@tangle-network/agent-eval 0.96.0 → 0.96.2

package/dist/multishot/index.js.map

@@ -1 +1 @@
-{"version":3,"sources":["../../src/multishot/router.ts","../../src/multishot/default-tools.ts","../../src/multishot/judges.ts","../../src/multishot/matrix.ts","../../src/multishot/types.ts","../../src/multishot/multishot.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 legacy 0-10 scale (scale is producer-defined; see the type doc).\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}\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: 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","// 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} 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  /** 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  /** 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        agentModel: opts.agentModel,\n        driverModel: opts.driverModel,\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(opts.judges.conversation, { transcript: sim.transcript, persona }),\n        opts.judges.codeReview\n          ? Promise.all(\n              codeArtifacts.map((artifact) =>\n                runJudge(opts.judges.codeReview!, { artifact, persona }).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(opts.judges.contentQuality!, { artifact, persona }).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","// 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\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","// 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 with profile's\n// 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  type MultishotMessage,\n  type MultishotPersona,\n  type MultishotResult,\n  type MultishotShape,\n  type MultishotToolDefinition,\n  type MultishotToolExecutor,\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  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 agentModel = opts.agentModel ?? 'openai/gpt-5.4'\n  const driverModel = opts.driverModel ?? 'openai/gpt-4o-mini'\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 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    const { message: agentMsg, usage: agentUsage } = await routerCompletion({\n      apiKey,\n      baseUrl,\n      model: agentModel,\n      messages: agentMessages,\n      tools,\n      temperature: 0.7,\n      maxTokens: 2500,\n      signal: opts.signal,\n    })\n    totalCostUsd += 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    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        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    // If the agent emitted tool_calls, give it a follow-up turn to integrate the results.\n    if (agentToolCalls.length > 0) {\n      const followUp = await routerCompletion({\n        apiKey,\n        baseUrl,\n        model: agentModel,\n        messages: agentMessages,\n        temperature: 0.7,\n        maxTokens: 2000,\n        signal: opts.signal,\n      })\n      totalCostUsd += estimateRouterCost(agentModel, followUp.usage)\n      const followUpText = (followUp.message.content ?? '').trim()\n      agentMessages.push({ role: 'assistant', content: followUpText })\n      transcript.push({ role: 'assistant', content: followUpText })\n    }\n\n    if (turn < maxTurns - 1) {\n      const driver = await driverTurn({\n        apiKey,\n        baseUrl,\n        persona: opts.persona,\n        shape: opts.shape,\n        transcript,\n        turn,\n        model: driverModel,\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  apiKey: string\n  baseUrl: string\n  persona: TPersona\n  shape: MultishotShape<TPersona>\n  transcript: MultishotMessage[]\n  turn: number\n  model: string\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    if (msg.role === 'assistant') driverMessages.push({ role: 'user', content: msg.content })\n    else if (msg.role === 'user') driverMessages.push({ role: 'assistant', content: msg.content })\n  }\n\n  // Driver must never go silent. Retry once on empty content; then fail loud.\n  for (let attempt = 0; attempt < 2; attempt++) {\n    const { message, usage } = await routerCompletion({\n      apiKey: opts.apiKey,\n      baseUrl: opts.baseUrl,\n      model: opts.model,\n      messages: driverMessages,\n      temperature: 0.9,\n      maxTokens: 600,\n      signal: opts.signal,\n    })\n    const content = (message.content ?? '').trim()\n    if (content.length > 0) return { content, costUsd: estimateRouterCost(opts.model, usage) }\n  }\n  throw new MultishotDriverEmptyError(opts.turn)\n}\n"],"mappings":";;;;;;AA4BA,eAAsB,iBACpB,KACmC;AACnC,QAAM,OAAgC;AAAA,IACpC,OAAO,IAAI;AAAA,IACX,UAAU,IAAI;AAAA,IACd,aAAa,IAAI,eAAe;AAAA,IAChC,YAAY,IAAI,aAAa;AAAA,EAC/B;AACA,MAAI,IAAI,OAAO,OAAQ,MAAK,QAAQ,IAAI;AACxC,QAAM,MAAM,GAAG,IAAI,QAAQ,QAAQ,QAAQ,EAAE,CAAC;AAC9C,QAAM,MAAM,MAAM,MAAM,KAAK;AAAA,IAC3B,QAAQ;AAAA,IACR,SAAS,EAAE,eAAe,UAAU,IAAI,MAAM,IAAI,gBAAgB,mBAAmB;AAAA,IACrF,MAAM,KAAK,UAAU,IAAI;AAAA,IACzB,QAAQ,IAAI;AAAA,EACd,CAAC;AACD,MAAI,CAAC,IAAI,IAAI;AACX,UAAM,OAAO,MAAM,IAAI,KAAK;AAC5B,UAAM,IAAI,MAAM,UAAU,IAAI,MAAM,KAAK,KAAK,MAAM,GAAG,GAAG,CAAC,EAAE;AAAA,EAC/D;AACA,QAAM,OAAQ,MAAM,IAAI,KAAK;AAI7B,QAAM,SAAS,KAAK,QAAQ,CAAC;AAC7B,MAAI,CAAC,OAAQ,OAAM,IAAI,MAAM,+BAA+B,KAAK,UAAU,IAAI,EAAE,MAAM,GAAG,GAAG,CAAC,EAAE;AAChG,SAAO,EAAE,SAAS,OAAO,SAAS,OAAO,KAAK,MAAM;AACtD;AAIO,SAAS,mBACd,OACA,OACQ;AACR,MAAI,CAAC,MAAO,QAAO;AACnB,QAAM,WAAW,MAAM,iBAAiB;AACxC,QAAM,YAAY,MAAM,qBAAqB;AAC7C,MAAI,UAAU;AACd,MAAI,WAAW;AACf,MAAI,MAAM,SAAS,aAAa,GAAG;AACjC,cAAU;AACV,eAAW;AAAA,EACb,WAAW,MAAM,SAAS,SAAS,KAAK,MAAM,SAAS,eAAe,GAAG;AACvE,cAAU;AACV,eAAW;AAAA,EACb,WAAW,MAAM,SAAS,MAAM,KAAK,MAAM,SAAS,KAAK,KAAK,MAAM,SAAS,UAAU,GAAG;AACxF,cAAU;AACV,eAAW;AAAA,EACb;AACA,UAAQ,WAAW,UAAU,YAAY,YAAY;AACvD;AAEO,SAAS,uBAA+B;AAC7C,UAAQ,QAAQ,IAAI,0BAA0B,kCAAkC;AAAA,IAC9E;AAAA,IACA;AAAA,EACF;AACF;AAEO,SAAS,sBAA8B;AAC5C,QAAM,MAAM,QAAQ,IAAI;AACxB,MAAI,CAAC,IAAK,OAAM,IAAI,MAAM,gEAAgE;AAC1F,SAAO;AACT;;;ACnFO,IAAM,2BAA2B;AACjC,IAAM,sBAAsB;AAgBnC,IAAM,4BACJ;AAEF,IAAM,uBACJ;AAEK,IAAM,iCAA0D;AAAA,EACrE,MAAM;AAAA,EACN,UAAU;AAAA,IACR,MAAM;AAAA,IACN,aACE;AAAA,IACF,YAAY;AAAA,MACV,MAAM;AAAA,MACN,YAAY;AAAA,QACV,UAAU,EAAE,MAAM,UAAU,aAAa,gCAAgC;AAAA,QACzE,OAAO;AAAA,UACL,MAAM;AAAA,UACN,aAAa;AAAA,QACf;AAAA,MACF;AAAA,MACA,UAAU,CAAC,UAAU;AAAA,IACvB;AAAA,EACF;AACF;AAEO,IAAM,6BAAsD;AAAA,EACjE,MAAM;AAAA,EACN,UAAU;AAAA,IACR,MAAM;AAAA,IACN,aACE;AAAA,IACF,YAAY;AAAA,MACV,MAAM;AAAA,MACN,YAAY;AAAA,QACV,MAAM,EAAE,MAAM,UAAU,aAAa,gCAAgC;AAAA,QACrE,UAAU;AAAA,UACR,MAAM;AAAA,UACN,aAAa;AAAA,QACf;AAAA,MACF;AAAA,MACA,UAAU,CAAC,MAAM;AAAA,IACnB;AAAA,EACF;AACF;AAEO,SAAS,uBACd,SAAkC,CAAC,GACZ;AACvB,QAAM,eAAe,OAAO,gBAAgB;AAC5C,QAAM,QAAQ,OAAO,SAAS;AAC9B,SAAO,OAAO,MAAM,QAAQ;AAC1B,UAAM,WAAW,OAAO,KAAK,YAAY,EAAE;AAC3C,UAAM,QAAQ,KAAK,QAAQ,OAAO,KAAK,KAAK,IAAI;AAChD,UAAM,EAAE,SAAS,MAAM,IAAI,MAAM,iBAAiB;AAAA,MAChD,QAAQ,IAAI;AAAA,MACZ,SAAS,IAAI;AAAA,MACb;AAAA,MACA,aAAa;AAAA,MACb,WAAW;AAAA,MACX,UAAU;AAAA,QACR,EAAE,MAAM,UAAU,SAAS,aAAa;AAAA,QACxC,EAAE,MAAM,QAAQ,SAAS,aAAa,QAAQ,GAAG,QAAQ;AAAA,SAAY,KAAK,KAAK,EAAE,GAAG;AAAA,MACtF;AAAA,MACA,QAAQ,IAAI;AAAA,IACd,CAAC;AACD,WAAO,EAAE,SAAS,QAAQ,WAAW,IAAI,SAAS,mBAAmB,OAAO,KAAK,EAAE;AAAA,EACrF;AACF;AAEO,SAAS,mBAAmB,SAA6B,CAAC,GAA0B;AACzF,QAAM,eAAe,OAAO,gBAAgB;AAC5C,QAAM,QAAQ,OAAO,SAAS;AAC9B,SAAO,OAAO,MAAM,QAAQ;AAC1B,UAAM,OAAO,OAAO,KAAK,QAAQ,EAAE;AACnC,UAAM,WAAW,KAAK,WAAW,OAAO,KAAK,QAAQ,IAAI;AACzD,UAAM,EAAE,SAAS,MAAM,IAAI,MAAM,iBAAiB;AAAA,MAChD,QAAQ,IAAI;AAAA,MACZ,SAAS,IAAI;AAAA,MACb;AAAA,MACA,aAAa;AAAA,MACb,WAAW;AAAA,MACX,UAAU;AAAA,QACR,EAAE,MAAM,UAAU,SAAS,GAAG,YAAY;AAAA;AAAA,YAAiB,QAAQ,GAAG;AAAA,QACtE,EAAE,MAAM,QAAQ,SAAS,YAAY,IAAI,GAAG;AAAA,MAC9C;AAAA,MACA,QAAQ,IAAI;AAAA,IACd,CAAC;AACD,WAAO,EAAE,SAAS,QAAQ,WAAW,IAAI,SAAS,mBAAmB,OAAO,KAAK,EAAE;AAAA,EACrF;AACF;AAgBO,SAAS,uBAAuB,SAA6B,CAAC,GAAuB;AAC1F,SAAO;AAAA,IACL,OAAO,CAAC,gCAAgC,0BAA0B;AAAA,IAClE,WAAW;AAAA,MACT,mBAAmB,uBAAuB,OAAO,QAAQ;AAAA,MACzD,eAAe,mBAAmB,OAAO,IAAI;AAAA,IAC/C;AAAA,IACA,iBAAiB,CAAC,SAChB,SAAS,sBAAsB,aAAa,SAAS,kBAAkB,SAAS;AAAA,EACpF;AACF;;;AC/HO,IAAM,sBAAsB;AA0BnC,eAAsB,SACpB,OACA,OACqB;AACrB,QAAM,SAAS,MAAM,UAAU,oBAAoB;AACnD,QAAM,UAAU,MAAM,WAAW,qBAAqB;AACtD,QAAM,QAAQ,MAAM,SAAS,QAAQ,IAAI,eAAe;AACxD,QAAM,SAAS,MAAM,YAAY,KAAK;AACtC,MAAI,MAAM;AACV,MAAI;AACF,UAAM,EAAE,QAAQ,IAAI,MAAM,iBAAiB;AAAA,MACzC;AAAA,MACA;AAAA,MACA;AAAA,MACA,aAAa;AAAA,MACb,WAAW;AAAA,MACX,UAAU;AAAA,QACR,EAAE,MAAM,UAAU,SAAS,MAAM,aAAa;AAAA,QAC9C,EAAE,MAAM,QAAQ,SAAS,OAAO;AAAA,MAClC;AAAA,IACF,CAAC;AACD,WAAO,QAAQ,WAAW,IAAI,KAAK;AAAA,EACrC,SAAS,KAAK;AAGZ,WAAO;AAAA,MACL,YAAY,CAAC;AAAA,MACb,WAAW;AAAA,MACX,QAAQ;AAAA,MACR,OAAO,SAAS,MAAM,IAAI,iBAAiB,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG,CAAC;AAAA,IAC7F;AAAA,EACF;AAEA,MAAI,SAAyC;AAC7C,MAAI;AACF,UAAM,UAAU,IACb,QAAQ,gBAAgB,EAAE,EAC1B,QAAQ,WAAW,EAAE,EACrB,KAAK;AACR,aAAS,KAAK,MAAM,OAAO;AAAA,EAC7B,QAAQ;AACN,WAAO;AAAA,MACL,YAAY,CAAC;AAAA,MACb,WAAW;AAAA,MACX,QAAQ;AAAA,MACR,OAAO,SAAS,MAAM,IAAI,uBAAuB,IAAI,MAAM,GAAG,GAAG,CAAC;AAAA,IACpE;AAAA,EACF;AAEA,QAAM,aAAqC,CAAC;AAC5C,MAAI,MAAM;AACV,aAAW,OAAO,MAAM,YAAY;AAClC,UAAM,IAAI,OAAO,OAAO,IAAI,GAAG,KAAK,CAAC;AACrC,UAAM,UAAU,OAAO,SAAS,CAAC,IAAI,KAAK,IAAI,GAAG,KAAK,IAAI,IAAI,CAAC,CAAC,IAAI;AACpE,eAAW,IAAI,GAAG,IAAI;AACtB,WAAO;AAAA,EACT;AACA,SAAO;AAAA,IACL;AAAA,IACA,WAAW,MAAM,WAAW,WAAW,IAAI,IAAI,MAAM,MAAM,WAAW;AAAA,IACtE,OAAO,OAAO,OAAO,UAAU,WAAW,OAAO,QAAQ;AAAA,EAC3D;AACF;AAIO,SAAS,iBAAiB,MAAyC;AACxE,SAAO,KAAK,IAAI,CAAC,MAAM,KAAK,EAAE,GAAG,KAAK,EAAE,WAAW,EAAE,EAAE,KAAK,IAAI;AAClE;AAGO,SAAS,iBAAiB,MAAyC;AACxE,QAAM,SAAS,KAAK,IAAI,CAAC,MAAM,IAAI,EAAE,GAAG,KAAK,EAAE,KAAK,GAAG;AACvD,SAAO;AAAA,GAA6D,MAAM;AAC5E;;;AC5GA,SAAS,WAAW,qBAAqB;AACzC,SAAS,YAAY;;;AC4Cd,IAAM,4BAAN,cAAwC,MAAM;AAAA,EACnD,YAA4B,MAAc;AACxC,UAAM,0DAA0D,IAAI,sBAAiB;AAD3D;AAE1B,SAAK,OAAO;AAAA,EACd;AAAA,EAH4B;AAI9B;;;ACXA,eAAsB,aACpB,MAC0B;AAC1B,QAAM,SAAS,KAAK,UAAU,oBAAoB;AAClD,QAAM,UAAU,KAAK,WAAW,qBAAqB;AACrD,QAAM,WAAW,KAAK,YAAY;AAClC,QAAM,aAAa,KAAK,cAAc;AACtC,QAAM,cAAc,KAAK,eAAe;AAExC,QAAM,SACJ,KAAK,SAAS,KAAK,gBACf;AAAA,IACE,OAAO,KAAK;AAAA,IACZ,WAAW,KAAK;AAAA,IAChB,iBAAiB,KAAK,oBAAoB,MAAM;AAAA,EAClD,IACA,uBAAuB;AAC7B,QAAM,QAAQ,KAAK,SAAS,OAAO;AACnC,QAAM,YAAY,KAAK,iBAAiB,OAAO;AAC/C,QAAM,kBAAkB,KAAK,mBAAmB,OAAO;AAEvD,QAAM,QAAQ,KAAK,IAAI;AACvB,QAAM,aAAiC,CAAC;AACxC,QAAM,YAAiC,CAAC;AACxC,MAAI,YAAY;AAChB,MAAI,eAAe;AAEnB,QAAM,SAAS,KAAK,MAAM,YAAY,KAAK,OAAO;AAClD,aAAW,KAAK,EAAE,MAAM,QAAQ,SAAS,OAAO,CAAC;AAEjD,QAAM,eAAe,KAAK,QAAQ,QAAQ,gBAAgB;AAC1D,QAAM,gBAAgD;AAAA,IACpD,EAAE,MAAM,UAAU,SAAS,aAAa;AAAA,IACxC,EAAE,MAAM,QAAQ,SAAS,OAAO;AAAA,EAClC;AAEA,WAAS,OAAO,GAAG,OAAO,UAAU,QAAQ;AAC1C,QAAI,KAAK,QAAQ,QAAS,OAAM,IAAI,MAAM,mBAAmB;AAE7D,UAAM,EAAE,SAAS,UAAU,OAAO,WAAW,IAAI,MAAM,iBAAiB;AAAA,MACtE;AAAA,MACA;AAAA,MACA,OAAO;AAAA,MACP,UAAU;AAAA,MACV;AAAA,MACA,aAAa;AAAA,MACb,WAAW;AAAA,MACX,QAAQ,KAAK;AAAA,IACf,CAAC;AACD,oBAAgB,mBAAmB,YAAY,UAAU;AAEzD,UAAM,aAAa,SAAS,WAAW,IAAI,KAAK;AAChD,UAAM,kBAAkB,SAAS,cAAc,CAAC,GAAG,IAAI,CAAC,QAAQ;AAAA,MAC9D,IAAI,GAAG;AAAA,MACP,MAAM,GAAG,SAAS;AAAA,MAClB,OAAO,MAAM;AACX,YAAI;AACF,iBAAO,KAAK,MAAM,GAAG,SAAS,SAAS;AAAA,QACzC,QAAQ;AACN,iBAAO,CAAC;AAAA,QACV;AAAA,MACF,GAAG;AAAA,IACL,EAAE;AAEF,kBAAc,KAAK;AAAA,MACjB,MAAM;AAAA,MACN,SAAS,aAAa;AAAA,MACtB,GAAI,SAAS,YAAY,SAAS,EAAE,YAAY,SAAS,WAAW,IAAI,CAAC;AAAA,IAC3E,CAAC;AACD,eAAW,KAAK;AAAA,MACd,MAAM;AAAA,MACN,SAAS;AAAA,MACT,WAAW,eAAe,SAAS,IAAI,iBAAiB;AAAA,IAC1D,CAAC;AAED,eAAW,MAAM,gBAAgB;AAC/B;AACA,UAAI,aAAa;AACjB,UAAI;AACF,cAAM,WAAW,UAAU,GAAG,IAAI;AAClC,YAAI,CAAC,UAAU;AACb,uBAAa,KAAK,UAAU,EAAE,OAAO,gBAAgB,GAAG,IAAI,GAAG,CAAC;AAAA,QAClE,OAAO;AACL,gBAAM,IAAI,MAAM,SAAS,GAAG,MAAM,EAAE,QAAQ,SAAS,QAAQ,KAAK,OAAO,CAAC;AAC1E,uBAAa,EAAE;AACf,0BAAgB,EAAE;AAClB,gBAAM,eAAe,gBAAgB,GAAG,IAAI;AAC5C,cAAI,cAAc;AAChB,sBAAU,KAAK;AAAA,cACb,MAAM;AAAA,cACN;AAAA,cACA,YAAY,EAAE,MAAM,GAAG,MAAM,MAAM,GAAG,KAAK;AAAA,cAC3C,SAAS;AAAA,YACX,CAAC;AAAA,UACH;AAAA,QACF;AAAA,MACF,SAAS,KAAK;AACZ,qBAAa,KAAK,UAAU,EAAE,OAAO,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG,EAAE,CAAC;AAAA,MACzF;AACA,oBAAc,KAAK,EAAE,MAAM,QAAQ,cAAc,GAAG,IAAI,SAAS,cAAc,OAAO,CAAC;AACvF,iBAAW,KAAK,EAAE,MAAM,QAAQ,SAAS,cAAc,QAAQ,YAAY,GAAG,GAAG,CAAC;AAAA,IACpF;AAGA,QAAI,eAAe,SAAS,GAAG;AAC7B,YAAM,WAAW,MAAM,iBAAiB;AAAA,QACtC;AAAA,QACA;AAAA,QACA,OAAO;AAAA,QACP,UAAU;AAAA,QACV,aAAa;AAAA,QACb,WAAW;AAAA,QACX,QAAQ,KAAK;AAAA,MACf,CAAC;AACD,sBAAgB,mBAAmB,YAAY,SAAS,KAAK;AAC7D,YAAM,gBAAgB,SAAS,QAAQ,WAAW,IAAI,KAAK;AAC3D,oBAAc,KAAK,EAAE,MAAM,aAAa,SAAS,aAAa,CAAC;AAC/D,iBAAW,KAAK,EAAE,MAAM,aAAa,SAAS,aAAa,CAAC;AAAA,IAC9D;AAEA,QAAI,OAAO,WAAW,GAAG;AACvB,YAAM,SAAS,MAAM,WAAW;AAAA,QAC9B;AAAA,QACA;AAAA,QACA,SAAS,KAAK;AAAA,QACd,OAAO,KAAK;AAAA,QACZ;AAAA,QACA;AAAA,QACA,OAAO;AAAA,QACP,QAAQ,KAAK;AAAA,MACf,CAAC;AACD,sBAAgB,OAAO;AACvB,oBAAc,KAAK,EAAE,MAAM,QAAQ,SAAS,OAAO,QAAQ,CAAC;AAC5D,iBAAW,KAAK,EAAE,MAAM,QAAQ,SAAS,OAAO,QAAQ,CAAC;AAAA,IAC3D;AAAA,EACF;AAEA,SAAO,EAAE,YAAY,WAAW,WAAW,YAAY,KAAK,IAAI,IAAI,OAAO,SAAS,aAAa;AACnG;AAEA,eAAe,WAA8C,MASX;AAChD,QAAM,eAAe,KAAK,MAAM,wBAAwB,KAAK,OAAO;AAKpE,QAAM,iBAAiD,CAAC,EAAE,MAAM,UAAU,SAAS,aAAa,CAAC;AACjG,aAAW,OAAO,KAAK,YAAY;AACjC,QAAI,IAAI,SAAS,OAAQ;AACzB,QAAI,IAAI,SAAS,YAAa,gBAAe,KAAK,EAAE,MAAM,QAAQ,SAAS,IAAI,QAAQ,CAAC;AAAA,aAC/E,IAAI,SAAS,OAAQ,gBAAe,KAAK,EAAE,MAAM,aAAa,SAAS,IAAI,QAAQ,CAAC;AAAA,EAC/F;AAGA,WAAS,UAAU,GAAG,UAAU,GAAG,WAAW;AAC5C,UAAM,EAAE,SAAS,MAAM,IAAI,MAAM,iBAAiB;AAAA,MAChD,QAAQ,KAAK;AAAA,MACb,SAAS,KAAK;AAAA,MACd,OAAO,KAAK;AAAA,MACZ,UAAU;AAAA,MACV,aAAa;AAAA,MACb,WAAW;AAAA,MACX,QAAQ,KAAK;AAAA,IACf,CAAC;AACD,UAAM,WAAW,QAAQ,WAAW,IAAI,KAAK;AAC7C,QAAI,QAAQ,SAAS,EAAG,QAAO,EAAE,SAAS,SAAS,mBAAmB,KAAK,OAAO,KAAK,EAAE;AAAA,EAC3F;AACA,QAAM,IAAI,0BAA0B,KAAK,IAAI;AAC/C;;;AFvHA,SAAS,6BAA6B,QAAkD;AACtF,MAAI,OAAO,WAAW,EAAG,QAAO;AAChC,QAAM,OAAO,OAAO,OAAO,CAAC,MAAM,CAAC,EAAE,MAAM;AAC3C,MAAI,KAAK,WAAW,EAAG,QAAO;AAC9B,SAAO,KAAK,OAAO,CAAC,KAAK,MAAM,MAAM,EAAE,WAAW,CAAC,IAAI,KAAK;AAC9D;AAeO,SAAS,qBAAqB,OAKnC;AACA,QAAM,gBAA0B,CAAC;AACjC,MAAI,CAAC,MAAM,aAAa,OAAQ,eAAc,KAAK,MAAM,aAAa,SAAS;AAE/E,QAAM,WAAW,MAAM,cAAc,6BAA6B,MAAM,WAAW,IAAI;AACvF,MAAI,OAAO,aAAa,SAAU,eAAc,KAAK,QAAQ;AAC7D,QAAM,cAAc,MAAM,iBACtB,6BAA6B,MAAM,cAAc,IACjD;AACJ,MAAI,OAAO,gBAAgB,SAAU,eAAc,KAAK,WAAW;AAEnE,SAAO;AAAA,IACL,WACE,cAAc,WAAW,IACrB,IACA,cAAc,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,cAAc;AAAA,IAC/D,eAAe,YAAY;AAAA,IAC3B,kBAAkB,eAAe;AAAA,IACjC,iBAAiB,cAAc,WAAW;AAAA,EAC5C;AACF;AAMA,eAAsB,mBACpB,MACmC;AACnC,QAAM,YAAY,IAAI,IAAI,KAAK,OAAO,qBAAqB,CAAC,MAAM,CAAC;AACnE,QAAM,eAAe,IAAI,IAAI,KAAK,OAAO,wBAAwB,CAAC,UAAU,CAAC;AAC7E,YAAU,KAAK,QAAQ,EAAE,WAAW,KAAK,CAAC;AAE1C,QAAM,SAAS,MAAM,eAA2B;AAAA,IAC9C,MAAM;AAAA,MACJ,EAAE,MAAM,WAAW,QAAQ,KAAK,SAAS;AAAA,MACzC,EAAE,MAAM,WAAW,QAAQ,KAAK,SAAS,IAAI,CAAC,OAAO,EAAE,IAAI,EAAE,IAAI,OAAO,EAAE,EAAE,EAAE;AAAA,IAChF;AAAA,IACA,MAAM,KAAK,QAAQ;AAAA,IACnB,gBAAgB,KAAK,kBAAkB;AAAA,IACvC,aAAa,KAAK;AAAA,IAClB,MAAM,QAAQ,MAAM;AAClB,YAAM,UAAU,KAAK,KAAK,SAAS;AACnC,YAAM,UAAU,KAAK,KAAK,SAAS;AACnC,YAAM,YAAY,OAAO,KAAK,KAAK,SAAS,MAAM,SAAS;AAC3D,YAAM,YAAY,OAAO,KAAK,KAAK,SAAS,MAAM,SAAS;AAE3D,YAAM,MAAM,MAAM,aAAa;AAAA,QAC7B;AAAA,QACA;AAAA,QACA,OAAO,KAAK;AAAA,QACZ,OAAO,KAAK;AAAA,QACZ,eAAe,KAAK;AAAA,QACpB,iBAAiB,KAAK;AAAA,QACtB,UAAU,KAAK;AAAA,QACf,YAAY,KAAK;AAAA,QACjB,aAAa,KAAK;AAAA,QAClB,QAAQ,KAAK;AAAA,QACb,SAAS,KAAK;AAAA,MAChB,CAAC;AAED,YAAM,gBAAgB,IAAI,UAAU,OAAO,CAAC,MAAM,UAAU,IAAI,EAAE,IAAI,CAAC;AACvE,YAAM,mBAAmB,IAAI,UAAU,OAAO,CAAC,MAAM,aAAa,IAAI,EAAE,IAAI,CAAC;AAE7E,YAAM,CAAC,cAAc,aAAa,cAAc,IAAI,MAAM,QAAQ,IAAI;AAAA,QACpE,SAAS,KAAK,OAAO,cAAc,EAAE,YAAY,IAAI,YAAY,QAAQ,CAAC;AAAA,QAC1E,KAAK,OAAO,aACR,QAAQ;AAAA,UACN,cAAc;AAAA,YAAI,CAAC,aACjB,SAAS,KAAK,OAAO,YAAa,EAAE,UAAU,QAAQ,CAAC,EAAE,KAAK,CAAC,OAAO;AAAA,cACpE,GAAG;AAAA,cACH,MAAM,SAAS;AAAA,cACf,MAAM,SAAS;AAAA,YACjB,EAAE;AAAA,UACJ;AAAA,QACF,IACA,QAAQ,QAAQ,CAAC,CAAuD;AAAA,QAC5E,KAAK,OAAO,iBACR,QAAQ;AAAA,UACN,iBAAiB;AAAA,YAAI,CAAC,aACpB,SAAS,KAAK,OAAO,gBAAiB,EAAE,UAAU,QAAQ,CAAC,EAAE,KAAK,CAAC,OAAO;AAAA,cACxE,GAAG;AAAA,cACH,MAAM,SAAS;AAAA,cACf,MAAM,SAAS;AAAA,YACjB,EAAE;AAAA,UACJ;AAAA,QACF,IACA,QAAQ,QAAQ,CAAC,CAAuD;AAAA,MAC9E,CAAC;AAED,YAAM,EAAE,WAAW,eAAe,kBAAkB,gBAAgB,IAAI,qBAAqB;AAAA,QAC3F;AAAA,QACA,aAAa,KAAK,OAAO,aAAa,cAAc;AAAA,QACpD,gBAAgB,KAAK,OAAO,iBAAiB,iBAAiB;AAAA,MAChE,CAAC;AAED,YAAM,YAAgC,EAAE,WAAW,aAAa;AAChE,UAAI,KAAK,OAAO;AACd,kBAAU,aAAa,EAAE,aAAa,aAAa,WAAW,cAAc;AAC9E,UAAI,KAAK,OAAO;AACd,kBAAU,iBAAiB,EAAE,aAAa,gBAAgB,WAAW,iBAAiB;AAExF,YAAM,UAAU,KAAK,KAAK,QAAQ,WAAW,WAAW,OAAO,KAAK,GAAG,EAAE;AACzE,gBAAU,SAAS,EAAE,WAAW,KAAK,CAAC;AACtC,oBAAc,KAAK,SAAS,iBAAiB,GAAG,KAAK,UAAU,IAAI,YAAY,MAAM,CAAC,CAAC;AACvF,oBAAc,KAAK,SAAS,gBAAgB,GAAG,KAAK,UAAU,IAAI,WAAW,MAAM,CAAC,CAAC;AACrF,oBAAc,KAAK,SAAS,aAAa,GAAG,KAAK,UAAU,WAAW,MAAM,CAAC,CAAC;AAE9E,YAAM,QAAQ,CAAC,SAAS,aAAa,UAAU,QAAQ,CAAC,CAAC,EAAE;AAC3D,UAAI,KAAK,OAAO,WAAY,OAAM,KAAK,QAAQ,cAAc,QAAQ,CAAC,CAAC,EAAE;AACzE,UAAI,KAAK,OAAO,eAAgB,OAAM,KAAK,WAAW,iBAAiB,QAAQ,CAAC,CAAC,EAAE;AACnF,UAAI,gBAAiB,OAAM,KAAK,mBAAmB;AAEnD,aAAO;AAAA,QACL,QAAQ;AAAA,UACN,OAAO,IAAI,WAAW;AAAA,UACtB,WAAW,IAAI;AAAA,UACf,eAAe,IAAI,UAAU;AAAA,QAC/B;AAAA,QACA,SAAS,EAAE,OAAO,aAAa,GAAG,OAAO,WAAW,OAAO,MAAM,KAAK,GAAG,EAAE;AAAA,QAC3E,SAAS,IAAI;AAAA,QACb,YAAY,IAAI;AAAA,MAClB;AAAA,IACF;AAAA,EACF,CAAC;AAGD,QAAM,UAAU;AAAA,IACd,OAAO,OAAO,QAAQ;AAAA,IACtB,UAAU,OAAO,QAAQ;AAAA,IACzB,WAAW,OAAO,QAAQ;AAAA,IAC1B,cAAc,OAAO,QAAQ;AAAA,IAC7B,YAAY,OAAO,QAAQ;AAAA,IAC3B,cAAc,OAAO,QAAQ;AAAA,IAC7B,cAAc,OAAO,QAAQ;AAAA,IAC7B,WAAW,OAAO,OAAO;AAAA,IACzB,WAAW,OAAO,OAAO;AAAA,EAC3B;AACA,gBAAc,KAAK,KAAK,QAAQ,cAAc,GAAG,KAAK,UAAU,SAAS,MAAM,CAAC,CAAC;AAEjF,QAAM,KAAe;AAAA,IACnB;AAAA,IACA;AAAA,IACA,cAAc,OAAO,QAAQ,UAAU,sBAAsB,OAAO,QAAQ,kBAAkB,KAAK,QAAQ,CAAC,CAAC,iBAAiB,OAAO,QAAQ,iBAAiB,QAAQ,CAAC,CAAC,iBAAiB,OAAO,QAAQ,aAAa,QAAQ,CAAC,CAAC,qBAAqB,OAAO,QAAQ,aAAa,KAAM,QAAQ,CAAC,CAAC;AAAA,IAChS;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA,GAAG,OAAO,QAAQ,OAAO,OAAO,WAAW,CAAC,CAAC,EAAE;AAAA,MAC7C,CAAC,CAAC,IAAI,CAAC,MACL,KAAK,EAAE,OAAO,EAAE,WAAW,KAAK,QAAQ,CAAC,CAAC,OAAO,EAAE,UAAU,QAAQ,CAAC,CAAC,OAAO,EAAE,aAAa,QAAQ,CAAC,CAAC;AAAA,IAC3G;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA,GAAG,OAAO,QAAQ,OAAO,OAAO,WAAW,CAAC,CAAC,EAAE;AAAA,MAC7C,CAAC,CAAC,IAAI,CAAC,MACL,KAAK,EAAE,OAAO,EAAE,WAAW,KAAK,QAAQ,CAAC,CAAC,OAAO,EAAE,UAAU,QAAQ,CAAC,CAAC,OAAO,EAAE,aAAa,QAAQ,CAAC,CAAC;AAAA,IAC3G;AAAA,IACA;AAAA,EACF;AACA,gBAAc,KAAK,KAAK,QAAQ,YAAY,GAAG,GAAG,KAAK,IAAI,CAAC;AAE5D,SAAO,EAAE,OAAO;AAClB;","names":[]}
+{"version":3,"sources":["../../src/multishot/router.ts","../../src/multishot/default-tools.ts","../../src/multishot/judges.ts","../../src/multishot/matrix.ts","../../src/multishot/types.ts","../../src/multishot/multishot.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 legacy 0-10 scale (scale is producer-defined; see the type doc).\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}\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: 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","// 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} 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  /** 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        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(opts.judges.conversation, { transcript: sim.transcript, persona }),\n        opts.judges.codeReview\n          ? Promise.all(\n              codeArtifacts.map((artifact) =>\n                runJudge(opts.judges.codeReview!, { artifact, persona }).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(opts.judges.contentQuality!, { artifact, persona }).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","// 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\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","// 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 with profile's\n// 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  type MultishotMessage,\n  type MultishotPersona,\n  type MultishotResult,\n  type MultishotShape,\n  type MultishotToolDefinition,\n  type MultishotToolExecutor,\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  /** Maximum tool calls the agent may dispatch inside one assistant turn. */\n  maxToolDispatches?: number\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\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 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 { message: agentMsg, usage: agentUsage } = await routerCompletion({\n        apiKey,\n        baseUrl,\n        model: agentModel,\n        messages: agentMessages,\n        tools,\n        temperature: 0.7,\n        maxTokens: dispatchesThisTurn === 0 ? 2500 : 2000,\n        signal: opts.signal,\n      })\n      totalCostUsd += 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          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        apiKey,\n        baseUrl,\n        persona: opts.persona,\n        shape: opts.shape,\n        transcript,\n        turn,\n        model: driverModel,\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  apiKey: string\n  baseUrl: string\n  persona: TPersona\n  shape: MultishotShape<TPersona>\n  transcript: MultishotMessage[]\n  turn: number\n  model: string\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    if (msg.role === 'assistant') driverMessages.push({ role: 'user', content: msg.content })\n    else if (msg.role === 'user') driverMessages.push({ role: 'assistant', content: msg.content })\n  }\n\n  // Driver must never go silent. Retry once on empty content; then fail loud.\n  for (let attempt = 0; attempt < 2; attempt++) {\n    const { message, usage } = await routerCompletion({\n      apiKey: opts.apiKey,\n      baseUrl: opts.baseUrl,\n      model: opts.model,\n      messages: driverMessages,\n      temperature: 0.9,\n      maxTokens: 600,\n      signal: opts.signal,\n    })\n    const content = (message.content ?? '').trim()\n    if (content.length > 0) return { content, costUsd: estimateRouterCost(opts.model, usage) }\n  }\n  throw new MultishotDriverEmptyError(opts.turn)\n}\n"],"mappings":";;;;;;AA4BA,eAAsB,iBACpB,KACmC;AACnC,QAAM,OAAgC;AAAA,IACpC,OAAO,IAAI;AAAA,IACX,UAAU,IAAI;AAAA,IACd,aAAa,IAAI,eAAe;AAAA,IAChC,YAAY,IAAI,aAAa;AAAA,EAC/B;AACA,MAAI,IAAI,OAAO,OAAQ,MAAK,QAAQ,IAAI;AACxC,QAAM,MAAM,GAAG,IAAI,QAAQ,QAAQ,QAAQ,EAAE,CAAC;AAC9C,QAAM,MAAM,MAAM,MAAM,KAAK;AAAA,IAC3B,QAAQ;AAAA,IACR,SAAS,EAAE,eAAe,UAAU,IAAI,MAAM,IAAI,gBAAgB,mBAAmB;AAAA,IACrF,MAAM,KAAK,UAAU,IAAI;AAAA,IACzB,QAAQ,IAAI;AAAA,EACd,CAAC;AACD,MAAI,CAAC,IAAI,IAAI;AACX,UAAM,OAAO,MAAM,IAAI,KAAK;AAC5B,UAAM,IAAI,MAAM,UAAU,IAAI,MAAM,KAAK,KAAK,MAAM,GAAG,GAAG,CAAC,EAAE;AAAA,EAC/D;AACA,QAAM,OAAQ,MAAM,IAAI,KAAK;AAI7B,QAAM,SAAS,KAAK,QAAQ,CAAC;AAC7B,MAAI,CAAC,OAAQ,OAAM,IAAI,MAAM,+BAA+B,KAAK,UAAU,IAAI,EAAE,MAAM,GAAG,GAAG,CAAC,EAAE;AAChG,SAAO,EAAE,SAAS,OAAO,SAAS,OAAO,KAAK,MAAM;AACtD;AAIO,SAAS,mBACd,OACA,OACQ;AACR,MAAI,CAAC,MAAO,QAAO;AACnB,QAAM,WAAW,MAAM,iBAAiB;AACxC,QAAM,YAAY,MAAM,qBAAqB;AAC7C,MAAI,UAAU;AACd,MAAI,WAAW;AACf,MAAI,MAAM,SAAS,aAAa,GAAG;AACjC,cAAU;AACV,eAAW;AAAA,EACb,WAAW,MAAM,SAAS,SAAS,KAAK,MAAM,SAAS,eAAe,GAAG;AACvE,cAAU;AACV,eAAW;AAAA,EACb,WAAW,MAAM,SAAS,MAAM,KAAK,MAAM,SAAS,KAAK,KAAK,MAAM,SAAS,UAAU,GAAG;AACxF,cAAU;AACV,eAAW;AAAA,EACb;AACA,UAAQ,WAAW,UAAU,YAAY,YAAY;AACvD;AAEO,SAAS,uBAA+B;AAC7C,UAAQ,QAAQ,IAAI,0BAA0B,kCAAkC;AAAA,IAC9E;AAAA,IACA;AAAA,EACF;AACF;AAEO,SAAS,sBAA8B;AAC5C,QAAM,MAAM,QAAQ,IAAI;AACxB,MAAI,CAAC,IAAK,OAAM,IAAI,MAAM,gEAAgE;AAC1F,SAAO;AACT;;;ACnFO,IAAM,2BAA2B;AACjC,IAAM,sBAAsB;AAgBnC,IAAM,4BACJ;AAEF,IAAM,uBACJ;AAEK,IAAM,iCAA0D;AAAA,EACrE,MAAM;AAAA,EACN,UAAU;AAAA,IACR,MAAM;AAAA,IACN,aACE;AAAA,IACF,YAAY;AAAA,MACV,MAAM;AAAA,MACN,YAAY;AAAA,QACV,UAAU,EAAE,MAAM,UAAU,aAAa,gCAAgC;AAAA,QACzE,OAAO;AAAA,UACL,MAAM;AAAA,UACN,aAAa;AAAA,QACf;AAAA,MACF;AAAA,MACA,UAAU,CAAC,UAAU;AAAA,IACvB;AAAA,EACF;AACF;AAEO,IAAM,6BAAsD;AAAA,EACjE,MAAM;AAAA,EACN,UAAU;AAAA,IACR,MAAM;AAAA,IACN,aACE;AAAA,IACF,YAAY;AAAA,MACV,MAAM;AAAA,MACN,YAAY;AAAA,QACV,MAAM,EAAE,MAAM,UAAU,aAAa,gCAAgC;AAAA,QACrE,UAAU;AAAA,UACR,MAAM;AAAA,UACN,aAAa;AAAA,QACf;AAAA,MACF;AAAA,MACA,UAAU,CAAC,MAAM;AAAA,IACnB;AAAA,EACF;AACF;AAEO,SAAS,uBACd,SAAkC,CAAC,GACZ;AACvB,QAAM,eAAe,OAAO,gBAAgB;AAC5C,QAAM,QAAQ,OAAO,SAAS;AAC9B,SAAO,OAAO,MAAM,QAAQ;AAC1B,UAAM,WAAW,OAAO,KAAK,YAAY,EAAE;AAC3C,UAAM,QAAQ,KAAK,QAAQ,OAAO,KAAK,KAAK,IAAI;AAChD,UAAM,EAAE,SAAS,MAAM,IAAI,MAAM,iBAAiB;AAAA,MAChD,QAAQ,IAAI;AAAA,MACZ,SAAS,IAAI;AAAA,MACb;AAAA,MACA,aAAa;AAAA,MACb,WAAW;AAAA,MACX,UAAU;AAAA,QACR,EAAE,MAAM,UAAU,SAAS,aAAa;AAAA,QACxC,EAAE,MAAM,QAAQ,SAAS,aAAa,QAAQ,GAAG,QAAQ;AAAA,SAAY,KAAK,KAAK,EAAE,GAAG;AAAA,MACtF;AAAA,MACA,QAAQ,IAAI;AAAA,IACd,CAAC;AACD,WAAO,EAAE,SAAS,QAAQ,WAAW,IAAI,SAAS,mBAAmB,OAAO,KAAK,EAAE;AAAA,EACrF;AACF;AAEO,SAAS,mBAAmB,SAA6B,CAAC,GAA0B;AACzF,QAAM,eAAe,OAAO,gBAAgB;AAC5C,QAAM,QAAQ,OAAO,SAAS;AAC9B,SAAO,OAAO,MAAM,QAAQ;AAC1B,UAAM,OAAO,OAAO,KAAK,QAAQ,EAAE;AACnC,UAAM,WAAW,KAAK,WAAW,OAAO,KAAK,QAAQ,IAAI;AACzD,UAAM,EAAE,SAAS,MAAM,IAAI,MAAM,iBAAiB;AAAA,MAChD,QAAQ,IAAI;AAAA,MACZ,SAAS,IAAI;AAAA,MACb;AAAA,MACA,aAAa;AAAA,MACb,WAAW;AAAA,MACX,UAAU;AAAA,QACR,EAAE,MAAM,UAAU,SAAS,GAAG,YAAY;AAAA;AAAA,YAAiB,QAAQ,GAAG;AAAA,QACtE,EAAE,MAAM,QAAQ,SAAS,YAAY,IAAI,GAAG;AAAA,MAC9C;AAAA,MACA,QAAQ,IAAI;AAAA,IACd,CAAC;AACD,WAAO,EAAE,SAAS,QAAQ,WAAW,IAAI,SAAS,mBAAmB,OAAO,KAAK,EAAE;AAAA,EACrF;AACF;AAgBO,SAAS,uBAAuB,SAA6B,CAAC,GAAuB;AAC1F,SAAO;AAAA,IACL,OAAO,CAAC,gCAAgC,0BAA0B;AAAA,IAClE,WAAW;AAAA,MACT,mBAAmB,uBAAuB,OAAO,QAAQ;AAAA,MACzD,eAAe,mBAAmB,OAAO,IAAI;AAAA,IAC/C;AAAA,IACA,iBAAiB,CAAC,SAChB,SAAS,sBAAsB,aAAa,SAAS,kBAAkB,SAAS;AAAA,EACpF;AACF;;;AC/HO,IAAM,sBAAsB;AA0BnC,eAAsB,SACpB,OACA,OACqB;AACrB,QAAM,SAAS,MAAM,UAAU,oBAAoB;AACnD,QAAM,UAAU,MAAM,WAAW,qBAAqB;AACtD,QAAM,QAAQ,MAAM,SAAS,QAAQ,IAAI,eAAe;AACxD,QAAM,SAAS,MAAM,YAAY,KAAK;AACtC,MAAI,MAAM;AACV,MAAI;AACF,UAAM,EAAE,QAAQ,IAAI,MAAM,iBAAiB;AAAA,MACzC;AAAA,MACA;AAAA,MACA;AAAA,MACA,aAAa;AAAA,MACb,WAAW;AAAA,MACX,UAAU;AAAA,QACR,EAAE,MAAM,UAAU,SAAS,MAAM,aAAa;AAAA,QAC9C,EAAE,MAAM,QAAQ,SAAS,OAAO;AAAA,MAClC;AAAA,IACF,CAAC;AACD,WAAO,QAAQ,WAAW,IAAI,KAAK;AAAA,EACrC,SAAS,KAAK;AAGZ,WAAO;AAAA,MACL,YAAY,CAAC;AAAA,MACb,WAAW;AAAA,MACX,QAAQ;AAAA,MACR,OAAO,SAAS,MAAM,IAAI,iBAAiB,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG,CAAC;AAAA,IAC7F;AAAA,EACF;AAEA,MAAI,SAAyC;AAC7C,MAAI;AACF,UAAM,UAAU,IACb,QAAQ,gBAAgB,EAAE,EAC1B,QAAQ,WAAW,EAAE,EACrB,KAAK;AACR,aAAS,KAAK,MAAM,OAAO;AAAA,EAC7B,QAAQ;AACN,WAAO;AAAA,MACL,YAAY,CAAC;AAAA,MACb,WAAW;AAAA,MACX,QAAQ;AAAA,MACR,OAAO,SAAS,MAAM,IAAI,uBAAuB,IAAI,MAAM,GAAG,GAAG,CAAC;AAAA,IACpE;AAAA,EACF;AAEA,QAAM,aAAqC,CAAC;AAC5C,MAAI,MAAM;AACV,aAAW,OAAO,MAAM,YAAY;AAClC,UAAM,IAAI,OAAO,OAAO,IAAI,GAAG,KAAK,CAAC;AACrC,UAAM,UAAU,OAAO,SAAS,CAAC,IAAI,KAAK,IAAI,GAAG,KAAK,IAAI,IAAI,CAAC,CAAC,IAAI;AACpE,eAAW,IAAI,GAAG,IAAI;AACtB,WAAO;AAAA,EACT;AACA,SAAO;AAAA,IACL;AAAA,IACA,WAAW,MAAM,WAAW,WAAW,IAAI,IAAI,MAAM,MAAM,WAAW;AAAA,IACtE,OAAO,OAAO,OAAO,UAAU,WAAW,OAAO,QAAQ;AAAA,EAC3D;AACF;AAIO,SAAS,iBAAiB,MAAyC;AACxE,SAAO,KAAK,IAAI,CAAC,MAAM,KAAK,EAAE,GAAG,KAAK,EAAE,WAAW,EAAE,EAAE,KAAK,IAAI;AAClE;AAGO,SAAS,iBAAiB,MAAyC;AACxE,QAAM,SAAS,KAAK,IAAI,CAAC,MAAM,IAAI,EAAE,GAAG,KAAK,EAAE,KAAK,GAAG;AACvD,SAAO;AAAA,GAA6D,MAAM;AAC5E;;;AC5GA,SAAS,WAAW,qBAAqB;AACzC,SAAS,YAAY;;;AC4Cd,IAAM,4BAAN,cAAwC,MAAM;AAAA,EACnD,YAA4B,MAAc;AACxC,UAAM,0DAA0D,IAAI,sBAAiB;AAD3D;AAE1B,SAAK,OAAO;AAAA,EACd;AAAA,EAH4B;AAI9B;;;ACTA,eAAsB,aACpB,MAC0B;AAC1B,QAAM,SAAS,KAAK,UAAU,oBAAoB;AAClD,QAAM,UAAU,KAAK,WAAW,qBAAqB;AACrD,QAAM,WAAW,KAAK,YAAY;AAClC,QAAM,oBAAoB,KAAK,qBAAqB;AACpD,QAAM,aAAa,KAAK,cAAc;AACtC,QAAM,cAAc,KAAK,eAAe;AAExC,QAAM,SACJ,KAAK,SAAS,KAAK,gBACf;AAAA,IACE,OAAO,KAAK;AAAA,IACZ,WAAW,KAAK;AAAA,IAChB,iBAAiB,KAAK,oBAAoB,MAAM;AAAA,EAClD,IACA,uBAAuB;AAC7B,QAAM,QAAQ,KAAK,SAAS,OAAO;AACnC,QAAM,YAAY,KAAK,iBAAiB,OAAO;AAC/C,QAAM,kBAAkB,KAAK,mBAAmB,OAAO;AAEvD,QAAM,QAAQ,KAAK,IAAI;AACvB,QAAM,aAAiC,CAAC;AACxC,QAAM,YAAiC,CAAC;AACxC,MAAI,YAAY;AAChB,MAAI,eAAe;AAEnB,QAAM,SAAS,KAAK,MAAM,YAAY,KAAK,OAAO;AAClD,aAAW,KAAK,EAAE,MAAM,QAAQ,SAAS,OAAO,CAAC;AAEjD,QAAM,eAAe,KAAK,QAAQ,QAAQ,gBAAgB;AAC1D,QAAM,gBAAgD;AAAA,IACpD,EAAE,MAAM,UAAU,SAAS,aAAa;AAAA,IACxC,EAAE,MAAM,QAAQ,SAAS,OAAO;AAAA,EAClC;AAEA,WAAS,OAAO,GAAG,OAAO,UAAU,QAAQ;AAC1C,QAAI,KAAK,QAAQ,QAAS,OAAM,IAAI,MAAM,mBAAmB;AAE7D,QAAI,qBAAqB;AACzB,WAAO,MAAM;AACX,YAAM,EAAE,SAAS,UAAU,OAAO,WAAW,IAAI,MAAM,iBAAiB;AAAA,QACtE;AAAA,QACA;AAAA,QACA,OAAO;AAAA,QACP,UAAU;AAAA,QACV;AAAA,QACA,aAAa;AAAA,QACb,WAAW,uBAAuB,IAAI,OAAO;AAAA,QAC7C,QAAQ,KAAK;AAAA,MACf,CAAC;AACD,sBAAgB,mBAAmB,YAAY,UAAU;AAEzD,YAAM,aAAa,SAAS,WAAW,IAAI,KAAK;AAChD,YAAM,kBAAkB,SAAS,cAAc,CAAC,GAAG,IAAI,CAAC,QAAQ;AAAA,QAC9D,IAAI,GAAG;AAAA,QACP,MAAM,GAAG,SAAS;AAAA,QAClB,OAAO,MAAM;AACX,cAAI;AACF,mBAAO,KAAK,MAAM,GAAG,SAAS,SAAS;AAAA,UACzC,QAAQ;AACN,mBAAO,CAAC;AAAA,UACV;AAAA,QACF,GAAG;AAAA,MACL,EAAE;AAEF,oBAAc,KAAK;AAAA,QACjB,MAAM;AAAA,QACN,SAAS,aAAa;AAAA,QACtB,GAAI,SAAS,YAAY,SAAS,EAAE,YAAY,SAAS,WAAW,IAAI,CAAC;AAAA,MAC3E,CAAC;AACD,iBAAW,KAAK;AAAA,QACd,MAAM;AAAA,QACN,SAAS;AAAA,QACT,WAAW,eAAe,SAAS,IAAI,iBAAiB;AAAA,MAC1D,CAAC;AAED,UAAI,eAAe,WAAW,EAAG;AACjC,4BAAsB,eAAe;AACrC,UAAI,qBAAqB,mBAAmB;AAC1C,cAAM,IAAI;AAAA,UACR,0CAA0C,kBAAkB,IAAI,iBAAiB,aAAa,IAAI;AAAA,QACpG;AAAA,MACF;AAEA,iBAAW,MAAM,gBAAgB;AAC/B;AACA,YAAI,aAAa;AACjB,YAAI;AACF,gBAAM,WAAW,UAAU,GAAG,IAAI;AAClC,cAAI,CAAC,UAAU;AACb,yBAAa,KAAK,UAAU,EAAE,OAAO,gBAAgB,GAAG,IAAI,GAAG,CAAC;AAAA,UAClE,OAAO;AACL,kBAAM,IAAI,MAAM,SAAS,GAAG,MAAM,EAAE,QAAQ,SAAS,QAAQ,KAAK,OAAO,CAAC;AAC1E,yBAAa,EAAE;AACf,4BAAgB,EAAE;AAClB,kBAAM,eAAe,gBAAgB,GAAG,IAAI;AAC5C,gBAAI,cAAc;AAChB,wBAAU,KAAK;AAAA,gBACb,MAAM;AAAA,gBACN;AAAA,gBACA,YAAY,EAAE,MAAM,GAAG,MAAM,MAAM,GAAG,KAAK;AAAA,gBAC3C,SAAS;AAAA,cACX,CAAC;AAAA,YACH;AAAA,UACF;AAAA,QACF,SAAS,KAAK;AACZ,uBAAa,KAAK,UAAU,EAAE,OAAO,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG,EAAE,CAAC;AAAA,QACzF;AACA,sBAAc,KAAK,EAAE,MAAM,QAAQ,cAAc,GAAG,IAAI,SAAS,cAAc,OAAO,CAAC;AACvF,mBAAW,KAAK,EAAE,MAAM,QAAQ,SAAS,cAAc,QAAQ,YAAY,GAAG,GAAG,CAAC;AAAA,MACpF;AAAA,IACF;AAEA,QAAI,OAAO,WAAW,GAAG;AACvB,YAAM,SAAS,MAAM,WAAW;AAAA,QAC9B;AAAA,QACA;AAAA,QACA,SAAS,KAAK;AAAA,QACd,OAAO,KAAK;AAAA,QACZ;AAAA,QACA;AAAA,QACA,OAAO;AAAA,QACP,QAAQ,KAAK;AAAA,MACf,CAAC;AACD,sBAAgB,OAAO;AACvB,oBAAc,KAAK,EAAE,MAAM,QAAQ,SAAS,OAAO,QAAQ,CAAC;AAC5D,iBAAW,KAAK,EAAE,MAAM,QAAQ,SAAS,OAAO,QAAQ,CAAC;AAAA,IAC3D;AAAA,EACF;AAEA,SAAO,EAAE,YAAY,WAAW,WAAW,YAAY,KAAK,IAAI,IAAI,OAAO,SAAS,aAAa;AACnG;AAEA,eAAe,WAA8C,MASX;AAChD,QAAM,eAAe,KAAK,MAAM,wBAAwB,KAAK,OAAO;AAKpE,QAAM,iBAAiD,CAAC,EAAE,MAAM,UAAU,SAAS,aAAa,CAAC;AACjG,aAAW,OAAO,KAAK,YAAY;AACjC,QAAI,IAAI,SAAS,OAAQ;AACzB,QAAI,IAAI,SAAS,YAAa,gBAAe,KAAK,EAAE,MAAM,QAAQ,SAAS,IAAI,QAAQ,CAAC;AAAA,aAC/E,IAAI,SAAS,OAAQ,gBAAe,KAAK,EAAE,MAAM,aAAa,SAAS,IAAI,QAAQ,CAAC;AAAA,EAC/F;AAGA,WAAS,UAAU,GAAG,UAAU,GAAG,WAAW;AAC5C,UAAM,EAAE,SAAS,MAAM,IAAI,MAAM,iBAAiB;AAAA,MAChD,QAAQ,KAAK;AAAA,MACb,SAAS,KAAK;AAAA,MACd,OAAO,KAAK;AAAA,MACZ,UAAU;AAAA,MACV,aAAa;AAAA,MACb,WAAW;AAAA,MACX,QAAQ,KAAK;AAAA,IACf,CAAC;AACD,UAAM,WAAW,QAAQ,WAAW,IAAI,KAAK;AAC7C,QAAI,QAAQ,SAAS,EAAG,QAAO,EAAE,SAAS,SAAS,mBAAmB,KAAK,OAAO,KAAK,EAAE;AAAA,EAC3F;AACA,QAAM,IAAI,0BAA0B,KAAK,IAAI;AAC/C;;;AFlHA,SAAS,6BAA6B,QAAkD;AACtF,MAAI,OAAO,WAAW,EAAG,QAAO;AAChC,QAAM,OAAO,OAAO,OAAO,CAAC,MAAM,CAAC,EAAE,MAAM;AAC3C,MAAI,KAAK,WAAW,EAAG,QAAO;AAC9B,SAAO,KAAK,OAAO,CAAC,KAAK,MAAM,MAAM,EAAE,WAAW,CAAC,IAAI,KAAK;AAC9D;AAeO,SAAS,qBAAqB,OAKnC;AACA,QAAM,gBAA0B,CAAC;AACjC,MAAI,CAAC,MAAM,aAAa,OAAQ,eAAc,KAAK,MAAM,aAAa,SAAS;AAE/E,QAAM,WAAW,MAAM,cAAc,6BAA6B,MAAM,WAAW,IAAI;AACvF,MAAI,OAAO,aAAa,SAAU,eAAc,KAAK,QAAQ;AAC7D,QAAM,cAAc,MAAM,iBACtB,6BAA6B,MAAM,cAAc,IACjD;AACJ,MAAI,OAAO,gBAAgB,SAAU,eAAc,KAAK,WAAW;AAEnE,SAAO;AAAA,IACL,WACE,cAAc,WAAW,IACrB,IACA,cAAc,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,cAAc;AAAA,IAC/D,eAAe,YAAY;AAAA,IAC3B,kBAAkB,eAAe;AAAA,IACjC,iBAAiB,cAAc,WAAW;AAAA,EAC5C;AACF;AAMA,eAAsB,mBACpB,MACmC;AACnC,QAAM,YAAY,IAAI,IAAI,KAAK,OAAO,qBAAqB,CAAC,MAAM,CAAC;AACnE,QAAM,eAAe,IAAI,IAAI,KAAK,OAAO,wBAAwB,CAAC,UAAU,CAAC;AAC7E,YAAU,KAAK,QAAQ,EAAE,WAAW,KAAK,CAAC;AAE1C,QAAM,SAAS,MAAM,eAA2B;AAAA,IAC9C,MAAM;AAAA,MACJ,EAAE,MAAM,WAAW,QAAQ,KAAK,SAAS;AAAA,MACzC,EAAE,MAAM,WAAW,QAAQ,KAAK,SAAS,IAAI,CAAC,OAAO,EAAE,IAAI,EAAE,IAAI,OAAO,EAAE,EAAE,EAAE;AAAA,IAChF;AAAA,IACA,MAAM,KAAK,QAAQ;AAAA,IACnB,gBAAgB,KAAK,kBAAkB;AAAA,IACvC,aAAa,KAAK;AAAA,IAClB,MAAM,QAAQ,MAAM;AAClB,YAAM,UAAU,KAAK,KAAK,SAAS;AACnC,YAAM,UAAU,KAAK,KAAK,SAAS;AACnC,YAAM,YAAY,OAAO,KAAK,KAAK,SAAS,MAAM,SAAS;AAC3D,YAAM,YAAY,OAAO,KAAK,KAAK,SAAS,MAAM,SAAS;AAE3D,YAAM,MAAM,MAAM,aAAa;AAAA,QAC7B;AAAA,QACA;AAAA,QACA,OAAO,KAAK;AAAA,QACZ,OAAO,KAAK;AAAA,QACZ,eAAe,KAAK;AAAA,QACpB,iBAAiB,KAAK;AAAA,QACtB,UAAU,KAAK;AAAA,QACf,mBAAmB,KAAK;AAAA,QACxB,YAAY,KAAK;AAAA,QACjB,aAAa,KAAK;AAAA,QAClB,QAAQ,KAAK;AAAA,QACb,SAAS,KAAK;AAAA,MAChB,CAAC;AAED,YAAM,gBAAgB,IAAI,UAAU,OAAO,CAAC,MAAM,UAAU,IAAI,EAAE,IAAI,CAAC;AACvE,YAAM,mBAAmB,IAAI,UAAU,OAAO,CAAC,MAAM,aAAa,IAAI,EAAE,IAAI,CAAC;AAE7E,YAAM,CAAC,cAAc,aAAa,cAAc,IAAI,MAAM,QAAQ,IAAI;AAAA,QACpE,SAAS,KAAK,OAAO,cAAc,EAAE,YAAY,IAAI,YAAY,QAAQ,CAAC;AAAA,QAC1E,KAAK,OAAO,aACR,QAAQ;AAAA,UACN,cAAc;AAAA,YAAI,CAAC,aACjB,SAAS,KAAK,OAAO,YAAa,EAAE,UAAU,QAAQ,CAAC,EAAE,KAAK,CAAC,OAAO;AAAA,cACpE,GAAG;AAAA,cACH,MAAM,SAAS;AAAA,cACf,MAAM,SAAS;AAAA,YACjB,EAAE;AAAA,UACJ;AAAA,QACF,IACA,QAAQ,QAAQ,CAAC,CAAuD;AAAA,QAC5E,KAAK,OAAO,iBACR,QAAQ;AAAA,UACN,iBAAiB;AAAA,YAAI,CAAC,aACpB,SAAS,KAAK,OAAO,gBAAiB,EAAE,UAAU,QAAQ,CAAC,EAAE,KAAK,CAAC,OAAO;AAAA,cACxE,GAAG;AAAA,cACH,MAAM,SAAS;AAAA,cACf,MAAM,SAAS;AAAA,YACjB,EAAE;AAAA,UACJ;AAAA,QACF,IACA,QAAQ,QAAQ,CAAC,CAAuD;AAAA,MAC9E,CAAC;AAED,YAAM,EAAE,WAAW,eAAe,kBAAkB,gBAAgB,IAAI,qBAAqB;AAAA,QAC3F;AAAA,QACA,aAAa,KAAK,OAAO,aAAa,cAAc;AAAA,QACpD,gBAAgB,KAAK,OAAO,iBAAiB,iBAAiB;AAAA,MAChE,CAAC;AAED,YAAM,YAAgC,EAAE,WAAW,aAAa;AAChE,UAAI,KAAK,OAAO;AACd,kBAAU,aAAa,EAAE,aAAa,aAAa,WAAW,cAAc;AAC9E,UAAI,KAAK,OAAO;AACd,kBAAU,iBAAiB,EAAE,aAAa,gBAAgB,WAAW,iBAAiB;AAExF,YAAM,UAAU,KAAK,KAAK,QAAQ,WAAW,WAAW,OAAO,KAAK,GAAG,EAAE;AACzE,gBAAU,SAAS,EAAE,WAAW,KAAK,CAAC;AACtC,oBAAc,KAAK,SAAS,iBAAiB,GAAG,KAAK,UAAU,IAAI,YAAY,MAAM,CAAC,CAAC;AACvF,oBAAc,KAAK,SAAS,gBAAgB,GAAG,KAAK,UAAU,IAAI,WAAW,MAAM,CAAC,CAAC;AACrF,oBAAc,KAAK,SAAS,aAAa,GAAG,KAAK,UAAU,WAAW,MAAM,CAAC,CAAC;AAE9E,YAAM,QAAQ,CAAC,SAAS,aAAa,UAAU,QAAQ,CAAC,CAAC,EAAE;AAC3D,UAAI,KAAK,OAAO,WAAY,OAAM,KAAK,QAAQ,cAAc,QAAQ,CAAC,CAAC,EAAE;AACzE,UAAI,KAAK,OAAO,eAAgB,OAAM,KAAK,WAAW,iBAAiB,QAAQ,CAAC,CAAC,EAAE;AACnF,UAAI,gBAAiB,OAAM,KAAK,mBAAmB;AAEnD,aAAO;AAAA,QACL,QAAQ;AAAA,UACN,OAAO,IAAI,WAAW;AAAA,UACtB,WAAW,IAAI;AAAA,UACf,eAAe,IAAI,UAAU;AAAA,QAC/B;AAAA,QACA,SAAS,EAAE,OAAO,aAAa,GAAG,OAAO,WAAW,OAAO,MAAM,KAAK,GAAG,EAAE;AAAA,QAC3E,SAAS,IAAI;AAAA,QACb,YAAY,IAAI;AAAA,MAClB;AAAA,IACF;AAAA,EACF,CAAC;AAGD,QAAM,UAAU;AAAA,IACd,OAAO,OAAO,QAAQ;AAAA,IACtB,UAAU,OAAO,QAAQ;AAAA,IACzB,WAAW,OAAO,QAAQ;AAAA,IAC1B,cAAc,OAAO,QAAQ;AAAA,IAC7B,YAAY,OAAO,QAAQ;AAAA,IAC3B,cAAc,OAAO,QAAQ;AAAA,IAC7B,cAAc,OAAO,QAAQ;AAAA,IAC7B,WAAW,OAAO,OAAO;AAAA,IACzB,WAAW,OAAO,OAAO;AAAA,EAC3B;AACA,gBAAc,KAAK,KAAK,QAAQ,cAAc,GAAG,KAAK,UAAU,SAAS,MAAM,CAAC,CAAC;AAEjF,QAAM,KAAe;AAAA,IACnB;AAAA,IACA;AAAA,IACA,cAAc,OAAO,QAAQ,UAAU,sBAAsB,OAAO,QAAQ,kBAAkB,KAAK,QAAQ,CAAC,CAAC,iBAAiB,OAAO,QAAQ,iBAAiB,QAAQ,CAAC,CAAC,iBAAiB,OAAO,QAAQ,aAAa,QAAQ,CAAC,CAAC,qBAAqB,OAAO,QAAQ,aAAa,KAAM,QAAQ,CAAC,CAAC;AAAA,IAChS;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA,GAAG,OAAO,QAAQ,OAAO,OAAO,WAAW,CAAC,CAAC,EAAE;AAAA,MAC7C,CAAC,CAAC,IAAI,CAAC,MACL,KAAK,EAAE,OAAO,EAAE,WAAW,KAAK,QAAQ,CAAC,CAAC,OAAO,EAAE,UAAU,QAAQ,CAAC,CAAC,OAAO,EAAE,aAAa,QAAQ,CAAC,CAAC;AAAA,IAC3G;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA,GAAG,OAAO,QAAQ,OAAO,OAAO,WAAW,CAAC,CAAC,EAAE;AAAA,MAC7C,CAAC,CAAC,IAAI,CAAC,MACL,KAAK,EAAE,OAAO,EAAE,WAAW,KAAK,QAAQ,CAAC,CAAC,OAAO,EAAE,UAAU,QAAQ,CAAC,CAAC,OAAO,EAAE,aAAa,QAAQ,CAAC,CAAC;AAAA,IAC3G;AAAA,IACA;AAAA,EACF;AACA,gBAAc,KAAK,KAAK,QAAQ,YAAY,GAAG,GAAG,KAAK,IAAI,CAAC;AAE5D,SAAO,EAAE,OAAO;AAClB;","names":[]}

package/dist/openapi.json

@@ -2,7 +2,7 @@
   "openapi": "3.1.0",
   "info": {
     "title": "@tangle-network/agent-eval — wire protocol",
-    "version": "0.96.0",
+    "version": "0.96.2",
     "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/pipelines/index.js

@@ -3,13 +3,14 @@ import {
   classifyFailure,
   compareToBaseline,
   computeToolUseMetrics
-} from "../chunk-NACM5RS7.js";
+} from "../chunk-IH7LYRHL.js";
 import {
   buildTrajectory
 } from "../chunk-RZTMDUO7.js";
 import {
-  interRaterReliability
-} from "../chunk-56GC6VYO.js";
+  interRaterReliability,
+  pearsonR
+} from "../chunk-HRGUJTER.js";
 import {
   aggregateLlm,
   argHash,
@@ -228,7 +229,7 @@ async function judgeAgreementView(store) {
           judgeB: judgeJ,
           dimension: dim,
           commonItems: common.length,
-          pearson: pearson(
+          pearson: pearsonR(
             common.map((c) => c[0]),
             common.map((c) => c[1])
           ),
@@ -243,21 +244,6 @@ async function judgeAgreementView(store) {
     judgeIds
   };
 }
-function pearson(a, b) {
-  if (a.length !== b.length || a.length < 2) return NaN;
-  const mA = a.reduce((s, v) => s + v, 0) / a.length;
-  const mB = b.reduce((s, v) => s + v, 0) / b.length;
-  let num = 0, denA = 0, denB = 0;
-  for (let i = 0; i < a.length; i++) {
-    const dA = a[i] - mA;
-    const dB = b[i] - mB;
-    num += dA * dB;
-    denA += dA * dA;
-    denB += dB * dB;
-  }
-  if (denA === 0 || denB === 0) return denA === 0 && denB === 0 ? 1 : 0;
-  return num / Math.sqrt(denA * denB);
-}
 
 // src/pipelines/regression.ts
 async function regressionView(store, metrics, options) {

package/dist/pipelines/index.js.map

@@ -1 +1 @@
-{"version":3,"sources":["../../src/pipelines/budget-breach.ts","../../src/pipelines/failure-cluster.ts","../../src/pipelines/first-divergence.ts","../../src/pipelines/judge-agreement.ts","../../src/pipelines/regression.ts","../../src/pipelines/stuck-loop.ts","../../src/pipelines/tool-waste.ts"],"sourcesContent":["/**\n * BudgetBreachView — aggregates breach events across the corpus.\n *\n * Answers: which dimensions get hit most often? Which scenarios are\n * underbudgeted? Which variants trigger the most breaches?\n */\n\nimport type { BudgetSpec } from '../trace/schema'\nimport type { TraceStore } from '../trace/store'\n\nexport interface BudgetBreachFinding {\n  runId: string\n  scenarioId: string\n  variantId?: string\n  dimension: keyof BudgetSpec\n  limit: number\n  consumed: number\n  excessRatio: number\n  timestamp: number\n}\n\nexport interface BudgetBreachReport {\n  findings: BudgetBreachFinding[]\n  byDimension: Record<string, number>\n  byScenario: Record<string, number>\n  byVariant: Record<string, number>\n  totalRuns: number\n  breachedRunRatio: number\n}\n\nexport async function budgetBreachView(\n  store: TraceStore,\n  options: { scenarioId?: string; variantId?: string } = {},\n): Promise<BudgetBreachReport> {\n  const runs = await store.listRuns({\n    scenarioId: options.scenarioId,\n    variantId: options.variantId,\n  })\n  const findings: BudgetBreachFinding[] = []\n  const byDimension: Record<string, number> = {}\n  const byScenario: Record<string, number> = {}\n  const byVariant: Record<string, number> = {}\n\n  for (const run of runs) {\n    const entries = await store.budget(run.runId)\n    for (const e of entries) {\n      if (!e.breached) continue\n      const excessRatio = e.limit > 0 ? e.consumed / e.limit : Infinity\n      findings.push({\n        runId: run.runId,\n        scenarioId: run.scenarioId,\n        variantId: run.variantId,\n        dimension: e.dimension,\n        limit: e.limit,\n        consumed: e.consumed,\n        excessRatio,\n        timestamp: e.timestamp,\n      })\n      byDimension[e.dimension] = (byDimension[e.dimension] ?? 0) + 1\n      byScenario[run.scenarioId] = (byScenario[run.scenarioId] ?? 0) + 1\n      if (run.variantId) byVariant[run.variantId] = (byVariant[run.variantId] ?? 0) + 1\n    }\n  }\n\n  const breachedRuns = new Set(findings.map((f) => f.runId))\n  return {\n    findings,\n    byDimension,\n    byScenario,\n    byVariant,\n    totalRuns: runs.length,\n    breachedRunRatio: runs.length > 0 ? breachedRuns.size / runs.length : 0,\n  }\n}\n","/**\n * FailureClusterView — groups failed runs by (failureClass, triggerTool,\n * argHash-prefix) so weekly reviews can prioritize the top-N clusters.\n *\n * Each cluster includes: N runs, scenarios affected, representative\n * error message, a proposed mitigation hint (rule → action table).\n */\n\nimport { classifyFailure, DEFAULT_RULES, type FailureRule } from '../failure-taxonomy'\nimport { argHash, toolSpans } from '../trace/query'\nimport type { FailureClass, Span } from '../trace/schema'\nimport type { TraceStore } from '../trace/store'\n\nexport interface FailureCluster {\n  failureClass: FailureClass\n  /** Tool name when the trigger was a tool span, else undefined. */\n  toolName?: string\n  /** First 16 chars of argHash — clusters similar args. */\n  argPrefix?: string\n  /**\n   * Source dimension when the trigger was a judge span (e.g. `'format'`,\n   * `'safety'`, `'correctness'`). Lets cross-template aggregators\n   * group failures by the dimension that fired without overloading\n   * `argPrefix`. Optional — clusters without this field deserialize cleanly.\n   */\n  dimension?: string\n  runCount: number\n  scenarioIds: string[]\n  exampleError?: string\n  exampleRunId: string\n}\n\nexport interface FailureClusterReport {\n  clusters: FailureCluster[]\n  totalFailures: number\n  totalRuns: number\n}\n\nexport async function failureClusterView(\n  store: TraceStore,\n  options: { rules?: FailureRule[]; minClusterSize?: number } = {},\n): Promise<FailureClusterReport> {\n  const rules = options.rules ?? DEFAULT_RULES\n  const minSize = options.minClusterSize ?? 1\n  const runs = await store.listRuns()\n\n  type Key = string\n  const clusters = new Map<Key, FailureCluster>()\n  let totalFailures = 0\n\n  for (const run of runs) {\n    if (run.status === 'completed' && run.outcome?.pass !== false) continue\n    totalFailures++\n    const spans = await store.spans({ runId: run.runId })\n    const events = await store.events({ runId: run.runId })\n    const cls = classifyFailure({ run, spans, events }, rules)\n\n    let toolName: string | undefined\n    let argPrefix: string | undefined\n    let dimension: string | undefined\n    if (cls.triggerSpanId) {\n      const trig = spans.find((s) => s.spanId === cls.triggerSpanId)\n      if (trig?.kind === 'tool') {\n        toolName = trig.toolName\n        argPrefix = argHash(trig.args).slice(0, 16)\n      } else if (trig?.kind === 'judge') {\n        dimension = trig.dimension\n      }\n    }\n    // Fallback: look at the last errored tool span\n    if (!toolName) {\n      const ts = await toolSpans(store, run.runId)\n      const errored = ts.filter((t) => t.status === 'error').pop()\n      if (errored) {\n        toolName = errored.toolName\n        argPrefix = argHash(errored.args).slice(0, 16)\n      }\n    }\n    // Secondary signal: any judge span on the failed run carries a\n    // dimension. Useful when the rule classified by judge score but\n    // didn't surface the trigger span (or surfaced a non-judge span).\n    if (!dimension) {\n      const judge = spans.find((s) => s.kind === 'judge' && typeof s.dimension === 'string')\n      if (judge?.kind === 'judge') dimension = judge.dimension\n    }\n\n    const key = `${cls.failureClass}|${toolName ?? ''}|${argPrefix ?? ''}|${dimension ?? ''}`\n    let cluster = clusters.get(key)\n    if (!cluster) {\n      cluster = {\n        failureClass: cls.failureClass,\n        toolName,\n        argPrefix,\n        dimension,\n        runCount: 0,\n        scenarioIds: [],\n        exampleRunId: run.runId,\n        exampleError: firstErrorMessage(spans) ?? cls.reason,\n      }\n      clusters.set(key, cluster)\n    }\n    cluster.runCount++\n    if (!cluster.scenarioIds.includes(run.scenarioId)) cluster.scenarioIds.push(run.scenarioId)\n  }\n\n  const arr = [...clusters.values()]\n    .filter((c) => c.runCount >= minSize)\n    .sort((a, b) => b.runCount - a.runCount)\n\n  return { clusters: arr, totalFailures, totalRuns: runs.length }\n}\n\nfunction firstErrorMessage(spans: Span[]): string | undefined {\n  const errored = spans.find((s) => s.status === 'error')\n  return errored?.error\n}\n","/**\n * FirstDivergenceView — aligns two trajectories by step index, reports\n * the first step where they differ.\n *\n * \"Differ\" is configurable — default is (kind, toolName if tool, model\n * if llm). Use this view to attribute \"why is variant B better?\" to a\n * specific step rather than an aggregate mean delta.\n */\n\nimport type { TraceStore } from '../trace/store'\nimport { buildTrajectory, type Trajectory, type TrajectoryStep } from '../trajectory'\n\nexport interface DivergenceReport {\n  runA: string\n  runB: string\n  firstDivergenceIndex: number | null\n  aStep?: TrajectoryStep\n  bStep?: TrajectoryStep\n  reason?: string\n  /** Common prefix length (steps that matched). */\n  commonPrefixLen: number\n}\n\nexport interface DivergenceOptions {\n  /** Returns true if two steps are considered equal. Default: kind + tool/model match. */\n  stepEquals?: (a: TrajectoryStep, b: TrajectoryStep) => boolean\n}\n\nexport async function firstDivergenceView(\n  store: TraceStore,\n  runA: string,\n  runB: string,\n  options: DivergenceOptions = {},\n): Promise<DivergenceReport> {\n  const [a, b] = await Promise.all([buildTrajectory(store, runA), buildTrajectory(store, runB)])\n  const eq = options.stepEquals ?? defaultStepEquals\n  const minLen = Math.min(a.steps.length, b.steps.length)\n  for (let i = 0; i < minLen; i++) {\n    const aStep = a.steps[i]!\n    const bStep = b.steps[i]!\n    if (!eq(aStep, bStep)) {\n      return {\n        runA,\n        runB,\n        firstDivergenceIndex: i,\n        aStep,\n        bStep,\n        reason: describeDifference(aStep, bStep),\n        commonPrefixLen: i,\n      }\n    }\n  }\n  if (a.steps.length === b.steps.length) {\n    return { runA, runB, firstDivergenceIndex: null, commonPrefixLen: minLen }\n  }\n  const longer: Trajectory = a.steps.length > b.steps.length ? a : b\n  const extra = longer.steps.length - minLen\n  // minLen === 0 means one trajectory is empty: divergence is the first step\n  // itself (index 0), and the absent side has no step to surface.\n  const reason =\n    minLen === 0\n      ? `one trajectory is empty; the other has ${extra} step(s) starting at index 0`\n      : `one trajectory has ${extra} more step(s) after index ${minLen - 1}`\n  return {\n    runA,\n    runB,\n    firstDivergenceIndex: minLen,\n    aStep: a.steps[minLen],\n    bStep: b.steps[minLen],\n    reason,\n    commonPrefixLen: minLen,\n  }\n}\n\nfunction defaultStepEquals(a: TrajectoryStep, b: TrajectoryStep): boolean {\n  if (a.span.kind !== b.span.kind) return false\n  if (a.span.kind === 'tool' && b.span.kind === 'tool') return a.span.toolName === b.span.toolName\n  if (a.span.kind === 'llm' && b.span.kind === 'llm') return a.span.model === b.span.model\n  if (a.span.kind === 'judge' && b.span.kind === 'judge')\n    return a.span.dimension === b.span.dimension\n  return a.span.name === b.span.name\n}\n\nfunction describeDifference(a: TrajectoryStep, b: TrajectoryStep): string {\n  if (a.span.kind !== b.span.kind) return `kind ${a.span.kind} vs ${b.span.kind}`\n  if (a.span.kind === 'tool' && b.span.kind === 'tool' && a.span.toolName !== b.span.toolName) {\n    return `tool ${a.span.toolName} vs ${b.span.toolName}`\n  }\n  if (a.span.kind === 'llm' && b.span.kind === 'llm' && a.span.model !== b.span.model) {\n    return `model ${a.span.model} vs ${b.span.model}`\n  }\n  return `name \"${a.span.name}\" vs \"${b.span.name}\"`\n}\n","/**\n * JudgeAgreementView — pairwise agreement between judges across the\n * corpus, grouped by dimension.\n *\n * Output drives two workflows:\n *   - Judge robustness audit: \"does Claude agree with GPT at κ ≥ 0.6?\"\n *   - Calibration tracking: κ vs golden human labels over time (by\n *     providing a `humanGoldenJudgeId`).\n */\n\nimport { interRaterReliability } from '../statistics'\nimport type { JudgeSpan } from '../trace/schema'\nimport type { TraceStore } from '../trace/store'\n\nexport interface JudgePair {\n  judgeA: string\n  judgeB: string\n  dimension: string\n  /** Number of (targetSpanId, dimension) tuples both judges scored. */\n  commonItems: number\n  pearson: number\n  krippendorff: number\n}\n\nexport interface JudgeAgreementReport {\n  pairs: JudgePair[]\n  dimensions: string[]\n  judgeIds: string[]\n}\n\nexport async function judgeAgreementView(store: TraceStore): Promise<JudgeAgreementReport> {\n  const all = (await store.spans({ kind: 'judge' })).filter(\n    (s): s is JudgeSpan => s.kind === 'judge',\n  )\n  if (all.length === 0) return { pairs: [], dimensions: [], judgeIds: [] }\n\n  const byDimension = new Map<string, JudgeSpan[]>()\n  for (const s of all) {\n    const arr = byDimension.get(s.dimension) ?? []\n    arr.push(s)\n    byDimension.set(s.dimension, arr)\n  }\n\n  const judgeIds = [...new Set(all.map((s) => s.judgeId))].sort()\n  const pairs: JudgePair[] = []\n  for (const [dim, spans] of byDimension) {\n    const byJudge = new Map<string, Map<string, number>>()\n    for (const s of spans) {\n      const m = byJudge.get(s.judgeId) ?? new Map<string, number>()\n      m.set(s.targetSpanId, s.score)\n      byJudge.set(s.judgeId, m)\n    }\n    const judgesHere = [...byJudge.keys()]\n    for (let i = 0; i < judgesHere.length; i++) {\n      for (let j = i + 1; j < judgesHere.length; j++) {\n        const judgeI = judgesHere[i]!\n        const judgeJ = judgesHere[j]!\n        const a = byJudge.get(judgeI)!\n        const b = byJudge.get(judgeJ)!\n        const common: Array<[number, number]> = []\n        for (const [target, scoreA] of a) {\n          const scoreB = b.get(target)\n          if (scoreB !== undefined) common.push([scoreA, scoreB])\n        }\n        if (common.length < 2) continue\n        const judgeScores = common.map(\n          ([scoreA, scoreB]) =>\n            [\n              { judgeName: judgeI, dimension: dim, score: scoreA, reasoning: '' },\n              { judgeName: judgeJ, dimension: dim, score: scoreB, reasoning: '' },\n            ] as const,\n        )\n        const k = interRaterReliability(\n          judgeScores[0]!.map((_, k2) => judgeScores.map((pair) => pair[k2]!)),\n        )\n        pairs.push({\n          judgeA: judgeI,\n          judgeB: judgeJ,\n          dimension: dim,\n          commonItems: common.length,\n          pearson: pearson(\n            common.map((c) => c[0]),\n            common.map((c) => c[1]),\n          ),\n          krippendorff: k,\n        })\n      }\n    }\n  }\n\n  return {\n    pairs: pairs.sort((a, b) => b.commonItems - a.commonItems),\n    dimensions: [...byDimension.keys()].sort(),\n    judgeIds,\n  }\n}\n\nfunction pearson(a: number[], b: number[]): number {\n  if (a.length !== b.length || a.length < 2) return NaN\n  const mA = a.reduce((s, v) => s + v, 0) / a.length\n  const mB = b.reduce((s, v) => s + v, 0) / b.length\n  let num = 0,\n    denA = 0,\n    denB = 0\n  for (let i = 0; i < a.length; i++) {\n    const dA = a[i]! - mA\n    const dB = b[i]! - mB\n    num += dA * dB\n    denA += dA * dA\n    denB += dB * dB\n  }\n  if (denA === 0 || denB === 0) return denA === 0 && denB === 0 ? 1 : 0\n  return num / Math.sqrt(denA * denB)\n}\n","/**\n * RegressionView — compares a candidate slice to a baseline slice on a\n * named metric. Delegates the statistics (Welch's t-test, Cohen's d,\n * IQR stability) to `baseline.ts`.\n *\n * This is the entry point for CI regression gates: \"given runs tagged\n * release=A and release=B, did any metric regress?\"\n */\n\nimport { type BaselineOptions, type BaselineReport, compareToBaseline } from '../baseline'\nimport { aggregateLlm, llmSpans, runFailureClass } from '../trace/query'\nimport type { Run } from '../trace/schema'\nimport type { RunFilter, TraceStore } from '../trace/store'\n\nexport interface RegressionSpec {\n  metric: string\n  higherIsBetter: boolean\n  /** Extract a scalar from a run. Default extractors handle common metrics. */\n  extract?: (run: Run, store: TraceStore) => Promise<number | null>\n}\n\nexport interface RegressionOptions extends BaselineOptions {\n  baseline: RunFilter\n  candidate: RunFilter\n}\n\nexport async function regressionView(\n  store: TraceStore,\n  metrics: RegressionSpec[],\n  options: RegressionOptions,\n): Promise<BaselineReport> {\n  const baselineRuns = await store.listRuns(options.baseline)\n  const candidateRuns = await store.listRuns(options.candidate)\n  const samples = await Promise.all(\n    metrics.map(async (m) => {\n      const extract = m.extract ?? defaultExtract(m.metric)\n      const baseline = await extractAll(baselineRuns, extract, store)\n      const candidate = await extractAll(candidateRuns, extract, store)\n      return { metric: m.metric, higherIsBetter: m.higherIsBetter, baseline, candidate }\n    }),\n  )\n  return compareToBaseline(samples, options)\n}\n\nasync function extractAll(\n  runs: Run[],\n  extract: (r: Run, s: TraceStore) => Promise<number | null>,\n  store: TraceStore,\n): Promise<number[]> {\n  const out: number[] = []\n  for (const r of runs) {\n    const v = await extract(r, store)\n    if (v !== null && Number.isFinite(v)) out.push(v)\n  }\n  return out\n}\n\nfunction defaultExtract(metric: string): (run: Run, store: TraceStore) => Promise<number | null> {\n  return async (run, store) => {\n    switch (metric) {\n      case 'score':\n      case 'overallScore':\n        return run.outcome?.score ?? null\n      case 'pass':\n        return run.outcome?.pass === true ? 1 : 0\n      case 'durationMs':\n        return run.endedAt && run.startedAt ? run.endedAt - run.startedAt : null\n      case 'costUsd': {\n        const llm = await llmSpans(store, run.runId)\n        return aggregateLlm(llm).costUsd\n      }\n      case 'inputTokens': {\n        const llm = await llmSpans(store, run.runId)\n        return aggregateLlm(llm).inputTokens\n      }\n      case 'outputTokens': {\n        const llm = await llmSpans(store, run.runId)\n        return aggregateLlm(llm).outputTokens\n      }\n      case 'failureClass': {\n        return runFailureClass(run) === 'success' ? 1 : 0\n      }\n      default:\n        return null\n    }\n  }\n}\n","/**\n * StuckLoopView — detects when an agent calls the same tool with the\n * same (or structurally similar) arguments ≥ N times in a short window.\n *\n * Rationale: agents that loop are the number-one production failure\n * mode on long-horizon flows. The view returns (runId, toolName,\n * argHash, occurrences, windowMs) for each detected loop plus a\n * fraction of runs affected.\n */\n\nimport { argHash, toolSpans } from '../trace/query'\nimport type { TraceStore } from '../trace/store'\n\nexport interface StuckLoopFinding {\n  runId: string\n  toolName: string\n  argHash: string\n  occurrences: number\n  spanIds: string[]\n  /** Milliseconds between first and last call in the loop. */\n  windowMs: number\n}\n\nexport interface StuckLoopReport {\n  findings: StuckLoopFinding[]\n  affectedRunRatio: number\n  totalRuns: number\n}\n\nexport interface StuckLoopOptions {\n  /** Minimum call count to flag a loop (default 3). */\n  minOccurrences?: number\n  /** Filter to a specific runId; omit to scan the entire corpus. */\n  runId?: string\n}\n\nexport async function stuckLoopView(\n  store: TraceStore,\n  options: StuckLoopOptions = {},\n): Promise<StuckLoopReport> {\n  const minOccurrences = options.minOccurrences ?? 3\n  const runs = options.runId\n    ? [{ runId: options.runId }]\n    : (await store.listRuns()).map((r) => ({ runId: r.runId }))\n\n  const findings: StuckLoopFinding[] = []\n  for (const { runId } of runs) {\n    const tools = await toolSpans(store, runId)\n    const byKey = new Map<string, { spans: typeof tools; argHash: string; toolName: string }>()\n    for (const t of tools) {\n      const h = argHash(t.args)\n      // NUL delimiter never appears in a tool name or a JSON argHash, so the\n      // key cannot collide; toolName is carried in the bucket rather than\n      // re-derived by splitting the key (which mislabeled any tool whose name\n      // contained the delimiter, e.g. \"shell|grep\").\n      const key = `${t.toolName}\\u0000${h}`\n      const bucket = byKey.get(key) ?? { spans: [], argHash: h, toolName: t.toolName }\n      bucket.spans.push(t)\n      byKey.set(key, bucket)\n    }\n    for (const { spans, argHash: h, toolName } of byKey.values()) {\n      if (spans.length < minOccurrences) continue\n      const sorted = [...spans].sort((a, b) => a.startedAt - b.startedAt)\n      const first = sorted[0]!.startedAt\n      const last = sorted[sorted.length - 1]!.startedAt\n      findings.push({\n        runId,\n        toolName,\n        argHash: h,\n        occurrences: sorted.length,\n        spanIds: sorted.map((s) => s.spanId),\n        windowMs: last - first,\n      })\n    }\n  }\n\n  const affectedRuns = new Set(findings.map((f) => f.runId))\n  return {\n    findings,\n    affectedRunRatio: runs.length > 0 ? affectedRuns.size / runs.length : 0,\n    totalRuns: runs.length,\n  }\n}\n","/**\n * ToolWasteView — fraction of tool calls whose results weren't used\n * downstream. Without a \"used\" signal we fall back to structural\n * proxies: error calls, duplicate calls, and tool calls followed by\n * zero subsequent LLM spans are all considered waste.\n *\n * Consumers can pass a `usageOracle` that inspects a tool span and\n * returns true iff the tool's result appears in a later LLM message,\n * artifact, or state mutation — that's the canonical definition; the\n * default heuristic is a reasonable fallback.\n */\n\nimport { computeToolUseMetrics } from '../tool-use-metrics'\nimport { llmSpans, toolSpans } from '../trace/query'\nimport type { LlmSpan, ToolSpan } from '../trace/schema'\nimport type { TraceStore } from '../trace/store'\n\nexport interface ToolWasteFinding {\n  runId: string\n  wastedCalls: number\n  totalCalls: number\n  wasteRate: number\n}\n\nexport interface ToolWasteReport {\n  byRun: ToolWasteFinding[]\n  overallWasteRate: number\n}\n\nexport interface ToolWasteOptions {\n  runId?: string\n  usageOracle?: (tool: ToolSpan, later: { llm: Awaited<ReturnType<typeof llmSpans>> }) => boolean\n}\n\nexport async function toolWasteView(\n  store: TraceStore,\n  options: ToolWasteOptions = {},\n): Promise<ToolWasteReport> {\n  const runs = options.runId ? [options.runId] : (await store.listRuns()).map((r) => r.runId)\n\n  const byRun: ToolWasteFinding[] = []\n  let totalCalls = 0\n  let totalWasted = 0\n  for (const runId of runs) {\n    const tools = await toolSpans(store, runId)\n    if (tools.length === 0) {\n      byRun.push({ runId, wastedCalls: 0, totalCalls: 0, wasteRate: 0 })\n      continue\n    }\n    const llms = await llmSpans(store, runId)\n    // Sort LLM spans once by start time, then build a suffix index of the\n    // concatenated message text. `suffixText[i]` is the haystack of every\n    // string message content in spans[i..]. Per tool we binary-search the\n    // first span started strictly after the tool, then test that one suffix\n    // — turning the per-tool O(llms × messages × content) scan into a single\n    // O(log llms) lookup over precomputed text.\n    const sortedLlm = [...llms].sort((a, b) => a.startedAt - b.startedAt)\n    const startTimes = sortedLlm.map((l) => l.startedAt)\n    const suffixText = buildSuffixText(sortedLlm)\n    let wasted = 0\n    for (const t of tools) {\n      if (t.status === 'error') {\n        wasted++\n        continue\n      }\n      // First LLM span started strictly after this tool (upper-bound search).\n      const cutoff = upperBound(startTimes, t.startedAt)\n      if (options.usageOracle) {\n        if (!options.usageOracle(t, { llm: sortedLlm.slice(cutoff) })) wasted++\n      } else {\n        // Default heuristic: a tool whose result is NOT mentioned in any\n        // later LLM input message is likely wasted. An empty/null result has\n        // no payload to propagate downstream — there is nothing to find in a\n        // later message, so it is not evidence of waste; skip it.\n        const resultStr = stringify(t.result)\n        if (resultStr === '') continue\n        const haystack = suffixText[cutoff] ?? ''\n        const used = haystack.includes(resultStr.slice(0, 120))\n        if (!used) wasted++\n      }\n    }\n    const wasteRate = wasted / tools.length\n    byRun.push({ runId, wastedCalls: wasted, totalCalls: tools.length, wasteRate })\n    totalCalls += tools.length\n    totalWasted += wasted\n  }\n  return { byRun, overallWasteRate: totalCalls > 0 ? totalWasted / totalCalls : 0 }\n}\n\n/**\n * Build per-position suffix haystacks: result[i] is the concatenation of every\n * string message content in spans[i..end]. Built back-to-front so each entry\n * reuses the next one — O(total message text) rather than O(spans²).\n */\nfunction buildSuffixText(spans: LlmSpan[]): string[] {\n  const result = new Array<string>(spans.length + 1)\n  result[spans.length] = ''\n  for (let i = spans.length - 1; i >= 0; i--) {\n    const own = spans[i]!.messages.map((m) =>\n      typeof m.content === 'string' ? m.content : '',\n    ).join('\\n')\n    result[i] = `${own}\\n${result[i + 1]}`\n  }\n  return result\n}\n\n/** Index of the first element strictly greater than `target` in a sorted array. */\nfunction upperBound(sorted: number[], target: number): number {\n  let lo = 0\n  let hi = sorted.length\n  while (lo < hi) {\n    const mid = (lo + hi) >>> 1\n    if (sorted[mid]! <= target) lo = mid + 1\n    else hi = mid\n  }\n  return lo\n}\n\nfunction stringify(v: unknown): string {\n  if (v === null || v === undefined) return ''\n  if (typeof v === 'string') return v\n  try {\n    return JSON.stringify(v)\n  } catch {\n    return String(v)\n  }\n}\n\n// Re-export for convenience in consumers that want both descriptive and usage metrics.\nexport { computeToolUseMetrics }\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;AA8BA,eAAsB,iBACpB,OACA,UAAuD,CAAC,GAC3B;AAC7B,QAAM,OAAO,MAAM,MAAM,SAAS;AAAA,IAChC,YAAY,QAAQ;AAAA,IACpB,WAAW,QAAQ;AAAA,EACrB,CAAC;AACD,QAAM,WAAkC,CAAC;AACzC,QAAM,cAAsC,CAAC;AAC7C,QAAM,aAAqC,CAAC;AAC5C,QAAM,YAAoC,CAAC;AAE3C,aAAW,OAAO,MAAM;AACtB,UAAM,UAAU,MAAM,MAAM,OAAO,IAAI,KAAK;AAC5C,eAAW,KAAK,SAAS;AACvB,UAAI,CAAC,EAAE,SAAU;AACjB,YAAM,cAAc,EAAE,QAAQ,IAAI,EAAE,WAAW,EAAE,QAAQ;AACzD,eAAS,KAAK;AAAA,QACZ,OAAO,IAAI;AAAA,QACX,YAAY,IAAI;AAAA,QAChB,WAAW,IAAI;AAAA,QACf,WAAW,EAAE;AAAA,QACb,OAAO,EAAE;AAAA,QACT,UAAU,EAAE;AAAA,QACZ;AAAA,QACA,WAAW,EAAE;AAAA,MACf,CAAC;AACD,kBAAY,EAAE,SAAS,KAAK,YAAY,EAAE,SAAS,KAAK,KAAK;AAC7D,iBAAW,IAAI,UAAU,KAAK,WAAW,IAAI,UAAU,KAAK,KAAK;AACjE,UAAI,IAAI,UAAW,WAAU,IAAI,SAAS,KAAK,UAAU,IAAI,SAAS,KAAK,KAAK;AAAA,IAClF;AAAA,EACF;AAEA,QAAM,eAAe,IAAI,IAAI,SAAS,IAAI,CAAC,MAAM,EAAE,KAAK,CAAC;AACzD,SAAO;AAAA,IACL;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA,WAAW,KAAK;AAAA,IAChB,kBAAkB,KAAK,SAAS,IAAI,aAAa,OAAO,KAAK,SAAS;AAAA,EACxE;AACF;;;ACnCA,eAAsB,mBACpB,OACA,UAA8D,CAAC,GAChC;AAC/B,QAAM,QAAQ,QAAQ,SAAS;AAC/B,QAAM,UAAU,QAAQ,kBAAkB;AAC1C,QAAM,OAAO,MAAM,MAAM,SAAS;AAGlC,QAAM,WAAW,oBAAI,IAAyB;AAC9C,MAAI,gBAAgB;AAEpB,aAAW,OAAO,MAAM;AACtB,QAAI,IAAI,WAAW,eAAe,IAAI,SAAS,SAAS,MAAO;AAC/D;AACA,UAAM,QAAQ,MAAM,MAAM,MAAM,EAAE,OAAO,IAAI,MAAM,CAAC;AACpD,UAAM,SAAS,MAAM,MAAM,OAAO,EAAE,OAAO,IAAI,MAAM,CAAC;AACtD,UAAM,MAAM,gBAAgB,EAAE,KAAK,OAAO,OAAO,GAAG,KAAK;AAEzD,QAAI;AACJ,QAAI;AACJ,QAAI;AACJ,QAAI,IAAI,eAAe;AACrB,YAAM,OAAO,MAAM,KAAK,CAAC,MAAM,EAAE,WAAW,IAAI,aAAa;AAC7D,UAAI,MAAM,SAAS,QAAQ;AACzB,mBAAW,KAAK;AAChB,oBAAY,QAAQ,KAAK,IAAI,EAAE,MAAM,GAAG,EAAE;AAAA,MAC5C,WAAW,MAAM,SAAS,SAAS;AACjC,oBAAY,KAAK;AAAA,MACnB;AAAA,IACF;AAEA,QAAI,CAAC,UAAU;AACb,YAAM,KAAK,MAAM,UAAU,OAAO,IAAI,KAAK;AAC3C,YAAM,UAAU,GAAG,OAAO,CAAC,MAAM,EAAE,WAAW,OAAO,EAAE,IAAI;AAC3D,UAAI,SAAS;AACX,mBAAW,QAAQ;AACnB,oBAAY,QAAQ,QAAQ,IAAI,EAAE,MAAM,GAAG,EAAE;AAAA,MAC/C;AAAA,IACF;AAIA,QAAI,CAAC,WAAW;AACd,YAAM,QAAQ,MAAM,KAAK,CAAC,MAAM,EAAE,SAAS,WAAW,OAAO,EAAE,cAAc,QAAQ;AACrF,UAAI,OAAO,SAAS,QAAS,aAAY,MAAM;AAAA,IACjD;AAEA,UAAM,MAAM,GAAG,IAAI,YAAY,IAAI,YAAY,EAAE,IAAI,aAAa,EAAE,IAAI,aAAa,EAAE;AACvF,QAAI,UAAU,SAAS,IAAI,GAAG;AAC9B,QAAI,CAAC,SAAS;AACZ,gBAAU;AAAA,QACR,cAAc,IAAI;AAAA,QAClB;AAAA,QACA;AAAA,QACA;AAAA,QACA,UAAU;AAAA,QACV,aAAa,CAAC;AAAA,QACd,cAAc,IAAI;AAAA,QAClB,cAAc,kBAAkB,KAAK,KAAK,IAAI;AAAA,MAChD;AACA,eAAS,IAAI,KAAK,OAAO;AAAA,IAC3B;AACA,YAAQ;AACR,QAAI,CAAC,QAAQ,YAAY,SAAS,IAAI,UAAU,EAAG,SAAQ,YAAY,KAAK,IAAI,UAAU;AAAA,EAC5F;AAEA,QAAM,MAAM,CAAC,GAAG,SAAS,OAAO,CAAC,EAC9B,OAAO,CAAC,MAAM,EAAE,YAAY,OAAO,EACnC,KAAK,CAAC,GAAG,MAAM,EAAE,WAAW,EAAE,QAAQ;AAEzC,SAAO,EAAE,UAAU,KAAK,eAAe,WAAW,KAAK,OAAO;AAChE;AAEA,SAAS,kBAAkB,OAAmC;AAC5D,QAAM,UAAU,MAAM,KAAK,CAAC,MAAM,EAAE,WAAW,OAAO;AACtD,SAAO,SAAS;AAClB;;;ACvFA,eAAsB,oBACpB,OACA,MACA,MACA,UAA6B,CAAC,GACH;AAC3B,QAAM,CAAC,GAAG,CAAC,IAAI,MAAM,QAAQ,IAAI,CAAC,gBAAgB,OAAO,IAAI,GAAG,gBAAgB,OAAO,IAAI,CAAC,CAAC;AAC7F,QAAM,KAAK,QAAQ,cAAc;AACjC,QAAM,SAAS,KAAK,IAAI,EAAE,MAAM,QAAQ,EAAE,MAAM,MAAM;AACtD,WAAS,IAAI,GAAG,IAAI,QAAQ,KAAK;AAC/B,UAAM,QAAQ,EAAE,MAAM,CAAC;AACvB,UAAM,QAAQ,EAAE,MAAM,CAAC;AACvB,QAAI,CAAC,GAAG,OAAO,KAAK,GAAG;AACrB,aAAO;AAAA,QACL;AAAA,QACA;AAAA,QACA,sBAAsB;AAAA,QACtB;AAAA,QACA;AAAA,QACA,QAAQ,mBAAmB,OAAO,KAAK;AAAA,QACvC,iBAAiB;AAAA,MACnB;AAAA,IACF;AAAA,EACF;AACA,MAAI,EAAE,MAAM,WAAW,EAAE,MAAM,QAAQ;AACrC,WAAO,EAAE,MAAM,MAAM,sBAAsB,MAAM,iBAAiB,OAAO;AAAA,EAC3E;AACA,QAAM,SAAqB,EAAE,MAAM,SAAS,EAAE,MAAM,SAAS,IAAI;AACjE,QAAM,QAAQ,OAAO,MAAM,SAAS;AAGpC,QAAM,SACJ,WAAW,IACP,0CAA0C,KAAK,iCAC/C,sBAAsB,KAAK,6BAA6B,SAAS,CAAC;AACxE,SAAO;AAAA,IACL;AAAA,IACA;AAAA,IACA,sBAAsB;AAAA,IACtB,OAAO,EAAE,MAAM,MAAM;AAAA,IACrB,OAAO,EAAE,MAAM,MAAM;AAAA,IACrB;AAAA,IACA,iBAAiB;AAAA,EACnB;AACF;AAEA,SAAS,kBAAkB,GAAmB,GAA4B;AACxE,MAAI,EAAE,KAAK,SAAS,EAAE,KAAK,KAAM,QAAO;AACxC,MAAI,EAAE,KAAK,SAAS,UAAU,EAAE,KAAK,SAAS,OAAQ,QAAO,EAAE,KAAK,aAAa,EAAE,KAAK;AACxF,MAAI,EAAE,KAAK,SAAS,SAAS,EAAE,KAAK,SAAS,MAAO,QAAO,EAAE,KAAK,UAAU,EAAE,KAAK;AACnF,MAAI,EAAE,KAAK,SAAS,WAAW,EAAE,KAAK,SAAS;AAC7C,WAAO,EAAE,KAAK,cAAc,EAAE,KAAK;AACrC,SAAO,EAAE,KAAK,SAAS,EAAE,KAAK;AAChC;AAEA,SAAS,mBAAmB,GAAmB,GAA2B;AACxE,MAAI,EAAE,KAAK,SAAS,EAAE,KAAK,KAAM,QAAO,QAAQ,EAAE,KAAK,IAAI,OAAO,EAAE,KAAK,IAAI;AAC7E,MAAI,EAAE,KAAK,SAAS,UAAU,EAAE,KAAK,SAAS,UAAU,EAAE,KAAK,aAAa,EAAE,KAAK,UAAU;AAC3F,WAAO,QAAQ,EAAE,KAAK,QAAQ,OAAO,EAAE,KAAK,QAAQ;AAAA,EACtD;AACA,MAAI,EAAE,KAAK,SAAS,SAAS,EAAE,KAAK,SAAS,SAAS,EAAE,KAAK,UAAU,EAAE,KAAK,OAAO;AACnF,WAAO,SAAS,EAAE,KAAK,KAAK,OAAO,EAAE,KAAK,KAAK;AAAA,EACjD;AACA,SAAO,SAAS,EAAE,KAAK,IAAI,SAAS,EAAE,KAAK,IAAI;AACjD;;;AC9DA,eAAsB,mBAAmB,OAAkD;AACzF,QAAM,OAAO,MAAM,MAAM,MAAM,EAAE,MAAM,QAAQ,CAAC,GAAG;AAAA,IACjD,CAAC,MAAsB,EAAE,SAAS;AAAA,EACpC;AACA,MAAI,IAAI,WAAW,EAAG,QAAO,EAAE,OAAO,CAAC,GAAG,YAAY,CAAC,GAAG,UAAU,CAAC,EAAE;AAEvE,QAAM,cAAc,oBAAI,IAAyB;AACjD,aAAW,KAAK,KAAK;AACnB,UAAM,MAAM,YAAY,IAAI,EAAE,SAAS,KAAK,CAAC;AAC7C,QAAI,KAAK,CAAC;AACV,gBAAY,IAAI,EAAE,WAAW,GAAG;AAAA,EAClC;AAEA,QAAM,WAAW,CAAC,GAAG,IAAI,IAAI,IAAI,IAAI,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC,EAAE,KAAK;AAC9D,QAAM,QAAqB,CAAC;AAC5B,aAAW,CAAC,KAAK,KAAK,KAAK,aAAa;AACtC,UAAM,UAAU,oBAAI,IAAiC;AACrD,eAAW,KAAK,OAAO;AACrB,YAAM,IAAI,QAAQ,IAAI,EAAE,OAAO,KAAK,oBAAI,IAAoB;AAC5D,QAAE,IAAI,EAAE,cAAc,EAAE,KAAK;AAC7B,cAAQ,IAAI,EAAE,SAAS,CAAC;AAAA,IAC1B;AACA,UAAM,aAAa,CAAC,GAAG,QAAQ,KAAK,CAAC;AACrC,aAAS,IAAI,GAAG,IAAI,WAAW,QAAQ,KAAK;AAC1C,eAAS,IAAI,IAAI,GAAG,IAAI,WAAW,QAAQ,KAAK;AAC9C,cAAM,SAAS,WAAW,CAAC;AAC3B,cAAM,SAAS,WAAW,CAAC;AAC3B,cAAM,IAAI,QAAQ,IAAI,MAAM;AAC5B,cAAM,IAAI,QAAQ,IAAI,MAAM;AAC5B,cAAM,SAAkC,CAAC;AACzC,mBAAW,CAAC,QAAQ,MAAM,KAAK,GAAG;AAChC,gBAAM,SAAS,EAAE,IAAI,MAAM;AAC3B,cAAI,WAAW,OAAW,QAAO,KAAK,CAAC,QAAQ,MAAM,CAAC;AAAA,QACxD;AACA,YAAI,OAAO,SAAS,EAAG;AACvB,cAAM,cAAc,OAAO;AAAA,UACzB,CAAC,CAAC,QAAQ,MAAM,MACd;AAAA,YACE,EAAE,WAAW,QAAQ,WAAW,KAAK,OAAO,QAAQ,WAAW,GAAG;AAAA,YAClE,EAAE,WAAW,QAAQ,WAAW,KAAK,OAAO,QAAQ,WAAW,GAAG;AAAA,UACpE;AAAA,QACJ;AACA,cAAM,IAAI;AAAA,UACR,YAAY,CAAC,EAAG,IAAI,CAAC,GAAG,OAAO,YAAY,IAAI,CAAC,SAAS,KAAK,EAAE,CAAE,CAAC;AAAA,QACrE;AACA,cAAM,KAAK;AAAA,UACT,QAAQ;AAAA,UACR,QAAQ;AAAA,UACR,WAAW;AAAA,UACX,aAAa,OAAO;AAAA,UACpB,SAAS;AAAA,YACP,OAAO,IAAI,CAAC,MAAM,EAAE,CAAC,CAAC;AAAA,YACtB,OAAO,IAAI,CAAC,MAAM,EAAE,CAAC,CAAC;AAAA,UACxB;AAAA,UACA,cAAc;AAAA,QAChB,CAAC;AAAA,MACH;AAAA,IACF;AAAA,EACF;AAEA,SAAO;AAAA,IACL,OAAO,MAAM,KAAK,CAAC,GAAG,MAAM,EAAE,cAAc,EAAE,WAAW;AAAA,IACzD,YAAY,CAAC,GAAG,YAAY,KAAK,CAAC,EAAE,KAAK;AAAA,IACzC;AAAA,EACF;AACF;AAEA,SAAS,QAAQ,GAAa,GAAqB;AACjD,MAAI,EAAE,WAAW,EAAE,UAAU,EAAE,SAAS,EAAG,QAAO;AAClD,QAAM,KAAK,EAAE,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,EAAE;AAC5C,QAAM,KAAK,EAAE,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,EAAE;AAC5C,MAAI,MAAM,GACR,OAAO,GACP,OAAO;AACT,WAAS,IAAI,GAAG,IAAI,EAAE,QAAQ,KAAK;AACjC,UAAM,KAAK,EAAE,CAAC,IAAK;AACnB,UAAM,KAAK,EAAE,CAAC,IAAK;AACnB,WAAO,KAAK;AACZ,YAAQ,KAAK;AACb,YAAQ,KAAK;AAAA,EACf;AACA,MAAI,SAAS,KAAK,SAAS,EAAG,QAAO,SAAS,KAAK,SAAS,IAAI,IAAI;AACpE,SAAO,MAAM,KAAK,KAAK,OAAO,IAAI;AACpC;;;ACvFA,eAAsB,eACpB,OACA,SACA,SACyB;AACzB,QAAM,eAAe,MAAM,MAAM,SAAS,QAAQ,QAAQ;AAC1D,QAAM,gBAAgB,MAAM,MAAM,SAAS,QAAQ,SAAS;AAC5D,QAAM,UAAU,MAAM,QAAQ;AAAA,IAC5B,QAAQ,IAAI,OAAO,MAAM;AACvB,YAAM,UAAU,EAAE,WAAW,eAAe,EAAE,MAAM;AACpD,YAAM,WAAW,MAAM,WAAW,cAAc,SAAS,KAAK;AAC9D,YAAM,YAAY,MAAM,WAAW,eAAe,SAAS,KAAK;AAChE,aAAO,EAAE,QAAQ,EAAE,QAAQ,gBAAgB,EAAE,gBAAgB,UAAU,UAAU;AAAA,IACnF,CAAC;AAAA,EACH;AACA,SAAO,kBAAkB,SAAS,OAAO;AAC3C;AAEA,eAAe,WACb,MACA,SACA,OACmB;AACnB,QAAM,MAAgB,CAAC;AACvB,aAAW,KAAK,MAAM;AACpB,UAAM,IAAI,MAAM,QAAQ,GAAG,KAAK;AAChC,QAAI,MAAM,QAAQ,OAAO,SAAS,CAAC,EAAG,KAAI,KAAK,CAAC;AAAA,EAClD;AACA,SAAO;AACT;AAEA,SAAS,eAAe,QAAyE;AAC/F,SAAO,OAAO,KAAK,UAAU;AAC3B,YAAQ,QAAQ;AAAA,MACd,KAAK;AAAA,MACL,KAAK;AACH,eAAO,IAAI,SAAS,SAAS;AAAA,MAC/B,KAAK;AACH,eAAO,IAAI,SAAS,SAAS,OAAO,IAAI;AAAA,MAC1C,KAAK;AACH,eAAO,IAAI,WAAW,IAAI,YAAY,IAAI,UAAU,IAAI,YAAY;AAAA,MACtE,KAAK,WAAW;AACd,cAAM,MAAM,MAAM,SAAS,OAAO,IAAI,KAAK;AAC3C,eAAO,aAAa,GAAG,EAAE;AAAA,MAC3B;AAAA,MACA,KAAK,eAAe;AAClB,cAAM,MAAM,MAAM,SAAS,OAAO,IAAI,KAAK;AAC3C,eAAO,aAAa,GAAG,EAAE;AAAA,MAC3B;AAAA,MACA,KAAK,gBAAgB;AACnB,cAAM,MAAM,MAAM,SAAS,OAAO,IAAI,KAAK;AAC3C,eAAO,aAAa,GAAG,EAAE;AAAA,MAC3B;AAAA,MACA,KAAK,gBAAgB;AACnB,eAAO,gBAAgB,GAAG,MAAM,YAAY,IAAI;AAAA,MAClD;AAAA,MACA;AACE,eAAO;AAAA,IACX;AAAA,EACF;AACF;;;AClDA,eAAsB,cACpB,OACA,UAA4B,CAAC,GACH;AAC1B,QAAM,iBAAiB,QAAQ,kBAAkB;AACjD,QAAM,OAAO,QAAQ,QACjB,CAAC,EAAE,OAAO,QAAQ,MAAM,CAAC,KACxB,MAAM,MAAM,SAAS,GAAG,IAAI,CAAC,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE;AAE5D,QAAM,WAA+B,CAAC;AACtC,aAAW,EAAE,MAAM,KAAK,MAAM;AAC5B,UAAM,QAAQ,MAAM,UAAU,OAAO,KAAK;AAC1C,UAAM,QAAQ,oBAAI,IAAwE;AAC1F,eAAW,KAAK,OAAO;AACrB,YAAM,IAAI,QAAQ,EAAE,IAAI;AAKxB,YAAM,MAAM,GAAG,EAAE,QAAQ,KAAS,CAAC;AACnC,YAAM,SAAS,MAAM,IAAI,GAAG,KAAK,EAAE,OAAO,CAAC,GAAG,SAAS,GAAG,UAAU,EAAE,SAAS;AAC/E,aAAO,MAAM,KAAK,CAAC;AACnB,YAAM,IAAI,KAAK,MAAM;AAAA,IACvB;AACA,eAAW,EAAE,OAAO,SAAS,GAAG,SAAS,KAAK,MAAM,OAAO,GAAG;AAC5D,UAAI,MAAM,SAAS,eAAgB;AACnC,YAAM,SAAS,CAAC,GAAG,KAAK,EAAE,KAAK,CAAC,GAAG,MAAM,EAAE,YAAY,EAAE,SAAS;AAClE,YAAM,QAAQ,OAAO,CAAC,EAAG;AACzB,YAAM,OAAO,OAAO,OAAO,SAAS,CAAC,EAAG;AACxC,eAAS,KAAK;AAAA,QACZ;AAAA,QACA;AAAA,QACA,SAAS;AAAA,QACT,aAAa,OAAO;AAAA,QACpB,SAAS,OAAO,IAAI,CAAC,MAAM,EAAE,MAAM;AAAA,QACnC,UAAU,OAAO;AAAA,MACnB,CAAC;AAAA,IACH;AAAA,EACF;AAEA,QAAM,eAAe,IAAI,IAAI,SAAS,IAAI,CAAC,MAAM,EAAE,KAAK,CAAC;AACzD,SAAO;AAAA,IACL;AAAA,IACA,kBAAkB,KAAK,SAAS,IAAI,aAAa,OAAO,KAAK,SAAS;AAAA,IACtE,WAAW,KAAK;AAAA,EAClB;AACF;;;AChDA,eAAsB,cACpB,OACA,UAA4B,CAAC,GACH;AAC1B,QAAM,OAAO,QAAQ,QAAQ,CAAC,QAAQ,KAAK,KAAK,MAAM,MAAM,SAAS,GAAG,IAAI,CAAC,MAAM,EAAE,KAAK;AAE1F,QAAM,QAA4B,CAAC;AACnC,MAAI,aAAa;AACjB,MAAI,cAAc;AAClB,aAAW,SAAS,MAAM;AACxB,UAAM,QAAQ,MAAM,UAAU,OAAO,KAAK;AAC1C,QAAI,MAAM,WAAW,GAAG;AACtB,YAAM,KAAK,EAAE,OAAO,aAAa,GAAG,YAAY,GAAG,WAAW,EAAE,CAAC;AACjE;AAAA,IACF;AACA,UAAM,OAAO,MAAM,SAAS,OAAO,KAAK;AAOxC,UAAM,YAAY,CAAC,GAAG,IAAI,EAAE,KAAK,CAAC,GAAG,MAAM,EAAE,YAAY,EAAE,SAAS;AACpE,UAAM,aAAa,UAAU,IAAI,CAAC,MAAM,EAAE,SAAS;AACnD,UAAM,aAAa,gBAAgB,SAAS;AAC5C,QAAI,SAAS;AACb,eAAW,KAAK,OAAO;AACrB,UAAI,EAAE,WAAW,SAAS;AACxB;AACA;AAAA,MACF;AAEA,YAAM,SAAS,WAAW,YAAY,EAAE,SAAS;AACjD,UAAI,QAAQ,aAAa;AACvB,YAAI,CAAC,QAAQ,YAAY,GAAG,EAAE,KAAK,UAAU,MAAM,MAAM,EAAE,CAAC,EAAG;AAAA,MACjE,OAAO;AAKL,cAAM,YAAY,UAAU,EAAE,MAAM;AACpC,YAAI,cAAc,GAAI;AACtB,cAAM,WAAW,WAAW,MAAM,KAAK;AACvC,cAAM,OAAO,SAAS,SAAS,UAAU,MAAM,GAAG,GAAG,CAAC;AACtD,YAAI,CAAC,KAAM;AAAA,MACb;AAAA,IACF;AACA,UAAM,YAAY,SAAS,MAAM;AACjC,UAAM,KAAK,EAAE,OAAO,aAAa,QAAQ,YAAY,MAAM,QAAQ,UAAU,CAAC;AAC9E,kBAAc,MAAM;AACpB,mBAAe;AAAA,EACjB;AACA,SAAO,EAAE,OAAO,kBAAkB,aAAa,IAAI,cAAc,aAAa,EAAE;AAClF;AAOA,SAAS,gBAAgB,OAA4B;AACnD,QAAM,SAAS,IAAI,MAAc,MAAM,SAAS,CAAC;AACjD,SAAO,MAAM,MAAM,IAAI;AACvB,WAAS,IAAI,MAAM,SAAS,GAAG,KAAK,GAAG,KAAK;AAC1C,UAAM,MAAM,MAAM,CAAC,EAAG,SAAS;AAAA,MAAI,CAAC,MAClC,OAAO,EAAE,YAAY,WAAW,EAAE,UAAU;AAAA,IAC9C,EAAE,KAAK,IAAI;AACX,WAAO,CAAC,IAAI,GAAG,GAAG;AAAA,EAAK,OAAO,IAAI,CAAC,CAAC;AAAA,EACtC;AACA,SAAO;AACT;AAGA,SAAS,WAAW,QAAkB,QAAwB;AAC5D,MAAI,KAAK;AACT,MAAI,KAAK,OAAO;AAChB,SAAO,KAAK,IAAI;AACd,UAAM,MAAO,KAAK,OAAQ;AAC1B,QAAI,OAAO,GAAG,KAAM,OAAQ,MAAK,MAAM;AAAA,QAClC,MAAK;AAAA,EACZ;AACA,SAAO;AACT;AAEA,SAAS,UAAU,GAAoB;AACrC,MAAI,MAAM,QAAQ,MAAM,OAAW,QAAO;AAC1C,MAAI,OAAO,MAAM,SAAU,QAAO;AAClC,MAAI;AACF,WAAO,KAAK,UAAU,CAAC;AAAA,EACzB,QAAQ;AACN,WAAO,OAAO,CAAC;AAAA,EACjB;AACF;","names":[]}
+{"version":3,"sources":["../../src/pipelines/budget-breach.ts","../../src/pipelines/failure-cluster.ts","../../src/pipelines/first-divergence.ts","../../src/pipelines/judge-agreement.ts","../../src/pipelines/regression.ts","../../src/pipelines/stuck-loop.ts","../../src/pipelines/tool-waste.ts"],"sourcesContent":["/**\n * BudgetBreachView — aggregates breach events across the corpus.\n *\n * Answers: which dimensions get hit most often? Which scenarios are\n * underbudgeted? Which variants trigger the most breaches?\n */\n\nimport type { BudgetSpec } from '../trace/schema'\nimport type { TraceStore } from '../trace/store'\n\nexport interface BudgetBreachFinding {\n  runId: string\n  scenarioId: string\n  variantId?: string\n  dimension: keyof BudgetSpec\n  limit: number\n  consumed: number\n  excessRatio: number\n  timestamp: number\n}\n\nexport interface BudgetBreachReport {\n  findings: BudgetBreachFinding[]\n  byDimension: Record<string, number>\n  byScenario: Record<string, number>\n  byVariant: Record<string, number>\n  totalRuns: number\n  breachedRunRatio: number\n}\n\nexport async function budgetBreachView(\n  store: TraceStore,\n  options: { scenarioId?: string; variantId?: string } = {},\n): Promise<BudgetBreachReport> {\n  const runs = await store.listRuns({\n    scenarioId: options.scenarioId,\n    variantId: options.variantId,\n  })\n  const findings: BudgetBreachFinding[] = []\n  const byDimension: Record<string, number> = {}\n  const byScenario: Record<string, number> = {}\n  const byVariant: Record<string, number> = {}\n\n  for (const run of runs) {\n    const entries = await store.budget(run.runId)\n    for (const e of entries) {\n      if (!e.breached) continue\n      const excessRatio = e.limit > 0 ? e.consumed / e.limit : Infinity\n      findings.push({\n        runId: run.runId,\n        scenarioId: run.scenarioId,\n        variantId: run.variantId,\n        dimension: e.dimension,\n        limit: e.limit,\n        consumed: e.consumed,\n        excessRatio,\n        timestamp: e.timestamp,\n      })\n      byDimension[e.dimension] = (byDimension[e.dimension] ?? 0) + 1\n      byScenario[run.scenarioId] = (byScenario[run.scenarioId] ?? 0) + 1\n      if (run.variantId) byVariant[run.variantId] = (byVariant[run.variantId] ?? 0) + 1\n    }\n  }\n\n  const breachedRuns = new Set(findings.map((f) => f.runId))\n  return {\n    findings,\n    byDimension,\n    byScenario,\n    byVariant,\n    totalRuns: runs.length,\n    breachedRunRatio: runs.length > 0 ? breachedRuns.size / runs.length : 0,\n  }\n}\n","/**\n * FailureClusterView — groups failed runs by (failureClass, triggerTool,\n * argHash-prefix) so weekly reviews can prioritize the top-N clusters.\n *\n * Each cluster includes: N runs, scenarios affected, representative\n * error message, a proposed mitigation hint (rule → action table).\n */\n\nimport { classifyFailure, DEFAULT_RULES, type FailureRule } from '../failure-taxonomy'\nimport { argHash, toolSpans } from '../trace/query'\nimport type { FailureClass, Span } from '../trace/schema'\nimport type { TraceStore } from '../trace/store'\n\nexport interface FailureCluster {\n  failureClass: FailureClass\n  /** Tool name when the trigger was a tool span, else undefined. */\n  toolName?: string\n  /** First 16 chars of argHash — clusters similar args. */\n  argPrefix?: string\n  /**\n   * Source dimension when the trigger was a judge span (e.g. `'format'`,\n   * `'safety'`, `'correctness'`). Lets cross-template aggregators\n   * group failures by the dimension that fired without overloading\n   * `argPrefix`. Optional — clusters without this field deserialize cleanly.\n   */\n  dimension?: string\n  runCount: number\n  scenarioIds: string[]\n  exampleError?: string\n  exampleRunId: string\n}\n\nexport interface FailureClusterReport {\n  clusters: FailureCluster[]\n  totalFailures: number\n  totalRuns: number\n}\n\nexport async function failureClusterView(\n  store: TraceStore,\n  options: { rules?: FailureRule[]; minClusterSize?: number } = {},\n): Promise<FailureClusterReport> {\n  const rules = options.rules ?? DEFAULT_RULES\n  const minSize = options.minClusterSize ?? 1\n  const runs = await store.listRuns()\n\n  type Key = string\n  const clusters = new Map<Key, FailureCluster>()\n  let totalFailures = 0\n\n  for (const run of runs) {\n    if (run.status === 'completed' && run.outcome?.pass !== false) continue\n    totalFailures++\n    const spans = await store.spans({ runId: run.runId })\n    const events = await store.events({ runId: run.runId })\n    const cls = classifyFailure({ run, spans, events }, rules)\n\n    let toolName: string | undefined\n    let argPrefix: string | undefined\n    let dimension: string | undefined\n    if (cls.triggerSpanId) {\n      const trig = spans.find((s) => s.spanId === cls.triggerSpanId)\n      if (trig?.kind === 'tool') {\n        toolName = trig.toolName\n        argPrefix = argHash(trig.args).slice(0, 16)\n      } else if (trig?.kind === 'judge') {\n        dimension = trig.dimension\n      }\n    }\n    // Fallback: look at the last errored tool span\n    if (!toolName) {\n      const ts = await toolSpans(store, run.runId)\n      const errored = ts.filter((t) => t.status === 'error').pop()\n      if (errored) {\n        toolName = errored.toolName\n        argPrefix = argHash(errored.args).slice(0, 16)\n      }\n    }\n    // Secondary signal: any judge span on the failed run carries a\n    // dimension. Useful when the rule classified by judge score but\n    // didn't surface the trigger span (or surfaced a non-judge span).\n    if (!dimension) {\n      const judge = spans.find((s) => s.kind === 'judge' && typeof s.dimension === 'string')\n      if (judge?.kind === 'judge') dimension = judge.dimension\n    }\n\n    const key = `${cls.failureClass}|${toolName ?? ''}|${argPrefix ?? ''}|${dimension ?? ''}`\n    let cluster = clusters.get(key)\n    if (!cluster) {\n      cluster = {\n        failureClass: cls.failureClass,\n        toolName,\n        argPrefix,\n        dimension,\n        runCount: 0,\n        scenarioIds: [],\n        exampleRunId: run.runId,\n        exampleError: firstErrorMessage(spans) ?? cls.reason,\n      }\n      clusters.set(key, cluster)\n    }\n    cluster.runCount++\n    if (!cluster.scenarioIds.includes(run.scenarioId)) cluster.scenarioIds.push(run.scenarioId)\n  }\n\n  const arr = [...clusters.values()]\n    .filter((c) => c.runCount >= minSize)\n    .sort((a, b) => b.runCount - a.runCount)\n\n  return { clusters: arr, totalFailures, totalRuns: runs.length }\n}\n\nfunction firstErrorMessage(spans: Span[]): string | undefined {\n  const errored = spans.find((s) => s.status === 'error')\n  return errored?.error\n}\n","/**\n * FirstDivergenceView — aligns two trajectories by step index, reports\n * the first step where they differ.\n *\n * \"Differ\" is configurable — default is (kind, toolName if tool, model\n * if llm). Use this view to attribute \"why is variant B better?\" to a\n * specific step rather than an aggregate mean delta.\n */\n\nimport type { TraceStore } from '../trace/store'\nimport { buildTrajectory, type Trajectory, type TrajectoryStep } from '../trajectory'\n\nexport interface DivergenceReport {\n  runA: string\n  runB: string\n  firstDivergenceIndex: number | null\n  aStep?: TrajectoryStep\n  bStep?: TrajectoryStep\n  reason?: string\n  /** Common prefix length (steps that matched). */\n  commonPrefixLen: number\n}\n\nexport interface DivergenceOptions {\n  /** Returns true if two steps are considered equal. Default: kind + tool/model match. */\n  stepEquals?: (a: TrajectoryStep, b: TrajectoryStep) => boolean\n}\n\nexport async function firstDivergenceView(\n  store: TraceStore,\n  runA: string,\n  runB: string,\n  options: DivergenceOptions = {},\n): Promise<DivergenceReport> {\n  const [a, b] = await Promise.all([buildTrajectory(store, runA), buildTrajectory(store, runB)])\n  const eq = options.stepEquals ?? defaultStepEquals\n  const minLen = Math.min(a.steps.length, b.steps.length)\n  for (let i = 0; i < minLen; i++) {\n    const aStep = a.steps[i]!\n    const bStep = b.steps[i]!\n    if (!eq(aStep, bStep)) {\n      return {\n        runA,\n        runB,\n        firstDivergenceIndex: i,\n        aStep,\n        bStep,\n        reason: describeDifference(aStep, bStep),\n        commonPrefixLen: i,\n      }\n    }\n  }\n  if (a.steps.length === b.steps.length) {\n    return { runA, runB, firstDivergenceIndex: null, commonPrefixLen: minLen }\n  }\n  const longer: Trajectory = a.steps.length > b.steps.length ? a : b\n  const extra = longer.steps.length - minLen\n  // minLen === 0 means one trajectory is empty: divergence is the first step\n  // itself (index 0), and the absent side has no step to surface.\n  const reason =\n    minLen === 0\n      ? `one trajectory is empty; the other has ${extra} step(s) starting at index 0`\n      : `one trajectory has ${extra} more step(s) after index ${minLen - 1}`\n  return {\n    runA,\n    runB,\n    firstDivergenceIndex: minLen,\n    aStep: a.steps[minLen],\n    bStep: b.steps[minLen],\n    reason,\n    commonPrefixLen: minLen,\n  }\n}\n\nfunction defaultStepEquals(a: TrajectoryStep, b: TrajectoryStep): boolean {\n  if (a.span.kind !== b.span.kind) return false\n  if (a.span.kind === 'tool' && b.span.kind === 'tool') return a.span.toolName === b.span.toolName\n  if (a.span.kind === 'llm' && b.span.kind === 'llm') return a.span.model === b.span.model\n  if (a.span.kind === 'judge' && b.span.kind === 'judge')\n    return a.span.dimension === b.span.dimension\n  return a.span.name === b.span.name\n}\n\nfunction describeDifference(a: TrajectoryStep, b: TrajectoryStep): string {\n  if (a.span.kind !== b.span.kind) return `kind ${a.span.kind} vs ${b.span.kind}`\n  if (a.span.kind === 'tool' && b.span.kind === 'tool' && a.span.toolName !== b.span.toolName) {\n    return `tool ${a.span.toolName} vs ${b.span.toolName}`\n  }\n  if (a.span.kind === 'llm' && b.span.kind === 'llm' && a.span.model !== b.span.model) {\n    return `model ${a.span.model} vs ${b.span.model}`\n  }\n  return `name \"${a.span.name}\" vs \"${b.span.name}\"`\n}\n","/**\n * JudgeAgreementView — pairwise agreement between judges across the\n * corpus, grouped by dimension.\n *\n * Output drives two workflows:\n *   - Judge robustness audit: \"does Claude agree with GPT at κ ≥ 0.6?\"\n *   - Calibration tracking: κ vs golden human labels over time (by\n *     providing a `humanGoldenJudgeId`).\n */\n\nimport { interRaterReliability, pearsonR } from '../statistics'\nimport type { JudgeSpan } from '../trace/schema'\nimport type { TraceStore } from '../trace/store'\n\nexport interface JudgePair {\n  judgeA: string\n  judgeB: string\n  dimension: string\n  /** Number of (targetSpanId, dimension) tuples both judges scored. */\n  commonItems: number\n  pearson: number\n  krippendorff: number\n}\n\nexport interface JudgeAgreementReport {\n  pairs: JudgePair[]\n  dimensions: string[]\n  judgeIds: string[]\n}\n\nexport async function judgeAgreementView(store: TraceStore): Promise<JudgeAgreementReport> {\n  const all = (await store.spans({ kind: 'judge' })).filter(\n    (s): s is JudgeSpan => s.kind === 'judge',\n  )\n  if (all.length === 0) return { pairs: [], dimensions: [], judgeIds: [] }\n\n  const byDimension = new Map<string, JudgeSpan[]>()\n  for (const s of all) {\n    const arr = byDimension.get(s.dimension) ?? []\n    arr.push(s)\n    byDimension.set(s.dimension, arr)\n  }\n\n  const judgeIds = [...new Set(all.map((s) => s.judgeId))].sort()\n  const pairs: JudgePair[] = []\n  for (const [dim, spans] of byDimension) {\n    const byJudge = new Map<string, Map<string, number>>()\n    for (const s of spans) {\n      const m = byJudge.get(s.judgeId) ?? new Map<string, number>()\n      m.set(s.targetSpanId, s.score)\n      byJudge.set(s.judgeId, m)\n    }\n    const judgesHere = [...byJudge.keys()]\n    for (let i = 0; i < judgesHere.length; i++) {\n      for (let j = i + 1; j < judgesHere.length; j++) {\n        const judgeI = judgesHere[i]!\n        const judgeJ = judgesHere[j]!\n        const a = byJudge.get(judgeI)!\n        const b = byJudge.get(judgeJ)!\n        const common: Array<[number, number]> = []\n        for (const [target, scoreA] of a) {\n          const scoreB = b.get(target)\n          if (scoreB !== undefined) common.push([scoreA, scoreB])\n        }\n        if (common.length < 2) continue\n        const judgeScores = common.map(\n          ([scoreA, scoreB]) =>\n            [\n              { judgeName: judgeI, dimension: dim, score: scoreA, reasoning: '' },\n              { judgeName: judgeJ, dimension: dim, score: scoreB, reasoning: '' },\n            ] as const,\n        )\n        const k = interRaterReliability(\n          judgeScores[0]!.map((_, k2) => judgeScores.map((pair) => pair[k2]!)),\n        )\n        pairs.push({\n          judgeA: judgeI,\n          judgeB: judgeJ,\n          dimension: dim,\n          commonItems: common.length,\n          pearson: pearsonR(\n            common.map((c) => c[0]),\n            common.map((c) => c[1]),\n          ),\n          krippendorff: k,\n        })\n      }\n    }\n  }\n\n  return {\n    pairs: pairs.sort((a, b) => b.commonItems - a.commonItems),\n    dimensions: [...byDimension.keys()].sort(),\n    judgeIds,\n  }\n}\n","/**\n * RegressionView — compares a candidate slice to a baseline slice on a\n * named metric. Delegates the statistics (Welch's t-test, Cohen's d,\n * IQR stability) to `baseline.ts`.\n *\n * This is the entry point for CI regression gates: \"given runs tagged\n * release=A and release=B, did any metric regress?\"\n */\n\nimport { type BaselineOptions, type BaselineReport, compareToBaseline } from '../baseline'\nimport { aggregateLlm, llmSpans, runFailureClass } from '../trace/query'\nimport type { Run } from '../trace/schema'\nimport type { RunFilter, TraceStore } from '../trace/store'\n\nexport interface RegressionSpec {\n  metric: string\n  higherIsBetter: boolean\n  /** Extract a scalar from a run. Default extractors handle common metrics. */\n  extract?: (run: Run, store: TraceStore) => Promise<number | null>\n}\n\nexport interface RegressionOptions extends BaselineOptions {\n  baseline: RunFilter\n  candidate: RunFilter\n}\n\nexport async function regressionView(\n  store: TraceStore,\n  metrics: RegressionSpec[],\n  options: RegressionOptions,\n): Promise<BaselineReport> {\n  const baselineRuns = await store.listRuns(options.baseline)\n  const candidateRuns = await store.listRuns(options.candidate)\n  const samples = await Promise.all(\n    metrics.map(async (m) => {\n      const extract = m.extract ?? defaultExtract(m.metric)\n      const baseline = await extractAll(baselineRuns, extract, store)\n      const candidate = await extractAll(candidateRuns, extract, store)\n      return { metric: m.metric, higherIsBetter: m.higherIsBetter, baseline, candidate }\n    }),\n  )\n  return compareToBaseline(samples, options)\n}\n\nasync function extractAll(\n  runs: Run[],\n  extract: (r: Run, s: TraceStore) => Promise<number | null>,\n  store: TraceStore,\n): Promise<number[]> {\n  const out: number[] = []\n  for (const r of runs) {\n    const v = await extract(r, store)\n    if (v !== null && Number.isFinite(v)) out.push(v)\n  }\n  return out\n}\n\nfunction defaultExtract(metric: string): (run: Run, store: TraceStore) => Promise<number | null> {\n  return async (run, store) => {\n    switch (metric) {\n      case 'score':\n      case 'overallScore':\n        return run.outcome?.score ?? null\n      case 'pass':\n        return run.outcome?.pass === true ? 1 : 0\n      case 'durationMs':\n        return run.endedAt && run.startedAt ? run.endedAt - run.startedAt : null\n      case 'costUsd': {\n        const llm = await llmSpans(store, run.runId)\n        return aggregateLlm(llm).costUsd\n      }\n      case 'inputTokens': {\n        const llm = await llmSpans(store, run.runId)\n        return aggregateLlm(llm).inputTokens\n      }\n      case 'outputTokens': {\n        const llm = await llmSpans(store, run.runId)\n        return aggregateLlm(llm).outputTokens\n      }\n      case 'failureClass': {\n        return runFailureClass(run) === 'success' ? 1 : 0\n      }\n      default:\n        return null\n    }\n  }\n}\n","/**\n * StuckLoopView — detects when an agent calls the same tool with the\n * same (or structurally similar) arguments ≥ N times in a short window.\n *\n * Rationale: agents that loop are the number-one production failure\n * mode on long-horizon flows. The view returns (runId, toolName,\n * argHash, occurrences, windowMs) for each detected loop plus a\n * fraction of runs affected.\n */\n\nimport { argHash, toolSpans } from '../trace/query'\nimport type { TraceStore } from '../trace/store'\n\nexport interface StuckLoopFinding {\n  runId: string\n  toolName: string\n  argHash: string\n  occurrences: number\n  spanIds: string[]\n  /** Milliseconds between first and last call in the loop. */\n  windowMs: number\n}\n\nexport interface StuckLoopReport {\n  findings: StuckLoopFinding[]\n  affectedRunRatio: number\n  totalRuns: number\n}\n\nexport interface StuckLoopOptions {\n  /** Minimum call count to flag a loop (default 3). */\n  minOccurrences?: number\n  /** Filter to a specific runId; omit to scan the entire corpus. */\n  runId?: string\n}\n\nexport async function stuckLoopView(\n  store: TraceStore,\n  options: StuckLoopOptions = {},\n): Promise<StuckLoopReport> {\n  const minOccurrences = options.minOccurrences ?? 3\n  const runs = options.runId\n    ? [{ runId: options.runId }]\n    : (await store.listRuns()).map((r) => ({ runId: r.runId }))\n\n  const findings: StuckLoopFinding[] = []\n  for (const { runId } of runs) {\n    const tools = await toolSpans(store, runId)\n    const byKey = new Map<string, { spans: typeof tools; argHash: string; toolName: string }>()\n    for (const t of tools) {\n      const h = argHash(t.args)\n      // NUL delimiter never appears in a tool name or a JSON argHash, so the\n      // key cannot collide; toolName is carried in the bucket rather than\n      // re-derived by splitting the key (which mislabeled any tool whose name\n      // contained the delimiter, e.g. \"shell|grep\").\n      const key = `${t.toolName}\\u0000${h}`\n      const bucket = byKey.get(key) ?? { spans: [], argHash: h, toolName: t.toolName }\n      bucket.spans.push(t)\n      byKey.set(key, bucket)\n    }\n    for (const { spans, argHash: h, toolName } of byKey.values()) {\n      if (spans.length < minOccurrences) continue\n      const sorted = [...spans].sort((a, b) => a.startedAt - b.startedAt)\n      const first = sorted[0]!.startedAt\n      const last = sorted[sorted.length - 1]!.startedAt\n      findings.push({\n        runId,\n        toolName,\n        argHash: h,\n        occurrences: sorted.length,\n        spanIds: sorted.map((s) => s.spanId),\n        windowMs: last - first,\n      })\n    }\n  }\n\n  const affectedRuns = new Set(findings.map((f) => f.runId))\n  return {\n    findings,\n    affectedRunRatio: runs.length > 0 ? affectedRuns.size / runs.length : 0,\n    totalRuns: runs.length,\n  }\n}\n","/**\n * ToolWasteView — fraction of tool calls whose results weren't used\n * downstream. Without a \"used\" signal we fall back to structural\n * proxies: error calls, duplicate calls, and tool calls followed by\n * zero subsequent LLM spans are all considered waste.\n *\n * Consumers can pass a `usageOracle` that inspects a tool span and\n * returns true iff the tool's result appears in a later LLM message,\n * artifact, or state mutation — that's the canonical definition; the\n * default heuristic is a reasonable fallback.\n */\n\nimport { computeToolUseMetrics } from '../tool-use-metrics'\nimport { llmSpans, toolSpans } from '../trace/query'\nimport type { LlmSpan, ToolSpan } from '../trace/schema'\nimport type { TraceStore } from '../trace/store'\n\nexport interface ToolWasteFinding {\n  runId: string\n  wastedCalls: number\n  totalCalls: number\n  wasteRate: number\n}\n\nexport interface ToolWasteReport {\n  byRun: ToolWasteFinding[]\n  overallWasteRate: number\n}\n\nexport interface ToolWasteOptions {\n  runId?: string\n  usageOracle?: (tool: ToolSpan, later: { llm: Awaited<ReturnType<typeof llmSpans>> }) => boolean\n}\n\nexport async function toolWasteView(\n  store: TraceStore,\n  options: ToolWasteOptions = {},\n): Promise<ToolWasteReport> {\n  const runs = options.runId ? [options.runId] : (await store.listRuns()).map((r) => r.runId)\n\n  const byRun: ToolWasteFinding[] = []\n  let totalCalls = 0\n  let totalWasted = 0\n  for (const runId of runs) {\n    const tools = await toolSpans(store, runId)\n    if (tools.length === 0) {\n      byRun.push({ runId, wastedCalls: 0, totalCalls: 0, wasteRate: 0 })\n      continue\n    }\n    const llms = await llmSpans(store, runId)\n    // Sort LLM spans once by start time, then build a suffix index of the\n    // concatenated message text. `suffixText[i]` is the haystack of every\n    // string message content in spans[i..]. Per tool we binary-search the\n    // first span started strictly after the tool, then test that one suffix\n    // — turning the per-tool O(llms × messages × content) scan into a single\n    // O(log llms) lookup over precomputed text.\n    const sortedLlm = [...llms].sort((a, b) => a.startedAt - b.startedAt)\n    const startTimes = sortedLlm.map((l) => l.startedAt)\n    const suffixText = buildSuffixText(sortedLlm)\n    let wasted = 0\n    for (const t of tools) {\n      if (t.status === 'error') {\n        wasted++\n        continue\n      }\n      // First LLM span started strictly after this tool (upper-bound search).\n      const cutoff = upperBound(startTimes, t.startedAt)\n      if (options.usageOracle) {\n        if (!options.usageOracle(t, { llm: sortedLlm.slice(cutoff) })) wasted++\n      } else {\n        // Default heuristic: a tool whose result is NOT mentioned in any\n        // later LLM input message is likely wasted. An empty/null result has\n        // no payload to propagate downstream — there is nothing to find in a\n        // later message, so it is not evidence of waste; skip it.\n        const resultStr = stringify(t.result)\n        if (resultStr === '') continue\n        const haystack = suffixText[cutoff] ?? ''\n        const used = haystack.includes(resultStr.slice(0, 120))\n        if (!used) wasted++\n      }\n    }\n    const wasteRate = wasted / tools.length\n    byRun.push({ runId, wastedCalls: wasted, totalCalls: tools.length, wasteRate })\n    totalCalls += tools.length\n    totalWasted += wasted\n  }\n  return { byRun, overallWasteRate: totalCalls > 0 ? totalWasted / totalCalls : 0 }\n}\n\n/**\n * Build per-position suffix haystacks: result[i] is the concatenation of every\n * string message content in spans[i..end]. Built back-to-front so each entry\n * reuses the next one — O(total message text) rather than O(spans²).\n */\nfunction buildSuffixText(spans: LlmSpan[]): string[] {\n  const result = new Array<string>(spans.length + 1)\n  result[spans.length] = ''\n  for (let i = spans.length - 1; i >= 0; i--) {\n    const own = spans[i]!.messages.map((m) =>\n      typeof m.content === 'string' ? m.content : '',\n    ).join('\\n')\n    result[i] = `${own}\\n${result[i + 1]}`\n  }\n  return result\n}\n\n/** Index of the first element strictly greater than `target` in a sorted array. */\nfunction upperBound(sorted: number[], target: number): number {\n  let lo = 0\n  let hi = sorted.length\n  while (lo < hi) {\n    const mid = (lo + hi) >>> 1\n    if (sorted[mid]! <= target) lo = mid + 1\n    else hi = mid\n  }\n  return lo\n}\n\nfunction stringify(v: unknown): string {\n  if (v === null || v === undefined) return ''\n  if (typeof v === 'string') return v\n  try {\n    return JSON.stringify(v)\n  } catch {\n    return String(v)\n  }\n}\n\n// Re-export for convenience in consumers that want both descriptive and usage metrics.\nexport { computeToolUseMetrics }\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;AA8BA,eAAsB,iBACpB,OACA,UAAuD,CAAC,GAC3B;AAC7B,QAAM,OAAO,MAAM,MAAM,SAAS;AAAA,IAChC,YAAY,QAAQ;AAAA,IACpB,WAAW,QAAQ;AAAA,EACrB,CAAC;AACD,QAAM,WAAkC,CAAC;AACzC,QAAM,cAAsC,CAAC;AAC7C,QAAM,aAAqC,CAAC;AAC5C,QAAM,YAAoC,CAAC;AAE3C,aAAW,OAAO,MAAM;AACtB,UAAM,UAAU,MAAM,MAAM,OAAO,IAAI,KAAK;AAC5C,eAAW,KAAK,SAAS;AACvB,UAAI,CAAC,EAAE,SAAU;AACjB,YAAM,cAAc,EAAE,QAAQ,IAAI,EAAE,WAAW,EAAE,QAAQ;AACzD,eAAS,KAAK;AAAA,QACZ,OAAO,IAAI;AAAA,QACX,YAAY,IAAI;AAAA,QAChB,WAAW,IAAI;AAAA,QACf,WAAW,EAAE;AAAA,QACb,OAAO,EAAE;AAAA,QACT,UAAU,EAAE;AAAA,QACZ;AAAA,QACA,WAAW,EAAE;AAAA,MACf,CAAC;AACD,kBAAY,EAAE,SAAS,KAAK,YAAY,EAAE,SAAS,KAAK,KAAK;AAC7D,iBAAW,IAAI,UAAU,KAAK,WAAW,IAAI,UAAU,KAAK,KAAK;AACjE,UAAI,IAAI,UAAW,WAAU,IAAI,SAAS,KAAK,UAAU,IAAI,SAAS,KAAK,KAAK;AAAA,IAClF;AAAA,EACF;AAEA,QAAM,eAAe,IAAI,IAAI,SAAS,IAAI,CAAC,MAAM,EAAE,KAAK,CAAC;AACzD,SAAO;AAAA,IACL;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA,WAAW,KAAK;AAAA,IAChB,kBAAkB,KAAK,SAAS,IAAI,aAAa,OAAO,KAAK,SAAS;AAAA,EACxE;AACF;;;ACnCA,eAAsB,mBACpB,OACA,UAA8D,CAAC,GAChC;AAC/B,QAAM,QAAQ,QAAQ,SAAS;AAC/B,QAAM,UAAU,QAAQ,kBAAkB;AAC1C,QAAM,OAAO,MAAM,MAAM,SAAS;AAGlC,QAAM,WAAW,oBAAI,IAAyB;AAC9C,MAAI,gBAAgB;AAEpB,aAAW,OAAO,MAAM;AACtB,QAAI,IAAI,WAAW,eAAe,IAAI,SAAS,SAAS,MAAO;AAC/D;AACA,UAAM,QAAQ,MAAM,MAAM,MAAM,EAAE,OAAO,IAAI,MAAM,CAAC;AACpD,UAAM,SAAS,MAAM,MAAM,OAAO,EAAE,OAAO,IAAI,MAAM,CAAC;AACtD,UAAM,MAAM,gBAAgB,EAAE,KAAK,OAAO,OAAO,GAAG,KAAK;AAEzD,QAAI;AACJ,QAAI;AACJ,QAAI;AACJ,QAAI,IAAI,eAAe;AACrB,YAAM,OAAO,MAAM,KAAK,CAAC,MAAM,EAAE,WAAW,IAAI,aAAa;AAC7D,UAAI,MAAM,SAAS,QAAQ;AACzB,mBAAW,KAAK;AAChB,oBAAY,QAAQ,KAAK,IAAI,EAAE,MAAM,GAAG,EAAE;AAAA,MAC5C,WAAW,MAAM,SAAS,SAAS;AACjC,oBAAY,KAAK;AAAA,MACnB;AAAA,IACF;AAEA,QAAI,CAAC,UAAU;AACb,YAAM,KAAK,MAAM,UAAU,OAAO,IAAI,KAAK;AAC3C,YAAM,UAAU,GAAG,OAAO,CAAC,MAAM,EAAE,WAAW,OAAO,EAAE,IAAI;AAC3D,UAAI,SAAS;AACX,mBAAW,QAAQ;AACnB,oBAAY,QAAQ,QAAQ,IAAI,EAAE,MAAM,GAAG,EAAE;AAAA,MAC/C;AAAA,IACF;AAIA,QAAI,CAAC,WAAW;AACd,YAAM,QAAQ,MAAM,KAAK,CAAC,MAAM,EAAE,SAAS,WAAW,OAAO,EAAE,cAAc,QAAQ;AACrF,UAAI,OAAO,SAAS,QAAS,aAAY,MAAM;AAAA,IACjD;AAEA,UAAM,MAAM,GAAG,IAAI,YAAY,IAAI,YAAY,EAAE,IAAI,aAAa,EAAE,IAAI,aAAa,EAAE;AACvF,QAAI,UAAU,SAAS,IAAI,GAAG;AAC9B,QAAI,CAAC,SAAS;AACZ,gBAAU;AAAA,QACR,cAAc,IAAI;AAAA,QAClB;AAAA,QACA;AAAA,QACA;AAAA,QACA,UAAU;AAAA,QACV,aAAa,CAAC;AAAA,QACd,cAAc,IAAI;AAAA,QAClB,cAAc,kBAAkB,KAAK,KAAK,IAAI;AAAA,MAChD;AACA,eAAS,IAAI,KAAK,OAAO;AAAA,IAC3B;AACA,YAAQ;AACR,QAAI,CAAC,QAAQ,YAAY,SAAS,IAAI,UAAU,EAAG,SAAQ,YAAY,KAAK,IAAI,UAAU;AAAA,EAC5F;AAEA,QAAM,MAAM,CAAC,GAAG,SAAS,OAAO,CAAC,EAC9B,OAAO,CAAC,MAAM,EAAE,YAAY,OAAO,EACnC,KAAK,CAAC,GAAG,MAAM,EAAE,WAAW,EAAE,QAAQ;AAEzC,SAAO,EAAE,UAAU,KAAK,eAAe,WAAW,KAAK,OAAO;AAChE;AAEA,SAAS,kBAAkB,OAAmC;AAC5D,QAAM,UAAU,MAAM,KAAK,CAAC,MAAM,EAAE,WAAW,OAAO;AACtD,SAAO,SAAS;AAClB;;;ACvFA,eAAsB,oBACpB,OACA,MACA,MACA,UAA6B,CAAC,GACH;AAC3B,QAAM,CAAC,GAAG,CAAC,IAAI,MAAM,QAAQ,IAAI,CAAC,gBAAgB,OAAO,IAAI,GAAG,gBAAgB,OAAO,IAAI,CAAC,CAAC;AAC7F,QAAM,KAAK,QAAQ,cAAc;AACjC,QAAM,SAAS,KAAK,IAAI,EAAE,MAAM,QAAQ,EAAE,MAAM,MAAM;AACtD,WAAS,IAAI,GAAG,IAAI,QAAQ,KAAK;AAC/B,UAAM,QAAQ,EAAE,MAAM,CAAC;AACvB,UAAM,QAAQ,EAAE,MAAM,CAAC;AACvB,QAAI,CAAC,GAAG,OAAO,KAAK,GAAG;AACrB,aAAO;AAAA,QACL;AAAA,QACA;AAAA,QACA,sBAAsB;AAAA,QACtB;AAAA,QACA;AAAA,QACA,QAAQ,mBAAmB,OAAO,KAAK;AAAA,QACvC,iBAAiB;AAAA,MACnB;AAAA,IACF;AAAA,EACF;AACA,MAAI,EAAE,MAAM,WAAW,EAAE,MAAM,QAAQ;AACrC,WAAO,EAAE,MAAM,MAAM,sBAAsB,MAAM,iBAAiB,OAAO;AAAA,EAC3E;AACA,QAAM,SAAqB,EAAE,MAAM,SAAS,EAAE,MAAM,SAAS,IAAI;AACjE,QAAM,QAAQ,OAAO,MAAM,SAAS;AAGpC,QAAM,SACJ,WAAW,IACP,0CAA0C,KAAK,iCAC/C,sBAAsB,KAAK,6BAA6B,SAAS,CAAC;AACxE,SAAO;AAAA,IACL;AAAA,IACA;AAAA,IACA,sBAAsB;AAAA,IACtB,OAAO,EAAE,MAAM,MAAM;AAAA,IACrB,OAAO,EAAE,MAAM,MAAM;AAAA,IACrB;AAAA,IACA,iBAAiB;AAAA,EACnB;AACF;AAEA,SAAS,kBAAkB,GAAmB,GAA4B;AACxE,MAAI,EAAE,KAAK,SAAS,EAAE,KAAK,KAAM,QAAO;AACxC,MAAI,EAAE,KAAK,SAAS,UAAU,EAAE,KAAK,SAAS,OAAQ,QAAO,EAAE,KAAK,aAAa,EAAE,KAAK;AACxF,MAAI,EAAE,KAAK,SAAS,SAAS,EAAE,KAAK,SAAS,MAAO,QAAO,EAAE,KAAK,UAAU,EAAE,KAAK;AACnF,MAAI,EAAE,KAAK,SAAS,WAAW,EAAE,KAAK,SAAS;AAC7C,WAAO,EAAE,KAAK,cAAc,EAAE,KAAK;AACrC,SAAO,EAAE,KAAK,SAAS,EAAE,KAAK;AAChC;AAEA,SAAS,mBAAmB,GAAmB,GAA2B;AACxE,MAAI,EAAE,KAAK,SAAS,EAAE,KAAK,KAAM,QAAO,QAAQ,EAAE,KAAK,IAAI,OAAO,EAAE,KAAK,IAAI;AAC7E,MAAI,EAAE,KAAK,SAAS,UAAU,EAAE,KAAK,SAAS,UAAU,EAAE,KAAK,aAAa,EAAE,KAAK,UAAU;AAC3F,WAAO,QAAQ,EAAE,KAAK,QAAQ,OAAO,EAAE,KAAK,QAAQ;AAAA,EACtD;AACA,MAAI,EAAE,KAAK,SAAS,SAAS,EAAE,KAAK,SAAS,SAAS,EAAE,KAAK,UAAU,EAAE,KAAK,OAAO;AACnF,WAAO,SAAS,EAAE,KAAK,KAAK,OAAO,EAAE,KAAK,KAAK;AAAA,EACjD;AACA,SAAO,SAAS,EAAE,KAAK,IAAI,SAAS,EAAE,KAAK,IAAI;AACjD;;;AC9DA,eAAsB,mBAAmB,OAAkD;AACzF,QAAM,OAAO,MAAM,MAAM,MAAM,EAAE,MAAM,QAAQ,CAAC,GAAG;AAAA,IACjD,CAAC,MAAsB,EAAE,SAAS;AAAA,EACpC;AACA,MAAI,IAAI,WAAW,EAAG,QAAO,EAAE,OAAO,CAAC,GAAG,YAAY,CAAC,GAAG,UAAU,CAAC,EAAE;AAEvE,QAAM,cAAc,oBAAI,IAAyB;AACjD,aAAW,KAAK,KAAK;AACnB,UAAM,MAAM,YAAY,IAAI,EAAE,SAAS,KAAK,CAAC;AAC7C,QAAI,KAAK,CAAC;AACV,gBAAY,IAAI,EAAE,WAAW,GAAG;AAAA,EAClC;AAEA,QAAM,WAAW,CAAC,GAAG,IAAI,IAAI,IAAI,IAAI,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC,EAAE,KAAK;AAC9D,QAAM,QAAqB,CAAC;AAC5B,aAAW,CAAC,KAAK,KAAK,KAAK,aAAa;AACtC,UAAM,UAAU,oBAAI,IAAiC;AACrD,eAAW,KAAK,OAAO;AACrB,YAAM,IAAI,QAAQ,IAAI,EAAE,OAAO,KAAK,oBAAI,IAAoB;AAC5D,QAAE,IAAI,EAAE,cAAc,EAAE,KAAK;AAC7B,cAAQ,IAAI,EAAE,SAAS,CAAC;AAAA,IAC1B;AACA,UAAM,aAAa,CAAC,GAAG,QAAQ,KAAK,CAAC;AACrC,aAAS,IAAI,GAAG,IAAI,WAAW,QAAQ,KAAK;AAC1C,eAAS,IAAI,IAAI,GAAG,IAAI,WAAW,QAAQ,KAAK;AAC9C,cAAM,SAAS,WAAW,CAAC;AAC3B,cAAM,SAAS,WAAW,CAAC;AAC3B,cAAM,IAAI,QAAQ,IAAI,MAAM;AAC5B,cAAM,IAAI,QAAQ,IAAI,MAAM;AAC5B,cAAM,SAAkC,CAAC;AACzC,mBAAW,CAAC,QAAQ,MAAM,KAAK,GAAG;AAChC,gBAAM,SAAS,EAAE,IAAI,MAAM;AAC3B,cAAI,WAAW,OAAW,QAAO,KAAK,CAAC,QAAQ,MAAM,CAAC;AAAA,QACxD;AACA,YAAI,OAAO,SAAS,EAAG;AACvB,cAAM,cAAc,OAAO;AAAA,UACzB,CAAC,CAAC,QAAQ,MAAM,MACd;AAAA,YACE,EAAE,WAAW,QAAQ,WAAW,KAAK,OAAO,QAAQ,WAAW,GAAG;AAAA,YAClE,EAAE,WAAW,QAAQ,WAAW,KAAK,OAAO,QAAQ,WAAW,GAAG;AAAA,UACpE;AAAA,QACJ;AACA,cAAM,IAAI;AAAA,UACR,YAAY,CAAC,EAAG,IAAI,CAAC,GAAG,OAAO,YAAY,IAAI,CAAC,SAAS,KAAK,EAAE,CAAE,CAAC;AAAA,QACrE;AACA,cAAM,KAAK;AAAA,UACT,QAAQ;AAAA,UACR,QAAQ;AAAA,UACR,WAAW;AAAA,UACX,aAAa,OAAO;AAAA,UACpB,SAAS;AAAA,YACP,OAAO,IAAI,CAAC,MAAM,EAAE,CAAC,CAAC;AAAA,YACtB,OAAO,IAAI,CAAC,MAAM,EAAE,CAAC,CAAC;AAAA,UACxB;AAAA,UACA,cAAc;AAAA,QAChB,CAAC;AAAA,MACH;AAAA,IACF;AAAA,EACF;AAEA,SAAO;AAAA,IACL,OAAO,MAAM,KAAK,CAAC,GAAG,MAAM,EAAE,cAAc,EAAE,WAAW;AAAA,IACzD,YAAY,CAAC,GAAG,YAAY,KAAK,CAAC,EAAE,KAAK;AAAA,IACzC;AAAA,EACF;AACF;;;ACrEA,eAAsB,eACpB,OACA,SACA,SACyB;AACzB,QAAM,eAAe,MAAM,MAAM,SAAS,QAAQ,QAAQ;AAC1D,QAAM,gBAAgB,MAAM,MAAM,SAAS,QAAQ,SAAS;AAC5D,QAAM,UAAU,MAAM,QAAQ;AAAA,IAC5B,QAAQ,IAAI,OAAO,MAAM;AACvB,YAAM,UAAU,EAAE,WAAW,eAAe,EAAE,MAAM;AACpD,YAAM,WAAW,MAAM,WAAW,cAAc,SAAS,KAAK;AAC9D,YAAM,YAAY,MAAM,WAAW,eAAe,SAAS,KAAK;AAChE,aAAO,EAAE,QAAQ,EAAE,QAAQ,gBAAgB,EAAE,gBAAgB,UAAU,UAAU;AAAA,IACnF,CAAC;AAAA,EACH;AACA,SAAO,kBAAkB,SAAS,OAAO;AAC3C;AAEA,eAAe,WACb,MACA,SACA,OACmB;AACnB,QAAM,MAAgB,CAAC;AACvB,aAAW,KAAK,MAAM;AACpB,UAAM,IAAI,MAAM,QAAQ,GAAG,KAAK;AAChC,QAAI,MAAM,QAAQ,OAAO,SAAS,CAAC,EAAG,KAAI,KAAK,CAAC;AAAA,EAClD;AACA,SAAO;AACT;AAEA,SAAS,eAAe,QAAyE;AAC/F,SAAO,OAAO,KAAK,UAAU;AAC3B,YAAQ,QAAQ;AAAA,MACd,KAAK;AAAA,MACL,KAAK;AACH,eAAO,IAAI,SAAS,SAAS;AAAA,MAC/B,KAAK;AACH,eAAO,IAAI,SAAS,SAAS,OAAO,IAAI;AAAA,MAC1C,KAAK;AACH,eAAO,IAAI,WAAW,IAAI,YAAY,IAAI,UAAU,IAAI,YAAY;AAAA,MACtE,KAAK,WAAW;AACd,cAAM,MAAM,MAAM,SAAS,OAAO,IAAI,KAAK;AAC3C,eAAO,aAAa,GAAG,EAAE;AAAA,MAC3B;AAAA,MACA,KAAK,eAAe;AAClB,cAAM,MAAM,MAAM,SAAS,OAAO,IAAI,KAAK;AAC3C,eAAO,aAAa,GAAG,EAAE;AAAA,MAC3B;AAAA,MACA,KAAK,gBAAgB;AACnB,cAAM,MAAM,MAAM,SAAS,OAAO,IAAI,KAAK;AAC3C,eAAO,aAAa,GAAG,EAAE;AAAA,MAC3B;AAAA,MACA,KAAK,gBAAgB;AACnB,eAAO,gBAAgB,GAAG,MAAM,YAAY,IAAI;AAAA,MAClD;AAAA,MACA;AACE,eAAO;AAAA,IACX;AAAA,EACF;AACF;;;AClDA,eAAsB,cACpB,OACA,UAA4B,CAAC,GACH;AAC1B,QAAM,iBAAiB,QAAQ,kBAAkB;AACjD,QAAM,OAAO,QAAQ,QACjB,CAAC,EAAE,OAAO,QAAQ,MAAM,CAAC,KACxB,MAAM,MAAM,SAAS,GAAG,IAAI,CAAC,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE;AAE5D,QAAM,WAA+B,CAAC;AACtC,aAAW,EAAE,MAAM,KAAK,MAAM;AAC5B,UAAM,QAAQ,MAAM,UAAU,OAAO,KAAK;AAC1C,UAAM,QAAQ,oBAAI,IAAwE;AAC1F,eAAW,KAAK,OAAO;AACrB,YAAM,IAAI,QAAQ,EAAE,IAAI;AAKxB,YAAM,MAAM,GAAG,EAAE,QAAQ,KAAS,CAAC;AACnC,YAAM,SAAS,MAAM,IAAI,GAAG,KAAK,EAAE,OAAO,CAAC,GAAG,SAAS,GAAG,UAAU,EAAE,SAAS;AAC/E,aAAO,MAAM,KAAK,CAAC;AACnB,YAAM,IAAI,KAAK,MAAM;AAAA,IACvB;AACA,eAAW,EAAE,OAAO,SAAS,GAAG,SAAS,KAAK,MAAM,OAAO,GAAG;AAC5D,UAAI,MAAM,SAAS,eAAgB;AACnC,YAAM,SAAS,CAAC,GAAG,KAAK,EAAE,KAAK,CAAC,GAAG,MAAM,EAAE,YAAY,EAAE,SAAS;AAClE,YAAM,QAAQ,OAAO,CAAC,EAAG;AACzB,YAAM,OAAO,OAAO,OAAO,SAAS,CAAC,EAAG;AACxC,eAAS,KAAK;AAAA,QACZ;AAAA,QACA;AAAA,QACA,SAAS;AAAA,QACT,aAAa,OAAO;AAAA,QACpB,SAAS,OAAO,IAAI,CAAC,MAAM,EAAE,MAAM;AAAA,QACnC,UAAU,OAAO;AAAA,MACnB,CAAC;AAAA,IACH;AAAA,EACF;AAEA,QAAM,eAAe,IAAI,IAAI,SAAS,IAAI,CAAC,MAAM,EAAE,KAAK,CAAC;AACzD,SAAO;AAAA,IACL;AAAA,IACA,kBAAkB,KAAK,SAAS,IAAI,aAAa,OAAO,KAAK,SAAS;AAAA,IACtE,WAAW,KAAK;AAAA,EAClB;AACF;;;AChDA,eAAsB,cACpB,OACA,UAA4B,CAAC,GACH;AAC1B,QAAM,OAAO,QAAQ,QAAQ,CAAC,QAAQ,KAAK,KAAK,MAAM,MAAM,SAAS,GAAG,IAAI,CAAC,MAAM,EAAE,KAAK;AAE1F,QAAM,QAA4B,CAAC;AACnC,MAAI,aAAa;AACjB,MAAI,cAAc;AAClB,aAAW,SAAS,MAAM;AACxB,UAAM,QAAQ,MAAM,UAAU,OAAO,KAAK;AAC1C,QAAI,MAAM,WAAW,GAAG;AACtB,YAAM,KAAK,EAAE,OAAO,aAAa,GAAG,YAAY,GAAG,WAAW,EAAE,CAAC;AACjE;AAAA,IACF;AACA,UAAM,OAAO,MAAM,SAAS,OAAO,KAAK;AAOxC,UAAM,YAAY,CAAC,GAAG,IAAI,EAAE,KAAK,CAAC,GAAG,MAAM,EAAE,YAAY,EAAE,SAAS;AACpE,UAAM,aAAa,UAAU,IAAI,CAAC,MAAM,EAAE,SAAS;AACnD,UAAM,aAAa,gBAAgB,SAAS;AAC5C,QAAI,SAAS;AACb,eAAW,KAAK,OAAO;AACrB,UAAI,EAAE,WAAW,SAAS;AACxB;AACA;AAAA,MACF;AAEA,YAAM,SAAS,WAAW,YAAY,EAAE,SAAS;AACjD,UAAI,QAAQ,aAAa;AACvB,YAAI,CAAC,QAAQ,YAAY,GAAG,EAAE,KAAK,UAAU,MAAM,MAAM,EAAE,CAAC,EAAG;AAAA,MACjE,OAAO;AAKL,cAAM,YAAY,UAAU,EAAE,MAAM;AACpC,YAAI,cAAc,GAAI;AACtB,cAAM,WAAW,WAAW,MAAM,KAAK;AACvC,cAAM,OAAO,SAAS,SAAS,UAAU,MAAM,GAAG,GAAG,CAAC;AACtD,YAAI,CAAC,KAAM;AAAA,MACb;AAAA,IACF;AACA,UAAM,YAAY,SAAS,MAAM;AACjC,UAAM,KAAK,EAAE,OAAO,aAAa,QAAQ,YAAY,MAAM,QAAQ,UAAU,CAAC;AAC9E,kBAAc,MAAM;AACpB,mBAAe;AAAA,EACjB;AACA,SAAO,EAAE,OAAO,kBAAkB,aAAa,IAAI,cAAc,aAAa,EAAE;AAClF;AAOA,SAAS,gBAAgB,OAA4B;AACnD,QAAM,SAAS,IAAI,MAAc,MAAM,SAAS,CAAC;AACjD,SAAO,MAAM,MAAM,IAAI;AACvB,WAAS,IAAI,MAAM,SAAS,GAAG,KAAK,GAAG,KAAK;AAC1C,UAAM,MAAM,MAAM,CAAC,EAAG,SAAS;AAAA,MAAI,CAAC,MAClC,OAAO,EAAE,YAAY,WAAW,EAAE,UAAU;AAAA,IAC9C,EAAE,KAAK,IAAI;AACX,WAAO,CAAC,IAAI,GAAG,GAAG;AAAA,EAAK,OAAO,IAAI,CAAC,CAAC;AAAA,EACtC;AACA,SAAO;AACT;AAGA,SAAS,WAAW,QAAkB,QAAwB;AAC5D,MAAI,KAAK;AACT,MAAI,KAAK,OAAO;AAChB,SAAO,KAAK,IAAI;AACd,UAAM,MAAO,KAAK,OAAQ;AAC1B,QAAI,OAAO,GAAG,KAAM,OAAQ,MAAK,MAAM;AAAA,QAClC,MAAK;AAAA,EACZ;AACA,SAAO;AACT;AAEA,SAAS,UAAU,GAAoB;AACrC,MAAI,MAAM,QAAQ,MAAM,OAAW,QAAO;AAC1C,MAAI,OAAO,MAAM,SAAU,QAAO;AAClC,MAAI;AACF,WAAO,KAAK,UAAU,CAAC;AAAA,EACzB,QAAQ;AACN,WAAO,OAAO,CAAC;AAAA,EACjB;AACF;","names":[]}

package/dist/{provenance-CIxfBnkl.d.ts → provenance-DJ--Jcbg.d.ts}

@@ -2,7 +2,7 @@ import { S as Scenario, f as Gate, G as GateResult, g as GateContext, C as Campa
 import { R as RedTeamCase } from './red-team-BWdoyleI.js';
 import { R as RunRecord } from './run-record-CP2ObebC.js';
 import { D as Direction } from './pareto-E-pembql.js';
-import { a as PairedBootstrapResult } from './statistics-CCJpTGOS.js';
+import { a as PairedBootstrapResult } from './statistics-xP-cWc5k.js';
 import { R as RunCampaignOptions, C as CampaignStorage } from './gepa-Dprxvz8r.js';
 import { HostedClient, TraceSpanEvent } from './hosted/index.js';
 

package/dist/reporting.d.ts

@@ -1,7 +1,7 @@
 export { R as RubricOutcomePair, a as RubricPredictiveValidityInput, b as RubricPredictiveValidityReport, c as RubricRanking, r as rubricPredictiveValidity } from './rubric-predictive-validity-C2hDKM8Z.js';
 export { B as BootstrapOptions, a as BootstrapResult, J as JudgeReplayGateArgs, R as ReleaseConfidenceAxis, b as ReleaseConfidenceAxisName, c as ReleaseConfidenceInput, d as ReleaseConfidenceIssue, e as ReleaseConfidenceMetrics, f as ReleaseConfidenceScorecard, g as ReleaseConfidenceStatus, h as ReleaseConfidenceThresholds, i as ReleaseTraceEvidence, j as RenderReleaseReportOptions, V as Verdict, k as assertReleaseConfidence, l as bootstrapCi, m as evaluateReleaseConfidence, n as judgeReplayGate, r as renderReleaseReport } from './release-report-pidWUMZ2.js';
 export { I as InterimReleaseConfidence, a as InterimReleaseConfidenceInput, P as PairedEvalueOptions, b as PairedEvalueSequence, c as PairedEvalueStep, S as SequentialDecision, e as evaluateInterimReleaseConfidence, p as pairedEvalueSequence } from './sequential-5iSVfzl2.js';
-export { P as PairedBootstrapOptions, a as PairedBootstrapResult, b as benjaminiHochberg, p as pairedBootstrap, w as wilcoxonSignedRank } from './statistics-CCJpTGOS.js';
+export { P as PairedBootstrapOptions, a as PairedBootstrapResult, b as benjaminiHochberg, p as pairedBootstrap, w as wilcoxonSignedRank } from './statistics-xP-cWc5k.js';
 export { G as GainDistributionBin, a as GainDistributionFigureSpec, b as GainDistributionOptions, P as ParetoFigureSpec, c as ParetoPoint, R as RESEARCH_REPORT_HARD_PAIR_FLOOR, d as ResearchReport, e as ResearchReportCandidate, f as ResearchReportDecision, g as ResearchReportMethodology, h as ResearchReportOptions, i as ResearchReportRecommendation, S as SummaryTable, j as SummaryTableOptions, k as SummaryTableRow, l as gainHistogram, p as paretoChart, r as researchReport, s as summaryTable } from './summary-report-CInXwsza.js';
 import './run-record-CP2ObebC.js';
 import '@tangle-network/agent-interface';

package/dist/reporting.js

@@ -4,10 +4,10 @@ import {
   evaluateReleaseConfidence,
   judgeReplayGate,
   renderReleaseReport
-} from "./chunk-QTMYW64F.js";
+} from "./chunk-VDGPPGE3.js";
 import {
   rubricPredictiveValidity
-} from "./chunk-YRZ4M5GS.js";
+} from "./chunk-NF7OZ4J7.js";
 import {
   evaluateInterimReleaseConfidence,
   pairedEvalueSequence
@@ -18,12 +18,12 @@ import {
   paretoChart,
   researchReport,
   summaryTable
-} from "./chunk-URWVKRCS.js";
+} from "./chunk-6GT4NI4V.js";
 import {
   benjaminiHochberg,
   pairedBootstrap,
   wilcoxonSignedRank
-} from "./chunk-56GC6VYO.js";
+} from "./chunk-HRGUJTER.js";
 import "./chunk-VSMTAMNK.js";
 import "./chunk-3BFEG2F6.js";
 import "./chunk-PZ5AY32C.js";

package/dist/rl.js

@@ -9,26 +9,26 @@ import {
   extractVerifiableReward,
   extractVerifiableRewardsFromRecords,
   filterDeterministicallyRewarded
-} from "./chunk-5HRORJQY.js";
+} from "./chunk-AIGWQEME.js";
 import {
   FileSystemOutcomeStore,
   InMemoryOutcomeStore
 } from "./chunk-3RF76KTD.js";
 import {
   runEvalCampaign
-} from "./chunk-X74V6ESX.js";
+} from "./chunk-OKQ2LAT7.js";
 import "./chunk-CWNP4DV4.js";
 import {
   rubricPredictiveValidity
-} from "./chunk-YRZ4M5GS.js";
+} from "./chunk-NF7OZ4J7.js";
 import {
   evaluateInterimReleaseConfidence
 } from "./chunk-MAZ26DC7.js";
-import "./chunk-URWVKRCS.js";
+import "./chunk-6GT4NI4V.js";
 import {
   benjaminiHochberg,
   wilcoxonSignedRank
-} from "./chunk-56GC6VYO.js";
+} from "./chunk-HRGUJTER.js";
 import {
   observationsFromRunRecords,
   thompsonCurriculum,

package/dist/{run-campaign-WXY7KI67.js → run-campaign-OWCFOEQG.js}

@@ -1,10 +1,10 @@
 import {
   runCampaign
-} from "./chunk-2T4EZACH.js";
-import "./chunk-56GC6VYO.js";
+} from "./chunk-HRGTA6U5.js";
+import "./chunk-HRGUJTER.js";
 import "./chunk-3BFEG2F6.js";
 import "./chunk-PZ5AY32C.js";
 export {
   runCampaign
 };
-//# sourceMappingURL=run-campaign-WXY7KI67.js.map
+//# sourceMappingURL=run-campaign-OWCFOEQG.js.map

package/dist/{statistics-CCJpTGOS.d.ts → statistics-xP-cWc5k.d.ts}

@@ -78,6 +78,22 @@ declare function cliffsDelta(before: number[], after: number[]): number;
  * <0.474 medium, else large.
  */
 declare function interpretCliffs(delta: number): CliffsMagnitude;
+/**
+ * Average-rank-with-ties transform (1-indexed). Tied values receive the mean
+ * of the ranks they span, the standard correction for Spearman's ρ.
+ */
+declare function ranks(xs: number[]): number[];
+/**
+ * Pearson product-moment correlation coefficient r ∈ [-1, 1] between two
+ * equal-length series. See the edge-case contract above: NaN for n < 2 or
+ * unequal lengths, 1 when both series are constant, 0 when exactly one is.
+ */
+declare function pearsonR(a: number[], b: number[]): number;
+/**
+ * Spearman's rank correlation ρ — Pearson over the average-rank-with-ties
+ * transform of each series. Same edge-case contract as {@link pearsonR}.
+ */
+declare function spearmanR(a: number[], b: number[]): number;
 interface WeightedCompositeInput {
     /** Per-dimension scores (typically 0..1). */
     dims: Record<string, number>;
@@ -430,4 +446,4 @@ declare function eProcess(opts?: EProcessOptions): EProcess;
  *  gate verdicts is non-reproducible by construction). */
 declare function mulberry32(seed: number): () => number;
 
-export { mulberry32 as A, normalizeScores as B, type CorpusAgreementReport as C, pairedMde as D, type EProcessState as E, pairedRiskDifference as F, pairedTTest as G, partialCredit as H, passAtK as I, requiredSampleSize as J, weightedComposite as K, weightedMean as L, type McNemarResult as M, wilson as N, type PairedBootstrapOptions as P, type RiskDifferenceResult as R, type WeightedCompositeInput as W, type PairedBootstrapResult as a, benjaminiHochberg as b, type CliffsMagnitude as c, type CorpusAgreementOptions as d, type CorpusAgreementPerDimension as e, type CorpusScoreRecord as f, type EProcess as g, type EProcessOptions as h, type EProcessStep as i, type ProportionInterval as j, type WeightedCompositeResult as k, bonferroni as l, cliffsDelta as m, cohensD as n, confidenceInterval as o, pairedBootstrap as p, corpusInterRaterAgreement as q, corpusInterRaterAgreementFromJudgeScores as r, eProcess as s, interRaterReliability as t, interpretCliffs as u, mannWhitneyU as v, wilcoxonSignedRank as w, mcnemar as x, mcnemarPower as y, mcnemarRequiredN as z };
+export { mulberry32 as A, normalizeScores as B, type CorpusAgreementReport as C, pairedMde as D, type EProcessState as E, pairedRiskDifference as F, pairedTTest as G, partialCredit as H, passAtK as I, pearsonR as J, ranks as K, requiredSampleSize as L, type McNemarResult as M, spearmanR as N, weightedComposite as O, type PairedBootstrapOptions as P, weightedMean as Q, type RiskDifferenceResult as R, wilson as S, type WeightedCompositeInput as W, type PairedBootstrapResult as a, benjaminiHochberg as b, type CliffsMagnitude as c, type CorpusAgreementOptions as d, type CorpusAgreementPerDimension as e, type CorpusScoreRecord as f, type EProcess as g, type EProcessOptions as h, type EProcessStep as i, type ProportionInterval as j, type WeightedCompositeResult as k, bonferroni as l, cliffsDelta as m, cohensD as n, confidenceInterval as o, pairedBootstrap as p, corpusInterRaterAgreement as q, corpusInterRaterAgreementFromJudgeScores as r, eProcess as s, interRaterReliability as t, interpretCliffs as u, mannWhitneyU as v, wilcoxonSignedRank as w, mcnemar as x, mcnemarPower as y, mcnemarRequiredN as z };