@tangle-network/agent-eval 0.95.1 → 0.96.0

package/dist/adapters/http.d.ts

@@ -1,4 +1,4 @@
-import { S as Scenario, D as DispatchFn, b as DispatchContext } from '../types-DQRY8ZT-.js';
+import { S as Scenario, D as DispatchFn, b as DispatchContext } from '../types-BMahhhio.js';
 import '../run-record-CP2ObebC.js';
 import '@tangle-network/agent-interface';
 import '../errors-CzMUYo7b.js';

package/dist/adapters/langchain.d.ts

@@ -1,4 +1,4 @@
-import { S as Scenario, J as JudgeScore, D as DispatchFn, a as JudgeConfig } from '../types-DQRY8ZT-.js';
+import { S as Scenario, J as JudgeScore, D as DispatchFn, a as JudgeConfig } from '../types-BMahhhio.js';
 import '../run-record-CP2ObebC.js';
 import '@tangle-network/agent-interface';
 import '../errors-CzMUYo7b.js';

package/dist/adapters/otel.d.ts

@@ -1,5 +1,5 @@
 import { TraceSpanEvent, HostedClient } from '../hosted/index.js';
-import '../types-DQRY8ZT-.js';
+import '../types-BMahhhio.js';
 import '../run-record-CP2ObebC.js';
 import '@tangle-network/agent-interface';
 import '../errors-CzMUYo7b.js';

package/dist/analyst/index.js

@@ -16,7 +16,7 @@ import {
   behavioralAnalyst,
   buildDefaultAnalystRegistry,
   deriveEfficiencyFindings
-} from "../chunk-AQ5WQAIV.js";
+} from "../chunk-3NHEO6ZC.js";
 import "../chunk-DJWX3GVS.js";
 import {
   ANALYST_SEVERITIES,

package/dist/campaign/index.d.ts

@@ -1,9 +1,9 @@
 import { A as AnalyzeTracesOptions, a as AnalyzeTracesInput, b as AnalyzeTracesResult } from '../analyst-C8HHvfJp.js';
-import { S as Scenario, M as MutableSurface, b as DispatchContext, a as JudgeConfig, n as GenerationRecord, G as Gate, J as JudgeScore, L as LabeledScenarioStore, s as LabeledScenarioWrite, t as LabeledScenarioSampleArgs, u as LabeledScenarioRecord, v as LabelTrust, c as SurfaceProposer, w as ProposedCandidate, x as ProposeContext, y as LabeledScenarioSource, C as CampaignResult, j as CodeSurface } from '../types-DQRY8ZT-.js';
-export { e as CampaignAggregates, f as CampaignArtifactWriter, g as CampaignCellResult, h as CampaignCostMeter, z as CampaignTokenUsage, i as CampaignTraceWriter, D as DispatchFn, k as GateContext, d as GateDecision, l as GateResult, m as GenerationCandidate, A as JudgeAggregate, o as JudgeDimension, p as Mutator, O as OptimizationProposer, q as OptimizerConfig, P as ParetoParent, R as RedactionStatus, B as ScenarioAggregate, r as SessionScript, T as TraceSpan, E as isProposedCandidate, F as labelTrustRank } from '../types-DQRY8ZT-.js';
-import { b as RunCampaignOptions, c as RunImprovementLoopOptions, C as CampaignStorage } from '../gepa-C1NCIZ9o.js';
-export { e as GepaProposerConstraints, G as GepaProposerOptions, O as OpenAutoPrOptions, h as OpenAutoPrResult, a as RunImprovementLoopResult, R as RunOptimizationOptions, j as RunOptimizationResult, k as countSentenceEdits, l as defaultRenderDiff, m as extractH2Sections, f as fsCampaignStorage, g as gepaProposer, i as inMemoryCampaignStorage, o as openAutoPr, r as runCampaign, d as runImprovementLoop, n as runOptimization, s as surfaceHash } from '../gepa-C1NCIZ9o.js';
-export { A as AxisEvidence, a as AxisVerdict, B as BuildEvidenceVectorOptions, k as BuildLoopProvenanceArgs, D as DefaultProductionGateOptions, l as EmitLoopProvenanceArgs, m as EmitLoopProvenanceResult, E as EvidenceVector, b as EvolutionaryProposerOptions, H as HeldOutGateOptions, n as LoopProvenanceBackend, o as LoopProvenanceCandidate, L as LoopProvenanceRecord, O as ObjectiveSource, P as ParetoSignificanceGateOptions, c as PromotionObjective, d as PromotionPolicy, R as RunEvalOptions, e as buildEvidenceVector, q as buildLoopProvenanceRecord, f as composeGate, g as defaultProductionGate, s as emitLoopProvenance, h as evolutionaryProposer, i as heldOutGate, t as loopProvenanceSpans, p as paretoPolicy, j as paretoSignificanceGate, u as provenanceRecordPath, v as provenanceSpansPath, r as runEval, w as surfaceContentHash } from '../provenance-CncDq9qE.js';
+import { S as Scenario, M as MutableSurface, b as DispatchContext, a as JudgeConfig, d as GenerationRecord, f as Gate, J as JudgeScore, L as LabeledScenarioStore, s as LabeledScenarioWrite, t as LabeledScenarioSampleArgs, u as LabeledScenarioRecord, v as LabelTrust, e as SurfaceProposer, w as ProposedCandidate, x as ProposeContext, y as LabeledScenarioSource, C as CampaignResult, n as CodeSurface } from '../types-BMahhhio.js';
+export { j as CampaignAggregates, k as CampaignArtifactWriter, l as CampaignCellResult, m as CampaignCostMeter, z as CampaignTokenUsage, c as CampaignTraceWriter, D as DispatchFn, g as GateContext, i as GateDecision, G as GateResult, o as GenerationCandidate, A as JudgeAggregate, p as JudgeDimension, h as Mutator, O as OptimizationProposer, q as OptimizerConfig, P as ParetoParent, R as RedactionStatus, B as ScenarioAggregate, r as SessionScript, T as TraceSpan, E as isProposedCandidate, F as labelTrustRank } from '../types-BMahhhio.js';
+import { R as RunCampaignOptions, c as RunImprovementLoopOptions, C as CampaignStorage } from '../gepa-Dprxvz8r.js';
+export { e as GepaProposerConstraints, G as GepaProposerOptions, O as OpenAutoPrOptions, h as OpenAutoPrResult, b as RunImprovementLoopResult, a as RunOptimizationOptions, j as RunOptimizationResult, k as countSentenceEdits, l as defaultRenderDiff, m as extractH2Sections, f as fsCampaignStorage, g as gepaProposer, i as inMemoryCampaignStorage, o as openAutoPr, r as runCampaign, d as runImprovementLoop, n as runOptimization, s as surfaceHash } from '../gepa-Dprxvz8r.js';
+export { A as AxisEvidence, a as AxisVerdict, B as BuildEvidenceVectorOptions, k as BuildLoopProvenanceArgs, D as DefaultProductionGateOptions, l as EmitLoopProvenanceArgs, m as EmitLoopProvenanceResult, E as EvidenceVector, b as EvolutionaryProposerOptions, H as HeldOutGateOptions, n as LoopProvenanceBackend, o as LoopProvenanceCandidate, L as LoopProvenanceRecord, O as ObjectiveSource, P as ParetoSignificanceGateOptions, c as PromotionObjective, d as PromotionPolicy, R as RunEvalOptions, e as buildEvidenceVector, q as buildLoopProvenanceRecord, f as composeGate, g as defaultProductionGate, s as emitLoopProvenance, h as evolutionaryProposer, i as heldOutGate, t as loopProvenanceSpans, p as paretoPolicy, j as paretoSignificanceGate, u as provenanceRecordPath, v as provenanceSpansPath, r as runEval, w as surfaceContentHash } from '../provenance-CIxfBnkl.js';
 import { S as SignedManifest, B as BackendIntegrityReport, C as CompletionRequirement, R as RuntimeEventLike, a as CompletionVerdict, P as ProducedState, b as CorrectnessChecker } from '../pre-registration-nfUdc9EQ.js';
 import { E as EProcessState, a as PairedBootstrapResult } from '../statistics-CCJpTGOS.js';
 import { L as LlmClientOptions } from '../llm-client-Bj7g0rqu.js';

package/dist/{chunk-AQ5WQAIV.js → chunk-3NHEO6ZC.js}

@@ -15,7 +15,7 @@ import {
 // src/trace-analyst/behavioral-metrics.ts
 var INPUT_GROWTH_FACTOR = 3;
 var MIN_TOOL_CALLS = 3;
-var VERIFY_RE = /verif|eval|inspect|check|assert|validat|review|confirm/i;
+var VERIFY_RE = /verif|eval|inspect|check|assert|validat|review|confirm|read|grep|glob|search|view|\blist\b|\bls\b|\bcat\b|\bfind\b|diff|status|\btest|lint|typecheck/i;
 function num(v) {
   return typeof v === "number" && Number.isFinite(v) ? v : null;
 }
@@ -204,4 +204,4 @@ export {
   behavioralAnalyst,
   buildDefaultAnalystRegistry
 };
-//# sourceMappingURL=chunk-AQ5WQAIV.js.map
+//# sourceMappingURL=chunk-3NHEO6ZC.js.map

package/dist/chunk-3NHEO6ZC.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 {\n  LLM_INPUT_TOKEN_ATTR_KEYS,\n  LLM_OUTPUT_TOKEN_ATTR_KEYS,\n  TOOL_NAME_ATTR_KEYS,\n} from '../trace/otlp-attributes'\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 that read or check state count as self-verification, not mutation.\n *  Covers the inspect verbs plus the read/search tools real harnesses use to\n *  verify (Claude Code Read/Grep/Glob, codex read_file/ls/cat, git status/diff,\n *  test/lint). A pure shell tool (Bash/exec_command) is intentionally NOT matched\n *  — its name can't tell a `pytest` from an `rm`. */\nconst VERIFY_RE =\n  /verif|eval|inspect|check|assert|validat|review|confirm|read|grep|glob|search|view|\\blist\\b|\\bls\\b|\\bcat\\b|\\bfind\\b|diff|status|\\btest|lint|typecheck/i\n\nfunction num(v: unknown): number | null {\n  return typeof v === 'number' && Number.isFinite(v) ? v : null\n}\nfunction numAttr(attrs: Record<string, unknown>, keys: readonly string[]): number | null {\n  for (const key of keys) {\n    const value = num(attrs[key])\n    if (value !== null) return value\n  }\n  return null\n}\nfunction inputTokensOf(s: TraceAnalystSpan): number | null {\n  return (\n    numAttr(s.attributes, LLM_INPUT_TOKEN_ATTR_KEYS) ?? num(s.attributes['llm.usage.input_tokens'])\n  )\n}\nfunction outputTokensOf(s: TraceAnalystSpan): number | null {\n  return (\n    numAttr(s.attributes, LLM_OUTPUT_TOKEN_ATTR_KEYS) ??\n    num(s.attributes['llm.usage.output_tokens'])\n  )\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  for (const key of TOOL_NAME_ATTR_KEYS) {\n    const t = s.attributes[key]\n    if (typeof t === 'string' && t.length > 0) return t\n  }\n  return 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":";;;;;;;;;;;;;;;AAiDA,IAAM,sBAAsB;AAE5B,IAAM,iBAAiB;AAMvB,IAAM,YACJ;AAEF,SAAS,IAAI,GAA2B;AACtC,SAAO,OAAO,MAAM,YAAY,OAAO,SAAS,CAAC,IAAI,IAAI;AAC3D;AACA,SAAS,QAAQ,OAAgC,MAAwC;AACvF,aAAW,OAAO,MAAM;AACtB,UAAM,QAAQ,IAAI,MAAM,GAAG,CAAC;AAC5B,QAAI,UAAU,KAAM,QAAO;AAAA,EAC7B;AACA,SAAO;AACT;AACA,SAAS,cAAc,GAAoC;AACzD,SACE,QAAQ,EAAE,YAAY,yBAAyB,KAAK,IAAI,EAAE,WAAW,wBAAwB,CAAC;AAElG;AACA,SAAS,eAAe,GAAoC;AAC1D,SACE,QAAQ,EAAE,YAAY,0BAA0B,KAChD,IAAI,EAAE,WAAW,yBAAyB,CAAC;AAE/C;AACA,SAAS,OAAO,GAAoC;AAClD,SAAO,IAAI,EAAE,WAAW,IAAI;AAC9B;AACA,SAAS,WAAW,GAAoC;AACtD,MAAI,EAAE,UAAW,QAAO,EAAE;AAC1B,aAAW,OAAO,qBAAqB;AACrC,UAAM,IAAI,EAAE,WAAW,GAAG;AAC1B,QAAI,OAAO,MAAM,YAAY,EAAE,SAAS,EAAG,QAAO;AAAA,EACpD;AACA,SAAO;AACT;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;;;AC5KA,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/contract/index.d.ts

@@ -1,9 +1,9 @@
-import { S as Scenario, M as MutableSurface, b as DispatchContext, a as JudgeConfig, c as SurfaceProposer, G as Gate, L as LabeledScenarioStore, C as CampaignResult, d as GateDecision } from '../types-DQRY8ZT-.js';
-export { e as CampaignAggregates, f as CampaignArtifactWriter, g as CampaignCellResult, h as CampaignCostMeter, i as CampaignTraceWriter, j as CodeSurface, D as Dispatch, k as GateContext, l as GateResult, m as GenerationCandidate, n as GenerationRecord, o as JudgeDimension, J as JudgeScore, p as Mutator, O as OptimizationProposer, q as OptimizerConfig, r as SessionScript } from '../types-DQRY8ZT-.js';
-import { L as LoopProvenanceRecord, R as RunEvalOptions } from '../provenance-CncDq9qE.js';
-export { A as AxisEvidence, a as AxisVerdict, B as BuildEvidenceVectorOptions, D as DefaultProductionGateOptions, E as EvidenceVector, b as EvolutionaryProposerOptions, H as HeldOutGateOptions, O as ObjectiveSource, P as ParetoSignificanceGateOptions, c as PromotionObjective, d as PromotionPolicy, e as buildEvidenceVector, f as composeGate, g as defaultProductionGate, h as evolutionaryProposer, i as heldOutGate, p as paretoPolicy, j as paretoSignificanceGate, r as runEval } from '../provenance-CncDq9qE.js';
-import { C as CampaignStorage, R as RunOptimizationOptions, a as RunImprovementLoopResult } from '../gepa-C1NCIZ9o.js';
-export { G as GepaProposerOptions, b as RunCampaignOptions, c as RunImprovementLoopOptions, f as fsCampaignStorage, g as gepaProposer, i as inMemoryCampaignStorage, r as runCampaign, d as runImprovementLoop } from '../gepa-C1NCIZ9o.js';
+import { S as Scenario, M as MutableSurface, b as DispatchContext, a as JudgeConfig, e as SurfaceProposer, f as Gate, L as LabeledScenarioStore, C as CampaignResult, i as GateDecision } from '../types-BMahhhio.js';
+export { j as CampaignAggregates, k as CampaignArtifactWriter, l as CampaignCellResult, m as CampaignCostMeter, c as CampaignTraceWriter, n as CodeSurface, D as Dispatch, g as GateContext, G as GateResult, o as GenerationCandidate, d as GenerationRecord, p as JudgeDimension, J as JudgeScore, h as Mutator, O as OptimizationProposer, q as OptimizerConfig, r as SessionScript } from '../types-BMahhhio.js';
+import { L as LoopProvenanceRecord, R as RunEvalOptions } from '../provenance-CIxfBnkl.js';
+export { A as AxisEvidence, a as AxisVerdict, B as BuildEvidenceVectorOptions, D as DefaultProductionGateOptions, E as EvidenceVector, b as EvolutionaryProposerOptions, H as HeldOutGateOptions, O as ObjectiveSource, P as ParetoSignificanceGateOptions, c as PromotionObjective, d as PromotionPolicy, e as buildEvidenceVector, f as composeGate, g as defaultProductionGate, h as evolutionaryProposer, i as heldOutGate, p as paretoPolicy, j as paretoSignificanceGate, r as runEval } from '../provenance-CIxfBnkl.js';
+import { C as CampaignStorage, a as RunOptimizationOptions, b as RunImprovementLoopResult } from '../gepa-Dprxvz8r.js';
+export { G as GepaProposerOptions, R as RunCampaignOptions, c as RunImprovementLoopOptions, f as fsCampaignStorage, g as gepaProposer, i as inMemoryCampaignStorage, r as runCampaign, d as runImprovementLoop } from '../gepa-Dprxvz8r.js';
 export { D as DeploymentOutcome, F as FileSystemOutcomeStore, a as FileSystemOutcomeStoreOptions, I as InMemoryOutcomeStore, b as OutcomeStore } from '../outcome-store-rnXLEqSn.js';
 import { HostedTenant, EvalRunCellScore, EvalRunGenerationSnapshot, EvalRunEvent, TraceSpanEvent } from '../hosted/index.js';
 import { R as RunRecord, a as RunSplitTag } from '../run-record-CP2ObebC.js';
@@ -11,7 +11,8 @@ import { I as InsightReport } from '../insight-report-BnRjTibG.js';
 export { F as FailureClusterInsight, a as InterRaterInsight, J as JudgeInsight, L as LiftInsight, O as OutcomeCorrelationInsight, R as Recommendation, b as ReleaseSummary, S as ScalarDistribution } from '../insight-report-BnRjTibG.js';
 export { D as DefaultAnalystRegistryOptions, c as buildDefaultAnalystRegistry } from '../default-registry-GyE8X5SP.js';
 export { c as AnalystFinding } from '../types-B5x54y6n.js';
-export { A as AnalyzeRunsOptions, a as analyzeRuns } from '../analyze-runs-DtT6F_6T.js';
+import { A as AnalyzeRunsOptions } from '../analyze-runs-DtT6F_6T.js';
+export { a as analyzeRuns } from '../analyze-runs-DtT6F_6T.js';
 export { C as CodeAgentSessionDiagnostic, a as CodeAgentSessionIntakeOptions, b as CodeAgentSessionIntakeResult, c as CodeAgentSessionMetrics, d as CodeAgentSessionSource, P as ParsedCodeAgentJsonl, f as fromClaudeCodeSession, e as fromCodexSession, g as fromKimiCodeSession, h as fromOpenCodeSession, i as fromPiSession, j as fromPigraphSession, p as parseCodeAgentJsonl } from '../code-agent-session-CPHRCb4-.js';
 import '../red-team-BWdoyleI.js';
 import '../dataset-BbGkaN2I.js';
@@ -358,6 +359,140 @@ interface DefinedAgentEval<TScenario extends Scenario, TArtifact> {
  */
 declare function defineAgentEval<TScenario extends Scenario, TArtifact>(defaults: DefineAgentEvalOptions<TScenario, TArtifact>): DefinedAgentEval<TScenario, TArtifact>;
 
+/**
+ * # `intake/run-record-dir` — load a directory or file of `RunRecord`s.
+ *
+ * The on-disk counterpart to the in-memory intake adapters: point it at a
+ * single `.json` (array) / `.jsonl` (one record per line) file or at a
+ * directory of such files, and it returns the substrate-canonical
+ * `RunRecord[]` ready for `analyzeRuns({ runs })`.
+ *
+ * Validation is at the boundary: each parsed object goes through
+ * `parseRunRecordSafe`. By default an invalid record fails loud with its
+ * file + index; pass `onInvalid: 'collect'` to keep the valid records and
+ * receive the rejects as structured diagnostics instead.
+ */
+
+/** A record that failed boundary validation, with enough context to fix it. */
+interface RunRecordRejection {
+    /** Absolute or caller-relative path to the file the record came from. */
+    file: string;
+    /** Zero-based position within the file (array index or JSONL line number). */
+    index: number;
+    /** The validator's message. */
+    reason: string;
+}
+interface FromRunRecordDirOptions {
+    /**
+     * How to treat a record that fails `parseRunRecordSafe`:
+     *   - `'throw'` (default) — fail loud on the first invalid record.
+     *   - `'collect'` — drop it, keep the rest, and return it under `rejected`.
+     */
+    onInvalid?: 'throw' | 'collect';
+    /**
+     * When the input is a directory, only files matching this predicate are
+     * read. Default: any file ending in `.json` or `.jsonl`. The `analysis.json`
+     * artifact `evalReportingSuite` writes is always skipped so a re-run never
+     * ingests its own output.
+     */
+    include?: (fileName: string) => boolean;
+    /**
+     * Recurse into subdirectories when the input is a directory. Default false —
+     * a flat run directory is the common case and recursion can silently pull in
+     * unrelated corpora.
+     */
+    recursive?: boolean;
+}
+interface FromRunRecordDirResult {
+    /** Records that passed boundary validation, in file-then-index order. */
+    runs: RunRecord[];
+    /** Records that failed validation. Empty unless `onInvalid: 'collect'`. */
+    rejected: RunRecordRejection[];
+    /** The files that were read, in the order they were processed. */
+    files: string[];
+}
+/**
+ * Resolve a file or directory path into validated `RunRecord[]`.
+ *
+ * A `.json` file must parse to a top-level array; a `.jsonl` file is one
+ * record per non-empty line. Directories are read shallowly by default
+ * (set `recursive` to descend); the `analysis.json` output artifact is
+ * always excluded.
+ */
+declare function fromRunRecordDir(path: string, options?: FromRunRecordDirOptions): Promise<FromRunRecordDirResult>;
+
+/**
+ * # `evalReportingSuite` — one call from runs (or a run dir) to `analysis.json`.
+ *
+ * A thin wrapper over the analysis primitive (`analyzeRuns`) and the on-disk
+ * intake adapter (`fromRunRecordDir`). It does NOT reimplement any statistics,
+ * distributions, or clustering — it resolves the input into validated
+ * `RunRecord[]`, calls `analyzeRuns` with the options you'd pass it directly,
+ * wraps the result in a small provenance envelope, and (optionally) writes a
+ * single `analysis.json` artifact.
+ *
+ * ```ts
+ * // From a directory of run files, write ./runs/analysis.json:
+ * const suite = await evalReportingSuite('./runs', { write: true })
+ * // From records already in memory, no write:
+ * const suite = await evalReportingSuite(records, { analyze: { decisionThreshold: 0.03 } })
+ * suite.report // the InsightReport — distributions, paired lift, findings rollup
+ * ```
+ */
+
+/** Either records in hand or a path to a `.json` / `.jsonl` file or a
+ *  directory of them. */
+type EvalReportingSuiteInput = RunRecord[] | string;
+interface EvalReportingSuiteOptions {
+    /** Forwarded verbatim to `analyzeRuns` (everything except `runs`, which the
+     *  suite supplies from the resolved input). Use this for split selection,
+     *  baseline/candidate ids, canaries, prior-period runs, the analyst registry,
+     *  etc. */
+    analyze?: Omit<AnalyzeRunsOptions, 'runs'>;
+    /** Loader options used only when the input is a path. */
+    load?: FromRunRecordDirOptions;
+    /**
+     * Write the suite result as a single `analysis.json`.
+     *   - `true` — write to `<dir>/analysis.json` when the input is a directory,
+     *     or alongside the input file; throws if the input is in-memory records
+     *     (no directory to anchor to — pass an explicit path instead).
+     *   - a string — write to exactly this path (a directory path gets
+     *     `analysis.json` appended; any other path is used verbatim).
+     *   - omitted / false — do not write.
+     */
+    write?: boolean | string;
+}
+/** The suite artifact — the `analyzeRuns` report plus provenance. This is the
+ *  exact shape serialized to `analysis.json`. */
+interface EvalReportingSuiteResult {
+    /** The analysis itself — distributions, paired stats/lift, failure rollup,
+     *  recommendations. Produced by `analyzeRuns`. */
+    report: InsightReport;
+    /** How the suite was run, so a reader can verify provenance. */
+    provenance: {
+        /** ISO timestamp the suite ran. */
+        generatedAt: string;
+        /** Number of records analyzed (mirrors `report.n`). */
+        runCount: number;
+        /** The source path when the input was a directory/file; null for
+         *  in-memory records. */
+        sourcePath: string | null;
+        /** Files read when loading from disk; empty for in-memory input. */
+        files: string[];
+        /** Records dropped at the validation boundary. Always empty unless
+         *  `load.onInvalid` was set to `'collect'`. */
+        rejected: FromRunRecordDirResult['rejected'];
+    };
+    /** The path `analysis.json` was written to, or null when `write` was unset. */
+    writtenTo: string | null;
+}
+/**
+ * Resolve runs (or a run dir/file), run `analyzeRuns`, and optionally persist a
+ * single `analysis.json`. The only analysis logic lives in `analyzeRuns`; this
+ * function is composition + I/O.
+ */
+declare function evalReportingSuite(input: EvalReportingSuiteInput, options?: EvalReportingSuiteOptions): Promise<EvalReportingSuiteResult>;
+
 /**
  * # `@tangle-network/agent-eval/contract` — eval-run diff primitive.
  *
@@ -687,4 +822,4 @@ interface FromOtelSpansOptions {
 }
 declare function fromOtelSpans(opts: FromOtelSpansOptions): RunRecord[];
 
-export { type AgentEvalAgent, type AgentEvalEvaluateOptions, type AgentEvalImproveOptions, type AgentTraceContributor, type AgentTraceContributorType, type AgentTraceConversation, type AgentTraceFile, type AgentTraceIndex, type AgentTraceRange, type AgentTraceRecord, type AuthoringProvenance, CampaignResult, CampaignStorage, type DefineAgentEvalOptions, type DefinedAgentEval, DispatchContext, type EvalCellScoreDelta, type EvalDimensionDelta, type EvalGenerationDiff, type EvalRunDiff, type FeedbackTableMeta, type FeedbackTableRow, type FromFeedbackTableOptions, type FromFeedbackTableResult, type FromOtelSpansOptions, Gate, GateDecision, HostedTenant, InsightReport, JudgeConfig, MutableSurface, type PartitionByAuthoringModelResult, RunEvalOptions, RunImprovementLoopResult, Scenario, type SelfImproveBudget, type SelfImproveLlm, type SelfImproveOptions, type SelfImproveProgressEvent, type SelfImproveResult, SurfaceProposer, defineAgentEval, diffGenerations, diffRunBaselineToWinner, diffRuns, fromFeedbackTable, fromOtelSpans, parseAgentTrace, partitionRunsByAuthoringModel, selfImprove };
+export { type AgentEvalAgent, type AgentEvalEvaluateOptions, type AgentEvalImproveOptions, type AgentTraceContributor, type AgentTraceContributorType, type AgentTraceConversation, type AgentTraceFile, type AgentTraceIndex, type AgentTraceRange, type AgentTraceRecord, AnalyzeRunsOptions, type AuthoringProvenance, CampaignResult, CampaignStorage, type DefineAgentEvalOptions, type DefinedAgentEval, DispatchContext, type EvalCellScoreDelta, type EvalDimensionDelta, type EvalGenerationDiff, type EvalReportingSuiteInput, type EvalReportingSuiteOptions, type EvalReportingSuiteResult, type EvalRunDiff, type FeedbackTableMeta, type FeedbackTableRow, type FromFeedbackTableOptions, type FromFeedbackTableResult, type FromOtelSpansOptions, type FromRunRecordDirOptions, type FromRunRecordDirResult, Gate, GateDecision, HostedTenant, InsightReport, JudgeConfig, MutableSurface, type PartitionByAuthoringModelResult, RunEvalOptions, RunImprovementLoopResult, type RunRecordRejection, Scenario, type SelfImproveBudget, type SelfImproveLlm, type SelfImproveOptions, type SelfImproveProgressEvent, type SelfImproveResult, SurfaceProposer, defineAgentEval, diffGenerations, diffRunBaselineToWinner, diffRuns, evalReportingSuite, fromFeedbackTable, fromOtelSpans, fromRunRecordDir, parseAgentTrace, partitionRunsByAuthoringModel, selfImprove };

package/dist/contract/index.js

@@ -37,7 +37,7 @@ import {
 import "../chunk-VI2UW6B6.js";
 import {
   buildDefaultAnalystRegistry
-} from "../chunk-AQ5WQAIV.js";
+} from "../chunk-3NHEO6ZC.js";
 import "../chunk-S4SYLDFX.js";
 import "../chunk-5HRORJQY.js";
 import {
@@ -57,6 +57,9 @@ import {
   LLM_OUTPUT_TOKEN_ATTR_KEYS
 } from "../chunk-YBIGNSCZ.js";
 import "../chunk-PC4UYEBM.js";
+import {
+  parseRunRecordSafe
+} from "../chunk-LO6IOIJ2.js";
 import "../chunk-VSMTAMNK.js";
 import "../chunk-3BFEG2F6.js";
 import "../chunk-PZ5AY32C.js";
@@ -474,6 +477,145 @@ function requirePositiveInteger(value, field) {
   return value;
 }
 
+// src/contract/eval-reporting-suite.ts
+import { mkdir, writeFile } from "fs/promises";
+import { dirname, join as join2 } from "path";
+
+// src/contract/intake/run-record-dir.ts
+import { readdir, readFile, stat } from "fs/promises";
+import { join } from "path";
+var ANALYSIS_ARTIFACT = "analysis.json";
+function defaultInclude(fileName) {
+  if (fileName === ANALYSIS_ARTIFACT) return false;
+  return fileName.endsWith(".json") || fileName.endsWith(".jsonl");
+}
+async function fromRunRecordDir(path, options = {}) {
+  const onInvalid = options.onInvalid ?? "throw";
+  const include = options.include ?? defaultInclude;
+  const stats = await stat(path);
+  const filePaths = stats.isDirectory() ? await collectFiles(path, include, options.recursive ?? false) : [path];
+  const runs = [];
+  const rejected = [];
+  for (const file of filePaths) {
+    const raw = await parseRecordFile(file);
+    for (const { index, value } of raw) {
+      const parsed = parseRunRecordSafe(value);
+      if (parsed.ok) {
+        runs.push(parsed.value);
+        continue;
+      }
+      const rejection = { file, index, reason: parsed.error.message };
+      if (onInvalid === "throw") {
+        throw new Error(
+          `fromRunRecordDir: invalid RunRecord in '${file}' at index ${index}: ${parsed.error.message}`
+        );
+      }
+      rejected.push(rejection);
+    }
+  }
+  return { runs, rejected, files: filePaths };
+}
+async function parseRecordFile(file) {
+  const text = await readFile(file, "utf8");
+  const trimmed = text.trim();
+  if (trimmed.length === 0) return [];
+  if (trimmed.startsWith("[")) {
+    const parsed = JSON.parse(trimmed);
+    if (!Array.isArray(parsed)) {
+      throw new Error(`fromRunRecordDir: file '${file}' did not parse to an array`);
+    }
+    return parsed.map((value, index) => ({ index, value }));
+  }
+  const out = [];
+  const lines = trimmed.split("\n");
+  for (let i = 0; i < lines.length; i++) {
+    const line = lines[i].trim();
+    if (line.length === 0) continue;
+    try {
+      out.push({ index: i, value: JSON.parse(line) });
+    } catch (err) {
+      throw new Error(
+        `fromRunRecordDir: file '${file}' line ${i + 1} is not valid JSON: ${err instanceof Error ? err.message : String(err)}`
+      );
+    }
+  }
+  return out;
+}
+async function collectFiles(dir, include, recursive) {
+  const entries = await readdir(dir, { withFileTypes: true });
+  const files = [];
+  const subdirs = [];
+  for (const entry of entries) {
+    if (entry.isDirectory()) {
+      if (recursive) subdirs.push(join(dir, entry.name));
+      continue;
+    }
+    if (include(entry.name)) files.push(join(dir, entry.name));
+  }
+  files.sort();
+  subdirs.sort();
+  for (const sub of subdirs) {
+    files.push(...await collectFiles(sub, include, recursive));
+  }
+  return files;
+}
+
+// src/contract/eval-reporting-suite.ts
+var ANALYSIS_ARTIFACT2 = "analysis.json";
+async function evalReportingSuite(input, options = {}) {
+  const fromPath = typeof input === "string";
+  let runs;
+  let files = [];
+  let rejected = [];
+  if (fromPath) {
+    const loaded = await fromRunRecordDir(input, options.load);
+    runs = loaded.runs;
+    files = loaded.files;
+    rejected = loaded.rejected;
+  } else {
+    runs = input;
+  }
+  if (runs.length === 0) {
+    throw new Error(
+      fromPath ? `evalReportingSuite: no RunRecords found at '${input}'` : "evalReportingSuite: no RunRecords to analyze"
+    );
+  }
+  const report = await analyzeRuns({ ...options.analyze, runs });
+  const result = {
+    report,
+    provenance: {
+      generatedAt: (/* @__PURE__ */ new Date()).toISOString(),
+      runCount: runs.length,
+      sourcePath: fromPath ? input : null,
+      files,
+      rejected
+    },
+    writtenTo: null
+  };
+  const target = resolveWriteTarget(options.write, fromPath ? input : null);
+  if (target) {
+    await mkdir(dirname(target), { recursive: true });
+    await writeFile(target, `${JSON.stringify(result, null, 2)}
+`, "utf8");
+    result.writtenTo = target;
+  }
+  return result;
+}
+function resolveWriteTarget(write, sourcePath) {
+  if (!write) return null;
+  if (typeof write === "string") {
+    const looksLikeDir = write.endsWith("/") || !write.endsWith(".json") && !write.endsWith(".jsonl");
+    return looksLikeDir ? join2(write, ANALYSIS_ARTIFACT2) : write;
+  }
+  if (sourcePath === null) {
+    throw new Error(
+      "evalReportingSuite: write:true needs a source path to anchor analysis.json \u2014 pass an explicit output path when analyzing in-memory records"
+    );
+  }
+  const isFile = sourcePath.endsWith(".json") || sourcePath.endsWith(".jsonl");
+  return isFile ? join2(dirname(sourcePath), ANALYSIS_ARTIFACT2) : join2(sourcePath, ANALYSIS_ARTIFACT2);
+}
+
 // src/contract/diff.ts
 function keyForCell(cell) {
   return JSON.stringify([cell.scenarioId, cell.rep]);
@@ -833,6 +975,7 @@ export {
   diffGenerations,
   diffRunBaselineToWinner,
   diffRuns,
+  evalReportingSuite,
   evolutionaryProposer,
   fromClaudeCodeSession,
   fromCodexSession,
@@ -842,6 +985,7 @@ export {
   fromOtelSpans,
   fromPiSession,
   fromPigraphSession,
+  fromRunRecordDir,
   fsCampaignStorage,
   gepaProposer,
   heldOutGate,

package/dist/contract/index.js.map

@@ -1 +1 @@
-{"version":3,"sources":["../../src/contract/self-improve.ts","../../src/contract/define-agent-eval.ts","../../src/contract/diff.ts","../../src/contract/intake/agent-trace.ts","../../src/contract/intake/feedback-table.ts","../../src/contract/intake/otel-spans.ts"],"sourcesContent":["/**\n * # `selfImprove()` - the one-call improvement loop.\n *\n * The cheapest possible call site to run a real closed-loop self-\n * improvement over your agent. Wraps `runImprovementLoop` with smart\n * defaults and a budget-shaped options API; every escape hatch the\n * substrate exposes is reachable from here without losing the\n * one-function feel.\n *\n * Defaults:\n *   - In-memory storage (no filesystem touch).\n *   - `gepaProposer` reflective mutation with copywriting-flavored primitives\n *     (override `proposer` or `mutationPrimitives` for any domain).\n *   - `defaultProductionGate` with `deltaThreshold: 0.05`.\n *   - Held-out split = 25% of scenarios, deterministic by id hash.\n *   - 3 generations × population 2 (raise via `budget` for more search).\n *   - `autoOnPromote: 'none'` (we don't open PRs unless you ask).\n *\n * Want one-click? Provide `agent` + `scenarios` + `judge`. Done.\n * Want distributed? Pass `cellPlacement` + an `httpDispatch`-backed\n * agent. Want a code-tier surface? Pass a `MutableSurface` + your own\n * `proposer`. Same function.\n */\n\nimport { createHash } from 'node:crypto'\nimport { defaultProductionGate } from '../campaign/gates/default-production-gate'\nimport {\n  type RunImprovementLoopResult,\n  runImprovementLoop,\n} from '../campaign/presets/run-improvement-loop'\nimport type { RunOptimizationOptions } from '../campaign/presets/run-optimization'\nimport { gepaProposer } from '../campaign/proposers/gepa'\nimport {\n  emitLoopProvenance,\n  type LoopProvenanceRecord,\n  surfaceContentHash,\n} from '../campaign/provenance'\nimport {\n  type CampaignStorage,\n  fsCampaignStorage,\n  inMemoryCampaignStorage,\n} from '../campaign/storage'\nimport type {\n  CampaignCellResult,\n  DispatchContext,\n  Gate,\n  JudgeConfig,\n  LabeledScenarioStore,\n  MutableSurface,\n  Scenario,\n  SurfaceProposer,\n} from '../campaign/types'\nimport { createHostedClient, type HostedTenant } from '../hosted/client'\nimport type { EvalRunCellScore, EvalRunEvent, EvalRunGenerationSnapshot } from '../hosted/types'\nimport type { JudgeScoresRecord, RunRecord } from '../run-record'\nimport { analyzeRuns } from './analyze-runs'\nimport type { InsightReport } from './insight-report'\n\nexport interface SelfImproveBudget {\n  /** Hard $ ceiling across all cells in baseline + every generation. Cells\n   *  beyond the ceiling are skipped (cost-aware, not aborted). */\n  dollars?: number\n  /** How many improvement generations to explore. Default 3. Set 0 to\n   *  skip improvement entirely (selfImprove becomes a baseline-only run). */\n  generations?: number\n  /** Candidates the proposer emits per generation. Default 2. */\n  populationSize?: number\n  /** Max concurrent cells across the loop. Default 2. */\n  maxConcurrency?: number\n  /** Fraction of `scenarios` held out from training, used for the gate.\n   *  Default 0.25. Ignored when `holdoutScenarios` is set explicitly. */\n  holdoutFraction?: number\n  /** Explicit held-out scenarios; overrides `holdoutFraction`. */\n  holdoutScenarios?: Scenario[]\n  /** Per-scenario replicates per cell — raises bootstrap-CI tightness. Default 1. */\n  reps?: number\n  /** Top-scoring candidates carried into the next generation. Default 2. */\n  promoteTopK?: number\n}\n\nexport interface SelfImproveLlm {\n  /** Endpoint base URL. Default Tangle Router. */\n  baseUrl?: string\n  /** Bearer token. Default `process.env.OPENAI_API_KEY`. */\n  apiKey?: string\n  /** Model id used by `gepaProposer` reflection. Default\n   *  `anthropic/claude-sonnet-4.6`. */\n  model?: string\n}\n\nexport type SelfImproveProgressEvent =\n  | { kind: 'baseline.started'; scenarios: number }\n  | { kind: 'baseline.completed'; compositeMean: number; durationMs: number }\n  | { kind: 'generation.started'; index: number; populationSize: number }\n  | { kind: 'generation.completed'; index: number; bestComposite: number; durationMs: number }\n  | { kind: 'gate.decided'; decision: string; lift: number }\n\nexport interface SelfImproveOptions<TScenario extends Scenario, TArtifact> {\n  /**\n   * Your agent — a function that takes the current `MutableSurface`\n   * (typically a system prompt the loop is optimizing) plus the\n   * scenario + cell ctx, and returns the artifact your judge scores.\n   *\n   * Same shape as `RunOptimizationOptions.dispatchWithSurface`. Wrap a\n   * plain `Dispatch` if you don't have a surface seam:\n   *\n   *   agent: (_surface, scenario, ctx) => yourPlainDispatch(scenario, ctx)\n   *\n   * That mode evaluates without mutating any surface — useful as a\n   * baseline-only run (set `budget.generations = 0`).\n   */\n  agent: (surface: MutableSurface, scenario: TScenario, ctx: DispatchContext) => Promise<TArtifact>\n\n  /** Scenarios to evaluate against. Train/holdout split is computed from\n   *  these unless `budget.holdoutScenarios` is set explicitly. */\n  scenarios: TScenario[]\n\n  /** Judge that scores artifacts. Bring your own; use `langchainJudge`\n   *  from `/adapters/langchain` for a Runnable-shaped one. */\n  judge: JudgeConfig<TArtifact, TScenario>\n\n  /** Starting surface — system prompt, JSON config, anything `MutableSurface`\n   *  accepts. The proposer mutates this each generation. */\n  baselineSurface: MutableSurface\n\n  /** Budget + loop shape. All fields optional. */\n  budget?: SelfImproveBudget\n\n  /** Custom surface proposer. Default is `gepaProposer` configured from `llm` +\n   *  `mutationPrimitives`. */\n  proposer?: SurfaceProposer\n\n  /** Default-proposer overrides — used when `proposer` is unset. */\n  mutationPrimitives?: string[]\n  proposerTarget?: string\n\n  /** Custom gate. Default is `defaultProductionGate` with\n   *  `deltaThreshold: 0.05` on the held-out split. */\n  gate?: Gate<TArtifact, TScenario>\n\n  /** LLM config consumed by the default `gepaProposer`. Ignored if you pass\n   *  your own `proposer`. */\n  llm?: SelfImproveLlm\n\n  /** Storage backend. Default is DURABLE: when a real (non-`mem://`) `runDir`\n   *  is available, the substrate defaults to `fsCampaignStorage()` so the\n   *  provenance record + OTel spans survive the call. Pass\n   *  `inMemoryCampaignStorage()` explicitly to opt OUT (tests, edge runtimes).\n   *  Default when `runDir` is `mem://...` (or unset): in-memory. */\n  storage?: CampaignStorage\n\n  /** Run directory (logical for in-memory storage, real path for fs).\n   *  Default `mem://selfImprove-<timestamp>` (in-memory, non-durable). Pass a\n   *  real path to persist the provenance record + spans. */\n  runDir?: string\n\n  /**\n   * Worker call records for backend provenance. The agent is opaque to the\n   * substrate (it returns an artifact, not token usage), so to capture an\n   * `assertRealBackend`-grade verdict + worker call count + model in the\n   * provenance record, the agent reports its per-call `RunRecord`s here.\n   * Called once after the loop; return the records the agent accumulated.\n   * When unset, backend provenance is derived from campaign cells (cost only;\n   * verdict will read `stub` without token usage — the honest signal that no\n   * token channel was wired).\n   */\n  collectWorkerRecords?: () => RunRecord[]\n\n  /** Fires once the durable provenance record + OTel spans are emitted.\n   *  Receives the structured record for inline assertions / custom routing. */\n  onProvenance?: (record: LoopProvenanceRecord) => void\n\n  /** Distributed execution seam — same as `RunCampaignOptions.cellPlacement`.\n   *  Returns an opaque placement key the substrate forwards to your agent\n   *  as `ctx.placement`. Combined with `httpDispatch` from\n   *  `/adapters/http`, fans cells across regions. */\n  cellPlacement?: (input: {\n    scenario: TScenario\n    rep: number\n    generation?: number\n  }) => string | undefined\n\n  /** Streaming hook — fires on baseline + each generation + gate decision.\n   *  Consumer routes events wherever (UI, dashboard, logs). */\n  onProgress?: (event: SelfImproveProgressEvent) => void\n\n  /** Auto-promotion behavior on a ship decision. Default `'none'` — we\n   *  return the winner; you ship it however you ship. `'pr'` opens a\n   *  GitHub PR via `openAutoPr`; requires `ghOwner` + `ghRepo`. */\n  autoOnPromote?: 'pr' | 'none'\n  ghOwner?: string\n  ghRepo?: string\n\n  /**\n   * Opt-in: ship eval-run events to a hosted orchestrator (ours, your\n   * self-hosted one, or any compatible implementation of the\n   * `docs/hosted-ingest-spec.md` wire format). When set, the substrate\n   * POSTs the final `EvalRunEvent` to `${endpoint}/v1/ingest/eval-runs`\n   * after the loop completes. Failures are logged but do not fail the\n   * loop — local result is always returned.\n   *\n   * For our orchestrator: `{ endpoint: 'https://orchestrator.tangle.tools/v1', apiKey, tenantId }`.\n   *\n   * For your self-hosted: any URL serving the wire format. See\n   * `examples/hosted-ingest-server/` for the reference receiver.\n   */\n  hostedTenant?: HostedTenant\n\n  /** Free-form labels attached to the hosted event (env, branch, model id,\n   *  etc.). Ignored when `hostedTenant` is unset. */\n  hostedLabels?: Record<string, string>\n\n  /** Capture every artifact + judge score to this store (labeled-example\n   *  corpus the proposer may read for few-shot, and the dataset you ship). Pass\n   *  `'off'` to disable. Default: off. */\n  labeledStore?: LabeledScenarioStore | 'off'\n\n  /** Capture-source tag for `labeledStore`. Default `'eval-run'`. */\n  captureSource?: 'production-trace' | 'eval-run' | 'manual' | 'red-team' | 'synthetic'\n\n  /**\n   * Per-cell backend-integrity expectation — the fail-loud guard. A cell that\n   * produced an artifact but reported `costUsd === 0` AND zero tokens is a\n   * stub. Modes: `'assert'` throws on the first such cell, `'warn'` logs it,\n   * `'off'` skips the check (offline/replay). Default `'assert'` — `selfImprove`\n   * is the real-run path, so a stub fails loud rather than scoring a clean 0.\n   */\n  expectUsage?: 'assert' | 'warn' | 'off'\n\n  /**\n   * Per-generation findings producer. After each\n   * generation is scored, this is called; whatever it returns REPLACES the\n   * proposer's `findings` for the next generation's `propose()`. Plug a\n   * trace-analyst registry / HALO here. When absent, findings stay\n   * `opts.findings`.\n   */\n  analyzeGeneration?: RunOptimizationOptions<TScenario, TArtifact>['analyzeGeneration']\n\n  /** Static findings forwarded to the proposer's `propose()` as `ctx.findings`\n   *  (a findings-grounded proposer consumes them). Default: none. */\n  findings?: unknown[]\n}\n\nexport interface SelfImproveResult<TScenario extends Scenario, TArtifact> {\n  /** Composite mean across all scenarios, baseline run. */\n  baseline: {\n    compositeMean: number\n    perScenario: Record<string, number>\n  }\n  /** Composite mean on the held-out set, winner run. */\n  winner: {\n    compositeMean: number\n    perScenario: Record<string, number>\n    surface: MutableSurface\n    /** Proposer label for the promoted change. Absent ⇒ winner == baseline or\n     *  a bare-surface mutator. */\n    label?: string\n    /** Proposer rationale — the \"because Z\" that motivated the promoted change.\n     *  Threaded from the proposer's `ProposedCandidate` through the loop.\n     *  Absent ⇒ winner == baseline. */\n    rationale?: string\n  }\n  /** `winner.compositeMean - baselineOnHoldout.compositeMean`. Positive\n   *  means the gate observed improvement. */\n  lift: number\n  /** The explicit baseline→winner unified diff. Always present (empty string\n   *  when winner == baseline). */\n  diff: string\n  /** Durable, queryable provenance record: candidate→cell→gate→promote chain +\n   *  rationale + diff + backend provenance. The artifact the hosted ingest\n   *  path stores; the +lift RECOMPUTES from `record.heldOutLift`. */\n  provenance: LoopProvenanceRecord\n  /** `defaultProductionGate.decide()` result. */\n  gateDecision: 'ship' | 'hold' | 'need_more_work' | 'model_ceiling' | 'arch_ceiling'\n  /** Number of generations actually explored (may be less than the\n   *  budget if the proposer gave up early). */\n  generationsExplored: number\n  /** Wall-clock total. */\n  durationMs: number\n  /** Total cost across baseline + every generation. */\n  totalCostUsd: number\n  /**\n   * Rigor packet: distributional summary, paired-bootstrap lift CI,\n   * judge stats, contamination check, recommendations. Wired through\n   * `analyzeRuns()` on the baseline + winner cells of the campaign.\n   * Hosted-tier dashboards render this as the v3-vs-v4 decision view.\n   */\n  insight: InsightReport\n  /**\n   * Raw substrate result for advanced inspection — full per-generation\n   * candidates, full campaign artifacts, all judge scores. Useful for\n   * debugging or reporting beyond the summary.\n   */\n  raw: RunImprovementLoopResult<TArtifact, TScenario>\n}\n\n/**\n * Deterministic train/holdout split by a stable hash of `scenario.id`,\n * so the same scenario set always splits the same way across runs.\n */\nfunction splitTrainHoldout<TScenario extends Scenario>(\n  scenarios: TScenario[],\n  fraction: number,\n): { train: TScenario[]; holdout: TScenario[] } {\n  // Stable fnv-1a-ish hash of the id for ordering.\n  function hash(s: string): number {\n    let h = 2166136261 >>> 0\n    for (let i = 0; i < s.length; i++) {\n      h ^= s.charCodeAt(i)\n      h = Math.imul(h, 16777619) >>> 0\n    }\n    return h\n  }\n  const sorted = [...scenarios].sort((a, b) => hash(a.id) - hash(b.id))\n  const nHoldout = Math.max(1, Math.min(sorted.length - 1, Math.round(sorted.length * fraction)))\n  return {\n    holdout: sorted.slice(0, nHoldout),\n    train: sorted.slice(nHoldout),\n  }\n}\n\nfunction meanComposite(byScenario: Record<string, { meanComposite: number }>): {\n  compositeMean: number\n  perScenario: Record<string, number>\n} {\n  const perScenario: Record<string, number> = {}\n  const values: number[] = []\n  for (const [id, agg] of Object.entries(byScenario)) {\n    perScenario[id] = agg.meanComposite\n    values.push(agg.meanComposite)\n  }\n  return {\n    compositeMean: values.length === 0 ? 0 : values.reduce((s, v) => s + v, 0) / values.length,\n    perScenario,\n  }\n}\n\nconst DEFAULT_MUTATION_PRIMITIVES = [\n  'Tighten the hook: lead with the specific user outcome.',\n  'Replace generic adjectives with specific verbs or proof numbers.',\n  \"Anchor every claim in something the scenario's brief literally supports.\",\n  'Honor the surface-shape constraint (length, register, audience vocabulary).',\n]\n\n/**\n * One-shot self-improvement loop. See module docstring for defaults +\n * extension points.\n *\n * @example Minimum:\n *\n *   const result = await selfImprove({\n *     agent: (surface, scenario, ctx) => myAgent(surface, scenario, ctx.signal),\n *     scenarios,\n *     judge,\n *     baselineSurface: DEFAULT_PROMPT,\n *   })\n *   console.log(`lift: ${result.lift.toFixed(3)} (${result.gateDecision})`)\n *\n * @example Distributed (workers in three regions):\n *\n *   await selfImprove({\n *     agent: httpDispatch({ resolveUrl: ({ placement }) => REGION_URLS[placement!] }),\n *     scenarios,\n *     judge,\n *     baselineSurface: DEFAULT_PROMPT,\n *     cellPlacement: ({ scenario }) => scenario.region,\n *     budget: { maxConcurrency: 12 },\n *   })\n */\nexport async function selfImprove<TScenario extends Scenario, TArtifact>(\n  opts: SelfImproveOptions<TScenario, TArtifact>,\n): Promise<SelfImproveResult<TScenario, TArtifact>> {\n  const startedAt = Date.now()\n\n  const budget = opts.budget ?? {}\n  const generations = budget.generations ?? 3\n  const populationSize = budget.populationSize ?? 2\n  const maxConcurrency = budget.maxConcurrency ?? 2\n  const holdoutFraction = budget.holdoutFraction ?? 0.25\n  const costCeiling = budget.dollars\n  const expectUsage = opts.expectUsage ?? 'assert'\n\n  const explicitHoldout = budget.holdoutScenarios\n  const { train, holdout } = explicitHoldout\n    ? {\n        train: opts.scenarios.filter((s) => !explicitHoldout.some((h) => h.id === s.id)),\n        holdout: explicitHoldout as TScenario[],\n      }\n    : splitTrainHoldout(opts.scenarios, holdoutFraction)\n\n  if (train.length === 0) {\n    throw new Error(\n      'selfImprove: train split is empty. Reduce holdoutFraction or pass more scenarios.',\n    )\n  }\n  if (holdout.length === 0) {\n    throw new Error('selfImprove: holdout split is empty. Pass more scenarios.')\n  }\n\n  const proposer: SurfaceProposer =\n    opts.proposer ??\n    gepaProposer({\n      llm: {\n        baseUrl: opts.llm?.baseUrl ?? 'https://router.tangle.tools/v1',\n        apiKey: opts.llm?.apiKey ?? process.env.OPENAI_API_KEY ?? '',\n      },\n      model: opts.llm?.model ?? 'anthropic/claude-sonnet-4.6',\n      target:\n        opts.proposerTarget ??\n        'agent surface (system prompt or config) being optimized by selfImprove',\n      mutationPrimitives: opts.mutationPrimitives ?? DEFAULT_MUTATION_PRIMITIVES,\n    })\n\n  const gate: Gate<TArtifact, TScenario> =\n    opts.gate ??\n    defaultProductionGate<TArtifact, TScenario>({\n      holdoutScenarios: holdout,\n      deltaThreshold: 0.05,\n    })\n\n  // Durable by default: a real (non-`mem://`) runDir means the caller wants\n  // persistence, so default to fs storage — the provenance record + spans\n  // survive the call. A `mem://` runDir (or none) stays in-memory. An explicit\n  // `storage` always wins (the opt-out path for tests / edge runtimes).\n  const runDir = opts.runDir ?? `mem://selfImprove-${startedAt}`\n  const isMemRunDir = runDir.startsWith('mem://')\n  const storage = opts.storage ?? (isMemRunDir ? inMemoryCampaignStorage() : fsCampaignStorage())\n\n  if (opts.onProgress) {\n    opts.onProgress({ kind: 'baseline.started', scenarios: opts.scenarios.length })\n  }\n\n  const result = await runImprovementLoop<TScenario, TArtifact>({\n    scenarios: train,\n    baselineSurface: opts.baselineSurface,\n    dispatchWithSurface: opts.agent,\n    proposer,\n    judges: [opts.judge],\n    populationSize,\n    maxGenerations: generations,\n    promoteTopK: budget.promoteTopK,\n    reps: budget.reps,\n    holdoutScenarios: holdout,\n    gate,\n    autoOnPromote: opts.autoOnPromote ?? 'none',\n    ghOwner: opts.ghOwner,\n    ghRepo: opts.ghRepo,\n    storage,\n    runDir,\n    maxConcurrency,\n    cellPlacement: opts.cellPlacement,\n    costCeiling,\n    expectUsage,\n    labeledStore: opts.labeledStore,\n    captureSource: opts.captureSource,\n    analyzeGeneration: opts.analyzeGeneration,\n    findings: opts.findings,\n  })\n\n  const baseline = meanComposite(result.baselineOnHoldout.aggregates.byScenario)\n  const winnerStats = meanComposite(result.winnerOnHoldout.aggregates.byScenario)\n\n  if (opts.onProgress) {\n    opts.onProgress({\n      kind: 'baseline.completed',\n      compositeMean: baseline.compositeMean,\n      durationMs: Date.now() - startedAt,\n    })\n    opts.onProgress({\n      kind: 'gate.decided',\n      decision: result.gateResult.decision,\n      lift: winnerStats.compositeMean - baseline.compositeMean,\n    })\n  }\n\n  const totalCost =\n    result.baselineCampaign.aggregates.totalCostUsd +\n    result.generations.reduce(\n      (sum, gen) =>\n        sum + gen.surfaces.reduce((s, sf) => s + sf.campaign.aggregates.totalCostUsd, 0),\n      0,\n    )\n\n  // Rigor packet: feed baseline + winner cells through analyzeRuns().\n  // The two candidates (`baseline` / `winner`) give the lift section a\n  // clean paired comparison; per-judge / per-dimension / cost-quality\n  // sections populate from the cells' judgeScores.\n  const insight = await analyzeRuns({\n    runs: [\n      ...cellsToRunRecords(result.baselineCampaign.cells, 'baseline', runDir, opts.baselineSurface),\n      ...cellsToRunRecords(result.winnerOnHoldout.cells, 'winner', runDir, result.winnerSurface),\n    ],\n    baselineCandidateId: 'baseline',\n    candidateCandidateId: 'winner',\n  })\n\n  // ── Durable provenance: candidate→cell→gate→promote chain + rationale +\n  // diff + backend provenance. Always emitted; the +lift recomputes from it.\n  const durationMs = Date.now() - startedAt\n  const workerRecords =\n    opts.collectWorkerRecords?.() ??\n    cellsToRunRecords(result.winnerOnHoldout.cells, 'winner', runDir, result.winnerSurface)\n  const { record: provenance } = await emitLoopProvenance<TArtifact, TScenario>({\n    runId: `${runDir}#${startedAt}`,\n    runDir,\n    timestamp: new Date(startedAt).toISOString(),\n    baselineSurface: opts.baselineSurface,\n    winnerSurface: result.winnerSurface,\n    winnerLabel: result.winnerLabel,\n    winnerRationale: result.winnerRationale,\n    diff: result.promotedDiff,\n    generations: result.generations.map((g) => ({\n      generationIndex: g.record.generationIndex,\n      candidates: g.record.candidates,\n      promoted: g.record.promoted,\n      surfaces: g.surfaces.map((s) => ({ surfaceHash: s.surfaceHash, surface: s.surface })),\n    })),\n    gate: result.gateResult,\n    baselineOnHoldout: result.baselineOnHoldout,\n    winnerOnHoldout: result.winnerOnHoldout,\n    workerRecords,\n    totalCostUsd: totalCost,\n    totalDurationMs: durationMs,\n    storage,\n    hostedClient: opts.hostedTenant ? createHostedClient(opts.hostedTenant) : undefined,\n  })\n  if (opts.onProvenance) opts.onProvenance(provenance)\n\n  const summary: SelfImproveResult<TScenario, TArtifact> = {\n    baseline,\n    winner: {\n      ...winnerStats,\n      surface: result.winnerSurface,\n      ...(result.winnerLabel ? { label: result.winnerLabel } : {}),\n      ...(result.winnerRationale ? { rationale: result.winnerRationale } : {}),\n    },\n    lift: winnerStats.compositeMean - baseline.compositeMean,\n    diff: result.promotedDiff,\n    provenance,\n    gateDecision: result.gateResult.decision,\n    generationsExplored: result.generations.length,\n    durationMs,\n    totalCostUsd: totalCost,\n    insight,\n    raw: result,\n  }\n\n  // Opt-in hosted ingest. Failures are logged but never fail the loop: the\n  // local result is always returned.\n  if (opts.hostedTenant) {\n    try {\n      await shipEvalRunToHosted(opts.hostedTenant, opts, summary, result, runDir)\n    } catch (err) {\n      const msg = err instanceof Error ? err.message : String(err)\n      // eslint-disable-next-line no-console -- intentional: hosted-ingest is best-effort\n      console.warn(`[agent-eval] hosted ingest failed (continuing): ${msg}`)\n    }\n  }\n\n  return summary\n}\n\nasync function shipEvalRunToHosted<TScenario extends Scenario, TArtifact>(\n  tenant: HostedTenant,\n  opts: SelfImproveOptions<TScenario, TArtifact>,\n  summary: SelfImproveResult<TScenario, TArtifact>,\n  raw: RunImprovementLoopResult<TArtifact, TScenario>,\n  runDir: string,\n): Promise<void> {\n  const client = createHostedClient(tenant)\n\n  function snapshotFromCampaign(\n    index: number,\n    surface: MutableSurface | undefined,\n    campaign: RunImprovementLoopResult<TArtifact, TScenario>['baselineCampaign'],\n    durationMs: number,\n  ): EvalRunGenerationSnapshot {\n    const cells: EvalRunCellScore[] = campaign.cells.map((cell) => {\n      const judgeScores = Object.values(cell.judgeScores)\n      const composite =\n        judgeScores.length === 0\n          ? 0\n          : judgeScores.reduce((s, j) => s + j.composite, 0) / judgeScores.length\n      return {\n        scenarioId: cell.scenarioId,\n        rep: cell.rep,\n        compositeMean: composite,\n        dimensions: Object.fromEntries(\n          Object.entries(cell.judgeScores).map(([name, score]) => [name, score.dimensions]),\n        ),\n        errorMessage: cell.error ?? undefined,\n      }\n    })\n    const compositeMean =\n      cells.length === 0 ? 0 : cells.reduce((s, c) => s + c.compositeMean, 0) / cells.length\n    return {\n      index,\n      surfaceHash:\n        typeof surface === 'string'\n          ? hashString(surface)\n          : hashString(JSON.stringify(surface ?? '')),\n      surface,\n      cells,\n      compositeMean,\n      costUsd: campaign.aggregates.totalCostUsd,\n      durationMs,\n    }\n  }\n\n  const generations: EvalRunGenerationSnapshot[] = []\n  // Baseline as generation 0.\n  generations.push(snapshotFromCampaign(0, opts.baselineSurface, raw.baselineCampaign, 0))\n  // Improvement generations as 1..N. Substrate stores per-surface campaigns\n  // per generation — we summarize the WINNING surface per generation here.\n  for (const gen of raw.generations) {\n    const winner = gen.surfaces.reduce(\n      (best, s) =>\n        s.campaign.aggregates.cellsExecuted > 0 &&\n        (best === undefined || averageComposite(s.campaign) > averageComposite(best.campaign))\n          ? s\n          : best,\n      gen.surfaces[0],\n    )\n    if (!winner) continue\n    generations.push(\n      snapshotFromCampaign(gen.record.generationIndex + 1, winner.surface, winner.campaign, 0),\n    )\n  }\n\n  const event: EvalRunEvent = {\n    runId: `${runDir}#${Date.now()}`,\n    runDir,\n    timestamp: new Date().toISOString(),\n    status: 'finished',\n    labels: opts.hostedLabels ?? {},\n    baseline: generations[0],\n    generations,\n    gateDecision: summary.gateDecision,\n    holdoutLift: summary.lift,\n    totalCostUsd: summary.totalCostUsd,\n    totalDurationMs: summary.durationMs,\n    insightReport: summary.insight,\n  }\n\n  await client.ingestEvalRun(event)\n}\n\nfunction averageComposite(\n  campaign: RunImprovementLoopResult<unknown, Scenario>['baselineCampaign'],\n): number {\n  const aggs = Object.values(campaign.aggregates.byScenario)\n  return aggs.length === 0 ? 0 : aggs.reduce((s, a) => s + a.meanComposite, 0) / aggs.length\n}\n\nfunction hashString(s: string): string {\n  let h = 2166136261 >>> 0\n  for (let i = 0; i < s.length; i++) {\n    h ^= s.charCodeAt(i)\n    h = Math.imul(h, 16777619) >>> 0\n  }\n  return h.toString(16).padStart(8, '0')\n}\n\n/**\n * Adapt campaign cells into the `RunRecord` shape `analyzeRuns()` consumes.\n * Each cell becomes one run; `candidateId` is the caller-supplied label so\n * baseline + winner pair cleanly on `(experimentId, seed)`.\n *\n * `promptHash` is the REAL sha256 content hash of the surface this cell ran\n * (baseline vs winner are byte-distinguishable + byte-identical-verifiable);\n * `configHash` is the sha256 of the candidate label so the two candidates'\n * config rows differ. Both were previously the literal `'sha256:cell'`, which\n * made baseline and winner indistinguishable in every downstream record.\n */\nfunction cellsToRunRecords<TArtifact>(\n  cells: ReadonlyArray<CampaignCellResult<TArtifact>>,\n  candidateId: 'baseline' | 'winner',\n  runId: string,\n  surface: MutableSurface,\n): RunRecord[] {\n  const promptHash = surfaceContentHash(surface)\n  const configHash = `sha256:${createHash('sha256').update(candidateId).digest('hex')}`\n  return cells.map((cell) => {\n    const perJudge: Record<string, Record<string, number>> = {}\n    const perDimMeanAccum: Record<string, { sum: number; n: number }> = {}\n    let compositeSum = 0\n    let compositeCount = 0\n    for (const [judgeId, score] of Object.entries(cell.judgeScores)) {\n      perJudge[judgeId] = { ...score.dimensions }\n      for (const [dim, value] of Object.entries(score.dimensions)) {\n        if (!Number.isFinite(value)) continue\n        const accum = perDimMeanAccum[dim] ?? { sum: 0, n: 0 }\n        accum.sum += value\n        accum.n += 1\n        perDimMeanAccum[dim] = accum\n      }\n      if (Number.isFinite(score.composite)) {\n        compositeSum += score.composite\n        compositeCount += 1\n      }\n    }\n    const perDimMean: Record<string, number> = {}\n    for (const [dim, { sum, n }] of Object.entries(perDimMeanAccum)) {\n      perDimMean[dim] = n === 0 ? 0 : sum / n\n    }\n    const composite = compositeCount === 0 ? 0 : compositeSum / compositeCount\n    const judgeScores: JudgeScoresRecord = {\n      perJudge,\n      perDimMean,\n      composite,\n    }\n    return {\n      runId: `${runId}::${candidateId}::${cell.cellId}`,\n      experimentId: runId,\n      candidateId,\n      // Pair on (scenarioId, rep) — analyzeRuns pairs on (experimentId, seed).\n      // Synthesize a stable seed for that pairing.\n      seed:\n        cell.rep * 1_000_000 +\n        hashString(cell.scenarioId)\n          .slice(0, 6)\n          .split('')\n          .reduce((a, c) => (a * 31 + c.charCodeAt(0)) >>> 0, 0),\n      model: 'campaign-cell',\n      promptHash,\n      configHash,\n      commitSha: 'cell',\n      wallMs: cell.durationMs,\n      costUsd: cell.costUsd,\n      tokenUsage: { input: 0, output: 0 },\n      outcome: {\n        holdoutScore: composite,\n        raw: {},\n        judgeScores,\n      },\n      splitTag: 'holdout',\n      ...(cell.error ? { failureMode: cell.error } : {}),\n    } satisfies RunRecord\n  })\n}\n","import type { RunEvalOptions } from '../campaign/presets/run-eval'\nimport { runEval } from '../campaign/presets/run-eval'\nimport { inMemoryCampaignStorage } from '../campaign/storage'\nimport type {\n  CampaignResult,\n  DispatchContext,\n  JudgeConfig,\n  MutableSurface,\n  Scenario,\n} from '../campaign/types'\nimport type { HostedTenant } from '../hosted/client'\nimport {\n  type SelfImproveBudget,\n  type SelfImproveLlm,\n  type SelfImproveOptions,\n  type SelfImproveResult,\n  selfImprove,\n} from './self-improve'\n\nexport type AgentEvalAgent<TScenario extends Scenario, TArtifact> = (\n  surface: MutableSurface,\n  scenario: TScenario,\n  ctx: DispatchContext,\n) => Promise<TArtifact>\n\nexport type DefineAgentEvalOptions<TScenario extends Scenario, TArtifact> = SelfImproveOptions<\n  TScenario,\n  TArtifact\n>\n\nexport interface AgentEvalEvaluateOptions<TScenario extends Scenario, TArtifact>\n  extends Omit<\n    RunEvalOptions<TScenario, TArtifact>,\n    'dispatch' | 'judges' | 'runDir' | 'scenarios'\n  > {\n  /** Scenario set to evaluate. Defaults to the scenarios passed to `defineAgentEval`. */\n  scenarios?: TScenario[]\n  /** Surface to evaluate. Defaults to the baseline surface passed to `defineAgentEval`. */\n  surface?: MutableSurface\n  /** Agent to evaluate. Defaults to the agent passed to `defineAgentEval`. */\n  agent?: AgentEvalAgent<TScenario, TArtifact>\n  /** Single judge override. Ignored when `judges` is set. */\n  judge?: JudgeConfig<TArtifact, TScenario>\n  /** Full judge list override. Defaults to the single judge passed to `defineAgentEval`. */\n  judges?: JudgeConfig<TArtifact, TScenario>[]\n  /** Logical or filesystem run directory. Defaults to an in-memory run. */\n  runDir?: string\n}\n\nexport type AgentEvalImproveOptions<TScenario extends Scenario, TArtifact> = Omit<\n  Partial<SelfImproveOptions<TScenario, TArtifact>>,\n  'budget' | 'hostedTenant' | 'llm'\n> & {\n  budget?: Partial<SelfImproveBudget>\n  hostedTenant?: Partial<HostedTenant>\n  llm?: Partial<SelfImproveLlm>\n}\n\nexport interface DefinedAgentEval<TScenario extends Scenario, TArtifact> {\n  /** The default scenarios used by `evaluate()` and `improve()`. */\n  readonly scenarios: readonly TScenario[]\n  /** The default baseline surface used by `evaluate()` and `improve()`. */\n  readonly baselineSurface: MutableSurface\n  /**\n   * Run one scored evaluation. Use this for a baseline score or to score one\n   * candidate surface without running an improvement loop.\n   */\n  evaluate(\n    opts?: AgentEvalEvaluateOptions<TScenario, TArtifact>,\n  ): Promise<CampaignResult<TArtifact, TScenario>>\n  /**\n   * Run the closed improvement loop. Per-call overrides replace the definition\n   * except for nested config objects (`budget`, `llm`, `hostedTenant`), which\n   * are merged field-by-field so callers can override one knob without\n   * repeating secrets or budget defaults.\n   */\n  improve(\n    opts?: AgentEvalImproveOptions<TScenario, TArtifact>,\n  ): Promise<SelfImproveResult<TScenario, TArtifact>>\n}\n\n/**\n * Define an agent eval once, then either score a surface with `evaluate()` or\n * run the closed loop with `improve()`.\n *\n * This is a DX wrapper only: it delegates to `runEval()` and `selfImprove()` and\n * returns their native result shapes.\n */\nexport function defineAgentEval<TScenario extends Scenario, TArtifact>(\n  defaults: DefineAgentEvalOptions<TScenario, TArtifact>,\n): DefinedAgentEval<TScenario, TArtifact> {\n  const defaultEvaluateOptions = evaluateDefaults(defaults)\n\n  return {\n    scenarios: defaults.scenarios,\n    baselineSurface: defaults.baselineSurface,\n\n    async evaluate(opts = {}) {\n      const { agent, judge, judges, runDir, scenarios, surface, ...campaignOpts } = opts\n      const selectedAgent = agent ?? defaults.agent\n      const selectedSurface = surface ?? defaults.baselineSurface\n      const selectedRunDir = runDir ?? defaults.runDir ?? `mem://defineAgentEval-${Date.now()}`\n      const selectedStorage =\n        campaignOpts.storage ??\n        defaultEvaluateOptions.storage ??\n        (selectedRunDir.startsWith('mem://') ? inMemoryCampaignStorage() : undefined)\n      const evalOptions: RunEvalOptions<TScenario, TArtifact> = {\n        ...defaultEvaluateOptions,\n        ...campaignOpts,\n        ...(selectedStorage ? { storage: selectedStorage } : {}),\n        runDir: selectedRunDir,\n        scenarios: scenarios ?? defaults.scenarios,\n        dispatch: (scenario, ctx) => selectedAgent(selectedSurface, scenario, ctx),\n        judges: evaluateJudges(judges, judge ?? defaults.judge),\n      }\n      if (evalOptions.reps !== undefined)\n        evalOptions.reps = requirePositiveInteger(evalOptions.reps, 'reps')\n      return runEval<TScenario, TArtifact>(evalOptions)\n    },\n\n    async improve(opts = {}) {\n      const {\n        budget: budgetOverride,\n        hostedTenant: hostedTenantOverride,\n        llm: llmOverride,\n        ...topLevelOverrides\n      } = opts\n      const merged = mergeDefined(defaults, topLevelOverrides)\n      const budget = mergeBudget(defaults.budget, budgetOverride)\n      const llm = mergeLlm(defaults.llm, llmOverride)\n      const hostedTenant = mergeHostedTenant(defaults.hostedTenant, hostedTenantOverride)\n      return selfImprove<TScenario, TArtifact>({\n        ...merged,\n        ...(budget ? { budget } : {}),\n        ...(llm ? { llm } : {}),\n        ...(hostedTenant ? { hostedTenant } : {}),\n      })\n    },\n  }\n}\n\ntype SharedEvaluateDefaults<TScenario extends Scenario, TArtifact> = Omit<\n  RunEvalOptions<TScenario, TArtifact>,\n  'dispatch' | 'judges' | 'runDir' | 'scenarios'\n>\n\nfunction evaluateDefaults<TScenario extends Scenario, TArtifact>(\n  defaults: DefineAgentEvalOptions<TScenario, TArtifact>,\n): SharedEvaluateDefaults<TScenario, TArtifact> {\n  const out: SharedEvaluateDefaults<TScenario, TArtifact> = {}\n  if (defaults.storage) out.storage = defaults.storage\n  if (defaults.labeledStore) out.labeledStore = defaults.labeledStore\n  if (defaults.captureSource) out.captureSource = defaults.captureSource\n  if (defaults.cellPlacement) out.cellPlacement = defaults.cellPlacement\n  if (defaults.expectUsage) out.expectUsage = defaults.expectUsage\n  if (defaults.budget?.dollars !== undefined) out.costCeiling = defaults.budget.dollars\n  if (defaults.budget?.maxConcurrency !== undefined)\n    out.maxConcurrency = defaults.budget.maxConcurrency\n  if (defaults.budget?.reps !== undefined)\n    out.reps = requirePositiveInteger(defaults.budget.reps, 'budget.reps')\n  return out\n}\n\nfunction mergeBudget(\n  defaults: SelfImproveBudget | undefined,\n  overrides: Partial<SelfImproveBudget> | undefined,\n): SelfImproveBudget | undefined {\n  const merged = mergeOptionalObject(defaults, overrides)\n  if (merged?.reps !== undefined) merged.reps = requirePositiveInteger(merged.reps, 'budget.reps')\n  return merged\n}\n\nfunction mergeLlm(\n  defaults: SelfImproveLlm | undefined,\n  overrides: Partial<SelfImproveLlm> | undefined,\n): SelfImproveLlm | undefined {\n  return mergeOptionalObject(defaults, overrides)\n}\n\nfunction mergeHostedTenant(\n  defaults: HostedTenant | undefined,\n  overrides: Partial<HostedTenant> | undefined,\n): HostedTenant | undefined {\n  const merged = mergeOptionalObject(defaults, overrides)\n  if (!merged) return undefined\n  if (!merged.endpoint?.trim() || !merged.apiKey?.trim() || !merged.tenantId?.trim()) {\n    throw new Error(\n      'defineAgentEval.improve: hostedTenant requires endpoint, apiKey, and tenantId after merging defaults and overrides',\n    )\n  }\n  return merged\n}\n\nfunction mergeDefined<T extends object>(defaults: T, overrides: Partial<T> | undefined): T {\n  if (!overrides) return defaults\n  const merged = { ...defaults } as Record<string, unknown>\n  for (const [key, value] of Object.entries(overrides)) {\n    if (value !== undefined) merged[key] = value\n  }\n  return merged as T\n}\n\nfunction mergeOptionalObject<T extends object>(\n  defaults: T | undefined,\n  overrides: Partial<T> | undefined,\n): T | undefined {\n  if (!defaults && !overrides) return undefined\n  return mergeDefined(defaults ?? ({} as T), overrides)\n}\n\nfunction evaluateJudges<TArtifact, TScenario extends Scenario>(\n  judges: JudgeConfig<TArtifact, TScenario>[] | undefined,\n  defaultJudge: JudgeConfig<TArtifact, TScenario>,\n): JudgeConfig<TArtifact, TScenario>[] {\n  if (judges !== undefined) {\n    if (judges.length === 0) {\n      throw new Error('defineAgentEval.evaluate: judges must not be empty')\n    }\n    return judges\n  }\n  return [defaultJudge]\n}\n\nfunction requirePositiveInteger(value: number, field: string): number {\n  if (!Number.isInteger(value) || value < 1) {\n    throw new Error(`defineAgentEval: ${field} must be a positive integer`)\n  }\n  return value\n}\n","/**\n * # `@tangle-network/agent-eval/contract` — eval-run diff primitive.\n *\n * The substrate side of the v-N-versus-v-N+1 dashboard view. Given two\n * `EvalRunEvent`s (or two `EvalRunGenerationSnapshot`s from one run), this\n * returns a normalised diff: per-cell composite + per-judge/per-dimension\n * deltas, surface-hash change, aggregate cost + duration shifts.\n *\n * Consumed by:\n *   - The hosted-tier dashboard (intelligence-web) — renders v3 vs v4\n *     comparisons of cells × judges × dimensions.\n *   - CI reporting — emits a \"shipped: composite +0.07, cost +$1.20\" line\n *     in PR review for autonomous-improvement runs.\n *   - Any downstream consumer that needs \"what actually changed\" without\n *     reimplementing the matching + arithmetic.\n *\n * Cells are matched on the natural composite key `(scenarioId, rep)`.\n * Unmatched cells surface as `removed` / `added` so callers can tell\n * \"this cell got worse\" from \"this cell wasn't run.\"\n */\n\nimport type { GateDecision } from '../campaign/types'\nimport type { EvalRunCellScore, EvalRunEvent, EvalRunGenerationSnapshot } from '../hosted/types'\n\n// ── Types ────────────────────────────────────────────────────────────\n\n/** Per-dimension delta. `before` / `after` are null when the judge did not\n *  emit a value for that side. `delta` is `after - before`; null when\n *  either side is null. */\nexport interface EvalDimensionDelta {\n  before: number | null\n  after: number | null\n  delta: number | null\n}\n\n/** Per-cell delta, keyed on `(scenarioId, rep)`. */\nexport interface EvalCellScoreDelta {\n  scenarioId: string\n  rep: number\n  compositeBefore: number\n  compositeAfter: number\n  compositeDelta: number\n  /** Per-judge → per-dimension deltas. Outer key = judge name from\n   *  `EvalRunCellScore.dimensions`; inner key = dimension name. */\n  dimensions: Record<string, Record<string, EvalDimensionDelta>>\n}\n\n/** Diff between two generation snapshots — the unit the dashboard renders\n *  for a single \"v3 vs v4\" comparison. */\nexport interface EvalGenerationDiff {\n  beforeIndex: number\n  afterIndex: number\n  beforeSurfaceHash: string\n  afterSurfaceHash: string\n  surfaceChanged: boolean\n  /** Cells present in both snapshots, matched on `(scenarioId, rep)`. */\n  matched: EvalCellScoreDelta[]\n  /** Cells present in `before` but missing from `after`. */\n  removed: EvalRunCellScore[]\n  /** Cells present in `after` but missing from `before`. */\n  added: EvalRunCellScore[]\n  /** Aggregate composite mean across all cells in the snapshot. */\n  compositeBefore: number\n  compositeAfter: number\n  compositeDelta: number\n  costUsdBefore: number\n  costUsdAfter: number\n  costUsdDelta: number\n  durationMsBefore: number\n  durationMsAfter: number\n  durationMsDelta: number\n}\n\n/** Diff between two full eval-runs. Includes both baseline-vs-baseline and\n *  winner-vs-winner generation diffs when both sides expose them, plus\n *  run-level metadata. */\nexport interface EvalRunDiff {\n  beforeRunId: string\n  afterRunId: string\n  beforeTimestamp: string\n  afterTimestamp: string\n  beforeGateDecision: GateDecision | null\n  afterGateDecision: GateDecision | null\n  beforeHoldoutLift: number | null\n  afterHoldoutLift: number | null\n  holdoutLiftDelta: number | null\n  beforeTotalCostUsd: number\n  afterTotalCostUsd: number\n  totalCostUsdDelta: number\n  beforeTotalDurationMs: number\n  afterTotalDurationMs: number\n  totalDurationMsDelta: number\n  /** Baseline-vs-baseline diff. Null when either run has no baseline. */\n  baselineDiff: EvalGenerationDiff | null\n  /** Highest-index-generation comparison. Null when either run has no\n   *  recorded generations (e.g. baseline-only or errored before any\n   *  generation completed). */\n  winnersDiff: EvalGenerationDiff | null\n}\n\n// ── Implementation ───────────────────────────────────────────────────\n\nfunction keyForCell(cell: EvalRunCellScore): string {\n  // JSON-tuple key — `scenarioId` may legitimately contain `::` or any other\n  // delimiter, so we use JSON.stringify on a 2-tuple to get an unambiguous,\n  // collision-free composite key.\n  return JSON.stringify([cell.scenarioId, cell.rep])\n}\n\n/** Build the per-dimension delta map for a matched cell. Each judge name +\n *  dimension name encountered on EITHER side appears in the result. */\nfunction diffDimensions(\n  before: EvalRunCellScore['dimensions'],\n  after: EvalRunCellScore['dimensions'],\n): EvalCellScoreDelta['dimensions'] {\n  const out: EvalCellScoreDelta['dimensions'] = {}\n  const judges = new Set<string>([...Object.keys(before), ...Object.keys(after)])\n  for (const judge of judges) {\n    const beforeDims = before[judge] ?? {}\n    const afterDims = after[judge] ?? {}\n    const dims = new Set<string>([...Object.keys(beforeDims), ...Object.keys(afterDims)])\n    const judgeOut: Record<string, EvalDimensionDelta> = {}\n    for (const dim of dims) {\n      // Coerce non-finite values (NaN, ±Infinity) to null so the diff never\n      // surfaces NaN/Infinity to the dashboard. A NaN score is a substrate\n      // bug from upstream; the diff treats it as \"no value\" rather than\n      // propagating the corruption.\n      const rawBefore = beforeDims[dim]\n      const rawAfter = afterDims[dim]\n      const b = typeof rawBefore === 'number' && Number.isFinite(rawBefore) ? rawBefore : null\n      const a = typeof rawAfter === 'number' && Number.isFinite(rawAfter) ? rawAfter : null\n      judgeOut[dim] = {\n        before: b,\n        after: a,\n        delta: b !== null && a !== null ? a - b : null,\n      }\n    }\n    out[judge] = judgeOut\n  }\n  return out\n}\n\n/**\n * Diff two generation snapshots. Cells are matched on `(scenarioId, rep)`;\n * unmatched cells surface in `added` / `removed`. Aggregate fields are\n * recomputed from the snapshot's stored fields, not re-derived from cells —\n * this keeps the diff consistent with whatever aggregation the substrate\n * actually reported.\n */\nexport function diffGenerations(\n  before: EvalRunGenerationSnapshot,\n  after: EvalRunGenerationSnapshot,\n): EvalGenerationDiff {\n  const beforeMap = new Map(before.cells.map((c) => [keyForCell(c), c]))\n  const afterMap = new Map(after.cells.map((c) => [keyForCell(c), c]))\n\n  const matched: EvalCellScoreDelta[] = []\n  const removed: EvalRunCellScore[] = []\n  const added: EvalRunCellScore[] = []\n\n  for (const [key, beforeCell] of beforeMap) {\n    const afterCell = afterMap.get(key)\n    if (!afterCell) {\n      removed.push(beforeCell)\n      continue\n    }\n    matched.push({\n      scenarioId: beforeCell.scenarioId,\n      rep: beforeCell.rep,\n      compositeBefore: beforeCell.compositeMean,\n      compositeAfter: afterCell.compositeMean,\n      compositeDelta: afterCell.compositeMean - beforeCell.compositeMean,\n      dimensions: diffDimensions(beforeCell.dimensions, afterCell.dimensions),\n    })\n  }\n  for (const [key, afterCell] of afterMap) {\n    if (!beforeMap.has(key)) added.push(afterCell)\n  }\n\n  return {\n    beforeIndex: before.index,\n    afterIndex: after.index,\n    beforeSurfaceHash: before.surfaceHash,\n    afterSurfaceHash: after.surfaceHash,\n    surfaceChanged: before.surfaceHash !== after.surfaceHash,\n    matched,\n    removed,\n    added,\n    compositeBefore: before.compositeMean,\n    compositeAfter: after.compositeMean,\n    compositeDelta: after.compositeMean - before.compositeMean,\n    costUsdBefore: before.costUsd,\n    costUsdAfter: after.costUsd,\n    costUsdDelta: after.costUsd - before.costUsd,\n    durationMsBefore: before.durationMs,\n    durationMsAfter: after.durationMs,\n    durationMsDelta: after.durationMs - before.durationMs,\n  }\n}\n\n/** Highest-index generation, or null if the run recorded none. */\nfunction winnerOf(run: EvalRunEvent): EvalRunGenerationSnapshot | null {\n  if (run.generations.length === 0) return null\n  let winner = run.generations[0] as EvalRunGenerationSnapshot\n  for (const gen of run.generations) {\n    if (gen.index > winner.index) winner = gen\n  }\n  return winner\n}\n\n/**\n * Diff two full eval-runs. Produces baseline-vs-baseline and\n * winner-vs-winner generation diffs when both sides expose them, plus\n * run-level cost / lift / gate-decision deltas.\n */\nexport function diffRuns(before: EvalRunEvent, after: EvalRunEvent): EvalRunDiff {\n  const beforeWinner = winnerOf(before)\n  const afterWinner = winnerOf(after)\n  const baselineDiff =\n    before.baseline && after.baseline ? diffGenerations(before.baseline, after.baseline) : null\n  const winnersDiff =\n    beforeWinner && afterWinner ? diffGenerations(beforeWinner, afterWinner) : null\n\n  const beforeLift = before.holdoutLift ?? null\n  const afterLift = after.holdoutLift ?? null\n\n  return {\n    beforeRunId: before.runId,\n    afterRunId: after.runId,\n    beforeTimestamp: before.timestamp,\n    afterTimestamp: after.timestamp,\n    beforeGateDecision: before.gateDecision ?? null,\n    afterGateDecision: after.gateDecision ?? null,\n    beforeHoldoutLift: beforeLift,\n    afterHoldoutLift: afterLift,\n    holdoutLiftDelta: beforeLift !== null && afterLift !== null ? afterLift - beforeLift : null,\n    beforeTotalCostUsd: before.totalCostUsd,\n    afterTotalCostUsd: after.totalCostUsd,\n    totalCostUsdDelta: after.totalCostUsd - before.totalCostUsd,\n    beforeTotalDurationMs: before.totalDurationMs,\n    afterTotalDurationMs: after.totalDurationMs,\n    totalDurationMsDelta: after.totalDurationMs - before.totalDurationMs,\n    baselineDiff,\n    winnersDiff,\n  }\n}\n\n/**\n * Within-run baseline → winning-generation diff. The natural \"what did the\n * improvement loop produce\" view for a single run. Returns null when the\n * run never reached a generation past baseline (errored early, or the gate\n * shipped the baseline as-is).\n */\nexport function diffRunBaselineToWinner(run: EvalRunEvent): EvalGenerationDiff | null {\n  if (!run.baseline) return null\n  const winner = winnerOf(run)\n  if (!winner || winner.index === run.baseline.index) return null\n  return diffGenerations(run.baseline, winner)\n}\n","/**\n * `fromAgentTrace` — provenance correlation from Cursor's Agent Trace spec\n * (https://github.com/cursor/agent-trace, RFC v0.1.0).\n *\n * Agent Trace is NOT a run/quality trace — it carries no outcome, score, or\n * cost. It records *code authorship*: for a VCS revision, which AI model /\n * conversation authored which file ranges. It explicitly disclaims quality\n * assessment — which is exactly what `analyzeRuns` adds.\n *\n * The two layers join on a key the substrate already has: a `RunRecord`\n * carries `commitSha`, and an Agent Trace record is keyed by\n * `vcs.revision`. So this adapter does not produce `RunRecord`s — it builds a\n * provenance index by commit and partitions existing runs by their authoring\n * model. Feed each cohort to `analyzeRuns` (or pass one as `baselineRuns`) to\n * answer the question no run-only trace can: *which authoring agent's code\n * fails / regresses / costs more.*\n *\n * Granularity is commit-level (the SHA join). Per-file/per-line correlation\n * would require runs to record which files they exercised — out of scope.\n */\n\nimport type { RunRecord } from '../../run-record'\n\n// ── Agent Trace record schema (the subset we read) ──────────────────────────\n\nexport type AgentTraceContributorType = 'human' | 'ai' | 'mixed' | 'unknown'\n\nexport interface AgentTraceContributor {\n  type: AgentTraceContributorType\n  /** models.dev id, e.g. `anthropic/claude-opus-4-5-20251101`. */\n  model_id?: string\n}\n\nexport interface AgentTraceRange {\n  start_line: number\n  end_line: number\n  content_hash?: string\n  /** Per-range contributor override (agent handoffs). Wins over the\n   *  conversation-level contributor for these lines. */\n  contributor?: AgentTraceContributor\n}\n\nexport interface AgentTraceConversation {\n  url?: string\n  contributor?: AgentTraceContributor\n  ranges: AgentTraceRange[]\n}\n\nexport interface AgentTraceFile {\n  path: string\n  conversations: AgentTraceConversation[]\n}\n\nexport interface AgentTraceRecord {\n  version: string\n  id: string\n  timestamp: string\n  vcs?: { type: string; revision: string }\n  tool?: { name?: string; version?: string }\n  files: AgentTraceFile[]\n}\n\n// ── Provenance index ─────────────────────────────────────────────────────────\n\n/** Authorship provenance for one VCS revision, aggregated across the record's\n *  files/conversations/ranges. */\nexport interface AuthoringProvenance {\n  commitSha: string\n  /** Unique AI model ids that authored code in this commit (type ai|mixed). */\n  aiModels: string[]\n  /** Tools that produced the records (e.g. `cursor`). */\n  tools: string[]\n  conversationCount: number\n  fileCount: number\n  /** Total attributed lines (sum of range spans). */\n  lineCount: number\n  /** True if any range was authored (in whole or part) by a human. */\n  humanInvolved: boolean\n}\n\nexport type AgentTraceIndex = Map<string, AuthoringProvenance>\n\nfunction rangeLines(r: AgentTraceRange): number {\n  return Math.max(0, r.end_line - r.start_line + 1)\n}\n\n/**\n * Build a commit → provenance index from Agent Trace records. Multiple records\n * for the same revision are merged. Records without `vcs.revision` are skipped\n * (the SHA is the join key — without it there is nothing to correlate against).\n */\nexport function parseAgentTrace(records: AgentTraceRecord[]): AgentTraceIndex {\n  interface Acc {\n    models: Set<string>\n    tools: Set<string>\n    files: Set<string>\n    conversationCount: number\n    lineCount: number\n    humanInvolved: boolean\n  }\n  const acc = new Map<string, Acc>()\n\n  for (const record of records) {\n    const sha = record.vcs?.revision\n    if (!sha) continue // the SHA is the join key — nothing to correlate without it\n\n    let a = acc.get(sha)\n    if (!a) {\n      a = {\n        models: new Set(),\n        tools: new Set(),\n        files: new Set(),\n        conversationCount: 0,\n        lineCount: 0,\n        humanInvolved: false,\n      }\n      acc.set(sha, a)\n    }\n\n    if (record.tool?.name) a.tools.add(record.tool.name)\n\n    for (const file of record.files ?? []) {\n      a.files.add(file.path)\n      for (const conv of file.conversations ?? []) {\n        a.conversationCount += 1\n        for (const range of conv.ranges ?? []) {\n          // Per-range contributor wins, else the conversation contributor.\n          const contributor = range.contributor ?? conv.contributor\n          a.lineCount += rangeLines(range)\n          if (!contributor) continue\n          if (contributor.type === 'human' || contributor.type === 'mixed') {\n            a.humanInvolved = true\n          }\n          if ((contributor.type === 'ai' || contributor.type === 'mixed') && contributor.model_id) {\n            a.models.add(contributor.model_id)\n          }\n        }\n      }\n    }\n  }\n\n  const index: AgentTraceIndex = new Map()\n  for (const [sha, a] of acc) {\n    index.set(sha, {\n      commitSha: sha,\n      aiModels: [...a.models].sort(),\n      tools: [...a.tools].sort(),\n      conversationCount: a.conversationCount,\n      fileCount: a.files.size,\n      lineCount: a.lineCount,\n      humanInvolved: a.humanInvolved,\n    })\n  }\n  return index\n}\n\n// ── Run ↔ provenance join ──────────────────────────────────────────────────\n\nexport interface PartitionByAuthoringModelResult {\n  /** Runs grouped by each AI model that authored code in the run's commit. A\n   *  run whose commit had multiple authoring models appears under EACH — the\n   *  cohorts overlap by construction at commit granularity. */\n  byModel: Map<string, RunRecord[]>\n  /** Runs whose `commitSha` had no Agent Trace provenance (no record, or no\n   *  AI authorship). Kept separate — never silently folded into a cohort. */\n  unattributed: RunRecord[]\n}\n\n/**\n * Partition runs by the AI model(s) that authored the code at each run's\n * `commitSha`. Feed `byModel.get(modelId)` to `analyzeRuns`, or compare two\n * model cohorts via `analyzeRuns({ runs: a, baselineRuns: b })` for a lift CI\n * on \"model A's code vs model B's code\".\n */\nexport function partitionRunsByAuthoringModel(\n  runs: RunRecord[],\n  index: AgentTraceIndex,\n): PartitionByAuthoringModelResult {\n  const byModel = new Map<string, RunRecord[]>()\n  const unattributed: RunRecord[] = []\n\n  for (const run of runs) {\n    const provenance = index.get(run.commitSha)\n    if (!provenance || provenance.aiModels.length === 0) {\n      unattributed.push(run)\n      continue\n    }\n    for (const model of provenance.aiModels) {\n      const cohort = byModel.get(model) ?? []\n      cohort.push(run)\n      byModel.set(model, cohort)\n    }\n  }\n\n  return { byModel, unattributed }\n}\n","/**\n * # `intake/feedback-table` — multi-rater approve/reject corpus → `RunRecord[]`.\n *\n * The generic shape behind Obsidian's `#approved` / `#rejected` tags, a\n * Google Sheet, a Postgres `feedback` table, or any CSV with ratings.\n *\n * Caller supplies one row per (run, rater) tuple plus per-run metadata; the\n * adapter rolls them up into the substrate-canonical `RunRecord` shape so\n * `analyzeRuns({ runs, raterScores })` can produce inter-rater agreement,\n * disagreement triage, and downstream recommendations.\n *\n * Per-run `RunRecord.outcome.searchScore` is the rater-mean rating\n * (normalised to 0..1 when scale is supplied); `outcome.raw` carries the\n * per-rater scores keyed by rater id for downstream attribution.\n */\n\nimport type { JudgeScoresRecord, RunOutcome, RunRecord, RunSplitTag } from '../../run-record'\n\nexport interface FeedbackTableRow {\n  /** Stable id for this run — the unit a rater scored. Drives pairing\n   *  across analysis primitives. */\n  runId: string\n  /** Identifier of the rater that produced this rating. */\n  rater: string\n  /** The rating itself. Accepts boolean (approve/reject), 0..1 scalar,\n   *  or any numeric scale — see `scale`. */\n  rating: number | boolean\n  /** Optional metadata carried through to `RunRecord.outcome.raw` and the\n   *  custom-shape metadata bag. */\n  metadata?: Record<string, unknown>\n}\n\nexport interface FeedbackTableMeta {\n  runId: string\n  /** When omitted, defaults to `'feedback-corpus'`. Used to group related\n   *  runs in `analyzeRuns()` lift analysis. */\n  experimentId?: string\n  /** When omitted, defaults to `runId` — each run is its own candidate. */\n  candidateId?: string\n  /** Cost in USD, when available. Set to 0 when unknown — the consumer's\n   *  cost analysis sections will collapse gracefully. */\n  costUsd?: number\n  /** Wall-clock ms, when available. Defaults to 0. */\n  wallMs?: number\n  /** Model identifier including snapshot. Default `unknown@unknown`. */\n  model?: string\n  /** Optional sha256 of the prompt; default `'sha256:unknown'`. */\n  promptHash?: string\n  /** Default `'sha256:unknown'`. */\n  configHash?: string\n  /** Default `'unknown'`. */\n  commitSha?: string\n  /** Default `'holdout'` — feedback corpora are by nature the holdout\n   *  signal a closed-loop improvement aims at. */\n  splitTag?: RunSplitTag\n  /** Free-form metadata available to consumers via the cast-out path on\n   *  the resulting RunRecord. */\n  extras?: Record<string, unknown>\n}\n\nexport interface FromFeedbackTableOptions {\n  /** Per-(run, rater) ratings. */\n  ratings: FeedbackTableRow[]\n  /** Per-run metadata. When a runId appears in `ratings` but not here, the\n   *  adapter synthesises minimal metadata with defaults documented above. */\n  meta?: FeedbackTableMeta[]\n  /** Rating scale. Provide `{ min, max }` for non-0..1 numeric scales.\n   *  Booleans are normalised: true → 1, false → 0. Default: assumes\n   *  ratings are already 0..1. */\n  scale?: { min: number; max: number }\n  /** When true, the rater scores are emitted into `raterScores` (a sibling\n   *  array `analyzeRuns()` accepts) instead of being averaged into the\n   *  run's `outcome.searchScore`. Default `true` — preserves rater-level\n   *  signal for inter-rater analysis. */\n  emitRaterScores?: boolean\n}\n\nexport interface FromFeedbackTableResult {\n  runs: RunRecord[]\n  /** Rater-level scores ready to pass into `analyzeRuns({ raterScores })`\n   *  for inter-rater agreement + disagreement triage. */\n  raterScores: Array<{ runId: string; rater: string; score: number }>\n}\n\nexport function fromFeedbackTable(opts: FromFeedbackTableOptions): FromFeedbackTableResult {\n  const { ratings, meta = [], scale, emitRaterScores = true } = opts\n  const metaByRun = new Map(meta.map((m) => [m.runId, m]))\n\n  // Normalise per-rating to a 0..1 score.\n  const normalise = (rating: number | boolean): number => {\n    if (typeof rating === 'boolean') return rating ? 1 : 0\n    if (!Number.isFinite(rating)) return Number.NaN\n    if (!scale) return rating\n    const { min, max } = scale\n    if (max === min) return rating\n    return (rating - min) / (max - min)\n  }\n\n  // Group ratings by runId.\n  const byRun = new Map<string, FeedbackTableRow[]>()\n  for (const row of ratings) {\n    const list = byRun.get(row.runId) ?? []\n    list.push(row)\n    byRun.set(row.runId, list)\n  }\n\n  const runs: RunRecord[] = []\n  const raterScores: FromFeedbackTableResult['raterScores'] = []\n\n  for (const [runId, rowsForRun] of byRun) {\n    const normalised = rowsForRun\n      .map((r) => ({ rater: r.rater, score: normalise(r.rating) }))\n      .filter((r) => Number.isFinite(r.score))\n    if (normalised.length === 0) continue\n\n    const meanScore = normalised.reduce((s, r) => s + r.score, 0) / normalised.length\n\n    const runMeta = metaByRun.get(runId) ?? ({ runId } as FeedbackTableMeta)\n\n    const judgeScores: JudgeScoresRecord = {\n      perJudge: Object.fromEntries(normalised.map((r) => [r.rater, { rating: r.score }])),\n      perDimMean: { rating: meanScore },\n      composite: meanScore,\n    }\n\n    const outcome: RunOutcome = {\n      // Feedback corpora ARE the holdout signal — score lands on\n      // `holdoutScore` so downstream substrate primitives (`paretoChart`,\n      // promotion gates) read it correctly by default.\n      holdoutScore: meanScore,\n      raw: Object.fromEntries(normalised.map((r) => [`rater:${r.rater}`, r.score])),\n      judgeScores,\n    }\n\n    runs.push({\n      runId,\n      experimentId: runMeta.experimentId ?? 'feedback-corpus',\n      candidateId: runMeta.candidateId ?? runId,\n      seed: 0,\n      model: runMeta.model ?? 'unknown@unknown',\n      promptHash: runMeta.promptHash ?? 'sha256:unknown',\n      configHash: runMeta.configHash ?? 'sha256:unknown',\n      commitSha: runMeta.commitSha ?? 'unknown',\n      wallMs: runMeta.wallMs ?? 0,\n      costUsd: runMeta.costUsd ?? 0,\n      tokenUsage: { input: 0, output: 0 },\n      outcome,\n      splitTag: runMeta.splitTag ?? 'holdout',\n    } as RunRecord)\n\n    if (emitRaterScores) {\n      for (const r of normalised) raterScores.push({ runId, rater: r.rater, score: r.score })\n    }\n  }\n\n  return { runs, raterScores }\n}\n","/**\n * # `intake/otel-spans` — OTel `TraceSpanEvent[]` → `RunRecord[]`.\n *\n * Turns an existing observability stream into the substrate-canonical\n * `RunRecord` shape so consumers with logs but no eval discipline can\n * call `analyzeRuns()` against their production traffic immediately.\n *\n * Pivot rule: spans are grouped by `tangle.runId` (the same attribute the\n * hosted-tier wire format uses) or, when absent, by `traceId`. One group\n * becomes one `RunRecord`. The root span (no `parentSpanId`) supplies:\n *\n *   - `runId` (the group key)\n *   - `wallMs` from `endTimeUnixNano - startTimeUnixNano`\n *   - `model` from `gen_ai.request.model` / `llm.model` / `tangle.model`\n *   - cost from `cost.usd` / `gen_ai.usage.cost_usd` / `tangle.cost.usd`\n *   - token usage from `gen_ai.usage.{input,output}_tokens`\n *   - `outcome.searchScore` from `tangle.score` / `eval.score` when\n *     present; `outcome.raw` collects every numeric attribute.\n *\n * Spans that ERRORed (`status.code === 'ERROR'`) populate `failureMode`\n * with their `name` so `analyzeRuns()`'s failure clustering sees them.\n */\n\nimport type { TraceSpanEvent } from '../../hosted/types'\nimport type {\n  JudgeScoresRecord,\n  RunOutcome,\n  RunRecord,\n  RunSplitTag,\n  RunTokenUsage,\n} from '../../run-record'\nimport {\n  LLM_INPUT_TOKEN_ATTR_KEYS,\n  LLM_MODEL_ATTR_KEYS,\n  LLM_OUTPUT_TOKEN_ATTR_KEYS,\n} from '../../trace/otlp-attributes'\n\nconst SCORE_KEYS = ['tangle.score', 'eval.score', 'score']\nconst MODEL_KEYS = ['tangle.model', ...LLM_MODEL_ATTR_KEYS, 'model']\nconst COST_KEYS = ['tangle.cost.usd', 'gen_ai.usage.cost_usd', 'cost.usd', 'cost']\nconst INPUT_TOKEN_KEYS = [...LLM_INPUT_TOKEN_ATTR_KEYS, 'tangle.tokens.in', 'tokens.in']\nconst OUTPUT_TOKEN_KEYS = [...LLM_OUTPUT_TOKEN_ATTR_KEYS, 'tangle.tokens.out', 'tokens.out']\nconst PROMPT_HASH_KEYS = ['tangle.prompt_hash', 'prompt.hash']\nconst CONFIG_HASH_KEYS = ['tangle.config_hash', 'config.hash']\n\nexport interface FromOtelSpansOptions {\n  spans: TraceSpanEvent[]\n  /** Default split tag for synthesized records. Defaults to `'holdout'`. */\n  defaultSplit?: RunSplitTag\n  /** Default `experimentId` when not present on any span. */\n  experimentId?: string\n}\n\nexport function fromOtelSpans(opts: FromOtelSpansOptions): RunRecord[] {\n  const { spans, defaultSplit = 'holdout', experimentId = 'otel-corpus' } = opts\n  const grouped = groupSpans(spans)\n\n  const runs: RunRecord[] = []\n  for (const [groupKey, groupSpans] of grouped) {\n    const root = findRoot(groupSpans)\n    if (!root) continue\n\n    const wallMs = Math.max(0, (root.endTimeUnixNano - root.startTimeUnixNano) / 1_000_000)\n    const model = readAttrString(groupSpans, MODEL_KEYS) ?? 'unknown@unknown'\n    const costUsd = readAttrNumber(groupSpans, COST_KEYS) ?? 0\n    const inputTokens = readAttrNumber(groupSpans, INPUT_TOKEN_KEYS) ?? 0\n    const outputTokens = readAttrNumber(groupSpans, OUTPUT_TOKEN_KEYS) ?? 0\n    const promptHash = readAttrString(groupSpans, PROMPT_HASH_KEYS) ?? 'sha256:unknown'\n    const configHash = readAttrString(groupSpans, CONFIG_HASH_KEYS) ?? 'sha256:unknown'\n    const score = readAttrNumber(groupSpans, SCORE_KEYS)\n\n    const rawNumeric = collectNumericAttrs(groupSpans)\n    const tokenUsage: RunTokenUsage = {\n      input: inputTokens,\n      output: outputTokens,\n    }\n\n    const judgeScores: JudgeScoresRecord | undefined =\n      score !== undefined\n        ? {\n            perJudge: { 'otel-derived': { score } },\n            perDimMean: { score },\n            composite: score,\n          }\n        : undefined\n\n    const errorSpan = groupSpans.find((s) => s.status?.code === 'ERROR')\n    const outcome: RunOutcome = {\n      ...(opts.defaultSplit === 'search' ? { searchScore: score } : { holdoutScore: score }),\n      raw: rawNumeric,\n      ...(judgeScores ? { judgeScores } : {}),\n    }\n\n    runs.push({\n      runId: groupKey,\n      experimentId,\n      candidateId: (root.attributes['tangle.candidateId'] as string | undefined) ?? 'otel-default',\n      seed: 0,\n      model,\n      promptHash,\n      configHash,\n      commitSha: (root.attributes['tangle.commit_sha'] as string | undefined) ?? 'unknown',\n      wallMs,\n      costUsd,\n      tokenUsage,\n      outcome,\n      splitTag: defaultSplit,\n      ...(errorSpan ? { failureMode: errorSpan.name } : {}),\n    } as RunRecord)\n  }\n  return runs\n}\n\n// ── Internal helpers ────────────────────────────────────────────────\n\nfunction groupSpans(spans: TraceSpanEvent[]): Map<string, TraceSpanEvent[]> {\n  const m = new Map<string, TraceSpanEvent[]>()\n  for (const span of spans) {\n    const key = (span['tangle.runId'] as string | undefined) ?? span.traceId\n    const list = m.get(key) ?? []\n    list.push(span)\n    m.set(key, list)\n  }\n  return m\n}\n\nfunction findRoot(group: TraceSpanEvent[]): TraceSpanEvent | undefined {\n  return group.find((s) => !s.parentSpanId) ?? group[0]\n}\n\nfunction readAttrString(spans: TraceSpanEvent[], keys: string[]): string | undefined {\n  for (const span of spans) {\n    for (const key of keys) {\n      const v = span.attributes[key]\n      if (typeof v === 'string' && v.length > 0) return v\n    }\n  }\n  return undefined\n}\n\nfunction readAttrNumber(spans: TraceSpanEvent[], keys: string[]): number | undefined {\n  for (const span of spans) {\n    for (const key of keys) {\n      const v = span.attributes[key]\n      if (typeof v === 'number' && Number.isFinite(v)) return v\n      if (typeof v === 'string') {\n        const parsed = Number(v)\n        if (Number.isFinite(parsed)) return parsed\n      }\n    }\n  }\n  return undefined\n}\n\nfunction collectNumericAttrs(spans: TraceSpanEvent[]): Record<string, number> {\n  const raw: Record<string, number> = {}\n  for (const span of spans) {\n    for (const [k, v] of Object.entries(span.attributes)) {\n      if (typeof v === 'number' && Number.isFinite(v)) raw[k] = v\n    }\n  }\n  return raw\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAwBA,SAAS,kBAAkB;AAoR3B,SAAS,kBACP,WACA,UAC8C;AAE9C,WAAS,KAAK,GAAmB;AAC/B,QAAI,IAAI,eAAe;AACvB,aAAS,IAAI,GAAG,IAAI,EAAE,QAAQ,KAAK;AACjC,WAAK,EAAE,WAAW,CAAC;AACnB,UAAI,KAAK,KAAK,GAAG,QAAQ,MAAM;AAAA,IACjC;AACA,WAAO;AAAA,EACT;AACA,QAAM,SAAS,CAAC,GAAG,SAAS,EAAE,KAAK,CAAC,GAAG,MAAM,KAAK,EAAE,EAAE,IAAI,KAAK,EAAE,EAAE,CAAC;AACpE,QAAM,WAAW,KAAK,IAAI,GAAG,KAAK,IAAI,OAAO,SAAS,GAAG,KAAK,MAAM,OAAO,SAAS,QAAQ,CAAC,CAAC;AAC9F,SAAO;AAAA,IACL,SAAS,OAAO,MAAM,GAAG,QAAQ;AAAA,IACjC,OAAO,OAAO,MAAM,QAAQ;AAAA,EAC9B;AACF;AAEA,SAAS,cAAc,YAGrB;AACA,QAAM,cAAsC,CAAC;AAC7C,QAAM,SAAmB,CAAC;AAC1B,aAAW,CAAC,IAAI,GAAG,KAAK,OAAO,QAAQ,UAAU,GAAG;AAClD,gBAAY,EAAE,IAAI,IAAI;AACtB,WAAO,KAAK,IAAI,aAAa;AAAA,EAC/B;AACA,SAAO;AAAA,IACL,eAAe,OAAO,WAAW,IAAI,IAAI,OAAO,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,OAAO;AAAA,IACpF;AAAA,EACF;AACF;AAEA,IAAM,8BAA8B;AAAA,EAClC;AAAA,EACA;AAAA,EACA;AAAA,EACA;AACF;AA2BA,eAAsB,YACpB,MACkD;AAClD,QAAM,YAAY,KAAK,IAAI;AAE3B,QAAM,SAAS,KAAK,UAAU,CAAC;AAC/B,QAAM,cAAc,OAAO,eAAe;AAC1C,QAAM,iBAAiB,OAAO,kBAAkB;AAChD,QAAM,iBAAiB,OAAO,kBAAkB;AAChD,QAAM,kBAAkB,OAAO,mBAAmB;AAClD,QAAM,cAAc,OAAO;AAC3B,QAAM,cAAc,KAAK,eAAe;AAExC,QAAM,kBAAkB,OAAO;AAC/B,QAAM,EAAE,OAAO,QAAQ,IAAI,kBACvB;AAAA,IACE,OAAO,KAAK,UAAU,OAAO,CAAC,MAAM,CAAC,gBAAgB,KAAK,CAAC,MAAM,EAAE,OAAO,EAAE,EAAE,CAAC;AAAA,IAC/E,SAAS;AAAA,EACX,IACA,kBAAkB,KAAK,WAAW,eAAe;AAErD,MAAI,MAAM,WAAW,GAAG;AACtB,UAAM,IAAI;AAAA,MACR;AAAA,IACF;AAAA,EACF;AACA,MAAI,QAAQ,WAAW,GAAG;AACxB,UAAM,IAAI,MAAM,2DAA2D;AAAA,EAC7E;AAEA,QAAM,WACJ,KAAK,YACL,aAAa;AAAA,IACX,KAAK;AAAA,MACH,SAAS,KAAK,KAAK,WAAW;AAAA,MAC9B,QAAQ,KAAK,KAAK,UAAU,QAAQ,IAAI,kBAAkB;AAAA,IAC5D;AAAA,IACA,OAAO,KAAK,KAAK,SAAS;AAAA,IAC1B,QACE,KAAK,kBACL;AAAA,IACF,oBAAoB,KAAK,sBAAsB;AAAA,EACjD,CAAC;AAEH,QAAM,OACJ,KAAK,QACL,sBAA4C;AAAA,IAC1C,kBAAkB;AAAA,IAClB,gBAAgB;AAAA,EAClB,CAAC;AAMH,QAAM,SAAS,KAAK,UAAU,qBAAqB,SAAS;AAC5D,QAAM,cAAc,OAAO,WAAW,QAAQ;AAC9C,QAAM,UAAU,KAAK,YAAY,cAAc,wBAAwB,IAAI,kBAAkB;AAE7F,MAAI,KAAK,YAAY;AACnB,SAAK,WAAW,EAAE,MAAM,oBAAoB,WAAW,KAAK,UAAU,OAAO,CAAC;AAAA,EAChF;AAEA,QAAM,SAAS,MAAM,mBAAyC;AAAA,IAC5D,WAAW;AAAA,IACX,iBAAiB,KAAK;AAAA,IACtB,qBAAqB,KAAK;AAAA,IAC1B;AAAA,IACA,QAAQ,CAAC,KAAK,KAAK;AAAA,IACnB;AAAA,IACA,gBAAgB;AAAA,IAChB,aAAa,OAAO;AAAA,IACpB,MAAM,OAAO;AAAA,IACb,kBAAkB;AAAA,IAClB;AAAA,IACA,eAAe,KAAK,iBAAiB;AAAA,IACrC,SAAS,KAAK;AAAA,IACd,QAAQ,KAAK;AAAA,IACb;AAAA,IACA;AAAA,IACA;AAAA,IACA,eAAe,KAAK;AAAA,IACpB;AAAA,IACA;AAAA,IACA,cAAc,KAAK;AAAA,IACnB,eAAe,KAAK;AAAA,IACpB,mBAAmB,KAAK;AAAA,IACxB,UAAU,KAAK;AAAA,EACjB,CAAC;AAED,QAAM,WAAW,cAAc,OAAO,kBAAkB,WAAW,UAAU;AAC7E,QAAM,cAAc,cAAc,OAAO,gBAAgB,WAAW,UAAU;AAE9E,MAAI,KAAK,YAAY;AACnB,SAAK,WAAW;AAAA,MACd,MAAM;AAAA,MACN,eAAe,SAAS;AAAA,MACxB,YAAY,KAAK,IAAI,IAAI;AAAA,IAC3B,CAAC;AACD,SAAK,WAAW;AAAA,MACd,MAAM;AAAA,MACN,UAAU,OAAO,WAAW;AAAA,MAC5B,MAAM,YAAY,gBAAgB,SAAS;AAAA,IAC7C,CAAC;AAAA,EACH;AAEA,QAAM,YACJ,OAAO,iBAAiB,WAAW,eACnC,OAAO,YAAY;AAAA,IACjB,CAAC,KAAK,QACJ,MAAM,IAAI,SAAS,OAAO,CAAC,GAAG,OAAO,IAAI,GAAG,SAAS,WAAW,cAAc,CAAC;AAAA,IACjF;AAAA,EACF;AAMF,QAAM,UAAU,MAAM,YAAY;AAAA,IAChC,MAAM;AAAA,MACJ,GAAG,kBAAkB,OAAO,iBAAiB,OAAO,YAAY,QAAQ,KAAK,eAAe;AAAA,MAC5F,GAAG,kBAAkB,OAAO,gBAAgB,OAAO,UAAU,QAAQ,OAAO,aAAa;AAAA,IAC3F;AAAA,IACA,qBAAqB;AAAA,IACrB,sBAAsB;AAAA,EACxB,CAAC;AAID,QAAM,aAAa,KAAK,IAAI,IAAI;AAChC,QAAM,gBACJ,KAAK,uBAAuB,KAC5B,kBAAkB,OAAO,gBAAgB,OAAO,UAAU,QAAQ,OAAO,aAAa;AACxF,QAAM,EAAE,QAAQ,WAAW,IAAI,MAAM,mBAAyC;AAAA,IAC5E,OAAO,GAAG,MAAM,IAAI,SAAS;AAAA,IAC7B;AAAA,IACA,WAAW,IAAI,KAAK,SAAS,EAAE,YAAY;AAAA,IAC3C,iBAAiB,KAAK;AAAA,IACtB,eAAe,OAAO;AAAA,IACtB,aAAa,OAAO;AAAA,IACpB,iBAAiB,OAAO;AAAA,IACxB,MAAM,OAAO;AAAA,IACb,aAAa,OAAO,YAAY,IAAI,CAAC,OAAO;AAAA,MAC1C,iBAAiB,EAAE,OAAO;AAAA,MAC1B,YAAY,EAAE,OAAO;AAAA,MACrB,UAAU,EAAE,OAAO;AAAA,MACnB,UAAU,EAAE,SAAS,IAAI,CAAC,OAAO,EAAE,aAAa,EAAE,aAAa,SAAS,EAAE,QAAQ,EAAE;AAAA,IACtF,EAAE;AAAA,IACF,MAAM,OAAO;AAAA,IACb,mBAAmB,OAAO;AAAA,IAC1B,iBAAiB,OAAO;AAAA,IACxB;AAAA,IACA,cAAc;AAAA,IACd,iBAAiB;AAAA,IACjB;AAAA,IACA,cAAc,KAAK,eAAe,mBAAmB,KAAK,YAAY,IAAI;AAAA,EAC5E,CAAC;AACD,MAAI,KAAK,aAAc,MAAK,aAAa,UAAU;AAEnD,QAAM,UAAmD;AAAA,IACvD;AAAA,IACA,QAAQ;AAAA,MACN,GAAG;AAAA,MACH,SAAS,OAAO;AAAA,MAChB,GAAI,OAAO,cAAc,EAAE,OAAO,OAAO,YAAY,IAAI,CAAC;AAAA,MAC1D,GAAI,OAAO,kBAAkB,EAAE,WAAW,OAAO,gBAAgB,IAAI,CAAC;AAAA,IACxE;AAAA,IACA,MAAM,YAAY,gBAAgB,SAAS;AAAA,IAC3C,MAAM,OAAO;AAAA,IACb;AAAA,IACA,cAAc,OAAO,WAAW;AAAA,IAChC,qBAAqB,OAAO,YAAY;AAAA,IACxC;AAAA,IACA,cAAc;AAAA,IACd;AAAA,IACA,KAAK;AAAA,EACP;AAIA,MAAI,KAAK,cAAc;AACrB,QAAI;AACF,YAAM,oBAAoB,KAAK,cAAc,MAAM,SAAS,QAAQ,MAAM;AAAA,IAC5E,SAAS,KAAK;AACZ,YAAM,MAAM,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;AAE3D,cAAQ,KAAK,mDAAmD,GAAG,EAAE;AAAA,IACvE;AAAA,EACF;AAEA,SAAO;AACT;AAEA,eAAe,oBACb,QACA,MACA,SACA,KACA,QACe;AACf,QAAM,SAAS,mBAAmB,MAAM;AAExC,WAAS,qBACP,OACA,SACA,UACA,YAC2B;AAC3B,UAAM,QAA4B,SAAS,MAAM,IAAI,CAAC,SAAS;AAC7D,YAAM,cAAc,OAAO,OAAO,KAAK,WAAW;AAClD,YAAM,YACJ,YAAY,WAAW,IACnB,IACA,YAAY,OAAO,CAAC,GAAG,MAAM,IAAI,EAAE,WAAW,CAAC,IAAI,YAAY;AACrE,aAAO;AAAA,QACL,YAAY,KAAK;AAAA,QACjB,KAAK,KAAK;AAAA,QACV,eAAe;AAAA,QACf,YAAY,OAAO;AAAA,UACjB,OAAO,QAAQ,KAAK,WAAW,EAAE,IAAI,CAAC,CAAC,MAAM,KAAK,MAAM,CAAC,MAAM,MAAM,UAAU,CAAC;AAAA,QAClF;AAAA,QACA,cAAc,KAAK,SAAS;AAAA,MAC9B;AAAA,IACF,CAAC;AACD,UAAM,gBACJ,MAAM,WAAW,IAAI,IAAI,MAAM,OAAO,CAAC,GAAG,MAAM,IAAI,EAAE,eAAe,CAAC,IAAI,MAAM;AAClF,WAAO;AAAA,MACL;AAAA,MACA,aACE,OAAO,YAAY,WACf,WAAW,OAAO,IAClB,WAAW,KAAK,UAAU,WAAW,EAAE,CAAC;AAAA,MAC9C;AAAA,MACA;AAAA,MACA;AAAA,MACA,SAAS,SAAS,WAAW;AAAA,MAC7B;AAAA,IACF;AAAA,EACF;AAEA,QAAM,cAA2C,CAAC;AAElD,cAAY,KAAK,qBAAqB,GAAG,KAAK,iBAAiB,IAAI,kBAAkB,CAAC,CAAC;AAGvF,aAAW,OAAO,IAAI,aAAa;AACjC,UAAM,SAAS,IAAI,SAAS;AAAA,MAC1B,CAAC,MAAM,MACL,EAAE,SAAS,WAAW,gBAAgB,MACrC,SAAS,UAAa,iBAAiB,EAAE,QAAQ,IAAI,iBAAiB,KAAK,QAAQ,KAChF,IACA;AAAA,MACN,IAAI,SAAS,CAAC;AAAA,IAChB;AACA,QAAI,CAAC,OAAQ;AACb,gBAAY;AAAA,MACV,qBAAqB,IAAI,OAAO,kBAAkB,GAAG,OAAO,SAAS,OAAO,UAAU,CAAC;AAAA,IACzF;AAAA,EACF;AAEA,QAAM,QAAsB;AAAA,IAC1B,OAAO,GAAG,MAAM,IAAI,KAAK,IAAI,CAAC;AAAA,IAC9B;AAAA,IACA,YAAW,oBAAI,KAAK,GAAE,YAAY;AAAA,IAClC,QAAQ;AAAA,IACR,QAAQ,KAAK,gBAAgB,CAAC;AAAA,IAC9B,UAAU,YAAY,CAAC;AAAA,IACvB;AAAA,IACA,cAAc,QAAQ;AAAA,IACtB,aAAa,QAAQ;AAAA,IACrB,cAAc,QAAQ;AAAA,IACtB,iBAAiB,QAAQ;AAAA,IACzB,eAAe,QAAQ;AAAA,EACzB;AAEA,QAAM,OAAO,cAAc,KAAK;AAClC;AAEA,SAAS,iBACP,UACQ;AACR,QAAM,OAAO,OAAO,OAAO,SAAS,WAAW,UAAU;AACzD,SAAO,KAAK,WAAW,IAAI,IAAI,KAAK,OAAO,CAAC,GAAG,MAAM,IAAI,EAAE,eAAe,CAAC,IAAI,KAAK;AACtF;AAEA,SAAS,WAAW,GAAmB;AACrC,MAAI,IAAI,eAAe;AACvB,WAAS,IAAI,GAAG,IAAI,EAAE,QAAQ,KAAK;AACjC,SAAK,EAAE,WAAW,CAAC;AACnB,QAAI,KAAK,KAAK,GAAG,QAAQ,MAAM;AAAA,EACjC;AACA,SAAO,EAAE,SAAS,EAAE,EAAE,SAAS,GAAG,GAAG;AACvC;AAaA,SAAS,kBACP,OACA,aACA,OACA,SACa;AACb,QAAM,aAAa,mBAAmB,OAAO;AAC7C,QAAM,aAAa,UAAU,WAAW,QAAQ,EAAE,OAAO,WAAW,EAAE,OAAO,KAAK,CAAC;AACnF,SAAO,MAAM,IAAI,CAAC,SAAS;AACzB,UAAM,WAAmD,CAAC;AAC1D,UAAM,kBAA8D,CAAC;AACrE,QAAI,eAAe;AACnB,QAAI,iBAAiB;AACrB,eAAW,CAAC,SAAS,KAAK,KAAK,OAAO,QAAQ,KAAK,WAAW,GAAG;AAC/D,eAAS,OAAO,IAAI,EAAE,GAAG,MAAM,WAAW;AAC1C,iBAAW,CAAC,KAAK,KAAK,KAAK,OAAO,QAAQ,MAAM,UAAU,GAAG;AAC3D,YAAI,CAAC,OAAO,SAAS,KAAK,EAAG;AAC7B,cAAM,QAAQ,gBAAgB,GAAG,KAAK,EAAE,KAAK,GAAG,GAAG,EAAE;AACrD,cAAM,OAAO;AACb,cAAM,KAAK;AACX,wBAAgB,GAAG,IAAI;AAAA,MACzB;AACA,UAAI,OAAO,SAAS,MAAM,SAAS,GAAG;AACpC,wBAAgB,MAAM;AACtB,0BAAkB;AAAA,MACpB;AAAA,IACF;AACA,UAAM,aAAqC,CAAC;AAC5C,eAAW,CAAC,KAAK,EAAE,KAAK,EAAE,CAAC,KAAK,OAAO,QAAQ,eAAe,GAAG;AAC/D,iBAAW,GAAG,IAAI,MAAM,IAAI,IAAI,MAAM;AAAA,IACxC;AACA,UAAM,YAAY,mBAAmB,IAAI,IAAI,eAAe;AAC5D,UAAM,cAAiC;AAAA,MACrC;AAAA,MACA;AAAA,MACA;AAAA,IACF;AACA,WAAO;AAAA,MACL,OAAO,GAAG,KAAK,KAAK,WAAW,KAAK,KAAK,MAAM;AAAA,MAC/C,cAAc;AAAA,MACd;AAAA;AAAA;AAAA,MAGA,MACE,KAAK,MAAM,MACX,WAAW,KAAK,UAAU,EACvB,MAAM,GAAG,CAAC,EACV,MAAM,EAAE,EACR,OAAO,CAAC,GAAG,MAAO,IAAI,KAAK,EAAE,WAAW,CAAC,MAAO,GAAG,CAAC;AAAA,MACzD,OAAO;AAAA,MACP;AAAA,MACA;AAAA,MACA,WAAW;AAAA,MACX,QAAQ,KAAK;AAAA,MACb,SAAS,KAAK;AAAA,MACd,YAAY,EAAE,OAAO,GAAG,QAAQ,EAAE;AAAA,MAClC,SAAS;AAAA,QACP,cAAc;AAAA,QACd,KAAK,CAAC;AAAA,QACN;AAAA,MACF;AAAA,MACA,UAAU;AAAA,MACV,GAAI,KAAK,QAAQ,EAAE,aAAa,KAAK,MAAM,IAAI,CAAC;AAAA,IAClD;AAAA,EACF,CAAC;AACH;;;AC3oBO,SAAS,gBACd,UACwC;AACxC,QAAM,yBAAyB,iBAAiB,QAAQ;AAExD,SAAO;AAAA,IACL,WAAW,SAAS;AAAA,IACpB,iBAAiB,SAAS;AAAA,IAE1B,MAAM,SAAS,OAAO,CAAC,GAAG;AACxB,YAAM,EAAE,OAAO,OAAO,QAAQ,QAAQ,WAAW,SAAS,GAAG,aAAa,IAAI;AAC9E,YAAM,gBAAgB,SAAS,SAAS;AACxC,YAAM,kBAAkB,WAAW,SAAS;AAC5C,YAAM,iBAAiB,UAAU,SAAS,UAAU,yBAAyB,KAAK,IAAI,CAAC;AACvF,YAAM,kBACJ,aAAa,WACb,uBAAuB,YACtB,eAAe,WAAW,QAAQ,IAAI,wBAAwB,IAAI;AACrE,YAAM,cAAoD;AAAA,QACxD,GAAG;AAAA,QACH,GAAG;AAAA,QACH,GAAI,kBAAkB,EAAE,SAAS,gBAAgB,IAAI,CAAC;AAAA,QACtD,QAAQ;AAAA,QACR,WAAW,aAAa,SAAS;AAAA,QACjC,UAAU,CAAC,UAAU,QAAQ,cAAc,iBAAiB,UAAU,GAAG;AAAA,QACzE,QAAQ,eAAe,QAAQ,SAAS,SAAS,KAAK;AAAA,MACxD;AACA,UAAI,YAAY,SAAS;AACvB,oBAAY,OAAO,uBAAuB,YAAY,MAAM,MAAM;AACpE,aAAO,QAA8B,WAAW;AAAA,IAClD;AAAA,IAEA,MAAM,QAAQ,OAAO,CAAC,GAAG;AACvB,YAAM;AAAA,QACJ,QAAQ;AAAA,QACR,cAAc;AAAA,QACd,KAAK;AAAA,QACL,GAAG;AAAA,MACL,IAAI;AACJ,YAAM,SAAS,aAAa,UAAU,iBAAiB;AACvD,YAAM,SAAS,YAAY,SAAS,QAAQ,cAAc;AAC1D,YAAM,MAAM,SAAS,SAAS,KAAK,WAAW;AAC9C,YAAM,eAAe,kBAAkB,SAAS,cAAc,oBAAoB;AAClF,aAAO,YAAkC;AAAA,QACvC,GAAG;AAAA,QACH,GAAI,SAAS,EAAE,OAAO,IAAI,CAAC;AAAA,QAC3B,GAAI,MAAM,EAAE,IAAI,IAAI,CAAC;AAAA,QACrB,GAAI,eAAe,EAAE,aAAa,IAAI,CAAC;AAAA,MACzC,CAAC;AAAA,IACH;AAAA,EACF;AACF;AAOA,SAAS,iBACP,UAC8C;AAC9C,QAAM,MAAoD,CAAC;AAC3D,MAAI,SAAS,QAAS,KAAI,UAAU,SAAS;AAC7C,MAAI,SAAS,aAAc,KAAI,eAAe,SAAS;AACvD,MAAI,SAAS,cAAe,KAAI,gBAAgB,SAAS;AACzD,MAAI,SAAS,cAAe,KAAI,gBAAgB,SAAS;AACzD,MAAI,SAAS,YAAa,KAAI,cAAc,SAAS;AACrD,MAAI,SAAS,QAAQ,YAAY,OAAW,KAAI,cAAc,SAAS,OAAO;AAC9E,MAAI,SAAS,QAAQ,mBAAmB;AACtC,QAAI,iBAAiB,SAAS,OAAO;AACvC,MAAI,SAAS,QAAQ,SAAS;AAC5B,QAAI,OAAO,uBAAuB,SAAS,OAAO,MAAM,aAAa;AACvE,SAAO;AACT;AAEA,SAAS,YACP,UACA,WAC+B;AAC/B,QAAM,SAAS,oBAAoB,UAAU,SAAS;AACtD,MAAI,QAAQ,SAAS,OAAW,QAAO,OAAO,uBAAuB,OAAO,MAAM,aAAa;AAC/F,SAAO;AACT;AAEA,SAAS,SACP,UACA,WAC4B;AAC5B,SAAO,oBAAoB,UAAU,SAAS;AAChD;AAEA,SAAS,kBACP,UACA,WAC0B;AAC1B,QAAM,SAAS,oBAAoB,UAAU,SAAS;AACtD,MAAI,CAAC,OAAQ,QAAO;AACpB,MAAI,CAAC,OAAO,UAAU,KAAK,KAAK,CAAC,OAAO,QAAQ,KAAK,KAAK,CAAC,OAAO,UAAU,KAAK,GAAG;AAClF,UAAM,IAAI;AAAA,MACR;AAAA,IACF;AAAA,EACF;AACA,SAAO;AACT;AAEA,SAAS,aAA+B,UAAa,WAAsC;AACzF,MAAI,CAAC,UAAW,QAAO;AACvB,QAAM,SAAS,EAAE,GAAG,SAAS;AAC7B,aAAW,CAAC,KAAK,KAAK,KAAK,OAAO,QAAQ,SAAS,GAAG;AACpD,QAAI,UAAU,OAAW,QAAO,GAAG,IAAI;AAAA,EACzC;AACA,SAAO;AACT;AAEA,SAAS,oBACP,UACA,WACe;AACf,MAAI,CAAC,YAAY,CAAC,UAAW,QAAO;AACpC,SAAO,aAAa,YAAa,CAAC,GAAS,SAAS;AACtD;AAEA,SAAS,eACP,QACA,cACqC;AACrC,MAAI,WAAW,QAAW;AACxB,QAAI,OAAO,WAAW,GAAG;AACvB,YAAM,IAAI,MAAM,oDAAoD;AAAA,IACtE;AACA,WAAO;AAAA,EACT;AACA,SAAO,CAAC,YAAY;AACtB;AAEA,SAAS,uBAAuB,OAAe,OAAuB;AACpE,MAAI,CAAC,OAAO,UAAU,KAAK,KAAK,QAAQ,GAAG;AACzC,UAAM,IAAI,MAAM,oBAAoB,KAAK,6BAA6B;AAAA,EACxE;AACA,SAAO;AACT;;;AC9HA,SAAS,WAAW,MAAgC;AAIlD,SAAO,KAAK,UAAU,CAAC,KAAK,YAAY,KAAK,GAAG,CAAC;AACnD;AAIA,SAAS,eACP,QACA,OACkC;AAClC,QAAM,MAAwC,CAAC;AAC/C,QAAM,SAAS,oBAAI,IAAY,CAAC,GAAG,OAAO,KAAK,MAAM,GAAG,GAAG,OAAO,KAAK,KAAK,CAAC,CAAC;AAC9E,aAAW,SAAS,QAAQ;AAC1B,UAAM,aAAa,OAAO,KAAK,KAAK,CAAC;AACrC,UAAM,YAAY,MAAM,KAAK,KAAK,CAAC;AACnC,UAAM,OAAO,oBAAI,IAAY,CAAC,GAAG,OAAO,KAAK,UAAU,GAAG,GAAG,OAAO,KAAK,SAAS,CAAC,CAAC;AACpF,UAAM,WAA+C,CAAC;AACtD,eAAW,OAAO,MAAM;AAKtB,YAAM,YAAY,WAAW,GAAG;AAChC,YAAM,WAAW,UAAU,GAAG;AAC9B,YAAM,IAAI,OAAO,cAAc,YAAY,OAAO,SAAS,SAAS,IAAI,YAAY;AACpF,YAAM,IAAI,OAAO,aAAa,YAAY,OAAO,SAAS,QAAQ,IAAI,WAAW;AACjF,eAAS,GAAG,IAAI;AAAA,QACd,QAAQ;AAAA,QACR,OAAO;AAAA,QACP,OAAO,MAAM,QAAQ,MAAM,OAAO,IAAI,IAAI;AAAA,MAC5C;AAAA,IACF;AACA,QAAI,KAAK,IAAI;AAAA,EACf;AACA,SAAO;AACT;AASO,SAAS,gBACd,QACA,OACoB;AACpB,QAAM,YAAY,IAAI,IAAI,OAAO,MAAM,IAAI,CAAC,MAAM,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC,CAAC;AACrE,QAAM,WAAW,IAAI,IAAI,MAAM,MAAM,IAAI,CAAC,MAAM,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC,CAAC;AAEnE,QAAM,UAAgC,CAAC;AACvC,QAAM,UAA8B,CAAC;AACrC,QAAM,QAA4B,CAAC;AAEnC,aAAW,CAAC,KAAK,UAAU,KAAK,WAAW;AACzC,UAAM,YAAY,SAAS,IAAI,GAAG;AAClC,QAAI,CAAC,WAAW;AACd,cAAQ,KAAK,UAAU;AACvB;AAAA,IACF;AACA,YAAQ,KAAK;AAAA,MACX,YAAY,WAAW;AAAA,MACvB,KAAK,WAAW;AAAA,MAChB,iBAAiB,WAAW;AAAA,MAC5B,gBAAgB,UAAU;AAAA,MAC1B,gBAAgB,UAAU,gBAAgB,WAAW;AAAA,MACrD,YAAY,eAAe,WAAW,YAAY,UAAU,UAAU;AAAA,IACxE,CAAC;AAAA,EACH;AACA,aAAW,CAAC,KAAK,SAAS,KAAK,UAAU;AACvC,QAAI,CAAC,UAAU,IAAI,GAAG,EAAG,OAAM,KAAK,SAAS;AAAA,EAC/C;AAEA,SAAO;AAAA,IACL,aAAa,OAAO;AAAA,IACpB,YAAY,MAAM;AAAA,IAClB,mBAAmB,OAAO;AAAA,IAC1B,kBAAkB,MAAM;AAAA,IACxB,gBAAgB,OAAO,gBAAgB,MAAM;AAAA,IAC7C;AAAA,IACA;AAAA,IACA;AAAA,IACA,iBAAiB,OAAO;AAAA,IACxB,gBAAgB,MAAM;AAAA,IACtB,gBAAgB,MAAM,gBAAgB,OAAO;AAAA,IAC7C,eAAe,OAAO;AAAA,IACtB,cAAc,MAAM;AAAA,IACpB,cAAc,MAAM,UAAU,OAAO;AAAA,IACrC,kBAAkB,OAAO;AAAA,IACzB,iBAAiB,MAAM;AAAA,IACvB,iBAAiB,MAAM,aAAa,OAAO;AAAA,EAC7C;AACF;AAGA,SAAS,SAAS,KAAqD;AACrE,MAAI,IAAI,YAAY,WAAW,EAAG,QAAO;AACzC,MAAI,SAAS,IAAI,YAAY,CAAC;AAC9B,aAAW,OAAO,IAAI,aAAa;AACjC,QAAI,IAAI,QAAQ,OAAO,MAAO,UAAS;AAAA,EACzC;AACA,SAAO;AACT;AAOO,SAAS,SAAS,QAAsB,OAAkC;AAC/E,QAAM,eAAe,SAAS,MAAM;AACpC,QAAM,cAAc,SAAS,KAAK;AAClC,QAAM,eACJ,OAAO,YAAY,MAAM,WAAW,gBAAgB,OAAO,UAAU,MAAM,QAAQ,IAAI;AACzF,QAAM,cACJ,gBAAgB,cAAc,gBAAgB,cAAc,WAAW,IAAI;AAE7E,QAAM,aAAa,OAAO,eAAe;AACzC,QAAM,YAAY,MAAM,eAAe;AAEvC,SAAO;AAAA,IACL,aAAa,OAAO;AAAA,IACpB,YAAY,MAAM;AAAA,IAClB,iBAAiB,OAAO;AAAA,IACxB,gBAAgB,MAAM;AAAA,IACtB,oBAAoB,OAAO,gBAAgB;AAAA,IAC3C,mBAAmB,MAAM,gBAAgB;AAAA,IACzC,mBAAmB;AAAA,IACnB,kBAAkB;AAAA,IAClB,kBAAkB,eAAe,QAAQ,cAAc,OAAO,YAAY,aAAa;AAAA,IACvF,oBAAoB,OAAO;AAAA,IAC3B,mBAAmB,MAAM;AAAA,IACzB,mBAAmB,MAAM,eAAe,OAAO;AAAA,IAC/C,uBAAuB,OAAO;AAAA,IAC9B,sBAAsB,MAAM;AAAA,IAC5B,sBAAsB,MAAM,kBAAkB,OAAO;AAAA,IACrD;AAAA,IACA;AAAA,EACF;AACF;AAQO,SAAS,wBAAwB,KAA8C;AACpF,MAAI,CAAC,IAAI,SAAU,QAAO;AAC1B,QAAM,SAAS,SAAS,GAAG;AAC3B,MAAI,CAAC,UAAU,OAAO,UAAU,IAAI,SAAS,MAAO,QAAO;AAC3D,SAAO,gBAAgB,IAAI,UAAU,MAAM;AAC7C;;;AChLA,SAAS,WAAW,GAA4B;AAC9C,SAAO,KAAK,IAAI,GAAG,EAAE,WAAW,EAAE,aAAa,CAAC;AAClD;AAOO,SAAS,gBAAgB,SAA8C;AAS5E,QAAM,MAAM,oBAAI,IAAiB;AAEjC,aAAW,UAAU,SAAS;AAC5B,UAAM,MAAM,OAAO,KAAK;AACxB,QAAI,CAAC,IAAK;AAEV,QAAI,IAAI,IAAI,IAAI,GAAG;AACnB,QAAI,CAAC,GAAG;AACN,UAAI;AAAA,QACF,QAAQ,oBAAI,IAAI;AAAA,QAChB,OAAO,oBAAI,IAAI;AAAA,QACf,OAAO,oBAAI,IAAI;AAAA,QACf,mBAAmB;AAAA,QACnB,WAAW;AAAA,QACX,eAAe;AAAA,MACjB;AACA,UAAI,IAAI,KAAK,CAAC;AAAA,IAChB;AAEA,QAAI,OAAO,MAAM,KAAM,GAAE,MAAM,IAAI,OAAO,KAAK,IAAI;AAEnD,eAAW,QAAQ,OAAO,SAAS,CAAC,GAAG;AACrC,QAAE,MAAM,IAAI,KAAK,IAAI;AACrB,iBAAW,QAAQ,KAAK,iBAAiB,CAAC,GAAG;AAC3C,UAAE,qBAAqB;AACvB,mBAAW,SAAS,KAAK,UAAU,CAAC,GAAG;AAErC,gBAAM,cAAc,MAAM,eAAe,KAAK;AAC9C,YAAE,aAAa,WAAW,KAAK;AAC/B,cAAI,CAAC,YAAa;AAClB,cAAI,YAAY,SAAS,WAAW,YAAY,SAAS,SAAS;AAChE,cAAE,gBAAgB;AAAA,UACpB;AACA,eAAK,YAAY,SAAS,QAAQ,YAAY,SAAS,YAAY,YAAY,UAAU;AACvF,cAAE,OAAO,IAAI,YAAY,QAAQ;AAAA,UACnC;AAAA,QACF;AAAA,MACF;AAAA,IACF;AAAA,EACF;AAEA,QAAM,QAAyB,oBAAI,IAAI;AACvC,aAAW,CAAC,KAAK,CAAC,KAAK,KAAK;AAC1B,UAAM,IAAI,KAAK;AAAA,MACb,WAAW;AAAA,MACX,UAAU,CAAC,GAAG,EAAE,MAAM,EAAE,KAAK;AAAA,MAC7B,OAAO,CAAC,GAAG,EAAE,KAAK,EAAE,KAAK;AAAA,MACzB,mBAAmB,EAAE;AAAA,MACrB,WAAW,EAAE,MAAM;AAAA,MACnB,WAAW,EAAE;AAAA,MACb,eAAe,EAAE;AAAA,IACnB,CAAC;AAAA,EACH;AACA,SAAO;AACT;AAoBO,SAAS,8BACd,MACA,OACiC;AACjC,QAAM,UAAU,oBAAI,IAAyB;AAC7C,QAAM,eAA4B,CAAC;AAEnC,aAAW,OAAO,MAAM;AACtB,UAAM,aAAa,MAAM,IAAI,IAAI,SAAS;AAC1C,QAAI,CAAC,cAAc,WAAW,SAAS,WAAW,GAAG;AACnD,mBAAa,KAAK,GAAG;AACrB;AAAA,IACF;AACA,eAAW,SAAS,WAAW,UAAU;AACvC,YAAM,SAAS,QAAQ,IAAI,KAAK,KAAK,CAAC;AACtC,aAAO,KAAK,GAAG;AACf,cAAQ,IAAI,OAAO,MAAM;AAAA,IAC3B;AAAA,EACF;AAEA,SAAO,EAAE,SAAS,aAAa;AACjC;;;AC/GO,SAAS,kBAAkB,MAAyD;AACzF,QAAM,EAAE,SAAS,OAAO,CAAC,GAAG,OAAO,kBAAkB,KAAK,IAAI;AAC9D,QAAM,YAAY,IAAI,IAAI,KAAK,IAAI,CAAC,MAAM,CAAC,EAAE,OAAO,CAAC,CAAC,CAAC;AAGvD,QAAM,YAAY,CAAC,WAAqC;AACtD,QAAI,OAAO,WAAW,UAAW,QAAO,SAAS,IAAI;AACrD,QAAI,CAAC,OAAO,SAAS,MAAM,EAAG,QAAO,OAAO;AAC5C,QAAI,CAAC,MAAO,QAAO;AACnB,UAAM,EAAE,KAAK,IAAI,IAAI;AACrB,QAAI,QAAQ,IAAK,QAAO;AACxB,YAAQ,SAAS,QAAQ,MAAM;AAAA,EACjC;AAGA,QAAM,QAAQ,oBAAI,IAAgC;AAClD,aAAW,OAAO,SAAS;AACzB,UAAM,OAAO,MAAM,IAAI,IAAI,KAAK,KAAK,CAAC;AACtC,SAAK,KAAK,GAAG;AACb,UAAM,IAAI,IAAI,OAAO,IAAI;AAAA,EAC3B;AAEA,QAAM,OAAoB,CAAC;AAC3B,QAAM,cAAsD,CAAC;AAE7D,aAAW,CAAC,OAAO,UAAU,KAAK,OAAO;AACvC,UAAM,aAAa,WAChB,IAAI,CAAC,OAAO,EAAE,OAAO,EAAE,OAAO,OAAO,UAAU,EAAE,MAAM,EAAE,EAAE,EAC3D,OAAO,CAAC,MAAM,OAAO,SAAS,EAAE,KAAK,CAAC;AACzC,QAAI,WAAW,WAAW,EAAG;AAE7B,UAAM,YAAY,WAAW,OAAO,CAAC,GAAG,MAAM,IAAI,EAAE,OAAO,CAAC,IAAI,WAAW;AAE3E,UAAM,UAAU,UAAU,IAAI,KAAK,KAAM,EAAE,MAAM;AAEjD,UAAM,cAAiC;AAAA,MACrC,UAAU,OAAO,YAAY,WAAW,IAAI,CAAC,MAAM,CAAC,EAAE,OAAO,EAAE,QAAQ,EAAE,MAAM,CAAC,CAAC,CAAC;AAAA,MAClF,YAAY,EAAE,QAAQ,UAAU;AAAA,MAChC,WAAW;AAAA,IACb;AAEA,UAAM,UAAsB;AAAA;AAAA;AAAA;AAAA,MAI1B,cAAc;AAAA,MACd,KAAK,OAAO,YAAY,WAAW,IAAI,CAAC,MAAM,CAAC,SAAS,EAAE,KAAK,IAAI,EAAE,KAAK,CAAC,CAAC;AAAA,MAC5E;AAAA,IACF;AAEA,SAAK,KAAK;AAAA,MACR;AAAA,MACA,cAAc,QAAQ,gBAAgB;AAAA,MACtC,aAAa,QAAQ,eAAe;AAAA,MACpC,MAAM;AAAA,MACN,OAAO,QAAQ,SAAS;AAAA,MACxB,YAAY,QAAQ,cAAc;AAAA,MAClC,YAAY,QAAQ,cAAc;AAAA,MAClC,WAAW,QAAQ,aAAa;AAAA,MAChC,QAAQ,QAAQ,UAAU;AAAA,MAC1B,SAAS,QAAQ,WAAW;AAAA,MAC5B,YAAY,EAAE,OAAO,GAAG,QAAQ,EAAE;AAAA,MAClC;AAAA,MACA,UAAU,QAAQ,YAAY;AAAA,IAChC,CAAc;AAEd,QAAI,iBAAiB;AACnB,iBAAW,KAAK,WAAY,aAAY,KAAK,EAAE,OAAO,OAAO,EAAE,OAAO,OAAO,EAAE,MAAM,CAAC;AAAA,IACxF;AAAA,EACF;AAEA,SAAO,EAAE,MAAM,YAAY;AAC7B;;;ACvHA,IAAM,aAAa,CAAC,gBAAgB,cAAc,OAAO;AACzD,IAAM,aAAa,CAAC,gBAAgB,GAAG,qBAAqB,OAAO;AACnE,IAAM,YAAY,CAAC,mBAAmB,yBAAyB,YAAY,MAAM;AACjF,IAAM,mBAAmB,CAAC,GAAG,2BAA2B,oBAAoB,WAAW;AACvF,IAAM,oBAAoB,CAAC,GAAG,4BAA4B,qBAAqB,YAAY;AAC3F,IAAM,mBAAmB,CAAC,sBAAsB,aAAa;AAC7D,IAAM,mBAAmB,CAAC,sBAAsB,aAAa;AAUtD,SAAS,cAAc,MAAyC;AACrE,QAAM,EAAE,OAAO,eAAe,WAAW,eAAe,cAAc,IAAI;AAC1E,QAAM,UAAU,WAAW,KAAK;AAEhC,QAAM,OAAoB,CAAC;AAC3B,aAAW,CAAC,UAAUA,WAAU,KAAK,SAAS;AAC5C,UAAM,OAAO,SAASA,WAAU;AAChC,QAAI,CAAC,KAAM;AAEX,UAAM,SAAS,KAAK,IAAI,IAAI,KAAK,kBAAkB,KAAK,qBAAqB,GAAS;AACtF,UAAM,QAAQ,eAAeA,aAAY,UAAU,KAAK;AACxD,UAAM,UAAU,eAAeA,aAAY,SAAS,KAAK;AACzD,UAAM,cAAc,eAAeA,aAAY,gBAAgB,KAAK;AACpE,UAAM,eAAe,eAAeA,aAAY,iBAAiB,KAAK;AACtE,UAAM,aAAa,eAAeA,aAAY,gBAAgB,KAAK;AACnE,UAAM,aAAa,eAAeA,aAAY,gBAAgB,KAAK;AACnE,UAAM,QAAQ,eAAeA,aAAY,UAAU;AAEnD,UAAM,aAAa,oBAAoBA,WAAU;AACjD,UAAM,aAA4B;AAAA,MAChC,OAAO;AAAA,MACP,QAAQ;AAAA,IACV;AAEA,UAAM,cACJ,UAAU,SACN;AAAA,MACE,UAAU,EAAE,gBAAgB,EAAE,MAAM,EAAE;AAAA,MACtC,YAAY,EAAE,MAAM;AAAA,MACpB,WAAW;AAAA,IACb,IACA;AAEN,UAAM,YAAYA,YAAW,KAAK,CAAC,MAAM,EAAE,QAAQ,SAAS,OAAO;AACnE,UAAM,UAAsB;AAAA,MAC1B,GAAI,KAAK,iBAAiB,WAAW,EAAE,aAAa,MAAM,IAAI,EAAE,cAAc,MAAM;AAAA,MACpF,KAAK;AAAA,MACL,GAAI,cAAc,EAAE,YAAY,IAAI,CAAC;AAAA,IACvC;AAEA,SAAK,KAAK;AAAA,MACR,OAAO;AAAA,MACP;AAAA,MACA,aAAc,KAAK,WAAW,oBAAoB,KAA4B;AAAA,MAC9E,MAAM;AAAA,MACN;AAAA,MACA;AAAA,MACA;AAAA,MACA,WAAY,KAAK,WAAW,mBAAmB,KAA4B;AAAA,MAC3E;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA,UAAU;AAAA,MACV,GAAI,YAAY,EAAE,aAAa,UAAU,KAAK,IAAI,CAAC;AAAA,IACrD,CAAc;AAAA,EAChB;AACA,SAAO;AACT;AAIA,SAAS,WAAW,OAAwD;AAC1E,QAAM,IAAI,oBAAI,IAA8B;AAC5C,aAAW,QAAQ,OAAO;AACxB,UAAM,MAAO,KAAK,cAAc,KAA4B,KAAK;AACjE,UAAM,OAAO,EAAE,IAAI,GAAG,KAAK,CAAC;AAC5B,SAAK,KAAK,IAAI;AACd,MAAE,IAAI,KAAK,IAAI;AAAA,EACjB;AACA,SAAO;AACT;AAEA,SAAS,SAAS,OAAqD;AACrE,SAAO,MAAM,KAAK,CAAC,MAAM,CAAC,EAAE,YAAY,KAAK,MAAM,CAAC;AACtD;AAEA,SAAS,eAAe,OAAyB,MAAoC;AACnF,aAAW,QAAQ,OAAO;AACxB,eAAW,OAAO,MAAM;AACtB,YAAM,IAAI,KAAK,WAAW,GAAG;AAC7B,UAAI,OAAO,MAAM,YAAY,EAAE,SAAS,EAAG,QAAO;AAAA,IACpD;AAAA,EACF;AACA,SAAO;AACT;AAEA,SAAS,eAAe,OAAyB,MAAoC;AACnF,aAAW,QAAQ,OAAO;AACxB,eAAW,OAAO,MAAM;AACtB,YAAM,IAAI,KAAK,WAAW,GAAG;AAC7B,UAAI,OAAO,MAAM,YAAY,OAAO,SAAS,CAAC,EAAG,QAAO;AACxD,UAAI,OAAO,MAAM,UAAU;AACzB,cAAM,SAAS,OAAO,CAAC;AACvB,YAAI,OAAO,SAAS,MAAM,EAAG,QAAO;AAAA,MACtC;AAAA,IACF;AAAA,EACF;AACA,SAAO;AACT;AAEA,SAAS,oBAAoB,OAAiD;AAC5E,QAAM,MAA8B,CAAC;AACrC,aAAW,QAAQ,OAAO;AACxB,eAAW,CAAC,GAAG,CAAC,KAAK,OAAO,QAAQ,KAAK,UAAU,GAAG;AACpD,UAAI,OAAO,MAAM,YAAY,OAAO,SAAS,CAAC,EAAG,KAAI,CAAC,IAAI;AAAA,IAC5D;AAAA,EACF;AACA,SAAO;AACT;","names":["groupSpans"]}
+{"version":3,"sources":["../../src/contract/self-improve.ts","../../src/contract/define-agent-eval.ts","../../src/contract/eval-reporting-suite.ts","../../src/contract/intake/run-record-dir.ts","../../src/contract/diff.ts","../../src/contract/intake/agent-trace.ts","../../src/contract/intake/feedback-table.ts","../../src/contract/intake/otel-spans.ts"],"sourcesContent":["/**\n * # `selfImprove()` - the one-call improvement loop.\n *\n * The cheapest possible call site to run a real closed-loop self-\n * improvement over your agent. Wraps `runImprovementLoop` with smart\n * defaults and a budget-shaped options API; every escape hatch the\n * substrate exposes is reachable from here without losing the\n * one-function feel.\n *\n * Defaults:\n *   - In-memory storage (no filesystem touch).\n *   - `gepaProposer` reflective mutation with copywriting-flavored primitives\n *     (override `proposer` or `mutationPrimitives` for any domain).\n *   - `defaultProductionGate` with `deltaThreshold: 0.05`.\n *   - Held-out split = 25% of scenarios, deterministic by id hash.\n *   - 3 generations × population 2 (raise via `budget` for more search).\n *   - `autoOnPromote: 'none'` (we don't open PRs unless you ask).\n *\n * Want one-click? Provide `agent` + `scenarios` + `judge`. Done.\n * Want distributed? Pass `cellPlacement` + an `httpDispatch`-backed\n * agent. Want a code-tier surface? Pass a `MutableSurface` + your own\n * `proposer`. Same function.\n */\n\nimport { createHash } from 'node:crypto'\nimport { defaultProductionGate } from '../campaign/gates/default-production-gate'\nimport {\n  type RunImprovementLoopResult,\n  runImprovementLoop,\n} from '../campaign/presets/run-improvement-loop'\nimport type { RunOptimizationOptions } from '../campaign/presets/run-optimization'\nimport { gepaProposer } from '../campaign/proposers/gepa'\nimport {\n  emitLoopProvenance,\n  type LoopProvenanceRecord,\n  surfaceContentHash,\n} from '../campaign/provenance'\nimport {\n  type CampaignStorage,\n  fsCampaignStorage,\n  inMemoryCampaignStorage,\n} from '../campaign/storage'\nimport type {\n  CampaignCellResult,\n  DispatchContext,\n  Gate,\n  JudgeConfig,\n  LabeledScenarioStore,\n  MutableSurface,\n  Scenario,\n  SurfaceProposer,\n} from '../campaign/types'\nimport { createHostedClient, type HostedTenant } from '../hosted/client'\nimport type { EvalRunCellScore, EvalRunEvent, EvalRunGenerationSnapshot } from '../hosted/types'\nimport type { JudgeScoresRecord, RunRecord } from '../run-record'\nimport { analyzeRuns } from './analyze-runs'\nimport type { InsightReport } from './insight-report'\n\nexport interface SelfImproveBudget {\n  /** Hard $ ceiling across all cells in baseline + every generation. Cells\n   *  beyond the ceiling are skipped (cost-aware, not aborted). */\n  dollars?: number\n  /** How many improvement generations to explore. Default 3. Set 0 to\n   *  skip improvement entirely (selfImprove becomes a baseline-only run). */\n  generations?: number\n  /** Candidates the proposer emits per generation. Default 2. */\n  populationSize?: number\n  /** Max concurrent cells across the loop. Default 2. */\n  maxConcurrency?: number\n  /** Fraction of `scenarios` held out from training, used for the gate.\n   *  Default 0.25. Ignored when `holdoutScenarios` is set explicitly. */\n  holdoutFraction?: number\n  /** Explicit held-out scenarios; overrides `holdoutFraction`. */\n  holdoutScenarios?: Scenario[]\n  /** Per-scenario replicates per cell — raises bootstrap-CI tightness. Default 1. */\n  reps?: number\n  /** Top-scoring candidates carried into the next generation. Default 2. */\n  promoteTopK?: number\n}\n\nexport interface SelfImproveLlm {\n  /** Endpoint base URL. Default Tangle Router. */\n  baseUrl?: string\n  /** Bearer token. Default `process.env.OPENAI_API_KEY`. */\n  apiKey?: string\n  /** Model id used by `gepaProposer` reflection. Default\n   *  `anthropic/claude-sonnet-4.6`. */\n  model?: string\n}\n\nexport type SelfImproveProgressEvent =\n  | { kind: 'baseline.started'; scenarios: number }\n  | { kind: 'baseline.completed'; compositeMean: number; durationMs: number }\n  | { kind: 'generation.started'; index: number; populationSize: number }\n  | { kind: 'generation.completed'; index: number; bestComposite: number; durationMs: number }\n  | { kind: 'gate.decided'; decision: string; lift: number }\n\nexport interface SelfImproveOptions<TScenario extends Scenario, TArtifact> {\n  /**\n   * Your agent — a function that takes the current `MutableSurface`\n   * (typically a system prompt the loop is optimizing) plus the\n   * scenario + cell ctx, and returns the artifact your judge scores.\n   *\n   * Same shape as `RunOptimizationOptions.dispatchWithSurface`. Wrap a\n   * plain `Dispatch` if you don't have a surface seam:\n   *\n   *   agent: (_surface, scenario, ctx) => yourPlainDispatch(scenario, ctx)\n   *\n   * That mode evaluates without mutating any surface — useful as a\n   * baseline-only run (set `budget.generations = 0`).\n   */\n  agent: (surface: MutableSurface, scenario: TScenario, ctx: DispatchContext) => Promise<TArtifact>\n\n  /** Scenarios to evaluate against. Train/holdout split is computed from\n   *  these unless `budget.holdoutScenarios` is set explicitly. */\n  scenarios: TScenario[]\n\n  /** Judge that scores artifacts. Bring your own; use `langchainJudge`\n   *  from `/adapters/langchain` for a Runnable-shaped one. */\n  judge: JudgeConfig<TArtifact, TScenario>\n\n  /** Starting surface — system prompt, JSON config, anything `MutableSurface`\n   *  accepts. The proposer mutates this each generation. */\n  baselineSurface: MutableSurface\n\n  /** Budget + loop shape. All fields optional. */\n  budget?: SelfImproveBudget\n\n  /** Custom surface proposer. Default is `gepaProposer` configured from `llm` +\n   *  `mutationPrimitives`. */\n  proposer?: SurfaceProposer\n\n  /** Default-proposer overrides — used when `proposer` is unset. */\n  mutationPrimitives?: string[]\n  proposerTarget?: string\n\n  /** Custom gate. Default is `defaultProductionGate` with\n   *  `deltaThreshold: 0.05` on the held-out split. */\n  gate?: Gate<TArtifact, TScenario>\n\n  /** LLM config consumed by the default `gepaProposer`. Ignored if you pass\n   *  your own `proposer`. */\n  llm?: SelfImproveLlm\n\n  /** Storage backend. Default is DURABLE: when a real (non-`mem://`) `runDir`\n   *  is available, the substrate defaults to `fsCampaignStorage()` so the\n   *  provenance record + OTel spans survive the call. Pass\n   *  `inMemoryCampaignStorage()` explicitly to opt OUT (tests, edge runtimes).\n   *  Default when `runDir` is `mem://...` (or unset): in-memory. */\n  storage?: CampaignStorage\n\n  /** Run directory (logical for in-memory storage, real path for fs).\n   *  Default `mem://selfImprove-<timestamp>` (in-memory, non-durable). Pass a\n   *  real path to persist the provenance record + spans. */\n  runDir?: string\n\n  /**\n   * Worker call records for backend provenance. The agent is opaque to the\n   * substrate (it returns an artifact, not token usage), so to capture an\n   * `assertRealBackend`-grade verdict + worker call count + model in the\n   * provenance record, the agent reports its per-call `RunRecord`s here.\n   * Called once after the loop; return the records the agent accumulated.\n   * When unset, backend provenance is derived from campaign cells (cost only;\n   * verdict will read `stub` without token usage — the honest signal that no\n   * token channel was wired).\n   */\n  collectWorkerRecords?: () => RunRecord[]\n\n  /** Fires once the durable provenance record + OTel spans are emitted.\n   *  Receives the structured record for inline assertions / custom routing. */\n  onProvenance?: (record: LoopProvenanceRecord) => void\n\n  /** Distributed execution seam — same as `RunCampaignOptions.cellPlacement`.\n   *  Returns an opaque placement key the substrate forwards to your agent\n   *  as `ctx.placement`. Combined with `httpDispatch` from\n   *  `/adapters/http`, fans cells across regions. */\n  cellPlacement?: (input: {\n    scenario: TScenario\n    rep: number\n    generation?: number\n  }) => string | undefined\n\n  /** Streaming hook — fires on baseline + each generation + gate decision.\n   *  Consumer routes events wherever (UI, dashboard, logs). */\n  onProgress?: (event: SelfImproveProgressEvent) => void\n\n  /** Auto-promotion behavior on a ship decision. Default `'none'` — we\n   *  return the winner; you ship it however you ship. `'pr'` opens a\n   *  GitHub PR via `openAutoPr`; requires `ghOwner` + `ghRepo`. */\n  autoOnPromote?: 'pr' | 'none'\n  ghOwner?: string\n  ghRepo?: string\n\n  /**\n   * Opt-in: ship eval-run events to a hosted orchestrator (ours, your\n   * self-hosted one, or any compatible implementation of the\n   * `docs/hosted-ingest-spec.md` wire format). When set, the substrate\n   * POSTs the final `EvalRunEvent` to `${endpoint}/v1/ingest/eval-runs`\n   * after the loop completes. Failures are logged but do not fail the\n   * loop — local result is always returned.\n   *\n   * For our orchestrator: `{ endpoint: 'https://orchestrator.tangle.tools/v1', apiKey, tenantId }`.\n   *\n   * For your self-hosted: any URL serving the wire format. See\n   * `examples/hosted-ingest-server/` for the reference receiver.\n   */\n  hostedTenant?: HostedTenant\n\n  /** Free-form labels attached to the hosted event (env, branch, model id,\n   *  etc.). Ignored when `hostedTenant` is unset. */\n  hostedLabels?: Record<string, string>\n\n  /** Capture every artifact + judge score to this store (labeled-example\n   *  corpus the proposer may read for few-shot, and the dataset you ship). Pass\n   *  `'off'` to disable. Default: off. */\n  labeledStore?: LabeledScenarioStore | 'off'\n\n  /** Capture-source tag for `labeledStore`. Default `'eval-run'`. */\n  captureSource?: 'production-trace' | 'eval-run' | 'manual' | 'red-team' | 'synthetic'\n\n  /**\n   * Per-cell backend-integrity expectation — the fail-loud guard. A cell that\n   * produced an artifact but reported `costUsd === 0` AND zero tokens is a\n   * stub. Modes: `'assert'` throws on the first such cell, `'warn'` logs it,\n   * `'off'` skips the check (offline/replay). Default `'assert'` — `selfImprove`\n   * is the real-run path, so a stub fails loud rather than scoring a clean 0.\n   */\n  expectUsage?: 'assert' | 'warn' | 'off'\n\n  /**\n   * Per-generation findings producer. After each\n   * generation is scored, this is called; whatever it returns REPLACES the\n   * proposer's `findings` for the next generation's `propose()`. Plug a\n   * trace-analyst registry / HALO here. When absent, findings stay\n   * `opts.findings`.\n   */\n  analyzeGeneration?: RunOptimizationOptions<TScenario, TArtifact>['analyzeGeneration']\n\n  /** Static findings forwarded to the proposer's `propose()` as `ctx.findings`\n   *  (a findings-grounded proposer consumes them). Default: none. */\n  findings?: unknown[]\n}\n\nexport interface SelfImproveResult<TScenario extends Scenario, TArtifact> {\n  /** Composite mean across all scenarios, baseline run. */\n  baseline: {\n    compositeMean: number\n    perScenario: Record<string, number>\n  }\n  /** Composite mean on the held-out set, winner run. */\n  winner: {\n    compositeMean: number\n    perScenario: Record<string, number>\n    surface: MutableSurface\n    /** Proposer label for the promoted change. Absent ⇒ winner == baseline or\n     *  a bare-surface mutator. */\n    label?: string\n    /** Proposer rationale — the \"because Z\" that motivated the promoted change.\n     *  Threaded from the proposer's `ProposedCandidate` through the loop.\n     *  Absent ⇒ winner == baseline. */\n    rationale?: string\n  }\n  /** `winner.compositeMean - baselineOnHoldout.compositeMean`. Positive\n   *  means the gate observed improvement. */\n  lift: number\n  /** The explicit baseline→winner unified diff. Always present (empty string\n   *  when winner == baseline). */\n  diff: string\n  /** Durable, queryable provenance record: candidate→cell→gate→promote chain +\n   *  rationale + diff + backend provenance. The artifact the hosted ingest\n   *  path stores; the +lift RECOMPUTES from `record.heldOutLift`. */\n  provenance: LoopProvenanceRecord\n  /** `defaultProductionGate.decide()` result. */\n  gateDecision: 'ship' | 'hold' | 'need_more_work' | 'model_ceiling' | 'arch_ceiling'\n  /** Number of generations actually explored (may be less than the\n   *  budget if the proposer gave up early). */\n  generationsExplored: number\n  /** Wall-clock total. */\n  durationMs: number\n  /** Total cost across baseline + every generation. */\n  totalCostUsd: number\n  /**\n   * Rigor packet: distributional summary, paired-bootstrap lift CI,\n   * judge stats, contamination check, recommendations. Wired through\n   * `analyzeRuns()` on the baseline + winner cells of the campaign.\n   * Hosted-tier dashboards render this as the v3-vs-v4 decision view.\n   */\n  insight: InsightReport\n  /**\n   * Raw substrate result for advanced inspection — full per-generation\n   * candidates, full campaign artifacts, all judge scores. Useful for\n   * debugging or reporting beyond the summary.\n   */\n  raw: RunImprovementLoopResult<TArtifact, TScenario>\n}\n\n/**\n * Deterministic train/holdout split by a stable hash of `scenario.id`,\n * so the same scenario set always splits the same way across runs.\n */\nfunction splitTrainHoldout<TScenario extends Scenario>(\n  scenarios: TScenario[],\n  fraction: number,\n): { train: TScenario[]; holdout: TScenario[] } {\n  // Stable fnv-1a-ish hash of the id for ordering.\n  function hash(s: string): number {\n    let h = 2166136261 >>> 0\n    for (let i = 0; i < s.length; i++) {\n      h ^= s.charCodeAt(i)\n      h = Math.imul(h, 16777619) >>> 0\n    }\n    return h\n  }\n  const sorted = [...scenarios].sort((a, b) => hash(a.id) - hash(b.id))\n  const nHoldout = Math.max(1, Math.min(sorted.length - 1, Math.round(sorted.length * fraction)))\n  return {\n    holdout: sorted.slice(0, nHoldout),\n    train: sorted.slice(nHoldout),\n  }\n}\n\nfunction meanComposite(byScenario: Record<string, { meanComposite: number }>): {\n  compositeMean: number\n  perScenario: Record<string, number>\n} {\n  const perScenario: Record<string, number> = {}\n  const values: number[] = []\n  for (const [id, agg] of Object.entries(byScenario)) {\n    perScenario[id] = agg.meanComposite\n    values.push(agg.meanComposite)\n  }\n  return {\n    compositeMean: values.length === 0 ? 0 : values.reduce((s, v) => s + v, 0) / values.length,\n    perScenario,\n  }\n}\n\nconst DEFAULT_MUTATION_PRIMITIVES = [\n  'Tighten the hook: lead with the specific user outcome.',\n  'Replace generic adjectives with specific verbs or proof numbers.',\n  \"Anchor every claim in something the scenario's brief literally supports.\",\n  'Honor the surface-shape constraint (length, register, audience vocabulary).',\n]\n\n/**\n * One-shot self-improvement loop. See module docstring for defaults +\n * extension points.\n *\n * @example Minimum:\n *\n *   const result = await selfImprove({\n *     agent: (surface, scenario, ctx) => myAgent(surface, scenario, ctx.signal),\n *     scenarios,\n *     judge,\n *     baselineSurface: DEFAULT_PROMPT,\n *   })\n *   console.log(`lift: ${result.lift.toFixed(3)} (${result.gateDecision})`)\n *\n * @example Distributed (workers in three regions):\n *\n *   await selfImprove({\n *     agent: httpDispatch({ resolveUrl: ({ placement }) => REGION_URLS[placement!] }),\n *     scenarios,\n *     judge,\n *     baselineSurface: DEFAULT_PROMPT,\n *     cellPlacement: ({ scenario }) => scenario.region,\n *     budget: { maxConcurrency: 12 },\n *   })\n */\nexport async function selfImprove<TScenario extends Scenario, TArtifact>(\n  opts: SelfImproveOptions<TScenario, TArtifact>,\n): Promise<SelfImproveResult<TScenario, TArtifact>> {\n  const startedAt = Date.now()\n\n  const budget = opts.budget ?? {}\n  const generations = budget.generations ?? 3\n  const populationSize = budget.populationSize ?? 2\n  const maxConcurrency = budget.maxConcurrency ?? 2\n  const holdoutFraction = budget.holdoutFraction ?? 0.25\n  const costCeiling = budget.dollars\n  const expectUsage = opts.expectUsage ?? 'assert'\n\n  const explicitHoldout = budget.holdoutScenarios\n  const { train, holdout } = explicitHoldout\n    ? {\n        train: opts.scenarios.filter((s) => !explicitHoldout.some((h) => h.id === s.id)),\n        holdout: explicitHoldout as TScenario[],\n      }\n    : splitTrainHoldout(opts.scenarios, holdoutFraction)\n\n  if (train.length === 0) {\n    throw new Error(\n      'selfImprove: train split is empty. Reduce holdoutFraction or pass more scenarios.',\n    )\n  }\n  if (holdout.length === 0) {\n    throw new Error('selfImprove: holdout split is empty. Pass more scenarios.')\n  }\n\n  const proposer: SurfaceProposer =\n    opts.proposer ??\n    gepaProposer({\n      llm: {\n        baseUrl: opts.llm?.baseUrl ?? 'https://router.tangle.tools/v1',\n        apiKey: opts.llm?.apiKey ?? process.env.OPENAI_API_KEY ?? '',\n      },\n      model: opts.llm?.model ?? 'anthropic/claude-sonnet-4.6',\n      target:\n        opts.proposerTarget ??\n        'agent surface (system prompt or config) being optimized by selfImprove',\n      mutationPrimitives: opts.mutationPrimitives ?? DEFAULT_MUTATION_PRIMITIVES,\n    })\n\n  const gate: Gate<TArtifact, TScenario> =\n    opts.gate ??\n    defaultProductionGate<TArtifact, TScenario>({\n      holdoutScenarios: holdout,\n      deltaThreshold: 0.05,\n    })\n\n  // Durable by default: a real (non-`mem://`) runDir means the caller wants\n  // persistence, so default to fs storage — the provenance record + spans\n  // survive the call. A `mem://` runDir (or none) stays in-memory. An explicit\n  // `storage` always wins (the opt-out path for tests / edge runtimes).\n  const runDir = opts.runDir ?? `mem://selfImprove-${startedAt}`\n  const isMemRunDir = runDir.startsWith('mem://')\n  const storage = opts.storage ?? (isMemRunDir ? inMemoryCampaignStorage() : fsCampaignStorage())\n\n  if (opts.onProgress) {\n    opts.onProgress({ kind: 'baseline.started', scenarios: opts.scenarios.length })\n  }\n\n  const result = await runImprovementLoop<TScenario, TArtifact>({\n    scenarios: train,\n    baselineSurface: opts.baselineSurface,\n    dispatchWithSurface: opts.agent,\n    proposer,\n    judges: [opts.judge],\n    populationSize,\n    maxGenerations: generations,\n    promoteTopK: budget.promoteTopK,\n    reps: budget.reps,\n    holdoutScenarios: holdout,\n    gate,\n    autoOnPromote: opts.autoOnPromote ?? 'none',\n    ghOwner: opts.ghOwner,\n    ghRepo: opts.ghRepo,\n    storage,\n    runDir,\n    maxConcurrency,\n    cellPlacement: opts.cellPlacement,\n    costCeiling,\n    expectUsage,\n    labeledStore: opts.labeledStore,\n    captureSource: opts.captureSource,\n    analyzeGeneration: opts.analyzeGeneration,\n    findings: opts.findings,\n  })\n\n  const baseline = meanComposite(result.baselineOnHoldout.aggregates.byScenario)\n  const winnerStats = meanComposite(result.winnerOnHoldout.aggregates.byScenario)\n\n  if (opts.onProgress) {\n    opts.onProgress({\n      kind: 'baseline.completed',\n      compositeMean: baseline.compositeMean,\n      durationMs: Date.now() - startedAt,\n    })\n    opts.onProgress({\n      kind: 'gate.decided',\n      decision: result.gateResult.decision,\n      lift: winnerStats.compositeMean - baseline.compositeMean,\n    })\n  }\n\n  const totalCost =\n    result.baselineCampaign.aggregates.totalCostUsd +\n    result.generations.reduce(\n      (sum, gen) =>\n        sum + gen.surfaces.reduce((s, sf) => s + sf.campaign.aggregates.totalCostUsd, 0),\n      0,\n    )\n\n  // Rigor packet: feed baseline + winner cells through analyzeRuns().\n  // The two candidates (`baseline` / `winner`) give the lift section a\n  // clean paired comparison; per-judge / per-dimension / cost-quality\n  // sections populate from the cells' judgeScores.\n  const insight = await analyzeRuns({\n    runs: [\n      ...cellsToRunRecords(result.baselineCampaign.cells, 'baseline', runDir, opts.baselineSurface),\n      ...cellsToRunRecords(result.winnerOnHoldout.cells, 'winner', runDir, result.winnerSurface),\n    ],\n    baselineCandidateId: 'baseline',\n    candidateCandidateId: 'winner',\n  })\n\n  // ── Durable provenance: candidate→cell→gate→promote chain + rationale +\n  // diff + backend provenance. Always emitted; the +lift recomputes from it.\n  const durationMs = Date.now() - startedAt\n  const workerRecords =\n    opts.collectWorkerRecords?.() ??\n    cellsToRunRecords(result.winnerOnHoldout.cells, 'winner', runDir, result.winnerSurface)\n  const { record: provenance } = await emitLoopProvenance<TArtifact, TScenario>({\n    runId: `${runDir}#${startedAt}`,\n    runDir,\n    timestamp: new Date(startedAt).toISOString(),\n    baselineSurface: opts.baselineSurface,\n    winnerSurface: result.winnerSurface,\n    winnerLabel: result.winnerLabel,\n    winnerRationale: result.winnerRationale,\n    diff: result.promotedDiff,\n    generations: result.generations.map((g) => ({\n      generationIndex: g.record.generationIndex,\n      candidates: g.record.candidates,\n      promoted: g.record.promoted,\n      surfaces: g.surfaces.map((s) => ({ surfaceHash: s.surfaceHash, surface: s.surface })),\n    })),\n    gate: result.gateResult,\n    baselineOnHoldout: result.baselineOnHoldout,\n    winnerOnHoldout: result.winnerOnHoldout,\n    workerRecords,\n    totalCostUsd: totalCost,\n    totalDurationMs: durationMs,\n    storage,\n    hostedClient: opts.hostedTenant ? createHostedClient(opts.hostedTenant) : undefined,\n  })\n  if (opts.onProvenance) opts.onProvenance(provenance)\n\n  const summary: SelfImproveResult<TScenario, TArtifact> = {\n    baseline,\n    winner: {\n      ...winnerStats,\n      surface: result.winnerSurface,\n      ...(result.winnerLabel ? { label: result.winnerLabel } : {}),\n      ...(result.winnerRationale ? { rationale: result.winnerRationale } : {}),\n    },\n    lift: winnerStats.compositeMean - baseline.compositeMean,\n    diff: result.promotedDiff,\n    provenance,\n    gateDecision: result.gateResult.decision,\n    generationsExplored: result.generations.length,\n    durationMs,\n    totalCostUsd: totalCost,\n    insight,\n    raw: result,\n  }\n\n  // Opt-in hosted ingest. Failures are logged but never fail the loop: the\n  // local result is always returned.\n  if (opts.hostedTenant) {\n    try {\n      await shipEvalRunToHosted(opts.hostedTenant, opts, summary, result, runDir)\n    } catch (err) {\n      const msg = err instanceof Error ? err.message : String(err)\n      // eslint-disable-next-line no-console -- intentional: hosted-ingest is best-effort\n      console.warn(`[agent-eval] hosted ingest failed (continuing): ${msg}`)\n    }\n  }\n\n  return summary\n}\n\nasync function shipEvalRunToHosted<TScenario extends Scenario, TArtifact>(\n  tenant: HostedTenant,\n  opts: SelfImproveOptions<TScenario, TArtifact>,\n  summary: SelfImproveResult<TScenario, TArtifact>,\n  raw: RunImprovementLoopResult<TArtifact, TScenario>,\n  runDir: string,\n): Promise<void> {\n  const client = createHostedClient(tenant)\n\n  function snapshotFromCampaign(\n    index: number,\n    surface: MutableSurface | undefined,\n    campaign: RunImprovementLoopResult<TArtifact, TScenario>['baselineCampaign'],\n    durationMs: number,\n  ): EvalRunGenerationSnapshot {\n    const cells: EvalRunCellScore[] = campaign.cells.map((cell) => {\n      const judgeScores = Object.values(cell.judgeScores)\n      const composite =\n        judgeScores.length === 0\n          ? 0\n          : judgeScores.reduce((s, j) => s + j.composite, 0) / judgeScores.length\n      return {\n        scenarioId: cell.scenarioId,\n        rep: cell.rep,\n        compositeMean: composite,\n        dimensions: Object.fromEntries(\n          Object.entries(cell.judgeScores).map(([name, score]) => [name, score.dimensions]),\n        ),\n        errorMessage: cell.error ?? undefined,\n      }\n    })\n    const compositeMean =\n      cells.length === 0 ? 0 : cells.reduce((s, c) => s + c.compositeMean, 0) / cells.length\n    return {\n      index,\n      surfaceHash:\n        typeof surface === 'string'\n          ? hashString(surface)\n          : hashString(JSON.stringify(surface ?? '')),\n      surface,\n      cells,\n      compositeMean,\n      costUsd: campaign.aggregates.totalCostUsd,\n      durationMs,\n    }\n  }\n\n  const generations: EvalRunGenerationSnapshot[] = []\n  // Baseline as generation 0.\n  generations.push(snapshotFromCampaign(0, opts.baselineSurface, raw.baselineCampaign, 0))\n  // Improvement generations as 1..N. Substrate stores per-surface campaigns\n  // per generation — we summarize the WINNING surface per generation here.\n  for (const gen of raw.generations) {\n    const winner = gen.surfaces.reduce(\n      (best, s) =>\n        s.campaign.aggregates.cellsExecuted > 0 &&\n        (best === undefined || averageComposite(s.campaign) > averageComposite(best.campaign))\n          ? s\n          : best,\n      gen.surfaces[0],\n    )\n    if (!winner) continue\n    generations.push(\n      snapshotFromCampaign(gen.record.generationIndex + 1, winner.surface, winner.campaign, 0),\n    )\n  }\n\n  const event: EvalRunEvent = {\n    runId: `${runDir}#${Date.now()}`,\n    runDir,\n    timestamp: new Date().toISOString(),\n    status: 'finished',\n    labels: opts.hostedLabels ?? {},\n    baseline: generations[0],\n    generations,\n    gateDecision: summary.gateDecision,\n    holdoutLift: summary.lift,\n    totalCostUsd: summary.totalCostUsd,\n    totalDurationMs: summary.durationMs,\n    insightReport: summary.insight,\n  }\n\n  await client.ingestEvalRun(event)\n}\n\nfunction averageComposite(\n  campaign: RunImprovementLoopResult<unknown, Scenario>['baselineCampaign'],\n): number {\n  const aggs = Object.values(campaign.aggregates.byScenario)\n  return aggs.length === 0 ? 0 : aggs.reduce((s, a) => s + a.meanComposite, 0) / aggs.length\n}\n\nfunction hashString(s: string): string {\n  let h = 2166136261 >>> 0\n  for (let i = 0; i < s.length; i++) {\n    h ^= s.charCodeAt(i)\n    h = Math.imul(h, 16777619) >>> 0\n  }\n  return h.toString(16).padStart(8, '0')\n}\n\n/**\n * Adapt campaign cells into the `RunRecord` shape `analyzeRuns()` consumes.\n * Each cell becomes one run; `candidateId` is the caller-supplied label so\n * baseline + winner pair cleanly on `(experimentId, seed)`.\n *\n * `promptHash` is the REAL sha256 content hash of the surface this cell ran\n * (baseline vs winner are byte-distinguishable + byte-identical-verifiable);\n * `configHash` is the sha256 of the candidate label so the two candidates'\n * config rows differ. Both were previously the literal `'sha256:cell'`, which\n * made baseline and winner indistinguishable in every downstream record.\n */\nfunction cellsToRunRecords<TArtifact>(\n  cells: ReadonlyArray<CampaignCellResult<TArtifact>>,\n  candidateId: 'baseline' | 'winner',\n  runId: string,\n  surface: MutableSurface,\n): RunRecord[] {\n  const promptHash = surfaceContentHash(surface)\n  const configHash = `sha256:${createHash('sha256').update(candidateId).digest('hex')}`\n  return cells.map((cell) => {\n    const perJudge: Record<string, Record<string, number>> = {}\n    const perDimMeanAccum: Record<string, { sum: number; n: number }> = {}\n    let compositeSum = 0\n    let compositeCount = 0\n    for (const [judgeId, score] of Object.entries(cell.judgeScores)) {\n      perJudge[judgeId] = { ...score.dimensions }\n      for (const [dim, value] of Object.entries(score.dimensions)) {\n        if (!Number.isFinite(value)) continue\n        const accum = perDimMeanAccum[dim] ?? { sum: 0, n: 0 }\n        accum.sum += value\n        accum.n += 1\n        perDimMeanAccum[dim] = accum\n      }\n      if (Number.isFinite(score.composite)) {\n        compositeSum += score.composite\n        compositeCount += 1\n      }\n    }\n    const perDimMean: Record<string, number> = {}\n    for (const [dim, { sum, n }] of Object.entries(perDimMeanAccum)) {\n      perDimMean[dim] = n === 0 ? 0 : sum / n\n    }\n    const composite = compositeCount === 0 ? 0 : compositeSum / compositeCount\n    const judgeScores: JudgeScoresRecord = {\n      perJudge,\n      perDimMean,\n      composite,\n    }\n    return {\n      runId: `${runId}::${candidateId}::${cell.cellId}`,\n      experimentId: runId,\n      candidateId,\n      // Pair on (scenarioId, rep) — analyzeRuns pairs on (experimentId, seed).\n      // Synthesize a stable seed for that pairing.\n      seed:\n        cell.rep * 1_000_000 +\n        hashString(cell.scenarioId)\n          .slice(0, 6)\n          .split('')\n          .reduce((a, c) => (a * 31 + c.charCodeAt(0)) >>> 0, 0),\n      model: 'campaign-cell',\n      promptHash,\n      configHash,\n      commitSha: 'cell',\n      wallMs: cell.durationMs,\n      costUsd: cell.costUsd,\n      tokenUsage: { input: 0, output: 0 },\n      outcome: {\n        holdoutScore: composite,\n        raw: {},\n        judgeScores,\n      },\n      splitTag: 'holdout',\n      ...(cell.error ? { failureMode: cell.error } : {}),\n    } satisfies RunRecord\n  })\n}\n","import type { RunEvalOptions } from '../campaign/presets/run-eval'\nimport { runEval } from '../campaign/presets/run-eval'\nimport { inMemoryCampaignStorage } from '../campaign/storage'\nimport type {\n  CampaignResult,\n  DispatchContext,\n  JudgeConfig,\n  MutableSurface,\n  Scenario,\n} from '../campaign/types'\nimport type { HostedTenant } from '../hosted/client'\nimport {\n  type SelfImproveBudget,\n  type SelfImproveLlm,\n  type SelfImproveOptions,\n  type SelfImproveResult,\n  selfImprove,\n} from './self-improve'\n\nexport type AgentEvalAgent<TScenario extends Scenario, TArtifact> = (\n  surface: MutableSurface,\n  scenario: TScenario,\n  ctx: DispatchContext,\n) => Promise<TArtifact>\n\nexport type DefineAgentEvalOptions<TScenario extends Scenario, TArtifact> = SelfImproveOptions<\n  TScenario,\n  TArtifact\n>\n\nexport interface AgentEvalEvaluateOptions<TScenario extends Scenario, TArtifact>\n  extends Omit<\n    RunEvalOptions<TScenario, TArtifact>,\n    'dispatch' | 'judges' | 'runDir' | 'scenarios'\n  > {\n  /** Scenario set to evaluate. Defaults to the scenarios passed to `defineAgentEval`. */\n  scenarios?: TScenario[]\n  /** Surface to evaluate. Defaults to the baseline surface passed to `defineAgentEval`. */\n  surface?: MutableSurface\n  /** Agent to evaluate. Defaults to the agent passed to `defineAgentEval`. */\n  agent?: AgentEvalAgent<TScenario, TArtifact>\n  /** Single judge override. Ignored when `judges` is set. */\n  judge?: JudgeConfig<TArtifact, TScenario>\n  /** Full judge list override. Defaults to the single judge passed to `defineAgentEval`. */\n  judges?: JudgeConfig<TArtifact, TScenario>[]\n  /** Logical or filesystem run directory. Defaults to an in-memory run. */\n  runDir?: string\n}\n\nexport type AgentEvalImproveOptions<TScenario extends Scenario, TArtifact> = Omit<\n  Partial<SelfImproveOptions<TScenario, TArtifact>>,\n  'budget' | 'hostedTenant' | 'llm'\n> & {\n  budget?: Partial<SelfImproveBudget>\n  hostedTenant?: Partial<HostedTenant>\n  llm?: Partial<SelfImproveLlm>\n}\n\nexport interface DefinedAgentEval<TScenario extends Scenario, TArtifact> {\n  /** The default scenarios used by `evaluate()` and `improve()`. */\n  readonly scenarios: readonly TScenario[]\n  /** The default baseline surface used by `evaluate()` and `improve()`. */\n  readonly baselineSurface: MutableSurface\n  /**\n   * Run one scored evaluation. Use this for a baseline score or to score one\n   * candidate surface without running an improvement loop.\n   */\n  evaluate(\n    opts?: AgentEvalEvaluateOptions<TScenario, TArtifact>,\n  ): Promise<CampaignResult<TArtifact, TScenario>>\n  /**\n   * Run the closed improvement loop. Per-call overrides replace the definition\n   * except for nested config objects (`budget`, `llm`, `hostedTenant`), which\n   * are merged field-by-field so callers can override one knob without\n   * repeating secrets or budget defaults.\n   */\n  improve(\n    opts?: AgentEvalImproveOptions<TScenario, TArtifact>,\n  ): Promise<SelfImproveResult<TScenario, TArtifact>>\n}\n\n/**\n * Define an agent eval once, then either score a surface with `evaluate()` or\n * run the closed loop with `improve()`.\n *\n * This is a DX wrapper only: it delegates to `runEval()` and `selfImprove()` and\n * returns their native result shapes.\n */\nexport function defineAgentEval<TScenario extends Scenario, TArtifact>(\n  defaults: DefineAgentEvalOptions<TScenario, TArtifact>,\n): DefinedAgentEval<TScenario, TArtifact> {\n  const defaultEvaluateOptions = evaluateDefaults(defaults)\n\n  return {\n    scenarios: defaults.scenarios,\n    baselineSurface: defaults.baselineSurface,\n\n    async evaluate(opts = {}) {\n      const { agent, judge, judges, runDir, scenarios, surface, ...campaignOpts } = opts\n      const selectedAgent = agent ?? defaults.agent\n      const selectedSurface = surface ?? defaults.baselineSurface\n      const selectedRunDir = runDir ?? defaults.runDir ?? `mem://defineAgentEval-${Date.now()}`\n      const selectedStorage =\n        campaignOpts.storage ??\n        defaultEvaluateOptions.storage ??\n        (selectedRunDir.startsWith('mem://') ? inMemoryCampaignStorage() : undefined)\n      const evalOptions: RunEvalOptions<TScenario, TArtifact> = {\n        ...defaultEvaluateOptions,\n        ...campaignOpts,\n        ...(selectedStorage ? { storage: selectedStorage } : {}),\n        runDir: selectedRunDir,\n        scenarios: scenarios ?? defaults.scenarios,\n        dispatch: (scenario, ctx) => selectedAgent(selectedSurface, scenario, ctx),\n        judges: evaluateJudges(judges, judge ?? defaults.judge),\n      }\n      if (evalOptions.reps !== undefined)\n        evalOptions.reps = requirePositiveInteger(evalOptions.reps, 'reps')\n      return runEval<TScenario, TArtifact>(evalOptions)\n    },\n\n    async improve(opts = {}) {\n      const {\n        budget: budgetOverride,\n        hostedTenant: hostedTenantOverride,\n        llm: llmOverride,\n        ...topLevelOverrides\n      } = opts\n      const merged = mergeDefined(defaults, topLevelOverrides)\n      const budget = mergeBudget(defaults.budget, budgetOverride)\n      const llm = mergeLlm(defaults.llm, llmOverride)\n      const hostedTenant = mergeHostedTenant(defaults.hostedTenant, hostedTenantOverride)\n      return selfImprove<TScenario, TArtifact>({\n        ...merged,\n        ...(budget ? { budget } : {}),\n        ...(llm ? { llm } : {}),\n        ...(hostedTenant ? { hostedTenant } : {}),\n      })\n    },\n  }\n}\n\ntype SharedEvaluateDefaults<TScenario extends Scenario, TArtifact> = Omit<\n  RunEvalOptions<TScenario, TArtifact>,\n  'dispatch' | 'judges' | 'runDir' | 'scenarios'\n>\n\nfunction evaluateDefaults<TScenario extends Scenario, TArtifact>(\n  defaults: DefineAgentEvalOptions<TScenario, TArtifact>,\n): SharedEvaluateDefaults<TScenario, TArtifact> {\n  const out: SharedEvaluateDefaults<TScenario, TArtifact> = {}\n  if (defaults.storage) out.storage = defaults.storage\n  if (defaults.labeledStore) out.labeledStore = defaults.labeledStore\n  if (defaults.captureSource) out.captureSource = defaults.captureSource\n  if (defaults.cellPlacement) out.cellPlacement = defaults.cellPlacement\n  if (defaults.expectUsage) out.expectUsage = defaults.expectUsage\n  if (defaults.budget?.dollars !== undefined) out.costCeiling = defaults.budget.dollars\n  if (defaults.budget?.maxConcurrency !== undefined)\n    out.maxConcurrency = defaults.budget.maxConcurrency\n  if (defaults.budget?.reps !== undefined)\n    out.reps = requirePositiveInteger(defaults.budget.reps, 'budget.reps')\n  return out\n}\n\nfunction mergeBudget(\n  defaults: SelfImproveBudget | undefined,\n  overrides: Partial<SelfImproveBudget> | undefined,\n): SelfImproveBudget | undefined {\n  const merged = mergeOptionalObject(defaults, overrides)\n  if (merged?.reps !== undefined) merged.reps = requirePositiveInteger(merged.reps, 'budget.reps')\n  return merged\n}\n\nfunction mergeLlm(\n  defaults: SelfImproveLlm | undefined,\n  overrides: Partial<SelfImproveLlm> | undefined,\n): SelfImproveLlm | undefined {\n  return mergeOptionalObject(defaults, overrides)\n}\n\nfunction mergeHostedTenant(\n  defaults: HostedTenant | undefined,\n  overrides: Partial<HostedTenant> | undefined,\n): HostedTenant | undefined {\n  const merged = mergeOptionalObject(defaults, overrides)\n  if (!merged) return undefined\n  if (!merged.endpoint?.trim() || !merged.apiKey?.trim() || !merged.tenantId?.trim()) {\n    throw new Error(\n      'defineAgentEval.improve: hostedTenant requires endpoint, apiKey, and tenantId after merging defaults and overrides',\n    )\n  }\n  return merged\n}\n\nfunction mergeDefined<T extends object>(defaults: T, overrides: Partial<T> | undefined): T {\n  if (!overrides) return defaults\n  const merged = { ...defaults } as Record<string, unknown>\n  for (const [key, value] of Object.entries(overrides)) {\n    if (value !== undefined) merged[key] = value\n  }\n  return merged as T\n}\n\nfunction mergeOptionalObject<T extends object>(\n  defaults: T | undefined,\n  overrides: Partial<T> | undefined,\n): T | undefined {\n  if (!defaults && !overrides) return undefined\n  return mergeDefined(defaults ?? ({} as T), overrides)\n}\n\nfunction evaluateJudges<TArtifact, TScenario extends Scenario>(\n  judges: JudgeConfig<TArtifact, TScenario>[] | undefined,\n  defaultJudge: JudgeConfig<TArtifact, TScenario>,\n): JudgeConfig<TArtifact, TScenario>[] {\n  if (judges !== undefined) {\n    if (judges.length === 0) {\n      throw new Error('defineAgentEval.evaluate: judges must not be empty')\n    }\n    return judges\n  }\n  return [defaultJudge]\n}\n\nfunction requirePositiveInteger(value: number, field: string): number {\n  if (!Number.isInteger(value) || value < 1) {\n    throw new Error(`defineAgentEval: ${field} must be a positive integer`)\n  }\n  return value\n}\n","/**\n * # `evalReportingSuite` — one call from runs (or a run dir) to `analysis.json`.\n *\n * A thin wrapper over the analysis primitive (`analyzeRuns`) and the on-disk\n * intake adapter (`fromRunRecordDir`). It does NOT reimplement any statistics,\n * distributions, or clustering — it resolves the input into validated\n * `RunRecord[]`, calls `analyzeRuns` with the options you'd pass it directly,\n * wraps the result in a small provenance envelope, and (optionally) writes a\n * single `analysis.json` artifact.\n *\n * ```ts\n * // From a directory of run files, write ./runs/analysis.json:\n * const suite = await evalReportingSuite('./runs', { write: true })\n * // From records already in memory, no write:\n * const suite = await evalReportingSuite(records, { analyze: { decisionThreshold: 0.03 } })\n * suite.report // the InsightReport — distributions, paired lift, findings rollup\n * ```\n */\n\nimport { mkdir, writeFile } from 'node:fs/promises'\nimport { dirname, join } from 'node:path'\nimport type { RunRecord } from '../run-record'\nimport { type AnalyzeRunsOptions, analyzeRuns } from './analyze-runs'\nimport type { InsightReport } from './insight-report'\nimport {\n  type FromRunRecordDirOptions,\n  type FromRunRecordDirResult,\n  fromRunRecordDir,\n} from './intake/run-record-dir'\n\n/** Either records in hand or a path to a `.json` / `.jsonl` file or a\n *  directory of them. */\nexport type EvalReportingSuiteInput = RunRecord[] | string\n\nexport interface EvalReportingSuiteOptions {\n  /** Forwarded verbatim to `analyzeRuns` (everything except `runs`, which the\n   *  suite supplies from the resolved input). Use this for split selection,\n   *  baseline/candidate ids, canaries, prior-period runs, the analyst registry,\n   *  etc. */\n  analyze?: Omit<AnalyzeRunsOptions, 'runs'>\n  /** Loader options used only when the input is a path. */\n  load?: FromRunRecordDirOptions\n  /**\n   * Write the suite result as a single `analysis.json`.\n   *   - `true` — write to `<dir>/analysis.json` when the input is a directory,\n   *     or alongside the input file; throws if the input is in-memory records\n   *     (no directory to anchor to — pass an explicit path instead).\n   *   - a string — write to exactly this path (a directory path gets\n   *     `analysis.json` appended; any other path is used verbatim).\n   *   - omitted / false — do not write.\n   */\n  write?: boolean | string\n}\n\n/** The suite artifact — the `analyzeRuns` report plus provenance. This is the\n *  exact shape serialized to `analysis.json`. */\nexport interface EvalReportingSuiteResult {\n  /** The analysis itself — distributions, paired stats/lift, failure rollup,\n   *  recommendations. Produced by `analyzeRuns`. */\n  report: InsightReport\n  /** How the suite was run, so a reader can verify provenance. */\n  provenance: {\n    /** ISO timestamp the suite ran. */\n    generatedAt: string\n    /** Number of records analyzed (mirrors `report.n`). */\n    runCount: number\n    /** The source path when the input was a directory/file; null for\n     *  in-memory records. */\n    sourcePath: string | null\n    /** Files read when loading from disk; empty for in-memory input. */\n    files: string[]\n    /** Records dropped at the validation boundary. Always empty unless\n     *  `load.onInvalid` was set to `'collect'`. */\n    rejected: FromRunRecordDirResult['rejected']\n  }\n  /** The path `analysis.json` was written to, or null when `write` was unset. */\n  writtenTo: string | null\n}\n\nconst ANALYSIS_ARTIFACT = 'analysis.json'\n\n/**\n * Resolve runs (or a run dir/file), run `analyzeRuns`, and optionally persist a\n * single `analysis.json`. The only analysis logic lives in `analyzeRuns`; this\n * function is composition + I/O.\n */\nexport async function evalReportingSuite(\n  input: EvalReportingSuiteInput,\n  options: EvalReportingSuiteOptions = {},\n): Promise<EvalReportingSuiteResult> {\n  const fromPath = typeof input === 'string'\n\n  let runs: RunRecord[]\n  let files: string[] = []\n  let rejected: FromRunRecordDirResult['rejected'] = []\n  if (fromPath) {\n    const loaded = await fromRunRecordDir(input, options.load)\n    runs = loaded.runs\n    files = loaded.files\n    rejected = loaded.rejected\n  } else {\n    runs = input\n  }\n\n  if (runs.length === 0) {\n    throw new Error(\n      fromPath\n        ? `evalReportingSuite: no RunRecords found at '${input}'`\n        : 'evalReportingSuite: no RunRecords to analyze',\n    )\n  }\n\n  const report = await analyzeRuns({ ...options.analyze, runs })\n\n  const result: EvalReportingSuiteResult = {\n    report,\n    provenance: {\n      generatedAt: new Date().toISOString(),\n      runCount: runs.length,\n      sourcePath: fromPath ? input : null,\n      files,\n      rejected,\n    },\n    writtenTo: null,\n  }\n\n  const target = resolveWriteTarget(options.write, fromPath ? input : null)\n  if (target) {\n    await mkdir(dirname(target), { recursive: true })\n    await writeFile(target, `${JSON.stringify(result, null, 2)}\\n`, 'utf8')\n    result.writtenTo = target\n  }\n\n  return result\n}\n\n/** Resolve where (if anywhere) to write `analysis.json`. Returns null when\n *  writing is disabled. Throws on `write: true` with in-memory input — there is\n *  no directory to anchor the artifact to, and silently inventing `cwd` would\n *  scatter files. */\nfunction resolveWriteTarget(\n  write: EvalReportingSuiteOptions['write'],\n  sourcePath: string | null,\n): string | null {\n  if (!write) return null\n\n  if (typeof write === 'string') {\n    const looksLikeDir =\n      write.endsWith('/') || (!write.endsWith('.json') && !write.endsWith('.jsonl'))\n    return looksLikeDir ? join(write, ANALYSIS_ARTIFACT) : write\n  }\n\n  // write === true\n  if (sourcePath === null) {\n    throw new Error(\n      'evalReportingSuite: write:true needs a source path to anchor analysis.json — pass an explicit output path when analyzing in-memory records',\n    )\n  }\n  const isFile = sourcePath.endsWith('.json') || sourcePath.endsWith('.jsonl')\n  return isFile ? join(dirname(sourcePath), ANALYSIS_ARTIFACT) : join(sourcePath, ANALYSIS_ARTIFACT)\n}\n","/**\n * # `intake/run-record-dir` — load a directory or file of `RunRecord`s.\n *\n * The on-disk counterpart to the in-memory intake adapters: point it at a\n * single `.json` (array) / `.jsonl` (one record per line) file or at a\n * directory of such files, and it returns the substrate-canonical\n * `RunRecord[]` ready for `analyzeRuns({ runs })`.\n *\n * Validation is at the boundary: each parsed object goes through\n * `parseRunRecordSafe`. By default an invalid record fails loud with its\n * file + index; pass `onInvalid: 'collect'` to keep the valid records and\n * receive the rejects as structured diagnostics instead.\n */\n\nimport { readdir, readFile, stat } from 'node:fs/promises'\nimport { join } from 'node:path'\nimport { parseRunRecordSafe, type RunRecord } from '../../run-record'\n\n/** A record that failed boundary validation, with enough context to fix it. */\nexport interface RunRecordRejection {\n  /** Absolute or caller-relative path to the file the record came from. */\n  file: string\n  /** Zero-based position within the file (array index or JSONL line number). */\n  index: number\n  /** The validator's message. */\n  reason: string\n}\n\nexport interface FromRunRecordDirOptions {\n  /**\n   * How to treat a record that fails `parseRunRecordSafe`:\n   *   - `'throw'` (default) — fail loud on the first invalid record.\n   *   - `'collect'` — drop it, keep the rest, and return it under `rejected`.\n   */\n  onInvalid?: 'throw' | 'collect'\n  /**\n   * When the input is a directory, only files matching this predicate are\n   * read. Default: any file ending in `.json` or `.jsonl`. The `analysis.json`\n   * artifact `evalReportingSuite` writes is always skipped so a re-run never\n   * ingests its own output.\n   */\n  include?: (fileName: string) => boolean\n  /**\n   * Recurse into subdirectories when the input is a directory. Default false —\n   * a flat run directory is the common case and recursion can silently pull in\n   * unrelated corpora.\n   */\n  recursive?: boolean\n}\n\nexport interface FromRunRecordDirResult {\n  /** Records that passed boundary validation, in file-then-index order. */\n  runs: RunRecord[]\n  /** Records that failed validation. Empty unless `onInvalid: 'collect'`. */\n  rejected: RunRecordRejection[]\n  /** The files that were read, in the order they were processed. */\n  files: string[]\n}\n\nconst ANALYSIS_ARTIFACT = 'analysis.json'\n\nfunction defaultInclude(fileName: string): boolean {\n  if (fileName === ANALYSIS_ARTIFACT) return false\n  return fileName.endsWith('.json') || fileName.endsWith('.jsonl')\n}\n\n/**\n * Resolve a file or directory path into validated `RunRecord[]`.\n *\n * A `.json` file must parse to a top-level array; a `.jsonl` file is one\n * record per non-empty line. Directories are read shallowly by default\n * (set `recursive` to descend); the `analysis.json` output artifact is\n * always excluded.\n */\nexport async function fromRunRecordDir(\n  path: string,\n  options: FromRunRecordDirOptions = {},\n): Promise<FromRunRecordDirResult> {\n  const onInvalid = options.onInvalid ?? 'throw'\n  const include = options.include ?? defaultInclude\n\n  const stats = await stat(path)\n  const filePaths = stats.isDirectory()\n    ? await collectFiles(path, include, options.recursive ?? false)\n    : [path]\n\n  const runs: RunRecord[] = []\n  const rejected: RunRecordRejection[] = []\n\n  for (const file of filePaths) {\n    const raw = await parseRecordFile(file)\n    for (const { index, value } of raw) {\n      const parsed = parseRunRecordSafe(value)\n      if (parsed.ok) {\n        runs.push(parsed.value)\n        continue\n      }\n      const rejection: RunRecordRejection = { file, index, reason: parsed.error.message }\n      if (onInvalid === 'throw') {\n        throw new Error(\n          `fromRunRecordDir: invalid RunRecord in '${file}' at index ${index}: ${parsed.error.message}`,\n        )\n      }\n      rejected.push(rejection)\n    }\n  }\n\n  return { runs, rejected, files: filePaths }\n}\n\n/** Read a single `.json` / `.jsonl` file into `{ index, value }` pairs. A\n *  malformed JSONL line throws with its line number rather than being skipped —\n *  silent line-dropping is how corpora quietly shrink. */\nasync function parseRecordFile(file: string): Promise<Array<{ index: number; value: unknown }>> {\n  const text = await readFile(file, 'utf8')\n  const trimmed = text.trim()\n  if (trimmed.length === 0) return []\n\n  if (trimmed.startsWith('[')) {\n    const parsed = JSON.parse(trimmed) as unknown\n    if (!Array.isArray(parsed)) {\n      throw new Error(`fromRunRecordDir: file '${file}' did not parse to an array`)\n    }\n    return parsed.map((value, index) => ({ index, value }))\n  }\n\n  const out: Array<{ index: number; value: unknown }> = []\n  const lines = trimmed.split('\\n')\n  for (let i = 0; i < lines.length; i++) {\n    const line = lines[i]!.trim()\n    if (line.length === 0) continue\n    try {\n      out.push({ index: i, value: JSON.parse(line) as unknown })\n    } catch (err) {\n      throw new Error(\n        `fromRunRecordDir: file '${file}' line ${i + 1} is not valid JSON: ${\n          err instanceof Error ? err.message : String(err)\n        }`,\n      )\n    }\n  }\n  return out\n}\n\n/** Sorted file list under a directory, filtered by `include`. Sorted so the\n *  resulting `RunRecord` order — and any downstream fingerprint — is stable\n *  across filesystems. */\nasync function collectFiles(\n  dir: string,\n  include: (fileName: string) => boolean,\n  recursive: boolean,\n): Promise<string[]> {\n  const entries = await readdir(dir, { withFileTypes: true })\n  const files: string[] = []\n  const subdirs: string[] = []\n  for (const entry of entries) {\n    if (entry.isDirectory()) {\n      if (recursive) subdirs.push(join(dir, entry.name))\n      continue\n    }\n    if (include(entry.name)) files.push(join(dir, entry.name))\n  }\n  files.sort()\n  subdirs.sort()\n  for (const sub of subdirs) {\n    files.push(...(await collectFiles(sub, include, recursive)))\n  }\n  return files\n}\n","/**\n * # `@tangle-network/agent-eval/contract` — eval-run diff primitive.\n *\n * The substrate side of the v-N-versus-v-N+1 dashboard view. Given two\n * `EvalRunEvent`s (or two `EvalRunGenerationSnapshot`s from one run), this\n * returns a normalised diff: per-cell composite + per-judge/per-dimension\n * deltas, surface-hash change, aggregate cost + duration shifts.\n *\n * Consumed by:\n *   - The hosted-tier dashboard (intelligence-web) — renders v3 vs v4\n *     comparisons of cells × judges × dimensions.\n *   - CI reporting — emits a \"shipped: composite +0.07, cost +$1.20\" line\n *     in PR review for autonomous-improvement runs.\n *   - Any downstream consumer that needs \"what actually changed\" without\n *     reimplementing the matching + arithmetic.\n *\n * Cells are matched on the natural composite key `(scenarioId, rep)`.\n * Unmatched cells surface as `removed` / `added` so callers can tell\n * \"this cell got worse\" from \"this cell wasn't run.\"\n */\n\nimport type { GateDecision } from '../campaign/types'\nimport type { EvalRunCellScore, EvalRunEvent, EvalRunGenerationSnapshot } from '../hosted/types'\n\n// ── Types ────────────────────────────────────────────────────────────\n\n/** Per-dimension delta. `before` / `after` are null when the judge did not\n *  emit a value for that side. `delta` is `after - before`; null when\n *  either side is null. */\nexport interface EvalDimensionDelta {\n  before: number | null\n  after: number | null\n  delta: number | null\n}\n\n/** Per-cell delta, keyed on `(scenarioId, rep)`. */\nexport interface EvalCellScoreDelta {\n  scenarioId: string\n  rep: number\n  compositeBefore: number\n  compositeAfter: number\n  compositeDelta: number\n  /** Per-judge → per-dimension deltas. Outer key = judge name from\n   *  `EvalRunCellScore.dimensions`; inner key = dimension name. */\n  dimensions: Record<string, Record<string, EvalDimensionDelta>>\n}\n\n/** Diff between two generation snapshots — the unit the dashboard renders\n *  for a single \"v3 vs v4\" comparison. */\nexport interface EvalGenerationDiff {\n  beforeIndex: number\n  afterIndex: number\n  beforeSurfaceHash: string\n  afterSurfaceHash: string\n  surfaceChanged: boolean\n  /** Cells present in both snapshots, matched on `(scenarioId, rep)`. */\n  matched: EvalCellScoreDelta[]\n  /** Cells present in `before` but missing from `after`. */\n  removed: EvalRunCellScore[]\n  /** Cells present in `after` but missing from `before`. */\n  added: EvalRunCellScore[]\n  /** Aggregate composite mean across all cells in the snapshot. */\n  compositeBefore: number\n  compositeAfter: number\n  compositeDelta: number\n  costUsdBefore: number\n  costUsdAfter: number\n  costUsdDelta: number\n  durationMsBefore: number\n  durationMsAfter: number\n  durationMsDelta: number\n}\n\n/** Diff between two full eval-runs. Includes both baseline-vs-baseline and\n *  winner-vs-winner generation diffs when both sides expose them, plus\n *  run-level metadata. */\nexport interface EvalRunDiff {\n  beforeRunId: string\n  afterRunId: string\n  beforeTimestamp: string\n  afterTimestamp: string\n  beforeGateDecision: GateDecision | null\n  afterGateDecision: GateDecision | null\n  beforeHoldoutLift: number | null\n  afterHoldoutLift: number | null\n  holdoutLiftDelta: number | null\n  beforeTotalCostUsd: number\n  afterTotalCostUsd: number\n  totalCostUsdDelta: number\n  beforeTotalDurationMs: number\n  afterTotalDurationMs: number\n  totalDurationMsDelta: number\n  /** Baseline-vs-baseline diff. Null when either run has no baseline. */\n  baselineDiff: EvalGenerationDiff | null\n  /** Highest-index-generation comparison. Null when either run has no\n   *  recorded generations (e.g. baseline-only or errored before any\n   *  generation completed). */\n  winnersDiff: EvalGenerationDiff | null\n}\n\n// ── Implementation ───────────────────────────────────────────────────\n\nfunction keyForCell(cell: EvalRunCellScore): string {\n  // JSON-tuple key — `scenarioId` may legitimately contain `::` or any other\n  // delimiter, so we use JSON.stringify on a 2-tuple to get an unambiguous,\n  // collision-free composite key.\n  return JSON.stringify([cell.scenarioId, cell.rep])\n}\n\n/** Build the per-dimension delta map for a matched cell. Each judge name +\n *  dimension name encountered on EITHER side appears in the result. */\nfunction diffDimensions(\n  before: EvalRunCellScore['dimensions'],\n  after: EvalRunCellScore['dimensions'],\n): EvalCellScoreDelta['dimensions'] {\n  const out: EvalCellScoreDelta['dimensions'] = {}\n  const judges = new Set<string>([...Object.keys(before), ...Object.keys(after)])\n  for (const judge of judges) {\n    const beforeDims = before[judge] ?? {}\n    const afterDims = after[judge] ?? {}\n    const dims = new Set<string>([...Object.keys(beforeDims), ...Object.keys(afterDims)])\n    const judgeOut: Record<string, EvalDimensionDelta> = {}\n    for (const dim of dims) {\n      // Coerce non-finite values (NaN, ±Infinity) to null so the diff never\n      // surfaces NaN/Infinity to the dashboard. A NaN score is a substrate\n      // bug from upstream; the diff treats it as \"no value\" rather than\n      // propagating the corruption.\n      const rawBefore = beforeDims[dim]\n      const rawAfter = afterDims[dim]\n      const b = typeof rawBefore === 'number' && Number.isFinite(rawBefore) ? rawBefore : null\n      const a = typeof rawAfter === 'number' && Number.isFinite(rawAfter) ? rawAfter : null\n      judgeOut[dim] = {\n        before: b,\n        after: a,\n        delta: b !== null && a !== null ? a - b : null,\n      }\n    }\n    out[judge] = judgeOut\n  }\n  return out\n}\n\n/**\n * Diff two generation snapshots. Cells are matched on `(scenarioId, rep)`;\n * unmatched cells surface in `added` / `removed`. Aggregate fields are\n * recomputed from the snapshot's stored fields, not re-derived from cells —\n * this keeps the diff consistent with whatever aggregation the substrate\n * actually reported.\n */\nexport function diffGenerations(\n  before: EvalRunGenerationSnapshot,\n  after: EvalRunGenerationSnapshot,\n): EvalGenerationDiff {\n  const beforeMap = new Map(before.cells.map((c) => [keyForCell(c), c]))\n  const afterMap = new Map(after.cells.map((c) => [keyForCell(c), c]))\n\n  const matched: EvalCellScoreDelta[] = []\n  const removed: EvalRunCellScore[] = []\n  const added: EvalRunCellScore[] = []\n\n  for (const [key, beforeCell] of beforeMap) {\n    const afterCell = afterMap.get(key)\n    if (!afterCell) {\n      removed.push(beforeCell)\n      continue\n    }\n    matched.push({\n      scenarioId: beforeCell.scenarioId,\n      rep: beforeCell.rep,\n      compositeBefore: beforeCell.compositeMean,\n      compositeAfter: afterCell.compositeMean,\n      compositeDelta: afterCell.compositeMean - beforeCell.compositeMean,\n      dimensions: diffDimensions(beforeCell.dimensions, afterCell.dimensions),\n    })\n  }\n  for (const [key, afterCell] of afterMap) {\n    if (!beforeMap.has(key)) added.push(afterCell)\n  }\n\n  return {\n    beforeIndex: before.index,\n    afterIndex: after.index,\n    beforeSurfaceHash: before.surfaceHash,\n    afterSurfaceHash: after.surfaceHash,\n    surfaceChanged: before.surfaceHash !== after.surfaceHash,\n    matched,\n    removed,\n    added,\n    compositeBefore: before.compositeMean,\n    compositeAfter: after.compositeMean,\n    compositeDelta: after.compositeMean - before.compositeMean,\n    costUsdBefore: before.costUsd,\n    costUsdAfter: after.costUsd,\n    costUsdDelta: after.costUsd - before.costUsd,\n    durationMsBefore: before.durationMs,\n    durationMsAfter: after.durationMs,\n    durationMsDelta: after.durationMs - before.durationMs,\n  }\n}\n\n/** Highest-index generation, or null if the run recorded none. */\nfunction winnerOf(run: EvalRunEvent): EvalRunGenerationSnapshot | null {\n  if (run.generations.length === 0) return null\n  let winner = run.generations[0] as EvalRunGenerationSnapshot\n  for (const gen of run.generations) {\n    if (gen.index > winner.index) winner = gen\n  }\n  return winner\n}\n\n/**\n * Diff two full eval-runs. Produces baseline-vs-baseline and\n * winner-vs-winner generation diffs when both sides expose them, plus\n * run-level cost / lift / gate-decision deltas.\n */\nexport function diffRuns(before: EvalRunEvent, after: EvalRunEvent): EvalRunDiff {\n  const beforeWinner = winnerOf(before)\n  const afterWinner = winnerOf(after)\n  const baselineDiff =\n    before.baseline && after.baseline ? diffGenerations(before.baseline, after.baseline) : null\n  const winnersDiff =\n    beforeWinner && afterWinner ? diffGenerations(beforeWinner, afterWinner) : null\n\n  const beforeLift = before.holdoutLift ?? null\n  const afterLift = after.holdoutLift ?? null\n\n  return {\n    beforeRunId: before.runId,\n    afterRunId: after.runId,\n    beforeTimestamp: before.timestamp,\n    afterTimestamp: after.timestamp,\n    beforeGateDecision: before.gateDecision ?? null,\n    afterGateDecision: after.gateDecision ?? null,\n    beforeHoldoutLift: beforeLift,\n    afterHoldoutLift: afterLift,\n    holdoutLiftDelta: beforeLift !== null && afterLift !== null ? afterLift - beforeLift : null,\n    beforeTotalCostUsd: before.totalCostUsd,\n    afterTotalCostUsd: after.totalCostUsd,\n    totalCostUsdDelta: after.totalCostUsd - before.totalCostUsd,\n    beforeTotalDurationMs: before.totalDurationMs,\n    afterTotalDurationMs: after.totalDurationMs,\n    totalDurationMsDelta: after.totalDurationMs - before.totalDurationMs,\n    baselineDiff,\n    winnersDiff,\n  }\n}\n\n/**\n * Within-run baseline → winning-generation diff. The natural \"what did the\n * improvement loop produce\" view for a single run. Returns null when the\n * run never reached a generation past baseline (errored early, or the gate\n * shipped the baseline as-is).\n */\nexport function diffRunBaselineToWinner(run: EvalRunEvent): EvalGenerationDiff | null {\n  if (!run.baseline) return null\n  const winner = winnerOf(run)\n  if (!winner || winner.index === run.baseline.index) return null\n  return diffGenerations(run.baseline, winner)\n}\n","/**\n * `fromAgentTrace` — provenance correlation from Cursor's Agent Trace spec\n * (https://github.com/cursor/agent-trace, RFC v0.1.0).\n *\n * Agent Trace is NOT a run/quality trace — it carries no outcome, score, or\n * cost. It records *code authorship*: for a VCS revision, which AI model /\n * conversation authored which file ranges. It explicitly disclaims quality\n * assessment — which is exactly what `analyzeRuns` adds.\n *\n * The two layers join on a key the substrate already has: a `RunRecord`\n * carries `commitSha`, and an Agent Trace record is keyed by\n * `vcs.revision`. So this adapter does not produce `RunRecord`s — it builds a\n * provenance index by commit and partitions existing runs by their authoring\n * model. Feed each cohort to `analyzeRuns` (or pass one as `baselineRuns`) to\n * answer the question no run-only trace can: *which authoring agent's code\n * fails / regresses / costs more.*\n *\n * Granularity is commit-level (the SHA join). Per-file/per-line correlation\n * would require runs to record which files they exercised — out of scope.\n */\n\nimport type { RunRecord } from '../../run-record'\n\n// ── Agent Trace record schema (the subset we read) ──────────────────────────\n\nexport type AgentTraceContributorType = 'human' | 'ai' | 'mixed' | 'unknown'\n\nexport interface AgentTraceContributor {\n  type: AgentTraceContributorType\n  /** models.dev id, e.g. `anthropic/claude-opus-4-5-20251101`. */\n  model_id?: string\n}\n\nexport interface AgentTraceRange {\n  start_line: number\n  end_line: number\n  content_hash?: string\n  /** Per-range contributor override (agent handoffs). Wins over the\n   *  conversation-level contributor for these lines. */\n  contributor?: AgentTraceContributor\n}\n\nexport interface AgentTraceConversation {\n  url?: string\n  contributor?: AgentTraceContributor\n  ranges: AgentTraceRange[]\n}\n\nexport interface AgentTraceFile {\n  path: string\n  conversations: AgentTraceConversation[]\n}\n\nexport interface AgentTraceRecord {\n  version: string\n  id: string\n  timestamp: string\n  vcs?: { type: string; revision: string }\n  tool?: { name?: string; version?: string }\n  files: AgentTraceFile[]\n}\n\n// ── Provenance index ─────────────────────────────────────────────────────────\n\n/** Authorship provenance for one VCS revision, aggregated across the record's\n *  files/conversations/ranges. */\nexport interface AuthoringProvenance {\n  commitSha: string\n  /** Unique AI model ids that authored code in this commit (type ai|mixed). */\n  aiModels: string[]\n  /** Tools that produced the records (e.g. `cursor`). */\n  tools: string[]\n  conversationCount: number\n  fileCount: number\n  /** Total attributed lines (sum of range spans). */\n  lineCount: number\n  /** True if any range was authored (in whole or part) by a human. */\n  humanInvolved: boolean\n}\n\nexport type AgentTraceIndex = Map<string, AuthoringProvenance>\n\nfunction rangeLines(r: AgentTraceRange): number {\n  return Math.max(0, r.end_line - r.start_line + 1)\n}\n\n/**\n * Build a commit → provenance index from Agent Trace records. Multiple records\n * for the same revision are merged. Records without `vcs.revision` are skipped\n * (the SHA is the join key — without it there is nothing to correlate against).\n */\nexport function parseAgentTrace(records: AgentTraceRecord[]): AgentTraceIndex {\n  interface Acc {\n    models: Set<string>\n    tools: Set<string>\n    files: Set<string>\n    conversationCount: number\n    lineCount: number\n    humanInvolved: boolean\n  }\n  const acc = new Map<string, Acc>()\n\n  for (const record of records) {\n    const sha = record.vcs?.revision\n    if (!sha) continue // the SHA is the join key — nothing to correlate without it\n\n    let a = acc.get(sha)\n    if (!a) {\n      a = {\n        models: new Set(),\n        tools: new Set(),\n        files: new Set(),\n        conversationCount: 0,\n        lineCount: 0,\n        humanInvolved: false,\n      }\n      acc.set(sha, a)\n    }\n\n    if (record.tool?.name) a.tools.add(record.tool.name)\n\n    for (const file of record.files ?? []) {\n      a.files.add(file.path)\n      for (const conv of file.conversations ?? []) {\n        a.conversationCount += 1\n        for (const range of conv.ranges ?? []) {\n          // Per-range contributor wins, else the conversation contributor.\n          const contributor = range.contributor ?? conv.contributor\n          a.lineCount += rangeLines(range)\n          if (!contributor) continue\n          if (contributor.type === 'human' || contributor.type === 'mixed') {\n            a.humanInvolved = true\n          }\n          if ((contributor.type === 'ai' || contributor.type === 'mixed') && contributor.model_id) {\n            a.models.add(contributor.model_id)\n          }\n        }\n      }\n    }\n  }\n\n  const index: AgentTraceIndex = new Map()\n  for (const [sha, a] of acc) {\n    index.set(sha, {\n      commitSha: sha,\n      aiModels: [...a.models].sort(),\n      tools: [...a.tools].sort(),\n      conversationCount: a.conversationCount,\n      fileCount: a.files.size,\n      lineCount: a.lineCount,\n      humanInvolved: a.humanInvolved,\n    })\n  }\n  return index\n}\n\n// ── Run ↔ provenance join ──────────────────────────────────────────────────\n\nexport interface PartitionByAuthoringModelResult {\n  /** Runs grouped by each AI model that authored code in the run's commit. A\n   *  run whose commit had multiple authoring models appears under EACH — the\n   *  cohorts overlap by construction at commit granularity. */\n  byModel: Map<string, RunRecord[]>\n  /** Runs whose `commitSha` had no Agent Trace provenance (no record, or no\n   *  AI authorship). Kept separate — never silently folded into a cohort. */\n  unattributed: RunRecord[]\n}\n\n/**\n * Partition runs by the AI model(s) that authored the code at each run's\n * `commitSha`. Feed `byModel.get(modelId)` to `analyzeRuns`, or compare two\n * model cohorts via `analyzeRuns({ runs: a, baselineRuns: b })` for a lift CI\n * on \"model A's code vs model B's code\".\n */\nexport function partitionRunsByAuthoringModel(\n  runs: RunRecord[],\n  index: AgentTraceIndex,\n): PartitionByAuthoringModelResult {\n  const byModel = new Map<string, RunRecord[]>()\n  const unattributed: RunRecord[] = []\n\n  for (const run of runs) {\n    const provenance = index.get(run.commitSha)\n    if (!provenance || provenance.aiModels.length === 0) {\n      unattributed.push(run)\n      continue\n    }\n    for (const model of provenance.aiModels) {\n      const cohort = byModel.get(model) ?? []\n      cohort.push(run)\n      byModel.set(model, cohort)\n    }\n  }\n\n  return { byModel, unattributed }\n}\n","/**\n * # `intake/feedback-table` — multi-rater approve/reject corpus → `RunRecord[]`.\n *\n * The generic shape behind Obsidian's `#approved` / `#rejected` tags, a\n * Google Sheet, a Postgres `feedback` table, or any CSV with ratings.\n *\n * Caller supplies one row per (run, rater) tuple plus per-run metadata; the\n * adapter rolls them up into the substrate-canonical `RunRecord` shape so\n * `analyzeRuns({ runs, raterScores })` can produce inter-rater agreement,\n * disagreement triage, and downstream recommendations.\n *\n * Per-run `RunRecord.outcome.searchScore` is the rater-mean rating\n * (normalised to 0..1 when scale is supplied); `outcome.raw` carries the\n * per-rater scores keyed by rater id for downstream attribution.\n */\n\nimport type { JudgeScoresRecord, RunOutcome, RunRecord, RunSplitTag } from '../../run-record'\n\nexport interface FeedbackTableRow {\n  /** Stable id for this run — the unit a rater scored. Drives pairing\n   *  across analysis primitives. */\n  runId: string\n  /** Identifier of the rater that produced this rating. */\n  rater: string\n  /** The rating itself. Accepts boolean (approve/reject), 0..1 scalar,\n   *  or any numeric scale — see `scale`. */\n  rating: number | boolean\n  /** Optional metadata carried through to `RunRecord.outcome.raw` and the\n   *  custom-shape metadata bag. */\n  metadata?: Record<string, unknown>\n}\n\nexport interface FeedbackTableMeta {\n  runId: string\n  /** When omitted, defaults to `'feedback-corpus'`. Used to group related\n   *  runs in `analyzeRuns()` lift analysis. */\n  experimentId?: string\n  /** When omitted, defaults to `runId` — each run is its own candidate. */\n  candidateId?: string\n  /** Cost in USD, when available. Set to 0 when unknown — the consumer's\n   *  cost analysis sections will collapse gracefully. */\n  costUsd?: number\n  /** Wall-clock ms, when available. Defaults to 0. */\n  wallMs?: number\n  /** Model identifier including snapshot. Default `unknown@unknown`. */\n  model?: string\n  /** Optional sha256 of the prompt; default `'sha256:unknown'`. */\n  promptHash?: string\n  /** Default `'sha256:unknown'`. */\n  configHash?: string\n  /** Default `'unknown'`. */\n  commitSha?: string\n  /** Default `'holdout'` — feedback corpora are by nature the holdout\n   *  signal a closed-loop improvement aims at. */\n  splitTag?: RunSplitTag\n  /** Free-form metadata available to consumers via the cast-out path on\n   *  the resulting RunRecord. */\n  extras?: Record<string, unknown>\n}\n\nexport interface FromFeedbackTableOptions {\n  /** Per-(run, rater) ratings. */\n  ratings: FeedbackTableRow[]\n  /** Per-run metadata. When a runId appears in `ratings` but not here, the\n   *  adapter synthesises minimal metadata with defaults documented above. */\n  meta?: FeedbackTableMeta[]\n  /** Rating scale. Provide `{ min, max }` for non-0..1 numeric scales.\n   *  Booleans are normalised: true → 1, false → 0. Default: assumes\n   *  ratings are already 0..1. */\n  scale?: { min: number; max: number }\n  /** When true, the rater scores are emitted into `raterScores` (a sibling\n   *  array `analyzeRuns()` accepts) instead of being averaged into the\n   *  run's `outcome.searchScore`. Default `true` — preserves rater-level\n   *  signal for inter-rater analysis. */\n  emitRaterScores?: boolean\n}\n\nexport interface FromFeedbackTableResult {\n  runs: RunRecord[]\n  /** Rater-level scores ready to pass into `analyzeRuns({ raterScores })`\n   *  for inter-rater agreement + disagreement triage. */\n  raterScores: Array<{ runId: string; rater: string; score: number }>\n}\n\nexport function fromFeedbackTable(opts: FromFeedbackTableOptions): FromFeedbackTableResult {\n  const { ratings, meta = [], scale, emitRaterScores = true } = opts\n  const metaByRun = new Map(meta.map((m) => [m.runId, m]))\n\n  // Normalise per-rating to a 0..1 score.\n  const normalise = (rating: number | boolean): number => {\n    if (typeof rating === 'boolean') return rating ? 1 : 0\n    if (!Number.isFinite(rating)) return Number.NaN\n    if (!scale) return rating\n    const { min, max } = scale\n    if (max === min) return rating\n    return (rating - min) / (max - min)\n  }\n\n  // Group ratings by runId.\n  const byRun = new Map<string, FeedbackTableRow[]>()\n  for (const row of ratings) {\n    const list = byRun.get(row.runId) ?? []\n    list.push(row)\n    byRun.set(row.runId, list)\n  }\n\n  const runs: RunRecord[] = []\n  const raterScores: FromFeedbackTableResult['raterScores'] = []\n\n  for (const [runId, rowsForRun] of byRun) {\n    const normalised = rowsForRun\n      .map((r) => ({ rater: r.rater, score: normalise(r.rating) }))\n      .filter((r) => Number.isFinite(r.score))\n    if (normalised.length === 0) continue\n\n    const meanScore = normalised.reduce((s, r) => s + r.score, 0) / normalised.length\n\n    const runMeta = metaByRun.get(runId) ?? ({ runId } as FeedbackTableMeta)\n\n    const judgeScores: JudgeScoresRecord = {\n      perJudge: Object.fromEntries(normalised.map((r) => [r.rater, { rating: r.score }])),\n      perDimMean: { rating: meanScore },\n      composite: meanScore,\n    }\n\n    const outcome: RunOutcome = {\n      // Feedback corpora ARE the holdout signal — score lands on\n      // `holdoutScore` so downstream substrate primitives (`paretoChart`,\n      // promotion gates) read it correctly by default.\n      holdoutScore: meanScore,\n      raw: Object.fromEntries(normalised.map((r) => [`rater:${r.rater}`, r.score])),\n      judgeScores,\n    }\n\n    runs.push({\n      runId,\n      experimentId: runMeta.experimentId ?? 'feedback-corpus',\n      candidateId: runMeta.candidateId ?? runId,\n      seed: 0,\n      model: runMeta.model ?? 'unknown@unknown',\n      promptHash: runMeta.promptHash ?? 'sha256:unknown',\n      configHash: runMeta.configHash ?? 'sha256:unknown',\n      commitSha: runMeta.commitSha ?? 'unknown',\n      wallMs: runMeta.wallMs ?? 0,\n      costUsd: runMeta.costUsd ?? 0,\n      tokenUsage: { input: 0, output: 0 },\n      outcome,\n      splitTag: runMeta.splitTag ?? 'holdout',\n    } as RunRecord)\n\n    if (emitRaterScores) {\n      for (const r of normalised) raterScores.push({ runId, rater: r.rater, score: r.score })\n    }\n  }\n\n  return { runs, raterScores }\n}\n","/**\n * # `intake/otel-spans` — OTel `TraceSpanEvent[]` → `RunRecord[]`.\n *\n * Turns an existing observability stream into the substrate-canonical\n * `RunRecord` shape so consumers with logs but no eval discipline can\n * call `analyzeRuns()` against their production traffic immediately.\n *\n * Pivot rule: spans are grouped by `tangle.runId` (the same attribute the\n * hosted-tier wire format uses) or, when absent, by `traceId`. One group\n * becomes one `RunRecord`. The root span (no `parentSpanId`) supplies:\n *\n *   - `runId` (the group key)\n *   - `wallMs` from `endTimeUnixNano - startTimeUnixNano`\n *   - `model` from `gen_ai.request.model` / `llm.model` / `tangle.model`\n *   - cost from `cost.usd` / `gen_ai.usage.cost_usd` / `tangle.cost.usd`\n *   - token usage from `gen_ai.usage.{input,output}_tokens`\n *   - `outcome.searchScore` from `tangle.score` / `eval.score` when\n *     present; `outcome.raw` collects every numeric attribute.\n *\n * Spans that ERRORed (`status.code === 'ERROR'`) populate `failureMode`\n * with their `name` so `analyzeRuns()`'s failure clustering sees them.\n */\n\nimport type { TraceSpanEvent } from '../../hosted/types'\nimport type {\n  JudgeScoresRecord,\n  RunOutcome,\n  RunRecord,\n  RunSplitTag,\n  RunTokenUsage,\n} from '../../run-record'\nimport {\n  LLM_INPUT_TOKEN_ATTR_KEYS,\n  LLM_MODEL_ATTR_KEYS,\n  LLM_OUTPUT_TOKEN_ATTR_KEYS,\n} from '../../trace/otlp-attributes'\n\nconst SCORE_KEYS = ['tangle.score', 'eval.score', 'score']\nconst MODEL_KEYS = ['tangle.model', ...LLM_MODEL_ATTR_KEYS, 'model']\nconst COST_KEYS = ['tangle.cost.usd', 'gen_ai.usage.cost_usd', 'cost.usd', 'cost']\nconst INPUT_TOKEN_KEYS = [...LLM_INPUT_TOKEN_ATTR_KEYS, 'tangle.tokens.in', 'tokens.in']\nconst OUTPUT_TOKEN_KEYS = [...LLM_OUTPUT_TOKEN_ATTR_KEYS, 'tangle.tokens.out', 'tokens.out']\nconst PROMPT_HASH_KEYS = ['tangle.prompt_hash', 'prompt.hash']\nconst CONFIG_HASH_KEYS = ['tangle.config_hash', 'config.hash']\n\nexport interface FromOtelSpansOptions {\n  spans: TraceSpanEvent[]\n  /** Default split tag for synthesized records. Defaults to `'holdout'`. */\n  defaultSplit?: RunSplitTag\n  /** Default `experimentId` when not present on any span. */\n  experimentId?: string\n}\n\nexport function fromOtelSpans(opts: FromOtelSpansOptions): RunRecord[] {\n  const { spans, defaultSplit = 'holdout', experimentId = 'otel-corpus' } = opts\n  const grouped = groupSpans(spans)\n\n  const runs: RunRecord[] = []\n  for (const [groupKey, groupSpans] of grouped) {\n    const root = findRoot(groupSpans)\n    if (!root) continue\n\n    const wallMs = Math.max(0, (root.endTimeUnixNano - root.startTimeUnixNano) / 1_000_000)\n    const model = readAttrString(groupSpans, MODEL_KEYS) ?? 'unknown@unknown'\n    const costUsd = readAttrNumber(groupSpans, COST_KEYS) ?? 0\n    const inputTokens = readAttrNumber(groupSpans, INPUT_TOKEN_KEYS) ?? 0\n    const outputTokens = readAttrNumber(groupSpans, OUTPUT_TOKEN_KEYS) ?? 0\n    const promptHash = readAttrString(groupSpans, PROMPT_HASH_KEYS) ?? 'sha256:unknown'\n    const configHash = readAttrString(groupSpans, CONFIG_HASH_KEYS) ?? 'sha256:unknown'\n    const score = readAttrNumber(groupSpans, SCORE_KEYS)\n\n    const rawNumeric = collectNumericAttrs(groupSpans)\n    const tokenUsage: RunTokenUsage = {\n      input: inputTokens,\n      output: outputTokens,\n    }\n\n    const judgeScores: JudgeScoresRecord | undefined =\n      score !== undefined\n        ? {\n            perJudge: { 'otel-derived': { score } },\n            perDimMean: { score },\n            composite: score,\n          }\n        : undefined\n\n    const errorSpan = groupSpans.find((s) => s.status?.code === 'ERROR')\n    const outcome: RunOutcome = {\n      ...(opts.defaultSplit === 'search' ? { searchScore: score } : { holdoutScore: score }),\n      raw: rawNumeric,\n      ...(judgeScores ? { judgeScores } : {}),\n    }\n\n    runs.push({\n      runId: groupKey,\n      experimentId,\n      candidateId: (root.attributes['tangle.candidateId'] as string | undefined) ?? 'otel-default',\n      seed: 0,\n      model,\n      promptHash,\n      configHash,\n      commitSha: (root.attributes['tangle.commit_sha'] as string | undefined) ?? 'unknown',\n      wallMs,\n      costUsd,\n      tokenUsage,\n      outcome,\n      splitTag: defaultSplit,\n      ...(errorSpan ? { failureMode: errorSpan.name } : {}),\n    } as RunRecord)\n  }\n  return runs\n}\n\n// ── Internal helpers ────────────────────────────────────────────────\n\nfunction groupSpans(spans: TraceSpanEvent[]): Map<string, TraceSpanEvent[]> {\n  const m = new Map<string, TraceSpanEvent[]>()\n  for (const span of spans) {\n    const key = (span['tangle.runId'] as string | undefined) ?? span.traceId\n    const list = m.get(key) ?? []\n    list.push(span)\n    m.set(key, list)\n  }\n  return m\n}\n\nfunction findRoot(group: TraceSpanEvent[]): TraceSpanEvent | undefined {\n  return group.find((s) => !s.parentSpanId) ?? group[0]\n}\n\nfunction readAttrString(spans: TraceSpanEvent[], keys: string[]): string | undefined {\n  for (const span of spans) {\n    for (const key of keys) {\n      const v = span.attributes[key]\n      if (typeof v === 'string' && v.length > 0) return v\n    }\n  }\n  return undefined\n}\n\nfunction readAttrNumber(spans: TraceSpanEvent[], keys: string[]): number | undefined {\n  for (const span of spans) {\n    for (const key of keys) {\n      const v = span.attributes[key]\n      if (typeof v === 'number' && Number.isFinite(v)) return v\n      if (typeof v === 'string') {\n        const parsed = Number(v)\n        if (Number.isFinite(parsed)) return parsed\n      }\n    }\n  }\n  return undefined\n}\n\nfunction collectNumericAttrs(spans: TraceSpanEvent[]): Record<string, number> {\n  const raw: Record<string, number> = {}\n  for (const span of spans) {\n    for (const [k, v] of Object.entries(span.attributes)) {\n      if (typeof v === 'number' && Number.isFinite(v)) raw[k] = v\n    }\n  }\n  return raw\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAwBA,SAAS,kBAAkB;AAoR3B,SAAS,kBACP,WACA,UAC8C;AAE9C,WAAS,KAAK,GAAmB;AAC/B,QAAI,IAAI,eAAe;AACvB,aAAS,IAAI,GAAG,IAAI,EAAE,QAAQ,KAAK;AACjC,WAAK,EAAE,WAAW,CAAC;AACnB,UAAI,KAAK,KAAK,GAAG,QAAQ,MAAM;AAAA,IACjC;AACA,WAAO;AAAA,EACT;AACA,QAAM,SAAS,CAAC,GAAG,SAAS,EAAE,KAAK,CAAC,GAAG,MAAM,KAAK,EAAE,EAAE,IAAI,KAAK,EAAE,EAAE,CAAC;AACpE,QAAM,WAAW,KAAK,IAAI,GAAG,KAAK,IAAI,OAAO,SAAS,GAAG,KAAK,MAAM,OAAO,SAAS,QAAQ,CAAC,CAAC;AAC9F,SAAO;AAAA,IACL,SAAS,OAAO,MAAM,GAAG,QAAQ;AAAA,IACjC,OAAO,OAAO,MAAM,QAAQ;AAAA,EAC9B;AACF;AAEA,SAAS,cAAc,YAGrB;AACA,QAAM,cAAsC,CAAC;AAC7C,QAAM,SAAmB,CAAC;AAC1B,aAAW,CAAC,IAAI,GAAG,KAAK,OAAO,QAAQ,UAAU,GAAG;AAClD,gBAAY,EAAE,IAAI,IAAI;AACtB,WAAO,KAAK,IAAI,aAAa;AAAA,EAC/B;AACA,SAAO;AAAA,IACL,eAAe,OAAO,WAAW,IAAI,IAAI,OAAO,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,OAAO;AAAA,IACpF;AAAA,EACF;AACF;AAEA,IAAM,8BAA8B;AAAA,EAClC;AAAA,EACA;AAAA,EACA;AAAA,EACA;AACF;AA2BA,eAAsB,YACpB,MACkD;AAClD,QAAM,YAAY,KAAK,IAAI;AAE3B,QAAM,SAAS,KAAK,UAAU,CAAC;AAC/B,QAAM,cAAc,OAAO,eAAe;AAC1C,QAAM,iBAAiB,OAAO,kBAAkB;AAChD,QAAM,iBAAiB,OAAO,kBAAkB;AAChD,QAAM,kBAAkB,OAAO,mBAAmB;AAClD,QAAM,cAAc,OAAO;AAC3B,QAAM,cAAc,KAAK,eAAe;AAExC,QAAM,kBAAkB,OAAO;AAC/B,QAAM,EAAE,OAAO,QAAQ,IAAI,kBACvB;AAAA,IACE,OAAO,KAAK,UAAU,OAAO,CAAC,MAAM,CAAC,gBAAgB,KAAK,CAAC,MAAM,EAAE,OAAO,EAAE,EAAE,CAAC;AAAA,IAC/E,SAAS;AAAA,EACX,IACA,kBAAkB,KAAK,WAAW,eAAe;AAErD,MAAI,MAAM,WAAW,GAAG;AACtB,UAAM,IAAI;AAAA,MACR;AAAA,IACF;AAAA,EACF;AACA,MAAI,QAAQ,WAAW,GAAG;AACxB,UAAM,IAAI,MAAM,2DAA2D;AAAA,EAC7E;AAEA,QAAM,WACJ,KAAK,YACL,aAAa;AAAA,IACX,KAAK;AAAA,MACH,SAAS,KAAK,KAAK,WAAW;AAAA,MAC9B,QAAQ,KAAK,KAAK,UAAU,QAAQ,IAAI,kBAAkB;AAAA,IAC5D;AAAA,IACA,OAAO,KAAK,KAAK,SAAS;AAAA,IAC1B,QACE,KAAK,kBACL;AAAA,IACF,oBAAoB,KAAK,sBAAsB;AAAA,EACjD,CAAC;AAEH,QAAM,OACJ,KAAK,QACL,sBAA4C;AAAA,IAC1C,kBAAkB;AAAA,IAClB,gBAAgB;AAAA,EAClB,CAAC;AAMH,QAAM,SAAS,KAAK,UAAU,qBAAqB,SAAS;AAC5D,QAAM,cAAc,OAAO,WAAW,QAAQ;AAC9C,QAAM,UAAU,KAAK,YAAY,cAAc,wBAAwB,IAAI,kBAAkB;AAE7F,MAAI,KAAK,YAAY;AACnB,SAAK,WAAW,EAAE,MAAM,oBAAoB,WAAW,KAAK,UAAU,OAAO,CAAC;AAAA,EAChF;AAEA,QAAM,SAAS,MAAM,mBAAyC;AAAA,IAC5D,WAAW;AAAA,IACX,iBAAiB,KAAK;AAAA,IACtB,qBAAqB,KAAK;AAAA,IAC1B;AAAA,IACA,QAAQ,CAAC,KAAK,KAAK;AAAA,IACnB;AAAA,IACA,gBAAgB;AAAA,IAChB,aAAa,OAAO;AAAA,IACpB,MAAM,OAAO;AAAA,IACb,kBAAkB;AAAA,IAClB;AAAA,IACA,eAAe,KAAK,iBAAiB;AAAA,IACrC,SAAS,KAAK;AAAA,IACd,QAAQ,KAAK;AAAA,IACb;AAAA,IACA;AAAA,IACA;AAAA,IACA,eAAe,KAAK;AAAA,IACpB;AAAA,IACA;AAAA,IACA,cAAc,KAAK;AAAA,IACnB,eAAe,KAAK;AAAA,IACpB,mBAAmB,KAAK;AAAA,IACxB,UAAU,KAAK;AAAA,EACjB,CAAC;AAED,QAAM,WAAW,cAAc,OAAO,kBAAkB,WAAW,UAAU;AAC7E,QAAM,cAAc,cAAc,OAAO,gBAAgB,WAAW,UAAU;AAE9E,MAAI,KAAK,YAAY;AACnB,SAAK,WAAW;AAAA,MACd,MAAM;AAAA,MACN,eAAe,SAAS;AAAA,MACxB,YAAY,KAAK,IAAI,IAAI;AAAA,IAC3B,CAAC;AACD,SAAK,WAAW;AAAA,MACd,MAAM;AAAA,MACN,UAAU,OAAO,WAAW;AAAA,MAC5B,MAAM,YAAY,gBAAgB,SAAS;AAAA,IAC7C,CAAC;AAAA,EACH;AAEA,QAAM,YACJ,OAAO,iBAAiB,WAAW,eACnC,OAAO,YAAY;AAAA,IACjB,CAAC,KAAK,QACJ,MAAM,IAAI,SAAS,OAAO,CAAC,GAAG,OAAO,IAAI,GAAG,SAAS,WAAW,cAAc,CAAC;AAAA,IACjF;AAAA,EACF;AAMF,QAAM,UAAU,MAAM,YAAY;AAAA,IAChC,MAAM;AAAA,MACJ,GAAG,kBAAkB,OAAO,iBAAiB,OAAO,YAAY,QAAQ,KAAK,eAAe;AAAA,MAC5F,GAAG,kBAAkB,OAAO,gBAAgB,OAAO,UAAU,QAAQ,OAAO,aAAa;AAAA,IAC3F;AAAA,IACA,qBAAqB;AAAA,IACrB,sBAAsB;AAAA,EACxB,CAAC;AAID,QAAM,aAAa,KAAK,IAAI,IAAI;AAChC,QAAM,gBACJ,KAAK,uBAAuB,KAC5B,kBAAkB,OAAO,gBAAgB,OAAO,UAAU,QAAQ,OAAO,aAAa;AACxF,QAAM,EAAE,QAAQ,WAAW,IAAI,MAAM,mBAAyC;AAAA,IAC5E,OAAO,GAAG,MAAM,IAAI,SAAS;AAAA,IAC7B;AAAA,IACA,WAAW,IAAI,KAAK,SAAS,EAAE,YAAY;AAAA,IAC3C,iBAAiB,KAAK;AAAA,IACtB,eAAe,OAAO;AAAA,IACtB,aAAa,OAAO;AAAA,IACpB,iBAAiB,OAAO;AAAA,IACxB,MAAM,OAAO;AAAA,IACb,aAAa,OAAO,YAAY,IAAI,CAAC,OAAO;AAAA,MAC1C,iBAAiB,EAAE,OAAO;AAAA,MAC1B,YAAY,EAAE,OAAO;AAAA,MACrB,UAAU,EAAE,OAAO;AAAA,MACnB,UAAU,EAAE,SAAS,IAAI,CAAC,OAAO,EAAE,aAAa,EAAE,aAAa,SAAS,EAAE,QAAQ,EAAE;AAAA,IACtF,EAAE;AAAA,IACF,MAAM,OAAO;AAAA,IACb,mBAAmB,OAAO;AAAA,IAC1B,iBAAiB,OAAO;AAAA,IACxB;AAAA,IACA,cAAc;AAAA,IACd,iBAAiB;AAAA,IACjB;AAAA,IACA,cAAc,KAAK,eAAe,mBAAmB,KAAK,YAAY,IAAI;AAAA,EAC5E,CAAC;AACD,MAAI,KAAK,aAAc,MAAK,aAAa,UAAU;AAEnD,QAAM,UAAmD;AAAA,IACvD;AAAA,IACA,QAAQ;AAAA,MACN,GAAG;AAAA,MACH,SAAS,OAAO;AAAA,MAChB,GAAI,OAAO,cAAc,EAAE,OAAO,OAAO,YAAY,IAAI,CAAC;AAAA,MAC1D,GAAI,OAAO,kBAAkB,EAAE,WAAW,OAAO,gBAAgB,IAAI,CAAC;AAAA,IACxE;AAAA,IACA,MAAM,YAAY,gBAAgB,SAAS;AAAA,IAC3C,MAAM,OAAO;AAAA,IACb;AAAA,IACA,cAAc,OAAO,WAAW;AAAA,IAChC,qBAAqB,OAAO,YAAY;AAAA,IACxC;AAAA,IACA,cAAc;AAAA,IACd;AAAA,IACA,KAAK;AAAA,EACP;AAIA,MAAI,KAAK,cAAc;AACrB,QAAI;AACF,YAAM,oBAAoB,KAAK,cAAc,MAAM,SAAS,QAAQ,MAAM;AAAA,IAC5E,SAAS,KAAK;AACZ,YAAM,MAAM,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;AAE3D,cAAQ,KAAK,mDAAmD,GAAG,EAAE;AAAA,IACvE;AAAA,EACF;AAEA,SAAO;AACT;AAEA,eAAe,oBACb,QACA,MACA,SACA,KACA,QACe;AACf,QAAM,SAAS,mBAAmB,MAAM;AAExC,WAAS,qBACP,OACA,SACA,UACA,YAC2B;AAC3B,UAAM,QAA4B,SAAS,MAAM,IAAI,CAAC,SAAS;AAC7D,YAAM,cAAc,OAAO,OAAO,KAAK,WAAW;AAClD,YAAM,YACJ,YAAY,WAAW,IACnB,IACA,YAAY,OAAO,CAAC,GAAG,MAAM,IAAI,EAAE,WAAW,CAAC,IAAI,YAAY;AACrE,aAAO;AAAA,QACL,YAAY,KAAK;AAAA,QACjB,KAAK,KAAK;AAAA,QACV,eAAe;AAAA,QACf,YAAY,OAAO;AAAA,UACjB,OAAO,QAAQ,KAAK,WAAW,EAAE,IAAI,CAAC,CAAC,MAAM,KAAK,MAAM,CAAC,MAAM,MAAM,UAAU,CAAC;AAAA,QAClF;AAAA,QACA,cAAc,KAAK,SAAS;AAAA,MAC9B;AAAA,IACF,CAAC;AACD,UAAM,gBACJ,MAAM,WAAW,IAAI,IAAI,MAAM,OAAO,CAAC,GAAG,MAAM,IAAI,EAAE,eAAe,CAAC,IAAI,MAAM;AAClF,WAAO;AAAA,MACL;AAAA,MACA,aACE,OAAO,YAAY,WACf,WAAW,OAAO,IAClB,WAAW,KAAK,UAAU,WAAW,EAAE,CAAC;AAAA,MAC9C;AAAA,MACA;AAAA,MACA;AAAA,MACA,SAAS,SAAS,WAAW;AAAA,MAC7B;AAAA,IACF;AAAA,EACF;AAEA,QAAM,cAA2C,CAAC;AAElD,cAAY,KAAK,qBAAqB,GAAG,KAAK,iBAAiB,IAAI,kBAAkB,CAAC,CAAC;AAGvF,aAAW,OAAO,IAAI,aAAa;AACjC,UAAM,SAAS,IAAI,SAAS;AAAA,MAC1B,CAAC,MAAM,MACL,EAAE,SAAS,WAAW,gBAAgB,MACrC,SAAS,UAAa,iBAAiB,EAAE,QAAQ,IAAI,iBAAiB,KAAK,QAAQ,KAChF,IACA;AAAA,MACN,IAAI,SAAS,CAAC;AAAA,IAChB;AACA,QAAI,CAAC,OAAQ;AACb,gBAAY;AAAA,MACV,qBAAqB,IAAI,OAAO,kBAAkB,GAAG,OAAO,SAAS,OAAO,UAAU,CAAC;AAAA,IACzF;AAAA,EACF;AAEA,QAAM,QAAsB;AAAA,IAC1B,OAAO,GAAG,MAAM,IAAI,KAAK,IAAI,CAAC;AAAA,IAC9B;AAAA,IACA,YAAW,oBAAI,KAAK,GAAE,YAAY;AAAA,IAClC,QAAQ;AAAA,IACR,QAAQ,KAAK,gBAAgB,CAAC;AAAA,IAC9B,UAAU,YAAY,CAAC;AAAA,IACvB;AAAA,IACA,cAAc,QAAQ;AAAA,IACtB,aAAa,QAAQ;AAAA,IACrB,cAAc,QAAQ;AAAA,IACtB,iBAAiB,QAAQ;AAAA,IACzB,eAAe,QAAQ;AAAA,EACzB;AAEA,QAAM,OAAO,cAAc,KAAK;AAClC;AAEA,SAAS,iBACP,UACQ;AACR,QAAM,OAAO,OAAO,OAAO,SAAS,WAAW,UAAU;AACzD,SAAO,KAAK,WAAW,IAAI,IAAI,KAAK,OAAO,CAAC,GAAG,MAAM,IAAI,EAAE,eAAe,CAAC,IAAI,KAAK;AACtF;AAEA,SAAS,WAAW,GAAmB;AACrC,MAAI,IAAI,eAAe;AACvB,WAAS,IAAI,GAAG,IAAI,EAAE,QAAQ,KAAK;AACjC,SAAK,EAAE,WAAW,CAAC;AACnB,QAAI,KAAK,KAAK,GAAG,QAAQ,MAAM;AAAA,EACjC;AACA,SAAO,EAAE,SAAS,EAAE,EAAE,SAAS,GAAG,GAAG;AACvC;AAaA,SAAS,kBACP,OACA,aACA,OACA,SACa;AACb,QAAM,aAAa,mBAAmB,OAAO;AAC7C,QAAM,aAAa,UAAU,WAAW,QAAQ,EAAE,OAAO,WAAW,EAAE,OAAO,KAAK,CAAC;AACnF,SAAO,MAAM,IAAI,CAAC,SAAS;AACzB,UAAM,WAAmD,CAAC;AAC1D,UAAM,kBAA8D,CAAC;AACrE,QAAI,eAAe;AACnB,QAAI,iBAAiB;AACrB,eAAW,CAAC,SAAS,KAAK,KAAK,OAAO,QAAQ,KAAK,WAAW,GAAG;AAC/D,eAAS,OAAO,IAAI,EAAE,GAAG,MAAM,WAAW;AAC1C,iBAAW,CAAC,KAAK,KAAK,KAAK,OAAO,QAAQ,MAAM,UAAU,GAAG;AAC3D,YAAI,CAAC,OAAO,SAAS,KAAK,EAAG;AAC7B,cAAM,QAAQ,gBAAgB,GAAG,KAAK,EAAE,KAAK,GAAG,GAAG,EAAE;AACrD,cAAM,OAAO;AACb,cAAM,KAAK;AACX,wBAAgB,GAAG,IAAI;AAAA,MACzB;AACA,UAAI,OAAO,SAAS,MAAM,SAAS,GAAG;AACpC,wBAAgB,MAAM;AACtB,0BAAkB;AAAA,MACpB;AAAA,IACF;AACA,UAAM,aAAqC,CAAC;AAC5C,eAAW,CAAC,KAAK,EAAE,KAAK,EAAE,CAAC,KAAK,OAAO,QAAQ,eAAe,GAAG;AAC/D,iBAAW,GAAG,IAAI,MAAM,IAAI,IAAI,MAAM;AAAA,IACxC;AACA,UAAM,YAAY,mBAAmB,IAAI,IAAI,eAAe;AAC5D,UAAM,cAAiC;AAAA,MACrC;AAAA,MACA;AAAA,MACA;AAAA,IACF;AACA,WAAO;AAAA,MACL,OAAO,GAAG,KAAK,KAAK,WAAW,KAAK,KAAK,MAAM;AAAA,MAC/C,cAAc;AAAA,MACd;AAAA;AAAA;AAAA,MAGA,MACE,KAAK,MAAM,MACX,WAAW,KAAK,UAAU,EACvB,MAAM,GAAG,CAAC,EACV,MAAM,EAAE,EACR,OAAO,CAAC,GAAG,MAAO,IAAI,KAAK,EAAE,WAAW,CAAC,MAAO,GAAG,CAAC;AAAA,MACzD,OAAO;AAAA,MACP;AAAA,MACA;AAAA,MACA,WAAW;AAAA,MACX,QAAQ,KAAK;AAAA,MACb,SAAS,KAAK;AAAA,MACd,YAAY,EAAE,OAAO,GAAG,QAAQ,EAAE;AAAA,MAClC,SAAS;AAAA,QACP,cAAc;AAAA,QACd,KAAK,CAAC;AAAA,QACN;AAAA,MACF;AAAA,MACA,UAAU;AAAA,MACV,GAAI,KAAK,QAAQ,EAAE,aAAa,KAAK,MAAM,IAAI,CAAC;AAAA,IAClD;AAAA,EACF,CAAC;AACH;;;AC3oBO,SAAS,gBACd,UACwC;AACxC,QAAM,yBAAyB,iBAAiB,QAAQ;AAExD,SAAO;AAAA,IACL,WAAW,SAAS;AAAA,IACpB,iBAAiB,SAAS;AAAA,IAE1B,MAAM,SAAS,OAAO,CAAC,GAAG;AACxB,YAAM,EAAE,OAAO,OAAO,QAAQ,QAAQ,WAAW,SAAS,GAAG,aAAa,IAAI;AAC9E,YAAM,gBAAgB,SAAS,SAAS;AACxC,YAAM,kBAAkB,WAAW,SAAS;AAC5C,YAAM,iBAAiB,UAAU,SAAS,UAAU,yBAAyB,KAAK,IAAI,CAAC;AACvF,YAAM,kBACJ,aAAa,WACb,uBAAuB,YACtB,eAAe,WAAW,QAAQ,IAAI,wBAAwB,IAAI;AACrE,YAAM,cAAoD;AAAA,QACxD,GAAG;AAAA,QACH,GAAG;AAAA,QACH,GAAI,kBAAkB,EAAE,SAAS,gBAAgB,IAAI,CAAC;AAAA,QACtD,QAAQ;AAAA,QACR,WAAW,aAAa,SAAS;AAAA,QACjC,UAAU,CAAC,UAAU,QAAQ,cAAc,iBAAiB,UAAU,GAAG;AAAA,QACzE,QAAQ,eAAe,QAAQ,SAAS,SAAS,KAAK;AAAA,MACxD;AACA,UAAI,YAAY,SAAS;AACvB,oBAAY,OAAO,uBAAuB,YAAY,MAAM,MAAM;AACpE,aAAO,QAA8B,WAAW;AAAA,IAClD;AAAA,IAEA,MAAM,QAAQ,OAAO,CAAC,GAAG;AACvB,YAAM;AAAA,QACJ,QAAQ;AAAA,QACR,cAAc;AAAA,QACd,KAAK;AAAA,QACL,GAAG;AAAA,MACL,IAAI;AACJ,YAAM,SAAS,aAAa,UAAU,iBAAiB;AACvD,YAAM,SAAS,YAAY,SAAS,QAAQ,cAAc;AAC1D,YAAM,MAAM,SAAS,SAAS,KAAK,WAAW;AAC9C,YAAM,eAAe,kBAAkB,SAAS,cAAc,oBAAoB;AAClF,aAAO,YAAkC;AAAA,QACvC,GAAG;AAAA,QACH,GAAI,SAAS,EAAE,OAAO,IAAI,CAAC;AAAA,QAC3B,GAAI,MAAM,EAAE,IAAI,IAAI,CAAC;AAAA,QACrB,GAAI,eAAe,EAAE,aAAa,IAAI,CAAC;AAAA,MACzC,CAAC;AAAA,IACH;AAAA,EACF;AACF;AAOA,SAAS,iBACP,UAC8C;AAC9C,QAAM,MAAoD,CAAC;AAC3D,MAAI,SAAS,QAAS,KAAI,UAAU,SAAS;AAC7C,MAAI,SAAS,aAAc,KAAI,eAAe,SAAS;AACvD,MAAI,SAAS,cAAe,KAAI,gBAAgB,SAAS;AACzD,MAAI,SAAS,cAAe,KAAI,gBAAgB,SAAS;AACzD,MAAI,SAAS,YAAa,KAAI,cAAc,SAAS;AACrD,MAAI,SAAS,QAAQ,YAAY,OAAW,KAAI,cAAc,SAAS,OAAO;AAC9E,MAAI,SAAS,QAAQ,mBAAmB;AACtC,QAAI,iBAAiB,SAAS,OAAO;AACvC,MAAI,SAAS,QAAQ,SAAS;AAC5B,QAAI,OAAO,uBAAuB,SAAS,OAAO,MAAM,aAAa;AACvE,SAAO;AACT;AAEA,SAAS,YACP,UACA,WAC+B;AAC/B,QAAM,SAAS,oBAAoB,UAAU,SAAS;AACtD,MAAI,QAAQ,SAAS,OAAW,QAAO,OAAO,uBAAuB,OAAO,MAAM,aAAa;AAC/F,SAAO;AACT;AAEA,SAAS,SACP,UACA,WAC4B;AAC5B,SAAO,oBAAoB,UAAU,SAAS;AAChD;AAEA,SAAS,kBACP,UACA,WAC0B;AAC1B,QAAM,SAAS,oBAAoB,UAAU,SAAS;AACtD,MAAI,CAAC,OAAQ,QAAO;AACpB,MAAI,CAAC,OAAO,UAAU,KAAK,KAAK,CAAC,OAAO,QAAQ,KAAK,KAAK,CAAC,OAAO,UAAU,KAAK,GAAG;AAClF,UAAM,IAAI;AAAA,MACR;AAAA,IACF;AAAA,EACF;AACA,SAAO;AACT;AAEA,SAAS,aAA+B,UAAa,WAAsC;AACzF,MAAI,CAAC,UAAW,QAAO;AACvB,QAAM,SAAS,EAAE,GAAG,SAAS;AAC7B,aAAW,CAAC,KAAK,KAAK,KAAK,OAAO,QAAQ,SAAS,GAAG;AACpD,QAAI,UAAU,OAAW,QAAO,GAAG,IAAI;AAAA,EACzC;AACA,SAAO;AACT;AAEA,SAAS,oBACP,UACA,WACe;AACf,MAAI,CAAC,YAAY,CAAC,UAAW,QAAO;AACpC,SAAO,aAAa,YAAa,CAAC,GAAS,SAAS;AACtD;AAEA,SAAS,eACP,QACA,cACqC;AACrC,MAAI,WAAW,QAAW;AACxB,QAAI,OAAO,WAAW,GAAG;AACvB,YAAM,IAAI,MAAM,oDAAoD;AAAA,IACtE;AACA,WAAO;AAAA,EACT;AACA,SAAO,CAAC,YAAY;AACtB;AAEA,SAAS,uBAAuB,OAAe,OAAuB;AACpE,MAAI,CAAC,OAAO,UAAU,KAAK,KAAK,QAAQ,GAAG;AACzC,UAAM,IAAI,MAAM,oBAAoB,KAAK,6BAA6B;AAAA,EACxE;AACA,SAAO;AACT;;;ACjNA,SAAS,OAAO,iBAAiB;AACjC,SAAS,SAAS,QAAAA,aAAY;;;ACN9B,SAAS,SAAS,UAAU,YAAY;AACxC,SAAS,YAAY;AA4CrB,IAAM,oBAAoB;AAE1B,SAAS,eAAe,UAA2B;AACjD,MAAI,aAAa,kBAAmB,QAAO;AAC3C,SAAO,SAAS,SAAS,OAAO,KAAK,SAAS,SAAS,QAAQ;AACjE;AAUA,eAAsB,iBACpB,MACA,UAAmC,CAAC,GACH;AACjC,QAAM,YAAY,QAAQ,aAAa;AACvC,QAAM,UAAU,QAAQ,WAAW;AAEnC,QAAM,QAAQ,MAAM,KAAK,IAAI;AAC7B,QAAM,YAAY,MAAM,YAAY,IAChC,MAAM,aAAa,MAAM,SAAS,QAAQ,aAAa,KAAK,IAC5D,CAAC,IAAI;AAET,QAAM,OAAoB,CAAC;AAC3B,QAAM,WAAiC,CAAC;AAExC,aAAW,QAAQ,WAAW;AAC5B,UAAM,MAAM,MAAM,gBAAgB,IAAI;AACtC,eAAW,EAAE,OAAO,MAAM,KAAK,KAAK;AAClC,YAAM,SAAS,mBAAmB,KAAK;AACvC,UAAI,OAAO,IAAI;AACb,aAAK,KAAK,OAAO,KAAK;AACtB;AAAA,MACF;AACA,YAAM,YAAgC,EAAE,MAAM,OAAO,QAAQ,OAAO,MAAM,QAAQ;AAClF,UAAI,cAAc,SAAS;AACzB,cAAM,IAAI;AAAA,UACR,2CAA2C,IAAI,cAAc,KAAK,KAAK,OAAO,MAAM,OAAO;AAAA,QAC7F;AAAA,MACF;AACA,eAAS,KAAK,SAAS;AAAA,IACzB;AAAA,EACF;AAEA,SAAO,EAAE,MAAM,UAAU,OAAO,UAAU;AAC5C;AAKA,eAAe,gBAAgB,MAAiE;AAC9F,QAAM,OAAO,MAAM,SAAS,MAAM,MAAM;AACxC,QAAM,UAAU,KAAK,KAAK;AAC1B,MAAI,QAAQ,WAAW,EAAG,QAAO,CAAC;AAElC,MAAI,QAAQ,WAAW,GAAG,GAAG;AAC3B,UAAM,SAAS,KAAK,MAAM,OAAO;AACjC,QAAI,CAAC,MAAM,QAAQ,MAAM,GAAG;AAC1B,YAAM,IAAI,MAAM,2BAA2B,IAAI,6BAA6B;AAAA,IAC9E;AACA,WAAO,OAAO,IAAI,CAAC,OAAO,WAAW,EAAE,OAAO,MAAM,EAAE;AAAA,EACxD;AAEA,QAAM,MAAgD,CAAC;AACvD,QAAM,QAAQ,QAAQ,MAAM,IAAI;AAChC,WAAS,IAAI,GAAG,IAAI,MAAM,QAAQ,KAAK;AACrC,UAAM,OAAO,MAAM,CAAC,EAAG,KAAK;AAC5B,QAAI,KAAK,WAAW,EAAG;AACvB,QAAI;AACF,UAAI,KAAK,EAAE,OAAO,GAAG,OAAO,KAAK,MAAM,IAAI,EAAa,CAAC;AAAA,IAC3D,SAAS,KAAK;AACZ,YAAM,IAAI;AAAA,QACR,2BAA2B,IAAI,UAAU,IAAI,CAAC,uBAC5C,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG,CACjD;AAAA,MACF;AAAA,IACF;AAAA,EACF;AACA,SAAO;AACT;AAKA,eAAe,aACb,KACA,SACA,WACmB;AACnB,QAAM,UAAU,MAAM,QAAQ,KAAK,EAAE,eAAe,KAAK,CAAC;AAC1D,QAAM,QAAkB,CAAC;AACzB,QAAM,UAAoB,CAAC;AAC3B,aAAW,SAAS,SAAS;AAC3B,QAAI,MAAM,YAAY,GAAG;AACvB,UAAI,UAAW,SAAQ,KAAK,KAAK,KAAK,MAAM,IAAI,CAAC;AACjD;AAAA,IACF;AACA,QAAI,QAAQ,MAAM,IAAI,EAAG,OAAM,KAAK,KAAK,KAAK,MAAM,IAAI,CAAC;AAAA,EAC3D;AACA,QAAM,KAAK;AACX,UAAQ,KAAK;AACb,aAAW,OAAO,SAAS;AACzB,UAAM,KAAK,GAAI,MAAM,aAAa,KAAK,SAAS,SAAS,CAAE;AAAA,EAC7D;AACA,SAAO;AACT;;;ADzFA,IAAMC,qBAAoB;AAO1B,eAAsB,mBACpB,OACA,UAAqC,CAAC,GACH;AACnC,QAAM,WAAW,OAAO,UAAU;AAElC,MAAI;AACJ,MAAI,QAAkB,CAAC;AACvB,MAAI,WAA+C,CAAC;AACpD,MAAI,UAAU;AACZ,UAAM,SAAS,MAAM,iBAAiB,OAAO,QAAQ,IAAI;AACzD,WAAO,OAAO;AACd,YAAQ,OAAO;AACf,eAAW,OAAO;AAAA,EACpB,OAAO;AACL,WAAO;AAAA,EACT;AAEA,MAAI,KAAK,WAAW,GAAG;AACrB,UAAM,IAAI;AAAA,MACR,WACI,+CAA+C,KAAK,MACpD;AAAA,IACN;AAAA,EACF;AAEA,QAAM,SAAS,MAAM,YAAY,EAAE,GAAG,QAAQ,SAAS,KAAK,CAAC;AAE7D,QAAM,SAAmC;AAAA,IACvC;AAAA,IACA,YAAY;AAAA,MACV,cAAa,oBAAI,KAAK,GAAE,YAAY;AAAA,MACpC,UAAU,KAAK;AAAA,MACf,YAAY,WAAW,QAAQ;AAAA,MAC/B;AAAA,MACA;AAAA,IACF;AAAA,IACA,WAAW;AAAA,EACb;AAEA,QAAM,SAAS,mBAAmB,QAAQ,OAAO,WAAW,QAAQ,IAAI;AACxE,MAAI,QAAQ;AACV,UAAM,MAAM,QAAQ,MAAM,GAAG,EAAE,WAAW,KAAK,CAAC;AAChD,UAAM,UAAU,QAAQ,GAAG,KAAK,UAAU,QAAQ,MAAM,CAAC,CAAC;AAAA,GAAM,MAAM;AACtE,WAAO,YAAY;AAAA,EACrB;AAEA,SAAO;AACT;AAMA,SAAS,mBACP,OACA,YACe;AACf,MAAI,CAAC,MAAO,QAAO;AAEnB,MAAI,OAAO,UAAU,UAAU;AAC7B,UAAM,eACJ,MAAM,SAAS,GAAG,KAAM,CAAC,MAAM,SAAS,OAAO,KAAK,CAAC,MAAM,SAAS,QAAQ;AAC9E,WAAO,eAAeC,MAAK,OAAOD,kBAAiB,IAAI;AAAA,EACzD;AAGA,MAAI,eAAe,MAAM;AACvB,UAAM,IAAI;AAAA,MACR;AAAA,IACF;AAAA,EACF;AACA,QAAM,SAAS,WAAW,SAAS,OAAO,KAAK,WAAW,SAAS,QAAQ;AAC3E,SAAO,SAASC,MAAK,QAAQ,UAAU,GAAGD,kBAAiB,IAAIC,MAAK,YAAYD,kBAAiB;AACnG;;;AE1DA,SAAS,WAAW,MAAgC;AAIlD,SAAO,KAAK,UAAU,CAAC,KAAK,YAAY,KAAK,GAAG,CAAC;AACnD;AAIA,SAAS,eACP,QACA,OACkC;AAClC,QAAM,MAAwC,CAAC;AAC/C,QAAM,SAAS,oBAAI,IAAY,CAAC,GAAG,OAAO,KAAK,MAAM,GAAG,GAAG,OAAO,KAAK,KAAK,CAAC,CAAC;AAC9E,aAAW,SAAS,QAAQ;AAC1B,UAAM,aAAa,OAAO,KAAK,KAAK,CAAC;AACrC,UAAM,YAAY,MAAM,KAAK,KAAK,CAAC;AACnC,UAAM,OAAO,oBAAI,IAAY,CAAC,GAAG,OAAO,KAAK,UAAU,GAAG,GAAG,OAAO,KAAK,SAAS,CAAC,CAAC;AACpF,UAAM,WAA+C,CAAC;AACtD,eAAW,OAAO,MAAM;AAKtB,YAAM,YAAY,WAAW,GAAG;AAChC,YAAM,WAAW,UAAU,GAAG;AAC9B,YAAM,IAAI,OAAO,cAAc,YAAY,OAAO,SAAS,SAAS,IAAI,YAAY;AACpF,YAAM,IAAI,OAAO,aAAa,YAAY,OAAO,SAAS,QAAQ,IAAI,WAAW;AACjF,eAAS,GAAG,IAAI;AAAA,QACd,QAAQ;AAAA,QACR,OAAO;AAAA,QACP,OAAO,MAAM,QAAQ,MAAM,OAAO,IAAI,IAAI;AAAA,MAC5C;AAAA,IACF;AACA,QAAI,KAAK,IAAI;AAAA,EACf;AACA,SAAO;AACT;AASO,SAAS,gBACd,QACA,OACoB;AACpB,QAAM,YAAY,IAAI,IAAI,OAAO,MAAM,IAAI,CAAC,MAAM,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC,CAAC;AACrE,QAAM,WAAW,IAAI,IAAI,MAAM,MAAM,IAAI,CAAC,MAAM,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC,CAAC;AAEnE,QAAM,UAAgC,CAAC;AACvC,QAAM,UAA8B,CAAC;AACrC,QAAM,QAA4B,CAAC;AAEnC,aAAW,CAAC,KAAK,UAAU,KAAK,WAAW;AACzC,UAAM,YAAY,SAAS,IAAI,GAAG;AAClC,QAAI,CAAC,WAAW;AACd,cAAQ,KAAK,UAAU;AACvB;AAAA,IACF;AACA,YAAQ,KAAK;AAAA,MACX,YAAY,WAAW;AAAA,MACvB,KAAK,WAAW;AAAA,MAChB,iBAAiB,WAAW;AAAA,MAC5B,gBAAgB,UAAU;AAAA,MAC1B,gBAAgB,UAAU,gBAAgB,WAAW;AAAA,MACrD,YAAY,eAAe,WAAW,YAAY,UAAU,UAAU;AAAA,IACxE,CAAC;AAAA,EACH;AACA,aAAW,CAAC,KAAK,SAAS,KAAK,UAAU;AACvC,QAAI,CAAC,UAAU,IAAI,GAAG,EAAG,OAAM,KAAK,SAAS;AAAA,EAC/C;AAEA,SAAO;AAAA,IACL,aAAa,OAAO;AAAA,IACpB,YAAY,MAAM;AAAA,IAClB,mBAAmB,OAAO;AAAA,IAC1B,kBAAkB,MAAM;AAAA,IACxB,gBAAgB,OAAO,gBAAgB,MAAM;AAAA,IAC7C;AAAA,IACA;AAAA,IACA;AAAA,IACA,iBAAiB,OAAO;AAAA,IACxB,gBAAgB,MAAM;AAAA,IACtB,gBAAgB,MAAM,gBAAgB,OAAO;AAAA,IAC7C,eAAe,OAAO;AAAA,IACtB,cAAc,MAAM;AAAA,IACpB,cAAc,MAAM,UAAU,OAAO;AAAA,IACrC,kBAAkB,OAAO;AAAA,IACzB,iBAAiB,MAAM;AAAA,IACvB,iBAAiB,MAAM,aAAa,OAAO;AAAA,EAC7C;AACF;AAGA,SAAS,SAAS,KAAqD;AACrE,MAAI,IAAI,YAAY,WAAW,EAAG,QAAO;AACzC,MAAI,SAAS,IAAI,YAAY,CAAC;AAC9B,aAAW,OAAO,IAAI,aAAa;AACjC,QAAI,IAAI,QAAQ,OAAO,MAAO,UAAS;AAAA,EACzC;AACA,SAAO;AACT;AAOO,SAAS,SAAS,QAAsB,OAAkC;AAC/E,QAAM,eAAe,SAAS,MAAM;AACpC,QAAM,cAAc,SAAS,KAAK;AAClC,QAAM,eACJ,OAAO,YAAY,MAAM,WAAW,gBAAgB,OAAO,UAAU,MAAM,QAAQ,IAAI;AACzF,QAAM,cACJ,gBAAgB,cAAc,gBAAgB,cAAc,WAAW,IAAI;AAE7E,QAAM,aAAa,OAAO,eAAe;AACzC,QAAM,YAAY,MAAM,eAAe;AAEvC,SAAO;AAAA,IACL,aAAa,OAAO;AAAA,IACpB,YAAY,MAAM;AAAA,IAClB,iBAAiB,OAAO;AAAA,IACxB,gBAAgB,MAAM;AAAA,IACtB,oBAAoB,OAAO,gBAAgB;AAAA,IAC3C,mBAAmB,MAAM,gBAAgB;AAAA,IACzC,mBAAmB;AAAA,IACnB,kBAAkB;AAAA,IAClB,kBAAkB,eAAe,QAAQ,cAAc,OAAO,YAAY,aAAa;AAAA,IACvF,oBAAoB,OAAO;AAAA,IAC3B,mBAAmB,MAAM;AAAA,IACzB,mBAAmB,MAAM,eAAe,OAAO;AAAA,IAC/C,uBAAuB,OAAO;AAAA,IAC9B,sBAAsB,MAAM;AAAA,IAC5B,sBAAsB,MAAM,kBAAkB,OAAO;AAAA,IACrD;AAAA,IACA;AAAA,EACF;AACF;AAQO,SAAS,wBAAwB,KAA8C;AACpF,MAAI,CAAC,IAAI,SAAU,QAAO;AAC1B,QAAM,SAAS,SAAS,GAAG;AAC3B,MAAI,CAAC,UAAU,OAAO,UAAU,IAAI,SAAS,MAAO,QAAO;AAC3D,SAAO,gBAAgB,IAAI,UAAU,MAAM;AAC7C;;;AChLA,SAAS,WAAW,GAA4B;AAC9C,SAAO,KAAK,IAAI,GAAG,EAAE,WAAW,EAAE,aAAa,CAAC;AAClD;AAOO,SAAS,gBAAgB,SAA8C;AAS5E,QAAM,MAAM,oBAAI,IAAiB;AAEjC,aAAW,UAAU,SAAS;AAC5B,UAAM,MAAM,OAAO,KAAK;AACxB,QAAI,CAAC,IAAK;AAEV,QAAI,IAAI,IAAI,IAAI,GAAG;AACnB,QAAI,CAAC,GAAG;AACN,UAAI;AAAA,QACF,QAAQ,oBAAI,IAAI;AAAA,QAChB,OAAO,oBAAI,IAAI;AAAA,QACf,OAAO,oBAAI,IAAI;AAAA,QACf,mBAAmB;AAAA,QACnB,WAAW;AAAA,QACX,eAAe;AAAA,MACjB;AACA,UAAI,IAAI,KAAK,CAAC;AAAA,IAChB;AAEA,QAAI,OAAO,MAAM,KAAM,GAAE,MAAM,IAAI,OAAO,KAAK,IAAI;AAEnD,eAAW,QAAQ,OAAO,SAAS,CAAC,GAAG;AACrC,QAAE,MAAM,IAAI,KAAK,IAAI;AACrB,iBAAW,QAAQ,KAAK,iBAAiB,CAAC,GAAG;AAC3C,UAAE,qBAAqB;AACvB,mBAAW,SAAS,KAAK,UAAU,CAAC,GAAG;AAErC,gBAAM,cAAc,MAAM,eAAe,KAAK;AAC9C,YAAE,aAAa,WAAW,KAAK;AAC/B,cAAI,CAAC,YAAa;AAClB,cAAI,YAAY,SAAS,WAAW,YAAY,SAAS,SAAS;AAChE,cAAE,gBAAgB;AAAA,UACpB;AACA,eAAK,YAAY,SAAS,QAAQ,YAAY,SAAS,YAAY,YAAY,UAAU;AACvF,cAAE,OAAO,IAAI,YAAY,QAAQ;AAAA,UACnC;AAAA,QACF;AAAA,MACF;AAAA,IACF;AAAA,EACF;AAEA,QAAM,QAAyB,oBAAI,IAAI;AACvC,aAAW,CAAC,KAAK,CAAC,KAAK,KAAK;AAC1B,UAAM,IAAI,KAAK;AAAA,MACb,WAAW;AAAA,MACX,UAAU,CAAC,GAAG,EAAE,MAAM,EAAE,KAAK;AAAA,MAC7B,OAAO,CAAC,GAAG,EAAE,KAAK,EAAE,KAAK;AAAA,MACzB,mBAAmB,EAAE;AAAA,MACrB,WAAW,EAAE,MAAM;AAAA,MACnB,WAAW,EAAE;AAAA,MACb,eAAe,EAAE;AAAA,IACnB,CAAC;AAAA,EACH;AACA,SAAO;AACT;AAoBO,SAAS,8BACd,MACA,OACiC;AACjC,QAAM,UAAU,oBAAI,IAAyB;AAC7C,QAAM,eAA4B,CAAC;AAEnC,aAAW,OAAO,MAAM;AACtB,UAAM,aAAa,MAAM,IAAI,IAAI,SAAS;AAC1C,QAAI,CAAC,cAAc,WAAW,SAAS,WAAW,GAAG;AACnD,mBAAa,KAAK,GAAG;AACrB;AAAA,IACF;AACA,eAAW,SAAS,WAAW,UAAU;AACvC,YAAM,SAAS,QAAQ,IAAI,KAAK,KAAK,CAAC;AACtC,aAAO,KAAK,GAAG;AACf,cAAQ,IAAI,OAAO,MAAM;AAAA,IAC3B;AAAA,EACF;AAEA,SAAO,EAAE,SAAS,aAAa;AACjC;;;AC/GO,SAAS,kBAAkB,MAAyD;AACzF,QAAM,EAAE,SAAS,OAAO,CAAC,GAAG,OAAO,kBAAkB,KAAK,IAAI;AAC9D,QAAM,YAAY,IAAI,IAAI,KAAK,IAAI,CAAC,MAAM,CAAC,EAAE,OAAO,CAAC,CAAC,CAAC;AAGvD,QAAM,YAAY,CAAC,WAAqC;AACtD,QAAI,OAAO,WAAW,UAAW,QAAO,SAAS,IAAI;AACrD,QAAI,CAAC,OAAO,SAAS,MAAM,EAAG,QAAO,OAAO;AAC5C,QAAI,CAAC,MAAO,QAAO;AACnB,UAAM,EAAE,KAAK,IAAI,IAAI;AACrB,QAAI,QAAQ,IAAK,QAAO;AACxB,YAAQ,SAAS,QAAQ,MAAM;AAAA,EACjC;AAGA,QAAM,QAAQ,oBAAI,IAAgC;AAClD,aAAW,OAAO,SAAS;AACzB,UAAM,OAAO,MAAM,IAAI,IAAI,KAAK,KAAK,CAAC;AACtC,SAAK,KAAK,GAAG;AACb,UAAM,IAAI,IAAI,OAAO,IAAI;AAAA,EAC3B;AAEA,QAAM,OAAoB,CAAC;AAC3B,QAAM,cAAsD,CAAC;AAE7D,aAAW,CAAC,OAAO,UAAU,KAAK,OAAO;AACvC,UAAM,aAAa,WAChB,IAAI,CAAC,OAAO,EAAE,OAAO,EAAE,OAAO,OAAO,UAAU,EAAE,MAAM,EAAE,EAAE,EAC3D,OAAO,CAAC,MAAM,OAAO,SAAS,EAAE,KAAK,CAAC;AACzC,QAAI,WAAW,WAAW,EAAG;AAE7B,UAAM,YAAY,WAAW,OAAO,CAAC,GAAG,MAAM,IAAI,EAAE,OAAO,CAAC,IAAI,WAAW;AAE3E,UAAM,UAAU,UAAU,IAAI,KAAK,KAAM,EAAE,MAAM;AAEjD,UAAM,cAAiC;AAAA,MACrC,UAAU,OAAO,YAAY,WAAW,IAAI,CAAC,MAAM,CAAC,EAAE,OAAO,EAAE,QAAQ,EAAE,MAAM,CAAC,CAAC,CAAC;AAAA,MAClF,YAAY,EAAE,QAAQ,UAAU;AAAA,MAChC,WAAW;AAAA,IACb;AAEA,UAAM,UAAsB;AAAA;AAAA;AAAA;AAAA,MAI1B,cAAc;AAAA,MACd,KAAK,OAAO,YAAY,WAAW,IAAI,CAAC,MAAM,CAAC,SAAS,EAAE,KAAK,IAAI,EAAE,KAAK,CAAC,CAAC;AAAA,MAC5E;AAAA,IACF;AAEA,SAAK,KAAK;AAAA,MACR;AAAA,MACA,cAAc,QAAQ,gBAAgB;AAAA,MACtC,aAAa,QAAQ,eAAe;AAAA,MACpC,MAAM;AAAA,MACN,OAAO,QAAQ,SAAS;AAAA,MACxB,YAAY,QAAQ,cAAc;AAAA,MAClC,YAAY,QAAQ,cAAc;AAAA,MAClC,WAAW,QAAQ,aAAa;AAAA,MAChC,QAAQ,QAAQ,UAAU;AAAA,MAC1B,SAAS,QAAQ,WAAW;AAAA,MAC5B,YAAY,EAAE,OAAO,GAAG,QAAQ,EAAE;AAAA,MAClC;AAAA,MACA,UAAU,QAAQ,YAAY;AAAA,IAChC,CAAc;AAEd,QAAI,iBAAiB;AACnB,iBAAW,KAAK,WAAY,aAAY,KAAK,EAAE,OAAO,OAAO,EAAE,OAAO,OAAO,EAAE,MAAM,CAAC;AAAA,IACxF;AAAA,EACF;AAEA,SAAO,EAAE,MAAM,YAAY;AAC7B;;;ACvHA,IAAM,aAAa,CAAC,gBAAgB,cAAc,OAAO;AACzD,IAAM,aAAa,CAAC,gBAAgB,GAAG,qBAAqB,OAAO;AACnE,IAAM,YAAY,CAAC,mBAAmB,yBAAyB,YAAY,MAAM;AACjF,IAAM,mBAAmB,CAAC,GAAG,2BAA2B,oBAAoB,WAAW;AACvF,IAAM,oBAAoB,CAAC,GAAG,4BAA4B,qBAAqB,YAAY;AAC3F,IAAM,mBAAmB,CAAC,sBAAsB,aAAa;AAC7D,IAAM,mBAAmB,CAAC,sBAAsB,aAAa;AAUtD,SAAS,cAAc,MAAyC;AACrE,QAAM,EAAE,OAAO,eAAe,WAAW,eAAe,cAAc,IAAI;AAC1E,QAAM,UAAU,WAAW,KAAK;AAEhC,QAAM,OAAoB,CAAC;AAC3B,aAAW,CAAC,UAAUE,WAAU,KAAK,SAAS;AAC5C,UAAM,OAAO,SAASA,WAAU;AAChC,QAAI,CAAC,KAAM;AAEX,UAAM,SAAS,KAAK,IAAI,IAAI,KAAK,kBAAkB,KAAK,qBAAqB,GAAS;AACtF,UAAM,QAAQ,eAAeA,aAAY,UAAU,KAAK;AACxD,UAAM,UAAU,eAAeA,aAAY,SAAS,KAAK;AACzD,UAAM,cAAc,eAAeA,aAAY,gBAAgB,KAAK;AACpE,UAAM,eAAe,eAAeA,aAAY,iBAAiB,KAAK;AACtE,UAAM,aAAa,eAAeA,aAAY,gBAAgB,KAAK;AACnE,UAAM,aAAa,eAAeA,aAAY,gBAAgB,KAAK;AACnE,UAAM,QAAQ,eAAeA,aAAY,UAAU;AAEnD,UAAM,aAAa,oBAAoBA,WAAU;AACjD,UAAM,aAA4B;AAAA,MAChC,OAAO;AAAA,MACP,QAAQ;AAAA,IACV;AAEA,UAAM,cACJ,UAAU,SACN;AAAA,MACE,UAAU,EAAE,gBAAgB,EAAE,MAAM,EAAE;AAAA,MACtC,YAAY,EAAE,MAAM;AAAA,MACpB,WAAW;AAAA,IACb,IACA;AAEN,UAAM,YAAYA,YAAW,KAAK,CAAC,MAAM,EAAE,QAAQ,SAAS,OAAO;AACnE,UAAM,UAAsB;AAAA,MAC1B,GAAI,KAAK,iBAAiB,WAAW,EAAE,aAAa,MAAM,IAAI,EAAE,cAAc,MAAM;AAAA,MACpF,KAAK;AAAA,MACL,GAAI,cAAc,EAAE,YAAY,IAAI,CAAC;AAAA,IACvC;AAEA,SAAK,KAAK;AAAA,MACR,OAAO;AAAA,MACP;AAAA,MACA,aAAc,KAAK,WAAW,oBAAoB,KAA4B;AAAA,MAC9E,MAAM;AAAA,MACN;AAAA,MACA;AAAA,MACA;AAAA,MACA,WAAY,KAAK,WAAW,mBAAmB,KAA4B;AAAA,MAC3E;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA,UAAU;AAAA,MACV,GAAI,YAAY,EAAE,aAAa,UAAU,KAAK,IAAI,CAAC;AAAA,IACrD,CAAc;AAAA,EAChB;AACA,SAAO;AACT;AAIA,SAAS,WAAW,OAAwD;AAC1E,QAAM,IAAI,oBAAI,IAA8B;AAC5C,aAAW,QAAQ,OAAO;AACxB,UAAM,MAAO,KAAK,cAAc,KAA4B,KAAK;AACjE,UAAM,OAAO,EAAE,IAAI,GAAG,KAAK,CAAC;AAC5B,SAAK,KAAK,IAAI;AACd,MAAE,IAAI,KAAK,IAAI;AAAA,EACjB;AACA,SAAO;AACT;AAEA,SAAS,SAAS,OAAqD;AACrE,SAAO,MAAM,KAAK,CAAC,MAAM,CAAC,EAAE,YAAY,KAAK,MAAM,CAAC;AACtD;AAEA,SAAS,eAAe,OAAyB,MAAoC;AACnF,aAAW,QAAQ,OAAO;AACxB,eAAW,OAAO,MAAM;AACtB,YAAM,IAAI,KAAK,WAAW,GAAG;AAC7B,UAAI,OAAO,MAAM,YAAY,EAAE,SAAS,EAAG,QAAO;AAAA,IACpD;AAAA,EACF;AACA,SAAO;AACT;AAEA,SAAS,eAAe,OAAyB,MAAoC;AACnF,aAAW,QAAQ,OAAO;AACxB,eAAW,OAAO,MAAM;AACtB,YAAM,IAAI,KAAK,WAAW,GAAG;AAC7B,UAAI,OAAO,MAAM,YAAY,OAAO,SAAS,CAAC,EAAG,QAAO;AACxD,UAAI,OAAO,MAAM,UAAU;AACzB,cAAM,SAAS,OAAO,CAAC;AACvB,YAAI,OAAO,SAAS,MAAM,EAAG,QAAO;AAAA,MACtC;AAAA,IACF;AAAA,EACF;AACA,SAAO;AACT;AAEA,SAAS,oBAAoB,OAAiD;AAC5E,QAAM,MAA8B,CAAC;AACrC,aAAW,QAAQ,OAAO;AACxB,eAAW,CAAC,GAAG,CAAC,KAAK,OAAO,QAAQ,KAAK,UAAU,GAAG;AACpD,UAAI,OAAO,MAAM,YAAY,OAAO,SAAS,CAAC,EAAG,KAAI,CAAC,IAAI;AAAA,IAC5D;AAAA,EACF;AACA,SAAO;AACT;","names":["join","ANALYSIS_ARTIFACT","join","groupSpans"]}

package/dist/{gepa-C1NCIZ9o.d.ts → gepa-Dprxvz8r.d.ts}

@@ -1,4 +1,4 @@
-import { S as Scenario, C as CampaignResult, l as GateResult, D as DispatchFn, a as JudgeConfig, L as LabeledScenarioStore, i as CampaignTraceWriter, n as GenerationRecord, M as MutableSurface, P as ParetoParent, c as SurfaceProposer, G as Gate } from './types-DQRY8ZT-.js';
+import { S as Scenario, C as CampaignResult, G as GateResult, D as DispatchFn, a as JudgeConfig, L as LabeledScenarioStore, c as CampaignTraceWriter, d as GenerationRecord, M as MutableSurface, P as ParetoParent, e as SurfaceProposer, f as Gate } from './types-BMahhhio.js';
 import { L as LlmClientOptions } from './llm-client-Bj7g0rqu.js';
 
 /**
@@ -411,4 +411,4 @@ declare function extractH2Sections(text: string): string[];
  *  whitespace as identical. Exported for tests + consumer-side validators. */
 declare function countSentenceEdits(baseline: string, candidate: string): number;
 
-export { type CampaignStorage as C, type GepaProposerOptions as G, type OpenAutoPrOptions as O, type RunOptimizationOptions as R, type RunImprovementLoopResult as a, type RunCampaignOptions as b, type RunImprovementLoopOptions as c, runImprovementLoop as d, type GepaProposerConstraints as e, fsCampaignStorage as f, gepaProposer as g, type OpenAutoPrResult as h, inMemoryCampaignStorage as i, type RunOptimizationResult as j, countSentenceEdits as k, defaultRenderDiff as l, extractH2Sections as m, runOptimization as n, openAutoPr as o, runCampaign as r, surfaceHash as s };
+export { type CampaignStorage as C, type GepaProposerOptions as G, type OpenAutoPrOptions as O, type RunCampaignOptions as R, type RunOptimizationOptions as a, type RunImprovementLoopResult as b, type RunImprovementLoopOptions as c, runImprovementLoop as d, type GepaProposerConstraints as e, fsCampaignStorage as f, gepaProposer as g, type OpenAutoPrResult as h, inMemoryCampaignStorage as i, type RunOptimizationResult as j, countSentenceEdits as k, defaultRenderDiff as l, extractH2Sections as m, runOptimization as n, openAutoPr as o, runCampaign as r, surfaceHash as s };

package/dist/hosted/index.d.ts

@@ -1,4 +1,4 @@
-import { M as MutableSurface, d as GateDecision } from '../types-DQRY8ZT-.js';
+import { M as MutableSurface, i as GateDecision } from '../types-BMahhhio.js';
 import { I as InsightReport } from '../insight-report-BnRjTibG.js';
 import '../run-record-CP2ObebC.js';
 import '@tangle-network/agent-interface';

package/dist/index.d.ts

@@ -27,7 +27,7 @@ export { CaptureFetchContext, CaptureFetchOptions, ExportableSpan, ExtractedUsag
 import { a as AnalyzeTracesInput, A as AnalyzeTracesOptions, b as AnalyzeTracesResult } from './analyst-C8HHvfJp.js';
 export { c as AnalyzeTracesTurnSnapshot, d as analyzeTraces } from './analyst-C8HHvfJp.js';
 export { D as DEFAULT_TRACE_ANALYST_BUDGETS, b as DatasetOverview, E as ErrorCluster, Q as QueryTracesPage, S as SearchSpanResult, c as SearchTraceResult, d as SpanMatchRecord, e as TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, T as TraceAnalysisStore, f as TraceAnalystByteBudgets, g as TraceAnalystFilters, a as TraceAnalystSpan, h as TraceAnalystSpanKind, i as TraceAnalystSpanStatus, j as TraceAnalystTraceSummary, V as ViewSpansResult, k as ViewTraceOversized, l as ViewTraceResult } from './store-C1YxJDEK.js';
-import { a as JudgeConfig, S as Scenario$1, G as Gate, J as JudgeScore } from './types-DQRY8ZT-.js';
+import { a as JudgeConfig, S as Scenario$1, f as Gate, J as JudgeScore } from './types-BMahhhio.js';
 import { A as AnalyzeRunsOptions } from './analyze-runs-DtT6F_6T.js';
 import { S as SteeringBundle } from './harness-optimizer-mOl9XX_O.js';
 export { D as DEFAULT_HARNESS_OBJECTIVES, H as HarnessAdapter, a as HarnessExperimentConfig, b as HarnessExperimentResult, c as HarnessIntervention, d as HarnessRunRequest, e as HarnessRunResult, f as HarnessScenario, g as HarnessSelection, h as HarnessVariant, i as HarnessVariantReport, M as MeasurementPolicy, j as SteeringDelta, k as SteeringRolePrompt, W as WorkflowTopology, m as mergeSteeringBundle, r as renderSteeringText, l as runHarnessExperiment, s as selectHarnessVariant, n as summarizeHarnessResults } from './harness-optimizer-mOl9XX_O.js';
@@ -72,7 +72,7 @@ export { C as CallbackResearcher, d as CallbackResearcherOptions, e as CampaignF
 export { G as GainDistributionBin, a as GainDistributionFigureSpec, b as GainDistributionOptions, m as GateDecision, n as GateEvidence, H as HeldOutGate, o as HeldOutGateConfig, q as HeldOutGateRejectionCode, P as ParetoFigureSpec, c as ParetoPoint, R as RESEARCH_REPORT_HARD_PAIR_FLOOR, d as ResearchReport, e as ResearchReportCandidate, f as ResearchReportDecision, g as ResearchReportMethodology, h as ResearchReportOptions, i as ResearchReportRecommendation, S as SummaryTable, j as SummaryTableOptions, k as SummaryTableRow, l as gainHistogram, p as paretoChart, r as researchReport, s as summaryTable } from './summary-report-CInXwsza.js';
 export { L as LockedJsonlAppender } from './testing-C21CHsq2.js';
 export { I as InterimReleaseConfidence, a as InterimReleaseConfidenceInput, P as PairedEvalueOptions, b as PairedEvalueSequence, c as PairedEvalueStep, S as SequentialDecision, e as evaluateInterimReleaseConfidence, p as pairedEvalueSequence } from './sequential-5iSVfzl2.js';
-import { e as GepaProposerConstraints, a as RunImprovementLoopResult } from './gepa-C1NCIZ9o.js';
+import { e as GepaProposerConstraints, b as RunImprovementLoopResult } from './gepa-Dprxvz8r.js';
 export { IntegrityResult, IntegrityViolation, JourneySpec, PerfBaseline, PerfGateResult, PerfRegression, PerfScenario, PerfStat, ScenarioAxes, assertRecordIntegrity, checkRecordIntegrity, expandMatrix, gatePerf, scenarioKey, summarizeRecords } from './perf/index.js';
 import '@ax-llm/ax';
 import 'zod';

package/dist/index.js

@@ -132,7 +132,7 @@ import {
 import {
   buildDefaultAnalystRegistry,
   computeTraceMetrics
-} from "./chunk-AQ5WQAIV.js";
+} from "./chunk-3NHEO6ZC.js";
 import {
   LockedJsonlAppender,
   Mutex

package/dist/multishot/index.d.ts

@@ -1,4 +1,4 @@
-import { J as JudgeScore } from '../types-DQRY8ZT-.js';
+import { J as JudgeScore } from '../types-BMahhhio.js';
 import { AgentProfile } from '@tangle-network/agent-interface';
 import { M as MatrixResult } from '../types-BUxNaJ8c.js';
 import '../run-record-CP2ObebC.js';

package/dist/openapi.json

@@ -2,7 +2,7 @@
   "openapi": "3.1.0",
   "info": {
     "title": "@tangle-network/agent-eval — wire protocol",
-    "version": "0.95.1",
+    "version": "0.96.0",
     "description": "HTTP and stdio RPC interface to agent-eval. The TypeScript runtime is the source of truth; this spec is the contract that cross-language clients (Python, Rust, Go) generate from.\n\nWire-protocol version: 1.0.0. Bumps on breaking changes to request/response schemas.",
     "contact": {
       "name": "Tangle Network",

package/dist/{provenance-CncDq9qE.d.ts → provenance-CIxfBnkl.d.ts}

@@ -1,9 +1,9 @@
-import { S as Scenario, G as Gate, l as GateResult, k as GateContext, C as CampaignResult, p as Mutator, c as SurfaceProposer, M as MutableSurface, d as GateDecision } from './types-DQRY8ZT-.js';
+import { S as Scenario, f as Gate, G as GateResult, g as GateContext, C as CampaignResult, h as Mutator, e as SurfaceProposer, M as MutableSurface, i as GateDecision } from './types-BMahhhio.js';
 import { R as RedTeamCase } from './red-team-BWdoyleI.js';
 import { R as RunRecord } from './run-record-CP2ObebC.js';
 import { D as Direction } from './pareto-E-pembql.js';
 import { a as PairedBootstrapResult } from './statistics-CCJpTGOS.js';
-import { b as RunCampaignOptions, C as CampaignStorage } from './gepa-C1NCIZ9o.js';
+import { R as RunCampaignOptions, C as CampaignStorage } from './gepa-Dprxvz8r.js';
 import { HostedClient, TraceSpanEvent } from './hosted/index.js';
 
 /**

package/dist/rl.d.ts

@@ -10,7 +10,7 @@ import { R as Researcher, F as FailureMode, S as SteeringChange, E as Experiment
 export { r as runEvalCampaign } from './researcher-Jr8ME1dZ.js';
 import { a as VerificationReport } from './multi-layer-verifier-DUZXrPDA.js';
 import { I as InterimReleaseConfidence } from './sequential-5iSVfzl2.js';
-import { C as CampaignResult } from './types-DQRY8ZT-.js';
+import { C as CampaignResult } from './types-BMahhhio.js';
 import '@tangle-network/agent-interface';
 import './errors-CzMUYo7b.js';
 import './schema-m0gsnbt3.js';

package/dist/{types-DQRY8ZT-.d.ts → types-BMahhhio.d.ts}

@@ -519,4 +519,4 @@ interface CampaignResult<TArtifact = unknown, TScenario extends Scenario = Scena
     scenarios: Array<Pick<TScenario, 'id' | 'kind'>>;
 }
 
-export { type JudgeAggregate as A, type ScenarioAggregate as B, type CampaignResult as C, type DispatchFn as D, isProposedCandidate as E, labelTrustRank as F, type Gate as G, type JudgeScore as J, type LabeledScenarioStore as L, type MutableSurface as M, type OptimizationProposer as O, type ParetoParent as P, type RedactionStatus as R, type Scenario as S, type TraceSpan as T, type JudgeConfig as a, type DispatchContext as b, type SurfaceProposer as c, type GateDecision as d, type CampaignAggregates as e, type CampaignArtifactWriter as f, type CampaignCellResult as g, type CampaignCostMeter as h, type CampaignTraceWriter as i, type CodeSurface as j, type GateContext as k, type GateResult as l, type GenerationCandidate as m, type GenerationRecord as n, type JudgeDimension as o, type Mutator as p, type OptimizerConfig as q, type SessionScript as r, type LabeledScenarioWrite as s, type LabeledScenarioSampleArgs as t, type LabeledScenarioRecord as u, type LabelTrust as v, type ProposedCandidate as w, type ProposeContext as x, type LabeledScenarioSource as y, type CampaignTokenUsage as z };
+export { type JudgeAggregate as A, type ScenarioAggregate as B, type CampaignResult as C, type DispatchFn as D, isProposedCandidate as E, labelTrustRank as F, type GateResult as G, type JudgeScore as J, type LabeledScenarioStore as L, type MutableSurface as M, type OptimizationProposer as O, type ParetoParent as P, type RedactionStatus as R, type Scenario as S, type TraceSpan as T, type JudgeConfig as a, type DispatchContext as b, type CampaignTraceWriter as c, type GenerationRecord as d, type SurfaceProposer as e, type Gate as f, type GateContext as g, type Mutator as h, type GateDecision as i, type CampaignAggregates as j, type CampaignArtifactWriter as k, type CampaignCellResult as l, type CampaignCostMeter as m, type CodeSurface as n, type GenerationCandidate as o, type JudgeDimension as p, type OptimizerConfig as q, type SessionScript as r, type LabeledScenarioWrite as s, type LabeledScenarioSampleArgs as t, type LabeledScenarioRecord as u, type LabelTrust as v, type ProposedCandidate as w, type ProposeContext as x, type LabeledScenarioSource as y, type CampaignTokenUsage as z };

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tangle-network/agent-eval",
-  "version": "0.95.1",
+  "version": "0.96.0",
   "description": "Evaluate and improve AI agents from runs, traces, judges, and feedback. Compare candidates, cluster failures, measure lift, and gate releases.",
   "homepage": "https://github.com/tangle-network/agent-eval#readme",
   "repository": {

package/dist/chunk-AQ5WQAIV.js.map

@@ -1 +0,0 @@
-{"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 {\n  LLM_INPUT_TOKEN_ATTR_KEYS,\n  LLM_OUTPUT_TOKEN_ATTR_KEYS,\n  TOOL_NAME_ATTR_KEYS,\n} from '../trace/otlp-attributes'\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 numAttr(attrs: Record<string, unknown>, keys: readonly string[]): number | null {\n  for (const key of keys) {\n    const value = num(attrs[key])\n    if (value !== null) return value\n  }\n  return null\n}\nfunction inputTokensOf(s: TraceAnalystSpan): number | null {\n  return (\n    numAttr(s.attributes, LLM_INPUT_TOKEN_ATTR_KEYS) ?? num(s.attributes['llm.usage.input_tokens'])\n  )\n}\nfunction outputTokensOf(s: TraceAnalystSpan): number | null {\n  return (\n    numAttr(s.attributes, LLM_OUTPUT_TOKEN_ATTR_KEYS) ??\n    num(s.attributes['llm.usage.output_tokens'])\n  )\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  for (const key of TOOL_NAME_ATTR_KEYS) {\n    const t = s.attributes[key]\n    if (typeof t === 'string' && t.length > 0) return t\n  }\n  return 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":";;;;;;;;;;;;;;;AAiDA,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,QAAQ,OAAgC,MAAwC;AACvF,aAAW,OAAO,MAAM;AACtB,UAAM,QAAQ,IAAI,MAAM,GAAG,CAAC;AAC5B,QAAI,UAAU,KAAM,QAAO;AAAA,EAC7B;AACA,SAAO;AACT;AACA,SAAS,cAAc,GAAoC;AACzD,SACE,QAAQ,EAAE,YAAY,yBAAyB,KAAK,IAAI,EAAE,WAAW,wBAAwB,CAAC;AAElG;AACA,SAAS,eAAe,GAAoC;AAC1D,SACE,QAAQ,EAAE,YAAY,0BAA0B,KAChD,IAAI,EAAE,WAAW,yBAAyB,CAAC;AAE/C;AACA,SAAS,OAAO,GAAoC;AAClD,SAAO,IAAI,EAAE,WAAW,IAAI;AAC9B;AACA,SAAS,WAAW,GAAoC;AACtD,MAAI,EAAE,UAAW,QAAO,EAAE;AAC1B,aAAW,OAAO,qBAAqB;AACrC,UAAM,IAAI,EAAE,WAAW,GAAG;AAC1B,QAAI,OAAO,MAAM,YAAY,EAAE,SAAS,EAAG,QAAO;AAAA,EACpD;AACA,SAAO;AACT;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;;;ACvKA,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":[]}