@tangle-network/agent-eval 0.137.0 → 0.138.0

package/dist/cli.js

@@ -1,6 +1,7 @@
 #!/usr/bin/env node
 import { o as runRolloutReleaseCli } from "./hf-dataset-DBJXXoY1.js";
-import { a as runRpcBatch, o as runRpcOnce, p as handleVersion, r as startServerAsync, s as buildOpenApi } from "./server-KjXZZUDX.js";
+import { n as runAnalystBenchmarkCommand } from "./benchmark-command-CMqVqReF.js";
+import { a as runRpcBatch, o as runRpcOnce, p as handleVersion, r as startServerAsync, s as buildOpenApi } from "./server-DLEvyW2z.js";
 import { writeFileSync } from "node:fs";
 //#region src/cli-config.ts
 function resolveCliLlmConfig(env = process.env) {
@@ -32,6 +33,7 @@ function nonEmpty(value) {
 *   agent-eval serve [--port 5005] [--host 127.0.0.1]
 *   agent-eval rpc <method>          # one request from stdin → one response on stdout
 *   agent-eval rpc-batch <method>    # JSONL stdin → JSONL stdout
+*   agent-eval analyst-benchmark ... # benchmark a real-model trace analyst
 *   agent-eval openapi [--out path]  # write OpenAPI spec
 *   agent-eval version
 *
@@ -85,6 +87,9 @@ Commands:
   rollout-release <ledger.jsonl...> --out <dir> [--formats sft,verifiers,rft,raw] [--include-proposers] [--push <org/name>]
         Build a HuggingFace-ready dataset dir from tangle.rollout.v1 ledgers:
         validate, fail-closed split filter, scrub, export formats + card.
+  analyst-benchmark --dataset <agentrx|codetracebench> --labels <path> --trace-dir <path> [--artifact-dir <path>] --out <dir> ...
+        Compare an empty baseline with a real-model trace analyst on public labels.
+        Run with --help for all required inputs and controls.
   version
         Print server + wire-protocol version JSON.
 
@@ -96,6 +101,7 @@ Without arguments, prints this help.`;
 async function main() {
 	const { command, positional, flags } = parseArgs(process.argv.slice(2));
 	assertKnownFlags(command, flags);
+	if (command === "analyst-benchmark" && flags.help === "true") return await runAnalystBenchmarkCommand(process.argv.slice(3));
 	if (flags.help === "true") {
 		process.stdout.write(`${HELP}\n`);
 		return 0;
@@ -148,6 +154,7 @@ async function main() {
 			return 0;
 		}
 		case "rollout-release": return await runRolloutReleaseCli(process.argv.slice(3));
+		case "analyst-benchmark": return await runAnalystBenchmarkCommand(process.argv.slice(3));
 		case "version":
 			process.stdout.write(`${JSON.stringify(handleVersion(), null, 2)}\n`);
 			return 0;
@@ -182,7 +189,7 @@ const FLAGS_BY_COMMAND = {
 	"": /* @__PURE__ */ new Set()
 };
 function assertKnownFlags(command, flags) {
-	if (command === "rollout-release") return;
+	if (command === "rollout-release" || command === "analyst-benchmark") return;
 	const allowed = FLAGS_BY_COMMAND[command];
 	if (!allowed) return;
 	const unknown = Object.keys(flags).filter((flag) => !allowed.has(flag));

package/dist/cli.js.map

@@ -1 +1 @@
-{"version":3,"file":"cli.js","names":[],"sources":["../src/cli-config.ts","../src/cli.ts"],"sourcesContent":["import type { LlmClientOptions } from './llm-client'\n\nexport interface CliLlmConfig {\n  client?: LlmClientOptions\n  model?: string\n}\n\nexport function resolveCliLlmConfig(env: NodeJS.ProcessEnv = process.env): CliLlmConfig {\n  const explicitBaseUrl = nonEmpty(env.AGENT_EVAL_LLM_BASE_URL)\n  const explicitApiKey = nonEmpty(env.AGENT_EVAL_LLM_API_KEY)\n  const openAiApiKey = nonEmpty(env.OPENAI_API_KEY)\n  const tangleApiKey = nonEmpty(env.TANGLE_API_KEY)\n  const baseUrl =\n    explicitBaseUrl ??\n    nonEmpty(env.OPENAI_BASE_URL) ??\n    nonEmpty(env.TANGLE_ROUTER_URL) ??\n    (openAiApiKey ? 'https://api.openai.com/v1' : undefined) ??\n    (tangleApiKey ? 'https://router.tangle.tools/v1' : undefined)\n  const apiKey = explicitApiKey ?? openAiApiKey ?? tangleApiKey\n  const model =\n    nonEmpty(env.AGENT_EVAL_LLM_MODEL) ?? nonEmpty(env.OPENAI_MODEL) ?? nonEmpty(env.TANGLE_MODEL)\n\n  const client =\n    baseUrl || apiKey\n      ? { ...(baseUrl ? { baseUrl } : {}), ...(apiKey ? { apiKey } : {}) }\n      : undefined\n  return { ...(client ? { client } : {}), ...(model ? { model } : {}) }\n}\n\nfunction nonEmpty(value: string | undefined): string | undefined {\n  const trimmed = value?.trim()\n  return trimmed ? trimmed : undefined\n}\n","#!/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 { resolveCliLlmConfig } from './cli-config'\nimport { runRolloutReleaseCli } from './rollout/release/hf-dataset'\nimport { handleVersion } from './wire/handlers'\nimport { buildOpenApi } from './wire/openapi'\nimport { runRpcBatch, runRpcOnce } from './wire/rpc'\nimport { startServerAsync } 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 raw = tok.slice(2)\n      const equalsAt = raw.indexOf('=')\n      if (equalsAt >= 0) {\n        flags[raw.slice(0, equalsAt)] = raw.slice(equalsAt + 1)\n        continue\n      }\n      const key = raw\n      if (key === 'help') {\n        flags[key] = 'true'\n        continue\n      }\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 if (tok === '-h') {\n      flags.help = 'true'\n    } else {\n      positional.push(tok)\n    }\n  }\n  return { command: command ?? 'help', positional, flags }\n}\n\nconst HELP = `agent-eval: evaluation RPC and HTTP server.\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. Methods: 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  rollout-release <ledger.jsonl...> --out <dir> [--formats sft,verifiers,rft,raw] [--include-proposers] [--push <org/name>]\n        Build a HuggingFace-ready dataset dir from tangle.rollout.v1 ledgers:\n        validate, fail-closed split filter, scrub, export formats + card.\n  version\n        Print server + wire-protocol version JSON.\n\nJudge provider:\n  Set AGENT_EVAL_LLM_BASE_URL, AGENT_EVAL_LLM_API_KEY, and AGENT_EVAL_LLM_MODEL.\n  OPENAI_* and TANGLE_* equivalents are also accepted.\n\nWithout arguments, prints this help.`\n\nasync function main(): Promise<number> {\n  const { command, positional, flags } = parseArgs(process.argv.slice(2))\n  assertKnownFlags(command, flags)\n\n  if (flags.help === 'true') {\n    process.stdout.write(`${HELP}\\n`)\n    return 0\n  }\n\n  switch (command) {\n    case 'serve': {\n      const port = parsePort(flags.port ?? '5005')\n      const host = flags.host ?? '127.0.0.1'\n      const llm = resolveCliLlmConfig()\n      const { server } = await startServerAsync({\n        port,\n        host,\n        llm: llm.client,\n        judgeModel: llm.model,\n        llmRouteRequirements: { requireExplicitBaseUrl: true },\n      })\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 ${sig}, shutting down`)\n        server.close(() => process.exit(0))\n        // Force exit after 5s if close hangs\n        setTimeout(() => process.exit(1), 5000).unref()\n      }\n      process.on('SIGINT', () => shutdown('SIGINT'))\n      process.on('SIGTERM', () => shutdown('SIGTERM'))\n      // Block forever\n      await new Promise(() => {})\n      return 0\n    }\n    case 'rpc': {\n      const [method] = positional\n      const llm = resolveCliLlmConfig()\n      return await runRpcOnce(method, {\n        llm: llm.client,\n        judgeModel: llm.model,\n        llmRouteRequirements: { requireExplicitBaseUrl: true },\n      })\n    }\n    case 'rpc-batch': {\n      const [method] = positional\n      const llm = resolveCliLlmConfig()\n      return await runRpcBatch(method, {\n        llm: llm.client,\n        judgeModel: llm.model,\n        llmRouteRequirements: { requireExplicitBaseUrl: true },\n      })\n    }\n    case 'openapi': {\n      const out = flags.out ?? 'openapi.json'\n      const spec = buildOpenApi(handleVersion().version)\n      writeFileSync(out, `${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 ${out}`)\n      return 0\n    }\n    case 'rollout-release': {\n      // The subcommand owns its own flag grammar (multi-value --formats,\n      // boolean --include-proposers) — pass raw argv through untouched.\n      return await runRolloutReleaseCli(process.argv.slice(3))\n    }\n    case 'version': {\n      process.stdout.write(`${JSON.stringify(handleVersion(), null, 2)}\\n`)\n      return 0\n    }\n    case '--version': {\n      process.stdout.write(`${handleVersion().version}\\n`)\n      return 0\n    }\n    case 'help':\n    case '--help':\n    case '-h':\n    case '':\n      process.stdout.write(`${HELP}\\n`)\n      return 0\n    default:\n      process.stderr.write(`unknown command: ${command}\\n${HELP}\\n`)\n      return 1\n  }\n}\n\nconst FLAGS_BY_COMMAND: Record<string, ReadonlySet<string>> = {\n  serve: new Set(['help', 'host', 'port']),\n  rpc: new Set(['help']),\n  'rpc-batch': new Set(['help']),\n  openapi: new Set(['help', 'out']),\n  version: new Set(['help']),\n  help: new Set(),\n  '--help': new Set(),\n  '-h': new Set(),\n  '--version': new Set(),\n  '': new Set(),\n}\n\nfunction assertKnownFlags(command: string, flags: Record<string, string>): void {\n  // rollout-release parses its own argv (multi-value flags).\n  if (command === 'rollout-release') return\n  const allowed = FLAGS_BY_COMMAND[command]\n  if (!allowed) return\n  const unknown = Object.keys(flags).filter((flag) => !allowed.has(flag))\n  if (unknown.length > 0) {\n    throw new Error(`unknown flag for ${command || 'help'}: --${unknown[0]}`)\n  }\n}\n\nfunction parsePort(raw: string): number {\n  const port = Number(raw)\n  if (!Number.isInteger(port) || port < 0 || port > 65_535) {\n    throw new Error(`--port must be an integer from 0 to 65535; received ${JSON.stringify(raw)}`)\n  }\n  return port\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":";;;;;AAOA,SAAgB,oBAAoB,MAAyB,QAAQ,KAAmB;CACtF,MAAM,kBAAkB,SAAS,IAAI,uBAAuB;CAC5D,MAAM,iBAAiB,SAAS,IAAI,sBAAsB;CAC1D,MAAM,eAAe,SAAS,IAAI,cAAc;CAChD,MAAM,eAAe,SAAS,IAAI,cAAc;CAChD,MAAM,UACJ,mBACA,SAAS,IAAI,eAAe,KAC5B,SAAS,IAAI,iBAAiB,MAC7B,eAAe,8BAA8B,KAAA,OAC7C,eAAe,mCAAmC,KAAA;CACrD,MAAM,SAAS,kBAAkB,gBAAgB;CACjD,MAAM,QACJ,SAAS,IAAI,oBAAoB,KAAK,SAAS,IAAI,YAAY,KAAK,SAAS,IAAI,YAAY;CAE/F,MAAM,SACJ,WAAW,SACP;EAAE,GAAI,UAAU,EAAE,QAAQ,IAAI,CAAC;EAAI,GAAI,SAAS,EAAE,OAAO,IAAI,CAAC;CAAG,IACjE,KAAA;CACN,OAAO;EAAE,GAAI,SAAS,EAAE,OAAO,IAAI,CAAC;EAAI,GAAI,QAAQ,EAAE,MAAM,IAAI,CAAC;CAAG;AACtE;AAEA,SAAS,SAAS,OAA+C;CAC/D,MAAM,UAAU,OAAO,KAAK;CAC5B,OAAO,UAAU,UAAU,KAAA;AAC7B;;;;;;;;;;;;;;;ACLA,SAAS,UAAU,MAAsB;CACvC,MAAM,CAAC,SAAS,GAAG,QAAQ;CAC3B,MAAM,aAAuB,CAAC;CAC9B,MAAM,QAAgC,CAAC;CACvC,KAAK,IAAI,IAAI,GAAG,IAAI,KAAK,QAAQ,KAAK;EACpC,MAAM,MAAM,KAAK;EACjB,IAAI,IAAI,WAAW,IAAI,GAAG;GACxB,MAAM,MAAM,IAAI,MAAM,CAAC;GACvB,MAAM,WAAW,IAAI,QAAQ,GAAG;GAChC,IAAI,YAAY,GAAG;IACjB,MAAM,IAAI,MAAM,GAAG,QAAQ,KAAK,IAAI,MAAM,WAAW,CAAC;IACtD;GACF;GACA,MAAM,MAAM;GACZ,IAAI,QAAQ,QAAQ;IAClB,MAAM,OAAO;IACb;GACF;GACA,MAAM,OAAO,KAAK,IAAI;GACtB,IAAI,QAAQ,QAAQ,CAAC,KAAK,WAAW,IAAI,GAAG;IAC1C,MAAM,OAAO;IACb;GACF,OACE,MAAM,OAAO;EAEjB,OAAO,IAAI,QAAQ,MACjB,MAAM,OAAO;OAEb,WAAW,KAAK,GAAG;CAEvB;CACA,OAAO;EAAE,SAAS,WAAW;EAAQ;EAAY;CAAM;AACzD;AAEA,MAAM,OAAO;;;;;;;;;;;;;;;;;;;;;;;AAwBb,eAAe,OAAwB;CACrC,MAAM,EAAE,SAAS,YAAY,UAAU,UAAU,QAAQ,KAAK,MAAM,CAAC,CAAC;CACtE,iBAAiB,SAAS,KAAK;CAE/B,IAAI,MAAM,SAAS,QAAQ;EACzB,QAAQ,OAAO,MAAM,GAAG,KAAK,GAAG;EAChC,OAAO;CACT;CAEA,QAAQ,SAAR;EACE,KAAK,SAAS;GACZ,MAAM,OAAO,UAAU,MAAM,QAAQ,MAAM;GAC3C,MAAM,OAAO,MAAM,QAAQ;GAC3B,MAAM,MAAM,oBAAoB;GAChC,MAAM,EAAE,WAAW,MAAM,iBAAiB;IACxC;IACA;IACA,KAAK,IAAI;IACT,YAAY,IAAI;IAChB,sBAAsB,EAAE,wBAAwB,KAAK;GACvD,CAAC;GAED,MAAM,YAAY,QAAgB;IAEhC,QAAQ,IAAI,yBAAyB,IAAI,gBAAgB;IACzD,OAAO,YAAY,QAAQ,KAAK,CAAC,CAAC;IAElC,iBAAiB,QAAQ,KAAK,CAAC,GAAG,GAAI,CAAC,CAAC,MAAM;GAChD;GACA,QAAQ,GAAG,gBAAgB,SAAS,QAAQ,CAAC;GAC7C,QAAQ,GAAG,iBAAiB,SAAS,SAAS,CAAC;GAE/C,MAAM,IAAI,cAAc,CAAC,CAAC;GAC1B,OAAO;EACT;EACA,KAAK,OAAO;GACV,MAAM,CAAC,UAAU;GACjB,MAAM,MAAM,oBAAoB;GAChC,OAAO,MAAM,WAAW,QAAQ;IAC9B,KAAK,IAAI;IACT,YAAY,IAAI;IAChB,sBAAsB,EAAE,wBAAwB,KAAK;GACvD,CAAC;EACH;EACA,KAAK,aAAa;GAChB,MAAM,CAAC,UAAU;GACjB,MAAM,MAAM,oBAAoB;GAChC,OAAO,MAAM,YAAY,QAAQ;IAC/B,KAAK,IAAI;IACT,YAAY,IAAI;IAChB,sBAAsB,EAAE,wBAAwB,KAAK;GACvD,CAAC;EACH;EACA,KAAK,WAAW;GACd,MAAM,MAAM,MAAM,OAAO;GACzB,MAAM,OAAO,aAAa,cAAc,CAAC,CAAC,OAAO;GACjD,cAAc,KAAK,GAAG,KAAK,UAAU,MAAM,MAAM,CAAC,EAAE,KAAK,OAAO;GAEhE,QAAQ,IAAI,0CAA0C,KAAK;GAC3D,OAAO;EACT;EACA,KAAK,mBAGH,OAAO,MAAM,qBAAqB,QAAQ,KAAK,MAAM,CAAC,CAAC;EAEzD,KAAK;GACH,QAAQ,OAAO,MAAM,GAAG,KAAK,UAAU,cAAc,GAAG,MAAM,CAAC,EAAE,GAAG;GACpE,OAAO;EAET,KAAK;GACH,QAAQ,OAAO,MAAM,GAAG,cAAc,CAAC,CAAC,QAAQ,GAAG;GACnD,OAAO;EAET,KAAK;EACL,KAAK;EACL,KAAK;EACL,KAAK;GACH,QAAQ,OAAO,MAAM,GAAG,KAAK,GAAG;GAChC,OAAO;EACT;GACE,QAAQ,OAAO,MAAM,oBAAoB,QAAQ,IAAI,KAAK,GAAG;GAC7D,OAAO;CACX;AACF;AAEA,MAAM,mBAAwD;CAC5D,uBAAO,IAAI,IAAI;EAAC;EAAQ;EAAQ;CAAM,CAAC;CACvC,qBAAK,IAAI,IAAI,CAAC,MAAM,CAAC;CACrB,6BAAa,IAAI,IAAI,CAAC,MAAM,CAAC;CAC7B,yBAAS,IAAI,IAAI,CAAC,QAAQ,KAAK,CAAC;CAChC,yBAAS,IAAI,IAAI,CAAC,MAAM,CAAC;CACzB,sBAAM,IAAI,IAAI;CACd,0BAAU,IAAI,IAAI;CAClB,sBAAM,IAAI,IAAI;CACd,6BAAa,IAAI,IAAI;CACrB,oBAAI,IAAI,IAAI;AACd;AAEA,SAAS,iBAAiB,SAAiB,OAAqC;CAE9E,IAAI,YAAY,mBAAmB;CACnC,MAAM,UAAU,iBAAiB;CACjC,IAAI,CAAC,SAAS;CACd,MAAM,UAAU,OAAO,KAAK,KAAK,CAAC,CAAC,QAAQ,SAAS,CAAC,QAAQ,IAAI,IAAI,CAAC;CACtE,IAAI,QAAQ,SAAS,GACnB,MAAM,IAAI,MAAM,oBAAoB,WAAW,OAAO,MAAM,QAAQ,IAAI;AAE5E;AAEA,SAAS,UAAU,KAAqB;CACtC,MAAM,OAAO,OAAO,GAAG;CACvB,IAAI,CAAC,OAAO,UAAU,IAAI,KAAK,OAAO,KAAK,OAAO,OAChD,MAAM,IAAI,MAAM,uDAAuD,KAAK,UAAU,GAAG,GAAG;CAE9F,OAAO;AACT;AAEA,KAAK,CAAC,CACH,MAAM,SAAS,QAAQ,KAAK,IAAI,CAAC,CAAC,CAClC,OAAO,QAAQ;CAEd,QAAQ,MAAM,2BAA2B,GAAG;CAC5C,QAAQ,KAAK,CAAC;AAChB,CAAC"}
+{"version":3,"file":"cli.js","names":[],"sources":["../src/cli-config.ts","../src/cli.ts"],"sourcesContent":["import type { LlmClientOptions } from './llm-client'\n\nexport interface CliLlmConfig {\n  client?: LlmClientOptions\n  model?: string\n}\n\nexport function resolveCliLlmConfig(env: NodeJS.ProcessEnv = process.env): CliLlmConfig {\n  const explicitBaseUrl = nonEmpty(env.AGENT_EVAL_LLM_BASE_URL)\n  const explicitApiKey = nonEmpty(env.AGENT_EVAL_LLM_API_KEY)\n  const openAiApiKey = nonEmpty(env.OPENAI_API_KEY)\n  const tangleApiKey = nonEmpty(env.TANGLE_API_KEY)\n  const baseUrl =\n    explicitBaseUrl ??\n    nonEmpty(env.OPENAI_BASE_URL) ??\n    nonEmpty(env.TANGLE_ROUTER_URL) ??\n    (openAiApiKey ? 'https://api.openai.com/v1' : undefined) ??\n    (tangleApiKey ? 'https://router.tangle.tools/v1' : undefined)\n  const apiKey = explicitApiKey ?? openAiApiKey ?? tangleApiKey\n  const model =\n    nonEmpty(env.AGENT_EVAL_LLM_MODEL) ?? nonEmpty(env.OPENAI_MODEL) ?? nonEmpty(env.TANGLE_MODEL)\n\n  const client =\n    baseUrl || apiKey\n      ? { ...(baseUrl ? { baseUrl } : {}), ...(apiKey ? { apiKey } : {}) }\n      : undefined\n  return { ...(client ? { client } : {}), ...(model ? { model } : {}) }\n}\n\nfunction nonEmpty(value: string | undefined): string | undefined {\n  const trimmed = value?.trim()\n  return trimmed ? trimmed : undefined\n}\n","#!/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 analyst-benchmark ... # benchmark a real-model trace analyst\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 { runAnalystBenchmarkCommand } from './analyst/benchmark-command'\nimport { resolveCliLlmConfig } from './cli-config'\nimport { runRolloutReleaseCli } from './rollout/release/hf-dataset'\nimport { handleVersion } from './wire/handlers'\nimport { buildOpenApi } from './wire/openapi'\nimport { runRpcBatch, runRpcOnce } from './wire/rpc'\nimport { startServerAsync } 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 raw = tok.slice(2)\n      const equalsAt = raw.indexOf('=')\n      if (equalsAt >= 0) {\n        flags[raw.slice(0, equalsAt)] = raw.slice(equalsAt + 1)\n        continue\n      }\n      const key = raw\n      if (key === 'help') {\n        flags[key] = 'true'\n        continue\n      }\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 if (tok === '-h') {\n      flags.help = 'true'\n    } else {\n      positional.push(tok)\n    }\n  }\n  return { command: command ?? 'help', positional, flags }\n}\n\nconst HELP = `agent-eval: evaluation RPC and HTTP server.\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. Methods: 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  rollout-release <ledger.jsonl...> --out <dir> [--formats sft,verifiers,rft,raw] [--include-proposers] [--push <org/name>]\n        Build a HuggingFace-ready dataset dir from tangle.rollout.v1 ledgers:\n        validate, fail-closed split filter, scrub, export formats + card.\n  analyst-benchmark --dataset <agentrx|codetracebench> --labels <path> --trace-dir <path> [--artifact-dir <path>] --out <dir> ...\n        Compare an empty baseline with a real-model trace analyst on public labels.\n        Run with --help for all required inputs and controls.\n  version\n        Print server + wire-protocol version JSON.\n\nJudge provider:\n  Set AGENT_EVAL_LLM_BASE_URL, AGENT_EVAL_LLM_API_KEY, and AGENT_EVAL_LLM_MODEL.\n  OPENAI_* and TANGLE_* equivalents are also accepted.\n\nWithout arguments, prints this help.`\n\nasync function main(): Promise<number> {\n  const { command, positional, flags } = parseArgs(process.argv.slice(2))\n  assertKnownFlags(command, flags)\n\n  if (command === 'analyst-benchmark' && flags.help === 'true') {\n    return await runAnalystBenchmarkCommand(process.argv.slice(3))\n  }\n  if (flags.help === 'true') {\n    process.stdout.write(`${HELP}\\n`)\n    return 0\n  }\n\n  switch (command) {\n    case 'serve': {\n      const port = parsePort(flags.port ?? '5005')\n      const host = flags.host ?? '127.0.0.1'\n      const llm = resolveCliLlmConfig()\n      const { server } = await startServerAsync({\n        port,\n        host,\n        llm: llm.client,\n        judgeModel: llm.model,\n        llmRouteRequirements: { requireExplicitBaseUrl: true },\n      })\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 ${sig}, shutting down`)\n        server.close(() => process.exit(0))\n        // Force exit after 5s if close hangs\n        setTimeout(() => process.exit(1), 5000).unref()\n      }\n      process.on('SIGINT', () => shutdown('SIGINT'))\n      process.on('SIGTERM', () => shutdown('SIGTERM'))\n      // Block forever\n      await new Promise(() => {})\n      return 0\n    }\n    case 'rpc': {\n      const [method] = positional\n      const llm = resolveCliLlmConfig()\n      return await runRpcOnce(method, {\n        llm: llm.client,\n        judgeModel: llm.model,\n        llmRouteRequirements: { requireExplicitBaseUrl: true },\n      })\n    }\n    case 'rpc-batch': {\n      const [method] = positional\n      const llm = resolveCliLlmConfig()\n      return await runRpcBatch(method, {\n        llm: llm.client,\n        judgeModel: llm.model,\n        llmRouteRequirements: { requireExplicitBaseUrl: true },\n      })\n    }\n    case 'openapi': {\n      const out = flags.out ?? 'openapi.json'\n      const spec = buildOpenApi(handleVersion().version)\n      writeFileSync(out, `${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 ${out}`)\n      return 0\n    }\n    case 'rollout-release': {\n      // The subcommand owns its own flag grammar (multi-value --formats,\n      // boolean --include-proposers) — pass raw argv through untouched.\n      return await runRolloutReleaseCli(process.argv.slice(3))\n    }\n    case 'analyst-benchmark': {\n      return await runAnalystBenchmarkCommand(process.argv.slice(3))\n    }\n    case 'version': {\n      process.stdout.write(`${JSON.stringify(handleVersion(), null, 2)}\\n`)\n      return 0\n    }\n    case '--version': {\n      process.stdout.write(`${handleVersion().version}\\n`)\n      return 0\n    }\n    case 'help':\n    case '--help':\n    case '-h':\n    case '':\n      process.stdout.write(`${HELP}\\n`)\n      return 0\n    default:\n      process.stderr.write(`unknown command: ${command}\\n${HELP}\\n`)\n      return 1\n  }\n}\n\nconst FLAGS_BY_COMMAND: Record<string, ReadonlySet<string>> = {\n  serve: new Set(['help', 'host', 'port']),\n  rpc: new Set(['help']),\n  'rpc-batch': new Set(['help']),\n  openapi: new Set(['help', 'out']),\n  version: new Set(['help']),\n  help: new Set(),\n  '--help': new Set(),\n  '-h': new Set(),\n  '--version': new Set(),\n  '': new Set(),\n}\n\nfunction assertKnownFlags(command: string, flags: Record<string, string>): void {\n  // These subcommands parse their own argv.\n  if (command === 'rollout-release' || command === 'analyst-benchmark') return\n  const allowed = FLAGS_BY_COMMAND[command]\n  if (!allowed) return\n  const unknown = Object.keys(flags).filter((flag) => !allowed.has(flag))\n  if (unknown.length > 0) {\n    throw new Error(`unknown flag for ${command || 'help'}: --${unknown[0]}`)\n  }\n}\n\nfunction parsePort(raw: string): number {\n  const port = Number(raw)\n  if (!Number.isInteger(port) || port < 0 || port > 65_535) {\n    throw new Error(`--port must be an integer from 0 to 65535; received ${JSON.stringify(raw)}`)\n  }\n  return port\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":";;;;;;AAOA,SAAgB,oBAAoB,MAAyB,QAAQ,KAAmB;CACtF,MAAM,kBAAkB,SAAS,IAAI,uBAAuB;CAC5D,MAAM,iBAAiB,SAAS,IAAI,sBAAsB;CAC1D,MAAM,eAAe,SAAS,IAAI,cAAc;CAChD,MAAM,eAAe,SAAS,IAAI,cAAc;CAChD,MAAM,UACJ,mBACA,SAAS,IAAI,eAAe,KAC5B,SAAS,IAAI,iBAAiB,MAC7B,eAAe,8BAA8B,KAAA,OAC7C,eAAe,mCAAmC,KAAA;CACrD,MAAM,SAAS,kBAAkB,gBAAgB;CACjD,MAAM,QACJ,SAAS,IAAI,oBAAoB,KAAK,SAAS,IAAI,YAAY,KAAK,SAAS,IAAI,YAAY;CAE/F,MAAM,SACJ,WAAW,SACP;EAAE,GAAI,UAAU,EAAE,QAAQ,IAAI,CAAC;EAAI,GAAI,SAAS,EAAE,OAAO,IAAI,CAAC;CAAG,IACjE,KAAA;CACN,OAAO;EAAE,GAAI,SAAS,EAAE,OAAO,IAAI,CAAC;EAAI,GAAI,QAAQ,EAAE,MAAM,IAAI,CAAC;CAAG;AACtE;AAEA,SAAS,SAAS,OAA+C;CAC/D,MAAM,UAAU,OAAO,KAAK;CAC5B,OAAO,UAAU,UAAU,KAAA;AAC7B;;;;;;;;;;;;;;;;ACHA,SAAS,UAAU,MAAsB;CACvC,MAAM,CAAC,SAAS,GAAG,QAAQ;CAC3B,MAAM,aAAuB,CAAC;CAC9B,MAAM,QAAgC,CAAC;CACvC,KAAK,IAAI,IAAI,GAAG,IAAI,KAAK,QAAQ,KAAK;EACpC,MAAM,MAAM,KAAK;EACjB,IAAI,IAAI,WAAW,IAAI,GAAG;GACxB,MAAM,MAAM,IAAI,MAAM,CAAC;GACvB,MAAM,WAAW,IAAI,QAAQ,GAAG;GAChC,IAAI,YAAY,GAAG;IACjB,MAAM,IAAI,MAAM,GAAG,QAAQ,KAAK,IAAI,MAAM,WAAW,CAAC;IACtD;GACF;GACA,MAAM,MAAM;GACZ,IAAI,QAAQ,QAAQ;IAClB,MAAM,OAAO;IACb;GACF;GACA,MAAM,OAAO,KAAK,IAAI;GACtB,IAAI,QAAQ,QAAQ,CAAC,KAAK,WAAW,IAAI,GAAG;IAC1C,MAAM,OAAO;IACb;GACF,OACE,MAAM,OAAO;EAEjB,OAAO,IAAI,QAAQ,MACjB,MAAM,OAAO;OAEb,WAAW,KAAK,GAAG;CAEvB;CACA,OAAO;EAAE,SAAS,WAAW;EAAQ;EAAY;CAAM;AACzD;AAEA,MAAM,OAAO;;;;;;;;;;;;;;;;;;;;;;;;;;AA2Bb,eAAe,OAAwB;CACrC,MAAM,EAAE,SAAS,YAAY,UAAU,UAAU,QAAQ,KAAK,MAAM,CAAC,CAAC;CACtE,iBAAiB,SAAS,KAAK;CAE/B,IAAI,YAAY,uBAAuB,MAAM,SAAS,QACpD,OAAO,MAAM,2BAA2B,QAAQ,KAAK,MAAM,CAAC,CAAC;CAE/D,IAAI,MAAM,SAAS,QAAQ;EACzB,QAAQ,OAAO,MAAM,GAAG,KAAK,GAAG;EAChC,OAAO;CACT;CAEA,QAAQ,SAAR;EACE,KAAK,SAAS;GACZ,MAAM,OAAO,UAAU,MAAM,QAAQ,MAAM;GAC3C,MAAM,OAAO,MAAM,QAAQ;GAC3B,MAAM,MAAM,oBAAoB;GAChC,MAAM,EAAE,WAAW,MAAM,iBAAiB;IACxC;IACA;IACA,KAAK,IAAI;IACT,YAAY,IAAI;IAChB,sBAAsB,EAAE,wBAAwB,KAAK;GACvD,CAAC;GAED,MAAM,YAAY,QAAgB;IAEhC,QAAQ,IAAI,yBAAyB,IAAI,gBAAgB;IACzD,OAAO,YAAY,QAAQ,KAAK,CAAC,CAAC;IAElC,iBAAiB,QAAQ,KAAK,CAAC,GAAG,GAAI,CAAC,CAAC,MAAM;GAChD;GACA,QAAQ,GAAG,gBAAgB,SAAS,QAAQ,CAAC;GAC7C,QAAQ,GAAG,iBAAiB,SAAS,SAAS,CAAC;GAE/C,MAAM,IAAI,cAAc,CAAC,CAAC;GAC1B,OAAO;EACT;EACA,KAAK,OAAO;GACV,MAAM,CAAC,UAAU;GACjB,MAAM,MAAM,oBAAoB;GAChC,OAAO,MAAM,WAAW,QAAQ;IAC9B,KAAK,IAAI;IACT,YAAY,IAAI;IAChB,sBAAsB,EAAE,wBAAwB,KAAK;GACvD,CAAC;EACH;EACA,KAAK,aAAa;GAChB,MAAM,CAAC,UAAU;GACjB,MAAM,MAAM,oBAAoB;GAChC,OAAO,MAAM,YAAY,QAAQ;IAC/B,KAAK,IAAI;IACT,YAAY,IAAI;IAChB,sBAAsB,EAAE,wBAAwB,KAAK;GACvD,CAAC;EACH;EACA,KAAK,WAAW;GACd,MAAM,MAAM,MAAM,OAAO;GACzB,MAAM,OAAO,aAAa,cAAc,CAAC,CAAC,OAAO;GACjD,cAAc,KAAK,GAAG,KAAK,UAAU,MAAM,MAAM,CAAC,EAAE,KAAK,OAAO;GAEhE,QAAQ,IAAI,0CAA0C,KAAK;GAC3D,OAAO;EACT;EACA,KAAK,mBAGH,OAAO,MAAM,qBAAqB,QAAQ,KAAK,MAAM,CAAC,CAAC;EAEzD,KAAK,qBACH,OAAO,MAAM,2BAA2B,QAAQ,KAAK,MAAM,CAAC,CAAC;EAE/D,KAAK;GACH,QAAQ,OAAO,MAAM,GAAG,KAAK,UAAU,cAAc,GAAG,MAAM,CAAC,EAAE,GAAG;GACpE,OAAO;EAET,KAAK;GACH,QAAQ,OAAO,MAAM,GAAG,cAAc,CAAC,CAAC,QAAQ,GAAG;GACnD,OAAO;EAET,KAAK;EACL,KAAK;EACL,KAAK;EACL,KAAK;GACH,QAAQ,OAAO,MAAM,GAAG,KAAK,GAAG;GAChC,OAAO;EACT;GACE,QAAQ,OAAO,MAAM,oBAAoB,QAAQ,IAAI,KAAK,GAAG;GAC7D,OAAO;CACX;AACF;AAEA,MAAM,mBAAwD;CAC5D,uBAAO,IAAI,IAAI;EAAC;EAAQ;EAAQ;CAAM,CAAC;CACvC,qBAAK,IAAI,IAAI,CAAC,MAAM,CAAC;CACrB,6BAAa,IAAI,IAAI,CAAC,MAAM,CAAC;CAC7B,yBAAS,IAAI,IAAI,CAAC,QAAQ,KAAK,CAAC;CAChC,yBAAS,IAAI,IAAI,CAAC,MAAM,CAAC;CACzB,sBAAM,IAAI,IAAI;CACd,0BAAU,IAAI,IAAI;CAClB,sBAAM,IAAI,IAAI;CACd,6BAAa,IAAI,IAAI;CACrB,oBAAI,IAAI,IAAI;AACd;AAEA,SAAS,iBAAiB,SAAiB,OAAqC;CAE9E,IAAI,YAAY,qBAAqB,YAAY,qBAAqB;CACtE,MAAM,UAAU,iBAAiB;CACjC,IAAI,CAAC,SAAS;CACd,MAAM,UAAU,OAAO,KAAK,KAAK,CAAC,CAAC,QAAQ,SAAS,CAAC,QAAQ,IAAI,IAAI,CAAC;CACtE,IAAI,QAAQ,SAAS,GACnB,MAAM,IAAI,MAAM,oBAAoB,WAAW,OAAO,MAAM,QAAQ,IAAI;AAE5E;AAEA,SAAS,UAAU,KAAqB;CACtC,MAAM,OAAO,OAAO,GAAG;CACvB,IAAI,CAAC,OAAO,UAAU,IAAI,KAAK,OAAO,KAAK,OAAO,OAChD,MAAM,IAAI,MAAM,uDAAuD,KAAK,UAAU,GAAG,GAAG;CAE9F,OAAO;AACT;AAEA,KAAK,CAAC,CACH,MAAM,SAAS,QAAQ,KAAK,IAAI,CAAC,CAAC,CAClC,OAAO,QAAQ;CAEd,QAAQ,MAAM,2BAA2B,GAAG;CAC5C,QAAQ,KAAK,CAAC;AAChB,CAAC"}

package/dist/{client-C8L6h6Wf.d.ts → client-BwPKohkJ.d.ts}

@@ -1,6 +1,6 @@
-import { l as RunTerminalOutcome } from "./run-record-BPCa2rQ8.js";
-import { _ as GateDecision, j as MutableSurface } from "./types-CTGbIm57.js";
-import { o as InsightReport } from "./insight-report-B9ooYH_g.js";
+import { l as RunTerminalOutcome } from "./run-record-DwHMk1Ai.js";
+import { _ as GateDecision, j as MutableSurface } from "./types-zFYez3PK.js";
+import { o as InsightReport } from "./insight-report-CXd8VBDR.js";
 //#region src/hosted/types.d.ts
 declare const HOSTED_WIRE_VERSION: '2026-07-24.v1';
 type HostedWireVersion = typeof HOSTED_WIRE_VERSION;
@@ -199,4 +199,4 @@ declare function hostedClientFromEnv(overrides?: Partial<HostedTenant> & {
 }): HostedClient | undefined;
 //#endregion
 export { UnixNanoTimestamp as _, hostedTenantFromEnv as a, EvalRunGenerationSnapshot as c, HostedIngestHeaders as d, HostedWireVersion as f, TraceSpanEvent as g, IngestTracesRequest as h, hostedClientFromEnv as i, EvalRunStatus as l, IngestResponse as m, HostedTenant as n, EvalRunCellScore as o, IngestEvalRunsRequest as p, createHostedClient as r, EvalRunEvent as s, HostedClient as t, HOSTED_WIRE_VERSION as u };
-//# sourceMappingURL=client-C8L6h6Wf.d.ts.map
+//# sourceMappingURL=client-BwPKohkJ.d.ts.map

package/dist/{client-C8L6h6Wf.d.ts.map → client-BwPKohkJ.d.ts.map}

@@ -1 +1 @@
-{"version":3,"file":"client-C8L6h6Wf.d.ts","names":[],"sources":["../src/hosted/types.ts","../src/hosted/client.ts"],"mappings":";;;;cAgCa;KACD,2BAA2B;;UAKtB;;EAEf;;EAEA;;EAEA,yBAAyB;;EAEzB;;;KAMU;UAQK;;EAEf;;EAEA;;EAEA;;EAEA,YAAY,eAAe;;EAE3B,iBAAiB;;EAEjB;;EAEA;;UAGe;;EAEf;;;EAGA;;;EAGA,UAAU;;EAEV,OAAO;;EAEP;;EAEA;;EAEA;;;;;;;UAQe;;EAEf;;EAEA;;EAEA;;EAEA,QAAQ;;EAER,QAAQ;;EAER,WAAW;;EAEX,aAAa;;EAEb,eAAe;;EAEf;;EAEA;;EAEA;;EAEA;;;;;;EAMA,gBAAgB;;;;;;KASN;;;;;;UAOK;EACf;EACA;EACA;EACA;EACA,mBAAmB;EACnB,iBAAiB;EACjB,YAAY;EACZ,SAAS;IACP,cAAc;IACd;IACA,aAAa;;EAEf;IAAW;IAAgC;;;EAE3C;;EAEA;;EAEA;;EAEA;;UAKe;EACf,aAAa;EACb,QAAQ;;UAGO;EACf,aAAa;EACb,OAAO;;UAGQ;;EAEf;;EAEA,UAAU;IAAQ;IAAe;;;;;UC1JlB;;EAEf;;EAEA;;EAEA;;EAEA,mBAAmB;;EAEnB;;EAEA;;UAGe;EACf,cAAc,OAAO,cAAc,0BAA0B,QAAQ;EACrE,eAAe,QAAQ,gBAAgB,0BAA0B,QAAQ;EACzE,aAAa,OAAO,kBAAkB,0BAA0B,QAAQ;WAC/D,QAAQ;WACR,aAAa;;iBAyGR,mBAAmB,QAAQ,eAAe;;;;;;;;;;;;;;;;;;;;;;;;;;;iBAsE1C,oBACd,YAAW,QAAQ;EAAkB,MAAM;IAC1C;iBAiBa,oBACd,YAAW,QAAQ;EAAkB,MAAM;IAC1C"}
+{"version":3,"file":"client-BwPKohkJ.d.ts","names":[],"sources":["../src/hosted/types.ts","../src/hosted/client.ts"],"mappings":";;;;cAgCa;KACD,2BAA2B;;UAKtB;;EAEf;;EAEA;;EAEA,yBAAyB;;EAEzB;;;KAMU;UAQK;;EAEf;;EAEA;;EAEA;;EAEA,YAAY,eAAe;;EAE3B,iBAAiB;;EAEjB;;EAEA;;UAGe;;EAEf;;;EAGA;;;EAGA,UAAU;;EAEV,OAAO;;EAEP;;EAEA;;EAEA;;;;;;;UAQe;;EAEf;;EAEA;;EAEA;;EAEA,QAAQ;;EAER,QAAQ;;EAER,WAAW;;EAEX,aAAa;;EAEb,eAAe;;EAEf;;EAEA;;EAEA;;EAEA;;;;;;EAMA,gBAAgB;;;;;;KASN;;;;;;UAOK;EACf;EACA;EACA;EACA;EACA,mBAAmB;EACnB,iBAAiB;EACjB,YAAY;EACZ,SAAS;IACP,cAAc;IACd;IACA,aAAa;;EAEf;IAAW;IAAgC;;;EAE3C;;EAEA;;EAEA;;EAEA;;UAKe;EACf,aAAa;EACb,QAAQ;;UAGO;EACf,aAAa;EACb,OAAO;;UAGQ;;EAEf;;EAEA,UAAU;IAAQ;IAAe;;;;;UC1JlB;;EAEf;;EAEA;;EAEA;;EAEA,mBAAmB;;EAEnB;;EAEA;;UAGe;EACf,cAAc,OAAO,cAAc,0BAA0B,QAAQ;EACrE,eAAe,QAAQ,gBAAgB,0BAA0B,QAAQ;EACzE,aAAa,OAAO,kBAAkB,0BAA0B,QAAQ;WAC/D,QAAQ;WACR,aAAa;;iBAyGR,mBAAmB,QAAQ,eAAe;;;;;;;;;;;;;;;;;;;;;;;;;;;iBAsE1C,oBACd,YAAW,QAAQ;EAAkB,MAAM;IAC1C;iBAiBa,oBACd,YAAW,QAAQ;EAAkB,MAAM;IAC1C"}

package/dist/{completion-verifier-DSyRNVzU.d.ts → completion-verifier-B4-IMYcS.d.ts}

@@ -1,6 +1,6 @@
 import { t as DefaultVerdict } from "./verdict-Dps8_okt.js";
-import { c as CostLedgerHandle } from "./cost-ledger-D2o6JOrL.js";
-import { A as ChatClient, _t as RawProviderSink } from "./types-CTvKfr5F.js";
+import { c as CostLedgerHandle } from "./cost-ledger-B1D3COAc.js";
+import { A as ChatClient, _t as RawProviderSink } from "./types-5q2T25iW.js";
 //#region src/artifact-validator.d.ts
 /**
  * Artifact validators.
@@ -237,4 +237,4 @@ declare function createTokenRecallChecker(opts?: {
 }): CorrectnessChecker;
 //#endregion
 export { jsonHasKeys as C, containsAll as S, ValidationContext as _, ProducedProposal as a, byteLengthRange as b, SatisfiedBy as c, createLlmCorrectnessChecker as d, createTokenRecallChecker as f, ArtifactValidator as g, Artifact as h, LlmCorrectnessCheckerOpts as i, TaskGold as l, verifyCompletion as m, CompletionVerdict as n, ProducedState as o, parseCorrectnessResponse as p, CorrectnessChecker as r, RequirementCheck as s, CompletionRequirement as t, completionVerdict as u, ValidationIssue as v, regexMatch as w, composeValidators as x, ValidationResult as y };
-//# sourceMappingURL=completion-verifier-DSyRNVzU.d.ts.map
+//# sourceMappingURL=completion-verifier-B4-IMYcS.d.ts.map

package/dist/{completion-verifier-DSyRNVzU.d.ts.map → completion-verifier-B4-IMYcS.d.ts.map}

@@ -1 +1 @@
-{"version":3,"file":"completion-verifier-DSyRNVzU.d.ts","names":[],"sources":["../src/artifact-validator.ts","../src/completion-verifier.ts"],"mappings":";;;;;;;;;;;;;;;;;UAciB;;EAEf;;EAEA;;EAEA;;EAEA,QAAQ;;EAER,WAAW;;UAGI;EACf;EACA;;EAEA,iBAAiB;;EAEjB,QAAQ;;UAGO;EACf;EACA;;EAEA;;UAGe;EACf;;EAEA;EACA,QAAQ;;EAER,WAAW;;UAGI;;EAEf;;EAEA;;EAEA,SAAS,UAAU,UAAU,SAAS,oBAAoB,QAAQ;;;;;;iBAWpD,kBACd,YAAY,qBACZ;EAAY;EAAe;IAC1B;;iBAgCa,WAAW,cAAc,SAAS,SAAS;;iBAkB3C,YAAY,cAAc,0BAA0B;;iBAuCpD,gBAAgB,cAAc,aAAa,cAAc;;iBAoBzD,YACd,cACA,oBACA;EAAY;IACX;;;;KCnJS;UAEK;;EAEf;;EAEA;;EAEA;;EAEA,cAAc;;UAGC;EACf;EACA,cAAc;;UAGC;EACf;EACA;EACA;;EAEA;;;UAIe;;EAEf,WAAW;;EAEX,WAAW;;EAEX;;UAGe;EACf;EACA;;EAEA;;;;;;EAMA;;EAEA;;;;;;;;;EASA;;EAEA;;EAEA;;;;;UAMe,0BAA0B;EACzC;EACA,cAAc;;EAEd;;EAEA;;EAEA;;;;;;;;;iBAUc,kBAAkB;EAChC;EACA,cAAc;IACZ;;;;;;KAoCQ,sBACV,aAAa,uBACb,oBACG;EAAU;EAAkB;;;;;;;;;iBAqLX,iBACpB,MAAM,UACN,OAAO,eACP,kBAAkB,qBACjB,QAAQ;UAoFM;EACf;;EAEA,aAAa;EACb;EACA,WAAW;EACX,SAAS;;EAET;;;;;;EAMA;;;;;;EAMA,UAAU;;;;;;;;;iBAUI,yBAAyB;EAAgB;EAAkB;;;;;;;;iBAiC3D,4BACd,MAAM,YACN,OAAM,4BACL;;;;;;;;;;;;;;;iBA6Ia,yBACd;EAAQ;EAAoB;IAC3B"}
+{"version":3,"file":"completion-verifier-B4-IMYcS.d.ts","names":[],"sources":["../src/artifact-validator.ts","../src/completion-verifier.ts"],"mappings":";;;;;;;;;;;;;;;;;UAciB;;EAEf;;EAEA;;EAEA;;EAEA,QAAQ;;EAER,WAAW;;UAGI;EACf;EACA;;EAEA,iBAAiB;;EAEjB,QAAQ;;UAGO;EACf;EACA;;EAEA;;UAGe;EACf;;EAEA;EACA,QAAQ;;EAER,WAAW;;UAGI;;EAEf;;EAEA;;EAEA,SAAS,UAAU,UAAU,SAAS,oBAAoB,QAAQ;;;;;;iBAWpD,kBACd,YAAY,qBACZ;EAAY;EAAe;IAC1B;;iBAgCa,WAAW,cAAc,SAAS,SAAS;;iBAkB3C,YAAY,cAAc,0BAA0B;;iBAuCpD,gBAAgB,cAAc,aAAa,cAAc;;iBAoBzD,YACd,cACA,oBACA;EAAY;IACX;;;;KCnJS;UAEK;;EAEf;;EAEA;;EAEA;;EAEA,cAAc;;UAGC;EACf;EACA,cAAc;;UAGC;EACf;EACA;EACA;;EAEA;;;UAIe;;EAEf,WAAW;;EAEX,WAAW;;EAEX;;UAGe;EACf;EACA;;EAEA;;;;;;EAMA;;EAEA;;;;;;;;;EASA;;EAEA;;EAEA;;;;;UAMe,0BAA0B;EACzC;EACA,cAAc;;EAEd;;EAEA;;EAEA;;;;;;;;;iBAUc,kBAAkB;EAChC;EACA,cAAc;IACZ;;;;;;KAoCQ,sBACV,aAAa,uBACb,oBACG;EAAU;EAAkB;;;;;;;;;iBAqLX,iBACpB,MAAM,UACN,OAAO,eACP,kBAAkB,qBACjB,QAAQ;UAoFM;EACf;;EAEA,aAAa;EACb;EACA,WAAW;EACX,SAAS;;EAET;;;;;;EAMA;;;;;;EAMA,UAAU;;;;;;;;;iBAUI,yBAAyB;EAAgB;EAAkB;;;;;;;;iBAiC3D,4BACd,MAAM,YACN,OAAM,4BACL;;;;;;;;;;;;;;;iBA6Ia,yBACd;EAAQ;EAAoB;IAC3B"}

package/dist/contract/index.d.ts

@@ -1,14 +1,14 @@
-import { c as CostLedgerHandle, f as CostLedgerSummary, m as CostReceipt } from "../cost-ledger-D2o6JOrL.js";
-import { a as RunRecord, n as RunCostProvenance, s as RunSplitTag } from "../run-record-BPCa2rQ8.js";
-import { A as ChatClient, F as CreateChatClientOpts, V as createChatClient } from "../types-CTvKfr5F.js";
-import { h as ProposalFindingOrigin, i as AnalystFinding, m as ProposalFinding, v as makeProposalFinding } from "../types-CKswbJGO.js";
-import { n as buildDefaultAnalystRegistry, t as DefaultAnalystRegistryOptions } from "../default-registry-Dc5D_Loc.js";
-import { C as JudgeDimension, H as SurfaceProposer, M as OptimizerConfig, R as Scenario, S as JudgeConfig, V as SessionScript, _ as GateDecision, a as CampaignResult, b as GenerationRecord, c as CampaignTraceWriter, d as DispatchContext, f as DispatchFn, g as GateContribution, h as GateContext, i as CampaignCostMeter, j as MutableSurface, k as LabeledScenarioStore, l as CodeSurface, m as GateCheckStatus, n as CampaignArtifactWriter, p as Gate, r as CampaignCellResult, t as CampaignAggregates, v as GateResult, w as JudgeScore, y as GenerationCandidate } from "../types-CTGbIm57.js";
-import { A as RunEvalOptions, B as OptimizerModelBudget, Bt as ComparisonCost, C as RunImprovementLoopOptions, Cn as ReferenceEquivalenceJudgeResult, D as RunOptimizationOptions, Dn as LlmJudgeDimension, E as PremeasuredOptimizationBaseline, En as runReferenceEquivalenceJudge, F as GepaOptimizationMethodConfig, Ft as ExternalTextOptimizerContext, G as ObjectiveSource, Gt as OptimizationMethodProvenance, H as AxisVerdict, Ht as OptimizationMethodComparison, I as GepaOptimizationRecipe, It as ExternalTextOptimizerResult, J as PromotionPolicy, K as ParetoSignificanceGateOptions, Kt as OptimizationMethodResult, L as GepaRunnerCommand, Lt as ExternalOptimizationExample, M as GepaAdaptiveEngineRun, Mt as composeGate, N as GepaEngineOptions, Nt as externalTextOptimizationMethod, On as LlmJudgeOptions, P as GepaEngineRun, Pt as ExternalTextOptimizationMethodConfig, R as gepaOptimizationMethod, Rt as ExternalTextEvaluationResponse, Sn as ReferenceEquivalenceJudgeOptions, T as runImprovementLoop, Tn as createReferenceEquivalenceJudge, U as BuildEvidenceVectorOptions, Ut as OptimizationMethodInput, V as AxisEvidence, Vt as OptimizationMethod, W as EvidenceVector, X as paretoPolicy, Xt as OptimizationTokenUsage, Y as buildEvidenceVector, Yt as OptimizationPackageSource, Z as paretoSignificanceGate, Zt as compareOptimizationMethods, bn as REFERENCE_EQUIVALENCE_JUDGE_VERSION, dt as DefaultProductionGateCheck, en as CampaignCellFailureReceipt, ft as DefaultProductionGateOptions, i as skillOptOptimizationMethod, in as RunCampaignOptions, j as runEval, kn as llmJudge, ln as fsCampaignStorage, lt as HeldOutGateOptions, mn as campaignSplitDigest, mt as defaultProductionGate, n as SkillOptRunnerCommand, on as runCampaign, ot as PowerPreflight, p as LoopProvenanceRecord, pt as DefaultProductionRewardHackingOptions, q as PromotionObjective, r as SkillOptTrainerConfig, sn as CampaignStorage, t as SkillOptOptimizationMethodConfig, un as inMemoryCampaignStorage, ut as heldOutGate, w as RunImprovementLoopResult, wn as ReferenceEquivalenceScenario, xn as ReferenceEquivalenceJudgeInput, yn as REFERENCE_EQUIVALENCE_INPUT_LIMITS, z as OpenAICompatibleOptimizerModel, zt as CompareOptimizationMethodsOptions } from "../skillopt-optimization-method-BpbnlvAZ.js";
-import { a as FailureClusterInsight, c as JudgeInsight, d as Recommendation, f as ReleaseSummary, i as FailureClassTally, l as LiftInsight, m as TokenUsageInsight, n as ExecutionErrorOutcomeCell, o as InsightReport, p as ScalarDistribution, r as ExecutionInsight, s as InterRaterInsight, t as CostProvenanceSummary, u as OutcomeCorrelationInsight } from "../insight-report-B9ooYH_g.js";
-import { c as EvalRunGenerationSnapshot, g as TraceSpanEvent, n as HostedTenant, o as EvalRunCellScore, s as EvalRunEvent } from "../client-C8L6h6Wf.js";
+import { c as CostLedgerHandle, f as CostLedgerSummary, m as CostReceipt } from "../cost-ledger-B1D3COAc.js";
+import { a as RunRecord, n as RunCostProvenance, s as RunSplitTag } from "../run-record-DwHMk1Ai.js";
+import { A as ChatClient, F as CreateChatClientOpts, V as createChatClient } from "../types-5q2T25iW.js";
+import { h as ProposalFindingOrigin, i as AnalystFinding, m as ProposalFinding, v as makeProposalFinding } from "../types-BtJhn8v6.js";
+import { n as buildDefaultAnalystRegistry, t as DefaultAnalystRegistryOptions } from "../default-registry-PUhIVRWz.js";
+import { C as JudgeDimension, H as SurfaceProposer, M as OptimizerConfig, R as Scenario, S as JudgeConfig, V as SessionScript, _ as GateDecision, a as CampaignResult, b as GenerationRecord, c as CampaignTraceWriter, d as DispatchContext, f as DispatchFn, g as GateContribution, h as GateContext, i as CampaignCostMeter, j as MutableSurface, k as LabeledScenarioStore, l as CodeSurface, m as GateCheckStatus, n as CampaignArtifactWriter, p as Gate, r as CampaignCellResult, t as CampaignAggregates, v as GateResult, w as JudgeScore, y as GenerationCandidate } from "../types-zFYez3PK.js";
+import { A as RunEvalOptions, B as OptimizerModelBudget, Bt as ComparisonCost, C as RunImprovementLoopOptions, Cn as ReferenceEquivalenceJudgeResult, D as RunOptimizationOptions, Dn as LlmJudgeDimension, E as PremeasuredOptimizationBaseline, En as runReferenceEquivalenceJudge, F as GepaOptimizationMethodConfig, Ft as ExternalTextOptimizerContext, G as ObjectiveSource, Gt as OptimizationMethodProvenance, H as AxisVerdict, Ht as OptimizationMethodComparison, I as GepaOptimizationRecipe, It as ExternalTextOptimizerResult, J as PromotionPolicy, K as ParetoSignificanceGateOptions, Kt as OptimizationMethodResult, L as GepaRunnerCommand, Lt as ExternalOptimizationExample, M as GepaAdaptiveEngineRun, Mt as composeGate, N as GepaEngineOptions, Nt as externalTextOptimizationMethod, On as LlmJudgeOptions, P as GepaEngineRun, Pt as ExternalTextOptimizationMethodConfig, R as gepaOptimizationMethod, Rt as ExternalTextEvaluationResponse, Sn as ReferenceEquivalenceJudgeOptions, T as runImprovementLoop, Tn as createReferenceEquivalenceJudge, U as BuildEvidenceVectorOptions, Ut as OptimizationMethodInput, V as AxisEvidence, Vt as OptimizationMethod, W as EvidenceVector, X as paretoPolicy, Xt as OptimizationTokenUsage, Y as buildEvidenceVector, Yt as OptimizationPackageSource, Z as paretoSignificanceGate, Zt as compareOptimizationMethods, bn as REFERENCE_EQUIVALENCE_JUDGE_VERSION, dt as DefaultProductionGateCheck, en as CampaignCellFailureReceipt, ft as DefaultProductionGateOptions, i as skillOptOptimizationMethod, in as RunCampaignOptions, j as runEval, kn as llmJudge, ln as fsCampaignStorage, lt as HeldOutGateOptions, mn as campaignSplitDigest, mt as defaultProductionGate, n as SkillOptRunnerCommand, on as runCampaign, ot as PowerPreflight, p as LoopProvenanceRecord, pt as DefaultProductionRewardHackingOptions, q as PromotionObjective, r as SkillOptTrainerConfig, sn as CampaignStorage, t as SkillOptOptimizationMethodConfig, un as inMemoryCampaignStorage, ut as heldOutGate, w as RunImprovementLoopResult, wn as ReferenceEquivalenceScenario, xn as ReferenceEquivalenceJudgeInput, yn as REFERENCE_EQUIVALENCE_INPUT_LIMITS, z as OpenAICompatibleOptimizerModel, zt as CompareOptimizationMethodsOptions } from "../skillopt-optimization-method-CzfnA8O-.js";
+import { a as FailureClusterInsight, c as JudgeInsight, d as Recommendation, f as ReleaseSummary, i as FailureClassTally, l as LiftInsight, m as TokenUsageInsight, n as ExecutionErrorOutcomeCell, o as InsightReport, p as ScalarDistribution, r as ExecutionInsight, s as InterRaterInsight, t as CostProvenanceSummary, u as OutcomeCorrelationInsight } from "../insight-report-CXd8VBDR.js";
+import { c as EvalRunGenerationSnapshot, g as TraceSpanEvent, n as HostedTenant, o as EvalRunCellScore, s as EvalRunEvent } from "../client-BwPKohkJ.js";
 import { i as InMemoryOutcomeStore, n as FileSystemOutcomeStore, o as OutcomeStore, r as FileSystemOutcomeStoreOptions, t as DeploymentOutcome } from "../outcome-store-BYHIuO0e.js";
-import { a as summarizeExecution, i as analyzeRuns, n as ExecutionReport, r as SummarizeExecutionOptions, t as AnalyzeRunsOptions } from "../analyze-runs-PVtnfjvA.js";
+import { a as summarizeExecution, i as analyzeRuns, n as ExecutionReport, r as SummarizeExecutionOptions, t as AnalyzeRunsOptions } from "../analyze-runs-CPYxfPWT.js";
 import { AgentCandidateBenchmarkCellRef, AgentCandidateBenchmarkSuiteInputs, AgentCandidateBenchmarkTask, AgentCandidateBenchmarkTaskMaterial, AgentCandidateBundle, AgentCandidateEvaluationPolicy, AgentCandidateExperiment, AgentCandidateExperimentMaterial, AgentCandidateExperimentMeasurement, AgentImprovementCost, AgentImprovementMeasuredComparison, AgentProfileImprovementExperiment, AgentProfileImprovementExperimentMaterial, AgentProfileImprovementMeasuredComparison, AgentProfileImprovementMeasurement, AgentProfileImprovementRunCell, AgentProfileImprovementRunReceipt, AgentProfileImprovementSuiteInputs, AgentProfileImprovementTask, AgentProfileImprovementTaskMaterial, CandidateExecutionEvidence, Sha256Digest } from "@tangle-network/agent-interface";
 //#region src/contract/self-improve.d.ts
 interface SelfImproveBudget {

package/dist/contract/index.js

@@ -1,23 +1,23 @@
 import { c as ValidationError } from "../errors-D-LKuDhb.js";
-import { R as createChatClient, t as buildDefaultAnalystRegistry } from "../default-registry-CLXbRt0f.js";
+import { G as createChatClient, t as buildDefaultAnalystRegistry } from "../default-registry-lp5R0lve.js";
 import { i as estimateCost, o as isModelPriced } from "../metrics-C9YY1OcL.js";
-import { i as CostLedger } from "../cost-ledger-D-5_-dhi.js";
+import { i as CostLedger } from "../cost-ledger-CHDLA0Ss.js";
 import { LLM_MODEL_ATTR_KEYS, SPAN_KIND_ATTR_KEYS } from "../trace-attributes.js";
-import { M as isOtlpModelCall, j as classifyOtlpSpanRole } from "../tools-DZk2Jn64.js";
-import { l as makeProposalFinding } from "../proposal-findings-DCawte-y.js";
-import { r as mapConcurrentRange } from "../concurrency-MUjT7VjM.js";
-import { B as surfaceContentHash, Ct as llmJudge, Ft as decidePairedPromotion, M as compareOptimizationMethods, O as composeGate, Q as inMemoryCampaignStorage, S as defaultProductionGate, T as heldoutSignificance, V as surfaceHash, X as createRunCostLedger, Y as runCampaign, Z as fsCampaignStorage, _ as buildEvidenceVector, a as emitLoopProvenance, b as powerPreflight, ct as campaignSplitDigest, d as runImprovementLoop, dt as REFERENCE_EQUIVALENCE_INPUT_LIMITS, ft as REFERENCE_EQUIVALENCE_JUDGE_VERSION, g as gepaOptimizationMethod, j as assertOptimizationResult, k as externalTextOptimizationMethod, mt as runReferenceEquivalenceJudge, o as loopProvenanceArgsFromResult, p as runEval, pt as createReferenceEquivalenceJudge, rt as resolveRunDir, t as skillOptOptimizationMethod, v as paretoPolicy, x as heldOutGate, y as paretoSignificanceGate } from "../skillopt-optimization-method-f4o9sUT4.js";
+import { A as classifyOtlpSpanRole, j as isOtlpModelCall } from "../store-otlp-BenKynPE.js";
+import { h as makeProposalFinding, i as inMemoryCampaignStorage, n as createRunCostLedger, r as fsCampaignStorage } from "../single-run-lock-D_bS5xhj.js";
+import { m as mapConcurrentRange } from "../ledger-core-C0Yx1I14.js";
+import { A as assertOptimizationResult, B as surfaceHash, Dt as decidePairedPromotion, J as runCampaign, O as composeGate, S as defaultProductionGate, T as heldoutSignificance, Y as resolveRunDir, _ as buildEvidenceVector, a as emitLoopProvenance, at as createReferenceEquivalenceJudge, b as powerPreflight, d as runImprovementLoop, et as campaignSplitDigest, g as gepaOptimizationMethod, ht as llmJudge, it as REFERENCE_EQUIVALENCE_JUDGE_VERSION, j as compareOptimizationMethods, k as externalTextOptimizationMethod, o as loopProvenanceArgsFromResult, ot as runReferenceEquivalenceJudge, p as runEval, rt as REFERENCE_EQUIVALENCE_INPUT_LIMITS, t as skillOptOptimizationMethod, v as paretoPolicy, x as heldOutGate, y as paretoSignificanceGate, z as surfaceContentHash } from "../skillopt-optimization-method-CjKMZy0d.js";
 import { E as pairedBootstrap } from "../statistics-ByxzSiOM.js";
 import { i as parseRunRecordSafe, r as modelHasSnapshot } from "../run-record-vRgqWmJw.js";
-import { a as recordAggregateMeasurements, i as readTaskFailureLabels, o as summarizeExecutionMeasurements, s as summarizeTraceErrors, t as extractUsage } from "../extract-usage-p-56bh8q.js";
+import { a as recordAggregateMeasurements, i as readTaskFailureLabels, o as summarizeExecutionMeasurements, s as summarizeTraceErrors, t as extractUsage } from "../extract-usage-CS391dOE.js";
 import { n as summarizeExecution, t as analyzeRuns } from "../analyze-runs-BScZvqMV.js";
 import { a as campaignCellExecutionEvidence, c as campaignCellToRunRecord, o as campaignCellJudgeDimensions, s as campaignCellTaskScore } from "../reward-hacking-GyN0kMd8.js";
 import { n as InMemoryOutcomeStore, t as FileSystemOutcomeStore } from "../outcome-store-ChBKlTd_.js";
 import { t as createHostedClient } from "../client-LIuo-KPv.js";
 import { dirname, join } from "node:path";
 import { createHash } from "node:crypto";
-import { mkdir, readFile, readdir, stat, writeFile } from "node:fs/promises";
 import { agentCandidateBenchmarkSuiteSchema, agentCandidateBenchmarkTaskSchema, agentCandidateBundleSchema, agentCandidateEvaluationPolicySchema, agentCandidateExperimentSchema, agentImprovementMeasuredComparisonSchema, agentProfileImprovementExperimentSchema, agentProfileImprovementMeasuredComparisonSchema, agentProfileImprovementRunCellSchema, agentProfileImprovementRunReceiptSchema, agentProfileImprovementSuiteInputsSchema, agentProfileImprovementSuiteSchema, agentProfileImprovementTaskSchema, candidateExecutionEvidenceSchema, canonicalCandidateDigest, canonicalCandidateJson, numbersApproximatelyEqual, omitTopLevelDigest } from "@tangle-network/agent-interface";
+import { mkdir, readFile, readdir, stat, writeFile } from "node:fs/promises";
 //#region src/contract/self-improve.ts
 /** Failed self-improvement run with an immutable receipt snapshot. */
 var SelfImproveRunError = class extends Error {

package/dist/control.d.ts

@@ -1,3 +1,3 @@
-import { $ as ControlRunResult, J as ControlActionOutcome, Q as ControlEvalResult, X as ControlContext, Y as ControlBudget, Z as ControlDecision, at as StopDecision, ct as runAgentControlLoop, dt as subjectiveEval, et as ControlRuntimeConfig, it as ControlStopPolicies, lt as stopOnNoProgress, nt as ControlSeverity, ot as allCriticalPassed, q as ControlActionFailureMode, rt as ControlStep, st as objectiveEval, tt as ControlRuntimeError, ut as stopOnRepeatedAction } from "./feedback-trajectory-N_F0PwHz.js";
-import { A as ActionExecutionPolicy, M as evaluateActionPolicy, _ as ProposeReviewReport, a as ProposeReviewControlAction, c as ProposeReviewControlState, g as ProposeReviewConfig, i as scoreFromEvals, j as ActionPolicyDecision, k as runProposeReview, n as RunEvidenceMetadata, o as ProposeReviewControlConfig, r as controlRunToRunRecord, s as ProposeReviewControlResult, t as ControlRunToRunRecordOptions, u as runProposeReviewAsControlLoop } from "./run-evidence-BDIircdA.js";
+import { $ as ControlRunResult, J as ControlActionOutcome, Q as ControlEvalResult, X as ControlContext, Y as ControlBudget, Z as ControlDecision, at as StopDecision, ct as runAgentControlLoop, dt as subjectiveEval, et as ControlRuntimeConfig, it as ControlStopPolicies, lt as stopOnNoProgress, nt as ControlSeverity, ot as allCriticalPassed, q as ControlActionFailureMode, rt as ControlStep, st as objectiveEval, tt as ControlRuntimeError, ut as stopOnRepeatedAction } from "./feedback-trajectory-CoNep7rl.js";
+import { A as ActionExecutionPolicy, M as evaluateActionPolicy, _ as ProposeReviewReport, a as ProposeReviewControlAction, c as ProposeReviewControlState, g as ProposeReviewConfig, i as scoreFromEvals, j as ActionPolicyDecision, k as runProposeReview, n as RunEvidenceMetadata, o as ProposeReviewControlConfig, r as controlRunToRunRecord, s as ProposeReviewControlResult, t as ControlRunToRunRecordOptions, u as runProposeReviewAsControlLoop } from "./run-evidence-CbE0A8Xg.js";
 export { type ActionExecutionPolicy, type ActionPolicyDecision, type ControlActionFailureMode, type ControlActionOutcome, type ControlBudget, type ControlContext, type ControlDecision, type ControlEvalResult, type ControlRunResult, type ControlRunToRunRecordOptions, type ControlRuntimeConfig, type ControlRuntimeError, type ControlSeverity, type ControlStep, type ControlStopPolicies, type ProposeReviewConfig, type ProposeReviewControlAction, type ProposeReviewControlConfig, type ProposeReviewControlResult, type ProposeReviewControlState, type ProposeReviewReport, type RunEvidenceMetadata, type StopDecision, allCriticalPassed, controlRunToRunRecord, evaluateActionPolicy, objectiveEval, runAgentControlLoop, runProposeReview, runProposeReviewAsControlLoop, scoreFromEvals, stopOnNoProgress, stopOnRepeatedAction, subjectiveEval };

package/dist/{cost-ledger-D2o6JOrL.d.ts → cost-ledger-B1D3COAc.d.ts}

@@ -55,7 +55,7 @@ interface CostReceipt extends CostCallBase, CostUsage {
   actualCostUsd?: number;
   /** Known non-provider amount supplied by an external executor. */
   estimatedCostUsd?: number;
-  error?: string;
+  error?: 'paid-call-failed';
 }
 interface CostReceiptInput extends CostUsage {
   model: string;
@@ -168,7 +168,7 @@ interface CostLedgerOptions {
   receipts?: readonly CostReceipt[];
 }
 declare class CostCeilingReachedError extends ValidationError {
-  constructor(ceilingUsd: number, committedAndReservedUsd: number, requestedUsd: number, phase: string, actor: string);
+  constructor(ceilingUsd: number, committedUsd: number, requestedUsd: number, phase: string, actor: string);
 }
 declare class CostAccountingIncompleteError extends ValidationError {}
 declare class CostReservationExceededError extends ValidationError {
@@ -211,7 +211,7 @@ declare class CostLedger {
   waitForIdle(options?: CostLedgerWaitOptions): Promise<boolean>;
   /** Settle a call left pending by a crashed process after reconciling with the provider. */
   reconcile(callId: string, observed: CostReceiptInput, options?: {
-    error?: string;
+    failed?: boolean;
   }): CostReceipt;
   list(filter?: CostLedgerFilter): CostReceipt[];
   /** Read pending calls without exposing mutable ledger state. */
@@ -222,6 +222,7 @@ declare class CostLedger {
   private execute;
   private captureLateOutcome;
   private releaseActiveCall;
+  private waitForReservationRelease;
   private commitOutcome;
   private captureFailure;
   private commitReceipt;
@@ -249,4 +250,4 @@ declare function costForUsage(model: string, usage: CostUsage): CostResult;
 declare function costForTokenPricing(pricing: CustomTokenPricing, usage: Pick<CostUsage, 'inputTokens' | 'outputTokens' | 'cachedTokens' | 'cacheWriteTokens'>): number;
 //#endregion
 export { PendingCostCall as C, costForUsage as D, costForTokenPricing as E, modelPriceKey as O, PaidCallResult as S, RunPaidCallInput as T, CostReservationExceededError as _, CostChannel as a, CustomTokenPricing as b, CostLedgerHandle as c, CostLedgerPersistenceError as d, CostLedgerSummary as f, CostReceiptInput as g, CostReceiptCaptureError as h, CostCeilingReachedError as i, CostLedgerOptions as l, CostReceipt as m, CostAccountingIncompleteError as n, CostLedger as o, CostProvenance as p, CostCallConflictError as r, CostLedgerFilter as s, ChannelRollup as t, CostLedgerPersistence as u, CostResult as v, PendingCostCallView as w, MaximumCharge as x, CostUsage as y };
-//# sourceMappingURL=cost-ledger-D2o6JOrL.d.ts.map
+//# sourceMappingURL=cost-ledger-B1D3COAc.d.ts.map

package/dist/{cost-ledger-D2o6JOrL.d.ts.map → cost-ledger-B1D3COAc.d.ts.map}

@@ -1 +1 @@
-{"version":3,"file":"cost-ledger-D2o6JOrL.d.ts","names":[],"sources":["../src/cost-ledger.ts"],"mappings":";;KAIY;;KAGA;EACN;EAAkB;;EAClB;EAAmB;;EACnB;EAAoB;;UAET;EACf;;EAEA;;EAEA;;EAEA;;EAEA;;UAGQ;EACR;EACA,SAAS;EACT;EACA;EACA;EACA;EACA,OAAO;EACP;;UAGe,wBAAwB;EACvC;;UAGe,4BAA4B;EAC3C;;UAGe,oBAAoB,cAAc;EACjD;EACA;EACA;EACA;;EAEA;IACE;IACA;IACA;IACA;;;EAGF;;EAEA;EACA;;UAKe,yBAAyB;EACxC;;EAEA,qBAAqB;EACrB;;EAEA;EACA;EACA;;;UAIe;;EAEf;;EAEA;;EAEA;EACA;;KAGU;EACN;;EAEA,oBAAoB;IAClB,KAAK;EACN;IAAkB;UAER,iBAAiB;EAChC;EACA,SAAS;EACT;EACA;;EAEA;EACA,OAAO;EACP,SAAS;;EAET,gBAAgB;;EAEhB,QAAQ,QAAQ,aAAa,iBAAiB,QAAQ;EACtD,QAAQ,OAAO,IAAI;EACnB,kBAAkB,OAAO,QAAQ;;KAGvB,eAAe;EACrB;EAAiB;EAAgB,OAAO;EAAG,SAAS;;EACpD;EAAkB;EAAiB,OAAO;EAAO,UAAU;;UAEhD;EACf,SAAS;EACT;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA,gBAAgB;EAChB,WAAW;EACX;EACA;EACA;EACA;EACA;;UAGe;EACf,UAAU;EACV;EACA,OAAO;;UAGQ;;EAEf;;EAEA,SAAS;;;UAIM;EACf;IAAU;IAAkB;;EAC5B,OAAO,0BAA0B;;UAGlB;EACf;EACA,cAAc;;EAEd,oBAAoB;;cAGT,gCAAgC;EAC3C,YACE,oBACA,iCACA,sBACA,eACA;;cAQS,sCAAsC;cAEtC,qCAAqC;EAChD,YAAY,eAAe,mBAAmB;;cAOnC,8BAA8B;WAChC;WACA,UAAU;EAEnB,YACE,iBACA;IAAW;IAAiB,UAAU;IAAa;;;cAQ1C,mCAAmC;WACrC;WACA,UAAU;EAEnB,YAAY,gBAAgB,iBAAiB,UAAU;;cAU5C,gCAAgC;WAClC;WACA,UAAU;WACV,cAAc;EAEvB,YAAY,gBAAgB,gBAAgB,uBAAuB,UAAU;;;cAgClE;mBACM;mBACA;mBACA;mBACA;UACT;UACA;UACA;WACC;mBACQ;EAEjB,YAAY,iBAAiB;EAwDvB,YAAY,GAAG,OAAO,iBAAiB,KAAK,QAAQ,eAAe;;EAmFnE,YAAY,UAAS,wBAA6B;;EA0BxD,UACE,gBACA,UAAU,kBACV;IAAW;MACV;EAaH,KAAK,SAAS,mBAAmB;;EAQjC,YAAY,SAAS,mBAAmB;EAcxC,QAAQ,SAAS,mBAAmB;EAkFpC,cAAc;EAYd;UAIc;UAoDN;UAsCA;UAKA;UAmBA;UAsBA;UAaA;UA+BA;UAUA;UAYA;UAMA;;;;;;;;KA0BE,mBAAmB,KAC7B,YACA,cAAc,8CAEd,QAAQ,KAAK;;iBAGC,cAAc;UAIb;EACf;EACA;;iBAGc,aAAa,eAAe,OAAO,YAAY;;iBAc/C,oBACd,SAAS,oBACT,OAAO,KAAK"}
+{"version":3,"file":"cost-ledger-B1D3COAc.d.ts","names":[],"sources":["../src/cost-ledger.ts"],"mappings":";;KAIY;;KAGA;EACN;EAAkB;;EAClB;EAAmB;;EACnB;EAAoB;;UAET;EACf;;EAEA;;EAEA;;EAEA;;EAEA;;UAGQ;EACR;EACA,SAAS;EACT;EACA;EACA;EACA;EACA,OAAO;EACP;;UAGe,wBAAwB;EACvC;;UAGe,4BAA4B;EAC3C;;UAGe,oBAAoB,cAAc;EACjD;EACA;EACA;EACA;;EAEA;IACE;IACA;IACA;IACA;;;EAGF;;EAEA;EACA;;UAKe,yBAAyB;EACxC;;EAEA,qBAAqB;EACrB;;EAEA;EACA;EACA;;;UAIe;;EAEf;;EAEA;;EAEA;EACA;;KAGU;EACN;;EAEA,oBAAoB;IAClB,KAAK;EACN;IAAkB;UAER,iBAAiB;EAChC;EACA,SAAS;EACT;EACA;;EAEA;EACA,OAAO;EACP,SAAS;;EAET,gBAAgB;;EAEhB,QAAQ,QAAQ,aAAa,iBAAiB,QAAQ;EACtD,QAAQ,OAAO,IAAI;EACnB,kBAAkB,OAAO,QAAQ;;KAGvB,eAAe;EACrB;EAAiB;EAAgB,OAAO;EAAG,SAAS;;EACpD;EAAkB;EAAiB,OAAO;EAAO,UAAU;;UAEhD;EACf,SAAS;EACT;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA,gBAAgB;EAChB,WAAW;EACX;EACA;EACA;EACA;EACA;;UAGe;EACf,UAAU;EACV;EACA,OAAO;;UAGQ;;EAEf;;EAEA,SAAS;;;UAIM;EACf;IAAU;IAAkB;;EAC5B,OAAO,0BAA0B;;UAGlB;EACf;EACA,cAAc;;EAEd,oBAAoB;;cAGT,gCAAgC;EAC3C,YACE,oBACA,sBACA,sBACA,eACA;;cAQS,sCAAsC;cAEtC,qCAAqC;EAChD,YAAY,eAAe,mBAAmB;;cAOnC,8BAA8B;WAChC;WACA,UAAU;EAEnB,YACE,iBACA;IAAW;IAAiB,UAAU;IAAa;;;cAQ1C,mCAAmC;WACrC;WACA,UAAU;EAEnB,YAAY,gBAAgB,iBAAiB,UAAU;;cAU5C,gCAAgC;WAClC;WACA,UAAU;WACV,cAAc;EAEvB,YAAY,gBAAgB,gBAAgB,uBAAuB,UAAU;;;cAgClE;mBACM;mBACA;mBACA;mBACA;UACT;UACA;UACA;WACC;mBACQ;EAEjB,YAAY,iBAAiB;EAwDvB,YAAY,GAAG,OAAO,iBAAiB,KAAK,QAAQ,eAAe;;EAyGnE,YAAY,UAAS,wBAA6B;;EA0BxD,UACE,gBACA,UAAU,kBACV;IAAW;MACV;EAaH,KAAK,SAAS,mBAAmB;;EAQjC,YAAY,SAAS,mBAAmB;EAcxC,QAAQ,SAAS,mBAAmB;EAkFpC,cAAc;EAYd;UAIc;UAoDN;UAsCA;UAKM;UAoBN;UAmBA;UAkBA;UAaA;UA+BA;UAUA;UAYA;UAMA;;;;;;;;KA0BE,mBAAmB,KAC7B,YACA,cAAc,8CAEd,QAAQ,KAAK;;iBAGC,cAAc;UAIb;EACf;EACA;;iBAGc,aAAa,eAAe,OAAO,YAAY;;iBAc/C,oBACd,SAAS,oBACT,OAAO,KAAK"}

package/dist/{cost-ledger-D-5_-dhi.js → cost-ledger-CHDLA0Ss.js}

@@ -3,8 +3,8 @@ import { i as estimateCost, o as isModelPriced, s as resolveModelPricing } from
 import { z } from "zod";
 //#region src/cost-ledger.ts
 var CostCeilingReachedError = class extends ValidationError {
-	constructor(ceilingUsd, committedAndReservedUsd, requestedUsd, phase, actor) {
-		super(`CostLedger: reserving ${requestedUsd} for '${actor}' during '${phase}' would exceed ceiling ${ceilingUsd} with ${committedAndReservedUsd} already committed or reserved`);
+	constructor(ceilingUsd, committedUsd, requestedUsd, phase, actor) {
+		super(`CostLedger: reserving ${requestedUsd} for '${actor}' during '${phase}' would exceed ceiling ${ceilingUsd} with ${committedUsd} already committed`);
 	}
 };
 var CostAccountingIncompleteError = class extends ValidationError {};
@@ -93,45 +93,55 @@ var CostLedger = class {
 				callId,
 				error: abortError(input.signal)
 			};
-			if (this.records.has(callId)) return {
-				succeeded: false,
-				callId,
-				error: new CostCallConflictError(`CostLedger: callId '${callId}' already exists`)
-			};
 			this.ensureCostLimitPersisted(callId);
-			const summary = this.summary();
-			if (summary.unresolvedCalls > 0) return {
-				succeeded: false,
-				callId,
-				error: new CostAccountingIncompleteError(`CostLedger: ${summary.unresolvedCalls} unresolved call(s) must be reconciled before new paid work`)
-			};
 			const maximumCostUsd = this.resolveMaximum(input.maximumCharge);
-			if (this.costCeilingUsd !== void 0 && this.hasIncompleteSettledCall()) return {
+			const existing = this.records.get(callId);
+			if (existing) return {
 				succeeded: false,
 				callId,
-				error: new CostAccountingIncompleteError(`CostLedger: accounting is incomplete; refusing paid call '${input.actor}' during '${input.phase}'`)
+				error: new CostCallConflictError(existing.status === "pending" ? `CostLedger: callId '${callId}' is unresolved and must be reconciled before retry` : `CostLedger: callId '${callId}' already exists`, { callId })
 			};
-			if (this.costCeilingUsd !== void 0) {
-				const committedAndReserved = summary.totalCostUsd + summary.reservedCostUsd;
-				if (committedAndReserved + maximumCostUsd > this.costCeilingUsd) return {
-					succeeded: false,
+			if (!pending) {
+				while (true) {
+					if (this.records.has(callId)) return {
+						succeeded: false,
+						callId,
+						error: new CostCallConflictError(`CostLedger: callId '${callId}' already exists`)
+					};
+					const summary = this.summary();
+					if (summary.unresolvedCalls > 0) return {
+						succeeded: false,
+						callId,
+						error: new CostAccountingIncompleteError(`CostLedger: ${summary.unresolvedCalls} unresolved call(s) must be reconciled before new paid work`)
+					};
+					if (this.costCeilingUsd === void 0) break;
+					if (this.hasIncompleteSettledCall()) return {
+						succeeded: false,
+						callId,
+						error: new CostAccountingIncompleteError(`CostLedger: accounting is incomplete; refusing paid call '${input.actor}' during '${input.phase}'`)
+					};
+					if (summary.totalCostUsd + maximumCostUsd > this.costCeilingUsd) return {
+						succeeded: false,
+						callId,
+						error: new CostCeilingReachedError(this.costCeilingUsd, summary.totalCostUsd, maximumCostUsd, input.phase, input.actor)
+					};
+					if (summary.totalCostUsd + summary.reservedCostUsd + maximumCostUsd <= this.costCeilingUsd) break;
+					await this.waitForReservationRelease(input.signal);
+				}
+				pending = {
+					status: "pending",
 					callId,
-					error: new CostCeilingReachedError(this.costCeilingUsd, committedAndReserved, maximumCostUsd, input.phase, input.actor)
+					channel: input.channel,
+					phase: input.phase,
+					actor: input.actor,
+					model: pendingModel(input),
+					...maximumCostUsd === void 0 ? {} : { maximumCostUsd },
+					...input.tags ? { tags: { ...input.tags } } : {},
+					timestamp: Date.now()
 				};
+				this.appendRecord(pending);
+				this.activeCallIds.add(callId);
 			}
-			pending = {
-				status: "pending",
-				callId,
-				channel: input.channel,
-				phase: input.phase,
-				actor: input.actor,
-				model: pendingModel(input),
-				...maximumCostUsd === void 0 ? {} : { maximumCostUsd },
-				...input.tags ? { tags: { ...input.tags } } : {},
-				timestamp: Date.now()
-			};
-			this.appendRecord(pending);
-			this.activeCallIds.add(callId);
 		} catch (error) {
 			return {
 				succeeded: false,
@@ -181,7 +191,7 @@ var CostLedger = class {
 		if (!pending) throw new CostCallConflictError(`CostLedger: no pending call '${callId}'`);
 		if (this.activeCallIds.has(callId)) throw new CostCallConflictError(`CostLedger: call '${callId}' is still active`);
 		this.ensureCostLimitPersisted(callId);
-		return this.commitReceipt(pending, observed, options.error);
+		return this.commitReceipt(pending, observed, options.failed === true);
 	}
 	list(filter) {
 		return [...this.records.values()].filter((record) => record.status === "settled").filter((receipt) => matches(receipt, filter)).map(cloneReceipt);
@@ -322,7 +332,7 @@ var CostLedger = class {
 			const error = toError(cause);
 			try {
 				const observed = input.receiptFromError?.(error);
-				this.commitReceipt(pending, observed ?? unknownReceipt(pending.model), error.message);
+				this.commitReceipt(pending, observed ?? unknownReceipt(pending.model), true);
 			} catch (receiptError) {
 				if (receiptError instanceof CostLedgerPersistenceError || receiptError instanceof CostCallConflictError) return;
 				this.captureFailure(pending, error, receiptError);
@@ -336,9 +346,28 @@ var CostLedger = class {
 		this.activeCallIds.delete(callId);
 		for (const resolve of [...this.idleWaiters]) resolve();
 	}
+	async waitForReservationRelease(signal) {
+		if (signal?.aborted) throw abortError(signal);
+		await new Promise((resolve, reject) => {
+			let finished = false;
+			const finish = (error) => {
+				if (finished) return;
+				finished = true;
+				this.idleWaiters.delete(onRelease);
+				signal?.removeEventListener("abort", onAbort);
+				if (error) reject(error);
+				else resolve();
+			};
+			const onRelease = () => finish();
+			const onAbort = () => finish(abortError(signal));
+			this.idleWaiters.add(onRelease);
+			signal?.addEventListener("abort", onAbort, { once: true });
+			if (this.activeCallIds.size === 0) onRelease();
+		});
+	}
 	commitOutcome(pending, error, observed) {
 		try {
-			const receipt = this.commitReceipt(pending, observed, error.message);
+			const receipt = this.commitReceipt(pending, observed, true);
 			return paidFailure(pending.callId, error, receipt);
 		} catch (receiptError) {
 			if (receiptError instanceof CostLedgerPersistenceError || receiptError instanceof CostCallConflictError) return paidFailure(pending.callId, receiptError);
@@ -347,15 +376,15 @@ var CostLedger = class {
 	}
 	captureFailure(pending, cause, receiptError) {
 		try {
-			const receipt = this.commitReceipt(pending, unknownReceipt(pending.model), toError(receiptError).message);
+			const receipt = this.commitReceipt(pending, unknownReceipt(pending.model), true);
 			return paidFailure(pending.callId, new CostReceiptCaptureError(pending.callId, cause, receiptError, receipt), receipt);
 		} catch (error) {
 			const typed = error instanceof Error ? error : toError(error);
 			return paidFailure(pending.callId, typed);
 		}
 	}
-	commitReceipt(pending, observed, error) {
-		const receipt = buildReceipt(pending, observed, error);
+	commitReceipt(pending, observed, failed = false) {
+		const receipt = buildReceipt(pending, observed, failed);
 		if (this.records.get(pending.callId)?.status !== "pending") throw new CostCallConflictError(`CostLedger: call '${pending.callId}' is not pending`);
 		this.appendRecord(receipt);
 		return cloneReceipt(receipt);
@@ -469,7 +498,7 @@ async function settle(promise, signal) {
 		}));
 	});
 }
-function buildReceipt(pending, observed, error) {
+function buildReceipt(pending, observed, failed = false) {
 	assertUsage(observed);
 	assertString(observed.model, "receipt.model");
 	if ([
@@ -514,7 +543,7 @@ function buildReceipt(pending, observed, error) {
 		...hasActual ? { actualCostUsd: observed.actualCostUsd } : {},
 		...hasExternalEstimate ? { estimatedCostUsd: observed.estimatedCostUsd } : {},
 		...pending.maximumCostUsd === void 0 ? {} : { maximumCostUsd: pending.maximumCostUsd },
-		...error ? { error } : {},
+		...failed ? { error: "paid-call-failed" } : {},
 		...pending.tags ? { tags: { ...pending.tags } } : {},
 		timestamp: pending.timestamp
 	}, "provider receipt");
@@ -555,7 +584,7 @@ const CostReceiptBaseShape = {
 	pricing: CostPricingSchema.optional(),
 	actualCostUsd: NonNegative.optional(),
 	estimatedCostUsd: NonNegative.optional(),
-	error: z.string().optional()
+	error: z.literal("paid-call-failed").optional()
 };
 const CostReceiptSchema = z.strictObject({
 	...CostReceiptBaseShape,
@@ -649,7 +678,7 @@ function parseEvents(serialized) {
 	let costCeilingUsd;
 	let lineNumber = 0;
 	try {
-		for (const line of serialized.split("\n")) {
+		for (const line of completeJournalPrefix(serialized).split("\n")) {
 			if (!line.trim()) continue;
 			lineNumber += 1;
 			const event = CostLedgerEventSchema.parse(JSON.parse(line));
@@ -674,6 +703,17 @@ function parseEvents(serialized) {
 		throw new ValidationError(`CostLedger: invalid persisted event ${lineNumber || 1}: ${validationMessage(cause)}`, { cause });
 	}
 }
+function completeJournalPrefix(serialized) {
+	if (!serialized || serialized.endsWith("\n")) return serialized;
+	const finalNewline = serialized.lastIndexOf("\n");
+	const tail = serialized.slice(finalNewline + 1);
+	try {
+		JSON.parse(tail);
+	} catch {
+		return finalNewline < 0 ? "" : serialized.slice(0, finalNewline + 1);
+	}
+	throw new ValidationError("CostLedger: complete final persisted event is missing its newline terminator");
+}
 function validateTransition(records, record) {
 	const current = records.get(record.callId);
 	if (!current) return;
@@ -681,7 +721,12 @@ function validateTransition(records, record) {
 	if (!sameAttribution(current, record)) throw new ValidationError(`CostLedger: receipt attribution changed for call '${record.callId}'`);
 }
 function sameAttribution(before, after) {
-	return before.channel === after.channel && before.phase === after.phase && before.actor === after.actor && before.maximumCostUsd === after.maximumCostUsd && before.timestamp === after.timestamp && JSON.stringify(before.tags ?? {}) === JSON.stringify(after.tags ?? {});
+	return before.channel === after.channel && before.phase === after.phase && before.actor === after.actor && before.maximumCostUsd === after.maximumCostUsd && before.timestamp === after.timestamp && sameTags(before.tags, after.tags);
+}
+function sameTags(left, right) {
+	const leftEntries = Object.entries(left ?? {}).sort(([a], [b]) => a.localeCompare(b));
+	const rightEntries = Object.entries(right ?? {}).sort(([a], [b]) => a.localeCompare(b));
+	return JSON.stringify(leftEntries) === JSON.stringify(rightEntries);
 }
 function parseReceipt(value, path) {
 	try {
@@ -836,4 +881,4 @@ function assertString(value, name) {
 //#endregion
 export { CostLedgerPersistenceError as a, costForTokenPricing as c, CostLedger as i, costForUsage as l, CostCallConflictError as n, CostReceiptCaptureError as o, CostCeilingReachedError as r, CostReservationExceededError as s, CostAccountingIncompleteError as t, modelPriceKey as u };
 
-//# sourceMappingURL=cost-ledger-D-5_-dhi.js.map
+//# sourceMappingURL=cost-ledger-CHDLA0Ss.js.map