@tangle-network/agent-eval 0.115.3 → 0.117.0

package/dist/cli.js

@@ -5,8 +5,10 @@ import {
   runRpcBatch,
   runRpcOnce,
   startServer
-} from "./chunk-4D5RVB3W.js";
-import "./chunk-GY4SYVPJ.js";
+} from "./chunk-LTVG32KX.js";
+import "./chunk-NJC7U437.js";
+import "./chunk-VCTY3W6J.js";
+import "./chunk-VI2UW6B6.js";
 import "./chunk-PC4UYEBM.js";
 import "./chunk-ONWEPEDO.js";
 import "./chunk-PZ5AY32C.js";

package/dist/cli.js.map

@@ -1 +1 @@
-{"version":3,"sources":["../src/cli.ts"],"sourcesContent":["#!/usr/bin/env node\n/**\n * agent-eval CLI.\n *\n *   agent-eval serve [--port 5005] [--host 127.0.0.1]\n *   agent-eval rpc <method>          # one request from stdin → one response on stdout\n *   agent-eval rpc-batch <method>    # JSONL stdin → JSONL stdout\n *   agent-eval openapi [--out path]  # write OpenAPI spec\n *   agent-eval version\n *\n * <method> is one of: judge, listRubrics, version. When omitted, the\n * stdin payload must be a full {method, params} envelope.\n */\nimport { writeFileSync } from 'node:fs'\nimport { handleVersion } from './wire/handlers'\nimport { buildOpenApi } from './wire/openapi'\nimport { runRpcBatch, runRpcOnce } from './wire/rpc'\nimport { startServer } from './wire/server'\n\ninterface Args {\n  command: string\n  positional: string[]\n  flags: Record<string, string>\n}\n\nfunction parseArgs(argv: string[]): Args {\n  const [command, ...rest] = argv\n  const positional: string[] = []\n  const flags: Record<string, string> = {}\n  for (let i = 0; i < rest.length; i++) {\n    const tok = rest[i]!\n    if (tok.startsWith('--')) {\n      const key = tok.slice(2)\n      const next = rest[i + 1]\n      if (next != null && !next.startsWith('--')) {\n        flags[key] = next\n        i++\n      } else {\n        flags[key] = 'true'\n      }\n    } else {\n      positional.push(tok)\n    }\n  }\n  return { command: command ?? 'help', positional, flags }\n}\n\nconst HELP = `agent-eval — wire-protocol entry point.\n\nCommands:\n  serve [--port 5005] [--host 127.0.0.1]\n        Start the HTTP server. POST /v1/judge, GET /v1/rubrics, GET /v1/version, GET /openapi.json.\n  rpc <method>\n        Read one JSON object from stdin (the params for <method>), write one\n        JSON object to stdout. Method ∈ {judge, listRubrics, version}.\n  rpc-batch <method>\n        Like 'rpc' but JSONL in / JSONL out.\n  openapi [--out openapi.json]\n        Write the OpenAPI 3.1 spec.\n  version\n        Print server + wire-protocol version JSON.\n\nWithout arguments, prints this help.`\n\nasync function main(): Promise<number> {\n  const { command, positional, flags } = parseArgs(process.argv.slice(2))\n\n  switch (command) {\n    case 'serve': {\n      const port = Number(flags.port ?? 5005)\n      const host = flags.host ?? '127.0.0.1'\n      const server = startServer({ port, host })\n      // Keep process alive on SIGINT/SIGTERM\n      const shutdown = (sig: string) => {\n        // eslint-disable-next-line no-console\n        console.log(`[agent-eval] received ${sig}, shutting down`)\n        server.close(() => process.exit(0))\n        // Force exit after 5s if close hangs\n        setTimeout(() => process.exit(1), 5000).unref()\n      }\n      process.on('SIGINT', () => shutdown('SIGINT'))\n      process.on('SIGTERM', () => shutdown('SIGTERM'))\n      // Block forever\n      await new Promise(() => {})\n      return 0\n    }\n    case 'rpc': {\n      const [method] = positional\n      return await runRpcOnce(method)\n    }\n    case 'rpc-batch': {\n      const [method] = positional\n      return await runRpcBatch(method)\n    }\n    case 'openapi': {\n      const out = flags.out ?? 'openapi.json'\n      const spec = buildOpenApi(handleVersion().version)\n      writeFileSync(out, `${JSON.stringify(spec, null, 2)}\\n`, 'utf-8')\n      // eslint-disable-next-line no-console\n      console.log(`[agent-eval] wrote OpenAPI 3.1 spec to ${out}`)\n      return 0\n    }\n    case 'version': {\n      process.stdout.write(`${JSON.stringify(handleVersion(), null, 2)}\\n`)\n      return 0\n    }\n    case 'help':\n    case '--help':\n    case '-h':\n    case '':\n      process.stdout.write(`${HELP}\\n`)\n      return 0\n    default:\n      process.stderr.write(`unknown command: ${command}\\n${HELP}\\n`)\n      return 1\n  }\n}\n\nmain()\n  .then((code) => process.exit(code))\n  .catch((err) => {\n    // eslint-disable-next-line no-console\n    console.error('[agent-eval] cli error:', err)\n    process.exit(1)\n  })\n"],"mappings":";;;;;;;;;;;;;;AAaA,SAAS,qBAAqB;AAY9B,SAAS,UAAU,MAAsB;AACvC,QAAM,CAAC,SAAS,GAAG,IAAI,IAAI;AAC3B,QAAM,aAAuB,CAAC;AAC9B,QAAM,QAAgC,CAAC;AACvC,WAAS,IAAI,GAAG,IAAI,KAAK,QAAQ,KAAK;AACpC,UAAM,MAAM,KAAK,CAAC;AAClB,QAAI,IAAI,WAAW,IAAI,GAAG;AACxB,YAAM,MAAM,IAAI,MAAM,CAAC;AACvB,YAAM,OAAO,KAAK,IAAI,CAAC;AACvB,UAAI,QAAQ,QAAQ,CAAC,KAAK,WAAW,IAAI,GAAG;AAC1C,cAAM,GAAG,IAAI;AACb;AAAA,MACF,OAAO;AACL,cAAM,GAAG,IAAI;AAAA,MACf;AAAA,IACF,OAAO;AACL,iBAAW,KAAK,GAAG;AAAA,IACrB;AAAA,EACF;AACA,SAAO,EAAE,SAAS,WAAW,QAAQ,YAAY,MAAM;AACzD;AAEA,IAAM,OAAO;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAiBb,eAAe,OAAwB;AACrC,QAAM,EAAE,SAAS,YAAY,MAAM,IAAI,UAAU,QAAQ,KAAK,MAAM,CAAC,CAAC;AAEtE,UAAQ,SAAS;AAAA,IACf,KAAK,SAAS;AACZ,YAAM,OAAO,OAAO,MAAM,QAAQ,IAAI;AACtC,YAAM,OAAO,MAAM,QAAQ;AAC3B,YAAM,SAAS,YAAY,EAAE,MAAM,KAAK,CAAC;AAEzC,YAAM,WAAW,CAAC,QAAgB;AAEhC,gBAAQ,IAAI,yBAAyB,GAAG,iBAAiB;AACzD,eAAO,MAAM,MAAM,QAAQ,KAAK,CAAC,CAAC;AAElC,mBAAW,MAAM,QAAQ,KAAK,CAAC,GAAG,GAAI,EAAE,MAAM;AAAA,MAChD;AACA,cAAQ,GAAG,UAAU,MAAM,SAAS,QAAQ,CAAC;AAC7C,cAAQ,GAAG,WAAW,MAAM,SAAS,SAAS,CAAC;AAE/C,YAAM,IAAI,QAAQ,MAAM;AAAA,MAAC,CAAC;AAC1B,aAAO;AAAA,IACT;AAAA,IACA,KAAK,OAAO;AACV,YAAM,CAAC,MAAM,IAAI;AACjB,aAAO,MAAM,WAAW,MAAM;AAAA,IAChC;AAAA,IACA,KAAK,aAAa;AAChB,YAAM,CAAC,MAAM,IAAI;AACjB,aAAO,MAAM,YAAY,MAAM;AAAA,IACjC;AAAA,IACA,KAAK,WAAW;AACd,YAAM,MAAM,MAAM,OAAO;AACzB,YAAM,OAAO,aAAa,cAAc,EAAE,OAAO;AACjD,oBAAc,KAAK,GAAG,KAAK,UAAU,MAAM,MAAM,CAAC,CAAC;AAAA,GAAM,OAAO;AAEhE,cAAQ,IAAI,0CAA0C,GAAG,EAAE;AAC3D,aAAO;AAAA,IACT;AAAA,IACA,KAAK,WAAW;AACd,cAAQ,OAAO,MAAM,GAAG,KAAK,UAAU,cAAc,GAAG,MAAM,CAAC,CAAC;AAAA,CAAI;AACpE,aAAO;AAAA,IACT;AAAA,IACA,KAAK;AAAA,IACL,KAAK;AAAA,IACL,KAAK;AAAA,IACL,KAAK;AACH,cAAQ,OAAO,MAAM,GAAG,IAAI;AAAA,CAAI;AAChC,aAAO;AAAA,IACT;AACE,cAAQ,OAAO,MAAM,oBAAoB,OAAO;AAAA,EAAK,IAAI;AAAA,CAAI;AAC7D,aAAO;AAAA,EACX;AACF;AAEA,KAAK,EACF,KAAK,CAAC,SAAS,QAAQ,KAAK,IAAI,CAAC,EACjC,MAAM,CAAC,QAAQ;AAEd,UAAQ,MAAM,2BAA2B,GAAG;AAC5C,UAAQ,KAAK,CAAC;AAChB,CAAC;","names":[]}
+{"version":3,"sources":["../src/cli.ts"],"sourcesContent":["#!/usr/bin/env node\n/**\n * agent-eval CLI.\n *\n *   agent-eval serve [--port 5005] [--host 127.0.0.1]\n *   agent-eval rpc <method>          # one request from stdin → one response on stdout\n *   agent-eval rpc-batch <method>    # JSONL stdin → JSONL stdout\n *   agent-eval openapi [--out path]  # write OpenAPI spec\n *   agent-eval version\n *\n * <method> is one of: judge, listRubrics, version. When omitted, the\n * stdin payload must be a full {method, params} envelope.\n */\nimport { writeFileSync } from 'node:fs'\nimport { handleVersion } from './wire/handlers'\nimport { buildOpenApi } from './wire/openapi'\nimport { runRpcBatch, runRpcOnce } from './wire/rpc'\nimport { startServer } from './wire/server'\n\ninterface Args {\n  command: string\n  positional: string[]\n  flags: Record<string, string>\n}\n\nfunction parseArgs(argv: string[]): Args {\n  const [command, ...rest] = argv\n  const positional: string[] = []\n  const flags: Record<string, string> = {}\n  for (let i = 0; i < rest.length; i++) {\n    const tok = rest[i]!\n    if (tok.startsWith('--')) {\n      const key = tok.slice(2)\n      const next = rest[i + 1]\n      if (next != null && !next.startsWith('--')) {\n        flags[key] = next\n        i++\n      } else {\n        flags[key] = 'true'\n      }\n    } else {\n      positional.push(tok)\n    }\n  }\n  return { command: command ?? 'help', positional, flags }\n}\n\nconst HELP = `agent-eval — wire-protocol entry point.\n\nCommands:\n  serve [--port 5005] [--host 127.0.0.1]\n        Start the HTTP server. POST /v1/judge, GET /v1/rubrics, GET /v1/version, GET /openapi.json.\n  rpc <method>\n        Read one JSON object from stdin (the params for <method>), write one\n        JSON object to stdout. Method ∈ {judge, listRubrics, version}.\n  rpc-batch <method>\n        Like 'rpc' but JSONL in / JSONL out.\n  openapi [--out openapi.json]\n        Write the OpenAPI 3.1 spec.\n  version\n        Print server + wire-protocol version JSON.\n\nWithout arguments, prints this help.`\n\nasync function main(): Promise<number> {\n  const { command, positional, flags } = parseArgs(process.argv.slice(2))\n\n  switch (command) {\n    case 'serve': {\n      const port = Number(flags.port ?? 5005)\n      const host = flags.host ?? '127.0.0.1'\n      const server = startServer({ port, host })\n      // Keep process alive on SIGINT/SIGTERM\n      const shutdown = (sig: string) => {\n        // eslint-disable-next-line no-console\n        console.log(`[agent-eval] received ${sig}, shutting down`)\n        server.close(() => process.exit(0))\n        // Force exit after 5s if close hangs\n        setTimeout(() => process.exit(1), 5000).unref()\n      }\n      process.on('SIGINT', () => shutdown('SIGINT'))\n      process.on('SIGTERM', () => shutdown('SIGTERM'))\n      // Block forever\n      await new Promise(() => {})\n      return 0\n    }\n    case 'rpc': {\n      const [method] = positional\n      return await runRpcOnce(method)\n    }\n    case 'rpc-batch': {\n      const [method] = positional\n      return await runRpcBatch(method)\n    }\n    case 'openapi': {\n      const out = flags.out ?? 'openapi.json'\n      const spec = buildOpenApi(handleVersion().version)\n      writeFileSync(out, `${JSON.stringify(spec, null, 2)}\\n`, 'utf-8')\n      // eslint-disable-next-line no-console\n      console.log(`[agent-eval] wrote OpenAPI 3.1 spec to ${out}`)\n      return 0\n    }\n    case 'version': {\n      process.stdout.write(`${JSON.stringify(handleVersion(), null, 2)}\\n`)\n      return 0\n    }\n    case 'help':\n    case '--help':\n    case '-h':\n    case '':\n      process.stdout.write(`${HELP}\\n`)\n      return 0\n    default:\n      process.stderr.write(`unknown command: ${command}\\n${HELP}\\n`)\n      return 1\n  }\n}\n\nmain()\n  .then((code) => process.exit(code))\n  .catch((err) => {\n    // eslint-disable-next-line no-console\n    console.error('[agent-eval] cli error:', err)\n    process.exit(1)\n  })\n"],"mappings":";;;;;;;;;;;;;;;;AAaA,SAAS,qBAAqB;AAY9B,SAAS,UAAU,MAAsB;AACvC,QAAM,CAAC,SAAS,GAAG,IAAI,IAAI;AAC3B,QAAM,aAAuB,CAAC;AAC9B,QAAM,QAAgC,CAAC;AACvC,WAAS,IAAI,GAAG,IAAI,KAAK,QAAQ,KAAK;AACpC,UAAM,MAAM,KAAK,CAAC;AAClB,QAAI,IAAI,WAAW,IAAI,GAAG;AACxB,YAAM,MAAM,IAAI,MAAM,CAAC;AACvB,YAAM,OAAO,KAAK,IAAI,CAAC;AACvB,UAAI,QAAQ,QAAQ,CAAC,KAAK,WAAW,IAAI,GAAG;AAC1C,cAAM,GAAG,IAAI;AACb;AAAA,MACF,OAAO;AACL,cAAM,GAAG,IAAI;AAAA,MACf;AAAA,IACF,OAAO;AACL,iBAAW,KAAK,GAAG;AAAA,IACrB;AAAA,EACF;AACA,SAAO,EAAE,SAAS,WAAW,QAAQ,YAAY,MAAM;AACzD;AAEA,IAAM,OAAO;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAiBb,eAAe,OAAwB;AACrC,QAAM,EAAE,SAAS,YAAY,MAAM,IAAI,UAAU,QAAQ,KAAK,MAAM,CAAC,CAAC;AAEtE,UAAQ,SAAS;AAAA,IACf,KAAK,SAAS;AACZ,YAAM,OAAO,OAAO,MAAM,QAAQ,IAAI;AACtC,YAAM,OAAO,MAAM,QAAQ;AAC3B,YAAM,SAAS,YAAY,EAAE,MAAM,KAAK,CAAC;AAEzC,YAAM,WAAW,CAAC,QAAgB;AAEhC,gBAAQ,IAAI,yBAAyB,GAAG,iBAAiB;AACzD,eAAO,MAAM,MAAM,QAAQ,KAAK,CAAC,CAAC;AAElC,mBAAW,MAAM,QAAQ,KAAK,CAAC,GAAG,GAAI,EAAE,MAAM;AAAA,MAChD;AACA,cAAQ,GAAG,UAAU,MAAM,SAAS,QAAQ,CAAC;AAC7C,cAAQ,GAAG,WAAW,MAAM,SAAS,SAAS,CAAC;AAE/C,YAAM,IAAI,QAAQ,MAAM;AAAA,MAAC,CAAC;AAC1B,aAAO;AAAA,IACT;AAAA,IACA,KAAK,OAAO;AACV,YAAM,CAAC,MAAM,IAAI;AACjB,aAAO,MAAM,WAAW,MAAM;AAAA,IAChC;AAAA,IACA,KAAK,aAAa;AAChB,YAAM,CAAC,MAAM,IAAI;AACjB,aAAO,MAAM,YAAY,MAAM;AAAA,IACjC;AAAA,IACA,KAAK,WAAW;AACd,YAAM,MAAM,MAAM,OAAO;AACzB,YAAM,OAAO,aAAa,cAAc,EAAE,OAAO;AACjD,oBAAc,KAAK,GAAG,KAAK,UAAU,MAAM,MAAM,CAAC,CAAC;AAAA,GAAM,OAAO;AAEhE,cAAQ,IAAI,0CAA0C,GAAG,EAAE;AAC3D,aAAO;AAAA,IACT;AAAA,IACA,KAAK,WAAW;AACd,cAAQ,OAAO,MAAM,GAAG,KAAK,UAAU,cAAc,GAAG,MAAM,CAAC,CAAC;AAAA,CAAI;AACpE,aAAO;AAAA,IACT;AAAA,IACA,KAAK;AAAA,IACL,KAAK;AAAA,IACL,KAAK;AAAA,IACL,KAAK;AACH,cAAQ,OAAO,MAAM,GAAG,IAAI;AAAA,CAAI;AAChC,aAAO;AAAA,IACT;AACE,cAAQ,OAAO,MAAM,oBAAoB,OAAO;AAAA,EAAK,IAAI;AAAA,CAAI;AAC7D,aAAO;AAAA,EACX;AACF;AAEA,KAAK,EACF,KAAK,CAAC,SAAS,QAAQ,KAAK,IAAI,CAAC,EACjC,MAAM,CAAC,QAAQ;AAEd,UAAQ,MAAM,2BAA2B,GAAG;AAC5C,UAAQ,KAAK,CAAC;AAChB,CAAC;","names":[]}

package/dist/{code-agent-session-D-g04tcy.d.ts → code-agent-session-CjZsVd19.d.ts}

@@ -1,4 +1,4 @@
-import { b as RunSplitTag, c as RunCostProvenance, R as RunRecord } from './run-record-B7RTi_ix.js';
+import { a as RunSplitTag, b as RunCostProvenance, R as RunRecord } from './run-record-BDH49H2E.js';
 
 type CodeAgentSessionSource = 'codex' | 'claude-code' | 'opencode' | 'kimi-code' | 'pi';
 interface ParsedCodeAgentJsonl {

package/dist/contract/index.d.ts

@@ -1,36 +1,38 @@
-import { S as Scenario, M as MutableSurface, D as DispatchContext, b as JudgeConfig, f as SurfaceProposer, g as Gate, L as LabeledScenarioStore, C as CampaignResult, h as GateDecision } from '../types-C5gJrOVT.js';
-export { i as CampaignAggregates, j as CampaignArtifactWriter, k as CampaignCellResult, l as CampaignCostMeter, d as CampaignTraceWriter, m as CodeSurface, c as Dispatch, n as GateContext, G as GateResult, o as GenerationCandidate, e as GenerationRecord, a as JudgeDimension, J as JudgeScore, p as Mutator, O as OptimizationProposer, q as OptimizerConfig, r as SessionScript } from '../types-C5gJrOVT.js';
-import { L as LoopProvenanceRecord, P as PowerPreflight, R as RunEvalOptions } from '../provenance-Bibyg1U9.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, c as ParetoSignificanceGateOptions, d as PromotionObjective, e as PromotionPolicy, f as buildEvidenceVector, g as composeGate, h as defaultProductionGate, i as evolutionaryProposer, j as heldOutGate, p as paretoPolicy, k as paretoSignificanceGate, r as runEval } from '../provenance-Bibyg1U9.js';
-import { R as RunOptimizationOptions, a as RunImprovementLoopResult } from '../gepa-dne9JDPL.js';
-export { G as GepaProposerOptions, b as RunCampaignOptions, c as RunImprovementLoopOptions, g as gepaProposer, r as runCampaign, d as runImprovementLoop } from '../gepa-dne9JDPL.js';
-import { C as CampaignStorage } from '../storage-Dw_f7WMt.js';
-export { f as fsCampaignStorage, i as inMemoryCampaignStorage } from '../storage-Dw_f7WMt.js';
+import { S as Scenario, M as MutableSurface, D as DispatchContext, b as JudgeConfig, c as SurfaceProposer, G as Gate, L as LabeledScenarioStore, C as CampaignResult, d as GateDecision } from '../types-BSw1rOUB.js';
+export { e as CampaignAggregates, f as CampaignArtifactWriter, g as CampaignCellResult, h as CampaignCostMeter, i as CampaignTraceWriter, j as CodeSurface, k as Dispatch, l as GateContext, m as GateResult, n as GenerationCandidate, o as GenerationRecord, a as JudgeDimension, J as JudgeScore, p as Mutator, O as OptimizationProposer, q as OptimizerConfig, r as SessionScript } from '../types-BSw1rOUB.js';
+import { L as LoopProvenanceRecord, P as PowerPreflight, R as RunEvalOptions } from '../provenance-DpjwyseI.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, c as ParetoSignificanceGateOptions, d as PromotionObjective, e as PromotionPolicy, f as buildEvidenceVector, g as composeGate, h as defaultProductionGate, i as evolutionaryProposer, j as heldOutGate, p as paretoPolicy, k as paretoSignificanceGate, r as runEval } from '../provenance-DpjwyseI.js';
+import { R as RunOptimizationOptions, a as RunImprovementLoopResult } from '../gepa-eESocoDi.js';
+export { G as GepaProposerOptions, b as REFERENCE_EQUIVALENCE_INPUT_LIMITS, c as REFERENCE_EQUIVALENCE_JUDGE_VERSION, d as ReferenceEquivalenceJudgeInput, e as ReferenceEquivalenceJudgeOptions, f as ReferenceEquivalenceJudgeResult, g as ReferenceEquivalenceScenario, h as RunCampaignOptions, i as RunImprovementLoopOptions, j as createReferenceEquivalenceJudge, k as gepaProposer, r as runCampaign, l as runImprovementLoop, m as runReferenceEquivalenceJudge } from '../gepa-eESocoDi.js';
+export { c as AnalystFinding, C as ChatClient, p as CreateChatClientOpts, V as createChatClient } from '../policy-edit-wG9uFEFm.js';
+import { C as CampaignStorage } from '../storage-DrX3v_5B.js';
+export { f as fsCampaignStorage, i as inMemoryCampaignStorage } from '../storage-DrX3v_5B.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, b as RunSplitTag } from '../run-record-B7RTi_ix.js';
-import { I as InsightReport } from '../insight-report-IwwvqZZv.js';
-export { C as CostProvenanceSummary, 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-IwwvqZZv.js';
-export { D as DefaultAnalystRegistryOptions, c as buildDefaultAnalystRegistry } from '../default-registry-DltpYR5u.js';
-export { A as AnalystFinding } from '../kind-factory-DcNg13sZ.js';
-import { A as AnalyzeRunsOptions } from '../analyze-runs-BYHg6Irm.js';
-export { a as analyzeRuns } from '../analyze-runs-BYHg6Irm.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-D-g04tcy.js';
-import '../statistics-oUbOJe-S.js';
+import { b as CostLedgerSummary, d as CostReceipt, C as CostLedger } from '../cost-ledger-DWy3XdJc.js';
+import { R as RunRecord, a as RunSplitTag } from '../run-record-BDH49H2E.js';
+import { I as InsightReport } from '../insight-report-DY4nDW9Q.js';
+export { C as CostProvenanceSummary, 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-DY4nDW9Q.js';
+export { D as DefaultAnalystRegistryOptions, c as buildDefaultAnalystRegistry } from '../default-registry-DaK8b3fv.js';
+import { A as AnalyzeRunsOptions } from '../analyze-runs--2x39HZ7.js';
+export { a as analyzeRuns } from '../analyze-runs--2x39HZ7.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-CjZsVd19.js';
+import '../llm-client-qoDd18Qz.js';
+import '../errors-oeQrLqXC.js';
+import '../raw-provider-sink-C46HDghv.js';
+import '../statistics-KUnG73jH.js';
 import '../judge-calibration-7C-IDmKr.js';
-import '../types-C7DGg5ex.js';
+import '../types-BkfcQnxV.js';
 import '@tangle-network/tcloud';
 import '../dataset-NENEzRgk.js';
-import '../errors-oeQrLqXC.js';
-import '../store-BsVi7ncX.js';
-import '../schema-SGWcK9wa.js';
-import '../llm-client-DyqEH4jH.js';
-import '../raw-provider-sink-C46HDghv.js';
+import '../store-DGqD0Pyo.js';
+import '../schema-B3Q3l9Z_.js';
+import '../store-C1YxJDEK.js';
 import '@tangle-network/agent-interface';
-import '../summary-report-BJ5aNwZ1.js';
-import '../failure-cluster-C48PiReX.js';
+import '../summary-report-C5bKFfm-.js';
+import '../failure-cluster-DOAcSJ87.js';
 import '@ax-llm/ax';
-import '../store-C1YxJDEK.js';
+import '../kind-factory-ClZmO25A.js';
 import 'zod';
 
 /**
@@ -58,8 +60,8 @@ import 'zod';
  */
 
 interface SelfImproveBudget {
-    /** Hard $ ceiling across all cells in baseline + every generation. Cells
-     *  beyond the ceiling are skipped (cost-aware, not aborted). */
+    /** Hard spend cap across the full run. Each paid call reserves its enforced
+     *  maximum before dispatch, so completed spend cannot cross this amount. */
     dollars?: number;
     /** How many improvement generations to explore. Default 3. Set 0 to
      *  skip improvement entirely (selfImprove becomes a baseline-only run). */
@@ -75,7 +77,8 @@ interface SelfImproveBudget {
     holdoutScenarios?: Scenario[];
     /** Per-scenario replicates per cell — raises bootstrap-CI tightness. Default 1. */
     reps?: number;
-    /** Top-scoring candidates carried into the next generation. Default 2. */
+    /** @deprecated Must be 1 when supplied. The loop promotes only a candidate
+     *  that replaces its single global incumbent. */
     promoteTopK?: number;
 }
 interface SelfImproveLlm {
@@ -282,8 +285,13 @@ interface SelfImproveResult<TScenario extends Scenario, TArtifact> {
     generationsExplored: number;
     /** Wall-clock total. */
     durationMs: number;
-    /** Total cost across baseline + every generation. */
+    /** Total newly observed cost across the full run. */
     totalCostUsd: number;
+    /** Canonical run-wide spend summary. */
+    cost: CostLedgerSummary;
+    /** Run-wide receipts across proposal, search, holdout, judging, analysis,
+     *  and promotion work, with phase and actor attribution. */
+    receipts: CostReceipt[];
     /**
      * Rigor packet: distributional summary, paired-bootstrap lift CI,
      * judge stats, contamination check, recommendations. Wired through
@@ -303,6 +311,12 @@ interface SelfImproveResult<TScenario extends Scenario, TArtifact> {
      */
     raw: RunImprovementLoopResult<TArtifact, TScenario>;
 }
+/** Failed self-improvement run with an immutable receipt snapshot. */
+declare class SelfImproveRunError extends Error {
+    readonly cost: CostLedgerSummary;
+    readonly receipts: CostReceipt[];
+    constructor(cause: unknown, ledger: CostLedger);
+}
 /**
  * One-shot self-improvement loop. See module docstring for defaults +
  * extension points.
@@ -841,4 +855,4 @@ 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, 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 };
+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, SelfImproveRunError, SurfaceProposer, defineAgentEval, diffGenerations, diffRunBaselineToWinner, diffRuns, evalReportingSuite, fromFeedbackTable, fromOtelSpans, fromRunRecordDir, parseAgentTrace, partitionRunsByAuthoringModel, selfImprove };

package/dist/contract/index.js

@@ -12,10 +12,14 @@ import {
 } from "../chunk-ZZUXHH3R.js";
 import {
   analyzeRuns
-} from "../chunk-J6P6PK2R.js";
+} from "../chunk-FQNLDL4D.js";
 import {
+  REFERENCE_EQUIVALENCE_INPUT_LIMITS,
+  REFERENCE_EQUIVALENCE_JUDGE_VERSION,
   buildEvidenceVector,
+  campaignMeanComposite,
   composeGate,
+  createReferenceEquivalenceJudge,
   defaultProductionGate,
   emitLoopProvenance,
   evolutionaryProposer,
@@ -26,19 +30,23 @@ import {
   powerPreflight,
   runEval,
   runImprovementLoop,
+  runReferenceEquivalenceJudge,
   surfaceContentHash,
   surfaceHash
-} from "../chunk-ADYLPOSX.js";
-import "../chunk-VI2UW6B6.js";
+} from "../chunk-HQPHZGL6.js";
 import {
+  createRunCostLedger,
   fsCampaignStorage,
   inMemoryCampaignStorage,
+  resolveRunDir,
   runCampaign
-} from "../chunk-FAOEFFRT.js";
+} from "../chunk-IDZTTFRR.js";
 import {
-  buildDefaultAnalystRegistry
-} from "../chunk-I6LVHOV3.js";
-import "../chunk-5S5NJ63F.js";
+  buildDefaultAnalystRegistry,
+  createChatClient
+} from "../chunk-VF3XSYTI.js";
+import "../chunk-HHWE3POT.js";
+import "../chunk-MGEHEHSN.js";
 import {
   FileSystemOutcomeStore,
   InMemoryOutcomeStore
@@ -46,25 +54,38 @@ import {
 import "../chunk-ARU2PZFM.js";
 import "../chunk-DPZAEKA6.js";
 import "../chunk-PJQFMIOX.js";
-import "../chunk-MHNQWM4I.js";
+import "../chunk-LQUTGLOZ.js";
 import "../chunk-GGE4NNQT.js";
 import {
   LLM_INPUT_TOKEN_ATTR_KEYS,
   LLM_MODEL_ATTR_KEYS,
   LLM_OUTPUT_TOKEN_ATTR_KEYS
-} from "../chunk-LNQEP766.js";
+} from "../chunk-S2F4J57L.js";
 import {
   parseRunRecordSafe
 } from "../chunk-5UF54T55.js";
 import "../chunk-XJYR7XFV.js";
 import "../chunk-VSMTAMNK.js";
-import "../chunk-GY4SYVPJ.js";
+import "../chunk-NJC7U437.js";
+import "../chunk-VCTY3W6J.js";
+import "../chunk-VI2UW6B6.js";
 import "../chunk-PC4UYEBM.js";
 import "../chunk-ONWEPEDO.js";
 import "../chunk-PZ5AY32C.js";
 
 // src/contract/self-improve.ts
 import { createHash } from "crypto";
+var SelfImproveRunError = class extends Error {
+  cost;
+  receipts;
+  constructor(cause, ledger) {
+    const original = cause instanceof Error ? cause : new Error(String(cause));
+    super(original.message, { cause: original });
+    this.name = "SelfImproveRunError";
+    this.cost = ledger.summary();
+    this.receipts = ledger.list();
+  }
+};
 function splitTrainHoldout(scenarios, fraction) {
   function hash(s) {
     let h = 2166136261 >>> 0;
@@ -95,12 +116,26 @@ function meanComposite(byScenario) {
 }
 async function selfImprove(opts) {
   const startedAt = Date.now();
+  const requestedRunDir = opts.runDir ?? `mem://selfImprove-${startedAt}`;
+  const runDir = resolveRunDir(requestedRunDir);
+  const storage = opts.storage ?? (runDir.startsWith("mem://") ? inMemoryCampaignStorage() : fsCampaignStorage());
+  const costLedger = createRunCostLedger({
+    storage,
+    runDir,
+    costCeilingUsd: opts.budget?.dollars
+  });
+  try {
+    return await runSelfImprove(opts, costLedger, startedAt, runDir, storage);
+  } catch (error) {
+    throw new SelfImproveRunError(error, costLedger);
+  }
+}
+async function runSelfImprove(opts, costLedger, startedAt, runDir, storage) {
   const budget = opts.budget ?? {};
   const generations = budget.generations ?? 3;
   const populationSize = budget.populationSize ?? 2;
   const maxConcurrency = budget.maxConcurrency ?? 2;
   const holdoutFraction = budget.holdoutFraction ?? 0.25;
-  const costCeiling = budget.dollars;
   const expectUsage = opts.expectUsage ?? "assert";
   const explicitHoldout = budget.holdoutScenarios;
   const { train, holdout } = explicitHoldout ? {
@@ -130,9 +165,6 @@ async function selfImprove(opts) {
     holdoutScenarios: holdout,
     deltaThreshold: 0.05
   });
-  const runDir = opts.runDir ?? `mem://selfImprove-${startedAt}`;
-  const isMemRunDir = runDir.startsWith("mem://");
-  const storage = opts.storage ?? (isMemRunDir ? inMemoryCampaignStorage() : fsCampaignStorage());
   if (opts.onProgress) {
     opts.onProgress({ kind: "baseline.started", scenarios: opts.scenarios.length });
   }
@@ -156,7 +188,7 @@ async function selfImprove(opts) {
     runDir,
     maxConcurrency,
     cellPlacement: opts.cellPlacement,
-    costCeiling,
+    costLedger,
     expectUsage,
     labeledStore: opts.labeledStore,
     captureSource: opts.captureSource,
@@ -200,10 +232,8 @@ async function selfImprove(opts) {
       lift: winnerStats.compositeMean - baseline.compositeMean
     });
   }
-  const totalCost = result.baselineCampaign.aggregates.totalCostUsd + result.generations.reduce(
-    (sum, gen) => sum + gen.surfaces.reduce((s, sf) => s + sf.campaign.aggregates.totalCostUsd, 0),
-    0
-  );
+  const cost = result.cost;
+  const totalCost = cost.totalCostUsd;
   const insight = await analyzeRuns({
     runs: [
       ...cellsToRunRecords(result.baselineCampaign.cells, "baseline", runDir, opts.baselineSurface),
@@ -223,6 +253,7 @@ async function selfImprove(opts) {
     winnerLabel: result.winnerLabel,
     winnerRationale: result.winnerRationale,
     diff: result.promotedDiff,
+    baselineSearchComposite: campaignMeanComposite(result.baselineCampaign),
     generations: result.generations.map((g) => ({
       generationIndex: g.record.generationIndex,
       candidates: g.record.candidates,
@@ -254,6 +285,8 @@ async function selfImprove(opts) {
     generationsExplored: result.generations.length,
     durationMs,
     totalCostUsd: totalCost,
+    cost,
+    receipts: costLedger.list(),
     insight,
     ...power ? { power } : {},
     raw: result
@@ -986,10 +1019,15 @@ function collectNumericAttrs(spans) {
 export {
   FileSystemOutcomeStore,
   InMemoryOutcomeStore,
+  REFERENCE_EQUIVALENCE_INPUT_LIMITS,
+  REFERENCE_EQUIVALENCE_JUDGE_VERSION,
+  SelfImproveRunError,
   analyzeRuns,
   buildDefaultAnalystRegistry,
   buildEvidenceVector,
   composeGate,
+  createChatClient,
+  createReferenceEquivalenceJudge,
   defaultProductionGate,
   defineAgentEval,
   diffGenerations,
@@ -1018,6 +1056,7 @@ export {
   runCampaign,
   runEval,
   runImprovementLoop,
+  runReferenceEquivalenceJudge,
   selfImprove
 };
 //# sourceMappingURL=index.js.map