@tangle-network/agent-eval 0.134.1 → 0.134.2

package/dist/multishot/index.js

@@ -1,8 +1,10 @@
+import { i as estimateCost, o as isModelPriced } from "../metrics-C9YY1OcL.js";
 import { t as runAgentMatrix } from "../matrix-BzQnu2S6.js";
 import { mkdirSync, writeFileSync } from "node:fs";
 import { join } from "node:path";
 //#region src/multishot/router.ts
 async function routerCompletion(req) {
+	const startedAt = Date.now();
 	const body = {
 		model: req.model,
 		messages: req.messages,
@@ -27,9 +29,14 @@ async function routerCompletion(req) {
 	const json = await res.json();
 	const choice = json.choices[0];
 	if (!choice) throw new Error(`router returned no choices: ${JSON.stringify(json).slice(0, 200)}`);
+	const rawCost = json._response_cost ?? json.cost_usd;
+	const costUsd = typeof rawCost === "number" && Number.isFinite(rawCost) && rawCost >= 0 ? rawCost : void 0;
 	return {
 		message: choice.message,
-		usage: json.usage
+		usage: json.usage,
+		...costUsd === void 0 ? {} : { costUsd },
+		model: typeof json.model === "string" && json.model ? json.model : req.model,
+		durationMs: Date.now() - startedAt
 	};
 }
 function estimateRouterCost(model, usage) {
@@ -179,8 +186,11 @@ async function runJudge(judge, input) {
 	const model = judge.model ?? process.env.JUDGE_MODEL ?? "openai/gpt-4o-mini";
 	const prompt = judge.buildPrompt(input);
 	let raw = "";
+	let llmCall;
+	let cost;
+	const startedAt = Date.now();
 	try {
-		const { message } = await routerCompletion({
+		const response = await routerCompletion({
 			apiKey,
 			baseUrl,
 			model,
@@ -194,13 +204,23 @@ async function runJudge(judge, input) {
 				content: prompt
 			}]
 		});
-		raw = (message.content ?? "").trim();
+		const call = judgeCallMetadata(response);
+		llmCall = call.llmCall;
+		cost = call.cost;
+		raw = (response.message.content ?? "").trim();
 	} catch (err) {
 		return {
-			dimensions: {},
-			composite: 0,
-			failed: true,
-			notes: `judge ${judge.name} call failed: ${err instanceof Error ? err.message : String(err)}`
+			score: {
+				dimensions: {},
+				composite: 0,
+				failed: true,
+				notes: `judge ${judge.name} call failed: ${err instanceof Error ? err.message : String(err)}`,
+				llmCall: unknownJudgeCall(model, Date.now() - startedAt)
+			},
+			cost: {
+				kind: "uncaptured",
+				usd: null
+			}
 		};
 	}
 	let parsed = null;
@@ -209,10 +229,14 @@ async function runJudge(judge, input) {
 		parsed = JSON.parse(cleaned);
 	} catch {
 		return {
-			dimensions: {},
-			composite: 0,
-			failed: true,
-			notes: `judge ${judge.name} returned non-JSON: ${raw.slice(0, 200)}`
+			score: {
+				dimensions: {},
+				composite: 0,
+				failed: true,
+				notes: `judge ${judge.name} returned non-JSON: ${raw.slice(0, 200)}`,
+				llmCall
+			},
+			cost
 		};
 	}
 	const dimensions = {};
@@ -224,11 +248,63 @@ async function runJudge(judge, input) {
 		sum += clamped;
 	}
 	return {
-		dimensions,
-		composite: judge.dimensions.length === 0 ? 0 : sum / judge.dimensions.length,
-		notes: typeof parsed.notes === "string" ? parsed.notes : ""
+		score: {
+			dimensions,
+			composite: judge.dimensions.length === 0 ? 0 : sum / judge.dimensions.length,
+			notes: typeof parsed.notes === "string" ? parsed.notes : "",
+			llmCall
+		},
+		cost
+	};
+}
+function judgeCallMetadata(response) {
+	const usage = canonicalUsage(response.usage);
+	return {
+		llmCall: {
+			usage,
+			costUsd: response.costUsd ?? null,
+			model: response.model,
+			durationMs: response.durationMs
+		},
+		cost: response.costUsd !== void 0 ? {
+			kind: "observed",
+			usd: response.costUsd
+		} : usage.captured === false || !isModelPriced(response.model) ? {
+			kind: "uncaptured",
+			usd: null
+		} : {
+			kind: "estimated",
+			usd: estimateCost(usage.promptTokens, usage.completionTokens, response.model)
+		}
 	};
 }
+function canonicalUsage(usage) {
+	const promptTokens = tokenCount(usage?.prompt_tokens);
+	const completionTokens = tokenCount(usage?.completion_tokens);
+	const captured = promptTokens !== void 0 && completionTokens !== void 0;
+	return {
+		promptTokens: promptTokens ?? 0,
+		completionTokens: completionTokens ?? 0,
+		totalTokens: (promptTokens ?? 0) + (completionTokens ?? 0),
+		captured
+	};
+}
+function unknownJudgeCall(model, durationMs) {
+	return {
+		usage: {
+			promptTokens: 0,
+			completionTokens: 0,
+			totalTokens: 0,
+			captured: false
+		},
+		costUsd: null,
+		model,
+		durationMs
+	};
+}
+function tokenCount(value) {
+	return typeof value === "number" && Number.isSafeInteger(value) && value >= 0 ? value : void 0;
+}
 /** Convenience: stringified dimension list for inclusion in a judge prompt.
 *  Returns lines like `- audience_fit: Does this match what the audience cares about? (0-10)`. */
 function renderDimensions(dims) {
@@ -526,7 +602,7 @@ async function runMultishotMatrix(opts) {
 			});
 			const codeArtifacts = sim.artifacts.filter((a) => codeTypes.has(a.type));
 			const contentArtifacts = sim.artifacts.filter((a) => contentTypes.has(a.type));
-			const [conversation, codeReviews, contentReviews] = await Promise.all([
+			const [conversationRun, codeReviewRuns, contentReviewRuns] = await Promise.all([
 				runJudge(withJudgeMaxTokens(opts.judges.conversation, opts.judgeMaxTokens), {
 					transcript: sim.transcript,
 					persona
@@ -534,20 +610,29 @@ async function runMultishotMatrix(opts) {
 				opts.judges.codeReview ? Promise.all(codeArtifacts.map((artifact) => runJudge(withJudgeMaxTokens(opts.judges.codeReview, opts.judgeMaxTokens), {
 					artifact,
 					persona
-				}).then((s) => ({
-					...s,
-					turn: artifact.turn,
-					type: artifact.type
+				}).then((result) => ({
+					score: {
+						...result.score,
+						turn: artifact.turn,
+						type: artifact.type
+					},
+					cost: result.cost
 				})))) : Promise.resolve([]),
 				opts.judges.contentQuality ? Promise.all(contentArtifacts.map((artifact) => runJudge(withJudgeMaxTokens(opts.judges.contentQuality, opts.judgeMaxTokens), {
 					artifact,
 					persona
-				}).then((s) => ({
-					...s,
-					turn: artifact.turn,
-					type: artifact.type
+				}).then((result) => ({
+					score: {
+						...result.score,
+						turn: artifact.turn,
+						type: artifact.type
+					},
+					cost: result.cost
 				})))) : Promise.resolve([])
 			]);
+			const conversation = conversationRun.score;
+			const codeReviews = codeReviewRuns.map((run) => run.score);
+			const contentReviews = contentReviewRuns.map((run) => run.score);
 			const { composite, codeComposite, contentComposite, allJudgesFailed } = computeCellComposite({
 				conversation,
 				codeReviews: opts.judges.codeReview ? codeReviews : void 0,
@@ -574,6 +659,13 @@ async function runMultishotMatrix(opts) {
 			if (opts.judges.codeReview) notes.push(`code=${codeComposite.toFixed(1)}`);
 			if (opts.judges.contentQuality) notes.push(`content=${contentComposite.toFixed(1)}`);
 			if (allJudgesFailed) notes.push("all-judges-failed");
+			const judgeRuns = [
+				conversationRun,
+				...codeReviewRuns,
+				...contentReviewRuns
+			];
+			const judgeCostUsd = judgeRuns.reduce((sum, run) => sum + (run.cost.usd ?? 0), 0);
+			if (judgeRuns.some((run) => run.cost.kind === "uncaptured")) notes.push("judge-cost-incomplete");
 			return {
 				output: {
 					turns: sim.transcript.length,
@@ -585,7 +677,7 @@ async function runMultishotMatrix(opts) {
 					score: composite,
 					notes: notes.join(" ")
 				},
-				costUsd: sim.costUsd,
+				costUsd: sim.costUsd + judgeCostUsd,
 				durationMs: sim.durationMs
 			};
 		}

package/dist/multishot/index.js.map

@@ -1 +1 @@
-{"version":3,"file":"index.js","names":[],"sources":["../../src/multishot/router.ts","../../src/multishot/default-tools.ts","../../src/multishot/judges.ts","../../src/multishot/types.ts","../../src/multishot/multishot.ts","../../src/multishot/matrix.ts"],"sourcesContent":["// Router fetch helper — single source of truth for OpenAI-compat calls\n// against the Tangle router. Used by the driver, agent, judges, and the\n// default tool executors.\n\nimport type { MultishotToolDefinition } from './types'\n\nexport interface RouterCompletionRequest {\n  apiKey: string\n  baseUrl: string\n  model: string\n  messages: Array<Record<string, unknown>>\n  tools?: MultishotToolDefinition[]\n  temperature?: number\n  maxTokens?: number\n  signal?: AbortSignal\n}\n\nexport interface RouterToolCall {\n  id: string\n  type: 'function'\n  function: { name: string; arguments: string }\n}\n\nexport interface RouterCompletionResponse {\n  message: { content?: string | null; tool_calls?: RouterToolCall[] }\n  usage?: { prompt_tokens?: number; completion_tokens?: number }\n}\n\nexport async function routerCompletion(\n  req: RouterCompletionRequest,\n): Promise<RouterCompletionResponse> {\n  const body: Record<string, unknown> = {\n    model: req.model,\n    messages: req.messages,\n    temperature: req.temperature ?? 0.7,\n    max_tokens: req.maxTokens ?? 2000,\n  }\n  if (req.tools?.length) body.tools = req.tools\n  const url = `${req.baseUrl.replace(/\\/+$/, '')}/chat/completions`\n  const res = await fetch(url, {\n    method: 'POST',\n    headers: { Authorization: `Bearer ${req.apiKey}`, 'Content-Type': 'application/json' },\n    body: JSON.stringify(body),\n    signal: req.signal,\n  })\n  if (!res.ok) {\n    const text = await res.text()\n    throw new Error(`router ${res.status}: ${text.slice(0, 300)}`)\n  }\n  const json = (await res.json()) as {\n    choices: Array<{ message: { content?: string | null; tool_calls?: RouterToolCall[] } }>\n    usage?: { prompt_tokens?: number; completion_tokens?: number }\n  }\n  const choice = json.choices[0]\n  if (!choice) throw new Error(`router returned no choices: ${JSON.stringify(json).slice(0, 200)}`)\n  return { message: choice.message, usage: json.usage }\n}\n\n// Rough per-model cost estimator. Used for cost-ceiling enforcement.\n// Underestimates Anthropic, overestimates oss models — fine for ceilings.\nexport function estimateRouterCost(\n  model: string,\n  usage?: { prompt_tokens?: number; completion_tokens?: number },\n): number {\n  if (!usage) return 0\n  const inputTok = usage.prompt_tokens ?? 0\n  const outputTok = usage.completion_tokens ?? 0\n  let inPer1k = 0.003\n  let outPer1k = 0.015\n  if (model.includes('gpt-4o-mini')) {\n    inPer1k = 0.00015\n    outPer1k = 0.0006\n  } else if (model.includes('gpt-5.4') || model.includes('claude-sonnet')) {\n    inPer1k = 0.003\n    outPer1k = 0.015\n  } else if (model.includes('kimi') || model.includes('glm') || model.includes('deepseek')) {\n    inPer1k = 0.0005\n    outPer1k = 0.002\n  }\n  return (inputTok * inPer1k + outputTok * outPer1k) / 1000\n}\n\nexport function defaultRouterBaseUrl(): string {\n  return (process.env.TANGLE_ROUTER_BASE_URL ?? 'https://router.tangle.tools/v1').replace(\n    /\\/+$/,\n    '',\n  )\n}\n\nexport function requireRouterApiKey(): string {\n  const key = process.env.TANGLE_API_KEY\n  if (!key) throw new Error('multishot requires TANGLE_API_KEY (router-scoped sk-tan-* key)')\n  return key\n}\n","// Default delegate_research + delegate_code tools and their inline executors.\n//\n// Consumers can override either by passing their own tools + executors to\n// runMultishot. The defaults are sufficient for most domains — point the\n// researcher system prompt at your domain's citation style and the coder\n// at your preferred language.\n\nimport { estimateRouterCost, routerCompletion } from './router'\nimport type { MultishotToolDefinition, MultishotToolExecutor } from './types'\n\nexport const DEFAULT_RESEARCHER_MODEL = 'openai/gpt-4o-mini'\nexport const DEFAULT_CODER_MODEL = 'openai/gpt-4o-mini'\n\nexport interface DefaultResearcherConfig {\n  /** Replace the system prompt to bias the researcher toward a domain's\n   *  citation style. Defaults to a generic \"cite sources by name\" prompt. */\n  systemPrompt?: string\n  model?: string\n}\n\nexport interface DefaultCoderConfig {\n  /** Replace the system prompt to bias the coder toward a language /\n   *  framework / artifact style. */\n  systemPrompt?: string\n  model?: string\n}\n\nconst GENERIC_RESEARCHER_SYSTEM =\n  'You are a research specialist. Return a markdown brief with 3-5 findings. Each finding cites a specific source by name. Add a confidence level (high/medium/low) per finding. No fluff, no preamble.'\n\nconst GENERIC_CODER_SYSTEM =\n  'You are an expert engineer. Output ONE fenced code block containing the complete solution. Inline-comment non-obvious decisions. No explanation outside the block.'\n\nexport const DEFAULT_DELEGATE_RESEARCH_TOOL: MultishotToolDefinition = {\n  type: 'function',\n  function: {\n    name: 'delegate_research',\n    description:\n      'Research a topic deeply via specialist. Returns evidence-bearing items with citations. Use for audience research, competitive intel, regulatory landscape, market data, citation-grounded analysis.',\n    parameters: {\n      type: 'object',\n      properties: {\n        question: { type: 'string', description: 'Specific question to research' },\n        scope: {\n          type: 'string',\n          description: 'Optional scope: time window, geography, jurisdiction, segment',\n        },\n      },\n      required: ['question'],\n    },\n  },\n}\n\nexport const DEFAULT_DELEGATE_CODE_TOOL: MultishotToolDefinition = {\n  type: 'function',\n  function: {\n    name: 'delegate_code',\n    description:\n      'Generate a runnable script, template, pipeline, or tool via specialist. Returns complete working code or structured markdown. Use for content pipelines, calc snippets, dashboards, compliance checklists, deadline trackers.',\n    parameters: {\n      type: 'object',\n      properties: {\n        goal: { type: 'string', description: 'What the code must accomplish' },\n        language: {\n          type: 'string',\n          description: 'Optional language preference (default: TypeScript)',\n        },\n      },\n      required: ['goal'],\n    },\n  },\n}\n\nexport function createResearchExecutor(\n  config: DefaultResearcherConfig = {},\n): MultishotToolExecutor {\n  const systemPrompt = config.systemPrompt ?? GENERIC_RESEARCHER_SYSTEM\n  const model = config.model ?? DEFAULT_RESEARCHER_MODEL\n  return async (args, ctx) => {\n    const question = String(args.question ?? '')\n    const scope = args.scope ? String(args.scope) : undefined\n    const { message, usage } = await routerCompletion({\n      apiKey: ctx.apiKey,\n      baseUrl: ctx.baseUrl,\n      model,\n      temperature: 0.3,\n      maxTokens: 1800,\n      messages: [\n        { role: 'system', content: systemPrompt },\n        { role: 'user', content: `Research: ${question}${scope ? `\\nScope: ${scope}` : ''}` },\n      ],\n      signal: ctx.signal,\n    })\n    return { content: message.content ?? '', costUsd: estimateRouterCost(model, usage) }\n  }\n}\n\nexport function createCodeExecutor(config: DefaultCoderConfig = {}): MultishotToolExecutor {\n  const systemPrompt = config.systemPrompt ?? GENERIC_CODER_SYSTEM\n  const model = config.model ?? DEFAULT_CODER_MODEL\n  return async (args, ctx) => {\n    const goal = String(args.goal ?? '')\n    const language = args.language ? String(args.language) : 'TypeScript'\n    const { message, usage } = await routerCompletion({\n      apiKey: ctx.apiKey,\n      baseUrl: ctx.baseUrl,\n      model,\n      temperature: 0.2,\n      maxTokens: 2000,\n      messages: [\n        { role: 'system', content: `${systemPrompt}\\n\\nLanguage: ${language}` },\n        { role: 'user', content: `Produce: ${goal}` },\n      ],\n      signal: ctx.signal,\n    })\n    return { content: message.content ?? '', costUsd: estimateRouterCost(model, usage) }\n  }\n}\n\nexport interface DefaultToolsConfig {\n  research?: DefaultResearcherConfig\n  code?: DefaultCoderConfig\n  /** When true (default), each tool result is recorded as a typed artifact:\n   *  research → type='research', code → type='code'. */\n  recordArtifacts?: boolean\n}\n\nexport interface DefaultToolsBundle {\n  tools: MultishotToolDefinition[]\n  executors: Record<string, MultishotToolExecutor>\n  artifactTypeFor: (toolName: string) => string | undefined\n}\n\nexport function defaultDelegationTools(config: DefaultToolsConfig = {}): DefaultToolsBundle {\n  return {\n    tools: [DEFAULT_DELEGATE_RESEARCH_TOOL, DEFAULT_DELEGATE_CODE_TOOL],\n    executors: {\n      delegate_research: createResearchExecutor(config.research),\n      delegate_code: createCodeExecutor(config.code),\n    },\n    artifactTypeFor: (name) =>\n      name === 'delegate_research' ? 'research' : name === 'delegate_code' ? 'code' : undefined,\n  }\n}\n\nexport { defaultRouterBaseUrl } from './router'\n","// Generic judge runner — domain consumers configure dimensions + prompts.\n//\n// Three judge slots are conventional for multishot eval:\n//   - conversation (scores the full transcript)\n//   - codeReview (scores each code artifact)\n//   - contentQuality (scores each non-code artifact)\n//\n// But the runJudge primitive is fully generic — any T → JudgeScore mapping.\n\nimport type { JudgeScore } from '../campaign/types'\nimport { defaultRouterBaseUrl, requireRouterApiKey, routerCompletion } from './router'\n\n// Canonical declaration lives in campaign/types.ts. Multishot emits the same\n// shape on its producer-defined 0-10 scale.\nexport type { JudgeScore } from '../campaign/types'\n\nexport const DEFAULT_JUDGE_MODEL = 'openai/gpt-4o-mini'\n\nexport interface JudgeDimension {\n  /** JSON field name + score key. */\n  key: string\n  /** Description shown in the judge's user prompt. */\n  description: string\n}\n\nexport interface JudgeConfig<TInput> {\n  /** Display name (for trace + log). */\n  name: string\n  /** Model used for this judge. */\n  model?: string\n  /** 0-10 scored dimensions. */\n  dimensions: JudgeDimension[]\n  /** Judge system prompt — sets persona + JSON-only constraint. */\n  systemPrompt: string\n  /** Build the user prompt from the typed input. Must include \"Respond with\n   *  ONLY this JSON: { ... }\" listing each dimension key. */\n  buildPrompt: (input: TInput) => string\n  /** Optional model + api overrides. */\n  apiKey?: string\n  baseUrl?: string\n  /** Maximum output tokens for the judge response. Defaults to 1500. */\n  maxTokens?: number\n}\n\nexport async function runJudge<TInput>(\n  judge: JudgeConfig<TInput>,\n  input: TInput,\n): Promise<JudgeScore> {\n  const apiKey = judge.apiKey ?? requireRouterApiKey()\n  const baseUrl = judge.baseUrl ?? defaultRouterBaseUrl()\n  const model = judge.model ?? process.env.JUDGE_MODEL ?? DEFAULT_JUDGE_MODEL\n  const prompt = judge.buildPrompt(input)\n  let raw = ''\n  try {\n    const { message } = await routerCompletion({\n      apiKey,\n      baseUrl,\n      model,\n      temperature: 0,\n      maxTokens: judge.maxTokens ?? 1500,\n      messages: [\n        { role: 'system', content: judge.systemPrompt },\n        { role: 'user', content: prompt },\n      ],\n    })\n    raw = (message.content ?? '').trim()\n  } catch (err) {\n    // failed:true lets consumers reading `.composite` keep working while\n    // aggregators exclude this score from means instead of averaging a zero.\n    return {\n      dimensions: {},\n      composite: 0,\n      failed: true,\n      notes: `judge ${judge.name} call failed: ${err instanceof Error ? err.message : String(err)}`,\n    }\n  }\n\n  let parsed: Record<string, unknown> | null = null\n  try {\n    const cleaned = raw\n      .replace(/^```json\\s*/i, '')\n      .replace(/```\\s*$/, '')\n      .trim()\n    parsed = JSON.parse(cleaned) as Record<string, unknown>\n  } catch {\n    return {\n      dimensions: {},\n      composite: 0,\n      failed: true,\n      notes: `judge ${judge.name} returned non-JSON: ${raw.slice(0, 200)}`,\n    }\n  }\n\n  const dimensions: Record<string, number> = {}\n  let sum = 0\n  for (const dim of judge.dimensions) {\n    const v = Number(parsed[dim.key] ?? 0)\n    const clamped = Number.isFinite(v) ? Math.max(0, Math.min(10, v)) : 0\n    dimensions[dim.key] = clamped\n    sum += clamped\n  }\n  return {\n    dimensions,\n    composite: judge.dimensions.length === 0 ? 0 : sum / judge.dimensions.length,\n    notes: typeof parsed.notes === 'string' ? parsed.notes : '',\n  }\n}\n\n/** Convenience: stringified dimension list for inclusion in a judge prompt.\n *  Returns lines like `- audience_fit: Does this match what the audience cares about? (0-10)`. */\nexport function renderDimensions(dims: readonly JudgeDimension[]): string {\n  return dims.map((d) => `- ${d.key}: ${d.description}`).join('\\n')\n}\n\n/** Convenience: build the \"Respond with ONLY this JSON\" footer for a judge prompt. */\nexport function renderJsonFooter(dims: readonly JudgeDimension[]): string {\n  const fields = dims.map((d) => `\"${d.key}\":N`).join(',')\n  return `Respond with ONLY this JSON (no markdown, no preamble):\\n{${fields},\"notes\":\"1-2 sentence critique\"}`\n}\n","// Public types for the multishot substrate.\n\nexport interface MultishotMessage {\n  role: 'user' | 'assistant' | 'tool'\n  content: string\n  toolCallId?: string\n  toolCalls?: Array<{ id: string; name: string; args: Record<string, unknown> }>\n}\n\nexport interface MultishotArtifact {\n  type: string\n  turn: number\n  invocation: { name: string; args: Record<string, unknown> }\n  content: string\n}\n\nexport interface MultishotResult {\n  transcript: MultishotMessage[]\n  artifacts: MultishotArtifact[]\n  toolCalls: number\n  durationMs: number\n  costUsd: number\n}\n\nexport interface MultishotToolDefinition {\n  type: 'function'\n  function: {\n    name: string\n    description: string\n    parameters: Record<string, unknown>\n  }\n}\n\n/** One chat-completion request the multishot loop issues for a single agent\n *  (or driver) inference step. Mirrors the OpenAI-compat body the loop would\n *  otherwise POST to the Tangle router. */\nexport interface MultishotTransportRequest {\n  model: string\n  messages: Array<Record<string, unknown>>\n  tools?: MultishotToolDefinition[]\n  temperature?: number\n  maxTokens?: number\n  signal?: AbortSignal\n}\n\nexport interface MultishotTransportToolCall {\n  id: string\n  type: 'function'\n  function: { name: string; arguments: string }\n}\n\nexport interface MultishotTransportResponse {\n  message: { content?: string | null; tool_calls?: MultishotTransportToolCall[] }\n  usage?: { prompt_tokens?: number; completion_tokens?: number }\n  /** Actual spend for this call. When omitted, the loop meters cost from\n   *  `usage` via the per-model router estimator (estimateRouterCost). */\n  costUsd?: number\n}\n\n/** Execution seam for one leg of the multishot loop. When provided, it\n *  replaces the internal router HTTP call for that leg — the loop still owns\n *  turn scheduling, tool dispatch, transcript capture, and cost metering.\n *  agent-eval has no dependency on agent-runtime; adapt agent-runtime's\n *  resolveAgentBackend (or any sandbox/cli-bridge/router client) into this\n *  signature product-side. */\nexport type MultishotTransport = (\n  req: MultishotTransportRequest,\n) => Promise<MultishotTransportResponse>\n\nexport type MultishotToolExecutor = (\n  args: Record<string, unknown>,\n  ctx: { apiKey: string; baseUrl: string; signal?: AbortSignal },\n) => Promise<{ content: string; costUsd: number }>\n\nexport interface MultishotPersona {\n  /** Stable identifier — used for per-cell artifact paths + matrix axis keys. */\n  id: string\n  /** Per-domain payload (income/profile/voice/etc.) shaped by the consumer. */\n  [k: string]: unknown\n}\n\nexport interface MultishotShape<TPersona extends MultishotPersona> {\n  /** Opening user message (turn 0) — the persona's first ask. */\n  buildOpener: (persona: TPersona) => string\n  /** System prompt the driver LLM uses to roleplay the persona. Should set\n   *  voice, goals, constraints, time-pressure, and the \"never go silent\" rule. */\n  buildDriverSystemPrompt: (persona: TPersona) => string\n}\n\nexport class MultishotDriverEmptyError extends Error {\n  constructor(public readonly turn: number) {\n    super(`multishot: driver returned empty content twice at turn ${turn} — failing loud`)\n    this.name = 'MultishotDriverEmptyError'\n  }\n}\n\nexport class MultishotFatalToolError extends Error {\n  constructor(message: string) {\n    super(message)\n    this.name = 'MultishotFatalToolError'\n  }\n}\n","// Multi-turn driver-agent simulation with inline tool execution.\n//\n// The driver = LLM acting as the persona (reactive, non-deterministic).\n// The agent = the product agent under test (router call by default, or an\n// injected transport — with profile's systemPrompt + the configured tools).\n// Tool calls execute inline via the configured executors and feed back\n// into the agent's message log so the agent integrates the result.\n\nimport type { AgentProfile } from '@tangle-network/agent-interface'\nimport { defaultDelegationTools } from './default-tools'\nimport {\n  defaultRouterBaseUrl,\n  estimateRouterCost,\n  requireRouterApiKey,\n  routerCompletion,\n} from './router'\nimport {\n  type MultishotArtifact,\n  MultishotDriverEmptyError,\n  MultishotFatalToolError,\n  type MultishotMessage,\n  type MultishotPersona,\n  type MultishotResult,\n  type MultishotShape,\n  type MultishotToolDefinition,\n  type MultishotToolExecutor,\n  type MultishotTransport,\n} from './types'\n\nexport interface RunMultishotOptions<TPersona extends MultishotPersona> {\n  profile: AgentProfile\n  persona: TPersona\n  shape: MultishotShape<TPersona>\n  /** Tool definitions advertised to the agent. Defaults to delegate_research + delegate_code. */\n  tools?: MultishotToolDefinition[]\n  /** Map from tool name → executor invoked inline when the agent emits a tool_call. */\n  toolExecutors?: Record<string, MultishotToolExecutor>\n  /** Map from tool name → artifact type label written into MultishotArtifact.type.\n   *  Tools without a mapping still execute, but their results aren't surfaced as\n   *  typed artifacts (only as tool messages in the transcript). */\n  artifactTypeFor?: (toolName: string) => string | undefined\n  maxTurns?: number\n  agentModel?: string\n  driverModel?: string\n  /** Fallback driver models tried when the primary simulated-user model returns empty twice. */\n  driverFallbackModels?: string[]\n  /** Maximum output tokens for the first agent call in each assistant turn. */\n  agentMaxTokens?: number\n  /** Maximum output tokens for agent follow-up calls after tool results. */\n  toolFollowupMaxTokens?: number\n  /** Maximum output tokens for each simulated-user driver response. */\n  driverMaxTokens?: number\n  /** Maximum tool calls the agent may dispatch inside one assistant turn. */\n  maxToolDispatches?: number\n  /** Execution seam for the agent leg. When provided, every agent inference\n   *  step goes through this function instead of the router HTTP call; the\n   *  string levers (agentModel, apiKey, baseUrl) stop applying to that leg.\n   *  apiKey/baseUrl are still resolved for tool executors and any leg\n   *  without an injected transport. */\n  agentTransport?: MultishotTransport\n  /** Execution seam for the simulated-user driver leg (symmetric to\n   *  agentTransport). Driver model fallback rotation still applies — the\n   *  transport receives each candidate model in turn. */\n  driverTransport?: MultishotTransport\n  apiKey?: string\n  baseUrl?: string\n  signal?: AbortSignal\n}\n\nexport async function runMultishot<TPersona extends MultishotPersona>(\n  opts: RunMultishotOptions<TPersona>,\n): Promise<MultishotResult> {\n  const apiKey = opts.apiKey ?? requireRouterApiKey()\n  const baseUrl = opts.baseUrl ?? defaultRouterBaseUrl()\n  const maxTurns = opts.maxTurns ?? 10\n  const maxToolDispatches = opts.maxToolDispatches ?? 4\n  const agentModel = opts.agentModel ?? 'openai/gpt-5.4'\n  const driverModel = opts.driverModel ?? 'openai/gpt-4o-mini'\n  const driverModels = [driverModel, ...(opts.driverFallbackModels ?? [])]\n  const agentMaxTokens = opts.agentMaxTokens ?? 2500\n  const toolFollowupMaxTokens = opts.toolFollowupMaxTokens ?? 2000\n  const driverMaxTokens = opts.driverMaxTokens ?? 600\n\n  const bundle =\n    opts.tools && opts.toolExecutors\n      ? {\n          tools: opts.tools,\n          executors: opts.toolExecutors,\n          artifactTypeFor: opts.artifactTypeFor ?? (() => undefined),\n        }\n      : defaultDelegationTools()\n  const tools = opts.tools ?? bundle.tools\n  const executors = opts.toolExecutors ?? bundle.executors\n  const artifactTypeFor = opts.artifactTypeFor ?? bundle.artifactTypeFor\n\n  const routerTransport: MultishotTransport = (req) => routerCompletion({ apiKey, baseUrl, ...req })\n  const agentTransport = opts.agentTransport ?? routerTransport\n  const driverTransport = opts.driverTransport ?? routerTransport\n\n  const start = Date.now()\n  const transcript: MultishotMessage[] = []\n  const artifacts: MultishotArtifact[] = []\n  let toolCalls = 0\n  let totalCostUsd = 0\n\n  const opener = opts.shape.buildOpener(opts.persona)\n  transcript.push({ role: 'user', content: opener })\n\n  const systemPrompt = opts.profile.prompt?.systemPrompt ?? ''\n  const agentMessages: Array<Record<string, unknown>> = [\n    { role: 'system', content: systemPrompt },\n    { role: 'user', content: opener },\n  ]\n\n  for (let turn = 0; turn < maxTurns; turn++) {\n    if (opts.signal?.aborted) throw new Error('multishot aborted')\n\n    let dispatchesThisTurn = 0\n    while (true) {\n      const {\n        message: agentMsg,\n        usage: agentUsage,\n        costUsd: agentCostUsd,\n      } = await agentTransport({\n        model: agentModel,\n        messages: agentMessages,\n        tools,\n        temperature: 0.7,\n        maxTokens: dispatchesThisTurn === 0 ? agentMaxTokens : toolFollowupMaxTokens,\n        signal: opts.signal,\n      })\n      totalCostUsd += agentCostUsd ?? estimateRouterCost(agentModel, agentUsage)\n\n      const agentText = (agentMsg.content ?? '').trim()\n      const agentToolCalls = (agentMsg.tool_calls ?? []).map((tc) => ({\n        id: tc.id,\n        name: tc.function.name,\n        args: (() => {\n          try {\n            return JSON.parse(tc.function.arguments) as Record<string, unknown>\n          } catch {\n            return {} as Record<string, unknown>\n          }\n        })(),\n      }))\n\n      agentMessages.push({\n        role: 'assistant',\n        content: agentText || null,\n        ...(agentMsg.tool_calls?.length ? { tool_calls: agentMsg.tool_calls } : {}),\n      })\n      transcript.push({\n        role: 'assistant',\n        content: agentText,\n        toolCalls: agentToolCalls.length > 0 ? agentToolCalls : undefined,\n      })\n\n      if (agentToolCalls.length === 0) break\n      dispatchesThisTurn += agentToolCalls.length\n      if (dispatchesThisTurn > maxToolDispatches) {\n        throw new Error(\n          `multishot: tool dispatch cap exceeded (${dispatchesThisTurn}/${maxToolDispatches}) on turn ${turn}`,\n        )\n      }\n\n      for (const tc of agentToolCalls) {\n        toolCalls++\n        let toolResult = ''\n        try {\n          const executor = executors[tc.name]\n          if (!executor) {\n            toolResult = JSON.stringify({ error: `unknown tool ${tc.name}` })\n          } else {\n            const r = await executor(tc.args, { apiKey, baseUrl, signal: opts.signal })\n            toolResult = r.content\n            totalCostUsd += r.costUsd\n            const artifactType = artifactTypeFor(tc.name)\n            if (artifactType) {\n              artifacts.push({\n                type: artifactType,\n                turn,\n                invocation: { name: tc.name, args: tc.args },\n                content: toolResult,\n              })\n            }\n          }\n        } catch (err) {\n          if (err instanceof MultishotFatalToolError) throw err\n          toolResult = JSON.stringify({ error: err instanceof Error ? err.message : String(err) })\n        }\n        agentMessages.push({ role: 'tool', tool_call_id: tc.id, content: toolResult || 'done' })\n        transcript.push({ role: 'tool', content: toolResult || 'done', toolCallId: tc.id })\n      }\n    }\n\n    if (turn < maxTurns - 1) {\n      const driver = await driverTurn({\n        transport: driverTransport,\n        persona: opts.persona,\n        shape: opts.shape,\n        transcript,\n        turn,\n        models: driverModels,\n        maxTokens: driverMaxTokens,\n        signal: opts.signal,\n      })\n      totalCostUsd += driver.costUsd\n      agentMessages.push({ role: 'user', content: driver.content })\n      transcript.push({ role: 'user', content: driver.content })\n    }\n  }\n\n  return { transcript, artifacts, toolCalls, durationMs: Date.now() - start, costUsd: totalCostUsd }\n}\n\nasync function driverTurn<TPersona extends MultishotPersona>(opts: {\n  transport: MultishotTransport\n  persona: TPersona\n  shape: MultishotShape<TPersona>\n  transcript: MultishotMessage[]\n  turn: number\n  models: string[]\n  maxTokens: number\n  signal?: AbortSignal\n}): Promise<{ content: string; costUsd: number }> {\n  const driverSystem = opts.shape.buildDriverSystemPrompt(opts.persona)\n\n  // Translate transcript to driver POV: agent's `assistant` messages become\n  // `user` (the agent talking TO the driver); the driver's prior `user`\n  // messages become `assistant` (the driver's prior responses).\n  const driverMessages: Array<Record<string, unknown>> = [{ role: 'system', content: driverSystem }]\n  for (const msg of opts.transcript) {\n    if (msg.role === 'tool') continue\n    const content = driverVisibleContent(msg)\n    if (!content) continue\n    if (msg.role === 'assistant') driverMessages.push({ role: 'user', content })\n    else if (msg.role === 'user') driverMessages.push({ role: 'assistant', content })\n  }\n\n  // Driver must never go silent. Retry once on empty content; then fail loud.\n  for (const model of opts.models) {\n    for (let attempt = 0; attempt < 2; attempt++) {\n      const { message, usage, costUsd } = await opts.transport({\n        model,\n        messages: driverMessages,\n        temperature: 0.9,\n        maxTokens: opts.maxTokens,\n        signal: opts.signal,\n      })\n      const content = (message.content ?? '').trim()\n      if (content.length > 0)\n        return { content, costUsd: costUsd ?? estimateRouterCost(model, usage) }\n    }\n  }\n  throw new MultishotDriverEmptyError(opts.turn)\n}\n\nfunction driverVisibleContent(msg: MultishotMessage): string | null {\n  const text = msg.content.trim()\n  if (text.length > 0) return text\n  if (msg.role !== 'assistant' || !msg.toolCalls?.length) return null\n\n  const toolNames = msg.toolCalls.map((call) => call.name.trim()).filter(Boolean)\n  if (toolNames.length === 0) return 'Agent called tools.'\n  return `Agent called ${toolNames.length === 1 ? 'tool' : 'tools'}: ${toolNames.join(', ')}.`\n}\n","// Multishot matrix wrapper — sweeps profiles × personas × reps, runs\n// the driver-agent loop per cell, applies up to three configured judges,\n// persists per-cell artifacts, and aggregates by axis.\n//\n// Uses runAgentMatrix from @tangle-network/agent-eval/matrix under the\n// hood so cell scheduling + concurrency + cost ceiling are unified with\n// other matrix consumers.\n\nimport { mkdirSync, writeFileSync } from 'node:fs'\nimport { join } from 'node:path'\nimport type { AgentProfile } from '@tangle-network/agent-interface'\nimport type { MatrixResult } from '../matrix'\nimport { runAgentMatrix } from '../matrix'\nimport { type JudgeConfig, type JudgeScore, runJudge } from './judges'\nimport { runMultishot } from './multishot'\nimport type {\n  MultishotArtifact,\n  MultishotMessage,\n  MultishotPersona,\n  MultishotShape,\n  MultishotToolDefinition,\n  MultishotToolExecutor,\n  MultishotTransport,\n} from './types'\n\nexport interface ConversationJudgeInput<TPersona extends MultishotPersona> {\n  transcript: MultishotMessage[]\n  persona: TPersona\n}\n\nexport interface ArtifactJudgeInput<TPersona extends MultishotPersona> {\n  artifact: MultishotArtifact\n  persona: TPersona\n}\n\nexport interface MultishotJudges<TPersona extends MultishotPersona> {\n  /** Scores the full transcript end-to-end (always runs). */\n  conversation: JudgeConfig<ConversationJudgeInput<TPersona>>\n  /** Scores each code-type artifact. Optional — omit when domain has no code artifacts. */\n  codeReview?: JudgeConfig<ArtifactJudgeInput<TPersona>>\n  /** Scores each non-code (research/content/template) artifact. Optional. */\n  contentQuality?: JudgeConfig<ArtifactJudgeInput<TPersona>>\n  /** Which artifact types route to codeReview. Defaults to ['code']. */\n  codeArtifactTypes?: string[]\n  /** Which artifact types route to contentQuality. Defaults to ['research']. */\n  contentArtifactTypes?: string[]\n}\n\nexport interface CellCompositeScore {\n  composite: number\n  conversation: JudgeScore\n  codeReview?: {\n    perArtifact: Array<JudgeScore & { turn: number; type: string }>\n    composite: number\n  }\n  contentQuality?: {\n    perArtifact: Array<JudgeScore & { turn: number; type: string }>\n    composite: number\n  }\n}\n\nexport interface RunMultishotMatrixOptions<TPersona extends MultishotPersona> {\n  /** AgentProfile axis (matrix primary). */\n  profiles: Array<{ id: string; value: AgentProfile }>\n  /** Persona axis. */\n  personas: TPersona[]\n  /** Persona-shaping callbacks. */\n  shape: MultishotShape<TPersona>\n  /** Judge configurations. */\n  judges: MultishotJudges<TPersona>\n  /** Tool definitions advertised to the agent. Defaults to delegate_research + delegate_code. */\n  tools?: MultishotToolDefinition[]\n  /** Map from tool name → inline executor. Must align with `tools`. */\n  toolExecutors?: Record<string, MultishotToolExecutor>\n  /** Tool name → artifact type label. Defaults to research/code mapping. */\n  artifactTypeFor?: (toolName: string) => string | undefined\n  /** Where per-cell artifacts land. Cells write to `<runDir>/<profileId>/<personaId>/rep-N/`. */\n  runDir: string\n  /** Replicates per (profile, persona) cell. */\n  reps?: number\n  /** Max conversation turns per cell. */\n  maxTurns?: number\n  /** Maximum tool calls the agent may dispatch inside one assistant turn. */\n  maxToolDispatches?: number\n  /** Max concurrent cells. */\n  maxConcurrency?: number\n  /** Total $ ceiling across the matrix; cells aborted past this. */\n  costCeiling?: number\n  /** Agent model. */\n  agentModel?: string\n  /** Driver model. */\n  driverModel?: string\n  /** Fallback driver models tried when the primary simulated-user model returns empty twice. */\n  driverFallbackModels?: string[]\n  /** Maximum output tokens for the first agent call in each assistant turn. */\n  agentMaxTokens?: number\n  /** Maximum output tokens for agent follow-up calls after tool results. */\n  toolFollowupMaxTokens?: number\n  /** Maximum output tokens for each simulated-user driver response. */\n  driverMaxTokens?: number\n  /** Maximum output tokens for each judge response. */\n  judgeMaxTokens?: number\n  /** Execution seam for the agent leg of every cell — replaces the router\n   *  HTTP call when provided (see RunMultishotOptions.agentTransport).\n   *  Judges are unaffected; configure those via MultishotJudges. */\n  agentTransport?: MultishotTransport\n  /** Execution seam for the simulated-user driver leg of every cell. */\n  driverTransport?: MultishotTransport\n  /** Pass-thru fields. */\n  apiKey?: string\n  baseUrl?: string\n}\n\ninterface CellOutput {\n  turns: number\n  toolCalls: number\n  artifactCount: number\n}\n\n/** Mean composite over non-failed scores. `0` when the list is empty (a\n *  configured judge with nothing to score contributes 0, matching the cell\n *  composite's long-standing semantics); `null` when scores exist but every\n *  one failed — no signal, so the slot must be EXCLUDED from the cell mean\n *  rather than dragging it to zero. */\nfunction meanCompositeExcludingFailed(scores: ReadonlyArray<JudgeScore>): number | null {\n  if (scores.length === 0) return 0\n  const live = scores.filter((s) => !s.failed)\n  if (live.length === 0) return null\n  return live.reduce((sum, s) => sum + s.composite, 0) / live.length\n}\n\nexport interface CellCompositeInput {\n  conversation: JudgeScore\n  /** Present iff the codeReview judge is configured. */\n  codeReviews?: ReadonlyArray<JudgeScore>\n  /** Present iff the contentQuality judge is configured. */\n  contentReviews?: ReadonlyArray<JudgeScore>\n}\n\n/** Cell composite = mean over configured judge slots, excluding failed\n *  scores: a failed conversation judge or an all-failed artifact slot carries\n *  no signal and is dropped from the mean. `composite` is 0 only when EVERY\n *  configured slot failed (`allJudgesFailed` distinguishes that from a real\n *  zero). Pure — exported for deterministic testing. */\nexport function computeCellComposite(input: CellCompositeInput): {\n  composite: number\n  codeComposite: number\n  contentComposite: number\n  allJudgesFailed: boolean\n} {\n  const contributions: number[] = []\n  if (!input.conversation.failed) contributions.push(input.conversation.composite)\n\n  const codeMean = input.codeReviews ? meanCompositeExcludingFailed(input.codeReviews) : undefined\n  if (typeof codeMean === 'number') contributions.push(codeMean)\n  const contentMean = input.contentReviews\n    ? meanCompositeExcludingFailed(input.contentReviews)\n    : undefined\n  if (typeof contentMean === 'number') contributions.push(contentMean)\n\n  return {\n    composite:\n      contributions.length === 0\n        ? 0\n        : contributions.reduce((s, v) => s + v, 0) / contributions.length,\n    codeComposite: codeMean ?? 0,\n    contentComposite: contentMean ?? 0,\n    allJudgesFailed: contributions.length === 0,\n  }\n}\n\nexport interface RunMultishotMatrixResult {\n  matrix: MatrixResult<CellOutput>\n}\n\nexport async function runMultishotMatrix<TPersona extends MultishotPersona>(\n  opts: RunMultishotMatrixOptions<TPersona>,\n): Promise<RunMultishotMatrixResult> {\n  const codeTypes = new Set(opts.judges.codeArtifactTypes ?? ['code'])\n  const contentTypes = new Set(opts.judges.contentArtifactTypes ?? ['research'])\n  mkdirSync(opts.runDir, { recursive: true })\n\n  const matrix = await runAgentMatrix<CellOutput>({\n    axes: [\n      { name: 'profile', values: opts.profiles },\n      { name: 'persona', values: opts.personas.map((p) => ({ id: p.id, value: p })) },\n    ],\n    reps: opts.reps ?? 1,\n    maxConcurrency: opts.maxConcurrency ?? 2,\n    costCeiling: opts.costCeiling,\n    async runCell(cell) {\n      const profile = cell.axes.profile?.value as AgentProfile\n      const persona = cell.axes.persona?.value as TPersona\n      const profileId = String(cell.axes.profile?.id ?? 'unknown')\n      const personaId = String(cell.axes.persona?.id ?? 'unknown')\n\n      const sim = await runMultishot({\n        profile,\n        persona,\n        shape: opts.shape,\n        tools: opts.tools,\n        toolExecutors: opts.toolExecutors,\n        artifactTypeFor: opts.artifactTypeFor,\n        maxTurns: opts.maxTurns,\n        maxToolDispatches: opts.maxToolDispatches,\n        agentModel: opts.agentModel,\n        driverModel: opts.driverModel,\n        driverFallbackModels: opts.driverFallbackModels,\n        agentMaxTokens: opts.agentMaxTokens,\n        toolFollowupMaxTokens: opts.toolFollowupMaxTokens,\n        driverMaxTokens: opts.driverMaxTokens,\n        agentTransport: opts.agentTransport,\n        driverTransport: opts.driverTransport,\n        apiKey: opts.apiKey,\n        baseUrl: opts.baseUrl,\n      })\n\n      const codeArtifacts = sim.artifacts.filter((a) => codeTypes.has(a.type))\n      const contentArtifacts = sim.artifacts.filter((a) => contentTypes.has(a.type))\n\n      const [conversation, codeReviews, contentReviews] = await Promise.all([\n        runJudge(withJudgeMaxTokens(opts.judges.conversation, opts.judgeMaxTokens), {\n          transcript: sim.transcript,\n          persona,\n        }),\n        opts.judges.codeReview\n          ? Promise.all(\n              codeArtifacts.map((artifact) =>\n                runJudge(withJudgeMaxTokens(opts.judges.codeReview!, opts.judgeMaxTokens), {\n                  artifact,\n                  persona,\n                }).then((s) => ({\n                  ...s,\n                  turn: artifact.turn,\n                  type: artifact.type,\n                })),\n              ),\n            )\n          : Promise.resolve([] as Array<JudgeScore & { turn: number; type: string }>),\n        opts.judges.contentQuality\n          ? Promise.all(\n              contentArtifacts.map((artifact) =>\n                runJudge(withJudgeMaxTokens(opts.judges.contentQuality!, opts.judgeMaxTokens), {\n                  artifact,\n                  persona,\n                }).then((s) => ({\n                  ...s,\n                  turn: artifact.turn,\n                  type: artifact.type,\n                })),\n              ),\n            )\n          : Promise.resolve([] as Array<JudgeScore & { turn: number; type: string }>),\n      ])\n\n      const { composite, codeComposite, contentComposite, allJudgesFailed } = computeCellComposite({\n        conversation,\n        codeReviews: opts.judges.codeReview ? codeReviews : undefined,\n        contentReviews: opts.judges.contentQuality ? contentReviews : undefined,\n      })\n\n      const cellScore: CellCompositeScore = { composite, conversation }\n      if (opts.judges.codeReview)\n        cellScore.codeReview = { perArtifact: codeReviews, composite: codeComposite }\n      if (opts.judges.contentQuality)\n        cellScore.contentQuality = { perArtifact: contentReviews, composite: contentComposite }\n\n      const cellDir = join(opts.runDir, profileId, personaId, `rep-${cell.rep}`)\n      mkdirSync(cellDir, { recursive: true })\n      writeFileSync(join(cellDir, 'transcript.json'), JSON.stringify(sim.transcript, null, 2))\n      writeFileSync(join(cellDir, 'artifacts.json'), JSON.stringify(sim.artifacts, null, 2))\n      writeFileSync(join(cellDir, 'scores.json'), JSON.stringify(cellScore, null, 2))\n\n      const notes = [`convo=${conversation.composite.toFixed(1)}`]\n      if (opts.judges.codeReview) notes.push(`code=${codeComposite.toFixed(1)}`)\n      if (opts.judges.contentQuality) notes.push(`content=${contentComposite.toFixed(1)}`)\n      if (allJudgesFailed) notes.push('all-judges-failed')\n\n      return {\n        output: {\n          turns: sim.transcript.length,\n          toolCalls: sim.toolCalls,\n          artifactCount: sim.artifacts.length,\n        },\n        verdict: { valid: composite >= 5, score: composite, notes: notes.join(' ') },\n        costUsd: sim.costUsd,\n        durationMs: sim.durationMs,\n      }\n    },\n  })\n\n  // Persist top-level summary.\n  const summary = {\n    cells: matrix.summary.totalCells,\n    passRate: matrix.summary.overallPassRate,\n    meanScore: matrix.summary.overallMeanScore,\n    totalCostUsd: matrix.summary.totalCostUsd,\n    durationMs: matrix.summary.durationMs,\n    runsExecuted: matrix.summary.runsExecuted,\n    cellsSkipped: matrix.summary.cellsSkipped,\n    byProfile: matrix.byAxis.profile,\n    byPersona: matrix.byAxis.persona,\n  }\n  writeFileSync(join(opts.runDir, 'summary.json'), JSON.stringify(summary, null, 2))\n\n  const md: string[] = [\n    `# Multishot matrix`,\n    ``,\n    `**Cells**: ${matrix.summary.totalCells} | **Pass rate**: ${(matrix.summary.overallPassRate * 100).toFixed(0)}% | **Mean**: ${matrix.summary.overallMeanScore.toFixed(2)} | **Cost**: $${matrix.summary.totalCostUsd.toFixed(2)} | **Duration**: ${(matrix.summary.durationMs / 1000).toFixed(0)}s`,\n    ``,\n    `## By profile`,\n    ``,\n    '| profile | pass | mean | cost |',\n    '|---|---|---|---|',\n    ...Object.entries(matrix.byAxis.profile ?? {}).map(\n      ([id, s]) =>\n        `| ${id} | ${(s.passRate * 100).toFixed(0)}% | ${s.meanScore.toFixed(2)} | $${s.totalCostUsd.toFixed(2)} |`,\n    ),\n    ``,\n    `## By persona`,\n    ``,\n    '| persona | pass | mean | cost |',\n    '|---|---|---|---|',\n    ...Object.entries(matrix.byAxis.persona ?? {}).map(\n      ([id, s]) =>\n        `| ${id} | ${(s.passRate * 100).toFixed(0)}% | ${s.meanScore.toFixed(2)} | $${s.totalCostUsd.toFixed(2)} |`,\n    ),\n    ``,\n  ]\n  writeFileSync(join(opts.runDir, 'summary.md'), md.join('\\n'))\n\n  return { matrix }\n}\n\nfunction withJudgeMaxTokens<TInput>(\n  judge: JudgeConfig<TInput>,\n  maxTokens: number | undefined,\n): JudgeConfig<TInput> {\n  if (maxTokens === undefined || judge.maxTokens !== undefined) return judge\n  return { ...judge, maxTokens }\n}\n"],"mappings":";;;;AA4BA,eAAsB,iBACpB,KACmC;CACnC,MAAM,OAAgC;EACpC,OAAO,IAAI;EACX,UAAU,IAAI;EACd,aAAa,IAAI,eAAe;EAChC,YAAY,IAAI,aAAa;CAC/B;CACA,IAAI,IAAI,OAAO,QAAQ,KAAK,QAAQ,IAAI;CACxC,MAAM,MAAM,GAAG,IAAI,QAAQ,QAAQ,QAAQ,EAAE,EAAE;CAC/C,MAAM,MAAM,MAAM,MAAM,KAAK;EAC3B,QAAQ;EACR,SAAS;GAAE,eAAe,UAAU,IAAI;GAAU,gBAAgB;EAAmB;EACrF,MAAM,KAAK,UAAU,IAAI;EACzB,QAAQ,IAAI;CACd,CAAC;CACD,IAAI,CAAC,IAAI,IAAI;EACX,MAAM,OAAO,MAAM,IAAI,KAAK;EAC5B,MAAM,IAAI,MAAM,UAAU,IAAI,OAAO,IAAI,KAAK,MAAM,GAAG,GAAG,GAAG;CAC/D;CACA,MAAM,OAAQ,MAAM,IAAI,KAAK;CAI7B,MAAM,SAAS,KAAK,QAAQ;CAC5B,IAAI,CAAC,QAAQ,MAAM,IAAI,MAAM,+BAA+B,KAAK,UAAU,IAAI,CAAC,CAAC,MAAM,GAAG,GAAG,GAAG;CAChG,OAAO;EAAE,SAAS,OAAO;EAAS,OAAO,KAAK;CAAM;AACtD;AAIA,SAAgB,mBACd,OACA,OACQ;CACR,IAAI,CAAC,OAAO,OAAO;CACnB,MAAM,WAAW,MAAM,iBAAiB;CACxC,MAAM,YAAY,MAAM,qBAAqB;CAC7C,IAAI,UAAU;CACd,IAAI,WAAW;CACf,IAAI,MAAM,SAAS,aAAa,GAAG;EACjC,UAAU;EACV,WAAW;CACb,OAAO,IAAI,MAAM,SAAS,SAAS,KAAK,MAAM,SAAS,eAAe,GAAG;EACvE,UAAU;EACV,WAAW;CACb,OAAO,IAAI,MAAM,SAAS,MAAM,KAAK,MAAM,SAAS,KAAK,KAAK,MAAM,SAAS,UAAU,GAAG;EACxF,UAAU;EACV,WAAW;CACb;CACA,QAAQ,WAAW,UAAU,YAAY,YAAY;AACvD;AAEA,SAAgB,uBAA+B;CAC7C,QAAQ,QAAQ,IAAI,0BAA0B,iCAAA,CAAkC,QAC9E,QACA,EACF;AACF;AAEA,SAAgB,sBAA8B;CAC5C,MAAM,MAAM,QAAQ,IAAI;CACxB,IAAI,CAAC,KAAK,MAAM,IAAI,MAAM,gEAAgE;CAC1F,OAAO;AACT;;;ACnFA,MAAa,2BAA2B;AACxC,MAAa,sBAAsB;AAgBnC,MAAM,4BACJ;AAEF,MAAM,uBACJ;AAEF,MAAa,iCAA0D;CACrE,MAAM;CACN,UAAU;EACR,MAAM;EACN,aACE;EACF,YAAY;GACV,MAAM;GACN,YAAY;IACV,UAAU;KAAE,MAAM;KAAU,aAAa;IAAgC;IACzE,OAAO;KACL,MAAM;KACN,aAAa;IACf;GACF;GACA,UAAU,CAAC,UAAU;EACvB;CACF;AACF;AAEA,MAAa,6BAAsD;CACjE,MAAM;CACN,UAAU;EACR,MAAM;EACN,aACE;EACF,YAAY;GACV,MAAM;GACN,YAAY;IACV,MAAM;KAAE,MAAM;KAAU,aAAa;IAAgC;IACrE,UAAU;KACR,MAAM;KACN,aAAa;IACf;GACF;GACA,UAAU,CAAC,MAAM;EACnB;CACF;AACF;AAEA,SAAgB,uBACd,SAAkC,CAAC,GACZ;CACvB,MAAM,eAAe,OAAO,gBAAgB;CAC5C,MAAM,QAAQ,OAAO,SAAA;CACrB,OAAO,OAAO,MAAM,QAAQ;EAC1B,MAAM,WAAW,OAAO,KAAK,YAAY,EAAE;EAC3C,MAAM,QAAQ,KAAK,QAAQ,OAAO,KAAK,KAAK,IAAI,KAAA;EAChD,MAAM,EAAE,SAAS,UAAU,MAAM,iBAAiB;GAChD,QAAQ,IAAI;GACZ,SAAS,IAAI;GACb;GACA,aAAa;GACb,WAAW;GACX,UAAU,CACR;IAAE,MAAM;IAAU,SAAS;GAAa,GACxC;IAAE,MAAM;IAAQ,SAAS,aAAa,WAAW,QAAQ,YAAY,UAAU;GAAK,CACtF;GACA,QAAQ,IAAI;EACd,CAAC;EACD,OAAO;GAAE,SAAS,QAAQ,WAAW;GAAI,SAAS,mBAAmB,OAAO,KAAK;EAAE;CACrF;AACF;AAEA,SAAgB,mBAAmB,SAA6B,CAAC,GAA0B;CACzF,MAAM,eAAe,OAAO,gBAAgB;CAC5C,MAAM,QAAQ,OAAO,SAAA;CACrB,OAAO,OAAO,MAAM,QAAQ;EAC1B,MAAM,OAAO,OAAO,KAAK,QAAQ,EAAE;EACnC,MAAM,WAAW,KAAK,WAAW,OAAO,KAAK,QAAQ,IAAI;EACzD,MAAM,EAAE,SAAS,UAAU,MAAM,iBAAiB;GAChD,QAAQ,IAAI;GACZ,SAAS,IAAI;GACb;GACA,aAAa;GACb,WAAW;GACX,UAAU,CACR;IAAE,MAAM;IAAU,SAAS,GAAG,aAAa,gBAAgB;GAAW,GACtE;IAAE,MAAM;IAAQ,SAAS,YAAY;GAAO,CAC9C;GACA,QAAQ,IAAI;EACd,CAAC;EACD,OAAO;GAAE,SAAS,QAAQ,WAAW;GAAI,SAAS,mBAAmB,OAAO,KAAK;EAAE;CACrF;AACF;AAgBA,SAAgB,uBAAuB,SAA6B,CAAC,GAAuB;CAC1F,OAAO;EACL,OAAO,CAAC,gCAAgC,0BAA0B;EAClE,WAAW;GACT,mBAAmB,uBAAuB,OAAO,QAAQ;GACzD,eAAe,mBAAmB,OAAO,IAAI;EAC/C;EACA,kBAAkB,SAChB,SAAS,sBAAsB,aAAa,SAAS,kBAAkB,SAAS,KAAA;CACpF;AACF;;;AC/HA,MAAa,sBAAsB;AA4BnC,eAAsB,SACpB,OACA,OACqB;CACrB,MAAM,SAAS,MAAM,UAAU,oBAAoB;CACnD,MAAM,UAAU,MAAM,WAAW,qBAAqB;CACtD,MAAM,QAAQ,MAAM,SAAS,QAAQ,IAAI,eAAA;CACzC,MAAM,SAAS,MAAM,YAAY,KAAK;CACtC,IAAI,MAAM;CACV,IAAI;EACF,MAAM,EAAE,YAAY,MAAM,iBAAiB;GACzC;GACA;GACA;GACA,aAAa;GACb,WAAW,MAAM,aAAa;GAC9B,UAAU,CACR;IAAE,MAAM;IAAU,SAAS,MAAM;GAAa,GAC9C;IAAE,MAAM;IAAQ,SAAS;GAAO,CAClC;EACF,CAAC;EACD,OAAO,QAAQ,WAAW,GAAA,CAAI,KAAK;CACrC,SAAS,KAAK;EAGZ,OAAO;GACL,YAAY,CAAC;GACb,WAAW;GACX,QAAQ;GACR,OAAO,SAAS,MAAM,KAAK,gBAAgB,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;EAC5F;CACF;CAEA,IAAI,SAAyC;CAC7C,IAAI;EACF,MAAM,UAAU,IACb,QAAQ,gBAAgB,EAAE,CAAC,CAC3B,QAAQ,WAAW,EAAE,CAAC,CACtB,KAAK;EACR,SAAS,KAAK,MAAM,OAAO;CAC7B,QAAQ;EACN,OAAO;GACL,YAAY,CAAC;GACb,WAAW;GACX,QAAQ;GACR,OAAO,SAAS,MAAM,KAAK,sBAAsB,IAAI,MAAM,GAAG,GAAG;EACnE;CACF;CAEA,MAAM,aAAqC,CAAC;CAC5C,IAAI,MAAM;CACV,KAAK,MAAM,OAAO,MAAM,YAAY;EAClC,MAAM,IAAI,OAAO,OAAO,IAAI,QAAQ,CAAC;EACrC,MAAM,UAAU,OAAO,SAAS,CAAC,IAAI,KAAK,IAAI,GAAG,KAAK,IAAI,IAAI,CAAC,CAAC,IAAI;EACpE,WAAW,IAAI,OAAO;EACtB,OAAO;CACT;CACA,OAAO;EACL;EACA,WAAW,MAAM,WAAW,WAAW,IAAI,IAAI,MAAM,MAAM,WAAW;EACtE,OAAO,OAAO,OAAO,UAAU,WAAW,OAAO,QAAQ;CAC3D;AACF;;;AAIA,SAAgB,iBAAiB,MAAyC;CACxE,OAAO,KAAK,KAAK,MAAM,KAAK,EAAE,IAAI,IAAI,EAAE,aAAa,CAAC,CAAC,KAAK,IAAI;AAClE;;AAGA,SAAgB,iBAAiB,MAAyC;CAExE,OAAO,6DADQ,KAAK,KAAK,MAAM,IAAI,EAAE,IAAI,IAAI,CAAC,CAAC,KAAK,GACqB,EAAE;AAC7E;;;AC7BA,IAAa,4BAAb,cAA+C,MAAM;CACvB;CAA5B,YAAY,MAA8B;EACxC,MAAM,0DAA0D,KAAK,gBAAgB;EAD3D,KAAA,OAAA;EAE1B,KAAK,OAAO;CACd;AACF;AAEA,IAAa,0BAAb,cAA6C,MAAM;CACjD,YAAY,SAAiB;EAC3B,MAAM,OAAO;EACb,KAAK,OAAO;CACd;AACF;;;AChCA,eAAsB,aACpB,MAC0B;CAC1B,MAAM,SAAS,KAAK,UAAU,oBAAoB;CAClD,MAAM,UAAU,KAAK,WAAW,qBAAqB;CACrD,MAAM,WAAW,KAAK,YAAY;CAClC,MAAM,oBAAoB,KAAK,qBAAqB;CACpD,MAAM,aAAa,KAAK,cAAc;CAEtC,MAAM,eAAe,CADD,KAAK,eAAe,sBACL,GAAI,KAAK,wBAAwB,CAAC,CAAE;CACvE,MAAM,iBAAiB,KAAK,kBAAkB;CAC9C,MAAM,wBAAwB,KAAK,yBAAyB;CAC5D,MAAM,kBAAkB,KAAK,mBAAmB;CAEhD,MAAM,SACJ,KAAK,SAAS,KAAK,gBACf;EACE,OAAO,KAAK;EACZ,WAAW,KAAK;EAChB,iBAAiB,KAAK,0BAA0B,KAAA;CAClD,IACA,uBAAuB;CAC7B,MAAM,QAAQ,KAAK,SAAS,OAAO;CACnC,MAAM,YAAY,KAAK,iBAAiB,OAAO;CAC/C,MAAM,kBAAkB,KAAK,mBAAmB,OAAO;CAEvD,MAAM,mBAAuC,QAAQ,iBAAiB;EAAE;EAAQ;EAAS,GAAG;CAAI,CAAC;CACjG,MAAM,iBAAiB,KAAK,kBAAkB;CAC9C,MAAM,kBAAkB,KAAK,mBAAmB;CAEhD,MAAM,QAAQ,KAAK,IAAI;CACvB,MAAM,aAAiC,CAAC;CACxC,MAAM,YAAiC,CAAC;CACxC,IAAI,YAAY;CAChB,IAAI,eAAe;CAEnB,MAAM,SAAS,KAAK,MAAM,YAAY,KAAK,OAAO;CAClD,WAAW,KAAK;EAAE,MAAM;EAAQ,SAAS;CAAO,CAAC;CAGjD,MAAM,gBAAgD,CACpD;EAAE,MAAM;EAAU,SAFC,KAAK,QAAQ,QAAQ,gBAAgB;CAEhB,GACxC;EAAE,MAAM;EAAQ,SAAS;CAAO,CAClC;CAEA,KAAK,IAAI,OAAO,GAAG,OAAO,UAAU,QAAQ;EAC1C,IAAI,KAAK,QAAQ,SAAS,MAAM,IAAI,MAAM,mBAAmB;EAE7D,IAAI,qBAAqB;EACzB,OAAO,MAAM;GACX,MAAM,EACJ,SAAS,UACT,OAAO,YACP,SAAS,iBACP,MAAM,eAAe;IACvB,OAAO;IACP,UAAU;IACV;IACA,aAAa;IACb,WAAW,uBAAuB,IAAI,iBAAiB;IACvD,QAAQ,KAAK;GACf,CAAC;GACD,gBAAgB,gBAAgB,mBAAmB,YAAY,UAAU;GAEzE,MAAM,aAAa,SAAS,WAAW,GAAA,CAAI,KAAK;GAChD,MAAM,kBAAkB,SAAS,cAAc,CAAC,EAAA,CAAG,KAAK,QAAQ;IAC9D,IAAI,GAAG;IACP,MAAM,GAAG,SAAS;IAClB,aAAa;KACX,IAAI;MACF,OAAO,KAAK,MAAM,GAAG,SAAS,SAAS;KACzC,QAAQ;MACN,OAAO,CAAC;KACV;IACF,EAAA,CAAG;GACL,EAAE;GAEF,cAAc,KAAK;IACjB,MAAM;IACN,SAAS,aAAa;IACtB,GAAI,SAAS,YAAY,SAAS,EAAE,YAAY,SAAS,WAAW,IAAI,CAAC;GAC3E,CAAC;GACD,WAAW,KAAK;IACd,MAAM;IACN,SAAS;IACT,WAAW,eAAe,SAAS,IAAI,iBAAiB,KAAA;GAC1D,CAAC;GAED,IAAI,eAAe,WAAW,GAAG;GACjC,sBAAsB,eAAe;GACrC,IAAI,qBAAqB,mBACvB,MAAM,IAAI,MACR,0CAA0C,mBAAmB,GAAG,kBAAkB,YAAY,MAChG;GAGF,KAAK,MAAM,MAAM,gBAAgB;IAC/B;IACA,IAAI,aAAa;IACjB,IAAI;KACF,MAAM,WAAW,UAAU,GAAG;KAC9B,IAAI,CAAC,UACH,aAAa,KAAK,UAAU,EAAE,OAAO,gBAAgB,GAAG,OAAO,CAAC;UAC3D;MACL,MAAM,IAAI,MAAM,SAAS,GAAG,MAAM;OAAE;OAAQ;OAAS,QAAQ,KAAK;MAAO,CAAC;MAC1E,aAAa,EAAE;MACf,gBAAgB,EAAE;MAClB,MAAM,eAAe,gBAAgB,GAAG,IAAI;MAC5C,IAAI,cACF,UAAU,KAAK;OACb,MAAM;OACN;OACA,YAAY;QAAE,MAAM,GAAG;QAAM,MAAM,GAAG;OAAK;OAC3C,SAAS;MACX,CAAC;KAEL;IACF,SAAS,KAAK;KACZ,IAAI,eAAe,yBAAyB,MAAM;KAClD,aAAa,KAAK,UAAU,EAAE,OAAO,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG,EAAE,CAAC;IACzF;IACA,cAAc,KAAK;KAAE,MAAM;KAAQ,cAAc,GAAG;KAAI,SAAS,cAAc;IAAO,CAAC;IACvF,WAAW,KAAK;KAAE,MAAM;KAAQ,SAAS,cAAc;KAAQ,YAAY,GAAG;IAAG,CAAC;GACpF;EACF;EAEA,IAAI,OAAO,WAAW,GAAG;GACvB,MAAM,SAAS,MAAM,WAAW;IAC9B,WAAW;IACX,SAAS,KAAK;IACd,OAAO,KAAK;IACZ;IACA;IACA,QAAQ;IACR,WAAW;IACX,QAAQ,KAAK;GACf,CAAC;GACD,gBAAgB,OAAO;GACvB,cAAc,KAAK;IAAE,MAAM;IAAQ,SAAS,OAAO;GAAQ,CAAC;GAC5D,WAAW,KAAK;IAAE,MAAM;IAAQ,SAAS,OAAO;GAAQ,CAAC;EAC3D;CACF;CAEA,OAAO;EAAE;EAAY;EAAW;EAAW,YAAY,KAAK,IAAI,IAAI;EAAO,SAAS;CAAa;AACnG;AAEA,eAAe,WAA8C,MASX;CAMhD,MAAM,iBAAiD,CAAC;EAAE,MAAM;EAAU,SALrD,KAAK,MAAM,wBAAwB,KAAK,OAKiC;CAAE,CAAC;CACjG,KAAK,MAAM,OAAO,KAAK,YAAY;EACjC,IAAI,IAAI,SAAS,QAAQ;EACzB,MAAM,UAAU,qBAAqB,GAAG;EACxC,IAAI,CAAC,SAAS;EACd,IAAI,IAAI,SAAS,aAAa,eAAe,KAAK;GAAE,MAAM;GAAQ;EAAQ,CAAC;OACtE,IAAI,IAAI,SAAS,QAAQ,eAAe,KAAK;GAAE,MAAM;GAAa;EAAQ,CAAC;CAClF;CAGA,KAAK,MAAM,SAAS,KAAK,QACvB,KAAK,IAAI,UAAU,GAAG,UAAU,GAAG,WAAW;EAC5C,MAAM,EAAE,SAAS,OAAO,YAAY,MAAM,KAAK,UAAU;GACvD;GACA,UAAU;GACV,aAAa;GACb,WAAW,KAAK;GAChB,QAAQ,KAAK;EACf,CAAC;EACD,MAAM,WAAW,QAAQ,WAAW,GAAA,CAAI,KAAK;EAC7C,IAAI,QAAQ,SAAS,GACnB,OAAO;GAAE;GAAS,SAAS,WAAW,mBAAmB,OAAO,KAAK;EAAE;CAC3E;CAEF,MAAM,IAAI,0BAA0B,KAAK,IAAI;AAC/C;AAEA,SAAS,qBAAqB,KAAsC;CAClE,MAAM,OAAO,IAAI,QAAQ,KAAK;CAC9B,IAAI,KAAK,SAAS,GAAG,OAAO;CAC5B,IAAI,IAAI,SAAS,eAAe,CAAC,IAAI,WAAW,QAAQ,OAAO;CAE/D,MAAM,YAAY,IAAI,UAAU,KAAK,SAAS,KAAK,KAAK,KAAK,CAAC,CAAC,CAAC,OAAO,OAAO;CAC9E,IAAI,UAAU,WAAW,GAAG,OAAO;CACnC,OAAO,gBAAgB,UAAU,WAAW,IAAI,SAAS,QAAQ,IAAI,UAAU,KAAK,IAAI,EAAE;AAC5F;;;;;;;;AC7IA,SAAS,6BAA6B,QAAkD;CACtF,IAAI,OAAO,WAAW,GAAG,OAAO;CAChC,MAAM,OAAO,OAAO,QAAQ,MAAM,CAAC,EAAE,MAAM;CAC3C,IAAI,KAAK,WAAW,GAAG,OAAO;CAC9B,OAAO,KAAK,QAAQ,KAAK,MAAM,MAAM,EAAE,WAAW,CAAC,IAAI,KAAK;AAC9D;;;;;;AAeA,SAAgB,qBAAqB,OAKnC;CACA,MAAM,gBAA0B,CAAC;CACjC,IAAI,CAAC,MAAM,aAAa,QAAQ,cAAc,KAAK,MAAM,aAAa,SAAS;CAE/E,MAAM,WAAW,MAAM,cAAc,6BAA6B,MAAM,WAAW,IAAI,KAAA;CACvF,IAAI,OAAO,aAAa,UAAU,cAAc,KAAK,QAAQ;CAC7D,MAAM,cAAc,MAAM,iBACtB,6BAA6B,MAAM,cAAc,IACjD,KAAA;CACJ,IAAI,OAAO,gBAAgB,UAAU,cAAc,KAAK,WAAW;CAEnE,OAAO;EACL,WACE,cAAc,WAAW,IACrB,IACA,cAAc,QAAQ,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,cAAc;EAC/D,eAAe,YAAY;EAC3B,kBAAkB,eAAe;EACjC,iBAAiB,cAAc,WAAW;CAC5C;AACF;AAMA,eAAsB,mBACpB,MACmC;CACnC,MAAM,YAAY,IAAI,IAAI,KAAK,OAAO,qBAAqB,CAAC,MAAM,CAAC;CACnE,MAAM,eAAe,IAAI,IAAI,KAAK,OAAO,wBAAwB,CAAC,UAAU,CAAC;CAC7E,UAAU,KAAK,QAAQ,EAAE,WAAW,KAAK,CAAC;CAE1C,MAAM,SAAS,MAAM,eAA2B;EAC9C,MAAM,CACJ;GAAE,MAAM;GAAW,QAAQ,KAAK;EAAS,GACzC;GAAE,MAAM;GAAW,QAAQ,KAAK,SAAS,KAAK,OAAO;IAAE,IAAI,EAAE;IAAI,OAAO;GAAE,EAAE;EAAE,CAChF;EACA,MAAM,KAAK,QAAQ;EACnB,gBAAgB,KAAK,kBAAkB;EACvC,aAAa,KAAK;EAClB,MAAM,QAAQ,MAAM;GAClB,MAAM,UAAU,KAAK,KAAK,SAAS;GACnC,MAAM,UAAU,KAAK,KAAK,SAAS;GACnC,MAAM,YAAY,OAAO,KAAK,KAAK,SAAS,MAAM,SAAS;GAC3D,MAAM,YAAY,OAAO,KAAK,KAAK,SAAS,MAAM,SAAS;GAE3D,MAAM,MAAM,MAAM,aAAa;IAC7B;IACA;IACA,OAAO,KAAK;IACZ,OAAO,KAAK;IACZ,eAAe,KAAK;IACpB,iBAAiB,KAAK;IACtB,UAAU,KAAK;IACf,mBAAmB,KAAK;IACxB,YAAY,KAAK;IACjB,aAAa,KAAK;IAClB,sBAAsB,KAAK;IAC3B,gBAAgB,KAAK;IACrB,uBAAuB,KAAK;IAC5B,iBAAiB,KAAK;IACtB,gBAAgB,KAAK;IACrB,iBAAiB,KAAK;IACtB,QAAQ,KAAK;IACb,SAAS,KAAK;GAChB,CAAC;GAED,MAAM,gBAAgB,IAAI,UAAU,QAAQ,MAAM,UAAU,IAAI,EAAE,IAAI,CAAC;GACvE,MAAM,mBAAmB,IAAI,UAAU,QAAQ,MAAM,aAAa,IAAI,EAAE,IAAI,CAAC;GAE7E,MAAM,CAAC,cAAc,aAAa,kBAAkB,MAAM,QAAQ,IAAI;IACpE,SAAS,mBAAmB,KAAK,OAAO,cAAc,KAAK,cAAc,GAAG;KAC1E,YAAY,IAAI;KAChB;IACF,CAAC;IACD,KAAK,OAAO,aACR,QAAQ,IACN,cAAc,KAAK,aACjB,SAAS,mBAAmB,KAAK,OAAO,YAAa,KAAK,cAAc,GAAG;KACzE;KACA;IACF,CAAC,CAAC,CAAC,MAAM,OAAO;KACd,GAAG;KACH,MAAM,SAAS;KACf,MAAM,SAAS;IACjB,EAAE,CACJ,CACF,IACA,QAAQ,QAAQ,CAAC,CAAuD;IAC5E,KAAK,OAAO,iBACR,QAAQ,IACN,iBAAiB,KAAK,aACpB,SAAS,mBAAmB,KAAK,OAAO,gBAAiB,KAAK,cAAc,GAAG;KAC7E;KACA;IACF,CAAC,CAAC,CAAC,MAAM,OAAO;KACd,GAAG;KACH,MAAM,SAAS;KACf,MAAM,SAAS;IACjB,EAAE,CACJ,CACF,IACA,QAAQ,QAAQ,CAAC,CAAuD;GAC9E,CAAC;GAED,MAAM,EAAE,WAAW,eAAe,kBAAkB,oBAAoB,qBAAqB;IAC3F;IACA,aAAa,KAAK,OAAO,aAAa,cAAc,KAAA;IACpD,gBAAgB,KAAK,OAAO,iBAAiB,iBAAiB,KAAA;GAChE,CAAC;GAED,MAAM,YAAgC;IAAE;IAAW;GAAa;GAChE,IAAI,KAAK,OAAO,YACd,UAAU,aAAa;IAAE,aAAa;IAAa,WAAW;GAAc;GAC9E,IAAI,KAAK,OAAO,gBACd,UAAU,iBAAiB;IAAE,aAAa;IAAgB,WAAW;GAAiB;GAExF,MAAM,UAAU,KAAK,KAAK,QAAQ,WAAW,WAAW,OAAO,KAAK,KAAK;GACzE,UAAU,SAAS,EAAE,WAAW,KAAK,CAAC;GACtC,cAAc,KAAK,SAAS,iBAAiB,GAAG,KAAK,UAAU,IAAI,YAAY,MAAM,CAAC,CAAC;GACvF,cAAc,KAAK,SAAS,gBAAgB,GAAG,KAAK,UAAU,IAAI,WAAW,MAAM,CAAC,CAAC;GACrF,cAAc,KAAK,SAAS,aAAa,GAAG,KAAK,UAAU,WAAW,MAAM,CAAC,CAAC;GAE9E,MAAM,QAAQ,CAAC,SAAS,aAAa,UAAU,QAAQ,CAAC,GAAG;GAC3D,IAAI,KAAK,OAAO,YAAY,MAAM,KAAK,QAAQ,cAAc,QAAQ,CAAC,GAAG;GACzE,IAAI,KAAK,OAAO,gBAAgB,MAAM,KAAK,WAAW,iBAAiB,QAAQ,CAAC,GAAG;GACnF,IAAI,iBAAiB,MAAM,KAAK,mBAAmB;GAEnD,OAAO;IACL,QAAQ;KACN,OAAO,IAAI,WAAW;KACtB,WAAW,IAAI;KACf,eAAe,IAAI,UAAU;IAC/B;IACA,SAAS;KAAE,OAAO,aAAa;KAAG,OAAO;KAAW,OAAO,MAAM,KAAK,GAAG;IAAE;IAC3E,SAAS,IAAI;IACb,YAAY,IAAI;GAClB;EACF;CACF,CAAC;CAGD,MAAM,UAAU;EACd,OAAO,OAAO,QAAQ;EACtB,UAAU,OAAO,QAAQ;EACzB,WAAW,OAAO,QAAQ;EAC1B,cAAc,OAAO,QAAQ;EAC7B,YAAY,OAAO,QAAQ;EAC3B,cAAc,OAAO,QAAQ;EAC7B,cAAc,OAAO,QAAQ;EAC7B,WAAW,OAAO,OAAO;EACzB,WAAW,OAAO,OAAO;CAC3B;CACA,cAAc,KAAK,KAAK,QAAQ,cAAc,GAAG,KAAK,UAAU,SAAS,MAAM,CAAC,CAAC;CAEjF,MAAM,KAAe;EACnB;EACA;EACA,cAAc,OAAO,QAAQ,WAAW,qBAAqB,OAAO,QAAQ,kBAAkB,IAAA,CAAK,QAAQ,CAAC,EAAE,gBAAgB,OAAO,QAAQ,iBAAiB,QAAQ,CAAC,EAAE,gBAAgB,OAAO,QAAQ,aAAa,QAAQ,CAAC,EAAE,oBAAoB,OAAO,QAAQ,aAAa,IAAA,CAAM,QAAQ,CAAC,EAAE;EACjS;EACA;EACA;EACA;EACA;EACA,GAAG,OAAO,QAAQ,OAAO,OAAO,WAAW,CAAC,CAAC,CAAC,CAAC,KAC5C,CAAC,IAAI,OACJ,KAAK,GAAG,MAAM,EAAE,WAAW,IAAA,CAAK,QAAQ,CAAC,EAAE,MAAM,EAAE,UAAU,QAAQ,CAAC,EAAE,MAAM,EAAE,aAAa,QAAQ,CAAC,EAAE,GAC5G;EACA;EACA;EACA;EACA;EACA;EACA,GAAG,OAAO,QAAQ,OAAO,OAAO,WAAW,CAAC,CAAC,CAAC,CAAC,KAC5C,CAAC,IAAI,OACJ,KAAK,GAAG,MAAM,EAAE,WAAW,IAAA,CAAK,QAAQ,CAAC,EAAE,MAAM,EAAE,UAAU,QAAQ,CAAC,EAAE,MAAM,EAAE,aAAa,QAAQ,CAAC,EAAE,GAC5G;EACA;CACF;CACA,cAAc,KAAK,KAAK,QAAQ,YAAY,GAAG,GAAG,KAAK,IAAI,CAAC;CAE5D,OAAO,EAAE,OAAO;AAClB;AAEA,SAAS,mBACP,OACA,WACqB;CACrB,IAAI,cAAc,KAAA,KAAa,MAAM,cAAc,KAAA,GAAW,OAAO;CACrE,OAAO;EAAE,GAAG;EAAO;CAAU;AAC/B"}
+{"version":3,"file":"index.js","names":[],"sources":["../../src/multishot/router.ts","../../src/multishot/default-tools.ts","../../src/multishot/judges.ts","../../src/multishot/types.ts","../../src/multishot/multishot.ts","../../src/multishot/matrix.ts"],"sourcesContent":["// Router fetch helper — single source of truth for OpenAI-compat calls\n// against the Tangle router. Used by the driver, agent, judges, and the\n// default tool executors.\n\nimport type { MultishotToolDefinition } from './types'\n\nexport interface RouterCompletionRequest {\n  apiKey: string\n  baseUrl: string\n  model: string\n  messages: Array<Record<string, unknown>>\n  tools?: MultishotToolDefinition[]\n  temperature?: number\n  maxTokens?: number\n  signal?: AbortSignal\n}\n\nexport interface RouterToolCall {\n  id: string\n  type: 'function'\n  function: { name: string; arguments: string }\n}\n\nexport interface RouterCompletionResponse {\n  message: { content?: string | null; tool_calls?: RouterToolCall[] }\n  usage?: { prompt_tokens?: number; completion_tokens?: number }\n  /** Provider-reported spend when the endpoint supplies it. */\n  costUsd?: number\n  /** Model echoed by the provider, falling back to the requested model. */\n  model: string\n  durationMs: number\n}\n\nexport async function routerCompletion(\n  req: RouterCompletionRequest,\n): Promise<RouterCompletionResponse> {\n  const startedAt = Date.now()\n  const body: Record<string, unknown> = {\n    model: req.model,\n    messages: req.messages,\n    temperature: req.temperature ?? 0.7,\n    max_tokens: req.maxTokens ?? 2000,\n  }\n  if (req.tools?.length) body.tools = req.tools\n  const url = `${req.baseUrl.replace(/\\/+$/, '')}/chat/completions`\n  const res = await fetch(url, {\n    method: 'POST',\n    headers: { Authorization: `Bearer ${req.apiKey}`, 'Content-Type': 'application/json' },\n    body: JSON.stringify(body),\n    signal: req.signal,\n  })\n  if (!res.ok) {\n    const text = await res.text()\n    throw new Error(`router ${res.status}: ${text.slice(0, 300)}`)\n  }\n  const json = (await res.json()) as {\n    choices: Array<{ message: { content?: string | null; tool_calls?: RouterToolCall[] } }>\n    usage?: { prompt_tokens?: number; completion_tokens?: number }\n    model?: unknown\n    _response_cost?: unknown\n    cost_usd?: unknown\n  }\n  const choice = json.choices[0]\n  if (!choice) throw new Error(`router returned no choices: ${JSON.stringify(json).slice(0, 200)}`)\n  const rawCost = json._response_cost ?? json.cost_usd\n  const costUsd =\n    typeof rawCost === 'number' && Number.isFinite(rawCost) && rawCost >= 0 ? rawCost : undefined\n  return {\n    message: choice.message,\n    usage: json.usage,\n    ...(costUsd === undefined ? {} : { costUsd }),\n    model: typeof json.model === 'string' && json.model ? json.model : req.model,\n    durationMs: Date.now() - startedAt,\n  }\n}\n\n// Rough per-model cost estimator. Used for cost-ceiling enforcement.\n// Underestimates Anthropic, overestimates oss models — fine for ceilings.\nexport function estimateRouterCost(\n  model: string,\n  usage?: { prompt_tokens?: number; completion_tokens?: number },\n): number {\n  if (!usage) return 0\n  const inputTok = usage.prompt_tokens ?? 0\n  const outputTok = usage.completion_tokens ?? 0\n  let inPer1k = 0.003\n  let outPer1k = 0.015\n  if (model.includes('gpt-4o-mini')) {\n    inPer1k = 0.00015\n    outPer1k = 0.0006\n  } else if (model.includes('gpt-5.4') || model.includes('claude-sonnet')) {\n    inPer1k = 0.003\n    outPer1k = 0.015\n  } else if (model.includes('kimi') || model.includes('glm') || model.includes('deepseek')) {\n    inPer1k = 0.0005\n    outPer1k = 0.002\n  }\n  return (inputTok * inPer1k + outputTok * outPer1k) / 1000\n}\n\nexport function defaultRouterBaseUrl(): string {\n  return (process.env.TANGLE_ROUTER_BASE_URL ?? 'https://router.tangle.tools/v1').replace(\n    /\\/+$/,\n    '',\n  )\n}\n\nexport function requireRouterApiKey(): string {\n  const key = process.env.TANGLE_API_KEY\n  if (!key) throw new Error('multishot requires TANGLE_API_KEY (router-scoped sk-tan-* key)')\n  return key\n}\n","// Default delegate_research + delegate_code tools and their inline executors.\n//\n// Consumers can override either by passing their own tools + executors to\n// runMultishot. The defaults are sufficient for most domains — point the\n// researcher system prompt at your domain's citation style and the coder\n// at your preferred language.\n\nimport { estimateRouterCost, routerCompletion } from './router'\nimport type { MultishotToolDefinition, MultishotToolExecutor } from './types'\n\nexport const DEFAULT_RESEARCHER_MODEL = 'openai/gpt-4o-mini'\nexport const DEFAULT_CODER_MODEL = 'openai/gpt-4o-mini'\n\nexport interface DefaultResearcherConfig {\n  /** Replace the system prompt to bias the researcher toward a domain's\n   *  citation style. Defaults to a generic \"cite sources by name\" prompt. */\n  systemPrompt?: string\n  model?: string\n}\n\nexport interface DefaultCoderConfig {\n  /** Replace the system prompt to bias the coder toward a language /\n   *  framework / artifact style. */\n  systemPrompt?: string\n  model?: string\n}\n\nconst GENERIC_RESEARCHER_SYSTEM =\n  'You are a research specialist. Return a markdown brief with 3-5 findings. Each finding cites a specific source by name. Add a confidence level (high/medium/low) per finding. No fluff, no preamble.'\n\nconst GENERIC_CODER_SYSTEM =\n  'You are an expert engineer. Output ONE fenced code block containing the complete solution. Inline-comment non-obvious decisions. No explanation outside the block.'\n\nexport const DEFAULT_DELEGATE_RESEARCH_TOOL: MultishotToolDefinition = {\n  type: 'function',\n  function: {\n    name: 'delegate_research',\n    description:\n      'Research a topic deeply via specialist. Returns evidence-bearing items with citations. Use for audience research, competitive intel, regulatory landscape, market data, citation-grounded analysis.',\n    parameters: {\n      type: 'object',\n      properties: {\n        question: { type: 'string', description: 'Specific question to research' },\n        scope: {\n          type: 'string',\n          description: 'Optional scope: time window, geography, jurisdiction, segment',\n        },\n      },\n      required: ['question'],\n    },\n  },\n}\n\nexport const DEFAULT_DELEGATE_CODE_TOOL: MultishotToolDefinition = {\n  type: 'function',\n  function: {\n    name: 'delegate_code',\n    description:\n      'Generate a runnable script, template, pipeline, or tool via specialist. Returns complete working code or structured markdown. Use for content pipelines, calc snippets, dashboards, compliance checklists, deadline trackers.',\n    parameters: {\n      type: 'object',\n      properties: {\n        goal: { type: 'string', description: 'What the code must accomplish' },\n        language: {\n          type: 'string',\n          description: 'Optional language preference (default: TypeScript)',\n        },\n      },\n      required: ['goal'],\n    },\n  },\n}\n\nexport function createResearchExecutor(\n  config: DefaultResearcherConfig = {},\n): MultishotToolExecutor {\n  const systemPrompt = config.systemPrompt ?? GENERIC_RESEARCHER_SYSTEM\n  const model = config.model ?? DEFAULT_RESEARCHER_MODEL\n  return async (args, ctx) => {\n    const question = String(args.question ?? '')\n    const scope = args.scope ? String(args.scope) : undefined\n    const { message, usage } = await routerCompletion({\n      apiKey: ctx.apiKey,\n      baseUrl: ctx.baseUrl,\n      model,\n      temperature: 0.3,\n      maxTokens: 1800,\n      messages: [\n        { role: 'system', content: systemPrompt },\n        { role: 'user', content: `Research: ${question}${scope ? `\\nScope: ${scope}` : ''}` },\n      ],\n      signal: ctx.signal,\n    })\n    return { content: message.content ?? '', costUsd: estimateRouterCost(model, usage) }\n  }\n}\n\nexport function createCodeExecutor(config: DefaultCoderConfig = {}): MultishotToolExecutor {\n  const systemPrompt = config.systemPrompt ?? GENERIC_CODER_SYSTEM\n  const model = config.model ?? DEFAULT_CODER_MODEL\n  return async (args, ctx) => {\n    const goal = String(args.goal ?? '')\n    const language = args.language ? String(args.language) : 'TypeScript'\n    const { message, usage } = await routerCompletion({\n      apiKey: ctx.apiKey,\n      baseUrl: ctx.baseUrl,\n      model,\n      temperature: 0.2,\n      maxTokens: 2000,\n      messages: [\n        { role: 'system', content: `${systemPrompt}\\n\\nLanguage: ${language}` },\n        { role: 'user', content: `Produce: ${goal}` },\n      ],\n      signal: ctx.signal,\n    })\n    return { content: message.content ?? '', costUsd: estimateRouterCost(model, usage) }\n  }\n}\n\nexport interface DefaultToolsConfig {\n  research?: DefaultResearcherConfig\n  code?: DefaultCoderConfig\n  /** When true (default), each tool result is recorded as a typed artifact:\n   *  research → type='research', code → type='code'. */\n  recordArtifacts?: boolean\n}\n\nexport interface DefaultToolsBundle {\n  tools: MultishotToolDefinition[]\n  executors: Record<string, MultishotToolExecutor>\n  artifactTypeFor: (toolName: string) => string | undefined\n}\n\nexport function defaultDelegationTools(config: DefaultToolsConfig = {}): DefaultToolsBundle {\n  return {\n    tools: [DEFAULT_DELEGATE_RESEARCH_TOOL, DEFAULT_DELEGATE_CODE_TOOL],\n    executors: {\n      delegate_research: createResearchExecutor(config.research),\n      delegate_code: createCodeExecutor(config.code),\n    },\n    artifactTypeFor: (name) =>\n      name === 'delegate_research' ? 'research' : name === 'delegate_code' ? 'code' : undefined,\n  }\n}\n\nexport { defaultRouterBaseUrl } from './router'\n","// Generic judge runner — domain consumers configure dimensions + prompts.\n//\n// Three judge slots are conventional for multishot eval:\n//   - conversation (scores the full transcript)\n//   - codeReview (scores each code artifact)\n//   - contentQuality (scores each non-code artifact)\n//\n// But the runJudge primitive is fully generic: any input maps to a score plus\n// explicit observed, estimated, or uncaptured cost.\n\nimport type { JudgeScore } from '../campaign/types'\nimport type { CostProvenance } from '../cost-ledger'\nimport type { LlmCallMetadata, LlmUsage } from '../llm-client'\nimport { estimateCost, isModelPriced } from '../metrics'\nimport {\n  defaultRouterBaseUrl,\n  type RouterCompletionResponse,\n  requireRouterApiKey,\n  routerCompletion,\n} from './router'\n\n// Canonical declaration lives in campaign/types.ts. Multishot emits the same\n// shape on its producer-defined 0-10 scale.\nexport type { JudgeScore } from '../campaign/types'\n\nexport const DEFAULT_JUDGE_MODEL = 'openai/gpt-4o-mini'\n\nexport interface JudgeDimension {\n  /** JSON field name + score key. */\n  key: string\n  /** Description shown in the judge's user prompt. */\n  description: string\n}\n\nexport interface JudgeConfig<TInput> {\n  /** Display name (for trace + log). */\n  name: string\n  /** Model used for this judge. */\n  model?: string\n  /** 0-10 scored dimensions. */\n  dimensions: JudgeDimension[]\n  /** Judge system prompt — sets persona + JSON-only constraint. */\n  systemPrompt: string\n  /** Build the user prompt from the typed input. Must include \"Respond with\n   *  ONLY this JSON: { ... }\" listing each dimension key. */\n  buildPrompt: (input: TInput) => string\n  /** Optional model + api overrides. */\n  apiKey?: string\n  baseUrl?: string\n  /** Maximum output tokens for the judge response. Defaults to 1500. */\n  maxTokens?: number\n}\n\nexport interface JudgeRunResult {\n  /** Semantic result; failed scores remain non-throwing so matrix aggregation can exclude them. */\n  score: JudgeScore\n  /** Cost of the completed call, separate from diagnostic provider metadata. */\n  cost: CostProvenance\n}\n\nexport async function runJudge<TInput>(\n  judge: JudgeConfig<TInput>,\n  input: TInput,\n): Promise<JudgeRunResult> {\n  const apiKey = judge.apiKey ?? requireRouterApiKey()\n  const baseUrl = judge.baseUrl ?? defaultRouterBaseUrl()\n  const model = judge.model ?? process.env.JUDGE_MODEL ?? DEFAULT_JUDGE_MODEL\n  const prompt = judge.buildPrompt(input)\n  let raw = ''\n  let llmCall: LlmCallMetadata\n  let cost: CostProvenance\n  const startedAt = Date.now()\n  try {\n    const response = await routerCompletion({\n      apiKey,\n      baseUrl,\n      model,\n      temperature: 0,\n      maxTokens: judge.maxTokens ?? 1500,\n      messages: [\n        { role: 'system', content: judge.systemPrompt },\n        { role: 'user', content: prompt },\n      ],\n    })\n    const call = judgeCallMetadata(response)\n    llmCall = call.llmCall\n    cost = call.cost\n    raw = (response.message.content ?? '').trim()\n  } catch (err) {\n    // failed:true lets consumers reading `.composite` keep working while\n    // aggregators exclude this score from means instead of averaging a zero.\n    return {\n      score: {\n        dimensions: {},\n        composite: 0,\n        failed: true,\n        notes: `judge ${judge.name} call failed: ${err instanceof Error ? err.message : String(err)}`,\n        llmCall: unknownJudgeCall(model, Date.now() - startedAt),\n      },\n      cost: { kind: 'uncaptured', usd: null },\n    }\n  }\n\n  let parsed: Record<string, unknown> | null = null\n  try {\n    const cleaned = raw\n      .replace(/^```json\\s*/i, '')\n      .replace(/```\\s*$/, '')\n      .trim()\n    parsed = JSON.parse(cleaned) as Record<string, unknown>\n  } catch {\n    return {\n      score: {\n        dimensions: {},\n        composite: 0,\n        failed: true,\n        notes: `judge ${judge.name} returned non-JSON: ${raw.slice(0, 200)}`,\n        llmCall,\n      },\n      cost,\n    }\n  }\n\n  const dimensions: Record<string, number> = {}\n  let sum = 0\n  for (const dim of judge.dimensions) {\n    const v = Number(parsed[dim.key] ?? 0)\n    const clamped = Number.isFinite(v) ? Math.max(0, Math.min(10, v)) : 0\n    dimensions[dim.key] = clamped\n    sum += clamped\n  }\n  return {\n    score: {\n      dimensions,\n      composite: judge.dimensions.length === 0 ? 0 : sum / judge.dimensions.length,\n      notes: typeof parsed.notes === 'string' ? parsed.notes : '',\n      llmCall,\n    },\n    cost,\n  }\n}\n\nfunction judgeCallMetadata(response: RouterCompletionResponse): {\n  llmCall: LlmCallMetadata\n  cost: CostProvenance\n} {\n  const usage = canonicalUsage(response.usage)\n  return {\n    llmCall: {\n      usage,\n      costUsd: response.costUsd ?? null,\n      model: response.model,\n      durationMs: response.durationMs,\n    },\n    cost:\n      response.costUsd !== undefined\n        ? { kind: 'observed', usd: response.costUsd }\n        : usage.captured === false || !isModelPriced(response.model)\n          ? { kind: 'uncaptured', usd: null }\n          : {\n              kind: 'estimated',\n              usd: estimateCost(usage.promptTokens, usage.completionTokens, response.model),\n            },\n  }\n}\n\nfunction canonicalUsage(usage: RouterCompletionResponse['usage']): LlmUsage {\n  const promptTokens = tokenCount(usage?.prompt_tokens)\n  const completionTokens = tokenCount(usage?.completion_tokens)\n  const captured = promptTokens !== undefined && completionTokens !== undefined\n  return {\n    promptTokens: promptTokens ?? 0,\n    completionTokens: completionTokens ?? 0,\n    totalTokens: (promptTokens ?? 0) + (completionTokens ?? 0),\n    captured,\n  }\n}\n\nfunction unknownJudgeCall(model: string, durationMs: number): LlmCallMetadata {\n  return {\n    usage: { promptTokens: 0, completionTokens: 0, totalTokens: 0, captured: false },\n    costUsd: null,\n    model,\n    durationMs,\n  }\n}\n\nfunction tokenCount(value: unknown): number | undefined {\n  return typeof value === 'number' && Number.isSafeInteger(value) && value >= 0 ? value : undefined\n}\n\n/** Convenience: stringified dimension list for inclusion in a judge prompt.\n *  Returns lines like `- audience_fit: Does this match what the audience cares about? (0-10)`. */\nexport function renderDimensions(dims: readonly JudgeDimension[]): string {\n  return dims.map((d) => `- ${d.key}: ${d.description}`).join('\\n')\n}\n\n/** Convenience: build the \"Respond with ONLY this JSON\" footer for a judge prompt. */\nexport function renderJsonFooter(dims: readonly JudgeDimension[]): string {\n  const fields = dims.map((d) => `\"${d.key}\":N`).join(',')\n  return `Respond with ONLY this JSON (no markdown, no preamble):\\n{${fields},\"notes\":\"1-2 sentence critique\"}`\n}\n","// Public types for the multishot substrate.\n\nexport interface MultishotMessage {\n  role: 'user' | 'assistant' | 'tool'\n  content: string\n  toolCallId?: string\n  toolCalls?: Array<{ id: string; name: string; args: Record<string, unknown> }>\n}\n\nexport interface MultishotArtifact {\n  type: string\n  turn: number\n  invocation: { name: string; args: Record<string, unknown> }\n  content: string\n}\n\nexport interface MultishotResult {\n  transcript: MultishotMessage[]\n  artifacts: MultishotArtifact[]\n  toolCalls: number\n  durationMs: number\n  costUsd: number\n}\n\nexport interface MultishotToolDefinition {\n  type: 'function'\n  function: {\n    name: string\n    description: string\n    parameters: Record<string, unknown>\n  }\n}\n\n/** One chat-completion request the multishot loop issues for a single agent\n *  (or driver) inference step. Mirrors the OpenAI-compat body the loop would\n *  otherwise POST to the Tangle router. */\nexport interface MultishotTransportRequest {\n  model: string\n  messages: Array<Record<string, unknown>>\n  tools?: MultishotToolDefinition[]\n  temperature?: number\n  maxTokens?: number\n  signal?: AbortSignal\n}\n\nexport interface MultishotTransportToolCall {\n  id: string\n  type: 'function'\n  function: { name: string; arguments: string }\n}\n\nexport interface MultishotTransportResponse {\n  message: { content?: string | null; tool_calls?: MultishotTransportToolCall[] }\n  usage?: { prompt_tokens?: number; completion_tokens?: number }\n  /** Actual spend for this call. When omitted, the loop meters cost from\n   *  `usage` via the per-model router estimator (estimateRouterCost). */\n  costUsd?: number\n}\n\n/** Execution seam for one leg of the multishot loop. When provided, it\n *  replaces the internal router HTTP call for that leg — the loop still owns\n *  turn scheduling, tool dispatch, transcript capture, and cost metering.\n *  agent-eval has no dependency on agent-runtime; adapt agent-runtime's\n *  resolveAgentBackend (or any sandbox/cli-bridge/router client) into this\n *  signature product-side. */\nexport type MultishotTransport = (\n  req: MultishotTransportRequest,\n) => Promise<MultishotTransportResponse>\n\nexport type MultishotToolExecutor = (\n  args: Record<string, unknown>,\n  ctx: { apiKey: string; baseUrl: string; signal?: AbortSignal },\n) => Promise<{ content: string; costUsd: number }>\n\nexport interface MultishotPersona {\n  /** Stable identifier — used for per-cell artifact paths + matrix axis keys. */\n  id: string\n  /** Per-domain payload (income/profile/voice/etc.) shaped by the consumer. */\n  [k: string]: unknown\n}\n\nexport interface MultishotShape<TPersona extends MultishotPersona> {\n  /** Opening user message (turn 0) — the persona's first ask. */\n  buildOpener: (persona: TPersona) => string\n  /** System prompt the driver LLM uses to roleplay the persona. Should set\n   *  voice, goals, constraints, time-pressure, and the \"never go silent\" rule. */\n  buildDriverSystemPrompt: (persona: TPersona) => string\n}\n\nexport class MultishotDriverEmptyError extends Error {\n  constructor(public readonly turn: number) {\n    super(`multishot: driver returned empty content twice at turn ${turn} — failing loud`)\n    this.name = 'MultishotDriverEmptyError'\n  }\n}\n\nexport class MultishotFatalToolError extends Error {\n  constructor(message: string) {\n    super(message)\n    this.name = 'MultishotFatalToolError'\n  }\n}\n","// Multi-turn driver-agent simulation with inline tool execution.\n//\n// The driver = LLM acting as the persona (reactive, non-deterministic).\n// The agent = the product agent under test (router call by default, or an\n// injected transport — with profile's systemPrompt + the configured tools).\n// Tool calls execute inline via the configured executors and feed back\n// into the agent's message log so the agent integrates the result.\n\nimport type { AgentProfile } from '@tangle-network/agent-interface'\nimport { defaultDelegationTools } from './default-tools'\nimport {\n  defaultRouterBaseUrl,\n  estimateRouterCost,\n  requireRouterApiKey,\n  routerCompletion,\n} from './router'\nimport {\n  type MultishotArtifact,\n  MultishotDriverEmptyError,\n  MultishotFatalToolError,\n  type MultishotMessage,\n  type MultishotPersona,\n  type MultishotResult,\n  type MultishotShape,\n  type MultishotToolDefinition,\n  type MultishotToolExecutor,\n  type MultishotTransport,\n} from './types'\n\nexport interface RunMultishotOptions<TPersona extends MultishotPersona> {\n  profile: AgentProfile\n  persona: TPersona\n  shape: MultishotShape<TPersona>\n  /** Tool definitions advertised to the agent. Defaults to delegate_research + delegate_code. */\n  tools?: MultishotToolDefinition[]\n  /** Map from tool name → executor invoked inline when the agent emits a tool_call. */\n  toolExecutors?: Record<string, MultishotToolExecutor>\n  /** Map from tool name → artifact type label written into MultishotArtifact.type.\n   *  Tools without a mapping still execute, but their results aren't surfaced as\n   *  typed artifacts (only as tool messages in the transcript). */\n  artifactTypeFor?: (toolName: string) => string | undefined\n  maxTurns?: number\n  agentModel?: string\n  driverModel?: string\n  /** Fallback driver models tried when the primary simulated-user model returns empty twice. */\n  driverFallbackModels?: string[]\n  /** Maximum output tokens for the first agent call in each assistant turn. */\n  agentMaxTokens?: number\n  /** Maximum output tokens for agent follow-up calls after tool results. */\n  toolFollowupMaxTokens?: number\n  /** Maximum output tokens for each simulated-user driver response. */\n  driverMaxTokens?: number\n  /** Maximum tool calls the agent may dispatch inside one assistant turn. */\n  maxToolDispatches?: number\n  /** Execution seam for the agent leg. When provided, every agent inference\n   *  step goes through this function instead of the router HTTP call; the\n   *  string levers (agentModel, apiKey, baseUrl) stop applying to that leg.\n   *  apiKey/baseUrl are still resolved for tool executors and any leg\n   *  without an injected transport. */\n  agentTransport?: MultishotTransport\n  /** Execution seam for the simulated-user driver leg (symmetric to\n   *  agentTransport). Driver model fallback rotation still applies — the\n   *  transport receives each candidate model in turn. */\n  driverTransport?: MultishotTransport\n  apiKey?: string\n  baseUrl?: string\n  signal?: AbortSignal\n}\n\nexport async function runMultishot<TPersona extends MultishotPersona>(\n  opts: RunMultishotOptions<TPersona>,\n): Promise<MultishotResult> {\n  const apiKey = opts.apiKey ?? requireRouterApiKey()\n  const baseUrl = opts.baseUrl ?? defaultRouterBaseUrl()\n  const maxTurns = opts.maxTurns ?? 10\n  const maxToolDispatches = opts.maxToolDispatches ?? 4\n  const agentModel = opts.agentModel ?? 'openai/gpt-5.4'\n  const driverModel = opts.driverModel ?? 'openai/gpt-4o-mini'\n  const driverModels = [driverModel, ...(opts.driverFallbackModels ?? [])]\n  const agentMaxTokens = opts.agentMaxTokens ?? 2500\n  const toolFollowupMaxTokens = opts.toolFollowupMaxTokens ?? 2000\n  const driverMaxTokens = opts.driverMaxTokens ?? 600\n\n  const bundle =\n    opts.tools && opts.toolExecutors\n      ? {\n          tools: opts.tools,\n          executors: opts.toolExecutors,\n          artifactTypeFor: opts.artifactTypeFor ?? (() => undefined),\n        }\n      : defaultDelegationTools()\n  const tools = opts.tools ?? bundle.tools\n  const executors = opts.toolExecutors ?? bundle.executors\n  const artifactTypeFor = opts.artifactTypeFor ?? bundle.artifactTypeFor\n\n  const routerTransport: MultishotTransport = (req) => routerCompletion({ apiKey, baseUrl, ...req })\n  const agentTransport = opts.agentTransport ?? routerTransport\n  const driverTransport = opts.driverTransport ?? routerTransport\n\n  const start = Date.now()\n  const transcript: MultishotMessage[] = []\n  const artifacts: MultishotArtifact[] = []\n  let toolCalls = 0\n  let totalCostUsd = 0\n\n  const opener = opts.shape.buildOpener(opts.persona)\n  transcript.push({ role: 'user', content: opener })\n\n  const systemPrompt = opts.profile.prompt?.systemPrompt ?? ''\n  const agentMessages: Array<Record<string, unknown>> = [\n    { role: 'system', content: systemPrompt },\n    { role: 'user', content: opener },\n  ]\n\n  for (let turn = 0; turn < maxTurns; turn++) {\n    if (opts.signal?.aborted) throw new Error('multishot aborted')\n\n    let dispatchesThisTurn = 0\n    while (true) {\n      const {\n        message: agentMsg,\n        usage: agentUsage,\n        costUsd: agentCostUsd,\n      } = await agentTransport({\n        model: agentModel,\n        messages: agentMessages,\n        tools,\n        temperature: 0.7,\n        maxTokens: dispatchesThisTurn === 0 ? agentMaxTokens : toolFollowupMaxTokens,\n        signal: opts.signal,\n      })\n      totalCostUsd += agentCostUsd ?? estimateRouterCost(agentModel, agentUsage)\n\n      const agentText = (agentMsg.content ?? '').trim()\n      const agentToolCalls = (agentMsg.tool_calls ?? []).map((tc) => ({\n        id: tc.id,\n        name: tc.function.name,\n        args: (() => {\n          try {\n            return JSON.parse(tc.function.arguments) as Record<string, unknown>\n          } catch {\n            return {} as Record<string, unknown>\n          }\n        })(),\n      }))\n\n      agentMessages.push({\n        role: 'assistant',\n        content: agentText || null,\n        ...(agentMsg.tool_calls?.length ? { tool_calls: agentMsg.tool_calls } : {}),\n      })\n      transcript.push({\n        role: 'assistant',\n        content: agentText,\n        toolCalls: agentToolCalls.length > 0 ? agentToolCalls : undefined,\n      })\n\n      if (agentToolCalls.length === 0) break\n      dispatchesThisTurn += agentToolCalls.length\n      if (dispatchesThisTurn > maxToolDispatches) {\n        throw new Error(\n          `multishot: tool dispatch cap exceeded (${dispatchesThisTurn}/${maxToolDispatches}) on turn ${turn}`,\n        )\n      }\n\n      for (const tc of agentToolCalls) {\n        toolCalls++\n        let toolResult = ''\n        try {\n          const executor = executors[tc.name]\n          if (!executor) {\n            toolResult = JSON.stringify({ error: `unknown tool ${tc.name}` })\n          } else {\n            const r = await executor(tc.args, { apiKey, baseUrl, signal: opts.signal })\n            toolResult = r.content\n            totalCostUsd += r.costUsd\n            const artifactType = artifactTypeFor(tc.name)\n            if (artifactType) {\n              artifacts.push({\n                type: artifactType,\n                turn,\n                invocation: { name: tc.name, args: tc.args },\n                content: toolResult,\n              })\n            }\n          }\n        } catch (err) {\n          if (err instanceof MultishotFatalToolError) throw err\n          toolResult = JSON.stringify({ error: err instanceof Error ? err.message : String(err) })\n        }\n        agentMessages.push({ role: 'tool', tool_call_id: tc.id, content: toolResult || 'done' })\n        transcript.push({ role: 'tool', content: toolResult || 'done', toolCallId: tc.id })\n      }\n    }\n\n    if (turn < maxTurns - 1) {\n      const driver = await driverTurn({\n        transport: driverTransport,\n        persona: opts.persona,\n        shape: opts.shape,\n        transcript,\n        turn,\n        models: driverModels,\n        maxTokens: driverMaxTokens,\n        signal: opts.signal,\n      })\n      totalCostUsd += driver.costUsd\n      agentMessages.push({ role: 'user', content: driver.content })\n      transcript.push({ role: 'user', content: driver.content })\n    }\n  }\n\n  return { transcript, artifacts, toolCalls, durationMs: Date.now() - start, costUsd: totalCostUsd }\n}\n\nasync function driverTurn<TPersona extends MultishotPersona>(opts: {\n  transport: MultishotTransport\n  persona: TPersona\n  shape: MultishotShape<TPersona>\n  transcript: MultishotMessage[]\n  turn: number\n  models: string[]\n  maxTokens: number\n  signal?: AbortSignal\n}): Promise<{ content: string; costUsd: number }> {\n  const driverSystem = opts.shape.buildDriverSystemPrompt(opts.persona)\n\n  // Translate transcript to driver POV: agent's `assistant` messages become\n  // `user` (the agent talking TO the driver); the driver's prior `user`\n  // messages become `assistant` (the driver's prior responses).\n  const driverMessages: Array<Record<string, unknown>> = [{ role: 'system', content: driverSystem }]\n  for (const msg of opts.transcript) {\n    if (msg.role === 'tool') continue\n    const content = driverVisibleContent(msg)\n    if (!content) continue\n    if (msg.role === 'assistant') driverMessages.push({ role: 'user', content })\n    else if (msg.role === 'user') driverMessages.push({ role: 'assistant', content })\n  }\n\n  // Driver must never go silent. Retry once on empty content; then fail loud.\n  for (const model of opts.models) {\n    for (let attempt = 0; attempt < 2; attempt++) {\n      const { message, usage, costUsd } = await opts.transport({\n        model,\n        messages: driverMessages,\n        temperature: 0.9,\n        maxTokens: opts.maxTokens,\n        signal: opts.signal,\n      })\n      const content = (message.content ?? '').trim()\n      if (content.length > 0)\n        return { content, costUsd: costUsd ?? estimateRouterCost(model, usage) }\n    }\n  }\n  throw new MultishotDriverEmptyError(opts.turn)\n}\n\nfunction driverVisibleContent(msg: MultishotMessage): string | null {\n  const text = msg.content.trim()\n  if (text.length > 0) return text\n  if (msg.role !== 'assistant' || !msg.toolCalls?.length) return null\n\n  const toolNames = msg.toolCalls.map((call) => call.name.trim()).filter(Boolean)\n  if (toolNames.length === 0) return 'Agent called tools.'\n  return `Agent called ${toolNames.length === 1 ? 'tool' : 'tools'}: ${toolNames.join(', ')}.`\n}\n","// Multishot matrix wrapper — sweeps profiles × personas × reps, runs\n// the driver-agent loop per cell, applies up to three configured judges,\n// persists per-cell artifacts, and aggregates by axis.\n//\n// Uses runAgentMatrix from @tangle-network/agent-eval/matrix under the\n// hood so cell scheduling + concurrency + cost ceiling are unified with\n// other matrix consumers.\n\nimport { mkdirSync, writeFileSync } from 'node:fs'\nimport { join } from 'node:path'\nimport type { AgentProfile } from '@tangle-network/agent-interface'\nimport type { CostProvenance } from '../cost-ledger'\nimport type { MatrixResult } from '../matrix'\nimport { runAgentMatrix } from '../matrix'\nimport { type JudgeConfig, type JudgeScore, runJudge } from './judges'\nimport { runMultishot } from './multishot'\nimport type {\n  MultishotArtifact,\n  MultishotMessage,\n  MultishotPersona,\n  MultishotShape,\n  MultishotToolDefinition,\n  MultishotToolExecutor,\n  MultishotTransport,\n} from './types'\n\nexport interface ConversationJudgeInput<TPersona extends MultishotPersona> {\n  transcript: MultishotMessage[]\n  persona: TPersona\n}\n\nexport interface ArtifactJudgeInput<TPersona extends MultishotPersona> {\n  artifact: MultishotArtifact\n  persona: TPersona\n}\n\nexport interface MultishotJudges<TPersona extends MultishotPersona> {\n  /** Scores the full transcript end-to-end (always runs). */\n  conversation: JudgeConfig<ConversationJudgeInput<TPersona>>\n  /** Scores each code-type artifact. Optional — omit when domain has no code artifacts. */\n  codeReview?: JudgeConfig<ArtifactJudgeInput<TPersona>>\n  /** Scores each non-code (research/content/template) artifact. Optional. */\n  contentQuality?: JudgeConfig<ArtifactJudgeInput<TPersona>>\n  /** Which artifact types route to codeReview. Defaults to ['code']. */\n  codeArtifactTypes?: string[]\n  /** Which artifact types route to contentQuality. Defaults to ['research']. */\n  contentArtifactTypes?: string[]\n}\n\nexport interface CellCompositeScore {\n  composite: number\n  conversation: JudgeScore\n  codeReview?: {\n    perArtifact: Array<JudgeScore & { turn: number; type: string }>\n    composite: number\n  }\n  contentQuality?: {\n    perArtifact: Array<JudgeScore & { turn: number; type: string }>\n    composite: number\n  }\n}\n\nexport interface RunMultishotMatrixOptions<TPersona extends MultishotPersona> {\n  /** AgentProfile axis (matrix primary). */\n  profiles: Array<{ id: string; value: AgentProfile }>\n  /** Persona axis. */\n  personas: TPersona[]\n  /** Persona-shaping callbacks. */\n  shape: MultishotShape<TPersona>\n  /** Judge configurations. */\n  judges: MultishotJudges<TPersona>\n  /** Tool definitions advertised to the agent. Defaults to delegate_research + delegate_code. */\n  tools?: MultishotToolDefinition[]\n  /** Map from tool name → inline executor. Must align with `tools`. */\n  toolExecutors?: Record<string, MultishotToolExecutor>\n  /** Tool name → artifact type label. Defaults to research/code mapping. */\n  artifactTypeFor?: (toolName: string) => string | undefined\n  /** Where per-cell artifacts land. Cells write to `<runDir>/<profileId>/<personaId>/rep-N/`. */\n  runDir: string\n  /** Replicates per (profile, persona) cell. */\n  reps?: number\n  /** Max conversation turns per cell. */\n  maxTurns?: number\n  /** Maximum tool calls the agent may dispatch inside one assistant turn. */\n  maxToolDispatches?: number\n  /** Max concurrent cells. */\n  maxConcurrency?: number\n  /** Total $ ceiling across the matrix; cells aborted past this. */\n  costCeiling?: number\n  /** Agent model. */\n  agentModel?: string\n  /** Driver model. */\n  driverModel?: string\n  /** Fallback driver models tried when the primary simulated-user model returns empty twice. */\n  driverFallbackModels?: string[]\n  /** Maximum output tokens for the first agent call in each assistant turn. */\n  agentMaxTokens?: number\n  /** Maximum output tokens for agent follow-up calls after tool results. */\n  toolFollowupMaxTokens?: number\n  /** Maximum output tokens for each simulated-user driver response. */\n  driverMaxTokens?: number\n  /** Maximum output tokens for each judge response. */\n  judgeMaxTokens?: number\n  /** Execution seam for the agent leg of every cell — replaces the router\n   *  HTTP call when provided (see RunMultishotOptions.agentTransport).\n   *  Judges are unaffected; configure those via MultishotJudges. */\n  agentTransport?: MultishotTransport\n  /** Execution seam for the simulated-user driver leg of every cell. */\n  driverTransport?: MultishotTransport\n  /** Pass-thru fields. */\n  apiKey?: string\n  baseUrl?: string\n}\n\ninterface CellOutput {\n  turns: number\n  toolCalls: number\n  artifactCount: number\n}\n\ninterface ArtifactJudgeRun {\n  score: JudgeScore & { turn: number; type: string }\n  cost: CostProvenance\n}\n\n/** Mean composite over non-failed scores. `0` when the list is empty (a\n *  configured judge with nothing to score contributes 0, matching the cell\n *  composite's long-standing semantics); `null` when scores exist but every\n *  one failed — no signal, so the slot must be EXCLUDED from the cell mean\n *  rather than dragging it to zero. */\nfunction meanCompositeExcludingFailed(scores: ReadonlyArray<JudgeScore>): number | null {\n  if (scores.length === 0) return 0\n  const live = scores.filter((s) => !s.failed)\n  if (live.length === 0) return null\n  return live.reduce((sum, s) => sum + s.composite, 0) / live.length\n}\n\nexport interface CellCompositeInput {\n  conversation: JudgeScore\n  /** Present iff the codeReview judge is configured. */\n  codeReviews?: ReadonlyArray<JudgeScore>\n  /** Present iff the contentQuality judge is configured. */\n  contentReviews?: ReadonlyArray<JudgeScore>\n}\n\n/** Cell composite = mean over configured judge slots, excluding failed\n *  scores: a failed conversation judge or an all-failed artifact slot carries\n *  no signal and is dropped from the mean. `composite` is 0 only when EVERY\n *  configured slot failed (`allJudgesFailed` distinguishes that from a real\n *  zero). Pure — exported for deterministic testing. */\nexport function computeCellComposite(input: CellCompositeInput): {\n  composite: number\n  codeComposite: number\n  contentComposite: number\n  allJudgesFailed: boolean\n} {\n  const contributions: number[] = []\n  if (!input.conversation.failed) contributions.push(input.conversation.composite)\n\n  const codeMean = input.codeReviews ? meanCompositeExcludingFailed(input.codeReviews) : undefined\n  if (typeof codeMean === 'number') contributions.push(codeMean)\n  const contentMean = input.contentReviews\n    ? meanCompositeExcludingFailed(input.contentReviews)\n    : undefined\n  if (typeof contentMean === 'number') contributions.push(contentMean)\n\n  return {\n    composite:\n      contributions.length === 0\n        ? 0\n        : contributions.reduce((s, v) => s + v, 0) / contributions.length,\n    codeComposite: codeMean ?? 0,\n    contentComposite: contentMean ?? 0,\n    allJudgesFailed: contributions.length === 0,\n  }\n}\n\nexport interface RunMultishotMatrixResult {\n  matrix: MatrixResult<CellOutput>\n}\n\nexport async function runMultishotMatrix<TPersona extends MultishotPersona>(\n  opts: RunMultishotMatrixOptions<TPersona>,\n): Promise<RunMultishotMatrixResult> {\n  const codeTypes = new Set(opts.judges.codeArtifactTypes ?? ['code'])\n  const contentTypes = new Set(opts.judges.contentArtifactTypes ?? ['research'])\n  mkdirSync(opts.runDir, { recursive: true })\n\n  const matrix = await runAgentMatrix<CellOutput>({\n    axes: [\n      { name: 'profile', values: opts.profiles },\n      { name: 'persona', values: opts.personas.map((p) => ({ id: p.id, value: p })) },\n    ],\n    reps: opts.reps ?? 1,\n    maxConcurrency: opts.maxConcurrency ?? 2,\n    costCeiling: opts.costCeiling,\n    async runCell(cell) {\n      const profile = cell.axes.profile?.value as AgentProfile\n      const persona = cell.axes.persona?.value as TPersona\n      const profileId = String(cell.axes.profile?.id ?? 'unknown')\n      const personaId = String(cell.axes.persona?.id ?? 'unknown')\n\n      const sim = await runMultishot({\n        profile,\n        persona,\n        shape: opts.shape,\n        tools: opts.tools,\n        toolExecutors: opts.toolExecutors,\n        artifactTypeFor: opts.artifactTypeFor,\n        maxTurns: opts.maxTurns,\n        maxToolDispatches: opts.maxToolDispatches,\n        agentModel: opts.agentModel,\n        driverModel: opts.driverModel,\n        driverFallbackModels: opts.driverFallbackModels,\n        agentMaxTokens: opts.agentMaxTokens,\n        toolFollowupMaxTokens: opts.toolFollowupMaxTokens,\n        driverMaxTokens: opts.driverMaxTokens,\n        agentTransport: opts.agentTransport,\n        driverTransport: opts.driverTransport,\n        apiKey: opts.apiKey,\n        baseUrl: opts.baseUrl,\n      })\n\n      const codeArtifacts = sim.artifacts.filter((a) => codeTypes.has(a.type))\n      const contentArtifacts = sim.artifacts.filter((a) => contentTypes.has(a.type))\n\n      const [conversationRun, codeReviewRuns, contentReviewRuns] = await Promise.all([\n        runJudge(withJudgeMaxTokens(opts.judges.conversation, opts.judgeMaxTokens), {\n          transcript: sim.transcript,\n          persona,\n        }),\n        opts.judges.codeReview\n          ? Promise.all(\n              codeArtifacts.map((artifact) =>\n                runJudge(withJudgeMaxTokens(opts.judges.codeReview!, opts.judgeMaxTokens), {\n                  artifact,\n                  persona,\n                }).then((result) => ({\n                  score: {\n                    ...result.score,\n                    turn: artifact.turn,\n                    type: artifact.type,\n                  },\n                  cost: result.cost,\n                })),\n              ),\n            )\n          : Promise.resolve([] as ArtifactJudgeRun[]),\n        opts.judges.contentQuality\n          ? Promise.all(\n              contentArtifacts.map((artifact) =>\n                runJudge(withJudgeMaxTokens(opts.judges.contentQuality!, opts.judgeMaxTokens), {\n                  artifact,\n                  persona,\n                }).then((result) => ({\n                  score: {\n                    ...result.score,\n                    turn: artifact.turn,\n                    type: artifact.type,\n                  },\n                  cost: result.cost,\n                })),\n              ),\n            )\n          : Promise.resolve([] as ArtifactJudgeRun[]),\n      ])\n      const conversation = conversationRun.score\n      const codeReviews = codeReviewRuns.map((run) => run.score)\n      const contentReviews = contentReviewRuns.map((run) => run.score)\n\n      const { composite, codeComposite, contentComposite, allJudgesFailed } = computeCellComposite({\n        conversation,\n        codeReviews: opts.judges.codeReview ? codeReviews : undefined,\n        contentReviews: opts.judges.contentQuality ? contentReviews : undefined,\n      })\n\n      const cellScore: CellCompositeScore = { composite, conversation }\n      if (opts.judges.codeReview)\n        cellScore.codeReview = { perArtifact: codeReviews, composite: codeComposite }\n      if (opts.judges.contentQuality)\n        cellScore.contentQuality = { perArtifact: contentReviews, composite: contentComposite }\n\n      const cellDir = join(opts.runDir, profileId, personaId, `rep-${cell.rep}`)\n      mkdirSync(cellDir, { recursive: true })\n      writeFileSync(join(cellDir, 'transcript.json'), JSON.stringify(sim.transcript, null, 2))\n      writeFileSync(join(cellDir, 'artifacts.json'), JSON.stringify(sim.artifacts, null, 2))\n      writeFileSync(join(cellDir, 'scores.json'), JSON.stringify(cellScore, null, 2))\n\n      const notes = [`convo=${conversation.composite.toFixed(1)}`]\n      if (opts.judges.codeReview) notes.push(`code=${codeComposite.toFixed(1)}`)\n      if (opts.judges.contentQuality) notes.push(`content=${contentComposite.toFixed(1)}`)\n      if (allJudgesFailed) notes.push('all-judges-failed')\n      const judgeRuns = [conversationRun, ...codeReviewRuns, ...contentReviewRuns]\n      const judgeCostUsd = judgeRuns.reduce((sum, run) => sum + (run.cost.usd ?? 0), 0)\n      if (judgeRuns.some((run) => run.cost.kind === 'uncaptured')) {\n        notes.push('judge-cost-incomplete')\n      }\n\n      return {\n        output: {\n          turns: sim.transcript.length,\n          toolCalls: sim.toolCalls,\n          artifactCount: sim.artifacts.length,\n        },\n        verdict: { valid: composite >= 5, score: composite, notes: notes.join(' ') },\n        costUsd: sim.costUsd + judgeCostUsd,\n        durationMs: sim.durationMs,\n      }\n    },\n  })\n\n  // Persist top-level summary.\n  const summary = {\n    cells: matrix.summary.totalCells,\n    passRate: matrix.summary.overallPassRate,\n    meanScore: matrix.summary.overallMeanScore,\n    totalCostUsd: matrix.summary.totalCostUsd,\n    durationMs: matrix.summary.durationMs,\n    runsExecuted: matrix.summary.runsExecuted,\n    cellsSkipped: matrix.summary.cellsSkipped,\n    byProfile: matrix.byAxis.profile,\n    byPersona: matrix.byAxis.persona,\n  }\n  writeFileSync(join(opts.runDir, 'summary.json'), JSON.stringify(summary, null, 2))\n\n  const md: string[] = [\n    `# Multishot matrix`,\n    ``,\n    `**Cells**: ${matrix.summary.totalCells} | **Pass rate**: ${(matrix.summary.overallPassRate * 100).toFixed(0)}% | **Mean**: ${matrix.summary.overallMeanScore.toFixed(2)} | **Cost**: $${matrix.summary.totalCostUsd.toFixed(2)} | **Duration**: ${(matrix.summary.durationMs / 1000).toFixed(0)}s`,\n    ``,\n    `## By profile`,\n    ``,\n    '| profile | pass | mean | cost |',\n    '|---|---|---|---|',\n    ...Object.entries(matrix.byAxis.profile ?? {}).map(\n      ([id, s]) =>\n        `| ${id} | ${(s.passRate * 100).toFixed(0)}% | ${s.meanScore.toFixed(2)} | $${s.totalCostUsd.toFixed(2)} |`,\n    ),\n    ``,\n    `## By persona`,\n    ``,\n    '| persona | pass | mean | cost |',\n    '|---|---|---|---|',\n    ...Object.entries(matrix.byAxis.persona ?? {}).map(\n      ([id, s]) =>\n        `| ${id} | ${(s.passRate * 100).toFixed(0)}% | ${s.meanScore.toFixed(2)} | $${s.totalCostUsd.toFixed(2)} |`,\n    ),\n    ``,\n  ]\n  writeFileSync(join(opts.runDir, 'summary.md'), md.join('\\n'))\n\n  return { matrix }\n}\n\nfunction withJudgeMaxTokens<TInput>(\n  judge: JudgeConfig<TInput>,\n  maxTokens: number | undefined,\n): JudgeConfig<TInput> {\n  if (maxTokens === undefined || judge.maxTokens !== undefined) return judge\n  return { ...judge, maxTokens }\n}\n"],"mappings":";;;;;AAiCA,eAAsB,iBACpB,KACmC;CACnC,MAAM,YAAY,KAAK,IAAI;CAC3B,MAAM,OAAgC;EACpC,OAAO,IAAI;EACX,UAAU,IAAI;EACd,aAAa,IAAI,eAAe;EAChC,YAAY,IAAI,aAAa;CAC/B;CACA,IAAI,IAAI,OAAO,QAAQ,KAAK,QAAQ,IAAI;CACxC,MAAM,MAAM,GAAG,IAAI,QAAQ,QAAQ,QAAQ,EAAE,EAAE;CAC/C,MAAM,MAAM,MAAM,MAAM,KAAK;EAC3B,QAAQ;EACR,SAAS;GAAE,eAAe,UAAU,IAAI;GAAU,gBAAgB;EAAmB;EACrF,MAAM,KAAK,UAAU,IAAI;EACzB,QAAQ,IAAI;CACd,CAAC;CACD,IAAI,CAAC,IAAI,IAAI;EACX,MAAM,OAAO,MAAM,IAAI,KAAK;EAC5B,MAAM,IAAI,MAAM,UAAU,IAAI,OAAO,IAAI,KAAK,MAAM,GAAG,GAAG,GAAG;CAC/D;CACA,MAAM,OAAQ,MAAM,IAAI,KAAK;CAO7B,MAAM,SAAS,KAAK,QAAQ;CAC5B,IAAI,CAAC,QAAQ,MAAM,IAAI,MAAM,+BAA+B,KAAK,UAAU,IAAI,CAAC,CAAC,MAAM,GAAG,GAAG,GAAG;CAChG,MAAM,UAAU,KAAK,kBAAkB,KAAK;CAC5C,MAAM,UACJ,OAAO,YAAY,YAAY,OAAO,SAAS,OAAO,KAAK,WAAW,IAAI,UAAU,KAAA;CACtF,OAAO;EACL,SAAS,OAAO;EAChB,OAAO,KAAK;EACZ,GAAI,YAAY,KAAA,IAAY,CAAC,IAAI,EAAE,QAAQ;EAC3C,OAAO,OAAO,KAAK,UAAU,YAAY,KAAK,QAAQ,KAAK,QAAQ,IAAI;EACvE,YAAY,KAAK,IAAI,IAAI;CAC3B;AACF;AAIA,SAAgB,mBACd,OACA,OACQ;CACR,IAAI,CAAC,OAAO,OAAO;CACnB,MAAM,WAAW,MAAM,iBAAiB;CACxC,MAAM,YAAY,MAAM,qBAAqB;CAC7C,IAAI,UAAU;CACd,IAAI,WAAW;CACf,IAAI,MAAM,SAAS,aAAa,GAAG;EACjC,UAAU;EACV,WAAW;CACb,OAAO,IAAI,MAAM,SAAS,SAAS,KAAK,MAAM,SAAS,eAAe,GAAG;EACvE,UAAU;EACV,WAAW;CACb,OAAO,IAAI,MAAM,SAAS,MAAM,KAAK,MAAM,SAAS,KAAK,KAAK,MAAM,SAAS,UAAU,GAAG;EACxF,UAAU;EACV,WAAW;CACb;CACA,QAAQ,WAAW,UAAU,YAAY,YAAY;AACvD;AAEA,SAAgB,uBAA+B;CAC7C,QAAQ,QAAQ,IAAI,0BAA0B,iCAAA,CAAkC,QAC9E,QACA,EACF;AACF;AAEA,SAAgB,sBAA8B;CAC5C,MAAM,MAAM,QAAQ,IAAI;CACxB,IAAI,CAAC,KAAK,MAAM,IAAI,MAAM,gEAAgE;CAC1F,OAAO;AACT;;;ACrGA,MAAa,2BAA2B;AACxC,MAAa,sBAAsB;AAgBnC,MAAM,4BACJ;AAEF,MAAM,uBACJ;AAEF,MAAa,iCAA0D;CACrE,MAAM;CACN,UAAU;EACR,MAAM;EACN,aACE;EACF,YAAY;GACV,MAAM;GACN,YAAY;IACV,UAAU;KAAE,MAAM;KAAU,aAAa;IAAgC;IACzE,OAAO;KACL,MAAM;KACN,aAAa;IACf;GACF;GACA,UAAU,CAAC,UAAU;EACvB;CACF;AACF;AAEA,MAAa,6BAAsD;CACjE,MAAM;CACN,UAAU;EACR,MAAM;EACN,aACE;EACF,YAAY;GACV,MAAM;GACN,YAAY;IACV,MAAM;KAAE,MAAM;KAAU,aAAa;IAAgC;IACrE,UAAU;KACR,MAAM;KACN,aAAa;IACf;GACF;GACA,UAAU,CAAC,MAAM;EACnB;CACF;AACF;AAEA,SAAgB,uBACd,SAAkC,CAAC,GACZ;CACvB,MAAM,eAAe,OAAO,gBAAgB;CAC5C,MAAM,QAAQ,OAAO,SAAA;CACrB,OAAO,OAAO,MAAM,QAAQ;EAC1B,MAAM,WAAW,OAAO,KAAK,YAAY,EAAE;EAC3C,MAAM,QAAQ,KAAK,QAAQ,OAAO,KAAK,KAAK,IAAI,KAAA;EAChD,MAAM,EAAE,SAAS,UAAU,MAAM,iBAAiB;GAChD,QAAQ,IAAI;GACZ,SAAS,IAAI;GACb;GACA,aAAa;GACb,WAAW;GACX,UAAU,CACR;IAAE,MAAM;IAAU,SAAS;GAAa,GACxC;IAAE,MAAM;IAAQ,SAAS,aAAa,WAAW,QAAQ,YAAY,UAAU;GAAK,CACtF;GACA,QAAQ,IAAI;EACd,CAAC;EACD,OAAO;GAAE,SAAS,QAAQ,WAAW;GAAI,SAAS,mBAAmB,OAAO,KAAK;EAAE;CACrF;AACF;AAEA,SAAgB,mBAAmB,SAA6B,CAAC,GAA0B;CACzF,MAAM,eAAe,OAAO,gBAAgB;CAC5C,MAAM,QAAQ,OAAO,SAAA;CACrB,OAAO,OAAO,MAAM,QAAQ;EAC1B,MAAM,OAAO,OAAO,KAAK,QAAQ,EAAE;EACnC,MAAM,WAAW,KAAK,WAAW,OAAO,KAAK,QAAQ,IAAI;EACzD,MAAM,EAAE,SAAS,UAAU,MAAM,iBAAiB;GAChD,QAAQ,IAAI;GACZ,SAAS,IAAI;GACb;GACA,aAAa;GACb,WAAW;GACX,UAAU,CACR;IAAE,MAAM;IAAU,SAAS,GAAG,aAAa,gBAAgB;GAAW,GACtE;IAAE,MAAM;IAAQ,SAAS,YAAY;GAAO,CAC9C;GACA,QAAQ,IAAI;EACd,CAAC;EACD,OAAO;GAAE,SAAS,QAAQ,WAAW;GAAI,SAAS,mBAAmB,OAAO,KAAK;EAAE;CACrF;AACF;AAgBA,SAAgB,uBAAuB,SAA6B,CAAC,GAAuB;CAC1F,OAAO;EACL,OAAO,CAAC,gCAAgC,0BAA0B;EAClE,WAAW;GACT,mBAAmB,uBAAuB,OAAO,QAAQ;GACzD,eAAe,mBAAmB,OAAO,IAAI;EAC/C;EACA,kBAAkB,SAChB,SAAS,sBAAsB,aAAa,SAAS,kBAAkB,SAAS,KAAA;CACpF;AACF;;;ACtHA,MAAa,sBAAsB;AAmCnC,eAAsB,SACpB,OACA,OACyB;CACzB,MAAM,SAAS,MAAM,UAAU,oBAAoB;CACnD,MAAM,UAAU,MAAM,WAAW,qBAAqB;CACtD,MAAM,QAAQ,MAAM,SAAS,QAAQ,IAAI,eAAA;CACzC,MAAM,SAAS,MAAM,YAAY,KAAK;CACtC,IAAI,MAAM;CACV,IAAI;CACJ,IAAI;CACJ,MAAM,YAAY,KAAK,IAAI;CAC3B,IAAI;EACF,MAAM,WAAW,MAAM,iBAAiB;GACtC;GACA;GACA;GACA,aAAa;GACb,WAAW,MAAM,aAAa;GAC9B,UAAU,CACR;IAAE,MAAM;IAAU,SAAS,MAAM;GAAa,GAC9C;IAAE,MAAM;IAAQ,SAAS;GAAO,CAClC;EACF,CAAC;EACD,MAAM,OAAO,kBAAkB,QAAQ;EACvC,UAAU,KAAK;EACf,OAAO,KAAK;EACZ,OAAO,SAAS,QAAQ,WAAW,GAAA,CAAI,KAAK;CAC9C,SAAS,KAAK;EAGZ,OAAO;GACL,OAAO;IACL,YAAY,CAAC;IACb,WAAW;IACX,QAAQ;IACR,OAAO,SAAS,MAAM,KAAK,gBAAgB,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;IAC1F,SAAS,iBAAiB,OAAO,KAAK,IAAI,IAAI,SAAS;GACzD;GACA,MAAM;IAAE,MAAM;IAAc,KAAK;GAAK;EACxC;CACF;CAEA,IAAI,SAAyC;CAC7C,IAAI;EACF,MAAM,UAAU,IACb,QAAQ,gBAAgB,EAAE,CAAC,CAC3B,QAAQ,WAAW,EAAE,CAAC,CACtB,KAAK;EACR,SAAS,KAAK,MAAM,OAAO;CAC7B,QAAQ;EACN,OAAO;GACL,OAAO;IACL,YAAY,CAAC;IACb,WAAW;IACX,QAAQ;IACR,OAAO,SAAS,MAAM,KAAK,sBAAsB,IAAI,MAAM,GAAG,GAAG;IACjE;GACF;GACA;EACF;CACF;CAEA,MAAM,aAAqC,CAAC;CAC5C,IAAI,MAAM;CACV,KAAK,MAAM,OAAO,MAAM,YAAY;EAClC,MAAM,IAAI,OAAO,OAAO,IAAI,QAAQ,CAAC;EACrC,MAAM,UAAU,OAAO,SAAS,CAAC,IAAI,KAAK,IAAI,GAAG,KAAK,IAAI,IAAI,CAAC,CAAC,IAAI;EACpE,WAAW,IAAI,OAAO;EACtB,OAAO;CACT;CACA,OAAO;EACL,OAAO;GACL;GACA,WAAW,MAAM,WAAW,WAAW,IAAI,IAAI,MAAM,MAAM,WAAW;GACtE,OAAO,OAAO,OAAO,UAAU,WAAW,OAAO,QAAQ;GACzD;EACF;EACA;CACF;AACF;AAEA,SAAS,kBAAkB,UAGzB;CACA,MAAM,QAAQ,eAAe,SAAS,KAAK;CAC3C,OAAO;EACL,SAAS;GACP;GACA,SAAS,SAAS,WAAW;GAC7B,OAAO,SAAS;GAChB,YAAY,SAAS;EACvB;EACA,MACE,SAAS,YAAY,KAAA,IACjB;GAAE,MAAM;GAAY,KAAK,SAAS;EAAQ,IAC1C,MAAM,aAAa,SAAS,CAAC,cAAc,SAAS,KAAK,IACvD;GAAE,MAAM;GAAc,KAAK;EAAK,IAChC;GACE,MAAM;GACN,KAAK,aAAa,MAAM,cAAc,MAAM,kBAAkB,SAAS,KAAK;EAC9E;CACV;AACF;AAEA,SAAS,eAAe,OAAoD;CAC1E,MAAM,eAAe,WAAW,OAAO,aAAa;CACpD,MAAM,mBAAmB,WAAW,OAAO,iBAAiB;CAC5D,MAAM,WAAW,iBAAiB,KAAA,KAAa,qBAAqB,KAAA;CACpE,OAAO;EACL,cAAc,gBAAgB;EAC9B,kBAAkB,oBAAoB;EACtC,cAAc,gBAAgB,MAAM,oBAAoB;EACxD;CACF;AACF;AAEA,SAAS,iBAAiB,OAAe,YAAqC;CAC5E,OAAO;EACL,OAAO;GAAE,cAAc;GAAG,kBAAkB;GAAG,aAAa;GAAG,UAAU;EAAM;EAC/E,SAAS;EACT;EACA;CACF;AACF;AAEA,SAAS,WAAW,OAAoC;CACtD,OAAO,OAAO,UAAU,YAAY,OAAO,cAAc,KAAK,KAAK,SAAS,IAAI,QAAQ,KAAA;AAC1F;;;AAIA,SAAgB,iBAAiB,MAAyC;CACxE,OAAO,KAAK,KAAK,MAAM,KAAK,EAAE,IAAI,IAAI,EAAE,aAAa,CAAC,CAAC,KAAK,IAAI;AAClE;;AAGA,SAAgB,iBAAiB,MAAyC;CAExE,OAAO,6DADQ,KAAK,KAAK,MAAM,IAAI,EAAE,IAAI,IAAI,CAAC,CAAC,KAAK,GACqB,EAAE;AAC7E;;;AChHA,IAAa,4BAAb,cAA+C,MAAM;CACvB;CAA5B,YAAY,MAA8B;EACxC,MAAM,0DAA0D,KAAK,gBAAgB;EAD3D,KAAA,OAAA;EAE1B,KAAK,OAAO;CACd;AACF;AAEA,IAAa,0BAAb,cAA6C,MAAM;CACjD,YAAY,SAAiB;EAC3B,MAAM,OAAO;EACb,KAAK,OAAO;CACd;AACF;;;AChCA,eAAsB,aACpB,MAC0B;CAC1B,MAAM,SAAS,KAAK,UAAU,oBAAoB;CAClD,MAAM,UAAU,KAAK,WAAW,qBAAqB;CACrD,MAAM,WAAW,KAAK,YAAY;CAClC,MAAM,oBAAoB,KAAK,qBAAqB;CACpD,MAAM,aAAa,KAAK,cAAc;CAEtC,MAAM,eAAe,CADD,KAAK,eAAe,sBACL,GAAI,KAAK,wBAAwB,CAAC,CAAE;CACvE,MAAM,iBAAiB,KAAK,kBAAkB;CAC9C,MAAM,wBAAwB,KAAK,yBAAyB;CAC5D,MAAM,kBAAkB,KAAK,mBAAmB;CAEhD,MAAM,SACJ,KAAK,SAAS,KAAK,gBACf;EACE,OAAO,KAAK;EACZ,WAAW,KAAK;EAChB,iBAAiB,KAAK,0BAA0B,KAAA;CAClD,IACA,uBAAuB;CAC7B,MAAM,QAAQ,KAAK,SAAS,OAAO;CACnC,MAAM,YAAY,KAAK,iBAAiB,OAAO;CAC/C,MAAM,kBAAkB,KAAK,mBAAmB,OAAO;CAEvD,MAAM,mBAAuC,QAAQ,iBAAiB;EAAE;EAAQ;EAAS,GAAG;CAAI,CAAC;CACjG,MAAM,iBAAiB,KAAK,kBAAkB;CAC9C,MAAM,kBAAkB,KAAK,mBAAmB;CAEhD,MAAM,QAAQ,KAAK,IAAI;CACvB,MAAM,aAAiC,CAAC;CACxC,MAAM,YAAiC,CAAC;CACxC,IAAI,YAAY;CAChB,IAAI,eAAe;CAEnB,MAAM,SAAS,KAAK,MAAM,YAAY,KAAK,OAAO;CAClD,WAAW,KAAK;EAAE,MAAM;EAAQ,SAAS;CAAO,CAAC;CAGjD,MAAM,gBAAgD,CACpD;EAAE,MAAM;EAAU,SAFC,KAAK,QAAQ,QAAQ,gBAAgB;CAEhB,GACxC;EAAE,MAAM;EAAQ,SAAS;CAAO,CAClC;CAEA,KAAK,IAAI,OAAO,GAAG,OAAO,UAAU,QAAQ;EAC1C,IAAI,KAAK,QAAQ,SAAS,MAAM,IAAI,MAAM,mBAAmB;EAE7D,IAAI,qBAAqB;EACzB,OAAO,MAAM;GACX,MAAM,EACJ,SAAS,UACT,OAAO,YACP,SAAS,iBACP,MAAM,eAAe;IACvB,OAAO;IACP,UAAU;IACV;IACA,aAAa;IACb,WAAW,uBAAuB,IAAI,iBAAiB;IACvD,QAAQ,KAAK;GACf,CAAC;GACD,gBAAgB,gBAAgB,mBAAmB,YAAY,UAAU;GAEzE,MAAM,aAAa,SAAS,WAAW,GAAA,CAAI,KAAK;GAChD,MAAM,kBAAkB,SAAS,cAAc,CAAC,EAAA,CAAG,KAAK,QAAQ;IAC9D,IAAI,GAAG;IACP,MAAM,GAAG,SAAS;IAClB,aAAa;KACX,IAAI;MACF,OAAO,KAAK,MAAM,GAAG,SAAS,SAAS;KACzC,QAAQ;MACN,OAAO,CAAC;KACV;IACF,EAAA,CAAG;GACL,EAAE;GAEF,cAAc,KAAK;IACjB,MAAM;IACN,SAAS,aAAa;IACtB,GAAI,SAAS,YAAY,SAAS,EAAE,YAAY,SAAS,WAAW,IAAI,CAAC;GAC3E,CAAC;GACD,WAAW,KAAK;IACd,MAAM;IACN,SAAS;IACT,WAAW,eAAe,SAAS,IAAI,iBAAiB,KAAA;GAC1D,CAAC;GAED,IAAI,eAAe,WAAW,GAAG;GACjC,sBAAsB,eAAe;GACrC,IAAI,qBAAqB,mBACvB,MAAM,IAAI,MACR,0CAA0C,mBAAmB,GAAG,kBAAkB,YAAY,MAChG;GAGF,KAAK,MAAM,MAAM,gBAAgB;IAC/B;IACA,IAAI,aAAa;IACjB,IAAI;KACF,MAAM,WAAW,UAAU,GAAG;KAC9B,IAAI,CAAC,UACH,aAAa,KAAK,UAAU,EAAE,OAAO,gBAAgB,GAAG,OAAO,CAAC;UAC3D;MACL,MAAM,IAAI,MAAM,SAAS,GAAG,MAAM;OAAE;OAAQ;OAAS,QAAQ,KAAK;MAAO,CAAC;MAC1E,aAAa,EAAE;MACf,gBAAgB,EAAE;MAClB,MAAM,eAAe,gBAAgB,GAAG,IAAI;MAC5C,IAAI,cACF,UAAU,KAAK;OACb,MAAM;OACN;OACA,YAAY;QAAE,MAAM,GAAG;QAAM,MAAM,GAAG;OAAK;OAC3C,SAAS;MACX,CAAC;KAEL;IACF,SAAS,KAAK;KACZ,IAAI,eAAe,yBAAyB,MAAM;KAClD,aAAa,KAAK,UAAU,EAAE,OAAO,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG,EAAE,CAAC;IACzF;IACA,cAAc,KAAK;KAAE,MAAM;KAAQ,cAAc,GAAG;KAAI,SAAS,cAAc;IAAO,CAAC;IACvF,WAAW,KAAK;KAAE,MAAM;KAAQ,SAAS,cAAc;KAAQ,YAAY,GAAG;IAAG,CAAC;GACpF;EACF;EAEA,IAAI,OAAO,WAAW,GAAG;GACvB,MAAM,SAAS,MAAM,WAAW;IAC9B,WAAW;IACX,SAAS,KAAK;IACd,OAAO,KAAK;IACZ;IACA;IACA,QAAQ;IACR,WAAW;IACX,QAAQ,KAAK;GACf,CAAC;GACD,gBAAgB,OAAO;GACvB,cAAc,KAAK;IAAE,MAAM;IAAQ,SAAS,OAAO;GAAQ,CAAC;GAC5D,WAAW,KAAK;IAAE,MAAM;IAAQ,SAAS,OAAO;GAAQ,CAAC;EAC3D;CACF;CAEA,OAAO;EAAE;EAAY;EAAW;EAAW,YAAY,KAAK,IAAI,IAAI;EAAO,SAAS;CAAa;AACnG;AAEA,eAAe,WAA8C,MASX;CAMhD,MAAM,iBAAiD,CAAC;EAAE,MAAM;EAAU,SALrD,KAAK,MAAM,wBAAwB,KAAK,OAKiC;CAAE,CAAC;CACjG,KAAK,MAAM,OAAO,KAAK,YAAY;EACjC,IAAI,IAAI,SAAS,QAAQ;EACzB,MAAM,UAAU,qBAAqB,GAAG;EACxC,IAAI,CAAC,SAAS;EACd,IAAI,IAAI,SAAS,aAAa,eAAe,KAAK;GAAE,MAAM;GAAQ;EAAQ,CAAC;OACtE,IAAI,IAAI,SAAS,QAAQ,eAAe,KAAK;GAAE,MAAM;GAAa;EAAQ,CAAC;CAClF;CAGA,KAAK,MAAM,SAAS,KAAK,QACvB,KAAK,IAAI,UAAU,GAAG,UAAU,GAAG,WAAW;EAC5C,MAAM,EAAE,SAAS,OAAO,YAAY,MAAM,KAAK,UAAU;GACvD;GACA,UAAU;GACV,aAAa;GACb,WAAW,KAAK;GAChB,QAAQ,KAAK;EACf,CAAC;EACD,MAAM,WAAW,QAAQ,WAAW,GAAA,CAAI,KAAK;EAC7C,IAAI,QAAQ,SAAS,GACnB,OAAO;GAAE;GAAS,SAAS,WAAW,mBAAmB,OAAO,KAAK;EAAE;CAC3E;CAEF,MAAM,IAAI,0BAA0B,KAAK,IAAI;AAC/C;AAEA,SAAS,qBAAqB,KAAsC;CAClE,MAAM,OAAO,IAAI,QAAQ,KAAK;CAC9B,IAAI,KAAK,SAAS,GAAG,OAAO;CAC5B,IAAI,IAAI,SAAS,eAAe,CAAC,IAAI,WAAW,QAAQ,OAAO;CAE/D,MAAM,YAAY,IAAI,UAAU,KAAK,SAAS,KAAK,KAAK,KAAK,CAAC,CAAC,CAAC,OAAO,OAAO;CAC9E,IAAI,UAAU,WAAW,GAAG,OAAO;CACnC,OAAO,gBAAgB,UAAU,WAAW,IAAI,SAAS,QAAQ,IAAI,UAAU,KAAK,IAAI,EAAE;AAC5F;;;;;;;;ACvIA,SAAS,6BAA6B,QAAkD;CACtF,IAAI,OAAO,WAAW,GAAG,OAAO;CAChC,MAAM,OAAO,OAAO,QAAQ,MAAM,CAAC,EAAE,MAAM;CAC3C,IAAI,KAAK,WAAW,GAAG,OAAO;CAC9B,OAAO,KAAK,QAAQ,KAAK,MAAM,MAAM,EAAE,WAAW,CAAC,IAAI,KAAK;AAC9D;;;;;;AAeA,SAAgB,qBAAqB,OAKnC;CACA,MAAM,gBAA0B,CAAC;CACjC,IAAI,CAAC,MAAM,aAAa,QAAQ,cAAc,KAAK,MAAM,aAAa,SAAS;CAE/E,MAAM,WAAW,MAAM,cAAc,6BAA6B,MAAM,WAAW,IAAI,KAAA;CACvF,IAAI,OAAO,aAAa,UAAU,cAAc,KAAK,QAAQ;CAC7D,MAAM,cAAc,MAAM,iBACtB,6BAA6B,MAAM,cAAc,IACjD,KAAA;CACJ,IAAI,OAAO,gBAAgB,UAAU,cAAc,KAAK,WAAW;CAEnE,OAAO;EACL,WACE,cAAc,WAAW,IACrB,IACA,cAAc,QAAQ,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,cAAc;EAC/D,eAAe,YAAY;EAC3B,kBAAkB,eAAe;EACjC,iBAAiB,cAAc,WAAW;CAC5C;AACF;AAMA,eAAsB,mBACpB,MACmC;CACnC,MAAM,YAAY,IAAI,IAAI,KAAK,OAAO,qBAAqB,CAAC,MAAM,CAAC;CACnE,MAAM,eAAe,IAAI,IAAI,KAAK,OAAO,wBAAwB,CAAC,UAAU,CAAC;CAC7E,UAAU,KAAK,QAAQ,EAAE,WAAW,KAAK,CAAC;CAE1C,MAAM,SAAS,MAAM,eAA2B;EAC9C,MAAM,CACJ;GAAE,MAAM;GAAW,QAAQ,KAAK;EAAS,GACzC;GAAE,MAAM;GAAW,QAAQ,KAAK,SAAS,KAAK,OAAO;IAAE,IAAI,EAAE;IAAI,OAAO;GAAE,EAAE;EAAE,CAChF;EACA,MAAM,KAAK,QAAQ;EACnB,gBAAgB,KAAK,kBAAkB;EACvC,aAAa,KAAK;EAClB,MAAM,QAAQ,MAAM;GAClB,MAAM,UAAU,KAAK,KAAK,SAAS;GACnC,MAAM,UAAU,KAAK,KAAK,SAAS;GACnC,MAAM,YAAY,OAAO,KAAK,KAAK,SAAS,MAAM,SAAS;GAC3D,MAAM,YAAY,OAAO,KAAK,KAAK,SAAS,MAAM,SAAS;GAE3D,MAAM,MAAM,MAAM,aAAa;IAC7B;IACA;IACA,OAAO,KAAK;IACZ,OAAO,KAAK;IACZ,eAAe,KAAK;IACpB,iBAAiB,KAAK;IACtB,UAAU,KAAK;IACf,mBAAmB,KAAK;IACxB,YAAY,KAAK;IACjB,aAAa,KAAK;IAClB,sBAAsB,KAAK;IAC3B,gBAAgB,KAAK;IACrB,uBAAuB,KAAK;IAC5B,iBAAiB,KAAK;IACtB,gBAAgB,KAAK;IACrB,iBAAiB,KAAK;IACtB,QAAQ,KAAK;IACb,SAAS,KAAK;GAChB,CAAC;GAED,MAAM,gBAAgB,IAAI,UAAU,QAAQ,MAAM,UAAU,IAAI,EAAE,IAAI,CAAC;GACvE,MAAM,mBAAmB,IAAI,UAAU,QAAQ,MAAM,aAAa,IAAI,EAAE,IAAI,CAAC;GAE7E,MAAM,CAAC,iBAAiB,gBAAgB,qBAAqB,MAAM,QAAQ,IAAI;IAC7E,SAAS,mBAAmB,KAAK,OAAO,cAAc,KAAK,cAAc,GAAG;KAC1E,YAAY,IAAI;KAChB;IACF,CAAC;IACD,KAAK,OAAO,aACR,QAAQ,IACN,cAAc,KAAK,aACjB,SAAS,mBAAmB,KAAK,OAAO,YAAa,KAAK,cAAc,GAAG;KACzE;KACA;IACF,CAAC,CAAC,CAAC,MAAM,YAAY;KACnB,OAAO;MACL,GAAG,OAAO;MACV,MAAM,SAAS;MACf,MAAM,SAAS;KACjB;KACA,MAAM,OAAO;IACf,EAAE,CACJ,CACF,IACA,QAAQ,QAAQ,CAAC,CAAuB;IAC5C,KAAK,OAAO,iBACR,QAAQ,IACN,iBAAiB,KAAK,aACpB,SAAS,mBAAmB,KAAK,OAAO,gBAAiB,KAAK,cAAc,GAAG;KAC7E;KACA;IACF,CAAC,CAAC,CAAC,MAAM,YAAY;KACnB,OAAO;MACL,GAAG,OAAO;MACV,MAAM,SAAS;MACf,MAAM,SAAS;KACjB;KACA,MAAM,OAAO;IACf,EAAE,CACJ,CACF,IACA,QAAQ,QAAQ,CAAC,CAAuB;GAC9C,CAAC;GACD,MAAM,eAAe,gBAAgB;GACrC,MAAM,cAAc,eAAe,KAAK,QAAQ,IAAI,KAAK;GACzD,MAAM,iBAAiB,kBAAkB,KAAK,QAAQ,IAAI,KAAK;GAE/D,MAAM,EAAE,WAAW,eAAe,kBAAkB,oBAAoB,qBAAqB;IAC3F;IACA,aAAa,KAAK,OAAO,aAAa,cAAc,KAAA;IACpD,gBAAgB,KAAK,OAAO,iBAAiB,iBAAiB,KAAA;GAChE,CAAC;GAED,MAAM,YAAgC;IAAE;IAAW;GAAa;GAChE,IAAI,KAAK,OAAO,YACd,UAAU,aAAa;IAAE,aAAa;IAAa,WAAW;GAAc;GAC9E,IAAI,KAAK,OAAO,gBACd,UAAU,iBAAiB;IAAE,aAAa;IAAgB,WAAW;GAAiB;GAExF,MAAM,UAAU,KAAK,KAAK,QAAQ,WAAW,WAAW,OAAO,KAAK,KAAK;GACzE,UAAU,SAAS,EAAE,WAAW,KAAK,CAAC;GACtC,cAAc,KAAK,SAAS,iBAAiB,GAAG,KAAK,UAAU,IAAI,YAAY,MAAM,CAAC,CAAC;GACvF,cAAc,KAAK,SAAS,gBAAgB,GAAG,KAAK,UAAU,IAAI,WAAW,MAAM,CAAC,CAAC;GACrF,cAAc,KAAK,SAAS,aAAa,GAAG,KAAK,UAAU,WAAW,MAAM,CAAC,CAAC;GAE9E,MAAM,QAAQ,CAAC,SAAS,aAAa,UAAU,QAAQ,CAAC,GAAG;GAC3D,IAAI,KAAK,OAAO,YAAY,MAAM,KAAK,QAAQ,cAAc,QAAQ,CAAC,GAAG;GACzE,IAAI,KAAK,OAAO,gBAAgB,MAAM,KAAK,WAAW,iBAAiB,QAAQ,CAAC,GAAG;GACnF,IAAI,iBAAiB,MAAM,KAAK,mBAAmB;GACnD,MAAM,YAAY;IAAC;IAAiB,GAAG;IAAgB,GAAG;GAAiB;GAC3E,MAAM,eAAe,UAAU,QAAQ,KAAK,QAAQ,OAAO,IAAI,KAAK,OAAO,IAAI,CAAC;GAChF,IAAI,UAAU,MAAM,QAAQ,IAAI,KAAK,SAAS,YAAY,GACxD,MAAM,KAAK,uBAAuB;GAGpC,OAAO;IACL,QAAQ;KACN,OAAO,IAAI,WAAW;KACtB,WAAW,IAAI;KACf,eAAe,IAAI,UAAU;IAC/B;IACA,SAAS;KAAE,OAAO,aAAa;KAAG,OAAO;KAAW,OAAO,MAAM,KAAK,GAAG;IAAE;IAC3E,SAAS,IAAI,UAAU;IACvB,YAAY,IAAI;GAClB;EACF;CACF,CAAC;CAGD,MAAM,UAAU;EACd,OAAO,OAAO,QAAQ;EACtB,UAAU,OAAO,QAAQ;EACzB,WAAW,OAAO,QAAQ;EAC1B,cAAc,OAAO,QAAQ;EAC7B,YAAY,OAAO,QAAQ;EAC3B,cAAc,OAAO,QAAQ;EAC7B,cAAc,OAAO,QAAQ;EAC7B,WAAW,OAAO,OAAO;EACzB,WAAW,OAAO,OAAO;CAC3B;CACA,cAAc,KAAK,KAAK,QAAQ,cAAc,GAAG,KAAK,UAAU,SAAS,MAAM,CAAC,CAAC;CAEjF,MAAM,KAAe;EACnB;EACA;EACA,cAAc,OAAO,QAAQ,WAAW,qBAAqB,OAAO,QAAQ,kBAAkB,IAAA,CAAK,QAAQ,CAAC,EAAE,gBAAgB,OAAO,QAAQ,iBAAiB,QAAQ,CAAC,EAAE,gBAAgB,OAAO,QAAQ,aAAa,QAAQ,CAAC,EAAE,oBAAoB,OAAO,QAAQ,aAAa,IAAA,CAAM,QAAQ,CAAC,EAAE;EACjS;EACA;EACA;EACA;EACA;EACA,GAAG,OAAO,QAAQ,OAAO,OAAO,WAAW,CAAC,CAAC,CAAC,CAAC,KAC5C,CAAC,IAAI,OACJ,KAAK,GAAG,MAAM,EAAE,WAAW,IAAA,CAAK,QAAQ,CAAC,EAAE,MAAM,EAAE,UAAU,QAAQ,CAAC,EAAE,MAAM,EAAE,aAAa,QAAQ,CAAC,EAAE,GAC5G;EACA;EACA;EACA;EACA;EACA;EACA,GAAG,OAAO,QAAQ,OAAO,OAAO,WAAW,CAAC,CAAC,CAAC,CAAC,KAC5C,CAAC,IAAI,OACJ,KAAK,GAAG,MAAM,EAAE,WAAW,IAAA,CAAK,QAAQ,CAAC,EAAE,MAAM,EAAE,UAAU,QAAQ,CAAC,EAAE,MAAM,EAAE,aAAa,QAAQ,CAAC,EAAE,GAC5G;EACA;CACF;CACA,cAAc,KAAK,KAAK,QAAQ,YAAY,GAAG,GAAG,KAAK,IAAI,CAAC;CAE5D,OAAO,EAAE,OAAO;AAClB;AAEA,SAAS,mBACP,OACA,WACqB;CACrB,IAAI,cAAc,KAAA,KAAa,MAAM,cAAc,KAAA,GAAW,OAAO;CACrE,OAAO;EAAE,GAAG;EAAO;CAAU;AAC/B"}

package/dist/openapi.json

@@ -2,7 +2,7 @@
   "openapi": "3.1.0",
   "info": {
     "title": "@tangle-network/agent-eval — wire protocol",
-    "version": "0.134.1",
+    "version": "0.134.2",
     "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/rl.js

@@ -7,7 +7,7 @@ import { t as isSplitEligible } from "./exporters-q9iL-2Jf.js";
 import { t as mintRolloutRows } from "./mint-BvkwcYZU.js";
 import { a as InMemoryTraceStore } from "./integrity-BzRbCHzi.js";
 import { c as campaignCellToRunRecord, i as filterDeterministicallyRewarded, n as extractVerifiableReward, r as extractVerifiableRewardsFromRecords, t as detectRewardHacking } from "./reward-hacking-DCdRK9TY.js";
-import { t as runEvalCampaign } from "./eval-campaign-CvPcvqXC.js";
+import { t as runEvalCampaign } from "./eval-campaign-CHFxPTVl.js";
 import { t as evaluateInterimReleaseConfidence } from "./sequential-Br0mAPHA.js";
 import { t as rubricPredictiveValidity } from "./rubric-predictive-validity-D6Q6n9oq.js";
 import { n as thompsonCurriculum, r as varianceBasedCurriculum, t as observationsFromRunRecords } from "./active-curriculum-C4mk67HP.js";

package/dist/{semantic-concept-judge-C0P1VTXD.js → semantic-concept-judge-Ca--u10C.js}

@@ -1,6 +1,6 @@
 import { a as NotFoundError } from "./errors-8YnH8WlF.js";
-import { i as CostLedger } from "./cost-ledger-BrJxbrMy.js";
-import { c as callLlmJson, f as maximumChargeForLlmRequest, l as costReceiptFromLlm, u as costReceiptFromLlmError } from "./llm-client-ClPW-dWB.js";
+import { i as CostLedger } from "./cost-ledger-ZAa_P4r0.js";
+import { c as callLlmJson, f as maximumChargeForLlmRequest, l as costReceiptFromLlm, u as costReceiptFromLlmError } from "./llm-client-BNcP4v08.js";
 import { a as clamp01, i as aggregateRunScore, s as computeFindingId } from "./proposal-findings-DCawte-y.js";
 import { t as Mutex } from "./concurrency-MUjT7VjM.js";
 import { appendFileSync, existsSync, mkdirSync, readFileSync, readdirSync, statSync } from "node:fs";
@@ -721,4 +721,4 @@ function createSemanticConceptJudge(options = {}) {
 //#endregion
 export { RunCritic as a, buildSkillUsageReport as c, defaultIsMaterial as d, diffFindings as f, runSemanticConceptJudge as i, emitSkillUsageFindings as l, SEMANTIC_CONCEPT_JUDGE_VERSION as n, SKILL_USAGE_ANALYST as o, LockedJsonlAppender as p, createSemanticConceptJudge as r, SkillUsageAnalyst as s, DEFAULT_COMPLEXITY_WEIGHTS as t, FindingsStore as u };
 
-//# sourceMappingURL=semantic-concept-judge-C0P1VTXD.js.map
+//# sourceMappingURL=semantic-concept-judge-Ca--u10C.js.map