@tangle-network/agent-eval 0.129.0 → 0.130.1

package/dist/chunk-7ZZMD7UK.js

@@ -1,386 +0,0 @@
-import {
-  FAILURE_CLASSES
-} from "./chunk-MA6HLL3S.js";
-import {
-  ValidationError
-} from "./chunk-ONWEPEDO.js";
-import {
-  LLM_CACHED_TOKEN_ATTR_KEYS,
-  LLM_CACHE_WRITE_TOKEN_ATTR_KEYS,
-  LLM_COST_ATTR_KEYS,
-  LLM_INPUT_TOKEN_ATTR_KEYS,
-  LLM_OUTPUT_TOKEN_ATTR_KEYS,
-  LLM_REASONING_TOKEN_ATTR_KEYS,
-  RUN_COST_ATTR_KEYS
-} from "./chunk-K4DBDHLK.js";
-
-// src/trace/error-classification.ts
-function summarizeTraceErrors(signals) {
-  const byId = /* @__PURE__ */ new Map();
-  for (const signal of signals) {
-    if (byId.has(signal.id)) {
-      throw new Error(`summarizeTraceErrors: duplicate span id '${signal.id}'`);
-    }
-    byId.set(signal.id, signal);
-  }
-  const propagated = /* @__PURE__ */ new Set();
-  for (const signal of signals) {
-    if (!signal.error) continue;
-    const visited = /* @__PURE__ */ new Set();
-    let parentId = signal.parentId;
-    while (parentId && !visited.has(parentId)) {
-      visited.add(parentId);
-      const parent = byId.get(parentId);
-      if (!parent) break;
-      if (parent.error) propagated.add(parent.id);
-      parentId = parent.parentId;
-    }
-  }
-  const summary = {
-    total: 0,
-    execution: 0,
-    process: 0,
-    guardrail: 0,
-    evaluation: 0,
-    propagated: 0,
-    unclassified: 0
-  };
-  for (const signal of signals) {
-    if (!signal.error) continue;
-    summary.total += 1;
-    if (signal.role === "GUARDRAIL") {
-      summary.guardrail += 1;
-    } else if (signal.role === "EVALUATOR") {
-      summary.evaluation += 1;
-    } else if (signal.processRoot) {
-      summary.process += 1;
-    } else if (propagated.has(signal.id)) {
-      summary.propagated += 1;
-    } else if (signal.role === "AGENT" || signal.role === "CHAIN" || signal.role === "LLM" || signal.role === "TOOL") {
-      summary.execution += 1;
-    } else {
-      summary.unclassified += 1;
-    }
-  }
-  return summary;
-}
-
-// src/trace/execution-measurements.ts
-var TOKEN_MEASUREMENT_KEY_GROUPS = [
-  LLM_INPUT_TOKEN_ATTR_KEYS,
-  LLM_OUTPUT_TOKEN_ATTR_KEYS,
-  LLM_REASONING_TOKEN_ATTR_KEYS,
-  LLM_CACHED_TOKEN_ATTR_KEYS,
-  LLM_CACHE_WRITE_TOKEN_ATTR_KEYS
-];
-var EXECUTION_MEASUREMENT_KEY_GROUPS = [
-  ...TOKEN_MEASUREMENT_KEY_GROUPS,
-  LLM_COST_ATTR_KEYS
-];
-function summarizeExecutionMeasurements(spans) {
-  const byId = /* @__PURE__ */ new Map();
-  for (const span of spans) {
-    if (byId.has(span.id)) {
-      throw new Error(`summarizeExecutionMeasurements: duplicate span id "${span.id}"`);
-    }
-    byId.set(span.id, span);
-  }
-  const tokenMeasurementKeys = TOKEN_MEASUREMENT_KEY_GROUPS.flat();
-  const candidates = spans.filter(
-    (span) => span.modelCall || !span.aggregate && readNumber(span.attributes, tokenMeasurementKeys) !== void 0
-  );
-  const candidateIds = new Set(candidates.map((span) => span.id));
-  const candidateChildren = /* @__PURE__ */ new Map();
-  for (const candidate of candidates) {
-    const parentId = nearestCandidateParent(candidate, byId, candidateIds);
-    if (!parentId) continue;
-    const children = candidateChildren.get(parentId) ?? [];
-    children.push(candidate);
-    candidateChildren.set(parentId, children);
-  }
-  const aggregateIds = classifyAggregateSpans(candidates, candidateChildren);
-  const untypedRunCostIds = new Set(
-    spans.filter(
-      (span) => !span.modelCall && !span.aggregate && !candidateIds.has(span.id) && readNumber(span.attributes, RUN_COST_ATTR_KEYS) !== void 0
-    ).map((span) => span.id)
-  );
-  const aggregateSourceIds = new Set(
-    spans.filter(
-      (span) => span.aggregate || aggregateIds.has(span.id) || untypedRunCostIds.has(span.id)
-    ).map((span) => span.id)
-  );
-  const calls = candidates.filter((span) => !aggregateIds.has(span.id));
-  const input = reconcileMeasurement(calls, byId, aggregateSourceIds, LLM_INPUT_TOKEN_ATTR_KEYS);
-  const reasoning = reconcileMeasurement(
-    calls,
-    byId,
-    aggregateSourceIds,
-    LLM_REASONING_TOKEN_ATTR_KEYS
-  );
-  const output = reconcileMeasurement(
-    calls,
-    byId,
-    aggregateSourceIds,
-    LLM_OUTPUT_TOKEN_ATTR_KEYS,
-    LLM_REASONING_TOKEN_ATTR_KEYS
-  );
-  const cached = reconcileMeasurement(calls, byId, aggregateSourceIds, LLM_CACHED_TOKEN_ATTR_KEYS);
-  const cacheWrite = reconcileMeasurement(
-    calls,
-    byId,
-    aggregateSourceIds,
-    LLM_CACHE_WRITE_TOKEN_ATTR_KEYS
-  );
-  const aggregate = summarizeAggregateMeasurements(
-    spans,
-    byId,
-    /* @__PURE__ */ new Set([...aggregateIds, ...untypedRunCostIds])
-  );
-  return {
-    tokenUsage: {
-      input: input.value ?? 0,
-      output: Math.max(output.value ?? 0, reasoning.value ?? 0),
-      ...reasoning.value !== void 0 ? { reasoning: reasoning.value } : {},
-      ...cached.value !== void 0 ? { cached: cached.value } : {},
-      ...cacheWrite.value !== void 0 ? { cacheWrite: cacheWrite.value } : {}
-    },
-    modelCallCount: calls.length,
-    callSpanIds: calls.map((span) => span.id),
-    cost: reconcileMeasurement(calls, byId, aggregateSourceIds, LLM_COST_ATTR_KEYS),
-    ...aggregate ? { aggregate } : {}
-  };
-}
-function recordAggregateMeasurements(raw, aggregate) {
-  if (!aggregate) return;
-  raw.aggregate_prompt_tokens = aggregate.tokenUsage.input;
-  raw.aggregate_completion_tokens = aggregate.tokenUsage.output;
-  if (aggregate.tokenUsage.reasoning !== void 0)
-    raw.aggregate_reasoning_tokens = aggregate.tokenUsage.reasoning;
-  if (aggregate.tokenUsage.cached !== void 0)
-    raw.aggregate_cached_tokens = aggregate.tokenUsage.cached;
-  if (aggregate.tokenUsage.cacheWrite !== void 0)
-    raw.aggregate_cache_write_tokens = aggregate.tokenUsage.cacheWrite;
-  if (aggregate.costUsd !== void 0) raw.aggregate_cost_usd = aggregate.costUsd;
-}
-function summarizeAggregateMeasurements(spans, byId, aggregateIds) {
-  const aggregates = spans.filter((span) => span.aggregate || aggregateIds.has(span.id));
-  const input = reconcileTopLevelMeasurement(aggregates, byId, LLM_INPUT_TOKEN_ATTR_KEYS);
-  const output = reconcileTopLevelMeasurement(aggregates, byId, LLM_OUTPUT_TOKEN_ATTR_KEYS);
-  const reasoning = reconcileTopLevelMeasurement(aggregates, byId, LLM_REASONING_TOKEN_ATTR_KEYS);
-  const cached = reconcileTopLevelMeasurement(aggregates, byId, LLM_CACHED_TOKEN_ATTR_KEYS);
-  const cacheWrite = reconcileTopLevelMeasurement(aggregates, byId, LLM_CACHE_WRITE_TOKEN_ATTR_KEYS);
-  const costUsd = reconcileTopLevelMeasurement(aggregates, byId, LLM_COST_ATTR_KEYS);
-  if (input === void 0 && output === void 0 && reasoning === void 0 && cached === void 0 && cacheWrite === void 0 && costUsd === void 0)
-    return void 0;
-  return {
-    tokenUsage: {
-      input: input ?? 0,
-      output: output ?? reasoning ?? 0,
-      ...reasoning !== void 0 ? { reasoning } : {},
-      ...cached !== void 0 ? { cached } : {},
-      ...cacheWrite !== void 0 ? { cacheWrite } : {}
-    },
-    ...costUsd !== void 0 ? { costUsd } : {}
-  };
-}
-function reconcileTopLevelMeasurement(spans, byId, keys) {
-  const selected = /* @__PURE__ */ new Map();
-  for (const span of spans) {
-    const value = readNumber(span.attributes, keys);
-    if (value !== void 0) selected.set(span.id, value);
-  }
-  for (const spanId of [...selected.keys()]) {
-    const span = byId.get(spanId);
-    if (!span) continue;
-    if (ancestorIds(span, byId).some((ancestorId) => selected.has(ancestorId))) {
-      selected.delete(spanId);
-    }
-  }
-  return selected.size > 0 ? [...selected.values()].reduce((total, value) => total + value, 0) : void 0;
-}
-function reconcileMeasurement(calls, byId, aggregateSourceIds, keys, fallbackKeys) {
-  const selected = /* @__PURE__ */ new Map();
-  const callIds = new Set(calls.map((call) => call.id));
-  let reportingCalls = 0;
-  for (const call of calls) {
-    const primary = nearestMeasurement(call, byId, callIds, aggregateSourceIds, keys);
-    const fallback = fallbackKeys ? nearestMeasurement(call, byId, callIds, aggregateSourceIds, fallbackKeys) : void 0;
-    const source = primary && fallback && primary.span.id === fallback.span.id ? { span: primary.span, value: Math.max(primary.value, fallback.value) } : primary ?? fallback;
-    if (!source) continue;
-    reportingCalls += 1;
-    selected.set(source.span.id, source.value);
-  }
-  for (const spanId of [...selected.keys()]) {
-    const span = byId.get(spanId);
-    if (!span) continue;
-    if (ancestorIds(span, byId).some(
-      (ancestorId) => selected.has(ancestorId) && !callIds.has(ancestorId)
-    )) {
-      selected.delete(spanId);
-    }
-  }
-  return {
-    ...selected.size > 0 ? { value: [...selected.values()].reduce((total, value) => total + value, 0) } : {},
-    reportingCalls,
-    complete: calls.length > 0 && reportingCalls === calls.length
-  };
-}
-function nearestMeasurement(call, byId, callIds, aggregateSourceIds, keys) {
-  let current = call;
-  const seen = /* @__PURE__ */ new Set();
-  while (current && !seen.has(current.id)) {
-    seen.add(current.id);
-    const value = readNumber(current.attributes, keys);
-    if (value !== void 0 && (current.id === call.id || !callIds.has(current.id) && aggregateSourceIds.has(current.id))) {
-      return { span: current, value };
-    }
-    current = current.parentId ? byId.get(current.parentId) : void 0;
-  }
-  return void 0;
-}
-function classifyAggregateSpans(candidates, childrenById) {
-  const aggregateIds = /* @__PURE__ */ new Set();
-  const summaries = /* @__PURE__ */ new Map();
-  const visiting = /* @__PURE__ */ new Set();
-  const visit = (span) => {
-    const cached = summaries.get(span.id);
-    if (cached) return cached;
-    if (visiting.has(span.id)) return emptyRetainedCallSummary();
-    visiting.add(span.id);
-    const descendants = emptyRetainedCallSummary();
-    for (const child of childrenById.get(span.id) ?? []) {
-      const childDescendants = visit(child);
-      if (!aggregateIds.has(child.id)) addRetainedCall(descendants, child);
-      addRetainedCallSummary(descendants, childDescendants);
-    }
-    if (descendants.callCount > 0 && (!span.modelCall || hasCompatibleDescendantMeasurements(span, descendants))) {
-      aggregateIds.add(span.id);
-    }
-    visiting.delete(span.id);
-    summaries.set(span.id, descendants);
-    return descendants;
-  };
-  for (const candidate of candidates) visit(candidate);
-  return aggregateIds;
-}
-function emptyRetainedCallSummary() {
-  return {
-    callCount: 0,
-    measurements: EXECUTION_MEASUREMENT_KEY_GROUPS.map(() => ({
-      total: 0,
-      reportingCalls: 0
-    }))
-  };
-}
-function addRetainedCall(summary, span) {
-  summary.callCount += 1;
-  for (let index = 0; index < EXECUTION_MEASUREMENT_KEY_GROUPS.length; index += 1) {
-    const value = readNumber(span.attributes, EXECUTION_MEASUREMENT_KEY_GROUPS[index]);
-    if (value === void 0) continue;
-    const measurement = summary.measurements[index];
-    measurement.total += value;
-    measurement.reportingCalls += 1;
-  }
-}
-function addRetainedCallSummary(target, source) {
-  target.callCount += source.callCount;
-  for (let index = 0; index < target.measurements.length; index += 1) {
-    const measurement = target.measurements[index];
-    const sourceMeasurement = source.measurements[index];
-    measurement.total += sourceMeasurement.total;
-    measurement.reportingCalls += sourceMeasurement.reportingCalls;
-  }
-}
-function hasCompatibleDescendantMeasurements(span, descendants) {
-  let parentMeasurements = 0;
-  let descendantMeasurements = 0;
-  for (let index = 0; index < EXECUTION_MEASUREMENT_KEY_GROUPS.length; index += 1) {
-    const keys = EXECUTION_MEASUREMENT_KEY_GROUPS[index];
-    const parentValue = readNumber(span.attributes, keys);
-    const measurement = descendants.measurements[index];
-    if (parentValue !== void 0) parentMeasurements += 1;
-    if (measurement.reportingCalls > 0) descendantMeasurements += 1;
-    if (parentValue === void 0 || measurement.reportingCalls === 0) continue;
-    if (measurement.reportingCalls !== descendants.callCount) continue;
-    if (Math.abs(parentValue - measurement.total) > 1e-12) return false;
-  }
-  return parentMeasurements === 0 || descendantMeasurements > 0;
-}
-function ancestorIds(span, byId) {
-  const ids = [];
-  const seen = /* @__PURE__ */ new Set();
-  let parentId = span.parentId;
-  while (parentId && !seen.has(parentId)) {
-    ids.push(parentId);
-    seen.add(parentId);
-    parentId = byId.get(parentId)?.parentId;
-  }
-  return ids;
-}
-function nearestCandidateParent(span, byId, candidateIds) {
-  const seen = /* @__PURE__ */ new Set();
-  let parentId = span.parentId;
-  while (parentId && !seen.has(parentId)) {
-    if (candidateIds.has(parentId)) return parentId;
-    seen.add(parentId);
-    parentId = byId.get(parentId)?.parentId;
-  }
-  return void 0;
-}
-function readNumber(attributes, keys) {
-  for (const key of keys) {
-    const value = attributes[key];
-    const parsed = typeof value === "number" ? value : typeof value === "string" && value.length > 0 ? Number(value) : Number.NaN;
-    if (Number.isFinite(parsed) && parsed >= 0) return parsed;
-  }
-  return void 0;
-}
-
-// src/trace/task-failure-attributes.ts
-var TASK_FAILURE_CLASS_ATTR = "tangle.task.failure_class";
-var TASK_FAILURE_MODE_ATTR = "tangle.task.failure_mode";
-function readTaskFailureLabels(roots, context) {
-  const failureClass = readConsistentRootString(roots, TASK_FAILURE_CLASS_ATTR, context);
-  const failureMode = readConsistentRootString(roots, TASK_FAILURE_MODE_ATTR, context);
-  if (failureClass !== void 0 && !FAILURE_CLASSES.includes(failureClass)) {
-    throw new ValidationError(
-      `${context}: ${TASK_FAILURE_CLASS_ATTR} must be one of ${FAILURE_CLASSES.join(", ")}`
-    );
-  }
-  if (failureMode !== void 0 && (failureClass === void 0 || failureClass === "success")) {
-    throw new ValidationError(
-      `${context}: ${TASK_FAILURE_MODE_ATTR} requires a non-success ${TASK_FAILURE_CLASS_ATTR}`
-    );
-  }
-  if (failureClass === void 0) return {};
-  if (failureClass === "success") return { failureClass };
-  return {
-    failureClass,
-    ...failureMode ? { failureMode } : {}
-  };
-}
-function readConsistentRootString(roots, key, context) {
-  const values = /* @__PURE__ */ new Set();
-  for (const root of roots) {
-    if (!Object.hasOwn(root.attributes, key)) continue;
-    const value = root.attributes[key];
-    if (typeof value !== "string" || value.trim().length === 0) {
-      throw new ValidationError(`${context}: ${key} must be a non-empty string`);
-    }
-    values.add(value);
-  }
-  if (values.size > 1) {
-    throw new ValidationError(
-      `${context}: conflicting ${key} values: ${[...values].sort().join(", ")}`
-    );
-  }
-  return values.values().next().value;
-}
-
-export {
-  summarizeTraceErrors,
-  summarizeExecutionMeasurements,
-  recordAggregateMeasurements,
-  readTaskFailureLabels
-};
-//# sourceMappingURL=chunk-7ZZMD7UK.js.map

package/dist/chunk-7ZZMD7UK.js.map

@@ -1 +0,0 @@
-{"version":3,"sources":["../src/trace/error-classification.ts","../src/trace/execution-measurements.ts","../src/trace/task-failure-attributes.ts"],"sourcesContent":["import type { OtlpSpanRole } from './otlp-attributes'\n\nexport type TraceErrorRole = OtlpSpanRole\n\nexport interface TraceErrorSignal {\n  id: string\n  parentId?: string\n  role: TraceErrorRole\n  error: boolean\n  processRoot: boolean\n}\n\nexport interface TraceErrorSummary {\n  total: number\n  execution: number\n  process: number\n  guardrail: number\n  evaluation: number\n  propagated: number\n  unclassified: number\n}\n\n/**\n * Classify errored spans without counting a propagated parent status as a\n * second execution failure.\n */\nexport function summarizeTraceErrors(signals: readonly TraceErrorSignal[]): TraceErrorSummary {\n  const byId = new Map<string, TraceErrorSignal>()\n  for (const signal of signals) {\n    if (byId.has(signal.id)) {\n      throw new Error(`summarizeTraceErrors: duplicate span id '${signal.id}'`)\n    }\n    byId.set(signal.id, signal)\n  }\n\n  const propagated = new Set<string>()\n  for (const signal of signals) {\n    if (!signal.error) continue\n    const visited = new Set<string>()\n    let parentId = signal.parentId\n    while (parentId && !visited.has(parentId)) {\n      visited.add(parentId)\n      const parent = byId.get(parentId)\n      if (!parent) break\n      if (parent.error) propagated.add(parent.id)\n      parentId = parent.parentId\n    }\n  }\n\n  const summary: TraceErrorSummary = {\n    total: 0,\n    execution: 0,\n    process: 0,\n    guardrail: 0,\n    evaluation: 0,\n    propagated: 0,\n    unclassified: 0,\n  }\n\n  for (const signal of signals) {\n    if (!signal.error) continue\n    summary.total += 1\n    if (signal.role === 'GUARDRAIL') {\n      summary.guardrail += 1\n    } else if (signal.role === 'EVALUATOR') {\n      summary.evaluation += 1\n    } else if (signal.processRoot) {\n      summary.process += 1\n    } else if (propagated.has(signal.id)) {\n      summary.propagated += 1\n    } else if (\n      signal.role === 'AGENT' ||\n      signal.role === 'CHAIN' ||\n      signal.role === 'LLM' ||\n      signal.role === 'TOOL'\n    ) {\n      summary.execution += 1\n    } else {\n      summary.unclassified += 1\n    }\n  }\n\n  return summary\n}\n","import type { RunTokenUsage } from '../run-record'\nimport {\n  LLM_CACHE_WRITE_TOKEN_ATTR_KEYS,\n  LLM_CACHED_TOKEN_ATTR_KEYS,\n  LLM_COST_ATTR_KEYS,\n  LLM_INPUT_TOKEN_ATTR_KEYS,\n  LLM_OUTPUT_TOKEN_ATTR_KEYS,\n  LLM_REASONING_TOKEN_ATTR_KEYS,\n  RUN_COST_ATTR_KEYS,\n} from './otlp-attributes'\n\nexport interface ExecutionMeasurementSpan {\n  id: string\n  parentId?: string\n  attributes: Record<string, unknown>\n  modelCall: boolean\n  aggregate: boolean\n}\n\nexport interface MeasurementCoverage {\n  value?: number\n  reportingCalls: number\n  complete: boolean\n}\n\nexport interface ExecutionMeasurements {\n  tokenUsage: RunTokenUsage\n  modelCallCount: number\n  callSpanIds: string[]\n  cost: MeasurementCoverage\n  aggregate?: {\n    tokenUsage: RunTokenUsage\n    costUsd?: number\n  }\n}\n\nconst TOKEN_MEASUREMENT_KEY_GROUPS = [\n  LLM_INPUT_TOKEN_ATTR_KEYS,\n  LLM_OUTPUT_TOKEN_ATTR_KEYS,\n  LLM_REASONING_TOKEN_ATTR_KEYS,\n  LLM_CACHED_TOKEN_ATTR_KEYS,\n  LLM_CACHE_WRITE_TOKEN_ATTR_KEYS,\n] as const\n\nconst EXECUTION_MEASUREMENT_KEY_GROUPS = [\n  ...TOKEN_MEASUREMENT_KEY_GROUPS,\n  LLM_COST_ATTR_KEYS,\n] as const\n\ninterface RetainedCallSummary {\n  callCount: number\n  measurements: Array<{\n    total: number\n    reportingCalls: number\n  }>\n}\n\n/**\n * Reconcile execution measurements across nested telemetry wrappers.\n * A measured parent is used only when a descendant call does not report the\n * same field, so aggregate wrappers neither duplicate complete child data nor\n * erase complementary parent fields.\n */\nexport function summarizeExecutionMeasurements(\n  spans: ExecutionMeasurementSpan[],\n): ExecutionMeasurements {\n  const byId = new Map<string, ExecutionMeasurementSpan>()\n  for (const span of spans) {\n    if (byId.has(span.id)) {\n      throw new Error(`summarizeExecutionMeasurements: duplicate span id \"${span.id}\"`)\n    }\n    byId.set(span.id, span)\n  }\n  const tokenMeasurementKeys = TOKEN_MEASUREMENT_KEY_GROUPS.flat()\n  const candidates = spans.filter(\n    (span) =>\n      span.modelCall ||\n      (!span.aggregate && readNumber(span.attributes, tokenMeasurementKeys) !== undefined),\n  )\n  const candidateIds = new Set(candidates.map((span) => span.id))\n  const candidateChildren = new Map<string, ExecutionMeasurementSpan[]>()\n  for (const candidate of candidates) {\n    const parentId = nearestCandidateParent(candidate, byId, candidateIds)\n    if (!parentId) continue\n    const children = candidateChildren.get(parentId) ?? []\n    children.push(candidate)\n    candidateChildren.set(parentId, children)\n  }\n  const aggregateIds = classifyAggregateSpans(candidates, candidateChildren)\n  const untypedRunCostIds = new Set(\n    spans\n      .filter(\n        (span) =>\n          !span.modelCall &&\n          !span.aggregate &&\n          !candidateIds.has(span.id) &&\n          readNumber(span.attributes, RUN_COST_ATTR_KEYS) !== undefined,\n      )\n      .map((span) => span.id),\n  )\n  const aggregateSourceIds = new Set(\n    spans\n      .filter(\n        (span) => span.aggregate || aggregateIds.has(span.id) || untypedRunCostIds.has(span.id),\n      )\n      .map((span) => span.id),\n  )\n\n  const calls = candidates.filter((span) => !aggregateIds.has(span.id))\n  const input = reconcileMeasurement(calls, byId, aggregateSourceIds, LLM_INPUT_TOKEN_ATTR_KEYS)\n  const reasoning = reconcileMeasurement(\n    calls,\n    byId,\n    aggregateSourceIds,\n    LLM_REASONING_TOKEN_ATTR_KEYS,\n  )\n  const output = reconcileMeasurement(\n    calls,\n    byId,\n    aggregateSourceIds,\n    LLM_OUTPUT_TOKEN_ATTR_KEYS,\n    LLM_REASONING_TOKEN_ATTR_KEYS,\n  )\n  const cached = reconcileMeasurement(calls, byId, aggregateSourceIds, LLM_CACHED_TOKEN_ATTR_KEYS)\n  const cacheWrite = reconcileMeasurement(\n    calls,\n    byId,\n    aggregateSourceIds,\n    LLM_CACHE_WRITE_TOKEN_ATTR_KEYS,\n  )\n  const aggregate = summarizeAggregateMeasurements(\n    spans,\n    byId,\n    new Set([...aggregateIds, ...untypedRunCostIds]),\n  )\n\n  return {\n    tokenUsage: {\n      input: input.value ?? 0,\n      output: Math.max(output.value ?? 0, reasoning.value ?? 0),\n      ...(reasoning.value !== undefined ? { reasoning: reasoning.value } : {}),\n      ...(cached.value !== undefined ? { cached: cached.value } : {}),\n      ...(cacheWrite.value !== undefined ? { cacheWrite: cacheWrite.value } : {}),\n    },\n    modelCallCount: calls.length,\n    callSpanIds: calls.map((span) => span.id),\n    cost: reconcileMeasurement(calls, byId, aggregateSourceIds, LLM_COST_ATTR_KEYS),\n    ...(aggregate ? { aggregate } : {}),\n  }\n}\n\nexport function recordAggregateMeasurements(\n  raw: Record<string, number>,\n  aggregate: ExecutionMeasurements['aggregate'],\n): void {\n  if (!aggregate) return\n  raw.aggregate_prompt_tokens = aggregate.tokenUsage.input\n  raw.aggregate_completion_tokens = aggregate.tokenUsage.output\n  if (aggregate.tokenUsage.reasoning !== undefined)\n    raw.aggregate_reasoning_tokens = aggregate.tokenUsage.reasoning\n  if (aggregate.tokenUsage.cached !== undefined)\n    raw.aggregate_cached_tokens = aggregate.tokenUsage.cached\n  if (aggregate.tokenUsage.cacheWrite !== undefined)\n    raw.aggregate_cache_write_tokens = aggregate.tokenUsage.cacheWrite\n  if (aggregate.costUsd !== undefined) raw.aggregate_cost_usd = aggregate.costUsd\n}\n\nfunction summarizeAggregateMeasurements(\n  spans: ExecutionMeasurementSpan[],\n  byId: Map<string, ExecutionMeasurementSpan>,\n  aggregateIds: Set<string>,\n): ExecutionMeasurements['aggregate'] {\n  const aggregates = spans.filter((span) => span.aggregate || aggregateIds.has(span.id))\n  const input = reconcileTopLevelMeasurement(aggregates, byId, LLM_INPUT_TOKEN_ATTR_KEYS)\n  const output = reconcileTopLevelMeasurement(aggregates, byId, LLM_OUTPUT_TOKEN_ATTR_KEYS)\n  const reasoning = reconcileTopLevelMeasurement(aggregates, byId, LLM_REASONING_TOKEN_ATTR_KEYS)\n  const cached = reconcileTopLevelMeasurement(aggregates, byId, LLM_CACHED_TOKEN_ATTR_KEYS)\n  const cacheWrite = reconcileTopLevelMeasurement(aggregates, byId, LLM_CACHE_WRITE_TOKEN_ATTR_KEYS)\n  const costUsd = reconcileTopLevelMeasurement(aggregates, byId, LLM_COST_ATTR_KEYS)\n  if (\n    input === undefined &&\n    output === undefined &&\n    reasoning === undefined &&\n    cached === undefined &&\n    cacheWrite === undefined &&\n    costUsd === undefined\n  )\n    return undefined\n  return {\n    tokenUsage: {\n      input: input ?? 0,\n      output: output ?? reasoning ?? 0,\n      ...(reasoning !== undefined ? { reasoning } : {}),\n      ...(cached !== undefined ? { cached } : {}),\n      ...(cacheWrite !== undefined ? { cacheWrite } : {}),\n    },\n    ...(costUsd !== undefined ? { costUsd } : {}),\n  }\n}\n\nfunction reconcileTopLevelMeasurement(\n  spans: ExecutionMeasurementSpan[],\n  byId: Map<string, ExecutionMeasurementSpan>,\n  keys: readonly string[],\n): number | undefined {\n  const selected = new Map<string, number>()\n  for (const span of spans) {\n    const value = readNumber(span.attributes, keys)\n    if (value !== undefined) selected.set(span.id, value)\n  }\n  for (const spanId of [...selected.keys()]) {\n    const span = byId.get(spanId)\n    if (!span) continue\n    if (ancestorIds(span, byId).some((ancestorId) => selected.has(ancestorId))) {\n      selected.delete(spanId)\n    }\n  }\n  return selected.size > 0\n    ? [...selected.values()].reduce((total, value) => total + value, 0)\n    : undefined\n}\n\nfunction reconcileMeasurement(\n  calls: ExecutionMeasurementSpan[],\n  byId: Map<string, ExecutionMeasurementSpan>,\n  aggregateSourceIds: Set<string>,\n  keys: readonly string[],\n  fallbackKeys?: readonly string[],\n): MeasurementCoverage {\n  const selected = new Map<string, number>()\n  const callIds = new Set(calls.map((call) => call.id))\n  let reportingCalls = 0\n\n  for (const call of calls) {\n    const primary = nearestMeasurement(call, byId, callIds, aggregateSourceIds, keys)\n    const fallback = fallbackKeys\n      ? nearestMeasurement(call, byId, callIds, aggregateSourceIds, fallbackKeys)\n      : undefined\n    const source =\n      primary && fallback && primary.span.id === fallback.span.id\n        ? { span: primary.span, value: Math.max(primary.value, fallback.value) }\n        : (primary ?? fallback)\n    if (!source) continue\n    reportingCalls += 1\n    selected.set(source.span.id, source.value)\n  }\n\n  for (const spanId of [...selected.keys()]) {\n    const span = byId.get(spanId)\n    if (!span) continue\n    if (\n      ancestorIds(span, byId).some(\n        (ancestorId) => selected.has(ancestorId) && !callIds.has(ancestorId),\n      )\n    ) {\n      selected.delete(spanId)\n    }\n  }\n\n  return {\n    ...(selected.size > 0\n      ? { value: [...selected.values()].reduce((total, value) => total + value, 0) }\n      : {}),\n    reportingCalls,\n    complete: calls.length > 0 && reportingCalls === calls.length,\n  }\n}\n\nfunction nearestMeasurement(\n  call: ExecutionMeasurementSpan,\n  byId: Map<string, ExecutionMeasurementSpan>,\n  callIds: Set<string>,\n  aggregateSourceIds: Set<string>,\n  keys: readonly string[],\n): { span: ExecutionMeasurementSpan; value: number } | undefined {\n  let current: ExecutionMeasurementSpan | undefined = call\n  const seen = new Set<string>()\n  while (current && !seen.has(current.id)) {\n    seen.add(current.id)\n    const value = readNumber(current.attributes, keys)\n    if (\n      value !== undefined &&\n      (current.id === call.id || (!callIds.has(current.id) && aggregateSourceIds.has(current.id)))\n    ) {\n      return { span: current, value }\n    }\n    current = current.parentId ? byId.get(current.parentId) : undefined\n  }\n  return undefined\n}\n\nfunction classifyAggregateSpans(\n  candidates: ExecutionMeasurementSpan[],\n  childrenById: Map<string, ExecutionMeasurementSpan[]>,\n): Set<string> {\n  const aggregateIds = new Set<string>()\n  const summaries = new Map<string, RetainedCallSummary>()\n  const visiting = new Set<string>()\n\n  const visit = (span: ExecutionMeasurementSpan): RetainedCallSummary => {\n    const cached = summaries.get(span.id)\n    if (cached) return cached\n    if (visiting.has(span.id)) return emptyRetainedCallSummary()\n    visiting.add(span.id)\n\n    const descendants = emptyRetainedCallSummary()\n    for (const child of childrenById.get(span.id) ?? []) {\n      const childDescendants = visit(child)\n      if (!aggregateIds.has(child.id)) addRetainedCall(descendants, child)\n      addRetainedCallSummary(descendants, childDescendants)\n    }\n\n    if (\n      descendants.callCount > 0 &&\n      (!span.modelCall || hasCompatibleDescendantMeasurements(span, descendants))\n    ) {\n      aggregateIds.add(span.id)\n    }\n\n    visiting.delete(span.id)\n    summaries.set(span.id, descendants)\n    return descendants\n  }\n\n  for (const candidate of candidates) visit(candidate)\n  return aggregateIds\n}\n\nfunction emptyRetainedCallSummary(): RetainedCallSummary {\n  return {\n    callCount: 0,\n    measurements: EXECUTION_MEASUREMENT_KEY_GROUPS.map(() => ({\n      total: 0,\n      reportingCalls: 0,\n    })),\n  }\n}\n\nfunction addRetainedCall(summary: RetainedCallSummary, span: ExecutionMeasurementSpan): void {\n  summary.callCount += 1\n  for (let index = 0; index < EXECUTION_MEASUREMENT_KEY_GROUPS.length; index += 1) {\n    const value = readNumber(span.attributes, EXECUTION_MEASUREMENT_KEY_GROUPS[index]!)\n    if (value === undefined) continue\n    const measurement = summary.measurements[index]!\n    measurement.total += value\n    measurement.reportingCalls += 1\n  }\n}\n\nfunction addRetainedCallSummary(target: RetainedCallSummary, source: RetainedCallSummary): void {\n  target.callCount += source.callCount\n  for (let index = 0; index < target.measurements.length; index += 1) {\n    const measurement = target.measurements[index]!\n    const sourceMeasurement = source.measurements[index]!\n    measurement.total += sourceMeasurement.total\n    measurement.reportingCalls += sourceMeasurement.reportingCalls\n  }\n}\n\nfunction hasCompatibleDescendantMeasurements(\n  span: ExecutionMeasurementSpan,\n  descendants: RetainedCallSummary,\n): boolean {\n  let parentMeasurements = 0\n  let descendantMeasurements = 0\n  for (let index = 0; index < EXECUTION_MEASUREMENT_KEY_GROUPS.length; index += 1) {\n    const keys = EXECUTION_MEASUREMENT_KEY_GROUPS[index]!\n    const parentValue = readNumber(span.attributes, keys)\n    const measurement = descendants.measurements[index]!\n    if (parentValue !== undefined) parentMeasurements += 1\n    if (measurement.reportingCalls > 0) descendantMeasurements += 1\n    if (parentValue === undefined || measurement.reportingCalls === 0) continue\n    if (measurement.reportingCalls !== descendants.callCount) continue\n    if (Math.abs(parentValue - measurement.total) > 1e-12) return false\n  }\n  return parentMeasurements === 0 || descendantMeasurements > 0\n}\n\nfunction ancestorIds(\n  span: ExecutionMeasurementSpan,\n  byId: Map<string, ExecutionMeasurementSpan>,\n): string[] {\n  const ids: string[] = []\n  const seen = new Set<string>()\n  let parentId = span.parentId\n  while (parentId && !seen.has(parentId)) {\n    ids.push(parentId)\n    seen.add(parentId)\n    parentId = byId.get(parentId)?.parentId\n  }\n  return ids\n}\n\nfunction nearestCandidateParent(\n  span: ExecutionMeasurementSpan,\n  byId: Map<string, ExecutionMeasurementSpan>,\n  candidateIds: Set<string>,\n): string | undefined {\n  const seen = new Set<string>()\n  let parentId = span.parentId\n  while (parentId && !seen.has(parentId)) {\n    if (candidateIds.has(parentId)) return parentId\n    seen.add(parentId)\n    parentId = byId.get(parentId)?.parentId\n  }\n  return undefined\n}\n\nfunction readNumber(\n  attributes: Record<string, unknown>,\n  keys: readonly string[],\n): number | undefined {\n  for (const key of keys) {\n    const value = attributes[key]\n    const parsed =\n      typeof value === 'number'\n        ? value\n        : typeof value === 'string' && value.length > 0\n          ? Number(value)\n          : Number.NaN\n    if (Number.isFinite(parsed) && parsed >= 0) return parsed\n  }\n  return undefined\n}\n","import { ValidationError } from '../errors'\nimport { FAILURE_CLASSES, type FailureClass } from './schema'\n\nconst TASK_FAILURE_CLASS_ATTR = 'tangle.task.failure_class'\nconst TASK_FAILURE_MODE_ATTR = 'tangle.task.failure_mode'\n\ninterface AttributeCarrier {\n  attributes: Record<string, unknown>\n}\n\nexport type TaskFailureLabels =\n  | { failureClass?: undefined; failureMode?: undefined }\n  | { failureClass: 'success'; failureMode?: undefined }\n  | { failureClass: Exclude<FailureClass, 'success'>; failureMode?: string }\n\nexport function readTaskFailureLabels(\n  roots: readonly AttributeCarrier[],\n  context: string,\n): TaskFailureLabels {\n  const failureClass = readConsistentRootString(roots, TASK_FAILURE_CLASS_ATTR, context)\n  const failureMode = readConsistentRootString(roots, TASK_FAILURE_MODE_ATTR, context)\n\n  if (failureClass !== undefined && !FAILURE_CLASSES.includes(failureClass as FailureClass)) {\n    throw new ValidationError(\n      `${context}: ${TASK_FAILURE_CLASS_ATTR} must be one of ${FAILURE_CLASSES.join(', ')}`,\n    )\n  }\n  if (failureMode !== undefined && (failureClass === undefined || failureClass === 'success')) {\n    throw new ValidationError(\n      `${context}: ${TASK_FAILURE_MODE_ATTR} requires a non-success ${TASK_FAILURE_CLASS_ATTR}`,\n    )\n  }\n\n  if (failureClass === undefined) return {}\n  if (failureClass === 'success') return { failureClass }\n  return {\n    failureClass: failureClass as Exclude<FailureClass, 'success'>,\n    ...(failureMode ? { failureMode } : {}),\n  }\n}\n\nfunction readConsistentRootString(\n  roots: readonly AttributeCarrier[],\n  key: string,\n  context: string,\n): string | undefined {\n  const values = new Set<string>()\n  for (const root of roots) {\n    if (!Object.hasOwn(root.attributes, key)) continue\n    const value = root.attributes[key]\n    if (typeof value !== 'string' || value.trim().length === 0) {\n      throw new ValidationError(`${context}: ${key} must be a non-empty string`)\n    }\n    values.add(value)\n  }\n\n  if (values.size > 1) {\n    throw new ValidationError(\n      `${context}: conflicting ${key} values: ${[...values].sort().join(', ')}`,\n    )\n  }\n  return values.values().next().value\n}\n"],"mappings":";;;;;;;;;;;;;;;;;AA0BO,SAAS,qBAAqB,SAAyD;AAC5F,QAAM,OAAO,oBAAI,IAA8B;AAC/C,aAAW,UAAU,SAAS;AAC5B,QAAI,KAAK,IAAI,OAAO,EAAE,GAAG;AACvB,YAAM,IAAI,MAAM,4CAA4C,OAAO,EAAE,GAAG;AAAA,IAC1E;AACA,SAAK,IAAI,OAAO,IAAI,MAAM;AAAA,EAC5B;AAEA,QAAM,aAAa,oBAAI,IAAY;AACnC,aAAW,UAAU,SAAS;AAC5B,QAAI,CAAC,OAAO,MAAO;AACnB,UAAM,UAAU,oBAAI,IAAY;AAChC,QAAI,WAAW,OAAO;AACtB,WAAO,YAAY,CAAC,QAAQ,IAAI,QAAQ,GAAG;AACzC,cAAQ,IAAI,QAAQ;AACpB,YAAM,SAAS,KAAK,IAAI,QAAQ;AAChC,UAAI,CAAC,OAAQ;AACb,UAAI,OAAO,MAAO,YAAW,IAAI,OAAO,EAAE;AAC1C,iBAAW,OAAO;AAAA,IACpB;AAAA,EACF;AAEA,QAAM,UAA6B;AAAA,IACjC,OAAO;AAAA,IACP,WAAW;AAAA,IACX,SAAS;AAAA,IACT,WAAW;AAAA,IACX,YAAY;AAAA,IACZ,YAAY;AAAA,IACZ,cAAc;AAAA,EAChB;AAEA,aAAW,UAAU,SAAS;AAC5B,QAAI,CAAC,OAAO,MAAO;AACnB,YAAQ,SAAS;AACjB,QAAI,OAAO,SAAS,aAAa;AAC/B,cAAQ,aAAa;AAAA,IACvB,WAAW,OAAO,SAAS,aAAa;AACtC,cAAQ,cAAc;AAAA,IACxB,WAAW,OAAO,aAAa;AAC7B,cAAQ,WAAW;AAAA,IACrB,WAAW,WAAW,IAAI,OAAO,EAAE,GAAG;AACpC,cAAQ,cAAc;AAAA,IACxB,WACE,OAAO,SAAS,WAChB,OAAO,SAAS,WAChB,OAAO,SAAS,SAChB,OAAO,SAAS,QAChB;AACA,cAAQ,aAAa;AAAA,IACvB,OAAO;AACL,cAAQ,gBAAgB;AAAA,IAC1B;AAAA,EACF;AAEA,SAAO;AACT;;;AC/CA,IAAM,+BAA+B;AAAA,EACnC;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AACF;AAEA,IAAM,mCAAmC;AAAA,EACvC,GAAG;AAAA,EACH;AACF;AAgBO,SAAS,+BACd,OACuB;AACvB,QAAM,OAAO,oBAAI,IAAsC;AACvD,aAAW,QAAQ,OAAO;AACxB,QAAI,KAAK,IAAI,KAAK,EAAE,GAAG;AACrB,YAAM,IAAI,MAAM,sDAAsD,KAAK,EAAE,GAAG;AAAA,IAClF;AACA,SAAK,IAAI,KAAK,IAAI,IAAI;AAAA,EACxB;AACA,QAAM,uBAAuB,6BAA6B,KAAK;AAC/D,QAAM,aAAa,MAAM;AAAA,IACvB,CAAC,SACC,KAAK,aACJ,CAAC,KAAK,aAAa,WAAW,KAAK,YAAY,oBAAoB,MAAM;AAAA,EAC9E;AACA,QAAM,eAAe,IAAI,IAAI,WAAW,IAAI,CAAC,SAAS,KAAK,EAAE,CAAC;AAC9D,QAAM,oBAAoB,oBAAI,IAAwC;AACtE,aAAW,aAAa,YAAY;AAClC,UAAM,WAAW,uBAAuB,WAAW,MAAM,YAAY;AACrE,QAAI,CAAC,SAAU;AACf,UAAM,WAAW,kBAAkB,IAAI,QAAQ,KAAK,CAAC;AACrD,aAAS,KAAK,SAAS;AACvB,sBAAkB,IAAI,UAAU,QAAQ;AAAA,EAC1C;AACA,QAAM,eAAe,uBAAuB,YAAY,iBAAiB;AACzE,QAAM,oBAAoB,IAAI;AAAA,IAC5B,MACG;AAAA,MACC,CAAC,SACC,CAAC,KAAK,aACN,CAAC,KAAK,aACN,CAAC,aAAa,IAAI,KAAK,EAAE,KACzB,WAAW,KAAK,YAAY,kBAAkB,MAAM;AAAA,IACxD,EACC,IAAI,CAAC,SAAS,KAAK,EAAE;AAAA,EAC1B;AACA,QAAM,qBAAqB,IAAI;AAAA,IAC7B,MACG;AAAA,MACC,CAAC,SAAS,KAAK,aAAa,aAAa,IAAI,KAAK,EAAE,KAAK,kBAAkB,IAAI,KAAK,EAAE;AAAA,IACxF,EACC,IAAI,CAAC,SAAS,KAAK,EAAE;AAAA,EAC1B;AAEA,QAAM,QAAQ,WAAW,OAAO,CAAC,SAAS,CAAC,aAAa,IAAI,KAAK,EAAE,CAAC;AACpE,QAAM,QAAQ,qBAAqB,OAAO,MAAM,oBAAoB,yBAAyB;AAC7F,QAAM,YAAY;AAAA,IAChB;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,EACF;AACA,QAAM,SAAS;AAAA,IACb;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,EACF;AACA,QAAM,SAAS,qBAAqB,OAAO,MAAM,oBAAoB,0BAA0B;AAC/F,QAAM,aAAa;AAAA,IACjB;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,EACF;AACA,QAAM,YAAY;AAAA,IAChB;AAAA,IACA;AAAA,IACA,oBAAI,IAAI,CAAC,GAAG,cAAc,GAAG,iBAAiB,CAAC;AAAA,EACjD;AAEA,SAAO;AAAA,IACL,YAAY;AAAA,MACV,OAAO,MAAM,SAAS;AAAA,MACtB,QAAQ,KAAK,IAAI,OAAO,SAAS,GAAG,UAAU,SAAS,CAAC;AAAA,MACxD,GAAI,UAAU,UAAU,SAAY,EAAE,WAAW,UAAU,MAAM,IAAI,CAAC;AAAA,MACtE,GAAI,OAAO,UAAU,SAAY,EAAE,QAAQ,OAAO,MAAM,IAAI,CAAC;AAAA,MAC7D,GAAI,WAAW,UAAU,SAAY,EAAE,YAAY,WAAW,MAAM,IAAI,CAAC;AAAA,IAC3E;AAAA,IACA,gBAAgB,MAAM;AAAA,IACtB,aAAa,MAAM,IAAI,CAAC,SAAS,KAAK,EAAE;AAAA,IACxC,MAAM,qBAAqB,OAAO,MAAM,oBAAoB,kBAAkB;AAAA,IAC9E,GAAI,YAAY,EAAE,UAAU,IAAI,CAAC;AAAA,EACnC;AACF;AAEO,SAAS,4BACd,KACA,WACM;AACN,MAAI,CAAC,UAAW;AAChB,MAAI,0BAA0B,UAAU,WAAW;AACnD,MAAI,8BAA8B,UAAU,WAAW;AACvD,MAAI,UAAU,WAAW,cAAc;AACrC,QAAI,6BAA6B,UAAU,WAAW;AACxD,MAAI,UAAU,WAAW,WAAW;AAClC,QAAI,0BAA0B,UAAU,WAAW;AACrD,MAAI,UAAU,WAAW,eAAe;AACtC,QAAI,+BAA+B,UAAU,WAAW;AAC1D,MAAI,UAAU,YAAY,OAAW,KAAI,qBAAqB,UAAU;AAC1E;AAEA,SAAS,+BACP,OACA,MACA,cACoC;AACpC,QAAM,aAAa,MAAM,OAAO,CAAC,SAAS,KAAK,aAAa,aAAa,IAAI,KAAK,EAAE,CAAC;AACrF,QAAM,QAAQ,6BAA6B,YAAY,MAAM,yBAAyB;AACtF,QAAM,SAAS,6BAA6B,YAAY,MAAM,0BAA0B;AACxF,QAAM,YAAY,6BAA6B,YAAY,MAAM,6BAA6B;AAC9F,QAAM,SAAS,6BAA6B,YAAY,MAAM,0BAA0B;AACxF,QAAM,aAAa,6BAA6B,YAAY,MAAM,+BAA+B;AACjG,QAAM,UAAU,6BAA6B,YAAY,MAAM,kBAAkB;AACjF,MACE,UAAU,UACV,WAAW,UACX,cAAc,UACd,WAAW,UACX,eAAe,UACf,YAAY;AAEZ,WAAO;AACT,SAAO;AAAA,IACL,YAAY;AAAA,MACV,OAAO,SAAS;AAAA,MAChB,QAAQ,UAAU,aAAa;AAAA,MAC/B,GAAI,cAAc,SAAY,EAAE,UAAU,IAAI,CAAC;AAAA,MAC/C,GAAI,WAAW,SAAY,EAAE,OAAO,IAAI,CAAC;AAAA,MACzC,GAAI,eAAe,SAAY,EAAE,WAAW,IAAI,CAAC;AAAA,IACnD;AAAA,IACA,GAAI,YAAY,SAAY,EAAE,QAAQ,IAAI,CAAC;AAAA,EAC7C;AACF;AAEA,SAAS,6BACP,OACA,MACA,MACoB;AACpB,QAAM,WAAW,oBAAI,IAAoB;AACzC,aAAW,QAAQ,OAAO;AACxB,UAAM,QAAQ,WAAW,KAAK,YAAY,IAAI;AAC9C,QAAI,UAAU,OAAW,UAAS,IAAI,KAAK,IAAI,KAAK;AAAA,EACtD;AACA,aAAW,UAAU,CAAC,GAAG,SAAS,KAAK,CAAC,GAAG;AACzC,UAAM,OAAO,KAAK,IAAI,MAAM;AAC5B,QAAI,CAAC,KAAM;AACX,QAAI,YAAY,MAAM,IAAI,EAAE,KAAK,CAAC,eAAe,SAAS,IAAI,UAAU,CAAC,GAAG;AAC1E,eAAS,OAAO,MAAM;AAAA,IACxB;AAAA,EACF;AACA,SAAO,SAAS,OAAO,IACnB,CAAC,GAAG,SAAS,OAAO,CAAC,EAAE,OAAO,CAAC,OAAO,UAAU,QAAQ,OAAO,CAAC,IAChE;AACN;AAEA,SAAS,qBACP,OACA,MACA,oBACA,MACA,cACqB;AACrB,QAAM,WAAW,oBAAI,IAAoB;AACzC,QAAM,UAAU,IAAI,IAAI,MAAM,IAAI,CAAC,SAAS,KAAK,EAAE,CAAC;AACpD,MAAI,iBAAiB;AAErB,aAAW,QAAQ,OAAO;AACxB,UAAM,UAAU,mBAAmB,MAAM,MAAM,SAAS,oBAAoB,IAAI;AAChF,UAAM,WAAW,eACb,mBAAmB,MAAM,MAAM,SAAS,oBAAoB,YAAY,IACxE;AACJ,UAAM,SACJ,WAAW,YAAY,QAAQ,KAAK,OAAO,SAAS,KAAK,KACrD,EAAE,MAAM,QAAQ,MAAM,OAAO,KAAK,IAAI,QAAQ,OAAO,SAAS,KAAK,EAAE,IACpE,WAAW;AAClB,QAAI,CAAC,OAAQ;AACb,sBAAkB;AAClB,aAAS,IAAI,OAAO,KAAK,IAAI,OAAO,KAAK;AAAA,EAC3C;AAEA,aAAW,UAAU,CAAC,GAAG,SAAS,KAAK,CAAC,GAAG;AACzC,UAAM,OAAO,KAAK,IAAI,MAAM;AAC5B,QAAI,CAAC,KAAM;AACX,QACE,YAAY,MAAM,IAAI,EAAE;AAAA,MACtB,CAAC,eAAe,SAAS,IAAI,UAAU,KAAK,CAAC,QAAQ,IAAI,UAAU;AAAA,IACrE,GACA;AACA,eAAS,OAAO,MAAM;AAAA,IACxB;AAAA,EACF;AAEA,SAAO;AAAA,IACL,GAAI,SAAS,OAAO,IAChB,EAAE,OAAO,CAAC,GAAG,SAAS,OAAO,CAAC,EAAE,OAAO,CAAC,OAAO,UAAU,QAAQ,OAAO,CAAC,EAAE,IAC3E,CAAC;AAAA,IACL;AAAA,IACA,UAAU,MAAM,SAAS,KAAK,mBAAmB,MAAM;AAAA,EACzD;AACF;AAEA,SAAS,mBACP,MACA,MACA,SACA,oBACA,MAC+D;AAC/D,MAAI,UAAgD;AACpD,QAAM,OAAO,oBAAI,IAAY;AAC7B,SAAO,WAAW,CAAC,KAAK,IAAI,QAAQ,EAAE,GAAG;AACvC,SAAK,IAAI,QAAQ,EAAE;AACnB,UAAM,QAAQ,WAAW,QAAQ,YAAY,IAAI;AACjD,QACE,UAAU,WACT,QAAQ,OAAO,KAAK,MAAO,CAAC,QAAQ,IAAI,QAAQ,EAAE,KAAK,mBAAmB,IAAI,QAAQ,EAAE,IACzF;AACA,aAAO,EAAE,MAAM,SAAS,MAAM;AAAA,IAChC;AACA,cAAU,QAAQ,WAAW,KAAK,IAAI,QAAQ,QAAQ,IAAI;AAAA,EAC5D;AACA,SAAO;AACT;AAEA,SAAS,uBACP,YACA,cACa;AACb,QAAM,eAAe,oBAAI,IAAY;AACrC,QAAM,YAAY,oBAAI,IAAiC;AACvD,QAAM,WAAW,oBAAI,IAAY;AAEjC,QAAM,QAAQ,CAAC,SAAwD;AACrE,UAAM,SAAS,UAAU,IAAI,KAAK,EAAE;AACpC,QAAI,OAAQ,QAAO;AACnB,QAAI,SAAS,IAAI,KAAK,EAAE,EAAG,QAAO,yBAAyB;AAC3D,aAAS,IAAI,KAAK,EAAE;AAEpB,UAAM,cAAc,yBAAyB;AAC7C,eAAW,SAAS,aAAa,IAAI,KAAK,EAAE,KAAK,CAAC,GAAG;AACnD,YAAM,mBAAmB,MAAM,KAAK;AACpC,UAAI,CAAC,aAAa,IAAI,MAAM,EAAE,EAAG,iBAAgB,aAAa,KAAK;AACnE,6BAAuB,aAAa,gBAAgB;AAAA,IACtD;AAEA,QACE,YAAY,YAAY,MACvB,CAAC,KAAK,aAAa,oCAAoC,MAAM,WAAW,IACzE;AACA,mBAAa,IAAI,KAAK,EAAE;AAAA,IAC1B;AAEA,aAAS,OAAO,KAAK,EAAE;AACvB,cAAU,IAAI,KAAK,IAAI,WAAW;AAClC,WAAO;AAAA,EACT;AAEA,aAAW,aAAa,WAAY,OAAM,SAAS;AACnD,SAAO;AACT;AAEA,SAAS,2BAAgD;AACvD,SAAO;AAAA,IACL,WAAW;AAAA,IACX,cAAc,iCAAiC,IAAI,OAAO;AAAA,MACxD,OAAO;AAAA,MACP,gBAAgB;AAAA,IAClB,EAAE;AAAA,EACJ;AACF;AAEA,SAAS,gBAAgB,SAA8B,MAAsC;AAC3F,UAAQ,aAAa;AACrB,WAAS,QAAQ,GAAG,QAAQ,iCAAiC,QAAQ,SAAS,GAAG;AAC/E,UAAM,QAAQ,WAAW,KAAK,YAAY,iCAAiC,KAAK,CAAE;AAClF,QAAI,UAAU,OAAW;AACzB,UAAM,cAAc,QAAQ,aAAa,KAAK;AAC9C,gBAAY,SAAS;AACrB,gBAAY,kBAAkB;AAAA,EAChC;AACF;AAEA,SAAS,uBAAuB,QAA6B,QAAmC;AAC9F,SAAO,aAAa,OAAO;AAC3B,WAAS,QAAQ,GAAG,QAAQ,OAAO,aAAa,QAAQ,SAAS,GAAG;AAClE,UAAM,cAAc,OAAO,aAAa,KAAK;AAC7C,UAAM,oBAAoB,OAAO,aAAa,KAAK;AACnD,gBAAY,SAAS,kBAAkB;AACvC,gBAAY,kBAAkB,kBAAkB;AAAA,EAClD;AACF;AAEA,SAAS,oCACP,MACA,aACS;AACT,MAAI,qBAAqB;AACzB,MAAI,yBAAyB;AAC7B,WAAS,QAAQ,GAAG,QAAQ,iCAAiC,QAAQ,SAAS,GAAG;AAC/E,UAAM,OAAO,iCAAiC,KAAK;AACnD,UAAM,cAAc,WAAW,KAAK,YAAY,IAAI;AACpD,UAAM,cAAc,YAAY,aAAa,KAAK;AAClD,QAAI,gBAAgB,OAAW,uBAAsB;AACrD,QAAI,YAAY,iBAAiB,EAAG,2BAA0B;AAC9D,QAAI,gBAAgB,UAAa,YAAY,mBAAmB,EAAG;AACnE,QAAI,YAAY,mBAAmB,YAAY,UAAW;AAC1D,QAAI,KAAK,IAAI,cAAc,YAAY,KAAK,IAAI,MAAO,QAAO;AAAA,EAChE;AACA,SAAO,uBAAuB,KAAK,yBAAyB;AAC9D;AAEA,SAAS,YACP,MACA,MACU;AACV,QAAM,MAAgB,CAAC;AACvB,QAAM,OAAO,oBAAI,IAAY;AAC7B,MAAI,WAAW,KAAK;AACpB,SAAO,YAAY,CAAC,KAAK,IAAI,QAAQ,GAAG;AACtC,QAAI,KAAK,QAAQ;AACjB,SAAK,IAAI,QAAQ;AACjB,eAAW,KAAK,IAAI,QAAQ,GAAG;AAAA,EACjC;AACA,SAAO;AACT;AAEA,SAAS,uBACP,MACA,MACA,cACoB;AACpB,QAAM,OAAO,oBAAI,IAAY;AAC7B,MAAI,WAAW,KAAK;AACpB,SAAO,YAAY,CAAC,KAAK,IAAI,QAAQ,GAAG;AACtC,QAAI,aAAa,IAAI,QAAQ,EAAG,QAAO;AACvC,SAAK,IAAI,QAAQ;AACjB,eAAW,KAAK,IAAI,QAAQ,GAAG;AAAA,EACjC;AACA,SAAO;AACT;AAEA,SAAS,WACP,YACA,MACoB;AACpB,aAAW,OAAO,MAAM;AACtB,UAAM,QAAQ,WAAW,GAAG;AAC5B,UAAM,SACJ,OAAO,UAAU,WACb,QACA,OAAO,UAAU,YAAY,MAAM,SAAS,IAC1C,OAAO,KAAK,IACZ,OAAO;AACf,QAAI,OAAO,SAAS,MAAM,KAAK,UAAU,EAAG,QAAO;AAAA,EACrD;AACA,SAAO;AACT;;;ACpaA,IAAM,0BAA0B;AAChC,IAAM,yBAAyB;AAWxB,SAAS,sBACd,OACA,SACmB;AACnB,QAAM,eAAe,yBAAyB,OAAO,yBAAyB,OAAO;AACrF,QAAM,cAAc,yBAAyB,OAAO,wBAAwB,OAAO;AAEnF,MAAI,iBAAiB,UAAa,CAAC,gBAAgB,SAAS,YAA4B,GAAG;AACzF,UAAM,IAAI;AAAA,MACR,GAAG,OAAO,KAAK,uBAAuB,mBAAmB,gBAAgB,KAAK,IAAI,CAAC;AAAA,IACrF;AAAA,EACF;AACA,MAAI,gBAAgB,WAAc,iBAAiB,UAAa,iBAAiB,YAAY;AAC3F,UAAM,IAAI;AAAA,MACR,GAAG,OAAO,KAAK,sBAAsB,2BAA2B,uBAAuB;AAAA,IACzF;AAAA,EACF;AAEA,MAAI,iBAAiB,OAAW,QAAO,CAAC;AACxC,MAAI,iBAAiB,UAAW,QAAO,EAAE,aAAa;AACtD,SAAO;AAAA,IACL;AAAA,IACA,GAAI,cAAc,EAAE,YAAY,IAAI,CAAC;AAAA,EACvC;AACF;AAEA,SAAS,yBACP,OACA,KACA,SACoB;AACpB,QAAM,SAAS,oBAAI,IAAY;AAC/B,aAAW,QAAQ,OAAO;AACxB,QAAI,CAAC,OAAO,OAAO,KAAK,YAAY,GAAG,EAAG;AAC1C,UAAM,QAAQ,KAAK,WAAW,GAAG;AACjC,QAAI,OAAO,UAAU,YAAY,MAAM,KAAK,EAAE,WAAW,GAAG;AAC1D,YAAM,IAAI,gBAAgB,GAAG,OAAO,KAAK,GAAG,6BAA6B;AAAA,IAC3E;AACA,WAAO,IAAI,KAAK;AAAA,EAClB;AAEA,MAAI,OAAO,OAAO,GAAG;AACnB,UAAM,IAAI;AAAA,MACR,GAAG,OAAO,iBAAiB,GAAG,YAAY,CAAC,GAAG,MAAM,EAAE,KAAK,EAAE,KAAK,IAAI,CAAC;AAAA,IACzE;AAAA,EACF;AACA,SAAO,OAAO,OAAO,EAAE,KAAK,EAAE;AAChC;","names":[]}

package/dist/chunk-BOD4O7OF.js

@@ -1,40 +0,0 @@
-// src/series-convergence.ts
-function analyzeSeries(values, options = {}) {
-  const window = options.window ?? 5;
-  const stableCv = options.stableCv ?? 0.05;
-  const driftRun = options.driftRun ?? 3;
-  if (values.length < Math.max(2, Math.min(window, 3))) {
-    return { state: "insufficient-data", windowMean: 0, windowCv: 0, tailRun: 0, stable: false };
-  }
-  const tail = values.slice(-window);
-  const mean = tail.reduce((a, b) => a + b, 0) / tail.length;
-  const variance = tail.reduce((acc, v) => acc + (v - mean) ** 2, 0) / tail.length;
-  const stdDev = Math.sqrt(variance);
-  const refMean = Math.abs(mean) > 1e-9 ? Math.abs(mean) : 1;
-  const cv = stdDev / refMean;
-  const stable = tail.length >= window && cv <= stableCv;
-  let tailRun = 0;
-  let direction = 0;
-  for (let i = values.length - 1; i > 0; i--) {
-    const delta = values[i] - values[i - 1];
-    if (delta === 0) break;
-    const dir = delta > 0 ? 1 : -1;
-    if (direction === 0) direction = dir;
-    if (dir !== direction) break;
-    tailRun += dir;
-  }
-  let state;
-  if (stable) {
-    state = "stabilized";
-  } else if (Math.abs(tailRun) >= driftRun) {
-    state = tailRun > 0 ? "drifting-up" : "drifting-down";
-  } else {
-    state = "noisy";
-  }
-  return { state, windowMean: mean, windowCv: cv, tailRun, stable };
-}
-
-export {
-  analyzeSeries
-};
-//# sourceMappingURL=chunk-BOD4O7OF.js.map

package/dist/chunk-BOD4O7OF.js.map

@@ -1 +0,0 @@
-{"version":3,"sources":["../src/series-convergence.ts"],"sourcesContent":["/**\n * Series convergence — detects whether a sequence of scalar measurements\n * is stabilizing, drifting, or noisy.\n *\n * Lifted from ADC convergence.ts. The per-turn `ConvergenceTracker` is\n * about progress *within* a single run; this module is about drift\n * *across* runs (e.g. \"are my nightly eval scores stabilizing?\").\n *\n * Three signals:\n *   - stabilized: last K values have low variance (< epsilon) — done\n *   - drifting:   recent trend is monotonic and beyond noise — regressing or improving\n *   - noisy:      neither — keep iterating, but flag as untrustworthy for gating\n */\n\nexport interface SeriesConvergenceOptions {\n  /** Window size for \"recent\" analysis (default 5). */\n  window?: number\n  /** Coefficient-of-variation threshold below which the window is stabilized (default 0.05 = 5%). */\n  stableCv?: number\n  /** Minimum monotone run length to call drift (default 3). */\n  driftRun?: number\n}\n\nexport interface SeriesConvergenceResult {\n  state: 'stabilized' | 'drifting-up' | 'drifting-down' | 'noisy' | 'insufficient-data'\n  windowMean: number\n  windowCv: number\n  /** Longest monotonic run at the tail of the series (positive for up, negative for down). */\n  tailRun: number\n  /** True when n ≥ window AND windowCv ≤ stableCv. */\n  stable: boolean\n}\n\nexport function analyzeSeries(\n  values: number[],\n  options: SeriesConvergenceOptions = {},\n): SeriesConvergenceResult {\n  const window = options.window ?? 5\n  const stableCv = options.stableCv ?? 0.05\n  const driftRun = options.driftRun ?? 3\n\n  if (values.length < Math.max(2, Math.min(window, 3))) {\n    return { state: 'insufficient-data', windowMean: 0, windowCv: 0, tailRun: 0, stable: false }\n  }\n\n  const tail = values.slice(-window)\n  const mean = tail.reduce((a, b) => a + b, 0) / tail.length\n  const variance = tail.reduce((acc, v) => acc + (v - mean) ** 2, 0) / tail.length\n  const stdDev = Math.sqrt(variance)\n  const refMean = Math.abs(mean) > 1e-9 ? Math.abs(mean) : 1\n  const cv = stdDev / refMean\n  const stable = tail.length >= window && cv <= stableCv\n\n  // Tail monotonic run: count how many consecutive strictly-increasing (or decreasing)\n  // steps end at the final value.\n  let tailRun = 0\n  let direction: 1 | -1 | 0 = 0\n  for (let i = values.length - 1; i > 0; i--) {\n    const delta = values[i]! - values[i - 1]!\n    if (delta === 0) break\n    const dir = delta > 0 ? 1 : -1\n    if (direction === 0) direction = dir\n    if (dir !== direction) break\n    tailRun += dir\n  }\n\n  let state: SeriesConvergenceResult['state']\n  if (stable) {\n    state = 'stabilized'\n  } else if (Math.abs(tailRun) >= driftRun) {\n    state = tailRun > 0 ? 'drifting-up' : 'drifting-down'\n  } else {\n    state = 'noisy'\n  }\n\n  return { state, windowMean: mean, windowCv: cv, tailRun, stable }\n}\n"],"mappings":";AAiCO,SAAS,cACd,QACA,UAAoC,CAAC,GACZ;AACzB,QAAM,SAAS,QAAQ,UAAU;AACjC,QAAM,WAAW,QAAQ,YAAY;AACrC,QAAM,WAAW,QAAQ,YAAY;AAErC,MAAI,OAAO,SAAS,KAAK,IAAI,GAAG,KAAK,IAAI,QAAQ,CAAC,CAAC,GAAG;AACpD,WAAO,EAAE,OAAO,qBAAqB,YAAY,GAAG,UAAU,GAAG,SAAS,GAAG,QAAQ,MAAM;AAAA,EAC7F;AAEA,QAAM,OAAO,OAAO,MAAM,CAAC,MAAM;AACjC,QAAM,OAAO,KAAK,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,KAAK;AACpD,QAAM,WAAW,KAAK,OAAO,CAAC,KAAK,MAAM,OAAO,IAAI,SAAS,GAAG,CAAC,IAAI,KAAK;AAC1E,QAAM,SAAS,KAAK,KAAK,QAAQ;AACjC,QAAM,UAAU,KAAK,IAAI,IAAI,IAAI,OAAO,KAAK,IAAI,IAAI,IAAI;AACzD,QAAM,KAAK,SAAS;AACpB,QAAM,SAAS,KAAK,UAAU,UAAU,MAAM;AAI9C,MAAI,UAAU;AACd,MAAI,YAAwB;AAC5B,WAAS,IAAI,OAAO,SAAS,GAAG,IAAI,GAAG,KAAK;AAC1C,UAAM,QAAQ,OAAO,CAAC,IAAK,OAAO,IAAI,CAAC;AACvC,QAAI,UAAU,EAAG;AACjB,UAAM,MAAM,QAAQ,IAAI,IAAI;AAC5B,QAAI,cAAc,EAAG,aAAY;AACjC,QAAI,QAAQ,UAAW;AACvB,eAAW;AAAA,EACb;AAEA,MAAI;AACJ,MAAI,QAAQ;AACV,YAAQ;AAAA,EACV,WAAW,KAAK,IAAI,OAAO,KAAK,UAAU;AACxC,YAAQ,UAAU,IAAI,gBAAgB;AAAA,EACxC,OAAO;AACL,YAAQ;AAAA,EACV;AAEA,SAAO,EAAE,OAAO,YAAY,MAAM,UAAU,IAAI,SAAS,OAAO;AAClE;","names":[]}