@tangle-network/agent-eval 0.93.0 → 0.94.0

package/dist/chunk-CWNP4DV4.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/llm-client.ts"],"sourcesContent":["/**\n * LLM client with graceful degrade.\n *\n * OpenAI-compatible `/v1/chat/completions` client with:\n *   - Exponential-backoff retry on 429 + 5xx gateway errors (502/503/504).\n *   - Retry on transient network errors (fetch failed, AbortError, ECONNRESET).\n *   - Graceful json_schema → json_object degrade on 400 with schema-reject body.\n *   - Fenced-JSON stripping (```json ... ```) for models that wrap structured output.\n *   - Configurable base URL + api key / bearer, works with LiteLLM proxies, OpenAI\n *     directly, cli-bridge subscriptions, and any router that speaks the spec.\n *\n * Usage:\n *   const { value, result } = await callLlmJson<MyType>(\n *     { model: 'gpt-4o', messages: [...], jsonSchema: { name: 'x', schema: {...} } },\n *     { baseUrl: 'https://router.tangle.tools/v1', apiKey: process.env.KEY },\n *   )\n *\n * This is THE llm-calling seam for agent-eval primitives that need structured\n * output (semantic concept judge, reviewer directives, critic scores). Primitives\n * that need free-form text use `callLlm` and parse output themselves.\n */\n\nimport { AgentEvalError, CaptureIntegrityError } from './errors'\nimport {\n  defaultProviderRedactor,\n  type ProviderRedactor,\n  providerFromBaseUrl,\n  type RawProviderEvent,\n  type RawProviderSink,\n} from './trace/raw-provider-sink'\n\n// ─── Types ──────────────────────────────────────────────────────────────\n\nexport interface LlmMessage {\n  role: 'system' | 'user' | 'assistant'\n  /**\n   * Either a plain text content string OR a multimodal content array\n   * (text + image_url parts) for vision-capable models.\n   */\n  content:\n    | string\n    | Array<\n        | { type: 'text'; text: string }\n        | { type: 'image_url'; image_url: { url: string; detail?: 'auto' | 'low' | 'high' } }\n      >\n}\n\nexport interface LlmCallRequest {\n  model: string\n  messages: LlmMessage[]\n  /** Optional JSON-mode response format (response_format: json_object). */\n  jsonMode?: boolean\n  /** Optional structured output via JSON Schema. Falls back to json_object on 400. */\n  jsonSchema?: { name: string; schema: Record<string, unknown> }\n  temperature?: number\n  maxTokens?: number\n  /** Per-call timeout, default 300s. */\n  timeoutMs?: number\n}\n\nexport interface LlmUsage {\n  promptTokens: number\n  completionTokens: number\n  totalTokens: number\n  /** Proxies populate this when prompt caching is on. */\n  cachedPromptTokens?: number\n}\n\nexport interface LlmCallResult {\n  /** The text content of the first choice. Empty string if none. */\n  content: string\n  usage: LlmUsage\n  /**\n   * Cost in USD. Pulled from proxy's `_response_cost` field when present;\n   * `null` when neither the proxy nor the caller can derive it.\n   */\n  costUsd: number | null\n  /** Model name actually used (echoed from response). */\n  model: string\n  /** Wall-clock duration of the HTTP call (last attempt, if retried). */\n  durationMs: number\n  /**\n   * `finish_reason` echoed from the first choice (`stop`, `length`,\n   * `content_filter`, `tool_calls`, ...). `null` when the provider omits it.\n   * Exposed so a free-form `callLlm` caller CAN detect a truncated answer\n   * (`length`) instead of treating a cut-off completion as complete. Note:\n   * `callLlm` does not itself reject on it — acting on this signal is the\n   * caller's responsibility (in-repo free-form drivers do not yet enforce it).\n   */\n  finishReason?: string | null\n  /**\n   * True when `content.trim()` is empty. An empty completion is a silent zero\n   * for free-form `callLlm` callers; this flag is the signal a caller can\n   * inspect to fail loud rather than proceed on an empty string. `callLlm`\n   * surfaces it but does not throw on it.\n   */\n  contentEmpty?: boolean\n  /** Raw response body. */\n  raw: Record<string, unknown>\n}\n\nexport class LlmCallError extends AgentEvalError {\n  constructor(\n    message: string,\n    public readonly status: number,\n    public readonly body: string,\n    public readonly model: string,\n  ) {\n    super('judge', message)\n  }\n}\n\nexport interface LlmClientOptions {\n  /** Base URL (without trailing slash). Must end at the `/v1` prefix. */\n  baseUrl?: string\n  /** Bearer token — either `apiKey` or `bearer` populates `Authorization: Bearer ...`. */\n  apiKey?: string\n  bearer?: string\n  /** Override for the `Authorization` header (e.g. `X-Auth: ...`). Takes precedence over apiKey/bearer. */\n  authHeader?: { name: string; value: string }\n  /** Default timeout in ms. Per-call can override. */\n  defaultTimeoutMs?: number\n  /**\n   * Caller-supplied abort signal — e.g. a campaign-wide cancel. Linked to\n   * each attempt's per-attempt timeout controller, so aborting it cancels\n   * the in-flight fetch. A caller abort is FATAL: it is not retried even\n   * though an AbortError otherwise matches the transient patterns.\n   */\n  signal?: AbortSignal\n  /**\n   * Cross-attempt wall-clock budget in ms, measured from the first attempt.\n   * Before launching each attempt the loop checks the remaining budget and\n   * stops retrying once it is exhausted, rather than waiting the full\n   * per-attempt timeout on every retry. Bounds total time independent of\n   * `maxRetries` × `timeoutMs`.\n   */\n  deadlineMs?: number\n  /** Max retry attempts on retriable errors. Default 3 (1 initial + 2 retries). */\n  maxRetries?: number\n  /** Fetch implementation — defaults to global `fetch`. Override for custom transport (e.g. tests). */\n  fetch?: typeof fetch\n  /**\n   * Optional raw HTTP capture sink. When provided, every request, response,\n   * and error (across all retry attempts) is recorded to the sink, with auth\n   * headers and credential-shaped body fields redacted by default. This is\n   * the layer-1 forensics primitive: structured `LlmSpan`s record intent,\n   * raw events record what actually crossed the wire.\n   */\n  rawSink?: RawProviderSink\n  /**\n   * Logical provider id attached to raw events. When omitted, derived from\n   * `baseUrl` via `providerFromBaseUrl`.\n   */\n  provider?: string\n  /** Trace context attached to raw events; populated by emitter-aware callers. */\n  traceContext?: { runId?: string; spanId?: string }\n  /** Override the redaction strategy for this call. Defaults to `defaultProviderRedactor`. */\n  redactor?: ProviderRedactor\n}\n\n// ─── Internals ──────────────────────────────────────────────────────────\n\nconst DEFAULT_BASE_URL = 'https://router.tangle.tools/v1'\n// Flagship / reasoning models routinely take several minutes on large prompts (a\n// reflection over many failures, a long tool transcript). A tight cap aborts a\n// legitimately-slow but healthy call — and because every retry attempt re-uses\n// the same window, such a model aborts on ALL attempts and the loop throws. The\n// default is generous enough to let those complete, bounded enough that a truly\n// hung call still fails over after retries, and tunable per deployment via\n// TANGLE_LLM_TIMEOUT_MS. Per-call `req.timeoutMs` / `opts.defaultTimeoutMs`\n// still win for callers that know their model's latency.\nconst DEFAULT_TIMEOUT_MS = Number(process.env.TANGLE_LLM_TIMEOUT_MS) || 300_000\nconst DEFAULT_MAX_RETRIES = Number(process.env.TANGLE_LLM_MAX_RETRIES) || 3\n\nconst RETRYABLE_STATUS = new Set([429, 502, 503, 504])\n\n/**\n * Transient transport/network error signatures, matched against an error's\n * name, message, and `code`. Covers fetch/undici network failures, aborts\n * and timeouts, and — critically — HTTP/2 transport faults a keep-alive\n * connection raises mid-response: `terminated`, `NGHTTP2_INTERNAL_ERROR`,\n * `UND_ERR_*`, `other side closed`. Those last ones carry no clean HTTP\n * status; unrecognised, they escape the retry loop and surface as an\n * uncaught rejection.\n */\nconst TRANSIENT_ERROR_PATTERNS: readonly RegExp[] = [\n  /AbortError/i,\n  /TimeoutError/i,\n  /this operation was aborted/i,\n  /fetch failed/i,\n  /ECONNRESET/i,\n  /ETIMEDOUT/i,\n  /EAI_AGAIN/i,\n  /socket hang up/i,\n  /stream.*ended.*unexpectedly/i,\n  /terminated/i,\n  /other side closed/i,\n  /NGHTTP2/i,\n  /UND_ERR/i,\n]\n\n/**\n * True when an error is a transient transport/network fault worth retrying,\n * as opposed to a deterministic failure (4xx schema reject, JSON parse) that\n * a retry cannot fix. Inspects `LlmCallError.status`, then the error's\n * name/message/code, then recurses into `error.cause` — undici nests the\n * real socket fault one or more levels under `.cause`.\n *\n * This is THE retry classifier for the package: `callLlm` and\n * `withJudgeRetry` both route through it, so a connection-class error is\n * treated identically whether it surfaces in the HTTP client or a\n * TCloud-backed judge.\n */\nexport function isTransientLlmError(err: unknown): boolean {\n  return classifyTransient(err, 0)\n}\n\nfunction classifyTransient(err: unknown, depth: number): boolean {\n  if (err instanceof LlmCallError) return RETRYABLE_STATUS.has(err.status)\n  if (!(err instanceof Error)) return false\n  // Foreign errors (e.g. a TCloud judge SDK error) can carry a numeric HTTP\n  // status without being an LlmCallError — a retryable status is decisive.\n  const status = (err as { status?: unknown }).status\n  if (typeof status === 'number' && RETRYABLE_STATUS.has(status)) return true\n  const code = (err as { code?: unknown }).code\n  const haystack = `${err.name}\\n${err.message}\\n${typeof code === 'string' ? code : ''}`\n  if (TRANSIENT_ERROR_PATTERNS.some((p) => p.test(haystack))) return true\n  const cause = (err as { cause?: unknown }).cause\n  if (depth < 4 && cause instanceof Error && cause !== err) {\n    return classifyTransient(cause, depth + 1)\n  }\n  return false\n}\n\nfunction parseRetryAfter(headers: Headers): number | null {\n  const h = headers.get('retry-after')\n  if (!h) return null\n  const asNumber = Number(h)\n  if (Number.isFinite(asNumber) && asNumber > 0) return asNumber * 1000\n  const asDate = Date.parse(h)\n  if (Number.isFinite(asDate)) return Math.max(0, asDate - Date.now())\n  return null\n}\n\n/** Exponential backoff: 500ms, 1s, 2s, 4s, ... capped at 16s. Attempt is 0-indexed. */\nexport function backoffMs(attempt: number): number {\n  return Math.min(500 * 2 ** attempt, 16_000)\n}\n\nfunction buildHeaders(opts: LlmClientOptions): Record<string, string> {\n  const headers: Record<string, string> = {\n    'Content-Type': 'application/json',\n    Accept: 'application/json',\n  }\n  if (opts.authHeader) {\n    headers[opts.authHeader.name] = opts.authHeader.value\n  } else if (opts.bearer || opts.apiKey) {\n    headers.Authorization = `Bearer ${opts.bearer ?? opts.apiKey}`\n  }\n  return headers\n}\n\nfunction isSchemaRejection(status: number, body: string): boolean {\n  if (status !== 400) return false\n  const lower = body.toLowerCase()\n  return (\n    lower.includes('response_format') ||\n    lower.includes('json_schema') ||\n    lower.includes('is unavailable') ||\n    lower.includes('not supported')\n  )\n}\n\nfunction buildBody(req: LlmCallRequest, forceJsonObject: boolean): Record<string, unknown> {\n  const body: Record<string, unknown> = {\n    model: req.model,\n    messages: req.messages,\n    temperature: req.temperature ?? 0,\n  }\n  if (req.maxTokens != null) {\n    if (usesMaxCompletionTokens(req.model)) body.max_completion_tokens = req.maxTokens\n    else body.max_tokens = req.maxTokens\n  }\n\n  if (req.jsonSchema && !forceJsonObject) {\n    body.response_format = {\n      type: 'json_schema',\n      json_schema: { name: req.jsonSchema.name, schema: req.jsonSchema.schema, strict: true },\n    }\n  } else if (req.jsonMode || req.jsonSchema) {\n    body.response_format = { type: 'json_object' }\n  }\n\n  return body\n}\n\nfunction usesMaxCompletionTokens(model: string): boolean {\n  return /^gpt-5(?:[.-]|$)/i.test(model)\n}\n\nasync function sleep(ms: number): Promise<void> {\n  return new Promise((resolve) => setTimeout(resolve, ms))\n}\n\n/**\n * Combine the per-attempt timeout signal with an optional caller signal into\n * one signal the fetch listens on. Prefers the native `AbortSignal.any`; falls\n * back to manual wiring on runtimes that predate it. The caller signal is also\n * propagated to the timeout controller so aborting it cancels the in-flight\n * fetch immediately.\n */\nfunction linkSignals(timeoutController: AbortController, caller?: AbortSignal): AbortSignal {\n  if (!caller) return timeoutController.signal\n  if (typeof (AbortSignal as { any?: unknown }).any === 'function') {\n    return AbortSignal.any([timeoutController.signal, caller])\n  }\n  if (caller.aborted) {\n    timeoutController.abort()\n  } else {\n    caller.addEventListener('abort', () => timeoutController.abort(), { once: true })\n  }\n  return timeoutController.signal\n}\n\n/** True once the cross-attempt wall-clock budget (if any) is exhausted. */\nfunction deadlineExceeded(start: number, deadlineMs: number | undefined): boolean {\n  return deadlineMs != null && Date.now() - start >= deadlineMs\n}\n\n// ─── Public API ─────────────────────────────────────────────────────────\n\n/**\n * Strip a ```json / ``` code fence if the model emitted one.\n * Idempotent for naked JSON. Some models (claude-code via router, certain\n * deepseek models) wrap output even under json_object.\n */\nexport function stripFencedJson(raw: string): string {\n  const trimmed = raw.trim()\n  const m = trimmed.match(/^```(?:json)?\\s*\\n?([\\s\\S]*?)\\n?```\\s*$/)\n  return m ? m[1]!.trim() : trimmed\n}\n\nexport function extractJsonPayload(raw: string): string {\n  const stripped = stripFencedJson(raw)\n  try {\n    JSON.parse(stripped)\n    return stripped\n  } catch {\n    // Continue with balanced extraction below.\n  }\n\n  const starts = [...stripped.matchAll(/[[{]/g)]\n    .map((match) => match.index)\n    .filter((index) => index != null)\n  for (const start of starts) {\n    const candidate = extractBalancedJson(stripped, start)\n    if (!candidate) continue\n    try {\n      JSON.parse(candidate)\n      return candidate\n    } catch {\n      // Keep scanning; earlier braces may belong to prose.\n    }\n  }\n\n  return stripped\n}\n\nfunction extractBalancedJson(input: string, start: number): string | null {\n  const opener = input[start]\n  const closer = opener === '{' ? '}' : opener === '[' ? ']' : null\n  if (!closer) return null\n\n  const stack: string[] = [closer]\n  let isInString = false\n  let isEscaped = false\n\n  for (let i = start + 1; i < input.length; i++) {\n    const char = input[i]!\n    if (isEscaped) {\n      isEscaped = false\n      continue\n    }\n    if (char === '\\\\') {\n      isEscaped = isInString\n      continue\n    }\n    if (char === '\"') {\n      isInString = !isInString\n      continue\n    }\n    if (isInString) continue\n\n    if (char === '{') stack.push('}')\n    else if (char === '[') stack.push(']')\n    else if (char === stack[stack.length - 1]) {\n      stack.pop()\n      if (stack.length === 0) return input.slice(start, i + 1)\n    }\n  }\n\n  return null\n}\n\n/**\n * Low-level call. Returns raw content + usage + cost. Retries on transient\n * failures; does NOT degrade schema here — callers that want graceful\n * degrade use `callLlmJson`.\n */\nexport async function callLlm(\n  req: LlmCallRequest,\n  opts: LlmClientOptions = {},\n): Promise<LlmCallResult> {\n  const baseUrl = (opts.baseUrl ?? DEFAULT_BASE_URL).replace(/\\/+$/, '')\n  const url = `${baseUrl}/chat/completions`\n  const endpoint = '/chat/completions'\n  const timeoutMs = req.timeoutMs ?? opts.defaultTimeoutMs ?? DEFAULT_TIMEOUT_MS\n  const maxRetries = opts.maxRetries ?? DEFAULT_MAX_RETRIES\n  const fetchFn = opts.fetch ?? globalThis.fetch\n  const headers = buildHeaders(opts)\n  const provider = opts.provider ?? providerFromBaseUrl(baseUrl)\n  const sink = opts.rawSink\n  const redactor = opts.redactor ?? defaultProviderRedactor\n  const traceContext = opts.traceContext\n  const callerSignal = opts.signal\n  const deadlineMs = opts.deadlineMs\n  const deadlineStart = Date.now()\n\n  let lastErr: unknown\n  for (let attempt = 0; attempt < maxRetries; attempt++) {\n    // A caller cancel is fatal — never retried. Checking before each attempt\n    // means an already-aborted signal short-circuits without firing fetch.\n    if (callerSignal?.aborted) {\n      throw new DOMException('callLlm aborted by caller signal', 'AbortError')\n    }\n    // Stop retrying once the cross-attempt budget is spent rather than burning\n    // a full per-attempt timeout on each remaining retry.\n    if (attempt > 0 && deadlineExceeded(deadlineStart, deadlineMs)) {\n      throw lastErr instanceof Error ? lastErr : new Error(String(lastErr))\n    }\n    const controller = new AbortController()\n    const attemptSignal = linkSignals(controller, callerSignal)\n    const timeoutHandle = setTimeout(() => controller.abort(), timeoutMs)\n    const started = Date.now()\n    const requestBody = buildBody(req, false)\n    let attemptErrorRecorded = false\n    if (sink) {\n      await recordRaw(sink, redactor, {\n        eventId: cryptoEventId(),\n        runId: traceContext?.runId,\n        spanId: traceContext?.spanId,\n        provider,\n        model: req.model,\n        endpoint,\n        baseUrl,\n        attemptIndex: attempt,\n        direction: 'request',\n        timestamp: started,\n        requestHeaders: headers,\n        requestBody,\n        redactedFields: [],\n      })\n    }\n\n    try {\n      const res = await fetchFn(url, {\n        method: 'POST',\n        headers,\n        body: JSON.stringify(requestBody),\n        signal: attemptSignal,\n      })\n      clearTimeout(timeoutHandle)\n      const responseHeaders = sink ? headersToObject(res.headers) : undefined\n\n      if (!res.ok) {\n        const body = await res.text()\n        if (sink) {\n          await recordRaw(sink, redactor, {\n            eventId: cryptoEventId(),\n            runId: traceContext?.runId,\n            spanId: traceContext?.spanId,\n            provider,\n            model: req.model,\n            endpoint,\n            baseUrl,\n            attemptIndex: attempt,\n            direction: 'error',\n            timestamp: Date.now(),\n            durationMs: Date.now() - started,\n            statusCode: res.status,\n            responseHeaders,\n            responseBody: body,\n            errorMessage: `HTTP ${res.status}`,\n            redactedFields: [],\n          })\n          attemptErrorRecorded = true\n        }\n        const err = new LlmCallError(\n          `LLM call ${res.status}: ${body.slice(0, 300)}`,\n          res.status,\n          body,\n          req.model,\n        )\n        if (\n          RETRYABLE_STATUS.has(res.status) &&\n          attempt < maxRetries - 1 &&\n          !deadlineExceeded(deadlineStart, deadlineMs)\n        ) {\n          lastErr = err\n          const retryAfter = parseRetryAfter(res.headers)\n          await sleep(retryAfter ?? backoffMs(attempt))\n          continue\n        }\n        throw err\n      }\n\n      const text = await res.text()\n      let json: Record<string, unknown>\n      try {\n        json = JSON.parse(text) as Record<string, unknown>\n      } catch (parseErr) {\n        if (sink) {\n          await recordRaw(sink, redactor, {\n            eventId: cryptoEventId(),\n            runId: traceContext?.runId,\n            spanId: traceContext?.spanId,\n            provider,\n            model: req.model,\n            endpoint,\n            baseUrl,\n            attemptIndex: attempt,\n            direction: 'error',\n            timestamp: Date.now(),\n            durationMs: Date.now() - started,\n            statusCode: res.status,\n            responseHeaders,\n            responseBody: text,\n            errorMessage: `non-JSON response: ${parseErr instanceof Error ? parseErr.message : String(parseErr)}`,\n            redactedFields: [],\n          })\n          attemptErrorRecorded = true\n        }\n        throw parseErr\n      }\n      if (sink) {\n        await recordRaw(sink, redactor, {\n          eventId: cryptoEventId(),\n          runId: traceContext?.runId,\n          spanId: traceContext?.spanId,\n          provider,\n          model: req.model,\n          endpoint,\n          baseUrl,\n          attemptIndex: attempt,\n          direction: 'response',\n          timestamp: Date.now(),\n          durationMs: Date.now() - started,\n          statusCode: res.status,\n          responseHeaders,\n          responseBody: json,\n          redactedFields: [],\n        })\n      }\n      const choice = (\n        json.choices as\n          | Array<{ message?: { content?: string }; finish_reason?: string | null }>\n          | undefined\n      )?.[0]\n      const usageRaw = (json.usage as Record<string, unknown> | undefined) ?? {}\n      const costFromProxy = (json._response_cost ?? json.cost_usd) as number | undefined\n      const content = choice?.message?.content ?? ''\n\n      return {\n        content,\n        finishReason: choice?.finish_reason ?? null,\n        contentEmpty: content.trim().length === 0,\n        usage: {\n          promptTokens: Number(usageRaw.prompt_tokens ?? 0),\n          completionTokens: Number(usageRaw.completion_tokens ?? 0),\n          totalTokens: Number(usageRaw.total_tokens ?? 0),\n          cachedPromptTokens:\n            usageRaw.prompt_tokens_details && typeof usageRaw.prompt_tokens_details === 'object'\n              ? Number(\n                  (usageRaw.prompt_tokens_details as Record<string, unknown>).cached_tokens ?? 0,\n                )\n              : undefined,\n        },\n        costUsd: typeof costFromProxy === 'number' ? costFromProxy : null,\n        model: (json.model as string) ?? req.model,\n        durationMs: Date.now() - started,\n        raw: json,\n      }\n    } catch (err) {\n      clearTimeout(timeoutHandle)\n      lastErr = err\n      // A caller cancel is fatal even though an AbortError matches the\n      // transient patterns — a cancelled call must surface immediately, not\n      // be retried against the same dead intent.\n      if (callerSignal?.aborted) {\n        if (sink && !attemptErrorRecorded) {\n          await recordRaw(sink, redactor, {\n            eventId: cryptoEventId(),\n            runId: traceContext?.runId,\n            spanId: traceContext?.spanId,\n            provider,\n            model: req.model,\n            endpoint,\n            baseUrl,\n            attemptIndex: attempt,\n            direction: 'error',\n            timestamp: Date.now(),\n            durationMs: Date.now() - started,\n            errorMessage: err instanceof Error ? err.message : String(err),\n            redactedFields: [],\n          })\n        }\n        throw err\n      }\n      if (sink && !attemptErrorRecorded) {\n        // Record only if neither the !res.ok branch nor the JSON.parse catch\n        // already produced an error event for this attempt. Covers network\n        // failures, timeouts, and aborts.\n        await recordRaw(sink, redactor, {\n          eventId: cryptoEventId(),\n          runId: traceContext?.runId,\n          spanId: traceContext?.spanId,\n          provider,\n          model: req.model,\n          endpoint,\n          baseUrl,\n          attemptIndex: attempt,\n          direction: 'error',\n          timestamp: Date.now(),\n          durationMs: Date.now() - started,\n          errorMessage: err instanceof Error ? err.message : String(err),\n          redactedFields: [],\n        })\n      }\n      if (\n        attempt < maxRetries - 1 &&\n        isTransientLlmError(err) &&\n        !deadlineExceeded(deadlineStart, deadlineMs)\n      ) {\n        await sleep(backoffMs(attempt))\n        continue\n      }\n      throw err\n    }\n  }\n  throw lastErr instanceof Error ? lastErr : new Error(String(lastErr))\n}\n\nasync function recordRaw(\n  sink: RawProviderSink,\n  redactor: ProviderRedactor,\n  event: RawProviderEvent,\n): Promise<void> {\n  // Errors from sinks must not crash the LLM call. Forensic capture is\n  // best-effort; the structured trace is the system of record.\n  try {\n    await sink.record(redactor(event))\n  } catch {\n    // Intentionally swallowed.\n  }\n}\n\nfunction headersToObject(h: Headers): Record<string, string> {\n  const out: Record<string, string> = {}\n  h.forEach((value, key) => {\n    out[key] = value\n  })\n  return out\n}\n\nfunction cryptoEventId(): string {\n  if (typeof globalThis.crypto?.randomUUID === 'function') return globalThis.crypto.randomUUID()\n  return `${Date.now().toString(36)}-${Math.random().toString(36).slice(2, 10)}`\n}\n\n/**\n * Structured-output call. Returns parsed JSON plus the raw result envelope.\n * Degrades `jsonSchema` → `jsonMode` on a 400 that names the schema param —\n * critical for deepseek-v3/v4, kimi-k2.6, and other models that don't accept\n * the `response_format.json_schema` shape but DO accept `json_object`.\n */\nexport async function callLlmJson<T = unknown>(\n  req: LlmCallRequest,\n  opts: LlmClientOptions = {},\n): Promise<{ value: T; result: LlmCallResult }> {\n  try {\n    const result = await callLlm({ ...req, jsonMode: req.jsonMode ?? !req.jsonSchema }, opts)\n    const value = parseJsonSafely<T>(result.content, result.model)\n    return { value, result }\n  } catch (err) {\n    if (err instanceof LlmCallError && isSchemaRejection(err.status, err.body) && req.jsonSchema) {\n      // Degrade to json_object + retry.\n      const degradedReq: LlmCallRequest = { ...req, jsonMode: true, jsonSchema: undefined }\n      const result = await callLlm(degradedReq, opts)\n      const value = parseJsonSafely<T>(result.content, result.model)\n      return { value, result }\n    }\n    throw err\n  }\n}\n\nfunction parseJsonSafely<T>(content: string, model: string): T {\n  const stripped = extractJsonPayload(content)\n  try {\n    return JSON.parse(stripped) as T\n  } catch (err) {\n    throw new Error(\n      `LLM returned non-JSON content (model=${model}): ${\n        err instanceof Error ? err.message : String(err)\n      }\\n--- raw content ---\\n${content.slice(0, 800)}`,\n    )\n  }\n}\n\n// ─── Route assertion ────────────────────────────────────────────────────\n\nexport type LlmRouteAssertionReason =\n  | 'no_explicit_base_url'\n  | 'base_url_blocked'\n  | 'base_url_not_allowed'\n  | 'no_auth'\n  | 'wrong_provider'\n\nexport class LlmRouteAssertionError extends CaptureIntegrityError {\n  constructor(\n    message: string,\n    public readonly reason: LlmRouteAssertionReason,\n    public readonly baseUrl: string,\n  ) {\n    super(message)\n  }\n}\n\nexport interface LlmRouteRequirements {\n  /**\n   * Throw if `opts.baseUrl` is undefined, i.e. the call would fall back to\n   * `DEFAULT_BASE_URL`. Set this for evaluation runs where silently using\n   * the public/free-tier router is a defect — the launch reviewer needs to\n   * know exactly which provider answered.\n   */\n  requireExplicitBaseUrl?: boolean\n  /**\n   * Allowlist of acceptable base URLs. Strings match by prefix\n   * (case-insensitive); RegExps test against the full base URL.\n   */\n  allowedBaseUrls?: Array<string | RegExp>\n  /** Blocklist that takes precedence over `allowedBaseUrls`. */\n  blockedBaseUrls?: Array<string | RegExp>\n  /** Throw if no auth header / api key is configured. */\n  requireAuth?: boolean\n  /**\n   * Logical provider id the configured `baseUrl` is expected to match (via\n   * `providerFromBaseUrl`). Mainly useful when paired with `requireExplicitBaseUrl`.\n   */\n  expectedProvider?: string\n}\n\n/**\n * Fail-loud assertion that the configured LLM client points at the route\n * the caller intends. Designed for the matrix-runner preflight: invoke\n * once before any LLM call to catch misconfiguration before a sweep burns\n * dollars on the wrong provider.\n *\n * Throws `LlmRouteAssertionError`. Pure — no I/O — so it's safe to call\n * from constructors and CI gates.\n */\nexport function assertLlmRoute(opts: LlmClientOptions, req: LlmRouteRequirements = {}): void {\n  const baseUrlExplicit = opts.baseUrl !== undefined\n  const baseUrl = (opts.baseUrl ?? DEFAULT_BASE_URL).replace(/\\/+$/, '')\n\n  if (req.requireExplicitBaseUrl && !baseUrlExplicit) {\n    throw new LlmRouteAssertionError(\n      `assertLlmRoute: requireExplicitBaseUrl set but opts.baseUrl is undefined; would fall back to ${DEFAULT_BASE_URL}.`,\n      'no_explicit_base_url',\n      baseUrl,\n    )\n  }\n\n  if (req.blockedBaseUrls?.some((p) => matchUrl(baseUrl, p))) {\n    throw new LlmRouteAssertionError(\n      `assertLlmRoute: baseUrl ${baseUrl} matches a blocked pattern.`,\n      'base_url_blocked',\n      baseUrl,\n    )\n  }\n\n  if (req.allowedBaseUrls && req.allowedBaseUrls.length > 0) {\n    const ok = req.allowedBaseUrls.some((p) => matchUrl(baseUrl, p))\n    if (!ok) {\n      throw new LlmRouteAssertionError(\n        `assertLlmRoute: baseUrl ${baseUrl} is not in the allowed list (${req.allowedBaseUrls.map(describePattern).join(', ')}).`,\n        'base_url_not_allowed',\n        baseUrl,\n      )\n    }\n  }\n\n  if (req.requireAuth && !opts.apiKey && !opts.bearer && !opts.authHeader) {\n    throw new LlmRouteAssertionError(\n      `assertLlmRoute: requireAuth set but no apiKey, bearer, or authHeader was supplied.`,\n      'no_auth',\n      baseUrl,\n    )\n  }\n\n  if (req.expectedProvider) {\n    const actual = opts.provider ?? providerFromBaseUrl(baseUrl)\n    if (actual !== req.expectedProvider) {\n      throw new LlmRouteAssertionError(\n        `assertLlmRoute: expected provider ${req.expectedProvider} but baseUrl ${baseUrl} resolves to ${actual}.`,\n        'wrong_provider',\n        baseUrl,\n      )\n    }\n  }\n}\n\nfunction matchUrl(url: string, pattern: string | RegExp): boolean {\n  if (pattern instanceof RegExp) return pattern.test(url)\n  return url.toLowerCase().startsWith(pattern.toLowerCase())\n}\n\nfunction describePattern(p: string | RegExp): string {\n  return p instanceof RegExp ? p.source : p\n}\n\n/**\n * Probe whether a model is reachable. Returns latency + null error on\n * success; `ok=false` + error message on any failure (HTTP, timeout,\n * network, parse). Designed for sweep preflights — fail loud at the\n * boundary before burning a 30-leaf run on a misconfigured router.\n *\n * Sends a tiny `ping` message with `maxTokens=64`. Reasoning models\n * (glm-5.1, deepseek-v4) can burn the entire budget on internal reasoning\n * for short prompts, so don't tighten this further. We don't validate\n * content; HTTP 200 means reachable.\n */\nexport async function probeLlm(\n  model: string,\n  opts: LlmClientOptions & { timeoutMs?: number } = {},\n): Promise<{ ok: boolean; latencyMs: number; error: string | null }> {\n  const start = Date.now()\n  try {\n    await callLlm(\n      {\n        model,\n        messages: [{ role: 'user', content: 'ping' }],\n        maxTokens: 64,\n        timeoutMs: opts.timeoutMs ?? 30_000,\n      },\n      opts,\n    )\n    return { ok: true, latencyMs: Date.now() - start, error: null }\n  } catch (err) {\n    return {\n      ok: false,\n      latencyMs: Date.now() - start,\n      error: err instanceof Error ? err.message : String(err),\n    }\n  }\n}\n\n/**\n * Stateful client — construct once with defaults, call many times.\n * Thin wrapper around the free functions; exists for callers that want\n * to inject a single configured instance into multiple primitives.\n */\nexport class LlmClient {\n  constructor(private readonly opts: LlmClientOptions = {}) {}\n\n  call(req: LlmCallRequest, per?: LlmClientOptions): Promise<LlmCallResult> {\n    return callLlm(req, { ...this.opts, ...per })\n  }\n\n  callJson<T = unknown>(\n    req: LlmCallRequest,\n    per?: LlmClientOptions,\n  ): Promise<{ value: T; result: LlmCallResult }> {\n    return callLlmJson<T>(req, { ...this.opts, ...per })\n  }\n}\n"],"mappings":";;;;;;;;;;AAqGO,IAAM,eAAN,cAA2B,eAAe;AAAA,EAC/C,YACE,SACgB,QACA,MACA,OAChB;AACA,UAAM,SAAS,OAAO;AAJN;AACA;AACA;AAAA,EAGlB;AAAA,EALkB;AAAA,EACA;AAAA,EACA;AAIpB;AAoDA,IAAM,mBAAmB;AASzB,IAAM,qBAAqB,OAAO,QAAQ,IAAI,qBAAqB,KAAK;AACxE,IAAM,sBAAsB,OAAO,QAAQ,IAAI,sBAAsB,KAAK;AAE1E,IAAM,mBAAmB,oBAAI,IAAI,CAAC,KAAK,KAAK,KAAK,GAAG,CAAC;AAWrD,IAAM,2BAA8C;AAAA,EAClD;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AACF;AAcO,SAAS,oBAAoB,KAAuB;AACzD,SAAO,kBAAkB,KAAK,CAAC;AACjC;AAEA,SAAS,kBAAkB,KAAc,OAAwB;AAC/D,MAAI,eAAe,aAAc,QAAO,iBAAiB,IAAI,IAAI,MAAM;AACvE,MAAI,EAAE,eAAe,OAAQ,QAAO;AAGpC,QAAM,SAAU,IAA6B;AAC7C,MAAI,OAAO,WAAW,YAAY,iBAAiB,IAAI,MAAM,EAAG,QAAO;AACvE,QAAM,OAAQ,IAA2B;AACzC,QAAM,WAAW,GAAG,IAAI,IAAI;AAAA,EAAK,IAAI,OAAO;AAAA,EAAK,OAAO,SAAS,WAAW,OAAO,EAAE;AACrF,MAAI,yBAAyB,KAAK,CAAC,MAAM,EAAE,KAAK,QAAQ,CAAC,EAAG,QAAO;AACnE,QAAM,QAAS,IAA4B;AAC3C,MAAI,QAAQ,KAAK,iBAAiB,SAAS,UAAU,KAAK;AACxD,WAAO,kBAAkB,OAAO,QAAQ,CAAC;AAAA,EAC3C;AACA,SAAO;AACT;AAEA,SAAS,gBAAgB,SAAiC;AACxD,QAAM,IAAI,QAAQ,IAAI,aAAa;AACnC,MAAI,CAAC,EAAG,QAAO;AACf,QAAM,WAAW,OAAO,CAAC;AACzB,MAAI,OAAO,SAAS,QAAQ,KAAK,WAAW,EAAG,QAAO,WAAW;AACjE,QAAM,SAAS,KAAK,MAAM,CAAC;AAC3B,MAAI,OAAO,SAAS,MAAM,EAAG,QAAO,KAAK,IAAI,GAAG,SAAS,KAAK,IAAI,CAAC;AACnE,SAAO;AACT;AAGO,SAAS,UAAU,SAAyB;AACjD,SAAO,KAAK,IAAI,MAAM,KAAK,SAAS,IAAM;AAC5C;AAEA,SAAS,aAAa,MAAgD;AACpE,QAAM,UAAkC;AAAA,IACtC,gBAAgB;AAAA,IAChB,QAAQ;AAAA,EACV;AACA,MAAI,KAAK,YAAY;AACnB,YAAQ,KAAK,WAAW,IAAI,IAAI,KAAK,WAAW;AAAA,EAClD,WAAW,KAAK,UAAU,KAAK,QAAQ;AACrC,YAAQ,gBAAgB,UAAU,KAAK,UAAU,KAAK,MAAM;AAAA,EAC9D;AACA,SAAO;AACT;AAEA,SAAS,kBAAkB,QAAgB,MAAuB;AAChE,MAAI,WAAW,IAAK,QAAO;AAC3B,QAAM,QAAQ,KAAK,YAAY;AAC/B,SACE,MAAM,SAAS,iBAAiB,KAChC,MAAM,SAAS,aAAa,KAC5B,MAAM,SAAS,gBAAgB,KAC/B,MAAM,SAAS,eAAe;AAElC;AAEA,SAAS,UAAU,KAAqB,iBAAmD;AACzF,QAAM,OAAgC;AAAA,IACpC,OAAO,IAAI;AAAA,IACX,UAAU,IAAI;AAAA,IACd,aAAa,IAAI,eAAe;AAAA,EAClC;AACA,MAAI,IAAI,aAAa,MAAM;AACzB,QAAI,wBAAwB,IAAI,KAAK,EAAG,MAAK,wBAAwB,IAAI;AAAA,QACpE,MAAK,aAAa,IAAI;AAAA,EAC7B;AAEA,MAAI,IAAI,cAAc,CAAC,iBAAiB;AACtC,SAAK,kBAAkB;AAAA,MACrB,MAAM;AAAA,MACN,aAAa,EAAE,MAAM,IAAI,WAAW,MAAM,QAAQ,IAAI,WAAW,QAAQ,QAAQ,KAAK;AAAA,IACxF;AAAA,EACF,WAAW,IAAI,YAAY,IAAI,YAAY;AACzC,SAAK,kBAAkB,EAAE,MAAM,cAAc;AAAA,EAC/C;AAEA,SAAO;AACT;AAEA,SAAS,wBAAwB,OAAwB;AACvD,SAAO,oBAAoB,KAAK,KAAK;AACvC;AAEA,eAAe,MAAM,IAA2B;AAC9C,SAAO,IAAI,QAAQ,CAAC,YAAY,WAAW,SAAS,EAAE,CAAC;AACzD;AASA,SAAS,YAAY,mBAAoC,QAAmC;AAC1F,MAAI,CAAC,OAAQ,QAAO,kBAAkB;AACtC,MAAI,OAAQ,YAAkC,QAAQ,YAAY;AAChE,WAAO,YAAY,IAAI,CAAC,kBAAkB,QAAQ,MAAM,CAAC;AAAA,EAC3D;AACA,MAAI,OAAO,SAAS;AAClB,sBAAkB,MAAM;AAAA,EAC1B,OAAO;AACL,WAAO,iBAAiB,SAAS,MAAM,kBAAkB,MAAM,GAAG,EAAE,MAAM,KAAK,CAAC;AAAA,EAClF;AACA,SAAO,kBAAkB;AAC3B;AAGA,SAAS,iBAAiB,OAAe,YAAyC;AAChF,SAAO,cAAc,QAAQ,KAAK,IAAI,IAAI,SAAS;AACrD;AASO,SAAS,gBAAgB,KAAqB;AACnD,QAAM,UAAU,IAAI,KAAK;AACzB,QAAM,IAAI,QAAQ,MAAM,yCAAyC;AACjE,SAAO,IAAI,EAAE,CAAC,EAAG,KAAK,IAAI;AAC5B;AAEO,SAAS,mBAAmB,KAAqB;AACtD,QAAM,WAAW,gBAAgB,GAAG;AACpC,MAAI;AACF,SAAK,MAAM,QAAQ;AACnB,WAAO;AAAA,EACT,QAAQ;AAAA,EAER;AAEA,QAAM,SAAS,CAAC,GAAG,SAAS,SAAS,OAAO,CAAC,EAC1C,IAAI,CAAC,UAAU,MAAM,KAAK,EAC1B,OAAO,CAAC,UAAU,SAAS,IAAI;AAClC,aAAW,SAAS,QAAQ;AAC1B,UAAM,YAAY,oBAAoB,UAAU,KAAK;AACrD,QAAI,CAAC,UAAW;AAChB,QAAI;AACF,WAAK,MAAM,SAAS;AACpB,aAAO;AAAA,IACT,QAAQ;AAAA,IAER;AAAA,EACF;AAEA,SAAO;AACT;AAEA,SAAS,oBAAoB,OAAe,OAA8B;AACxE,QAAM,SAAS,MAAM,KAAK;AAC1B,QAAM,SAAS,WAAW,MAAM,MAAM,WAAW,MAAM,MAAM;AAC7D,MAAI,CAAC,OAAQ,QAAO;AAEpB,QAAM,QAAkB,CAAC,MAAM;AAC/B,MAAI,aAAa;AACjB,MAAI,YAAY;AAEhB,WAAS,IAAI,QAAQ,GAAG,IAAI,MAAM,QAAQ,KAAK;AAC7C,UAAM,OAAO,MAAM,CAAC;AACpB,QAAI,WAAW;AACb,kBAAY;AACZ;AAAA,IACF;AACA,QAAI,SAAS,MAAM;AACjB,kBAAY;AACZ;AAAA,IACF;AACA,QAAI,SAAS,KAAK;AAChB,mBAAa,CAAC;AACd;AAAA,IACF;AACA,QAAI,WAAY;AAEhB,QAAI,SAAS,IAAK,OAAM,KAAK,GAAG;AAAA,aACvB,SAAS,IAAK,OAAM,KAAK,GAAG;AAAA,aAC5B,SAAS,MAAM,MAAM,SAAS,CAAC,GAAG;AACzC,YAAM,IAAI;AACV,UAAI,MAAM,WAAW,EAAG,QAAO,MAAM,MAAM,OAAO,IAAI,CAAC;AAAA,IACzD;AAAA,EACF;AAEA,SAAO;AACT;AAOA,eAAsB,QACpB,KACA,OAAyB,CAAC,GACF;AACxB,QAAM,WAAW,KAAK,WAAW,kBAAkB,QAAQ,QAAQ,EAAE;AACrE,QAAM,MAAM,GAAG,OAAO;AACtB,QAAM,WAAW;AACjB,QAAM,YAAY,IAAI,aAAa,KAAK,oBAAoB;AAC5D,QAAM,aAAa,KAAK,cAAc;AACtC,QAAM,UAAU,KAAK,SAAS,WAAW;AACzC,QAAM,UAAU,aAAa,IAAI;AACjC,QAAM,WAAW,KAAK,YAAY,oBAAoB,OAAO;AAC7D,QAAM,OAAO,KAAK;AAClB,QAAM,WAAW,KAAK,YAAY;AAClC,QAAM,eAAe,KAAK;AAC1B,QAAM,eAAe,KAAK;AAC1B,QAAM,aAAa,KAAK;AACxB,QAAM,gBAAgB,KAAK,IAAI;AAE/B,MAAI;AACJ,WAAS,UAAU,GAAG,UAAU,YAAY,WAAW;AAGrD,QAAI,cAAc,SAAS;AACzB,YAAM,IAAI,aAAa,oCAAoC,YAAY;AAAA,IACzE;AAGA,QAAI,UAAU,KAAK,iBAAiB,eAAe,UAAU,GAAG;AAC9D,YAAM,mBAAmB,QAAQ,UAAU,IAAI,MAAM,OAAO,OAAO,CAAC;AAAA,IACtE;AACA,UAAM,aAAa,IAAI,gBAAgB;AACvC,UAAM,gBAAgB,YAAY,YAAY,YAAY;AAC1D,UAAM,gBAAgB,WAAW,MAAM,WAAW,MAAM,GAAG,SAAS;AACpE,UAAM,UAAU,KAAK,IAAI;AACzB,UAAM,cAAc,UAAU,KAAK,KAAK;AACxC,QAAI,uBAAuB;AAC3B,QAAI,MAAM;AACR,YAAM,UAAU,MAAM,UAAU;AAAA,QAC9B,SAAS,cAAc;AAAA,QACvB,OAAO,cAAc;AAAA,QACrB,QAAQ,cAAc;AAAA,QACtB;AAAA,QACA,OAAO,IAAI;AAAA,QACX;AAAA,QACA;AAAA,QACA,cAAc;AAAA,QACd,WAAW;AAAA,QACX,WAAW;AAAA,QACX,gBAAgB;AAAA,QAChB;AAAA,QACA,gBAAgB,CAAC;AAAA,MACnB,CAAC;AAAA,IACH;AAEA,QAAI;AACF,YAAM,MAAM,MAAM,QAAQ,KAAK;AAAA,QAC7B,QAAQ;AAAA,QACR;AAAA,QACA,MAAM,KAAK,UAAU,WAAW;AAAA,QAChC,QAAQ;AAAA,MACV,CAAC;AACD,mBAAa,aAAa;AAC1B,YAAM,kBAAkB,OAAO,gBAAgB,IAAI,OAAO,IAAI;AAE9D,UAAI,CAAC,IAAI,IAAI;AACX,cAAM,OAAO,MAAM,IAAI,KAAK;AAC5B,YAAI,MAAM;AACR,gBAAM,UAAU,MAAM,UAAU;AAAA,YAC9B,SAAS,cAAc;AAAA,YACvB,OAAO,cAAc;AAAA,YACrB,QAAQ,cAAc;AAAA,YACtB;AAAA,YACA,OAAO,IAAI;AAAA,YACX;AAAA,YACA;AAAA,YACA,cAAc;AAAA,YACd,WAAW;AAAA,YACX,WAAW,KAAK,IAAI;AAAA,YACpB,YAAY,KAAK,IAAI,IAAI;AAAA,YACzB,YAAY,IAAI;AAAA,YAChB;AAAA,YACA,cAAc;AAAA,YACd,cAAc,QAAQ,IAAI,MAAM;AAAA,YAChC,gBAAgB,CAAC;AAAA,UACnB,CAAC;AACD,iCAAuB;AAAA,QACzB;AACA,cAAM,MAAM,IAAI;AAAA,UACd,YAAY,IAAI,MAAM,KAAK,KAAK,MAAM,GAAG,GAAG,CAAC;AAAA,UAC7C,IAAI;AAAA,UACJ;AAAA,UACA,IAAI;AAAA,QACN;AACA,YACE,iBAAiB,IAAI,IAAI,MAAM,KAC/B,UAAU,aAAa,KACvB,CAAC,iBAAiB,eAAe,UAAU,GAC3C;AACA,oBAAU;AACV,gBAAM,aAAa,gBAAgB,IAAI,OAAO;AAC9C,gBAAM,MAAM,cAAc,UAAU,OAAO,CAAC;AAC5C;AAAA,QACF;AACA,cAAM;AAAA,MACR;AAEA,YAAM,OAAO,MAAM,IAAI,KAAK;AAC5B,UAAI;AACJ,UAAI;AACF,eAAO,KAAK,MAAM,IAAI;AAAA,MACxB,SAAS,UAAU;AACjB,YAAI,MAAM;AACR,gBAAM,UAAU,MAAM,UAAU;AAAA,YAC9B,SAAS,cAAc;AAAA,YACvB,OAAO,cAAc;AAAA,YACrB,QAAQ,cAAc;AAAA,YACtB;AAAA,YACA,OAAO,IAAI;AAAA,YACX;AAAA,YACA;AAAA,YACA,cAAc;AAAA,YACd,WAAW;AAAA,YACX,WAAW,KAAK,IAAI;AAAA,YACpB,YAAY,KAAK,IAAI,IAAI;AAAA,YACzB,YAAY,IAAI;AAAA,YAChB;AAAA,YACA,cAAc;AAAA,YACd,cAAc,sBAAsB,oBAAoB,QAAQ,SAAS,UAAU,OAAO,QAAQ,CAAC;AAAA,YACnG,gBAAgB,CAAC;AAAA,UACnB,CAAC;AACD,iCAAuB;AAAA,QACzB;AACA,cAAM;AAAA,MACR;AACA,UAAI,MAAM;AACR,cAAM,UAAU,MAAM,UAAU;AAAA,UAC9B,SAAS,cAAc;AAAA,UACvB,OAAO,cAAc;AAAA,UACrB,QAAQ,cAAc;AAAA,UACtB;AAAA,UACA,OAAO,IAAI;AAAA,UACX;AAAA,UACA;AAAA,UACA,cAAc;AAAA,UACd,WAAW;AAAA,UACX,WAAW,KAAK,IAAI;AAAA,UACpB,YAAY,KAAK,IAAI,IAAI;AAAA,UACzB,YAAY,IAAI;AAAA,UAChB;AAAA,UACA,cAAc;AAAA,UACd,gBAAgB,CAAC;AAAA,QACnB,CAAC;AAAA,MACH;AACA,YAAM,SACJ,KAAK,UAGH,CAAC;AACL,YAAM,WAAY,KAAK,SAAiD,CAAC;AACzE,YAAM,gBAAiB,KAAK,kBAAkB,KAAK;AACnD,YAAM,UAAU,QAAQ,SAAS,WAAW;AAE5C,aAAO;AAAA,QACL;AAAA,QACA,cAAc,QAAQ,iBAAiB;AAAA,QACvC,cAAc,QAAQ,KAAK,EAAE,WAAW;AAAA,QACxC,OAAO;AAAA,UACL,cAAc,OAAO,SAAS,iBAAiB,CAAC;AAAA,UAChD,kBAAkB,OAAO,SAAS,qBAAqB,CAAC;AAAA,UACxD,aAAa,OAAO,SAAS,gBAAgB,CAAC;AAAA,UAC9C,oBACE,SAAS,yBAAyB,OAAO,SAAS,0BAA0B,WACxE;AAAA,YACG,SAAS,sBAAkD,iBAAiB;AAAA,UAC/E,IACA;AAAA,QACR;AAAA,QACA,SAAS,OAAO,kBAAkB,WAAW,gBAAgB;AAAA,QAC7D,OAAQ,KAAK,SAAoB,IAAI;AAAA,QACrC,YAAY,KAAK,IAAI,IAAI;AAAA,QACzB,KAAK;AAAA,MACP;AAAA,IACF,SAAS,KAAK;AACZ,mBAAa,aAAa;AAC1B,gBAAU;AAIV,UAAI,cAAc,SAAS;AACzB,YAAI,QAAQ,CAAC,sBAAsB;AACjC,gBAAM,UAAU,MAAM,UAAU;AAAA,YAC9B,SAAS,cAAc;AAAA,YACvB,OAAO,cAAc;AAAA,YACrB,QAAQ,cAAc;AAAA,YACtB;AAAA,YACA,OAAO,IAAI;AAAA,YACX;AAAA,YACA;AAAA,YACA,cAAc;AAAA,YACd,WAAW;AAAA,YACX,WAAW,KAAK,IAAI;AAAA,YACpB,YAAY,KAAK,IAAI,IAAI;AAAA,YACzB,cAAc,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;AAAA,YAC7D,gBAAgB,CAAC;AAAA,UACnB,CAAC;AAAA,QACH;AACA,cAAM;AAAA,MACR;AACA,UAAI,QAAQ,CAAC,sBAAsB;AAIjC,cAAM,UAAU,MAAM,UAAU;AAAA,UAC9B,SAAS,cAAc;AAAA,UACvB,OAAO,cAAc;AAAA,UACrB,QAAQ,cAAc;AAAA,UACtB;AAAA,UACA,OAAO,IAAI;AAAA,UACX;AAAA,UACA;AAAA,UACA,cAAc;AAAA,UACd,WAAW;AAAA,UACX,WAAW,KAAK,IAAI;AAAA,UACpB,YAAY,KAAK,IAAI,IAAI;AAAA,UACzB,cAAc,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;AAAA,UAC7D,gBAAgB,CAAC;AAAA,QACnB,CAAC;AAAA,MACH;AACA,UACE,UAAU,aAAa,KACvB,oBAAoB,GAAG,KACvB,CAAC,iBAAiB,eAAe,UAAU,GAC3C;AACA,cAAM,MAAM,UAAU,OAAO,CAAC;AAC9B;AAAA,MACF;AACA,YAAM;AAAA,IACR;AAAA,EACF;AACA,QAAM,mBAAmB,QAAQ,UAAU,IAAI,MAAM,OAAO,OAAO,CAAC;AACtE;AAEA,eAAe,UACb,MACA,UACA,OACe;AAGf,MAAI;AACF,UAAM,KAAK,OAAO,SAAS,KAAK,CAAC;AAAA,EACnC,QAAQ;AAAA,EAER;AACF;AAEA,SAAS,gBAAgB,GAAoC;AAC3D,QAAM,MAA8B,CAAC;AACrC,IAAE,QAAQ,CAAC,OAAO,QAAQ;AACxB,QAAI,GAAG,IAAI;AAAA,EACb,CAAC;AACD,SAAO;AACT;AAEA,SAAS,gBAAwB;AAC/B,MAAI,OAAO,WAAW,QAAQ,eAAe,WAAY,QAAO,WAAW,OAAO,WAAW;AAC7F,SAAO,GAAG,KAAK,IAAI,EAAE,SAAS,EAAE,CAAC,IAAI,KAAK,OAAO,EAAE,SAAS,EAAE,EAAE,MAAM,GAAG,EAAE,CAAC;AAC9E;AAQA,eAAsB,YACpB,KACA,OAAyB,CAAC,GACoB;AAC9C,MAAI;AACF,UAAM,SAAS,MAAM,QAAQ,EAAE,GAAG,KAAK,UAAU,IAAI,YAAY,CAAC,IAAI,WAAW,GAAG,IAAI;AACxF,UAAM,QAAQ,gBAAmB,OAAO,SAAS,OAAO,KAAK;AAC7D,WAAO,EAAE,OAAO,OAAO;AAAA,EACzB,SAAS,KAAK;AACZ,QAAI,eAAe,gBAAgB,kBAAkB,IAAI,QAAQ,IAAI,IAAI,KAAK,IAAI,YAAY;AAE5F,YAAM,cAA8B,EAAE,GAAG,KAAK,UAAU,MAAM,YAAY,OAAU;AACpF,YAAM,SAAS,MAAM,QAAQ,aAAa,IAAI;AAC9C,YAAM,QAAQ,gBAAmB,OAAO,SAAS,OAAO,KAAK;AAC7D,aAAO,EAAE,OAAO,OAAO;AAAA,IACzB;AACA,UAAM;AAAA,EACR;AACF;AAEA,SAAS,gBAAmB,SAAiB,OAAkB;AAC7D,QAAM,WAAW,mBAAmB,OAAO;AAC3C,MAAI;AACF,WAAO,KAAK,MAAM,QAAQ;AAAA,EAC5B,SAAS,KAAK;AACZ,UAAM,IAAI;AAAA,MACR,wCAAwC,KAAK,MAC3C,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG,CACjD;AAAA;AAAA,EAA0B,QAAQ,MAAM,GAAG,GAAG,CAAC;AAAA,IACjD;AAAA,EACF;AACF;AAWO,IAAM,yBAAN,cAAqC,sBAAsB;AAAA,EAChE,YACE,SACgB,QACA,SAChB;AACA,UAAM,OAAO;AAHG;AACA;AAAA,EAGlB;AAAA,EAJkB;AAAA,EACA;AAIpB;AAmCO,SAAS,eAAe,MAAwB,MAA4B,CAAC,GAAS;AAC3F,QAAM,kBAAkB,KAAK,YAAY;AACzC,QAAM,WAAW,KAAK,WAAW,kBAAkB,QAAQ,QAAQ,EAAE;AAErE,MAAI,IAAI,0BAA0B,CAAC,iBAAiB;AAClD,UAAM,IAAI;AAAA,MACR,gGAAgG,gBAAgB;AAAA,MAChH;AAAA,MACA;AAAA,IACF;AAAA,EACF;AAEA,MAAI,IAAI,iBAAiB,KAAK,CAAC,MAAM,SAAS,SAAS,CAAC,CAAC,GAAG;AAC1D,UAAM,IAAI;AAAA,MACR,2BAA2B,OAAO;AAAA,MAClC;AAAA,MACA;AAAA,IACF;AAAA,EACF;AAEA,MAAI,IAAI,mBAAmB,IAAI,gBAAgB,SAAS,GAAG;AACzD,UAAM,KAAK,IAAI,gBAAgB,KAAK,CAAC,MAAM,SAAS,SAAS,CAAC,CAAC;AAC/D,QAAI,CAAC,IAAI;AACP,YAAM,IAAI;AAAA,QACR,2BAA2B,OAAO,gCAAgC,IAAI,gBAAgB,IAAI,eAAe,EAAE,KAAK,IAAI,CAAC;AAAA,QACrH;AAAA,QACA;AAAA,MACF;AAAA,IACF;AAAA,EACF;AAEA,MAAI,IAAI,eAAe,CAAC,KAAK,UAAU,CAAC,KAAK,UAAU,CAAC,KAAK,YAAY;AACvE,UAAM,IAAI;AAAA,MACR;AAAA,MACA;AAAA,MACA;AAAA,IACF;AAAA,EACF;AAEA,MAAI,IAAI,kBAAkB;AACxB,UAAM,SAAS,KAAK,YAAY,oBAAoB,OAAO;AAC3D,QAAI,WAAW,IAAI,kBAAkB;AACnC,YAAM,IAAI;AAAA,QACR,qCAAqC,IAAI,gBAAgB,gBAAgB,OAAO,gBAAgB,MAAM;AAAA,QACtG;AAAA,QACA;AAAA,MACF;AAAA,IACF;AAAA,EACF;AACF;AAEA,SAAS,SAAS,KAAa,SAAmC;AAChE,MAAI,mBAAmB,OAAQ,QAAO,QAAQ,KAAK,GAAG;AACtD,SAAO,IAAI,YAAY,EAAE,WAAW,QAAQ,YAAY,CAAC;AAC3D;AAEA,SAAS,gBAAgB,GAA4B;AACnD,SAAO,aAAa,SAAS,EAAE,SAAS;AAC1C;AAaA,eAAsB,SACpB,OACA,OAAkD,CAAC,GACgB;AACnE,QAAM,QAAQ,KAAK,IAAI;AACvB,MAAI;AACF,UAAM;AAAA,MACJ;AAAA,QACE;AAAA,QACA,UAAU,CAAC,EAAE,MAAM,QAAQ,SAAS,OAAO,CAAC;AAAA,QAC5C,WAAW;AAAA,QACX,WAAW,KAAK,aAAa;AAAA,MAC/B;AAAA,MACA;AAAA,IACF;AACA,WAAO,EAAE,IAAI,MAAM,WAAW,KAAK,IAAI,IAAI,OAAO,OAAO,KAAK;AAAA,EAChE,SAAS,KAAK;AACZ,WAAO;AAAA,MACL,IAAI;AAAA,MACJ,WAAW,KAAK,IAAI,IAAI;AAAA,MACxB,OAAO,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;AAAA,IACxD;AAAA,EACF;AACF;AAOO,IAAM,YAAN,MAAgB;AAAA,EACrB,YAA6B,OAAyB,CAAC,GAAG;AAA7B;AAAA,EAA8B;AAAA,EAA9B;AAAA,EAE7B,KAAK,KAAqB,KAAgD;AACxE,WAAO,QAAQ,KAAK,EAAE,GAAG,KAAK,MAAM,GAAG,IAAI,CAAC;AAAA,EAC9C;AAAA,EAEA,SACE,KACA,KAC8C;AAC9C,WAAO,YAAe,KAAK,EAAE,GAAG,KAAK,MAAM,GAAG,IAAI,CAAC;AAAA,EACrD;AACF;","names":[]}

package/dist/{chunk-QS3RBQPI.js → chunk-D5KBVULY.js}

@@ -1,6 +1,6 @@
 import {
   callLlmJson
-} from "./chunk-CVVHBFGN.js";
+} from "./chunk-CWNP4DV4.js";
 
 // src/wire/schemas.ts
 import { extendZodWithOpenApi } from "@asteasolutions/zod-to-openapi";
@@ -1002,4 +1002,4 @@ export {
   startServer,
   startServerAsync
 };
-//# sourceMappingURL=chunk-QS3RBQPI.js.map
+//# sourceMappingURL=chunk-D5KBVULY.js.map

package/dist/{chunk-ZFIBGEOL.js → chunk-EGPMSBEZ.js}

@@ -1,10 +1,10 @@
 import {
   runCampaign,
   summarizeBackendIntegrity
-} from "./chunk-TWS7AZEY.js";
+} from "./chunk-2K6UUZ7P.js";
 import {
   callLlm
-} from "./chunk-CVVHBFGN.js";
+} from "./chunk-CWNP4DV4.js";
 import {
   DEFAULT_REDACTION_RULES
 } from "./chunk-GGE4NNQT.js";
@@ -1443,7 +1443,7 @@ async function runImprovementLoop(opts) {
   const dispatchTimeoutMs = opts.dispatchTimeoutMs ?? DEFAULT_DISPATCH_TIMEOUT_MS;
   const optimization = await runOptimization({ ...opts, dispatchTimeoutMs });
   const winnerIsBaseline = optimization.winnerSurfaceHash === surfaceHash(opts.baselineSurface);
-  const { runCampaign: runCampaign2 } = await import("./run-campaign-RDGAM5KJ.js");
+  const { runCampaign: runCampaign2 } = await import("./run-campaign-7WNXMDSN.js");
   const baselineOnHoldout = await runCampaign2({
     ...opts,
     dispatchTimeoutMs,
@@ -1844,4 +1844,4 @@ export {
   provenanceSpansPath,
   emitLoopProvenance
 };
-//# sourceMappingURL=chunk-ZFIBGEOL.js.map
+//# sourceMappingURL=chunk-EGPMSBEZ.js.map

package/dist/{chunk-GSH6QNNS.js → chunk-KW53MSA5.js}

@@ -1,9 +1,9 @@
 import {
   assertLlmRoute
-} from "./chunk-CVVHBFGN.js";
+} from "./chunk-CWNP4DV4.js";
 import {
   researchReport
-} from "./chunk-CY6U5S3X.js";
+} from "./chunk-URWVKRCS.js";
 import {
   RunIntegrityError,
   assertRunCaptured
@@ -100,8 +100,11 @@ async function runEvalCampaign(opts) {
   const integrityReports = [];
   const failedRuns = [];
   let cursor = 0;
+  let aborting = false;
+  const genuineErrors = [];
+  const openRuns = /* @__PURE__ */ new Map();
   async function worker() {
-    while (true) {
+    while (!aborting) {
       const i = cursor++;
       if (i >= cells.length) return;
       const cell = cells[i];
@@ -114,7 +117,9 @@ async function runEvalCampaign(opts) {
           failedRuns.push(err.failed);
           if (err.integrity) integrityReports.push(err.integrity);
         } else {
-          throw err;
+          genuineErrors.push(err);
+          aborting = true;
+          return;
         }
       }
     }
@@ -142,6 +147,7 @@ async function runEvalCampaign(opts) {
       now: opts.now,
       onRunComplete: opts.onRunComplete
     });
+    openRuns.set(runId, emitter);
     const llmOpts = {
       ...opts.llmOpts,
       rawSink,
@@ -161,90 +167,101 @@ async function runEvalCampaign(opts) {
       rawSink,
       llmOpts
     };
-    const wallStart = now();
-    let outcome;
     try {
-      outcome = await opts.runner(ctx);
-    } catch (err) {
-      const message = err instanceof Error ? err.message : String(err);
+      const wallStart = now();
+      let outcome;
       try {
-        await emitter.abortRun(message);
-      } catch {
+        outcome = await opts.runner(ctx);
+      } catch (err) {
+        const message = err instanceof Error ? err.message : String(err);
+        await finalizeAbort(emitter, runId, message);
+        throw new CellExecutionError({
+          runId,
+          variantId: cell.variant.id,
+          scenarioId: cell.scenario.scenarioId,
+          seed: cell.seed,
+          reason: "runner_threw",
+          error: message
+        });
+      }
+      const wallMs = now() - wallStart;
+      const integrityReport = await assertRunCaptured(store, runId, { ...integrity, rawSink });
+      if (!integrityReport.ok) {
+        switch (onIntegrityFailure) {
+          case "throw":
+            throw new RunIntegrityError(integrityReport);
+          case "mark_failed":
+            throw new CellExecutionError(
+              {
+                runId,
+                variantId: cell.variant.id,
+                scenarioId: cell.scenario.scenarioId,
+                seed: cell.seed,
+                reason: "integrity_failed",
+                error: integrityReport.issues.map((i) => i.code).join(", ")
+              },
+              integrityReport
+            );
+          case "log":
+            break;
+        }
       }
-      throw new CellExecutionError({
+      const recordOutcome = {
+        raw: outcome.raw ?? {}
+      };
+      if (splitTag === "holdout") recordOutcome.holdoutScore = outcome.score;
+      else recordOutcome.searchScore = outcome.score;
+      if (outcome.judgeScores !== void 0) recordOutcome.judgeScores = outcome.judgeScores;
+      const record = {
+        runId,
+        experimentId: opts.campaignId,
+        candidateId: cell.variant.id,
+        seed: cell.seed,
+        model: outcome.model,
+        promptHash: outcome.promptHash,
+        configHash: outcome.configHash,
+        commitSha: opts.commitSha,
+        wallMs,
+        costUsd: outcome.costUsd,
+        tokenUsage: outcome.tokenUsage,
+        judgeMetadata: outcome.judgeMetadata,
+        outcome: recordOutcome,
+        ...outcome.failureClass ? { failureClass: outcome.failureClass } : {},
+        failureMode: outcome.failureMode,
+        splitTag,
+        scenarioId: cell.scenario.scenarioId
+      };
+      const profileSource = outcome.agentProfile ?? (typeof opts.agentProfile === "function" ? await opts.agentProfile({
+        campaignId: opts.campaignId,
         runId,
         variantId: cell.variant.id,
         scenarioId: cell.scenario.scenarioId,
         seed: cell.seed,
-        reason: "runner_threw",
-        error: message
-      });
-    }
-    const wallMs = now() - wallStart;
-    const integrityReport = await assertRunCaptured(store, runId, { ...integrity, rawSink });
-    if (!integrityReport.ok) {
-      switch (onIntegrityFailure) {
-        case "throw":
-          throw new RunIntegrityError(integrityReport);
-        case "mark_failed":
-          throw new CellExecutionError(
-            {
-              runId,
-              variantId: cell.variant.id,
-              scenarioId: cell.scenario.scenarioId,
-              seed: cell.seed,
-              reason: "integrity_failed",
-              error: integrityReport.issues.map((i) => i.code).join(", ")
-            },
-            integrityReport
-          );
-        case "log":
-          break;
+        variant: cell.variant.payload,
+        scenarioTags: cell.scenario.tags ?? {}
+      }) : opts.agentProfile);
+      if (profileSource !== void 0) {
+        const agentProfile = await resolveAgentProfileCell(profileSource);
+        assertAgentProfileMatchesRun(agentProfile, outcome.model, outcome.promptHash);
+        record.agentProfile = agentProfile;
       }
+      return { record: validateRunRecord(record), integrity: integrityReport };
+    } finally {
+      openRuns.delete(runId);
     }
-    const recordOutcome = {
-      raw: outcome.raw ?? {}
-    };
-    if (splitTag === "holdout") recordOutcome.holdoutScore = outcome.score;
-    else recordOutcome.searchScore = outcome.score;
-    if (outcome.judgeScores !== void 0) recordOutcome.judgeScores = outcome.judgeScores;
-    const record = {
-      runId,
-      experimentId: opts.campaignId,
-      candidateId: cell.variant.id,
-      seed: cell.seed,
-      model: outcome.model,
-      promptHash: outcome.promptHash,
-      configHash: outcome.configHash,
-      commitSha: opts.commitSha,
-      wallMs,
-      costUsd: outcome.costUsd,
-      tokenUsage: outcome.tokenUsage,
-      judgeMetadata: outcome.judgeMetadata,
-      outcome: recordOutcome,
-      ...outcome.failureClass ? { failureClass: outcome.failureClass } : {},
-      failureMode: outcome.failureMode,
-      splitTag,
-      scenarioId: cell.scenario.scenarioId
-    };
-    const profileSource = outcome.agentProfile ?? (typeof opts.agentProfile === "function" ? await opts.agentProfile({
-      campaignId: opts.campaignId,
-      runId,
-      variantId: cell.variant.id,
-      scenarioId: cell.scenario.scenarioId,
-      seed: cell.seed,
-      variant: cell.variant.payload,
-      scenarioTags: cell.scenario.tags ?? {}
-    }) : opts.agentProfile);
-    if (profileSource !== void 0) {
-      const agentProfile = await resolveAgentProfileCell(profileSource);
-      assertAgentProfileMatchesRun(agentProfile, outcome.model, outcome.promptHash);
-      record.agentProfile = agentProfile;
-    }
-    return { record: validateRunRecord(record), integrity: integrityReport };
   }
   const workers = Array.from({ length: Math.min(concurrency, cells.length) }, () => worker());
-  await Promise.all(workers);
+  await Promise.allSettled(workers);
+  for (const [runId, emitter] of openRuns) {
+    await finalizeAbort(emitter, runId, "campaign aborted: genuine error in a sibling run");
+  }
+  openRuns.clear();
+  if (genuineErrors.length > 0) {
+    throw genuineErrors.length === 1 ? genuineErrors[0] : new AggregateError(
+      genuineErrors,
+      `runEvalCampaign: ${genuineErrors.length} runs failed with genuine (non-run-level) errors`
+    );
+  }
   let report;
   if (opts.report) {
     const reportOpts = {
@@ -278,6 +295,12 @@ var CellExecutionError = class extends Error {
     this.integrity = integrity;
   }
 };
+async function finalizeAbort(emitter, runId, reason) {
+  const existing = await emitter.traceStore.getRun(runId);
+  if (existing === void 0) return;
+  if (existing.status !== "running") return;
+  await emitter.abortRun(reason);
+}
 function defaultRawSinkFactory(workDir) {
   return (params) => {
     if (!workDir) {
@@ -329,4 +352,4 @@ function defaultRunId(params) {
 export {
   runEvalCampaign
 };
-//# sourceMappingURL=chunk-GSH6QNNS.js.map
+//# sourceMappingURL=chunk-KW53MSA5.js.map

package/dist/chunk-KW53MSA5.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/eval-campaign.ts"],"sourcesContent":["/**\n * EvalCampaign — opinionated matrix runner that wires the four\n * capture-integrity directives by construction.\n *\n * The canonical benchmark shape — matrix runner → for each\n * (variant, scenario, seed) → start a TraceEmitter → call LLMs → end the\n * run → analyze — has a bug class at the integration boundary: raw\n * events not captured, route silently wrong, integrity not asserted,\n * analyst never run. The directives in `SKILL.md § Capture integrity`\n * are the mitigations.\n *\n * `EvalCampaign` is the structural fix — consumers don't wire the\n * integrity surface themselves; the campaign owns it. Specifically:\n *\n *   - calls `assertLlmRoute` once at preflight before any work runs\n *   - constructs a per-run `TraceStore` and `RawProviderSink` via factories\n *   - constructs the `TraceEmitter` with `onRunComplete: [analyst hook]`\n *   - hands the runner an `LlmClientOptions` pre-wired with the sink and\n *     trace context — the runner can't accidentally call an LLM without\n *     capturing the raw HTTP envelope\n *   - calls `assertRunCaptured` after every `endRun` and routes failures\n *     through a configurable policy (`throw` / `mark_failed` / `log`)\n *   - assembles per-run `RunRecord`s and runs `researchReport` at the end\n *     so the campaign artifact is launch-decision-grade by default\n *   - embeds the campaign fingerprint (a SHA-256 over the canonicalised\n *     run set) and optional `preregistrationHash` in the report\n *\n * The runner contract is intentionally narrow: produce a `CampaignRunOutcome`\n * given a fully-wired `CampaignRunContext`. Everything orchestration-shaped\n * lives in the campaign. This is the inversion-of-control point — consumers\n * stop writing matrix runners and start writing scenario-runners.\n *\n * Out of scope for v1 (tracked in `docs/research-report-methodology.md`):\n *\n *   - Distributed/cluster execution (concurrency is local async)\n *   - Adaptive sampling / sequential interim looks\n *   - Resume from partial state across crashes\n *   - LLM-call retry beyond what `LlmClient` already does\n */\n\nimport {\n  type AgentProfileCell,\n  type AgentProfileCellInput,\n  buildAgentProfileCell,\n  verifyAgentProfileCell,\n} from './agent-profile-cell'\nimport { assertLlmRoute, type LlmClientOptions, type LlmRouteRequirements } from './llm-client'\nimport { canonicalize, hashJson } from './pre-registration'\nimport type {\n  JudgeScoresRecord,\n  RunJudgeMetadata,\n  RunOutcome,\n  RunRecord,\n  RunSplitTag,\n  RunTokenUsage,\n} from './run-record'\nimport { validateRunRecord } from './run-record'\nimport { type ResearchReport, type ResearchReportOptions, researchReport } from './summary-report'\nimport type { RunCompleteHook } from './trace/emitter'\nimport { TraceEmitter } from './trace/emitter'\nimport {\n  assertRunCaptured,\n  RunIntegrityError,\n  type RunIntegrityExpectations,\n  type RunIntegrityReport,\n} from './trace/integrity'\nimport { FileSystemRawProviderSink, type RawProviderSink } from './trace/raw-provider-sink'\nimport type { FailureClass } from './trace/schema'\nimport type { TraceStore } from './trace/store'\n\n// ── Public types ─────────────────────────────────────────────────────────\n\nexport interface CampaignVariant<V> {\n  id: string\n  payload: V\n}\n\nexport interface CampaignScenario {\n  scenarioId: string\n  /** Free-form metadata propagated to runs and reports. */\n  tags?: Record<string, string>\n}\n\nexport interface CampaignRunContext<V> {\n  /** Stable run id. The campaign generates this; the runner does not. */\n  runId: string\n  /** Logical experiment id (campaignId by default; overridable per-run via opts). */\n  experimentId: string\n  variant: V\n  variantId: string\n  scenarioId: string\n  scenarioTags: Record<string, string>\n  seed: number\n  splitTag: RunSplitTag\n  /**\n   * The TraceEmitter for this run, with `onRunComplete` hooks pre-wired\n   * (analyst auto-execution if configured, plus integrity check). The\n   * runner MUST call `emitter.startRun` before doing any work and either\n   * `emitter.endRun` or `emitter.abortRun` before returning.\n   */\n  emitter: TraceEmitter\n  store: TraceStore\n  rawSink: RawProviderSink\n  /**\n   * Pre-wired LLM client options — `rawSink` and `traceContext` are populated\n   * so any `callLlm(req, ctx.llmOpts)` automatically captures raw HTTP. The\n   * runner can spread additional fields if needed.\n   */\n  llmOpts: LlmClientOptions\n}\n\nexport interface CampaignRunOutcome {\n  /** Did the run pass? Mirrors `RunOutcome.pass` semantics. */\n  pass: boolean\n  /** Score for the run on its split. Maps to `searchScore` or `holdoutScore`. */\n  score: number\n  /** Mandatory cost in USD. Use 0 + raw.cost_unknown=1 only if truly unknown. */\n  costUsd: number\n  tokenUsage: RunTokenUsage\n  /** Snapshot model id (e.g. `claude-sonnet-4-6@2025-04-15`). */\n  model: string\n  /** sha256 of the effective prompt sent to the model. */\n  promptHash: string\n  /** sha256 of the effective config (model, temperature, tools, judges, splits). */\n  configHash: string\n  /** Optional extra numeric metrics to land in `outcome.raw`. */\n  raw?: Record<string, number>\n  /** Canonical cross-agent failure class from the shared `FAILURE_CLASSES`\n   *  taxonomy. Propagated to `RunRecord.failureClass` so campaign runs\n   *  aggregate failures in the same vocabulary as every other producer. */\n  failureClass?: FailureClass\n  /** Optional free-form failure detail, scoped under `failureClass`. */\n  failureMode?: string\n  /** Optional judge metadata when a judge was used. */\n  judgeMetadata?: RunJudgeMetadata\n  /**\n   * Optional per-judge / per-dim breakdown for ensemble-judged runs.\n   * Propagated to `outcome.judgeScores` on the resulting `RunRecord`.\n   * Single-judge or scalar-only runs leave this unset.\n   */\n  judgeScores?: JudgeScoresRecord\n  /**\n   * Agent profile cell observed by the runner. When supplied, it overrides\n   * `EvalCampaignOptions.agentProfile` for this run and must match the\n   * outcome's `model` and `promptHash`.\n   */\n  agentProfile?: AgentProfileCell | AgentProfileCellInput\n}\n\nexport type CampaignRunner<V> = (ctx: CampaignRunContext<V>) => Promise<CampaignRunOutcome>\n\nexport type CampaignIntegrityPolicy = 'throw' | 'mark_failed' | 'log'\n\nexport interface EvalCampaignOptions<V> {\n  /**\n   * Stable id for the campaign. Used as the default `experimentId` on\n   * every run, and folded into the campaign fingerprint.\n   */\n  campaignId: string\n  variants: CampaignVariant<V>[]\n  scenarios: CampaignScenario[]\n  /** Default `[0, 1, 2]`. */\n  seeds?: number[]\n  /** Default `'holdout'` — the split that anchors a launch decision. */\n  splitTag?: RunSplitTag\n  /** Git SHA the campaign is run against. Mandatory; `RunRecord` rejects unset. */\n  commitSha: string\n  /**\n   * LLM client config. Augmented per-run with `rawSink` and `traceContext`\n   * before being passed to the runner. The campaign asserts this config\n   * matches `routeRequirements` once at preflight.\n   */\n  llmOpts: LlmClientOptions\n  /**\n   * Default `{ requireExplicitBaseUrl: true, requireAuth: true }` — fail\n   * loud if the campaign would silently fall back to the public router or\n   * run unauthenticated. Override with an empty object to disable.\n   */\n  routeRequirements?: LlmRouteRequirements\n  /**\n   * Per-run TraceStore factory. Common shape: a fresh store per run keyed\n   * on `runId`. Implementations that share a store across the campaign\n   * are valid — the campaign only writes through `emitter`.\n   */\n  storeFactory: (params: CampaignFactoryParams) => TraceStore\n  /**\n   * Per-run RawProviderSink factory. Defaults to `FileSystemRawProviderSink`\n   * rooted at `${workDir}/raw-events/${runId}` if `workDir` is supplied;\n   * otherwise required. Forensic capture is non-negotiable in a campaign\n   * run — pass `NoopRawProviderSink` explicitly if you want to opt out.\n   */\n  rawSinkFactory?: (params: CampaignFactoryParams) => RawProviderSink\n  /**\n   * Filesystem root for default `rawSinkFactory`. Ignored if\n   * `rawSinkFactory` is supplied.\n   */\n  workDir?: string\n  /**\n   * Extra `onRunComplete` hooks the campaign appends (after its own\n   * integrity-check hook). Pass `traceAnalystOnRunComplete(...)` here.\n   */\n  onRunComplete?: RunCompleteHook[]\n  /**\n   * Per-run integrity expectations. Defaults to:\n   *   `{ llmSpansMin: 1, requireRawCoverageOfLlmSpans: true, requireOutcome: true }`.\n   * Override (e.g. `{ llmSpansMin: 0 }`) for runs that don't call LLMs.\n   */\n  integrity?: RunIntegrityExpectations\n  /** Behaviour when integrity fails. Default `'mark_failed'`. */\n  onIntegrityFailure?: CampaignIntegrityPolicy\n  /**\n   * Per-run runner. Receives a fully-wired context; produces an outcome\n   * the campaign converts into a `RunRecord`.\n   */\n  runner: CampaignRunner<V>\n  /**\n   * If set, the campaign computes `researchReport` at the end. `comparator`\n   * is a `variantId`. Other fields are forwarded verbatim.\n   */\n  report?: { comparator?: string } & Omit<\n    ResearchReportOptions,\n    'comparator' | 'preregistrationHash' | 'generatedAt'\n  >\n  /**\n   * Hash of a signed `HypothesisManifest` (see `pre-registration.ts`).\n   * Embedded in the campaign fingerprint and the research report.\n   */\n  preregistrationHash?: string\n  /** Local concurrency. Default `1` (sequential). */\n  concurrency?: number\n  /**\n   * Override the time source. Tests pass a mock to make wallMs deterministic.\n   */\n  now?: () => number\n  /** Override the runId generator. Tests pin this. */\n  runId?: (params: CampaignFactoryParams) => string\n  /**\n   * Agent profile cell for campaign runs. Static profiles can pass an object;\n   * routers or variant-specific harnesses can pass a factory. The campaign\n   * stamps the built cell onto every `RunRecord` and rejects profile/model or\n   * profile/prompt contradictions.\n   */\n  agentProfile?:\n    | AgentProfileCell\n    | AgentProfileCellInput\n    | ((\n        params: CampaignFactoryParams & {\n          variant: V\n          scenarioTags: Record<string, string>\n        },\n      ) =>\n        | AgentProfileCell\n        | AgentProfileCellInput\n        | Promise<AgentProfileCell | AgentProfileCellInput>)\n}\n\nexport interface CampaignFactoryParams {\n  campaignId: string\n  runId: string\n  variantId: string\n  scenarioId: string\n  seed: number\n}\n\nexport interface FailedRun {\n  runId: string\n  variantId: string\n  scenarioId: string\n  seed: number\n  reason: string\n  error?: string\n}\n\nexport interface EvalCampaignResult {\n  campaignId: string\n  /** SHA-256 over canonicalised `(variantIds, scenarioIds, seeds, comparator, splitTag, baseUrl, provider, preregistrationHash)`. */\n  campaignFingerprint: string\n  preregistrationHash: string | null\n  /** Successful runs only. Failed runs land in `failedRuns`. */\n  runs: RunRecord[]\n  /** Integrity reports for every successful run. */\n  integrityReports: RunIntegrityReport[]\n  failedRuns: FailedRun[]\n  /** Computed when `report` is set on options. */\n  report?: ResearchReport\n  startedAt: string\n  endedAt: string\n}\n\n// ── Implementation ───────────────────────────────────────────────────────\n\nconst DEFAULT_INTEGRITY: RunIntegrityExpectations = {\n  llmSpansMin: 1,\n  requireRawCoverageOfLlmSpans: true,\n  requireOutcome: true,\n}\n\nconst DEFAULT_ROUTE: LlmRouteRequirements = {\n  requireExplicitBaseUrl: true,\n  requireAuth: true,\n}\n\nexport async function runEvalCampaign<V>(\n  opts: EvalCampaignOptions<V>,\n): Promise<EvalCampaignResult> {\n  // ── Preflight ──────────────────────────────────────────────────────\n  assertLlmRoute(opts.llmOpts, opts.routeRequirements ?? DEFAULT_ROUTE)\n\n  if (opts.variants.length === 0) {\n    throw new Error('runEvalCampaign: variants must be non-empty.')\n  }\n  if (opts.scenarios.length === 0) {\n    throw new Error('runEvalCampaign: scenarios must be non-empty.')\n  }\n  const variantIds = new Set<string>()\n  for (const v of opts.variants) {\n    if (variantIds.has(v.id)) {\n      throw new Error(`runEvalCampaign: duplicate variant id \"${v.id}\".`)\n    }\n    variantIds.add(v.id)\n  }\n  const scenarioIds = new Set<string>()\n  for (const s of opts.scenarios) {\n    if (scenarioIds.has(s.scenarioId)) {\n      throw new Error(`runEvalCampaign: duplicate scenarioId \"${s.scenarioId}\".`)\n    }\n    scenarioIds.add(s.scenarioId)\n  }\n  if (opts.report?.comparator && !variantIds.has(opts.report.comparator)) {\n    throw new Error(\n      `runEvalCampaign: report.comparator \"${opts.report.comparator}\" is not a configured variantId.`,\n    )\n  }\n  if (!opts.commitSha) {\n    throw new Error('runEvalCampaign: commitSha is required (every RunRecord needs it).')\n  }\n\n  const seeds = opts.seeds ?? [0, 1, 2]\n  const splitTag: RunSplitTag = opts.splitTag ?? 'holdout'\n  const concurrency = Math.max(1, opts.concurrency ?? 1)\n  const integrity = { ...DEFAULT_INTEGRITY, ...(opts.integrity ?? {}) }\n  const onIntegrityFailure: CampaignIntegrityPolicy = opts.onIntegrityFailure ?? 'mark_failed'\n  const now = opts.now ?? (() => Date.now())\n  const baseUrl = (opts.llmOpts.baseUrl ?? '').replace(/\\/+$/, '')\n  const provider = opts.llmOpts.provider ?? null\n  const preregistrationHash = opts.preregistrationHash ?? null\n\n  const rawSinkFactory = opts.rawSinkFactory ?? defaultRawSinkFactory(opts.workDir)\n\n  // ── Fingerprint ────────────────────────────────────────────────────\n  const campaignFingerprint = await hashJson(\n    canonicalize({\n      campaignId: opts.campaignId,\n      variants: opts.variants.map((v) => v.id).sort(),\n      scenarios: opts.scenarios.map((s) => s.scenarioId).sort(),\n      seeds: [...seeds].sort((a, b) => a - b),\n      splitTag,\n      comparator: opts.report?.comparator ?? null,\n      baseUrl,\n      provider,\n      preregistrationHash,\n    }),\n  )\n\n  // ── Plan the matrix ────────────────────────────────────────────────\n  type Cell = { variant: CampaignVariant<V>; scenario: CampaignScenario; seed: number }\n  const cells: Cell[] = []\n  for (const variant of opts.variants) {\n    for (const scenario of opts.scenarios) {\n      for (const seed of seeds) {\n        cells.push({ variant, scenario, seed })\n      }\n    }\n  }\n\n  const startedAt = new Date(now()).toISOString()\n  const runs: RunRecord[] = []\n  const integrityReports: RunIntegrityReport[] = []\n  const failedRuns: FailedRun[] = []\n\n  // ── Execute (bounded-concurrency worker pool) ──────────────────────\n  // A genuine (non-CellExecutionError) error from any worker is a bug, not a\n  // run-level failure. We must NOT let `Promise.all` reject mid-flight and\n  // orphan the other workers' in-progress runs (emitters never finalized,\n  // sinks/handles leak, partial work silently discarded). Instead: capture the\n  // first genuine error, stop dispatching new cells so in-flight workers wind\n  // down, finalize every still-open run, then re-throw the aggregated error.\n  let cursor = 0\n  let aborting = false\n  const genuineErrors: unknown[] = []\n  // Emitters for runs that are currently mid-flight, keyed by runId. A worker\n  // registers its emitter before invoking the runner and removes it once the\n  // run is finalized (via endRun/abortRun, success or failure). Anything left\n  // here after the pool settles is an orphan we must abort.\n  const openRuns = new Map<string, TraceEmitter>()\n\n  async function worker(): Promise<void> {\n    while (!aborting) {\n      const i = cursor++\n      if (i >= cells.length) return\n      const cell = cells[i]!\n      try {\n        const result = await runOneCell(cell)\n        runs.push(result.record)\n        integrityReports.push(result.integrity)\n      } catch (err) {\n        if (err instanceof CellExecutionError) {\n          failedRuns.push(err.failed)\n          if (err.integrity) integrityReports.push(err.integrity)\n        } else {\n          // Genuine bug — not a runner failure, not an integrity failure.\n          // Capture it and stop dispatching so peers wind down gracefully;\n          // re-thrown after the pool settles. Do not surface here — that would\n          // reject Promise.all and orphan the other workers' runs.\n          genuineErrors.push(err)\n          aborting = true\n          return\n        }\n      }\n    }\n  }\n\n  async function runOneCell(\n    cell: Cell,\n  ): Promise<{ record: RunRecord; integrity: RunIntegrityReport }> {\n    const runId = (opts.runId ?? defaultRunId)({\n      campaignId: opts.campaignId,\n      runId: '', // unused by default generator\n      variantId: cell.variant.id,\n      scenarioId: cell.scenario.scenarioId,\n      seed: cell.seed,\n    })\n    const factoryParams: CampaignFactoryParams = {\n      campaignId: opts.campaignId,\n      runId,\n      variantId: cell.variant.id,\n      scenarioId: cell.scenario.scenarioId,\n      seed: cell.seed,\n    }\n    const store = opts.storeFactory(factoryParams)\n    const rawSink = rawSinkFactory(factoryParams)\n\n    const emitter = new TraceEmitter(store, {\n      runId,\n      now: opts.now,\n      onRunComplete: opts.onRunComplete,\n    })\n    // Track this run as open so a genuine error elsewhere in the pool can\n    // finalize it instead of orphaning it. Removed in the finally below.\n    openRuns.set(runId, emitter)\n\n    const llmOpts: LlmClientOptions = {\n      ...opts.llmOpts,\n      rawSink,\n      traceContext: { runId },\n    }\n\n    const ctx: CampaignRunContext<V> = {\n      runId,\n      experimentId: opts.campaignId,\n      variant: cell.variant.payload,\n      variantId: cell.variant.id,\n      scenarioId: cell.scenario.scenarioId,\n      scenarioTags: cell.scenario.tags ?? {},\n      seed: cell.seed,\n      splitTag,\n      emitter,\n      store,\n      rawSink,\n      llmOpts,\n    }\n\n    try {\n      const wallStart = now()\n      let outcome: CampaignRunOutcome\n      try {\n        outcome = await opts.runner(ctx)\n      } catch (err) {\n        const message = err instanceof Error ? err.message : String(err)\n        // The runner threw mid-execution. Abort the run so the emitter\n        // finalizes. The only benign abortRun failure is \"the runner never\n        // started the run\" (nothing to finalize). A store-write failure (disk\n        // full, FS error) is a genuine diagnostic — surface it rather than\n        // masking it as a plain runner failure.\n        await finalizeAbort(emitter, runId, message)\n        throw new CellExecutionError({\n          runId,\n          variantId: cell.variant.id,\n          scenarioId: cell.scenario.scenarioId,\n          seed: cell.seed,\n          reason: 'runner_threw',\n          error: message,\n        })\n      }\n      const wallMs = now() - wallStart\n\n      const integrityReport = await assertRunCaptured(store, runId, { ...integrity, rawSink })\n      if (!integrityReport.ok) {\n        switch (onIntegrityFailure) {\n          case 'throw':\n            throw new RunIntegrityError(integrityReport)\n          case 'mark_failed':\n            throw new CellExecutionError(\n              {\n                runId,\n                variantId: cell.variant.id,\n                scenarioId: cell.scenario.scenarioId,\n                seed: cell.seed,\n                reason: 'integrity_failed',\n                error: integrityReport.issues.map((i) => i.code).join(', '),\n              },\n              integrityReport,\n            )\n          case 'log':\n            // Caller wants the run admitted with a flagged report; fall through.\n            break\n        }\n      }\n\n      const recordOutcome: RunOutcome = {\n        raw: outcome.raw ?? {},\n      }\n      if (splitTag === 'holdout') recordOutcome.holdoutScore = outcome.score\n      else recordOutcome.searchScore = outcome.score\n      if (outcome.judgeScores !== undefined) recordOutcome.judgeScores = outcome.judgeScores\n\n      const record: RunRecord = {\n        runId,\n        experimentId: opts.campaignId,\n        candidateId: cell.variant.id,\n        seed: cell.seed,\n        model: outcome.model,\n        promptHash: outcome.promptHash,\n        configHash: outcome.configHash,\n        commitSha: opts.commitSha,\n        wallMs,\n        costUsd: outcome.costUsd,\n        tokenUsage: outcome.tokenUsage,\n        judgeMetadata: outcome.judgeMetadata,\n        outcome: recordOutcome,\n        ...(outcome.failureClass ? { failureClass: outcome.failureClass } : {}),\n        failureMode: outcome.failureMode,\n        splitTag,\n        scenarioId: cell.scenario.scenarioId,\n      }\n      const profileSource =\n        outcome.agentProfile ??\n        (typeof opts.agentProfile === 'function'\n          ? await opts.agentProfile({\n              campaignId: opts.campaignId,\n              runId,\n              variantId: cell.variant.id,\n              scenarioId: cell.scenario.scenarioId,\n              seed: cell.seed,\n              variant: cell.variant.payload,\n              scenarioTags: cell.scenario.tags ?? {},\n            })\n          : opts.agentProfile)\n      if (profileSource !== undefined) {\n        const agentProfile = await resolveAgentProfileCell(profileSource)\n        assertAgentProfileMatchesRun(agentProfile, outcome.model, outcome.promptHash)\n        record.agentProfile = agentProfile\n      }\n      return { record: validateRunRecord(record), integrity: integrityReport }\n    } finally {\n      // This run's worker has finished with it (success, run-level failure, or\n      // genuine error). It is no longer the pool's job to finalize — drop it so\n      // the post-settle sweep doesn't double-abort a finalized run.\n      openRuns.delete(runId)\n    }\n  }\n\n  const workers = Array.from({ length: Math.min(concurrency, cells.length) }, () => worker())\n  // allSettled (not all): a genuine error in one worker must not reject the\n  // pool mid-flight and orphan the others. Each worker captures its own\n  // genuine error into `genuineErrors` and returns; we re-throw below.\n  await Promise.allSettled(workers)\n\n  // Finalize any run still open after the pool wound down. With the\n  // stop-dispatch flag these are the runs that were mid-flight in peer workers\n  // when the first genuine error fired — abort them so their emitters finalize\n  // (hooks fire, store records a terminal status) instead of leaking.\n  for (const [runId, emitter] of openRuns) {\n    await finalizeAbort(emitter, runId, 'campaign aborted: genuine error in a sibling run')\n  }\n  openRuns.clear()\n\n  if (genuineErrors.length > 0) {\n    throw genuineErrors.length === 1\n      ? genuineErrors[0]\n      : new AggregateError(\n          genuineErrors,\n          `runEvalCampaign: ${genuineErrors.length} runs failed with genuine (non-run-level) errors`,\n        )\n  }\n\n  // ── Optional research report ───────────────────────────────────────\n  let report: ResearchReport | undefined\n  if (opts.report) {\n    const reportOpts: ResearchReportOptions = {\n      ...opts.report,\n      comparator: opts.report.comparator,\n      split: splitTag === 'dev' ? 'search' : splitTag,\n      generatedAt: new Date(now()).toISOString(),\n      preregistrationHash: preregistrationHash ?? undefined,\n    }\n    report = await researchReport(runs, reportOpts)\n  }\n\n  const endedAt = new Date(now()).toISOString()\n\n  return {\n    campaignId: opts.campaignId,\n    campaignFingerprint,\n    preregistrationHash,\n    runs,\n    integrityReports,\n    failedRuns,\n    report,\n    startedAt,\n    endedAt,\n  }\n}\n\n// ── Internal ─────────────────────────────────────────────────────────────\n\nclass CellExecutionError extends Error {\n  readonly failed: FailedRun\n  readonly integrity?: RunIntegrityReport\n  constructor(failed: FailedRun, integrity?: RunIntegrityReport) {\n    super(`cell ${failed.variantId}/${failed.scenarioId}@${failed.seed} failed: ${failed.reason}`)\n    this.failed = failed\n    this.integrity = integrity\n  }\n}\n\n/**\n * Abort a run whose owning work threw or was orphaned by a sibling's genuine\n * error, finalizing the emitter so hooks fire and the store records a terminal\n * status. Safe to call unconditionally: it only aborts runs that are still\n * `running`. Two no-op cases are intentional and benign:\n *\n *   - the run was never started (absent from the store) — nothing to finalize\n *   - the run is already terminal (`completed` / `failed` / `aborted`) —\n *     finalizing again would overwrite the real outcome (e.g. flip a passed run\n *     to `aborted` with `{ pass: false, notes: reason }`), so we leave it alone\n *\n * Any OTHER failure of the store read or the abort write (disk full, FS fault,\n * backend down) is a genuine diagnostic and propagates rather than being\n * swallowed.\n */\nexport async function finalizeAbort(\n  emitter: TraceEmitter,\n  runId: string,\n  reason: string,\n): Promise<void> {\n  const existing = await emitter.traceStore.getRun(runId)\n  if (existing === undefined) return // run never started; nothing to abort\n  if (existing.status !== 'running') return // already finalized; never overwrite a real outcome\n  await emitter.abortRun(reason)\n}\n\nfunction defaultRawSinkFactory(workDir: string | undefined) {\n  return (params: CampaignFactoryParams): RawProviderSink => {\n    if (!workDir) {\n      throw new Error(\n        'runEvalCampaign: rawSinkFactory not supplied and workDir not set. Pass either to enable raw provider capture, or pass `new NoopRawProviderSink()` via rawSinkFactory to opt out explicitly.',\n      )\n    }\n    return new FileSystemRawProviderSink({\n      dir: `${workDir}/raw-events/${params.runId}`,\n    })\n  }\n}\n\nasync function resolveAgentProfileCell(\n  input: AgentProfileCell | AgentProfileCellInput,\n): Promise<AgentProfileCell> {\n  if (isAgentProfileCell(input)) {\n    if (!(await verifyAgentProfileCell(input))) {\n      throw new Error(`runEvalCampaign: agentProfile.cellId does not match its content`)\n    }\n    return input\n  }\n  return buildAgentProfileCell(input)\n}\n\nfunction isAgentProfileCell(\n  input: AgentProfileCell | AgentProfileCellInput,\n): input is AgentProfileCell {\n  return 'schemaVersion' in input && 'cellId' in input\n}\n\nfunction assertAgentProfileMatchesRun(\n  profile: AgentProfileCell,\n  model: string,\n  promptHash: string,\n): void {\n  if (profile.model !== undefined && profile.model !== model) {\n    throw new Error(\n      `runEvalCampaign: agentProfile.model \"${profile.model}\" does not match outcome.model \"${model}\"`,\n    )\n  }\n  if (profile.promptHash !== undefined && profile.promptHash !== promptHash) {\n    throw new Error(\n      `runEvalCampaign: agentProfile.promptHash \"${profile.promptHash}\" does not match outcome.promptHash \"${promptHash}\"`,\n    )\n  }\n}\n\nfunction defaultRunId(params: CampaignFactoryParams): string {\n  // Stable across re-runs: fingerprint of (campaignId, variantId, scenarioId, seed).\n  // Caller can override via opts.runId for non-deterministic IDs.\n  const base = `${params.campaignId}::${params.variantId}::${params.scenarioId}::${params.seed}`\n  // Lightweight hex: we don't need crypto-grade here, just stability + uniqueness.\n  let h1 = 0x811c9dc5\n  let h2 = 0x12345678\n  for (let i = 0; i < base.length; i++) {\n    const c = base.charCodeAt(i)\n    h1 = Math.imul(h1 ^ c, 0x01000193) >>> 0\n    h2 = Math.imul(h2 ^ c, 0x9e3779b1) >>> 0\n  }\n  return `run-${h1.toString(16).padStart(8, '0')}${h2.toString(16).padStart(8, '0')}`\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;AAmSA,IAAM,oBAA8C;AAAA,EAClD,aAAa;AAAA,EACb,8BAA8B;AAAA,EAC9B,gBAAgB;AAClB;AAEA,IAAM,gBAAsC;AAAA,EAC1C,wBAAwB;AAAA,EACxB,aAAa;AACf;AAEA,eAAsB,gBACpB,MAC6B;AAE7B,iBAAe,KAAK,SAAS,KAAK,qBAAqB,aAAa;AAEpE,MAAI,KAAK,SAAS,WAAW,GAAG;AAC9B,UAAM,IAAI,MAAM,8CAA8C;AAAA,EAChE;AACA,MAAI,KAAK,UAAU,WAAW,GAAG;AAC/B,UAAM,IAAI,MAAM,+CAA+C;AAAA,EACjE;AACA,QAAM,aAAa,oBAAI,IAAY;AACnC,aAAW,KAAK,KAAK,UAAU;AAC7B,QAAI,WAAW,IAAI,EAAE,EAAE,GAAG;AACxB,YAAM,IAAI,MAAM,0CAA0C,EAAE,EAAE,IAAI;AAAA,IACpE;AACA,eAAW,IAAI,EAAE,EAAE;AAAA,EACrB;AACA,QAAM,cAAc,oBAAI,IAAY;AACpC,aAAW,KAAK,KAAK,WAAW;AAC9B,QAAI,YAAY,IAAI,EAAE,UAAU,GAAG;AACjC,YAAM,IAAI,MAAM,0CAA0C,EAAE,UAAU,IAAI;AAAA,IAC5E;AACA,gBAAY,IAAI,EAAE,UAAU;AAAA,EAC9B;AACA,MAAI,KAAK,QAAQ,cAAc,CAAC,WAAW,IAAI,KAAK,OAAO,UAAU,GAAG;AACtE,UAAM,IAAI;AAAA,MACR,uCAAuC,KAAK,OAAO,UAAU;AAAA,IAC/D;AAAA,EACF;AACA,MAAI,CAAC,KAAK,WAAW;AACnB,UAAM,IAAI,MAAM,oEAAoE;AAAA,EACtF;AAEA,QAAM,QAAQ,KAAK,SAAS,CAAC,GAAG,GAAG,CAAC;AACpC,QAAM,WAAwB,KAAK,YAAY;AAC/C,QAAM,cAAc,KAAK,IAAI,GAAG,KAAK,eAAe,CAAC;AACrD,QAAM,YAAY,EAAE,GAAG,mBAAmB,GAAI,KAAK,aAAa,CAAC,EAAG;AACpE,QAAM,qBAA8C,KAAK,sBAAsB;AAC/E,QAAM,MAAM,KAAK,QAAQ,MAAM,KAAK,IAAI;AACxC,QAAM,WAAW,KAAK,QAAQ,WAAW,IAAI,QAAQ,QAAQ,EAAE;AAC/D,QAAM,WAAW,KAAK,QAAQ,YAAY;AAC1C,QAAM,sBAAsB,KAAK,uBAAuB;AAExD,QAAM,iBAAiB,KAAK,kBAAkB,sBAAsB,KAAK,OAAO;AAGhF,QAAM,sBAAsB,MAAM;AAAA,IAChC,aAAa;AAAA,MACX,YAAY,KAAK;AAAA,MACjB,UAAU,KAAK,SAAS,IAAI,CAAC,MAAM,EAAE,EAAE,EAAE,KAAK;AAAA,MAC9C,WAAW,KAAK,UAAU,IAAI,CAAC,MAAM,EAAE,UAAU,EAAE,KAAK;AAAA,MACxD,OAAO,CAAC,GAAG,KAAK,EAAE,KAAK,CAAC,GAAG,MAAM,IAAI,CAAC;AAAA,MACtC;AAAA,MACA,YAAY,KAAK,QAAQ,cAAc;AAAA,MACvC;AAAA,MACA;AAAA,MACA;AAAA,IACF,CAAC;AAAA,EACH;AAIA,QAAM,QAAgB,CAAC;AACvB,aAAW,WAAW,KAAK,UAAU;AACnC,eAAW,YAAY,KAAK,WAAW;AACrC,iBAAW,QAAQ,OAAO;AACxB,cAAM,KAAK,EAAE,SAAS,UAAU,KAAK,CAAC;AAAA,MACxC;AAAA,IACF;AAAA,EACF;AAEA,QAAM,YAAY,IAAI,KAAK,IAAI,CAAC,EAAE,YAAY;AAC9C,QAAM,OAAoB,CAAC;AAC3B,QAAM,mBAAyC,CAAC;AAChD,QAAM,aAA0B,CAAC;AASjC,MAAI,SAAS;AACb,MAAI,WAAW;AACf,QAAM,gBAA2B,CAAC;AAKlC,QAAM,WAAW,oBAAI,IAA0B;AAE/C,iBAAe,SAAwB;AACrC,WAAO,CAAC,UAAU;AAChB,YAAM,IAAI;AACV,UAAI,KAAK,MAAM,OAAQ;AACvB,YAAM,OAAO,MAAM,CAAC;AACpB,UAAI;AACF,cAAM,SAAS,MAAM,WAAW,IAAI;AACpC,aAAK,KAAK,OAAO,MAAM;AACvB,yBAAiB,KAAK,OAAO,SAAS;AAAA,MACxC,SAAS,KAAK;AACZ,YAAI,eAAe,oBAAoB;AACrC,qBAAW,KAAK,IAAI,MAAM;AAC1B,cAAI,IAAI,UAAW,kBAAiB,KAAK,IAAI,SAAS;AAAA,QACxD,OAAO;AAKL,wBAAc,KAAK,GAAG;AACtB,qBAAW;AACX;AAAA,QACF;AAAA,MACF;AAAA,IACF;AAAA,EACF;AAEA,iBAAe,WACb,MAC+D;AAC/D,UAAM,SAAS,KAAK,SAAS,cAAc;AAAA,MACzC,YAAY,KAAK;AAAA,MACjB,OAAO;AAAA;AAAA,MACP,WAAW,KAAK,QAAQ;AAAA,MACxB,YAAY,KAAK,SAAS;AAAA,MAC1B,MAAM,KAAK;AAAA,IACb,CAAC;AACD,UAAM,gBAAuC;AAAA,MAC3C,YAAY,KAAK;AAAA,MACjB;AAAA,MACA,WAAW,KAAK,QAAQ;AAAA,MACxB,YAAY,KAAK,SAAS;AAAA,MAC1B,MAAM,KAAK;AAAA,IACb;AACA,UAAM,QAAQ,KAAK,aAAa,aAAa;AAC7C,UAAM,UAAU,eAAe,aAAa;AAE5C,UAAM,UAAU,IAAI,aAAa,OAAO;AAAA,MACtC;AAAA,MACA,KAAK,KAAK;AAAA,MACV,eAAe,KAAK;AAAA,IACtB,CAAC;AAGD,aAAS,IAAI,OAAO,OAAO;AAE3B,UAAM,UAA4B;AAAA,MAChC,GAAG,KAAK;AAAA,MACR;AAAA,MACA,cAAc,EAAE,MAAM;AAAA,IACxB;AAEA,UAAM,MAA6B;AAAA,MACjC;AAAA,MACA,cAAc,KAAK;AAAA,MACnB,SAAS,KAAK,QAAQ;AAAA,MACtB,WAAW,KAAK,QAAQ;AAAA,MACxB,YAAY,KAAK,SAAS;AAAA,MAC1B,cAAc,KAAK,SAAS,QAAQ,CAAC;AAAA,MACrC,MAAM,KAAK;AAAA,MACX;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,IACF;AAEA,QAAI;AACF,YAAM,YAAY,IAAI;AACtB,UAAI;AACJ,UAAI;AACF,kBAAU,MAAM,KAAK,OAAO,GAAG;AAAA,MACjC,SAAS,KAAK;AACZ,cAAM,UAAU,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;AAM/D,cAAM,cAAc,SAAS,OAAO,OAAO;AAC3C,cAAM,IAAI,mBAAmB;AAAA,UAC3B;AAAA,UACA,WAAW,KAAK,QAAQ;AAAA,UACxB,YAAY,KAAK,SAAS;AAAA,UAC1B,MAAM,KAAK;AAAA,UACX,QAAQ;AAAA,UACR,OAAO;AAAA,QACT,CAAC;AAAA,MACH;AACA,YAAM,SAAS,IAAI,IAAI;AAEvB,YAAM,kBAAkB,MAAM,kBAAkB,OAAO,OAAO,EAAE,GAAG,WAAW,QAAQ,CAAC;AACvF,UAAI,CAAC,gBAAgB,IAAI;AACvB,gBAAQ,oBAAoB;AAAA,UAC1B,KAAK;AACH,kBAAM,IAAI,kBAAkB,eAAe;AAAA,UAC7C,KAAK;AACH,kBAAM,IAAI;AAAA,cACR;AAAA,gBACE;AAAA,gBACA,WAAW,KAAK,QAAQ;AAAA,gBACxB,YAAY,KAAK,SAAS;AAAA,gBAC1B,MAAM,KAAK;AAAA,gBACX,QAAQ;AAAA,gBACR,OAAO,gBAAgB,OAAO,IAAI,CAAC,MAAM,EAAE,IAAI,EAAE,KAAK,IAAI;AAAA,cAC5D;AAAA,cACA;AAAA,YACF;AAAA,UACF,KAAK;AAEH;AAAA,QACJ;AAAA,MACF;AAEA,YAAM,gBAA4B;AAAA,QAChC,KAAK,QAAQ,OAAO,CAAC;AAAA,MACvB;AACA,UAAI,aAAa,UAAW,eAAc,eAAe,QAAQ;AAAA,UAC5D,eAAc,cAAc,QAAQ;AACzC,UAAI,QAAQ,gBAAgB,OAAW,eAAc,cAAc,QAAQ;AAE3E,YAAM,SAAoB;AAAA,QACxB;AAAA,QACA,cAAc,KAAK;AAAA,QACnB,aAAa,KAAK,QAAQ;AAAA,QAC1B,MAAM,KAAK;AAAA,QACX,OAAO,QAAQ;AAAA,QACf,YAAY,QAAQ;AAAA,QACpB,YAAY,QAAQ;AAAA,QACpB,WAAW,KAAK;AAAA,QAChB;AAAA,QACA,SAAS,QAAQ;AAAA,QACjB,YAAY,QAAQ;AAAA,QACpB,eAAe,QAAQ;AAAA,QACvB,SAAS;AAAA,QACT,GAAI,QAAQ,eAAe,EAAE,cAAc,QAAQ,aAAa,IAAI,CAAC;AAAA,QACrE,aAAa,QAAQ;AAAA,QACrB;AAAA,QACA,YAAY,KAAK,SAAS;AAAA,MAC5B;AACA,YAAM,gBACJ,QAAQ,iBACP,OAAO,KAAK,iBAAiB,aAC1B,MAAM,KAAK,aAAa;AAAA,QACtB,YAAY,KAAK;AAAA,QACjB;AAAA,QACA,WAAW,KAAK,QAAQ;AAAA,QACxB,YAAY,KAAK,SAAS;AAAA,QAC1B,MAAM,KAAK;AAAA,QACX,SAAS,KAAK,QAAQ;AAAA,QACtB,cAAc,KAAK,SAAS,QAAQ,CAAC;AAAA,MACvC,CAAC,IACD,KAAK;AACX,UAAI,kBAAkB,QAAW;AAC/B,cAAM,eAAe,MAAM,wBAAwB,aAAa;AAChE,qCAA6B,cAAc,QAAQ,OAAO,QAAQ,UAAU;AAC5E,eAAO,eAAe;AAAA,MACxB;AACA,aAAO,EAAE,QAAQ,kBAAkB,MAAM,GAAG,WAAW,gBAAgB;AAAA,IACzE,UAAE;AAIA,eAAS,OAAO,KAAK;AAAA,IACvB;AAAA,EACF;AAEA,QAAM,UAAU,MAAM,KAAK,EAAE,QAAQ,KAAK,IAAI,aAAa,MAAM,MAAM,EAAE,GAAG,MAAM,OAAO,CAAC;AAI1F,QAAM,QAAQ,WAAW,OAAO;AAMhC,aAAW,CAAC,OAAO,OAAO,KAAK,UAAU;AACvC,UAAM,cAAc,SAAS,OAAO,kDAAkD;AAAA,EACxF;AACA,WAAS,MAAM;AAEf,MAAI,cAAc,SAAS,GAAG;AAC5B,UAAM,cAAc,WAAW,IAC3B,cAAc,CAAC,IACf,IAAI;AAAA,MACF;AAAA,MACA,oBAAoB,cAAc,MAAM;AAAA,IAC1C;AAAA,EACN;AAGA,MAAI;AACJ,MAAI,KAAK,QAAQ;AACf,UAAM,aAAoC;AAAA,MACxC,GAAG,KAAK;AAAA,MACR,YAAY,KAAK,OAAO;AAAA,MACxB,OAAO,aAAa,QAAQ,WAAW;AAAA,MACvC,aAAa,IAAI,KAAK,IAAI,CAAC,EAAE,YAAY;AAAA,MACzC,qBAAqB,uBAAuB;AAAA,IAC9C;AACA,aAAS,MAAM,eAAe,MAAM,UAAU;AAAA,EAChD;AAEA,QAAM,UAAU,IAAI,KAAK,IAAI,CAAC,EAAE,YAAY;AAE5C,SAAO;AAAA,IACL,YAAY,KAAK;AAAA,IACjB;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,EACF;AACF;AAIA,IAAM,qBAAN,cAAiC,MAAM;AAAA,EAC5B;AAAA,EACA;AAAA,EACT,YAAY,QAAmB,WAAgC;AAC7D,UAAM,QAAQ,OAAO,SAAS,IAAI,OAAO,UAAU,IAAI,OAAO,IAAI,YAAY,OAAO,MAAM,EAAE;AAC7F,SAAK,SAAS;AACd,SAAK,YAAY;AAAA,EACnB;AACF;AAiBA,eAAsB,cACpB,SACA,OACA,QACe;AACf,QAAM,WAAW,MAAM,QAAQ,WAAW,OAAO,KAAK;AACtD,MAAI,aAAa,OAAW;AAC5B,MAAI,SAAS,WAAW,UAAW;AACnC,QAAM,QAAQ,SAAS,MAAM;AAC/B;AAEA,SAAS,sBAAsB,SAA6B;AAC1D,SAAO,CAAC,WAAmD;AACzD,QAAI,CAAC,SAAS;AACZ,YAAM,IAAI;AAAA,QACR;AAAA,MACF;AAAA,IACF;AACA,WAAO,IAAI,0BAA0B;AAAA,MACnC,KAAK,GAAG,OAAO,eAAe,OAAO,KAAK;AAAA,IAC5C,CAAC;AAAA,EACH;AACF;AAEA,eAAe,wBACb,OAC2B;AAC3B,MAAI,mBAAmB,KAAK,GAAG;AAC7B,QAAI,CAAE,MAAM,uBAAuB,KAAK,GAAI;AAC1C,YAAM,IAAI,MAAM,iEAAiE;AAAA,IACnF;AACA,WAAO;AAAA,EACT;AACA,SAAO,sBAAsB,KAAK;AACpC;AAEA,SAAS,mBACP,OAC2B;AAC3B,SAAO,mBAAmB,SAAS,YAAY;AACjD;AAEA,SAAS,6BACP,SACA,OACA,YACM;AACN,MAAI,QAAQ,UAAU,UAAa,QAAQ,UAAU,OAAO;AAC1D,UAAM,IAAI;AAAA,MACR,wCAAwC,QAAQ,KAAK,mCAAmC,KAAK;AAAA,IAC/F;AAAA,EACF;AACA,MAAI,QAAQ,eAAe,UAAa,QAAQ,eAAe,YAAY;AACzE,UAAM,IAAI;AAAA,MACR,6CAA6C,QAAQ,UAAU,wCAAwC,UAAU;AAAA,IACnH;AAAA,EACF;AACF;AAEA,SAAS,aAAa,QAAuC;AAG3D,QAAM,OAAO,GAAG,OAAO,UAAU,KAAK,OAAO,SAAS,KAAK,OAAO,UAAU,KAAK,OAAO,IAAI;AAE5F,MAAI,KAAK;AACT,MAAI,KAAK;AACT,WAAS,IAAI,GAAG,IAAI,KAAK,QAAQ,KAAK;AACpC,UAAM,IAAI,KAAK,WAAW,CAAC;AAC3B,SAAK,KAAK,KAAK,KAAK,GAAG,QAAU,MAAM;AACvC,SAAK,KAAK,KAAK,KAAK,GAAG,UAAU,MAAM;AAAA,EACzC;AACA,SAAO,OAAO,GAAG,SAAS,EAAE,EAAE,SAAS,GAAG,GAAG,CAAC,GAAG,GAAG,SAAS,EAAE,EAAE,SAAS,GAAG,GAAG,CAAC;AACnF;","names":[]}

package/dist/{chunk-UMMZHCPB.js → chunk-MIFZUPEK.js}

@@ -2,7 +2,7 @@ import {
   AnalystRegistry,
   DEFAULT_TRACE_ANALYST_KINDS,
   createTraceAnalystKind
-} from "./chunk-D3V5B42D.js";
+} from "./chunk-Q5LIB7BC.js";
 import {
   makeFinding
 } from "./chunk-45EEMHTC.js";
@@ -57,16 +57,18 @@ function computeTraceMetrics(spans) {
   if (inputTokenTrajectory.length >= 3) {
     const first = inputTokenTrajectory[0];
     const last = inputTokenTrajectory[inputTokenTrajectory.length - 1];
-    const growth = first > 0 ? last / first : 0;
+    const growthFromZero = first === 0 && last > 0;
+    const growth = growthFromZero ? Infinity : first > 0 ? last / first : 0;
     if (last > first && growth >= INPUT_GROWTH_FACTOR) {
+      const growthLabel = growthFromZero ? "0\u2192nonzero (unbounded)" : `${growth.toFixed(1)}x`;
       signals.push({
         code: "monotonic-input-growth",
         severity: "high",
-        detail: `LLM input tokens grew ${growth.toFixed(1)}x (${first}\u2192${last}) across ${inputTokenTrajectory.length} calls \u2014 full history re-sent each step with no compression.`,
+        detail: `LLM input tokens grew ${growthLabel} (${first}\u2192${last}) across ${inputTokenTrajectory.length} calls \u2014 full history re-sent each step with no compression.`,
         evidence: {
           first,
           last,
-          growth_x: Number(growth.toFixed(2)),
+          growth_x: growthFromZero ? "unbounded" : Number(growth.toFixed(2)),
           calls: inputTokenTrajectory.length
         }
       });
@@ -187,4 +189,4 @@ export {
   behavioralAnalyst,
   buildDefaultAnalystRegistry
 };
-//# sourceMappingURL=chunk-UMMZHCPB.js.map
+//# sourceMappingURL=chunk-MIFZUPEK.js.map

package/dist/chunk-MIFZUPEK.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/trace-analyst/behavioral-metrics.ts","../src/analyst/behavioral-analyst.ts","../src/analyst/default-registry.ts"],"sourcesContent":["/**\n * Deterministic behavioral metrics over OTLP spans — pure arithmetic, no LLM.\n *\n * These are the model-independent multiplier: the four trace-quality signals a\n * tolerant analyzer (e.g. HALO) re-derives per run inside the model — token\n * growth, output decay, tool monoculture, missing self-verification — computed\n * here once, in TypeScript, with zero model judgment. A finding that falls out\n * of arithmetic is trivially model-agnostic and cannot hallucinate the trend.\n *\n * General, not trace-specific: the detectors key off token trajectories and\n * tool usage present in any agentic OTLP trace, not any one benchmark.\n */\n\nimport type { TraceAnalystSpan } from './types'\n\nexport type SuboptimalCode =\n  | 'monotonic-input-growth'\n  | 'output-length-decay'\n  | 'single-tool-dependency'\n  | 'no-self-verification'\n\nexport interface SuboptimalSignal {\n  code: SuboptimalCode\n  severity: 'high' | 'medium' | 'low'\n  /** Human-readable claim, with the backing numbers inlined. */\n  detail: string\n  /** The exact figures the detector fired on — auditable, no model in the loop. */\n  evidence: Record<string, number | string | boolean>\n}\n\nexport interface BehavioralMetrics {\n  llmCallCount: number\n  inputTokenTrajectory: number[]\n  outputTokenTrajectory: number[]\n  toolHistogram: Record<string, number>\n  totalToolCalls: number\n  distinctTools: number\n  /** distinct/total tool calls; 1.0 when there are no tool calls. */\n  toolDiversityRatio: number\n  hasSelfVerification: boolean\n  signals: SuboptimalSignal[]\n}\n\n/** ≥ this input-token growth ratio across a run, with no compression, fires. */\nconst INPUT_GROWTH_FACTOR = 3\n/** Tool-usage signals need at least this many calls to be meaningful. */\nconst MIN_TOOL_CALLS = 3\n/** Tool names matching this are self-verification, not state mutation. */\nconst VERIFY_RE = /verif|eval|inspect|check|assert|validat|review|confirm/i\n\nfunction num(v: unknown): number | null {\n  return typeof v === 'number' && Number.isFinite(v) ? v : null\n}\nfunction inputTokensOf(s: TraceAnalystSpan): number | null {\n  return num(s.attributes['llm.input_tokens']) ?? num(s.attributes['llm.usage.input_tokens'])\n}\nfunction outputTokensOf(s: TraceAnalystSpan): number | null {\n  return num(s.attributes['llm.output_tokens']) ?? num(s.attributes['llm.usage.output_tokens'])\n}\nfunction stepOf(s: TraceAnalystSpan): number | null {\n  return num(s.attributes.step)\n}\nfunction toolNameOf(s: TraceAnalystSpan): string | null {\n  if (s.tool_name) return s.tool_name\n  const t = s.attributes['tool.name']\n  return typeof t === 'string' && t.length > 0 ? t : null\n}\n\n/**\n * Reduce a span list to behavioral metrics + fired suboptimality signals.\n * Pure + deterministic: same spans → same output, on any machine, no model.\n */\nexport function computeTraceMetrics(spans: readonly TraceAnalystSpan[]): BehavioralMetrics {\n  // Order by step (when present) then start_time so trajectories reflect run order.\n  const ordered = [...spans].sort((a, b) => {\n    const sa = stepOf(a)\n    const sb = stepOf(b)\n    if (sa !== null && sb !== null && sa !== sb) return sa - sb\n    return a.start_time.localeCompare(b.start_time)\n  })\n\n  const inputTokenTrajectory: number[] = []\n  const outputTokenTrajectory: number[] = []\n  const toolHistogram: Record<string, number> = {}\n  let hasSelfVerification = false\n\n  for (const s of ordered) {\n    const inT = inputTokensOf(s)\n    if (inT !== null) inputTokenTrajectory.push(inT)\n    const outT = outputTokensOf(s)\n    if (outT !== null) outputTokenTrajectory.push(outT)\n    const tool = toolNameOf(s)\n    if (tool) {\n      toolHistogram[tool] = (toolHistogram[tool] ?? 0) + 1\n      if (VERIFY_RE.test(tool)) hasSelfVerification = true\n    }\n  }\n\n  const totalToolCalls = Object.values(toolHistogram).reduce((a, b) => a + b, 0)\n  const distinctTools = Object.keys(toolHistogram).length\n  const toolDiversityRatio = totalToolCalls === 0 ? 1 : distinctTools / totalToolCalls\n\n  const signals: SuboptimalSignal[] = []\n\n  if (inputTokenTrajectory.length >= 3) {\n    const first = inputTokenTrajectory[0]!\n    const last = inputTokenTrajectory[inputTokenTrajectory.length - 1]!\n    // first === 0 with later growth is an unbounded ratio (0→huge context blowup);\n    // treat it as infinite so the signal fires, and report it as such instead of\n    // dividing by zero for the displayed factor.\n    const growthFromZero = first === 0 && last > 0\n    const growth = growthFromZero ? Infinity : first > 0 ? last / first : 0\n    if (last > first && growth >= INPUT_GROWTH_FACTOR) {\n      const growthLabel = growthFromZero ? '0→nonzero (unbounded)' : `${growth.toFixed(1)}x`\n      signals.push({\n        code: 'monotonic-input-growth',\n        severity: 'high',\n        detail: `LLM input tokens grew ${growthLabel} (${first}→${last}) across ${inputTokenTrajectory.length} calls — full history re-sent each step with no compression.`,\n        evidence: {\n          first,\n          last,\n          growth_x: growthFromZero ? 'unbounded' : Number(growth.toFixed(2)),\n          calls: inputTokenTrajectory.length,\n        },\n      })\n    }\n  }\n\n  if (outputTokenTrajectory.length >= 3) {\n    const first = outputTokenTrajectory[0]!\n    const last = outputTokenTrajectory[outputTokenTrajectory.length - 1]!\n    if (last < first) {\n      signals.push({\n        code: 'output-length-decay',\n        severity: 'medium',\n        detail: `LLM output tokens shrank ${first}→${last} over ${outputTokenTrajectory.length} calls — less planning/reasoning per step as context grows.`,\n        evidence: { first, last, calls: outputTokenTrajectory.length },\n      })\n    }\n  }\n\n  if (totalToolCalls >= MIN_TOOL_CALLS && distinctTools === 1) {\n    const only = Object.keys(toolHistogram)[0]!\n    signals.push({\n      code: 'single-tool-dependency',\n      severity: 'medium',\n      detail: `All ${totalToolCalls} tool calls are \\`${only}\\` — no tool diversity and no fallback path.`,\n      evidence: { tool: only, calls: totalToolCalls, distinct_tools: 1 },\n    })\n  }\n\n  if (totalToolCalls >= MIN_TOOL_CALLS && !hasSelfVerification) {\n    signals.push({\n      code: 'no-self-verification',\n      severity: 'medium',\n      detail: `${totalToolCalls} tool calls and none verify/inspect/check state — the agent never validates its own actions.`,\n      evidence: { tool_calls: totalToolCalls, verification_calls: 0 },\n    })\n  }\n\n  return {\n    llmCallCount: inputTokenTrajectory.length,\n    inputTokenTrajectory,\n    outputTokenTrajectory,\n    toolHistogram,\n    totalToolCalls,\n    distinctTools,\n    toolDiversityRatio,\n    hasSelfVerification,\n    signals,\n  }\n}\n","/**\n * `behavioralAnalyst` — a DETERMINISTIC analyst (cost.kind = 'deterministic',\n * never calls the LLM). It produces the efficiency/behavioral findings a\n * tolerant agentic analyzer (HALO) re-derives per run inside the model —\n * context bloat, output decay, tool monoculture, missing self-verification —\n * directly from arithmetic over spans (`computeTraceMetrics`).\n *\n * Why it matters: these findings are model-agnostic BY CONSTRUCTION (no model\n * in the loop), so they cannot return 0 on a weak model the way the Ax-RLM\n * does — and they are strictly more reliable than HALO, which spends tokens\n * re-deriving the same numbers and can hallucinate the trend. The agentic\n * RLM kinds remain for SEMANTIC findings that genuinely need a model; this\n * analyst owns the behavioral class.\n */\n\nimport {\n  type BehavioralMetrics,\n  computeTraceMetrics,\n  type SuboptimalCode,\n} from '../trace-analyst/behavioral-metrics'\nimport type { TraceAnalysisStore } from '../trace-analyst/store'\nimport type { TraceAnalystSpan } from '../trace-analyst/types'\nimport { type Analyst, type AnalystFinding, makeFinding } from './types'\n\nconst RECOMMENDED_ACTION: Record<SuboptimalCode, string> = {\n  'monotonic-input-growth':\n    'Add a context-budget instruction: once prior context exceeds a threshold, summarize earlier steps into a short status line instead of re-sending full history.',\n  'output-length-decay':\n    'Require a minimum planning/reasoning budget per step so late steps do not degrade into terse, error-prone commands.',\n  'single-tool-dependency':\n    'Direct the agent to use the full toolset (verify / inspect / alternate actions), not a single execute call, and to plan a fallback when a call returns an unexpected result.',\n  'no-self-verification':\n    'After every state-mutating action, verify the result (eval / inspect / assert) before proceeding.',\n}\n\nconst ANALYST_ID = 'efficiency-behavioral'\n\n/**\n * Map computed signals → structured AnalystFindings. Pure: no LLM, no clock\n * dependence beyond `produced_at` (overridable for deterministic tests).\n */\nexport function deriveEfficiencyFindings(\n  metrics: BehavioralMetrics,\n  opts: { analystId?: string; producedAt?: string } = {},\n): AnalystFinding[] {\n  const analystId = opts.analystId ?? ANALYST_ID\n  return metrics.signals.map((sig) =>\n    makeFinding({\n      analyst_id: analystId,\n      area: 'efficiency',\n      subject: sig.code, // kebab — passes the cluster grammar; stable key for diffFindings\n      claim: sig.detail,\n      severity: sig.severity,\n      // Deterministic arithmetic over spans, not a model judgment → certain.\n      confidence: 1,\n      evidence_refs: [\n        {\n          kind: 'metric',\n          uri: `metric://efficiency/${sig.code}`,\n          excerpt: JSON.stringify(sig.evidence),\n        },\n      ],\n      recommended_action: RECOMMENDED_ACTION[sig.code],\n      metadata: { deterministic: true, evidence: sig.evidence },\n      ...(opts.producedAt ? { produced_at: opts.producedAt } : {}),\n    }),\n  )\n}\n\n/** The deterministic behavioral/efficiency analyst (no LLM, any-model). */\nexport function behavioralAnalyst(): Analyst<TraceAnalysisStore> {\n  return {\n    id: ANALYST_ID,\n    description:\n      'Deterministic behavioral/efficiency findings over OTLP spans — token-growth, output-decay, tool-monoculture, missing self-verification. Zero LLM; model-agnostic by construction.',\n    inputKind: 'trace-store',\n    cost: { kind: 'deterministic' },\n    version: '1.0.0',\n    async analyze(store) {\n      const overview = await store.getOverview()\n      const spans: TraceAnalystSpan[] = []\n      for (const traceId of overview.sample_trace_ids) {\n        const viewed = await store.viewTrace({ trace_id: traceId })\n        if (viewed.spans) spans.push(...viewed.spans)\n      }\n      return deriveEfficiencyFindings(computeTraceMetrics(spans))\n    },\n  }\n}\n","/**\n * `buildDefaultAnalystRegistry` — the canonical analyst suite, so consumers\n * stop hand-wiring `new AnalystRegistry()` + per-kind `createTraceAnalystKind`.\n *\n * The deterministic `behavioralAnalyst` is ALWAYS registered (it needs no\n * model and is model-agnostic by construction). The agentic RLM kinds are\n * registered only when an `ai` service is supplied — so a caller with no LLM\n * still gets the full behavioral/efficiency diagnosis, and the substrate's\n * \"any model (including no model)\" guarantee holds at the suite level.\n */\n\nimport type { AxAIService } from '@ax-llm/ax'\nimport { behavioralAnalyst } from './behavioral-analyst'\nimport { createTraceAnalystKind, type TraceAnalystKindSpec } from './kind-factory'\nimport { DEFAULT_TRACE_ANALYST_KINDS } from './kinds'\nimport { AnalystRegistry, type AnalystRegistryOptions } from './registry'\n\nexport interface DefaultAnalystRegistryOptions {\n  /** Ax service for the agentic RLM kinds. Omit → only the deterministic analyst. */\n  ai?: AxAIService\n  /** Model for the agentic kinds (falls back to the ai service default). */\n  model?: string\n  /** Which agentic kinds to register when `ai` is present. Default = the shipped suite. */\n  kinds?: readonly TraceAnalystKindSpec[]\n  /** Set false to omit the deterministic behavioral analyst (default: include). */\n  includeBehavioral?: boolean\n  /** Forwarded to the AnalystRegistry constructor (signal, tags, priorFindings). */\n  registry?: AnalystRegistryOptions\n}\n\nexport function buildDefaultAnalystRegistry(\n  opts: DefaultAnalystRegistryOptions = {},\n): AnalystRegistry {\n  const registry = new AnalystRegistry(opts.registry)\n  if (opts.includeBehavioral !== false) {\n    registry.register(behavioralAnalyst())\n  }\n  if (opts.ai) {\n    const kinds = opts.kinds ?? DEFAULT_TRACE_ANALYST_KINDS\n    for (const spec of kinds) {\n      registry.register(createTraceAnalystKind(spec, { ai: opts.ai, model: opts.model }))\n    }\n  }\n  return registry\n}\n"],"mappings":";;;;;;;;;;AA4CA,IAAM,sBAAsB;AAE5B,IAAM,iBAAiB;AAEvB,IAAM,YAAY;AAElB,SAAS,IAAI,GAA2B;AACtC,SAAO,OAAO,MAAM,YAAY,OAAO,SAAS,CAAC,IAAI,IAAI;AAC3D;AACA,SAAS,cAAc,GAAoC;AACzD,SAAO,IAAI,EAAE,WAAW,kBAAkB,CAAC,KAAK,IAAI,EAAE,WAAW,wBAAwB,CAAC;AAC5F;AACA,SAAS,eAAe,GAAoC;AAC1D,SAAO,IAAI,EAAE,WAAW,mBAAmB,CAAC,KAAK,IAAI,EAAE,WAAW,yBAAyB,CAAC;AAC9F;AACA,SAAS,OAAO,GAAoC;AAClD,SAAO,IAAI,EAAE,WAAW,IAAI;AAC9B;AACA,SAAS,WAAW,GAAoC;AACtD,MAAI,EAAE,UAAW,QAAO,EAAE;AAC1B,QAAM,IAAI,EAAE,WAAW,WAAW;AAClC,SAAO,OAAO,MAAM,YAAY,EAAE,SAAS,IAAI,IAAI;AACrD;AAMO,SAAS,oBAAoB,OAAuD;AAEzF,QAAM,UAAU,CAAC,GAAG,KAAK,EAAE,KAAK,CAAC,GAAG,MAAM;AACxC,UAAM,KAAK,OAAO,CAAC;AACnB,UAAM,KAAK,OAAO,CAAC;AACnB,QAAI,OAAO,QAAQ,OAAO,QAAQ,OAAO,GAAI,QAAO,KAAK;AACzD,WAAO,EAAE,WAAW,cAAc,EAAE,UAAU;AAAA,EAChD,CAAC;AAED,QAAM,uBAAiC,CAAC;AACxC,QAAM,wBAAkC,CAAC;AACzC,QAAM,gBAAwC,CAAC;AAC/C,MAAI,sBAAsB;AAE1B,aAAW,KAAK,SAAS;AACvB,UAAM,MAAM,cAAc,CAAC;AAC3B,QAAI,QAAQ,KAAM,sBAAqB,KAAK,GAAG;AAC/C,UAAM,OAAO,eAAe,CAAC;AAC7B,QAAI,SAAS,KAAM,uBAAsB,KAAK,IAAI;AAClD,UAAM,OAAO,WAAW,CAAC;AACzB,QAAI,MAAM;AACR,oBAAc,IAAI,KAAK,cAAc,IAAI,KAAK,KAAK;AACnD,UAAI,UAAU,KAAK,IAAI,EAAG,uBAAsB;AAAA,IAClD;AAAA,EACF;AAEA,QAAM,iBAAiB,OAAO,OAAO,aAAa,EAAE,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC;AAC7E,QAAM,gBAAgB,OAAO,KAAK,aAAa,EAAE;AACjD,QAAM,qBAAqB,mBAAmB,IAAI,IAAI,gBAAgB;AAEtE,QAAM,UAA8B,CAAC;AAErC,MAAI,qBAAqB,UAAU,GAAG;AACpC,UAAM,QAAQ,qBAAqB,CAAC;AACpC,UAAM,OAAO,qBAAqB,qBAAqB,SAAS,CAAC;AAIjE,UAAM,iBAAiB,UAAU,KAAK,OAAO;AAC7C,UAAM,SAAS,iBAAiB,WAAW,QAAQ,IAAI,OAAO,QAAQ;AACtE,QAAI,OAAO,SAAS,UAAU,qBAAqB;AACjD,YAAM,cAAc,iBAAiB,+BAA0B,GAAG,OAAO,QAAQ,CAAC,CAAC;AACnF,cAAQ,KAAK;AAAA,QACX,MAAM;AAAA,QACN,UAAU;AAAA,QACV,QAAQ,yBAAyB,WAAW,KAAK,KAAK,SAAI,IAAI,YAAY,qBAAqB,MAAM;AAAA,QACrG,UAAU;AAAA,UACR;AAAA,UACA;AAAA,UACA,UAAU,iBAAiB,cAAc,OAAO,OAAO,QAAQ,CAAC,CAAC;AAAA,UACjE,OAAO,qBAAqB;AAAA,QAC9B;AAAA,MACF,CAAC;AAAA,IACH;AAAA,EACF;AAEA,MAAI,sBAAsB,UAAU,GAAG;AACrC,UAAM,QAAQ,sBAAsB,CAAC;AACrC,UAAM,OAAO,sBAAsB,sBAAsB,SAAS,CAAC;AACnE,QAAI,OAAO,OAAO;AAChB,cAAQ,KAAK;AAAA,QACX,MAAM;AAAA,QACN,UAAU;AAAA,QACV,QAAQ,4BAA4B,KAAK,SAAI,IAAI,SAAS,sBAAsB,MAAM;AAAA,QACtF,UAAU,EAAE,OAAO,MAAM,OAAO,sBAAsB,OAAO;AAAA,MAC/D,CAAC;AAAA,IACH;AAAA,EACF;AAEA,MAAI,kBAAkB,kBAAkB,kBAAkB,GAAG;AAC3D,UAAM,OAAO,OAAO,KAAK,aAAa,EAAE,CAAC;AACzC,YAAQ,KAAK;AAAA,MACX,MAAM;AAAA,MACN,UAAU;AAAA,MACV,QAAQ,OAAO,cAAc,qBAAqB,IAAI;AAAA,MACtD,UAAU,EAAE,MAAM,MAAM,OAAO,gBAAgB,gBAAgB,EAAE;AAAA,IACnE,CAAC;AAAA,EACH;AAEA,MAAI,kBAAkB,kBAAkB,CAAC,qBAAqB;AAC5D,YAAQ,KAAK;AAAA,MACX,MAAM;AAAA,MACN,UAAU;AAAA,MACV,QAAQ,GAAG,cAAc;AAAA,MACzB,UAAU,EAAE,YAAY,gBAAgB,oBAAoB,EAAE;AAAA,IAChE,CAAC;AAAA,EACH;AAEA,SAAO;AAAA,IACL,cAAc,qBAAqB;AAAA,IACnC;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,EACF;AACF;;;ACnJA,IAAM,qBAAqD;AAAA,EACzD,0BACE;AAAA,EACF,uBACE;AAAA,EACF,0BACE;AAAA,EACF,wBACE;AACJ;AAEA,IAAM,aAAa;AAMZ,SAAS,yBACd,SACA,OAAoD,CAAC,GACnC;AAClB,QAAM,YAAY,KAAK,aAAa;AACpC,SAAO,QAAQ,QAAQ;AAAA,IAAI,CAAC,QAC1B,YAAY;AAAA,MACV,YAAY;AAAA,MACZ,MAAM;AAAA,MACN,SAAS,IAAI;AAAA;AAAA,MACb,OAAO,IAAI;AAAA,MACX,UAAU,IAAI;AAAA;AAAA,MAEd,YAAY;AAAA,MACZ,eAAe;AAAA,QACb;AAAA,UACE,MAAM;AAAA,UACN,KAAK,uBAAuB,IAAI,IAAI;AAAA,UACpC,SAAS,KAAK,UAAU,IAAI,QAAQ;AAAA,QACtC;AAAA,MACF;AAAA,MACA,oBAAoB,mBAAmB,IAAI,IAAI;AAAA,MAC/C,UAAU,EAAE,eAAe,MAAM,UAAU,IAAI,SAAS;AAAA,MACxD,GAAI,KAAK,aAAa,EAAE,aAAa,KAAK,WAAW,IAAI,CAAC;AAAA,IAC5D,CAAC;AAAA,EACH;AACF;AAGO,SAAS,oBAAiD;AAC/D,SAAO;AAAA,IACL,IAAI;AAAA,IACJ,aACE;AAAA,IACF,WAAW;AAAA,IACX,MAAM,EAAE,MAAM,gBAAgB;AAAA,IAC9B,SAAS;AAAA,IACT,MAAM,QAAQ,OAAO;AACnB,YAAM,WAAW,MAAM,MAAM,YAAY;AACzC,YAAM,QAA4B,CAAC;AACnC,iBAAW,WAAW,SAAS,kBAAkB;AAC/C,cAAM,SAAS,MAAM,MAAM,UAAU,EAAE,UAAU,QAAQ,CAAC;AAC1D,YAAI,OAAO,MAAO,OAAM,KAAK,GAAG,OAAO,KAAK;AAAA,MAC9C;AACA,aAAO,yBAAyB,oBAAoB,KAAK,CAAC;AAAA,IAC5D;AAAA,EACF;AACF;;;AC1DO,SAAS,4BACd,OAAsC,CAAC,GACtB;AACjB,QAAM,WAAW,IAAI,gBAAgB,KAAK,QAAQ;AAClD,MAAI,KAAK,sBAAsB,OAAO;AACpC,aAAS,SAAS,kBAAkB,CAAC;AAAA,EACvC;AACA,MAAI,KAAK,IAAI;AACX,UAAM,QAAQ,KAAK,SAAS;AAC5B,eAAW,QAAQ,OAAO;AACxB,eAAS,SAAS,uBAAuB,MAAM,EAAE,IAAI,KAAK,IAAI,OAAO,KAAK,MAAM,CAAC,CAAC;AAAA,IACpF;AAAA,EACF;AACA,SAAO;AACT;","names":[]}

package/dist/{chunk-XY4DDNEG.js → chunk-MPQWFX6Y.js}

@@ -1,7 +1,7 @@
 import {
   LlmClient,
   callLlmJson
-} from "./chunk-CVVHBFGN.js";
+} from "./chunk-CWNP4DV4.js";
 import {
   computeFindingId
 } from "./chunk-45EEMHTC.js";
@@ -890,4 +890,4 @@ export {
   runSemanticConceptJudge,
   createSemanticConceptJudge
 };
-//# sourceMappingURL=chunk-XY4DDNEG.js.map
+//# sourceMappingURL=chunk-MPQWFX6Y.js.map

package/dist/{chunk-L3JOU6XM.js → chunk-NACM5RS7.js}

@@ -1,11 +1,11 @@
 import {
   cohensD
-} from "./chunk-LMZQ2Z4U.js";
+} from "./chunk-56GC6VYO.js";
 import {
   argHash,
   groupBy,
   toolSpans
-} from "./chunk-47X6LRCE.js";
+} from "./chunk-UFSG7ACU.js";
 
 // src/failure-taxonomy.ts
 var DEFAULT_RULES = [
@@ -552,4 +552,4 @@ export {
   iqr,
   welchsTTest
 };
-//# sourceMappingURL=chunk-L3JOU6XM.js.map
+//# sourceMappingURL=chunk-NACM5RS7.js.map

package/dist/{chunk-D3V5B42D.js → chunk-Q5LIB7BC.js}

@@ -1,13 +1,13 @@
 import {
   callLlm
-} from "./chunk-CVVHBFGN.js";
+} from "./chunk-CWNP4DV4.js";
 import {
   makeFinding
 } from "./chunk-45EEMHTC.js";
 import {
   TraceFileMissingError,
   buildTraceAnalystTools
-} from "./chunk-QAY5UIJO.js";
+} from "./chunk-QMUEXQJS.js";
 
 // src/analyst/finding-subject.ts
 import { z } from "zod";
@@ -821,6 +821,9 @@ var AnalystRegistry = class {
     const allFindings = [];
     let totalCost = 0;
     let remainingUsd = budget?.totalUsd;
+    const runnableCount = selected.filter(
+      (a) => this.routeInput(a, inputs).kind !== "missing"
+    ).length;
     for (const analyst of selected) {
       const t0 = Date.now();
       const input = this.routeInput(analyst, inputs);
@@ -842,7 +845,7 @@ var AnalystRegistry = class {
       const perBudget = allocateBudget(budget, {
         analyst,
         remainingUsd,
-        runningCount: selected.length
+        runningCount: runnableCount
       });
       const ctx = {
         runId,
@@ -1012,4 +1015,4 @@ export {
   DEFAULT_TRACE_ANALYST_KINDS,
   AnalystRegistry
 };
-//# sourceMappingURL=chunk-D3V5B42D.js.map
+//# sourceMappingURL=chunk-Q5LIB7BC.js.map