    const satisfied = structurallyPresent && correct !== false\n    requirements.push({\n      reqId: req.reqId,\n      title: req.title,\n      structurallyPresent,\n      correct,\n      satisfied,\n      evidence,\n    })\n  }\n\n  return completionVerdict({ taskId: gold.taskId, requirements })\n}\n\nexport interface LlmCorrectnessCheckerOpts {\n  model?: string\n  /** Max chars of artifact content sent to the checker. */\n  maxContentChars?: number\n}\n\n/** Parse the correctness checker's model response. Fails loud on a bad shape. */\nexport function parseCorrectnessResponse(raw: string): { correct: boolean; reason: string } {\n  const match = raw.match(/\\{[\\s\\S]*\\}/)\n  if (!match) {\n    throw new Error(`correctness checker: no JSON object in model response: ${raw.slice(0, 200)}`)\n  }\n  const parsed = JSON.parse(match[0]) as { correct?: unknown; reason?: unknown }\n  if (typeof parsed.correct !== 'boolean') {\n    throw new Error(`correctness checker: 'correct' is not a boolean in: ${match[0].slice(0, 200)}`)\n  }\n  return { correct: parsed.correct, reason: typeof parsed.reason === 'string' ? parsed.reason : '' }\n}\n\n/**\n * Production `CorrectnessChecker` — one LLM call per matched artifact,\n * deterministic (temperature 0), structured JSON out. Judges fulfilment\n * only: a plan, a gesture, or a description of what should be done does not\n * fulfil a requirement — the artifact must BE the deliverable.\n */\nexport function createLlmCorrectnessChecker(\n  tc: TCloud,\n  opts: LlmCorrectnessCheckerOpts = {},\n): CorrectnessChecker {\n  const model = opts.model ?? 'claude-sonnet-4-6'\n  const maxContentChars = opts.maxContentChars ?? 8000\n  return async (requirement, content) => {\n    const resp = await tc.chat({\n      model,\n      messages: [\n        {\n          role: 'system',\n          content:\n            'You verify whether a produced work artifact actually fulfils a stated requirement. Judge fulfilment only — is the deliverable substantively present and on-point — not polish. A plan to do it later, a vague gesture, or a description of what should be done does NOT fulfil a requirement; the artifact must BE the deliverable. Respond with a single JSON object: {\"correct\": boolean, \"reason\": string (<= 30 words)}.',\n        },\n        {\n          role: 'user',\n          content: `Requirement: ${requirement.title}\\n${\n            requirement.category ? `Category: ${requirement.category}\\n` : ''\n          }\\nProduced artifact:\\n${content.slice(0, maxContentChars)}`,\n        },\n      ],\n      temperature: 0,\n      maxTokens: 200,\n    })\n    const raw =\n      (resp as { choices?: { message?: { content?: string } }[] }).choices?.[0]?.message?.content ??\n      ''\n    return parseCorrectnessResponse(raw)\n  }\n}\n\n/** Stopwords for requirement-title tokenization — drops the imperative verbs\n *  ('review', 'update', …) common to deliverable titles so recall keys on the\n *  substantive nouns, not the boilerplate ask. */\nconst TITLE_STOPWORDS = new Set([\n  'the',\n  'a',\n  'an',\n  'and',\n  'or',\n  'for',\n  'to',\n  'of',\n  'in',\n  'on',\n  'with',\n  'review',\n  'update',\n  'new',\n  'proposed',\n])\n\n/**\n * Deterministic `CorrectnessChecker` — the no-LLM counterpart to\n * `createLlmCorrectnessChecker`. A produced item fulfils a requirement when its\n * content is substantive (≥ `minContentLength` chars) AND recalls ≥ `minRecall`\n * of the requirement title's significant tokens. No network — the default gate\n * for apps and tests without an LLM judge. Pass to `verifyCompletion` as the\n * checker.\n */\nexport function createTokenRecallChecker(\n  opts: { minRecall?: number; minContentLength?: number } = {},\n): CorrectnessChecker {\n  const minRecall = opts.minRecall ?? 0.5\n  const minLen = opts.minContentLength ?? 120\n  return async (requirement, content) => {\n    const body = content.trim()\n    if (body.length < minLen)\n      return {\n        correct: false,\n        reason: `content too thin (${body.length} chars) to be the deliverable`,\n      }\n    const titleTokens = requirement.title\n      .toLowerCase()\n      .split(/[^a-z0-9]+/)\n      .filter((t) => t.length > 2 && !TITLE_STOPWORDS.has(t))\n    if (titleTokens.length === 0)\n      return {\n        correct: true,\n        reason: 'requirement title has no significant tokens — structural match accepted',\n      }\n    const lower = body.toLowerCase()\n    const hits = titleTokens.filter((t) => lower.includes(t)).length\n    const recall = hits / titleTokens.length\n    return recall >= minRecall\n      ? {\n          correct: true,\n          reason: `content recalls ${hits}/${titleTokens.length} requirement tokens`,\n        }\n      : {\n          correct: false,\n          reason: `content recalls only ${hits}/${titleTokens.length} requirement tokens`,\n        }\n  }\n}\n","/**\n * Produced-state extraction — normalize a run's runtime event stream into the\n * typed `ProducedState` the completion oracle consumes.\n *\n * `ProducedState` answers \"what did the agent actually produce\" — vault\n * artifacts, proposals, tool calls. The runtime emits these as a stream of\n * events; this module is the single normalization point from that stream to\n * the shape `verifyCompletion` expects.\n *\n * Input is structurally typed (`RuntimeEventLike`) so this module does not\n * depend on agent-runtime — agent-runtime's `RuntimeStreamEvent` satisfies it\n * structurally. The `content` on `ArtifactEventLike` and the whole\n * `proposal_created` variant are the runtime-side enrichments this contract\n * requires; the runtime emits them, this module consumes them.\n */\n\nimport type { Artifact } from './artifact-validator'\nimport type { ProducedProposal, ProducedState } from './completion-verifier'\n\n/** A tool the agent invoked. */\nexport interface ToolCallEventLike {\n  type: 'tool_call'\n  toolName: string\n}\n\n/**\n * An artifact the agent produced. `content` is the enriched field — the\n * runtime's base `artifact` event carries only metadata; the completion\n * oracle needs the body to verify the deliverable, so the runtime emits it.\n */\nexport interface ArtifactEventLike {\n  type: 'artifact'\n  artifactId: string\n  name?: string\n  mimeType?: string\n  uri?: string\n  content?: string\n}\n\n/** A proposal / filing the agent created. */\nexport interface ProposalEventLike {\n  type: 'proposal_created'\n  proposalId: string\n  title: string\n  status?: 'pending' | 'approved' | 'rejected'\n}\n\n/**\n * The subset of runtime stream events `extractProducedState` consumes.\n * agent-runtime's full `RuntimeStreamEvent` union satisfies this structurally;\n * the `{ type: string }` catch-all keeps the input permissive so callers can\n * pass the whole unfiltered telemetry stream — unrecognized events are skipped.\n */\nexport type RuntimeEventLike =\n  | ToolCallEventLike\n  | ArtifactEventLike\n  | ProposalEventLike\n  | { type: string }\n\nfunction artifactKind(mimeType: string | undefined): string {\n  if (!mimeType) return 'file'\n  if (mimeType.includes('json')) return 'json'\n  if (mimeType.startsWith('text/')) return 'text'\n  return 'file'\n}\n\n/**\n * Normalize a run's runtime event stream into `ProducedState`.\n *\n * Pure and total — unrecognized event types are skipped. `toolCalls` is\n * deduplicated by name in first-seen order (completion cares about a tool's\n * presence, not its call count). An artifact with neither a name nor a uri\n * still yields an entry keyed by its `artifactId` so it is never silently\n * dropped; an artifact with no `content` yields empty content, which the\n * completion oracle's structural check then rejects on its own.\n */\nexport function extractProducedState(events: readonly RuntimeEventLike[]): ProducedState {\n  const artifacts: Artifact[] = []\n  const proposals: ProducedProposal[] = []\n  const toolCalls: string[] = []\n  const seenTools = new Set<string>()\n\n  for (const ev of events) {\n    if (ev.type === 'tool_call') {\n      const name = (ev as ToolCallEventLike).toolName\n      if (name && !seenTools.has(name)) {\n        seenTools.add(name)\n        toolCalls.push(name)\n      }\n    } else if (ev.type === 'artifact') {\n      const a = ev as ArtifactEventLike\n      artifacts.push({\n        kind: artifactKind(a.mimeType),\n        path: a.name ?? a.uri ?? a.artifactId,\n        content: a.content ?? '',\n      })\n    } else if (ev.type === 'proposal_created') {\n      const p = ev as ProposalEventLike\n      proposals.push({ id: p.proposalId, title: p.title, status: p.status ?? 'pending' })\n    }\n  }\n\n  return { artifacts, proposals, toolCalls }\n}\n","/**\n * @stable\n *\n * AgentProfile — the eval harness's unit of variation.\n *\n * A profile pins everything that changes agent behaviour for a benchmark\n * cell: the model, the active skills, the prompt version, the available\n * tools. Vary the profile — swap a model, add a skill — and re-run the suite\n * to benchmark the change. The scorecard keys a cell on\n * `(scenarioId, profileHash)`, so the model is not a separate axis: it lives\n * inside the profile, and two profiles with the same model but different\n * skills are different cells.\n *\n * `agentProfileHash` is the profile's behaviour identity. Two profiles that\n * produce the same agent behaviour share a hash (and a scorecard cell);\n * reordering `skills` or `tools` does not change it; the human-facing `id`\n * label does not affect it.\n */\n\nimport { createHash } from 'node:crypto'\nimport { ValidationError } from './errors'\nimport { canonicalize } from './pre-registration'\n\nexport interface AgentProfile {\n  /** Human-facing label, e.g. `sonnet-legal-skills-v3`. Not part of the hash. */\n  id: string\n  /** Model snapshot id this profile pins, e.g. `claude-sonnet-4-6@2025-04-15`. */\n  model: string\n  /** Skill ids/versions active in this profile — the primary behaviour lever. */\n  skills?: string[]\n  /** Prompt version identifier. */\n  promptVersion?: string\n  /** Tool ids available to the agent. */\n  tools?: string[]\n  /** Any other behaviour-bearing knobs that should fingerprint into the hash. */\n  metadata?: Record<string, string | number | boolean>\n}\n\n/**\n * Deterministic behaviour identity of a profile — a sha256 over the\n * behaviour-bearing fields. `skills` and `tools` are order-insensitive; the\n * `id` label is excluded. Throws on a profile with no `model` — an unkeyable\n * profile must fail loud rather than collapse into a blank-model cell.\n */\nexport function agentProfileHash(profile: AgentProfile): string {\n  if (typeof profile.model !== 'string' || profile.model.trim().length === 0) {\n    throw new ValidationError(`AgentProfile \"${profile.id}\" has no model — cannot hash`)\n  }\n  const behaviour = {\n    model: profile.model.trim(),\n    skills: [...(profile.skills ?? [])].sort(),\n    promptVersion: profile.promptVersion ?? null,\n    tools: [...(profile.tools ?? [])].sort(),\n    metadata: profile.metadata ?? {},\n  }\n  return createHash('sha256')\n    .update(JSON.stringify(canonicalize(behaviour)))\n    .digest('hex')\n}\n"],"mappings":";;;;;;;;AAmGO,SAAS,kBAAkB,OAGZ;AACpB,MAAI,MAAM,aAAa,WAAW,GAAG;AACnC,UAAM,IAAI;AAAA,MACR,4BAA4B,MAAM,MAAM;AAAA,IAC1C;AAAA,EACF;AACA,QAAM,iBAAiB,MAAM,aAAa,OAAO,CAAC,MAAM,EAAE,SAAS,EAAE;AACrE,QAAM,iBAAiB,iBAAiB,MAAM,aAAa;AAC3D,QAAM,gBAAgB,mBAAmB,MAAM,aAAa;AAC5D,SAAO;AAAA,IACL,QAAQ,MAAM;AAAA,IACd,cAAc,MAAM;AAAA,IACpB;AAAA,IACA;AAAA,IACA,OAAO;AAAA,IACP,OAAO;AAAA,EACT;AACF;AAYA,IAAM,YAAY,oBAAI,IAAI;AAAA,EACxB;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AACF,CAAC;AAED,IAAM,kBAAkB;AACxB,IAAM,oBAAoB;AAE1B,SAAS,OAAO,GAAwB;AACtC,SAAO,IAAI;AAAA,IACT,EACG,YAAY,EACZ,MAAM,YAAY,EAClB,OAAO,CAAC,MAAM,EAAE,SAAS,KAAK,CAAC,UAAU,IAAI,CAAC,CAAC;AAAA,EACpD;AACF;AAOA,SAAS,YAAY,iBAAyB,eAA+B;AAC3E,QAAM,MAAM,OAAO,eAAe;AAClC,MAAI,IAAI,SAAS,EAAG,QAAO;AAC3B,QAAM,OAAO,OAAO,aAAa;AACjC,MAAI,MAAM;AACV,aAAW,KAAK,IAAK,KAAI,KAAK,IAAI,CAAC,EAAG;AACtC,SAAO,MAAM,IAAI;AACnB;AAYA,SAAS,mBACP,KACA,UACA,WACa;AACb,QAAM,UAAU,GAAG,IAAI,KAAK,IAAI,IAAI,YAAY,EAAE;AAClD,QAAM,MAAmB,CAAC;AAC1B,YAAU,QAAQ,CAAC,GAAG,MAAM;AAC1B,SAAK,EAAE,WAAW,IAAI,KAAK,EAAE,SAAS,kBAAmB;AACzD,QAAI,QAAQ,YAAY,SAAS,GAAG,EAAE,QAAQ,EAAE,IAAI,EAAE,IAAI,EAAE;AAC5D,QAAI,IAAI,YAAY,EAAE,QAAQ,IAAI,SAAS,YAAY,MAAM,EAAE,KAAK,YAAY,GAAG;AACjF,cAAQ,KAAK,IAAI,OAAO,CAAC;AAAA,IAC3B;AACA,QAAI,QAAQ,gBAAiB;AAC7B,QAAI,KAAK;AAAA,MACP;AAAA,MACA,SAAS,YAAY,CAAC;AAAA,MACtB;AAAA,MACA,UAAU,aAAa,EAAE,QAAQ,EAAE,IAAI,2BAA2B,MAAM,QAAQ,CAAC,CAAC;AAAA,MAClF,SAAS,EAAE,WAAW;AAAA,IACxB,CAAC;AAAA,EACH,CAAC;AACD,SAAO;AACT;AAEA,SAAS,mBACP,KACA,UACA,WACa;AACb,QAAM,UAAU,GAAG,IAAI,KAAK,IAAI,IAAI,YAAY,EAAE;AAClD,QAAM,MAAmB,CAAC;AAC1B,aAAW,KAAK,WAAW;AAEzB,QAAI,EAAE,WAAW,WAAY;AAC7B,UAAM,QAAQ,YAAY,SAAS,EAAE,KAAK;AAC1C,QAAI,QAAQ,gBAAiB;AAC7B,UAAM,OAAO,EAAE,WAAW;AAC1B,QAAI,KAAK;AAAA,MACP;AAAA,MACA,SAAS,YAAY,EAAE,EAAE;AAAA,MACzB;AAAA,MACA,UAAU,sBAAsB,EAAE,KAAK,2BAA2B,MAAM,QAAQ,CAAC,CAAC;AAAA,MAClF,SAAS,KAAK,KAAK,EAAE,UAAU,oBAAoB,OAAO;AAAA,IAC5D,CAAC;AAAA,EACH;AACA,SAAO;AACT;AAEA,SAAS,mBACP,KACA,UACA,WACa;AACb,QAAM,MAAmB,CAAC;AAC1B,YAAU,QAAQ,CAAC,MAAM,MAAM;AAC7B,UAAM,QAAQ,YAAY,IAAI,OAAO,IAAI;AACzC,QAAI,QAAQ,gBAAiB;AAC7B,QAAI,KAAK;AAAA,MACP;AAAA,MACA,SAAS,QAAQ,CAAC;AAAA,MAClB;AAAA,MACA,UAAU,cAAc,IAAI,2BAA2B,MAAM,QAAQ,CAAC,CAAC;AAAA,MACvE,SAAS;AAAA,IACX,CAAC;AAAA,EACH,CAAC;AACD,SAAO;AACT;AASA,eAAsB,iBACpB,MACA,OACA,kBAC4B;AAC5B,MAAI,KAAK,aAAa,WAAW,GAAG;AAClC,UAAM,IAAI;AAAA,MACR,2BAA2B,KAAK,MAAM;AAAA,IACxC;AAAA,EACF;AAKA,QAAM,aAA0B,CAAC;AACjC,OAAK,aAAa,QAAQ,CAAC,KAAK,MAAM;AACpC,UAAM,KAAK,IAAI,eAAe;AAC9B,QAAI,OAAO,cAAc,OAAO,OAAO;AACrC,iBAAW,KAAK,GAAG,mBAAmB,KAAK,GAAG,MAAM,SAAS,CAAC;AAAA,IAChE;AACA,QAAI,OAAO,cAAc,OAAO,OAAO;AACrC,iBAAW,KAAK,GAAG,mBAAmB,KAAK,GAAG,MAAM,SAAS,CAAC;AAAA,IAChE;AACA,QAAI,OAAO,eAAe,OAAO,OAAO;AACtC,iBAAW,KAAK,GAAG,mBAAmB,KAAK,GAAG,MAAM,SAAS,CAAC;AAAA,IAChE;AAAA,EACF,CAAC;AACD,aAAW,KAAK,CAAC,GAAG,MAAM,EAAE,QAAQ,EAAE,KAAK;AAE3C,QAAM,WAAW,oBAAI,IAAuB;AAC5C,QAAM,YAAY,oBAAI,IAAY;AAClC,aAAW,KAAK,YAAY;AAC1B,QAAI,SAAS,IAAI,EAAE,QAAQ,KAAK,UAAU,IAAI,EAAE,OAAO,EAAG;AAC1D,aAAS,IAAI,EAAE,UAAU,CAAC;AAC1B,cAAU,IAAI,EAAE,OAAO;AAAA,EACzB;AAEA,QAAM,eAAmC,CAAC;AAC1C,WAAS,IAAI,GAAG,IAAI,KAAK,aAAa,QAAQ,KAAK;AACjD,UAAM,MAAM,KAAK,aAAa,CAAC;AAC/B,UAAM,QAAQ,SAAS,IAAI,CAAC;AAC5B,UAAM,WAAqB,CAAC;AAC5B,QAAI,UAA0B;AAE9B,QAAI,OAAO;AACT,eAAS,KAAK,MAAM,QAAQ;AAC5B,UAAI,MAAM,YAAY,MAAM;AAC1B,cAAM,IAAI,MAAM,iBAAiB,KAAK,MAAM,OAAO;AACnD,kBAAU,EAAE;AACZ,iBAAS,KAAK,gBAAgB,EAAE,UAAU,SAAS,MAAM,WAAM,EAAE,MAAM,EAAE;AAAA,MAC3E,OAAO;AACL,iBAAS,KAAK,kEAA6D;AAAA,MAC7E;AAAA,IACF,OAAO;AACL,YAAM,KAAK,IAAI,eAAe;AAC9B,YAAM,OAAO,OAAO,QAAQ,gCAAgC;AAC5D,eAAS,KAAK,eAAe,IAAI,2BAA2B;AAAA,IAC9D;AAEA,UAAM,sBAAsB,UAAU;AACtC,UAAM,YAAY,uBAAuB,YAAY;AACrD,iBAAa,KAAK;AAAA,MAChB,OAAO,IAAI;AAAA,MACX,OAAO,IAAI;AAAA,MACX;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,IACF,CAAC;AAAA,EACH;AAEA,SAAO,kBAAkB,EAAE,QAAQ,KAAK,QAAQ,aAAa,CAAC;AAChE;AASO,SAAS,yBAAyB,KAAmD;AAC1F,QAAM,QAAQ,IAAI,MAAM,aAAa;AACrC,MAAI,CAAC,OAAO;AACV,UAAM,IAAI,MAAM,0DAA0D,IAAI,MAAM,GAAG,GAAG,CAAC,EAAE;AAAA,EAC/F;AACA,QAAM,SAAS,KAAK,MAAM,MAAM,CAAC,CAAC;AAClC,MAAI,OAAO,OAAO,YAAY,WAAW;AACvC,UAAM,IAAI,MAAM,uDAAuD,MAAM,CAAC,EAAE,MAAM,GAAG,GAAG,CAAC,EAAE;AAAA,EACjG;AACA,SAAO,EAAE,SAAS,OAAO,SAAS,QAAQ,OAAO,OAAO,WAAW,WAAW,OAAO,SAAS,GAAG;AACnG;AAQO,SAAS,4BACd,IACA,OAAkC,CAAC,GACf;AACpB,QAAM,QAAQ,KAAK,SAAS;AAC5B,QAAM,kBAAkB,KAAK,mBAAmB;AAChD,SAAO,OAAO,aAAa,YAAY;AACrC,UAAM,OAAO,MAAM,GAAG,KAAK;AAAA,MACzB;AAAA,MACA,UAAU;AAAA,QACR;AAAA,UACE,MAAM;AAAA,UACN,SACE;AAAA,QACJ;AAAA,QACA;AAAA,UACE,MAAM;AAAA,UACN,SAAS,gBAAgB,YAAY,KAAK;AAAA,EACxC,YAAY,WAAW,aAAa,YAAY,QAAQ;AAAA,IAAO,EACjE;AAAA;AAAA,EAAyB,QAAQ,MAAM,GAAG,eAAe,CAAC;AAAA,QAC5D;AAAA,MACF;AAAA,MACA,aAAa;AAAA,MACb,WAAW;AAAA,IACb,CAAC;AACD,UAAM,MACH,KAA4D,UAAU,CAAC,GAAG,SAAS,WACpF;AACF,WAAO,yBAAyB,GAAG;AAAA,EACrC;AACF;AAKA,IAAM,kBAAkB,oBAAI,IAAI;AAAA,EAC9B;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AACF,CAAC;AAUM,SAAS,yBACd,OAA0D,CAAC,GACvC;AACpB,QAAM,YAAY,KAAK,aAAa;AACpC,QAAM,SAAS,KAAK,oBAAoB;AACxC,SAAO,OAAO,aAAa,YAAY;AACrC,UAAM,OAAO,QAAQ,KAAK;AAC1B,QAAI,KAAK,SAAS;AAChB,aAAO;AAAA,QACL,SAAS;AAAA,QACT,QAAQ,qBAAqB,KAAK,MAAM;AAAA,MAC1C;AACF,UAAM,cAAc,YAAY,MAC7B,YAAY,EACZ,MAAM,YAAY,EAClB,OAAO,CAAC,MAAM,EAAE,SAAS,KAAK,CAAC,gBAAgB,IAAI,CAAC,CAAC;AACxD,QAAI,YAAY,WAAW;AACzB,aAAO;AAAA,QACL,SAAS;AAAA,QACT,QAAQ;AAAA,MACV;AACF,UAAM,QAAQ,KAAK,YAAY;AAC/B,UAAM,OAAO,YAAY,OAAO,CAAC,MAAM,MAAM,SAAS,CAAC,CAAC,EAAE;AAC1D,UAAM,SAAS,OAAO,YAAY;AAClC,WAAO,UAAU,YACb;AAAA,MACE,SAAS;AAAA,MACT,QAAQ,mBAAmB,IAAI,IAAI,YAAY,MAAM;AAAA,IACvD,IACA;AAAA,MACE,SAAS;AAAA,MACT,QAAQ,wBAAwB,IAAI,IAAI,YAAY,MAAM;AAAA,IAC5D;AAAA,EACN;AACF;;;ACzYA,SAAS,aAAa,UAAsC;AAC1D,MAAI,CAAC,SAAU,QAAO;AACtB,MAAI,SAAS,SAAS,MAAM,EAAG,QAAO;AACtC,MAAI,SAAS,WAAW,OAAO,EAAG,QAAO;AACzC,SAAO;AACT;AAYO,SAAS,qBAAqB,QAAoD;AACvF,QAAM,YAAwB,CAAC;AAC/B,QAAM,YAAgC,CAAC;AACvC,QAAM,YAAsB,CAAC;AAC7B,QAAM,YAAY,oBAAI,IAAY;AAElC,aAAW,MAAM,QAAQ;AACvB,QAAI,GAAG,SAAS,aAAa;AAC3B,YAAM,OAAQ,GAAyB;AACvC,UAAI,QAAQ,CAAC,UAAU,IAAI,IAAI,GAAG;AAChC,kBAAU,IAAI,IAAI;AAClB,kBAAU,KAAK,IAAI;AAAA,MACrB;AAAA,IACF,WAAW,GAAG,SAAS,YAAY;AACjC,YAAM,IAAI;AACV,gBAAU,KAAK;AAAA,QACb,MAAM,aAAa,EAAE,QAAQ;AAAA,QAC7B,MAAM,EAAE,QAAQ,EAAE,OAAO,EAAE;AAAA,QAC3B,SAAS,EAAE,WAAW;AAAA,MACxB,CAAC;AAAA,IACH,WAAW,GAAG,SAAS,oBAAoB;AACzC,YAAM,IAAI;AACV,gBAAU,KAAK,EAAE,IAAI,EAAE,YAAY,OAAO,EAAE,OAAO,QAAQ,EAAE,UAAU,UAAU,CAAC;AAAA,IACpF;AAAA,EACF;AAEA,SAAO,EAAE,WAAW,WAAW,UAAU;AAC3C;;;ACpFA,SAAS,kBAAkB;AAyBpB,SAAS,iBAAiB,SAA+B;AAC9D,MAAI,OAAO,QAAQ,UAAU,YAAY,QAAQ,MAAM,KAAK,EAAE,WAAW,GAAG;AAC1E,UAAM,IAAI,gBAAgB,iBAAiB,QAAQ,EAAE,mCAA8B;AAAA,EACrF;AACA,QAAM,YAAY;AAAA,IAChB,OAAO,QAAQ,MAAM,KAAK;AAAA,IAC1B,QAAQ,CAAC,GAAI,QAAQ,UAAU,CAAC,CAAE,EAAE,KAAK;AAAA,IACzC,eAAe,QAAQ,iBAAiB;AAAA,IACxC,OAAO,CAAC,GAAI,QAAQ,SAAS,CAAC,CAAE,EAAE,KAAK;AAAA,IACvC,UAAU,QAAQ,YAAY,CAAC;AAAA,EACjC;AACA,SAAO,WAAW,QAAQ,EACvB,OAAO,KAAK,UAAU,aAAa,SAAS,CAAC,CAAC,EAC9C,OAAO,KAAK;AACjB;","names":[]}