@tangle-network/agent-eval 1.0.2 → 1.0.3

package/CHANGELOG.md

@@ -4,6 +4,13 @@ All notable changes to `@tangle-network/agent-eval` and its sibling `agent-eval-
 
 ---
 
+## [1.0.3] — 2026-10-29
+
+- chore(release): declare the existing deployment runner for workflow lint
+- fix(release): bind bootstrap to the existing qualified action revision
+- fix(release): use the qualified pnpm bootstrap in publication workflows
+- Refactor sandbox integration scenarios into the eval harness (#1161)
+
 ## [1.0.2] — 2026-10-29
 
 - Read archived Runtime bytes without changing logical run identity (#1159)

package/dist/builder-eval/index.d.ts

@@ -1,7 +1,7 @@
 import { f as Run, o as FailureClass } from "../schema-CkIvnmLU.js";
 import { s as TraceStore } from "../store-CMs5wGXP.js";
 import { o as TraceEmitter } from "../emitter-mAXC5PSN.js";
-import { n as SandboxDriver, r as SandboxHarnessResult, t as HarnessConfig } from "../sandbox-harness-B3Kx1fOA.js";
+import { c as SandboxHarnessResult, n as HarnessConfig, s as SandboxDriver } from "../sandbox-harness--VUvMji7.js";
 interface TestGradedScenario {
   id: string;
   description?: string;

package/dist/builder-eval/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","names":[],"sources":["../../src/test-graded-scenario.ts","../../src/builder-eval/builder-session.ts","../../src/builder-eval/three-layer-eval.ts","../../src/builder-eval/correlation.ts","../../src/builder-eval/project-registry.ts"],"mappings":";;;;UAmBiB;EACf;EACA;EACA,SAAS;;EAET;;EAEA;;EAEA,OAAO;;UAUQ;EACf;EACA,UAAU;EACV,SAAS;EACT;EACA;EACA,eAAe;;iBCjBA;EACf;EACA;;EAEA,OAAO;;iBAGQ;EACf,SAAS;EACT,SAAS;;EAET;;iBAGe;EACf,UAAU;;EAEV,SAAS;;qBAGE;UACH;UACA;WACC;WACA;UACD;UACA;UACA;EAEI,YAAA,OAAO,YAAY,MAAM,oBAAoB,SAAS;;EAS5D,UAAU,sBAAwC;;MAYpD,WAAW;;;;;EAST,KAAK,SAAS,cAAc;IAAU;IAAe,QAAQ;;;;;;;;EA2B7D,eAAe,SAAS,wBAAwB,QAAQ;;;EAqBxD,gBAAgB,eAAe,qBAAqB;;EAcpD,QAAQ;IAAW;IAAe;IAAgB;MAAmB;;;;;;;;EAerE,gBAAgB,qBAAqB,QAAQ;;;;;;EAsB7C,iBAAiB;IACrB;IACA;IACA;IACA;MACE;MAqBA;MAGA;;;;;;;wBAUgB,qBACpB,OAAO,YACP,oBACC;EACD;EACA,UAAU;EACV,iBAAiB;EACjB,eAAe;EACf,oBAAoB;;YC5MV;iBAEK;EACf;;;;;;;EAOA,MAAM;EACN;;EAEA;EACA;;EAEA;EACA;;EAEA;EACA;;;;;;EAMA;;wBAGoB,aACpB,OAAO,YACP,oBACC,QAAQ;;wBAuCW,iBAAiB,OAAO,aAAa,QAAQ;iBC/ElD;EACf;EACA;EACA;;iBAGe;;EAEf,cAAc;EACd,gBAAgB;EAChB,iBAAiB;;EAEjB;;wBAGc,gBAAgB,SAAS,4BAA4B;iBCpBpD;EACf;EACA;EACA;EACA;EACA;EACA;EACA;IAAkB;IAAe;;;iBAGlB;EACf;EACA;EACA;EACA;EACA;EACA,QAAQ;EACR,UAAU;;EAEV;EACA;EACA;EACA;;iBAGe;EACf,KAAK;EACL;;qBAGW;UACS;EAAA,YAAA,OAAO;EAErB,gBAAgB,QAAQ;EAgCxB,gBAAgB,oBAAoB,QAAQ;EAgB5C,aAAa,oBAAoB,QAAQ"}
+{"version":3,"file":"index.d.ts","names":[],"sources":["../../src/test-graded-scenario.ts","../../src/builder-eval/builder-session.ts","../../src/builder-eval/three-layer-eval.ts","../../src/builder-eval/correlation.ts","../../src/builder-eval/project-registry.ts"],"mappings":";;;;UAMiB;EACf;EACA;EACA,SAAS;;EAET;;EAEA;;EAEA,OAAO;;UAUQ;EACf;EACA,UAAU;EACV,SAAS;EACT;EACA;EACA,eAAe;;iBCJA;EACf;EACA;;EAEA,OAAO;;iBAGQ;EACf,SAAS;EACT,SAAS;;EAET;;iBAGe;EACf,UAAU;;EAEV,SAAS;;qBAGE;UACH;UACA;WACC;WACA;UACD;UACA;UACA;EAEI,YAAA,OAAO,YAAY,MAAM,oBAAoB,SAAS;;EAS5D,UAAU,sBAAwC;;MAYpD,WAAW;;;;;EAST,KAAK,SAAS,cAAc;IAAU;IAAe,QAAQ;;;;;;;;EA2B7D,eAAe,SAAS,wBAAwB,QAAQ;;;EAqBxD,gBAAgB,eAAe,qBAAqB;;EAcpD,QAAQ;IAAW;IAAe;IAAgB;MAAmB;;;;;;;;EAerE,gBAAgB,qBAAqB,QAAQ;;;;;;EAsB7C,iBAAiB;IACrB;IACA;IACA;IACA;MACE;MAqBA;MAGA;;;;;;;wBAUgB,qBACpB,OAAO,YACP,oBACC;EACD;EACA,UAAU;EACV,iBAAiB;EACjB,eAAe;EACf,oBAAoB;;YC5MV;iBAEK;EACf;;;;;;;EAOA,MAAM;EACN;;EAEA;EACA;;EAEA;EACA;;EAEA;EACA;;;;;;EAMA;;wBAGoB,aACpB,OAAO,YACP,oBACC,QAAQ;;wBAuCW,iBAAiB,OAAO,aAAa,QAAQ;iBC/ElD;EACf;EACA;EACA;;iBAGe;;EAEf,cAAc;EACd,gBAAgB;EAChB,iBAAiB;;EAEjB;;wBAGc,gBAAgB,SAAS,4BAA4B;iBCpBpD;EACf;EACA;EACA;EACA;EACA;EACA;EACA;IAAkB;IAAe;;;iBAGlB;EACf;EACA;EACA;EACA;EACA;EACA,QAAQ;EACR,UAAU;;EAEV;EACA;EACA;EACA;;iBAGe;EACf,KAAK;EACL;;qBAGW;UACS;EAAA,YAAA,OAAO;EAErB,gBAAgB,QAAQ;EAgCxB,gBAAgB,oBAAoB,QAAQ;EAgB5C,aAAa,oBAAoB,QAAQ"}

package/dist/builder-eval/index.js

@@ -2,131 +2,7 @@ import { i as pearsonR, o as spearmanR } from "../descriptive-BTzLL-aj.js";
 import { i as judgeSpans } from "../query-CGXzO9d4.js";
 import { t as newRecordId } from "../record-id-CWZXFh58.js";
 import { t as TraceEmitter } from "../emitter-DhQ65mza.js";
-import { n as runBoundedProcess } from "../bounded-process-BZrWXIKW.js";
-// src/sandbox-harness.ts
-/**
-* SandboxHarness — executes a scenario in an isolated environment and
-* emits a rich SandboxSpan into the trace.
-*
-* Two built-in drivers:
-*   - `SubprocessSandboxDriver` — runs the phase in a local cwd with env
-*     vars, through `runBoundedProcess` (deadline, process-group kill,
-*     output cap). Fast, no dependencies, fine for unit tests and CI gates.
-*   - `DockerSandboxDriver` — lifted from tangle-router's sandbox path;
-*     shells out to `docker run`. Stronger isolation, slower startup.
-*
-* Consumers implement `SandboxDriver` for custom backends (Firecracker,
-* Cloudflare sandbox product, etc.). The harness doesn't care which.
-*/
-var SubprocessSandboxDriver = class {
-	id = "subprocess";
-	defaultCwd;
-	defaultEnv;
-	constructor(options = {}) {
-		this.defaultCwd = options.cwd;
-		this.defaultEnv = options.env;
-	}
-	async exec(phase, command, config) {
-		const effectiveCwd = config.cwd ?? this.defaultCwd;
-		const effectiveEnv = {
-			...process.env,
-			...this.defaultEnv ?? {},
-			...config.env ?? {}
-		};
-		const result = await runBoundedProcess({
-			command,
-			shell: true,
-			cwd: effectiveCwd,
-			env: effectiveEnv,
-			envMode: "replace",
-			timeoutMs: config.timeoutMs ?? 6e5,
-			maxOutputBytes: config.maxOutputBytes ?? 16777216
-		});
-		const parsed = !result.killedByTimeout && phase === "test" && config.testParser ? config.testParser.parse(result.stdout, result.stderr, result.exitCode) : void 0;
-		return {
-			phase,
-			exitCode: result.exitCode,
-			stdout: result.stdout,
-			stderr: result.runnerError ? result.stderr + result.runnerError : result.stderr,
-			wallMs: result.wallMs,
-			testsTotal: parsed?.testsTotal,
-			testsPassed: parsed?.testsPassed,
-			killedByTimeout: result.killedByTimeout || void 0,
-			outputTruncated: result.outputTruncated || void 0
-		};
-	}
-};
-var SandboxHarness = class {
-	driver;
-	constructor(driver = new SubprocessSandboxDriver()) {
-		this.driver = driver;
-	}
-	async run(config, emitter) {
-		const handle = await emitter.sandbox({
-			name: `sandbox(${this.driver.id})`,
-			image: config.image,
-			command: [
-				config.setupCommand,
-				config.runCommand,
-				config.testCommand
-			].filter(Boolean).join(" && ")
-		});
-		const result = {
-			passed: false,
-			totalWallMs: 0,
-			score: 0
-		};
-		try {
-			if (config.setupCommand) {
-				result.setup = await this.driver.exec("setup", config.setupCommand, config);
-				result.totalWallMs += result.setup.wallMs;
-				if (result.setup.killedByTimeout || result.setup.exitCode !== 0) {
-					const reason = result.setup.killedByTimeout ? `setup timed out after ${result.setup.wallMs}ms` : `setup failed (exit ${result.setup.exitCode})`;
-					await handle.fail(reason, {
-						exitCode: result.setup.exitCode,
-						wallMs: result.totalWallMs
-					});
-					return result;
-				}
-			}
-			if (config.runCommand) {
-				result.run = await this.driver.exec("run", config.runCommand, config);
-				result.totalWallMs += result.run.wallMs;
-				if (result.run.killedByTimeout || result.run.exitCode !== 0) {
-					const reason = result.run.killedByTimeout ? `run timed out after ${result.run.wallMs}ms` : `run failed (exit ${result.run.exitCode})`;
-					await handle.fail(reason, {
-						exitCode: result.run.exitCode,
-						wallMs: result.totalWallMs
-					});
-					return result;
-				}
-			}
-			if (config.testCommand) {
-				result.test = await this.driver.exec("test", config.testCommand, config);
-				result.totalWallMs += result.test.wallMs;
-				const passed = !result.test.killedByTimeout && result.test.exitCode === 0;
-				result.passed = passed;
-				if (result.test.testsTotal !== void 0 && result.test.testsTotal > 0) result.score = (result.test.testsPassed ?? 0) / result.test.testsTotal;
-				else result.score = passed ? 1 : 0;
-				await handle.end({
-					exitCode: result.test.exitCode,
-					testsTotal: result.test.testsTotal,
-					testsPassed: result.test.testsPassed,
-					wallMs: result.totalWallMs,
-					status: passed ? "ok" : "error"
-				});
-			} else {
-				result.passed = true;
-				result.score = 1;
-				await handle.end({ wallMs: result.totalWallMs });
-			}
-		} catch (err) {
-			await handle.fail(err instanceof Error ? err : String(err));
-			throw err;
-		}
-		return result;
-	}
-};
+import { t as SandboxHarness } from "../sandbox-harness-BuacGaVn.js";
 // src/test-graded-scenario.ts
 async function runTestGradedScenario(scenario, store, options = {}) {
 	const emitter = new TraceEmitter(store);
@@ -138,12 +14,13 @@ async function runTestGradedScenario(scenario, store, options = {}) {
 		...options.provenance
 	});
 	const result = await new SandboxHarness(options.driver).run(scenario.harness, emitter);
-	const threshold = scenario.passThreshold ?? 1;
-	const pass = result.passed && result.score >= threshold;
-	const setupFailed = result.setup !== void 0 && result.setup.exitCode !== 0;
-	const runFailed = result.run !== void 0 && result.run.exitCode !== 0;
-	const testFailed = result.test !== void 0 && result.test.exitCode !== 0;
-	const failureClass = pass ? "success" : setupFailed || runFailed ? "sandbox_failure" : testFailed ? "format_drift" : "unknown";
+	const pass = result.passed && result.score >= (scenario.passThreshold ?? 1);
+	const failure = [
+		"setup",
+		"run",
+		"test"
+	].find((phase) => result[phase] && result[phase].exitCode !== 0);
+	const failureClass = pass ? "success" : failure === "test" ? "format_drift" : failure ? "sandbox_failure" : "unknown";
 	await emitter.endRun({
 		pass,
 		score: result.score,
@@ -160,13 +37,10 @@ async function runTestGradedScenario(scenario, store, options = {}) {
 	};
 }
 function reasonForFailure(result) {
-	if (result.setup && result.setup.exitCode !== 0) return `setup failed: exit ${result.setup.exitCode}`;
-	if (result.run && result.run.exitCode !== 0) return `run failed: exit ${result.run.exitCode}`;
-	if (result.test) {
-		if (result.test.testsTotal !== void 0) return `tests: ${result.test.testsPassed ?? 0}/${result.test.testsTotal}`;
-		return `test exit ${result.test.exitCode}`;
-	}
-	return "no test command";
+	const failed = ["setup", "run"].find((phase) => result[phase] && result[phase].exitCode !== 0);
+	if (failed) return `${failed} failed: exit ${result[failed].exitCode}`;
+	const test = result.test;
+	return !test ? "no test command" : test.testsTotal !== void 0 ? `tests: ${test.testsPassed ?? 0}/${test.testsTotal}` : `test exit ${test.exitCode}`;
 }
 // src/builder-eval/builder-session.ts
 /**

package/dist/builder-eval/index.js.map

@@ -1 +1 @@
-{"version":3,"file":"index.js","names":[],"sources":["../../src/sandbox-harness.ts","../../src/test-graded-scenario.ts","../../src/builder-eval/builder-session.ts","../../src/builder-eval/correlation.ts","../../src/builder-eval/project-registry.ts","../../src/builder-eval/three-layer-eval.ts"],"sourcesContent":["/**\n * SandboxHarness — executes a scenario in an isolated environment and\n * emits a rich SandboxSpan into the trace.\n *\n * Two built-in drivers:\n *   - `SubprocessSandboxDriver` — runs the phase in a local cwd with env\n *     vars, through `runBoundedProcess` (deadline, process-group kill,\n *     output cap). Fast, no dependencies, fine for unit tests and CI gates.\n *   - `DockerSandboxDriver` — lifted from tangle-router's sandbox path;\n *     shells out to `docker run`. Stronger isolation, slower startup.\n *\n * Consumers implement `SandboxDriver` for custom backends (Firecracker,\n * Cloudflare sandbox product, etc.). The harness doesn't care which.\n */\n\nimport { runBoundedProcess } from './bounded-process'\n\nimport type { TraceEmitter } from './trace/emitter'\nimport type { SandboxSpan } from './trace/schema'\n\nexport interface HarnessConfig {\n  /** Setup command (e.g. \"pnpm install\"). Non-zero exit fails the run. */\n  setupCommand?: string\n  /** Run command (e.g. \"pnpm build\"). */\n  runCommand?: string\n  /** Test command (e.g. \"pnpm test --run\"). Drives the test count + pass count. */\n  testCommand?: string\n  /** Absolute cwd for the subprocess driver. Ignored by docker driver. */\n  cwd?: string\n  /** Max wall-clock per phase in ms. Default 10 minutes. */\n  timeoutMs?: number\n  /**\n   * Cap on captured stdout+stderr bytes per phase. A runaway process can\n   * otherwise grow the in-memory buffer without bound. Once hit, further\n   * output is dropped and `outputTruncated` is set. Default 16 MiB.\n   */\n  maxOutputBytes?: number\n  /** Image for the docker driver. */\n  image?: string\n  /** Extra env vars (validated; shell-escaped). */\n  env?: Record<string, string>\n  /** Parser for the test output — maps stdout/stderr/exit code → pass count. */\n  testParser?: TestOutputParser\n}\n\nexport interface TestOutputParser {\n  id: string\n  parse(\n    stdout: string,\n    stderr: string,\n    exitCode: number,\n  ): { testsTotal: number; testsPassed: number } | undefined\n}\n\nexport interface SandboxResult {\n  phase: 'setup' | 'run' | 'test'\n  exitCode: number\n  stdout: string\n  stderr: string\n  wallMs: number\n  testsTotal?: number\n  testsPassed?: number\n  /**\n   * True when the process was killed because it exceeded `timeoutMs`. A\n   * SIGKILLed child can still close with exit code 0; callers MUST treat\n   * a timed-out phase as a hard failure regardless of `exitCode`, never\n   * as a pass. `undefined`/`false` means the process completed on its own.\n   */\n  killedByTimeout?: boolean\n  /**\n   * True when captured stdout/stderr hit `maxOutputBytes` and further\n   * output was dropped. The result is still returned (the process was\n   * not killed for this), but downstream parsers see truncated text.\n   */\n  outputTruncated?: boolean\n}\n\nexport interface SandboxDriver {\n  id: string\n  exec(\n    phase: SandboxResult['phase'],\n    command: string,\n    config: HarnessConfig,\n  ): Promise<SandboxResult>\n}\n\n// ── Parsers ──────────────────────────────────────────────────────────\n\n/** Vitest default summary line: \"Tests  X passed | Y failed\". */\nexport const vitestTestParser: TestOutputParser = {\n  id: 'vitest',\n  parse(stdout) {\n    const m = stdout.match(/Tests\\s+(\\d+)\\s+(passed|failed)(?:\\s*\\|\\s*(\\d+)\\s+(passed|failed))?/i)\n    if (!m) return undefined\n    let passed = 0\n    let failed = 0\n    const a = parseInt(m[1]!, 10)\n    const aLabel = m[2]!.toLowerCase()\n    if (aLabel === 'passed') passed += a\n    else failed += a\n    if (m[3] && m[4]) {\n      const b = parseInt(m[3], 10)\n      if (m[4].toLowerCase() === 'passed') passed += b\n      else failed += b\n    }\n    return { testsTotal: passed + failed, testsPassed: passed }\n  },\n}\n\n/** Pytest default: \"collected N items\" + \" X passed, Y failed\". */\nexport const pytestTestParser: TestOutputParser = {\n  id: 'pytest',\n  parse(stdout) {\n    const total = stdout.match(/collected\\s+(\\d+)\\s+items?/i)\n    const passed = stdout.match(/(\\d+)\\s+passed/)\n    if (!total || !passed) return undefined\n    return { testsTotal: parseInt(total[1]!, 10), testsPassed: parseInt(passed[1]!, 10) }\n  },\n}\n\n/** Jest: \"Tests: X passed, Y total\" (and optional failed). */\nexport const jestTestParser: TestOutputParser = {\n  id: 'jest',\n  parse(stdout) {\n    const m = stdout.match(/Tests:\\s+(?:(\\d+)\\s+failed[^,]*,\\s*)?(\\d+)\\s+passed,\\s+(\\d+)\\s+total/i)\n    if (!m) return undefined\n    return { testsTotal: parseInt(m[3]!, 10), testsPassed: parseInt(m[2]!, 10) }\n  },\n}\n\n/** Composite parser — tries a list of parsers in order. */\nexport function composeParsers(...parsers: TestOutputParser[]): TestOutputParser {\n  return {\n    id: parsers.map((p) => p.id).join('|'),\n    parse(stdout, stderr, exitCode) {\n      for (const p of parsers) {\n        const res = p.parse(stdout, stderr, exitCode)\n        if (res) return res\n      }\n      return undefined\n    },\n  }\n}\n\n// ── Drivers ──────────────────────────────────────────────────────────\n\ninterface SubprocessSandboxDriverOptions {\n  /**\n   * Default cwd for all `exec` calls. Used when the per-call `HarnessConfig`\n   * does not set its own `cwd`. Lets callers bind the driver to a working\n   * directory once instead of spreading cwd into every harness config —\n   * useful when the harness config is constructed far from the call site\n   * (e.g. starter-foundry's promoter passes a static HarnessConfig per\n   * family taxonomy but needs a per-run composed-scaffold cwd).\n   */\n  cwd?: string\n  /**\n   * Default env merged into every `exec` call's env (per-call `HarnessConfig.env`\n   * still wins on key collision). Same ergonomic rationale as `cwd` above.\n   */\n  env?: Record<string, string>\n}\n\nexport class SubprocessSandboxDriver implements SandboxDriver {\n  id = 'subprocess'\n  private defaultCwd?: string\n  private defaultEnv?: Record<string, string>\n\n  constructor(options: SubprocessSandboxDriverOptions = {}) {\n    this.defaultCwd = options.cwd\n    this.defaultEnv = options.env\n  }\n\n  async exec(\n    phase: SandboxResult['phase'],\n    command: string,\n    config: HarnessConfig,\n  ): Promise<SandboxResult> {\n    // Per-call config wins; fall back to constructor defaults. Honoring\n    // the constructor `cwd` keeps the subprocess from inheriting Node's\n    // cwd when only the constructor arg is supplied.\n    const effectiveCwd = config.cwd ?? this.defaultCwd\n    // The parent environment is merged here, so the runner is handed the\n    // complete environment and passes it through unchanged.\n    const effectiveEnv = { ...process.env, ...(this.defaultEnv ?? {}), ...(config.env ?? {}) }\n    const result = await runBoundedProcess({\n      command,\n      shell: true,\n      cwd: effectiveCwd,\n      env: effectiveEnv,\n      envMode: 'replace',\n      timeoutMs: config.timeoutMs ?? 10 * 60_000,\n      maxOutputBytes: config.maxOutputBytes ?? 16 * 1024 * 1024,\n    })\n    // A timed-out phase never reaches the parser: the output of a killed\n    // process is a fragment, and a fragment that happens to parse would\n    // report a pass for a run that never finished.\n    const parsed =\n      !result.killedByTimeout && phase === 'test' && config.testParser\n        ? config.testParser.parse(result.stdout, result.stderr, result.exitCode)\n        : undefined\n    return {\n      phase,\n      exitCode: result.exitCode,\n      stdout: result.stdout,\n      stderr: result.runnerError ? result.stderr + result.runnerError : result.stderr,\n      wallMs: result.wallMs,\n      testsTotal: parsed?.testsTotal,\n      testsPassed: parsed?.testsPassed,\n      killedByTimeout: result.killedByTimeout || undefined,\n      outputTruncated: result.outputTruncated || undefined,\n    }\n  }\n}\n\n\n\n\n\n// ── Harness orchestration ────────────────────────────────────────────\n\nexport interface SandboxHarnessResult {\n  passed: boolean\n  setup?: SandboxResult\n  run?: SandboxResult\n  test?: SandboxResult\n  totalWallMs: number\n  /** Final score — 0 when no tests; otherwise testsPassed/testsTotal. */\n  score: number\n}\n\nexport class SandboxHarness {\n  private driver: SandboxDriver\n  constructor(driver: SandboxDriver = new SubprocessSandboxDriver()) {\n    this.driver = driver\n  }\n\n  async run(config: HarnessConfig, emitter: TraceEmitter): Promise<SandboxHarnessResult> {\n    const handle = await emitter.sandbox({\n      name: `sandbox(${this.driver.id})`,\n      image: config.image,\n      command: [config.setupCommand, config.runCommand, config.testCommand]\n        .filter(Boolean)\n        .join(' && '),\n    })\n    const result: SandboxHarnessResult = { passed: false, totalWallMs: 0, score: 0 }\n    try {\n      if (config.setupCommand) {\n        result.setup = await this.driver.exec('setup', config.setupCommand, config)\n        result.totalWallMs += result.setup.wallMs\n        if (result.setup.killedByTimeout || result.setup.exitCode !== 0) {\n          const reason = result.setup.killedByTimeout\n            ? `setup timed out after ${result.setup.wallMs}ms`\n            : `setup failed (exit ${result.setup.exitCode})`\n          await handle.fail(reason, {\n            exitCode: result.setup.exitCode,\n            wallMs: result.totalWallMs,\n          } as Partial<SandboxSpan>)\n          return result\n        }\n      }\n      if (config.runCommand) {\n        result.run = await this.driver.exec('run', config.runCommand, config)\n        result.totalWallMs += result.run.wallMs\n        if (result.run.killedByTimeout || result.run.exitCode !== 0) {\n          const reason = result.run.killedByTimeout\n            ? `run timed out after ${result.run.wallMs}ms`\n            : `run failed (exit ${result.run.exitCode})`\n          await handle.fail(reason, {\n            exitCode: result.run.exitCode,\n            wallMs: result.totalWallMs,\n          } as Partial<SandboxSpan>)\n          return result\n        }\n      }\n      if (config.testCommand) {\n        result.test = await this.driver.exec('test', config.testCommand, config)\n        result.totalWallMs += result.test.wallMs\n        // A timed-out test phase is a hard fail regardless of the exit code\n        // a SIGKILLed child happens to report — never let it score as a pass.\n        const passed = !result.test.killedByTimeout && result.test.exitCode === 0\n        result.passed = passed\n        if (result.test.testsTotal !== undefined && result.test.testsTotal > 0) {\n          result.score = (result.test.testsPassed ?? 0) / result.test.testsTotal\n        } else {\n          result.score = passed ? 1 : 0\n        }\n        await handle.end({\n          exitCode: result.test.exitCode,\n          testsTotal: result.test.testsTotal,\n          testsPassed: result.test.testsPassed,\n          wallMs: result.totalWallMs,\n          status: passed ? 'ok' : 'error',\n        } as Partial<SandboxSpan>)\n      } else {\n        result.passed = true\n        result.score = 1\n        await handle.end({ wallMs: result.totalWallMs } as Partial<SandboxSpan>)\n      }\n    } catch (err) {\n      await handle.fail(err instanceof Error ? err : String(err))\n      throw err\n    }\n    return result\n  }\n}\n","/**\n * TestGradedScenario — a scenario whose score comes from a test suite.\n *\n * This is the SWE-bench pattern generalized. The scenario ships:\n *   - fixture data (setup instructions)\n *   - a test command the harness runs\n *   - optional assertion overrides\n *\n * The runner emits a run, delegates to SandboxHarness, records the\n * outcome, and returns a structured verdict. Consumers bind their own\n * agent execution to this contract.\n */\n\nimport type { HarnessConfig, SandboxDriver, SandboxHarnessResult } from './sandbox-harness'\nimport { SandboxHarness } from './sandbox-harness'\nimport { TraceEmitter } from './trace/emitter'\nimport type { FailureClass, Run } from './trace/schema'\nimport type { TraceStore } from './trace/store'\n\nexport interface TestGradedScenario {\n  id: string\n  description?: string\n  harness: HarnessConfig\n  /** Optional pass threshold in 0..1 (default 1.0 = all tests must pass). */\n  passThreshold?: number\n  /** Provenance for dataset tracking. */\n  datasetVersion?: string\n  /** Free-form tags (difficulty, category, etc.). */\n  tags?: Record<string, string>\n}\n\ninterface TestGradedRunOptions {\n  variantId?: string\n  driver?: SandboxDriver\n  /** Metadata recorded on the Run (codeSha, promptSha, modelFingerprint, seed). */\n  provenance?: Pick<Run, 'codeSha' | 'promptSha' | 'modelFingerprint' | 'seed' | 'envFingerprint'>\n}\n\nexport interface TestGradedRunResult {\n  runId: string\n  scenario: TestGradedScenario\n  harness: SandboxHarnessResult\n  pass: boolean\n  score: number\n  failureClass?: FailureClass\n}\n\nexport async function runTestGradedScenario(\n  scenario: TestGradedScenario,\n  store: TraceStore,\n  options: TestGradedRunOptions = {},\n): Promise<TestGradedRunResult> {\n  const emitter = new TraceEmitter(store)\n  await emitter.startRun({\n    scenarioId: scenario.id,\n    variantId: options.variantId,\n    datasetVersion: scenario.datasetVersion,\n    tags: scenario.tags,\n    ...options.provenance,\n  })\n  const harness = new SandboxHarness(options.driver)\n  const result = await harness.run(scenario.harness, emitter)\n  const threshold = scenario.passThreshold ?? 1.0\n  const pass = result.passed && result.score >= threshold\n  const setupFailed = result.setup !== undefined && result.setup.exitCode !== 0\n  const runFailed = result.run !== undefined && result.run.exitCode !== 0\n  const testFailed = result.test !== undefined && result.test.exitCode !== 0\n  const failureClass: FailureClass | undefined = pass\n    ? 'success'\n    : setupFailed || runFailed\n      ? 'sandbox_failure'\n      : testFailed\n        ? 'format_drift'\n        : 'unknown'\n  await emitter.endRun({\n    pass,\n    score: result.score,\n    failureClass,\n    notes: pass ? undefined : reasonForFailure(result),\n  })\n  return {\n    runId: emitter.runId,\n    scenario,\n    harness: result,\n    pass,\n    score: result.score,\n    failureClass,\n  }\n}\n\nfunction reasonForFailure(result: SandboxHarnessResult): string {\n  if (result.setup && result.setup.exitCode !== 0)\n    return `setup failed: exit ${result.setup.exitCode}`\n  if (result.run && result.run.exitCode !== 0) return `run failed: exit ${result.run.exitCode}`\n  if (result.test) {\n    if (result.test.testsTotal !== undefined) {\n      return `tests: ${result.test.testsPassed ?? 0}/${result.test.testsTotal}`\n    }\n    return `test exit ${result.test.exitCode}`\n  }\n  return 'no test command'\n}\n","/**\n * BuilderSession — ties a builder-of-builders workflow together.\n *\n * Models agent-builder's shape: Project → Chat → Edit → Ship → App →\n * AppAgent. Each layer is a Run (linked via parentRunId). The\n * framework-enforced invariants:\n *\n *   - One Project → many Chats; chatId scopes runs within a project.\n *   - One Chat = one builder Run with `layer='builder'`.\n *   - One Ship = one child Run with `layer='app-build'` + SandboxHarness.\n *   - One AppScenario = one grandchild Run with `layer='app-runtime'`.\n *\n * Consumers obtain a BuilderSession, call `startChat`, drive the\n * builder agent (emitting spans), and call `ship` / `runAppScenario`\n * as the workflow progresses. The session reconstructs itself from\n * trace data via `resume(store, projectId)`.\n */\n\nimport { newRecordId } from '../record-id'\nimport type { HarnessConfig, SandboxDriver, SandboxHarnessResult } from '../sandbox-harness'\nimport { SandboxHarness } from '../sandbox-harness'\nimport type { TestGradedRunResult, TestGradedScenario } from '../test-graded-scenario'\nimport { runTestGradedScenario } from '../test-graded-scenario'\nimport { TraceEmitter } from '../trace/emitter'\nimport type { Run } from '../trace/schema'\nimport type { TraceStore } from '../trace/store'\n\nexport interface BuilderSessionInit {\n  projectId: string\n  chatId?: string\n  /** Free-form: user's task description, project name, etc. Stored on the builder Run. */\n  tags?: Record<string, string>\n}\n\nexport interface ShipOptions {\n  harness: HarnessConfig\n  driver?: SandboxDriver\n  /** scenarioId of this app-build run. Defaults to `${projectId}/build`. */\n  scenarioId?: string\n}\n\nexport interface RunAppScenarioOptions {\n  scenario: TestGradedScenario\n  /** Harness driver override; defaults to the one the session was created with. */\n  driver?: SandboxDriver\n}\n\nexport class BuilderSession {\n  private store: TraceStore\n  private builderEmitter: TraceEmitter\n  readonly projectId: string\n  readonly chatId: string\n  private builderRunId?: string\n  private lastBuildRunId?: string\n  private defaultDriver?: SandboxDriver\n\n  constructor(store: TraceStore, init: BuilderSessionInit, driver?: SandboxDriver) {\n    this.store = store\n    this.projectId = init.projectId\n    this.chatId = init.chatId ?? newRecordId()\n    this.defaultDriver = driver\n    this.builderEmitter = new TraceEmitter(store)\n  }\n\n  /** Start the builder (L0) run for this chat. Returns the runId. */\n  async startChat(scenarioId = `${this.projectId}/chat`): Promise<string> {\n    await this.builderEmitter.startRun({\n      scenarioId,\n      projectId: this.projectId,\n      chatId: this.chatId,\n      layer: 'builder',\n    })\n    this.builderRunId = this.builderEmitter.runId\n    return this.builderRunId\n  }\n\n  /** The emitter for builder-level spans (edits, LLM calls, tool invocations). */\n  get emitter(): TraceEmitter {\n    if (!this.builderRunId) throw new Error('BuilderSession.emitter: call startChat() first')\n    return this.builderEmitter\n  }\n\n  /**\n   * Ship the project's generated app: run the sandbox harness as a child\n   * Run (`layer='app-build'`). Returns the build result + runId.\n   */\n  async ship(options: ShipOptions): Promise<{ runId: string; result: SandboxHarnessResult }> {\n    if (!this.builderRunId) throw new Error('BuilderSession.ship: call startChat() first')\n    const buildEmitter = new TraceEmitter(this.store)\n    await buildEmitter.startRun({\n      scenarioId: options.scenarioId ?? `${this.projectId}/build`,\n      projectId: this.projectId,\n      chatId: this.chatId,\n      parentRunId: this.builderRunId,\n      layer: 'app-build',\n    })\n    const harness = new SandboxHarness(options.driver ?? this.defaultDriver)\n    const result = await harness.run(options.harness, buildEmitter)\n    await buildEmitter.endRun({\n      pass: result.passed,\n      score: result.score,\n      failureClass: result.passed ? 'success' : 'sandbox_failure',\n    })\n    this.lastBuildRunId = buildEmitter.runId\n    return { runId: buildEmitter.runId, result }\n  }\n\n  /**\n   * Run a domain scenario against the just-built app as a grandchild Run\n   * (`layer='app-runtime'`). The `ship` call must precede this so the\n   * parent is set correctly; if no build exists yet the session attaches\n   * directly to the builder run (useful for prototypes).\n   */\n  async runAppScenario(options: RunAppScenarioOptions): Promise<TestGradedRunResult> {\n    const parentRunId = this.lastBuildRunId ?? this.builderRunId\n    if (!parentRunId)\n      throw new Error('BuilderSession.runAppScenario: call startChat() + ship() first')\n    const { scenario, driver } = options\n    const result = await runTestGradedScenario(scenario, this.store, {\n      driver: driver ?? this.defaultDriver,\n      provenance: { codeSha: undefined, promptSha: undefined, modelFingerprint: undefined },\n    })\n    // Attach to the parent chain by updating the stored Run in place.\n    await this.store.updateRun(result.runId, {\n      parentRunId,\n      projectId: this.projectId,\n      chatId: this.chatId,\n      layer: 'app-runtime',\n    })\n    return result\n  }\n\n  /** Record an end-of-chat meta score (judge verdict on whether the builder\n   *  served the user's intent). Accepts a numeric score + optional rationale. */\n  async recordMetaScore(score: number, rationale?: string): Promise<void> {\n    if (!this.builderRunId)\n      throw new Error('BuilderSession.recordMetaScore: call startChat() first')\n    await this.builderEmitter.recordJudge({\n      judgeId: 'builder-meta',\n      targetSpanId: this.builderRunId, // attach to the builder run itself\n      dimension: 'user_intent_satisfaction',\n      score,\n      rationale,\n      name: 'builder-meta',\n    })\n  }\n\n  /** Close the builder Run with a final outcome. */\n  async endChat(outcome: { pass: boolean; score?: number; notes?: string }): Promise<void> {\n    await this.builderEmitter.endRun({\n      pass: outcome.pass,\n      score: outcome.score,\n      notes: outcome.notes,\n    })\n  }\n\n  /**\n   * Inline app-runtime run — for cases where the \"scenario\" isn't a\n   * SWE-bench-style test suite but a live agent interaction (LLM chat,\n   * domain flow). Returns an emitter bound to a fresh Run in the\n   * `app-runtime` layer; caller emits spans inside and calls\n   * `.endRun()` with the final verdict.\n   */\n  async startAppRuntime(scenarioId: string): Promise<TraceEmitter> {\n    const parentRunId = this.lastBuildRunId ?? this.builderRunId\n    if (!parentRunId)\n      throw new Error(\n        'BuilderSession.startAppRuntime: call startChat() + (optionally) ship() first',\n      )\n    const emitter = new TraceEmitter(this.store)\n    await emitter.startRun({\n      scenarioId,\n      projectId: this.projectId,\n      chatId: this.chatId,\n      parentRunId,\n      layer: 'app-runtime',\n    })\n    return emitter\n  }\n\n  /**\n   * Lightweight \"ship marker\" — record an app-build Run with a caller-\n   * provided verdict. Use when there isn't a sandbox harness to run but\n   * you still want to mark the build state at publish time.\n   */\n  async recordShipMarker(args: {\n    pass: boolean\n    score: number\n    scenarioId?: string\n    notes?: string\n  }): Promise<string> {\n    if (!this.builderRunId)\n      throw new Error('BuilderSession.recordShipMarker: call startChat() first')\n    const emitter = new TraceEmitter(this.store)\n    await emitter.startRun({\n      scenarioId: args.scenarioId ?? `${this.projectId}/ship`,\n      projectId: this.projectId,\n      chatId: this.chatId,\n      parentRunId: this.builderRunId,\n      layer: 'app-build',\n    })\n    await emitter.endRun({\n      pass: args.pass,\n      score: args.score,\n      failureClass: args.pass ? 'success' : 'sandbox_failure',\n      notes: args.notes,\n    })\n    this.lastBuildRunId = emitter.runId\n    return emitter.runId\n  }\n\n  get lastBuildRunIdValue(): string | undefined {\n    return this.lastBuildRunId\n  }\n  get builderRunIdValue(): string | undefined {\n    return this.builderRunId\n  }\n}\n\n/**\n * Reconstruct the most recent BuilderSession state for a given project —\n * returns { builderRunId, lastBuildRunId, chatRuns }. For chat-first UIs\n * this is how a resumed session finds its place in the edit history.\n */\nexport async function resumeBuilderSession(\n  store: TraceStore,\n  projectId: string,\n): Promise<{\n  projectId: string\n  chatRuns: Run[]\n  lastBuilderRun?: Run\n  lastBuildRun?: Run\n  lastAppRuntimeRuns: Run[]\n}> {\n  const runs = await store.listRuns({ projectId })\n  const chatRuns = runs\n    .filter((r) => r.layer === 'builder')\n    .sort((a, b) => b.startedAt - a.startedAt)\n  const buildRuns = runs\n    .filter((r) => r.layer === 'app-build')\n    .sort((a, b) => b.startedAt - a.startedAt)\n  const appRuntimeRuns = runs\n    .filter((r) => r.layer === 'app-runtime')\n    .sort((a, b) => b.startedAt - a.startedAt)\n  return {\n    projectId,\n    chatRuns,\n    lastBuilderRun: chatRuns[0],\n    lastBuildRun: buildRuns[0],\n    lastAppRuntimeRuns: appRuntimeRuns,\n  }\n}\n","/**\n * Meta-eval correlation — the highest-leverage signal in the framework.\n *\n * Given a corpus of three-layer project reports, compute how well each\n * pair of layers correlates. The question we care about most:\n *\n *   Does `metaScore` (what the builder thinks it did) predict\n *   `runtimeScore` (what the user actually gets)?\n *\n * If r < ~0.4, the builder's self-scoring is broken — it's optimizing\n * for something other than real-world success. If r > 0.7, meta_score\n * is a usable proxy and can drive CI gates cheaply.\n *\n * Non-parametric rank correlation (Spearman) is also reported because\n * meta scores are often ordinal-ish.\n */\n\nimport { pearsonR, spearmanR } from '../statistics'\nimport type { ThreeLayerProjectReport } from './three-layer-eval'\n\nexport interface LayerCorrelation {\n  n: number\n  pearson: number\n  spearman: number\n}\n\nexport interface CorrelationReport {\n  /** Pairs present in the corpus (layers with ≥ 2 matched data points). */\n  metaVsBuild?: LayerCorrelation\n  metaVsRuntime?: LayerCorrelation\n  buildVsRuntime?: LayerCorrelation\n  /** Number of complete projects (all 3 scores present). */\n  completeProjects: number\n}\n\nexport function correlateLayers(reports: ThreeLayerProjectReport[]): CorrelationReport {\n  const completeProjects = reports.filter((r) => r.complete).length\n  return {\n    metaVsBuild: pairwise(\n      reports,\n      (r) => r.metaScore,\n      (r) => r.buildScore,\n    ),\n    metaVsRuntime: pairwise(\n      reports,\n      (r) => r.metaScore,\n      (r) => r.runtimeScore,\n    ),\n    buildVsRuntime: pairwise(\n      reports,\n      (r) => r.buildScore,\n      (r) => r.runtimeScore,\n    ),\n    completeProjects,\n  }\n}\n\nfunction pairwise(\n  reports: ThreeLayerProjectReport[],\n  a: (r: ThreeLayerProjectReport) => number | null,\n  b: (r: ThreeLayerProjectReport) => number | null,\n): LayerCorrelation | undefined {\n  const xs: number[] = []\n  const ys: number[] = []\n  for (const r of reports) {\n    const x = a(r)\n    const y = b(r)\n    if (x !== null && y !== null && Number.isFinite(x) && Number.isFinite(y)) {\n      xs.push(x)\n      ys.push(y)\n    }\n  }\n  if (xs.length < 2) return undefined\n  return {\n    n: xs.length,\n    pearson: pearsonR(xs, ys),\n    spearman: spearmanR(xs, ys),\n  }\n}\n","/**\n * ProjectRegistry — project-level aggregation over the trace corpus.\n *\n * Thin reader over TraceStore that answers the questions a chat-first,\n * resumable UI needs:\n *   - listProjects() → project IDs with latest activity\n *   - projectTimeline(id) → chats + builds + runtime runs, chronological\n *   - projectChats(id) → chat-level summaries (turn count, outcome)\n *\n * All queries are pure reads; no state duplication.\n */\n\nimport type { Run } from '../trace/schema'\nimport type { TraceStore } from '../trace/store'\n\nexport interface ProjectSummary {\n  projectId: string\n  chatCount: number\n  buildCount: number\n  appRuntimeCount: number\n  lastActivityAt: number\n  latestChatId?: string\n  latestOutcome?: { pass: boolean; score?: number }\n}\n\nexport interface ChatSummary {\n  chatId: string\n  projectId: string\n  builderRunId: string\n  startedAt: number\n  endedAt?: number\n  status: Run['status']\n  outcome?: Run['outcome']\n  /** Counts of spans emitted during the chat. */\n  llmTurns?: number\n  toolCalls?: number\n  buildRunId?: string\n  appRuntimeRunIds: string[]\n}\n\nexport interface ProjectTimelineEntry {\n  run: Run\n  layerBucket: 'chat' | 'build' | 'runtime' | 'other'\n}\n\nexport class ProjectRegistry {\n  constructor(private store: TraceStore) {}\n\n  async listProjects(): Promise<ProjectSummary[]> {\n    const runs = await this.store.listRuns()\n    const byProject = new Map<string, Run[]>()\n    for (const r of runs) {\n      if (!r.projectId) continue\n      const arr = byProject.get(r.projectId) ?? []\n      arr.push(r)\n      byProject.set(r.projectId, arr)\n    }\n    const summaries: ProjectSummary[] = []\n    for (const [projectId, projectRuns] of byProject) {\n      const sorted = projectRuns.slice().sort((a, b) => b.startedAt - a.startedAt)\n      const chats = projectRuns.filter((r) => r.layer === 'builder')\n      const builds = projectRuns.filter((r) => r.layer === 'app-build')\n      const runtimes = projectRuns.filter((r) => r.layer === 'app-runtime')\n      const latest = sorted[0]\n      if (!latest) continue\n      summaries.push({\n        projectId,\n        chatCount: chats.length,\n        buildCount: builds.length,\n        appRuntimeCount: runtimes.length,\n        lastActivityAt: latest.startedAt,\n        latestChatId: chats[0]?.chatId,\n        latestOutcome: latest.outcome\n          ? { pass: latest.outcome.pass ?? false, score: latest.outcome.score }\n          : undefined,\n      })\n    }\n    return summaries.sort((a, b) => b.lastActivityAt - a.lastActivityAt)\n  }\n\n  async projectTimeline(projectId: string): Promise<ProjectTimelineEntry[]> {\n    const runs = await this.store.listRuns({ projectId })\n    const ordered = runs.slice().sort((a, b) => a.startedAt - b.startedAt)\n    return ordered.map((run) => ({\n      run,\n      layerBucket:\n        run.layer === 'builder'\n          ? 'chat'\n          : run.layer === 'app-build'\n            ? 'build'\n            : run.layer === 'app-runtime'\n              ? 'runtime'\n              : 'other',\n    }))\n  }\n\n  async projectChats(projectId: string): Promise<ChatSummary[]> {\n    const builderRuns = (await this.store.listRuns({ projectId, layer: 'builder' })).sort(\n      (a, b) => b.startedAt - a.startedAt,\n    )\n    const childrenFor = async (runId: string) => this.store.listRuns({ parentRunId: runId })\n    const out: ChatSummary[] = []\n    for (const run of builderRuns) {\n      const spans = await this.store.spans({ runId: run.runId })\n      const children = await childrenFor(run.runId)\n      const build = children.find((c) => c.layer === 'app-build')\n      const runtime: string[] = []\n      // Runtime runs can be grandchildren (attached to the build) or siblings\n      // when shipped skipped.\n      if (build) {\n        const grands = await childrenFor(build.runId)\n        for (const g of grands) if (g.layer === 'app-runtime') runtime.push(g.runId)\n      }\n      for (const c of children) if (c.layer === 'app-runtime') runtime.push(c.runId)\n      out.push({\n        chatId: run.chatId ?? run.runId,\n        projectId,\n        builderRunId: run.runId,\n        startedAt: run.startedAt,\n        endedAt: run.endedAt,\n        status: run.status,\n        outcome: run.outcome,\n        llmTurns: spans.filter((s) => s.kind === 'llm').length,\n        toolCalls: spans.filter((s) => s.kind === 'tool').length,\n        buildRunId: build?.runId,\n        appRuntimeRunIds: runtime,\n      })\n    }\n    return out\n  }\n}\n","/**\n * Three-layer evaluation — the canonical scoring breakdown for\n * builder-of-builders workflows.\n *\n *   meta_score:    did the builder understand + satisfy user intent?\n *                  (judge verdict attached to the builder run)\n *   build_score:   did the generated scaffold build + pass its own tests?\n *                  (outcome.score on the app-build child run)\n *   runtime_score: did the generated agent pass its domain scenarios?\n *                  (mean outcome.score over app-runtime grandchild runs)\n *\n * Returns a structured report per project. The cross-layer correlation\n * is the highest-leverage signal the framework computes — if\n * meta_score doesn't predict runtime_score, the builder's self-scoring\n * is broken.\n *\n * Scaffold-only mode: when a project has no `app-runtime` runs (e.g. a\n * scaffold-builder eval that grades compose + build without driving a\n * runtime scenario), `kind` is `'scaffold-only'` and `complete` measures\n * meta + build only. Consumers can tell the two apart without having to\n * interpret null-runtime as either \"not yet computed\" or \"N/A for this\n * project shape\".\n */\n\nimport { judgeSpans } from '../trace/query'\nimport type { Run } from '../trace/schema'\nimport type { TraceStore } from '../trace/store'\n\nexport type ProjectKind = 'full' | 'scaffold-only'\n\nexport interface ThreeLayerProjectReport {\n  projectId: string\n  /**\n   * `'full'` when the project has at least one `app-runtime` run;\n   * `'scaffold-only'` when it only has meta + build layers. Lets\n   * downstream consumers treat a null runtime score as expected\n   * (scaffold-only) vs. missing (full, pipeline broke).\n   */\n  kind: ProjectKind\n  builderRunId?: string\n  /** Judge-verdict score on the builder run (0..1 after normalization). */\n  metaScore: number | null\n  buildRunId?: string\n  /** 0..1 from the sandbox harness (testsPassed / testsTotal). */\n  buildScore: number | null\n  appRuntimeRunIds: string[]\n  /** Mean of outcome.score over app-runtime runs, 0..1. Always null in scaffold-only mode. */\n  runtimeScore: number | null\n  runtimePassRate: number | null\n  /**\n   * Layer-aware completeness:\n   *   - `kind='full'`: all three layers scored\n   *   - `kind='scaffold-only'`: meta + build scored (runtime not applicable)\n   */\n  complete: boolean\n}\n\nexport async function scoreProject(\n  store: TraceStore,\n  projectId: string,\n): Promise<ThreeLayerProjectReport> {\n  const allRuns = await store.listRuns({ projectId })\n  const builder = latestByLayer(allRuns, 'builder')\n  const build = latestByLayer(allRuns, 'app-build')\n  const runtime = allRuns.filter((r) => r.layer === 'app-runtime')\n\n  const metaScore = builder ? await extractMetaScore(store, builder.runId) : null\n  const buildScore = build?.outcome?.score ?? null\n  const runtimeScores = runtime\n    .map((r) => r.outcome?.score)\n    .filter((s): s is number => typeof s === 'number')\n  const runtimeScore =\n    runtimeScores.length > 0\n      ? runtimeScores.reduce((a, b) => a + b, 0) / runtimeScores.length\n      : null\n  const runtimePassed = runtime.filter((r) => r.outcome?.pass === true).length\n  const runtimePassRate = runtime.length > 0 ? runtimePassed / runtime.length : null\n\n  const kind: ProjectKind = runtime.length === 0 ? 'scaffold-only' : 'full'\n  const complete =\n    kind === 'scaffold-only'\n      ? metaScore !== null && buildScore !== null\n      : metaScore !== null && buildScore !== null && runtimeScore !== null\n\n  return {\n    projectId,\n    kind,\n    builderRunId: builder?.runId,\n    metaScore,\n    buildRunId: build?.runId,\n    buildScore,\n    appRuntimeRunIds: runtime.map((r) => r.runId),\n    runtimeScore,\n    runtimePassRate,\n    complete,\n  }\n}\n\n/** Aggregate scoring across every project in a corpus. */\nexport async function scoreAllProjects(store: TraceStore): Promise<ThreeLayerProjectReport[]> {\n  const runs = await store.listRuns()\n  const projectIds = [...new Set(runs.map((r) => r.projectId).filter((p): p is string => !!p))]\n  return Promise.all(projectIds.map((p) => scoreProject(store, p)))\n}\n\nfunction latestByLayer(runs: Run[], layer: Run['layer']): Run | undefined {\n  const filtered = runs.filter((r) => r.layer === layer).sort((a, b) => b.startedAt - a.startedAt)\n  return filtered[0]\n}\n\nasync function extractMetaScore(store: TraceStore, builderRunId: string): Promise<number | null> {\n  const js = await judgeSpans(store, builderRunId)\n  const meta = js.find(\n    (s) => s.judgeId === 'builder-meta' && s.dimension === 'user_intent_satisfaction',\n  )\n  if (!meta) return null\n  // Normalize score to 0..1. Accept 0-1 natively; 0-10 scale is also common.\n  if (meta.score >= 0 && meta.score <= 1) return meta.score\n  if (meta.score >= 0 && meta.score <= 10) return meta.score / 10\n  return null\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;AAmKA,IAAa,0BAAb,MAA8D;CAC5D,KAAK;CACL;CACA;CAEA,YAAY,UAA0C,CAAC,GAAG;EACxD,KAAK,aAAa,QAAQ;EAC1B,KAAK,aAAa,QAAQ;CAC5B;CAEA,MAAM,KACJ,OACA,SACA,QACwB;EAIxB,MAAM,eAAe,OAAO,OAAO,KAAK;EAGxC,MAAM,eAAe;GAAE,GAAG,QAAQ;GAAK,GAAI,KAAK,cAAc,CAAC;GAAI,GAAI,OAAO,OAAO,CAAC;EAAG;EACzF,MAAM,SAAS,MAAM,kBAAkB;GACrC;GACA,OAAO;GACP,KAAK;GACL,KAAK;GACL,SAAS;GACT,WAAW,OAAO,aAAa;GAC/B,gBAAgB,OAAO,kBAAkB;EAC3C,CAAC;EAID,MAAM,SACJ,CAAC,OAAO,mBAAmB,UAAU,UAAU,OAAO,aAClD,OAAO,WAAW,MAAM,OAAO,QAAQ,OAAO,QAAQ,OAAO,QAAQ,IACrE,KAAA;EACN,OAAO;GACL;GACA,UAAU,OAAO;GACjB,QAAQ,OAAO;GACf,QAAQ,OAAO,cAAc,OAAO,SAAS,OAAO,cAAc,OAAO;GACzE,QAAQ,OAAO;GACf,YAAY,QAAQ;GACpB,aAAa,QAAQ;GACrB,iBAAiB,OAAO,mBAAmB,KAAA;GAC3C,iBAAiB,OAAO,mBAAmB,KAAA;EAC7C;CACF;AACF;AAkBA,IAAa,iBAAb,MAA4B;CAC1B;CACA,YAAY,SAAwB,IAAI,wBAAwB,GAAG;EACjE,KAAK,SAAS;CAChB;CAEA,MAAM,IAAI,QAAuB,SAAsD;EACrF,MAAM,SAAS,MAAM,QAAQ,QAAQ;GACnC,MAAM,WAAW,KAAK,OAAO,GAAG;GAChC,OAAO,OAAO;GACd,SAAS;IAAC,OAAO;IAAc,OAAO;IAAY,OAAO;GAAW,CAAC,CAClE,OAAO,OAAO,CAAC,CACf,KAAK,MAAM;EAChB,CAAC;EACD,MAAM,SAA+B;GAAE,QAAQ;GAAO,aAAa;GAAG,OAAO;EAAE;EAC/E,IAAI;GACF,IAAI,OAAO,cAAc;IACvB,OAAO,QAAQ,MAAM,KAAK,OAAO,KAAK,SAAS,OAAO,cAAc,MAAM;IAC1E,OAAO,eAAe,OAAO,MAAM;IACnC,IAAI,OAAO,MAAM,mBAAmB,OAAO,MAAM,aAAa,GAAG;KAC/D,MAAM,SAAS,OAAO,MAAM,kBACxB,yBAAyB,OAAO,MAAM,OAAO,MAC7C,sBAAsB,OAAO,MAAM,SAAS;KAChD,MAAM,OAAO,KAAK,QAAQ;MACxB,UAAU,OAAO,MAAM;MACvB,QAAQ,OAAO;KACjB,CAAyB;KACzB,OAAO;IACT;GACF;GACA,IAAI,OAAO,YAAY;IACrB,OAAO,MAAM,MAAM,KAAK,OAAO,KAAK,OAAO,OAAO,YAAY,MAAM;IACpE,OAAO,eAAe,OAAO,IAAI;IACjC,IAAI,OAAO,IAAI,mBAAmB,OAAO,IAAI,aAAa,GAAG;KAC3D,MAAM,SAAS,OAAO,IAAI,kBACtB,uBAAuB,OAAO,IAAI,OAAO,MACzC,oBAAoB,OAAO,IAAI,SAAS;KAC5C,MAAM,OAAO,KAAK,QAAQ;MACxB,UAAU,OAAO,IAAI;MACrB,QAAQ,OAAO;KACjB,CAAyB;KACzB,OAAO;IACT;GACF;GACA,IAAI,OAAO,aAAa;IACtB,OAAO,OAAO,MAAM,KAAK,OAAO,KAAK,QAAQ,OAAO,aAAa,MAAM;IACvE,OAAO,eAAe,OAAO,KAAK;IAGlC,MAAM,SAAS,CAAC,OAAO,KAAK,mBAAmB,OAAO,KAAK,aAAa;IACxE,OAAO,SAAS;IAChB,IAAI,OAAO,KAAK,eAAe,KAAA,KAAa,OAAO,KAAK,aAAa,GACnE,OAAO,SAAS,OAAO,KAAK,eAAe,KAAK,OAAO,KAAK;SAE5D,OAAO,QAAQ,SAAS,IAAI;IAE9B,MAAM,OAAO,IAAI;KACf,UAAU,OAAO,KAAK;KACtB,YAAY,OAAO,KAAK;KACxB,aAAa,OAAO,KAAK;KACzB,QAAQ,OAAO;KACf,QAAQ,SAAS,OAAO;IAC1B,CAAyB;GAC3B,OAAO;IACL,OAAO,SAAS;IAChB,OAAO,QAAQ;IACf,MAAM,OAAO,IAAI,EAAE,QAAQ,OAAO,YAAY,CAAyB;GACzE;EACF,SAAS,KAAK;GACZ,MAAM,OAAO,KAAK,eAAe,QAAQ,MAAM,OAAO,GAAG,CAAC;GAC1D,MAAM;EACR;EACA,OAAO;CACT;AACF;;AClQA,eAAsB,sBACpB,UACA,OACA,UAAgC,CAAC,GACH;CAC9B,MAAM,UAAU,IAAI,aAAa,KAAK;CACtC,MAAM,QAAQ,SAAS;EACrB,YAAY,SAAS;EACrB,WAAW,QAAQ;EACnB,gBAAgB,SAAS;EACzB,MAAM,SAAS;EACf,GAAG,QAAQ;CACb,CAAC;CAED,MAAM,SAAS,MAAM,IADD,eAAe,QAAQ,MAChB,CAAC,CAAC,IAAI,SAAS,SAAS,OAAO;CAC1D,MAAM,YAAY,SAAS,iBAAiB;CAC5C,MAAM,OAAO,OAAO,UAAU,OAAO,SAAS;CAC9C,MAAM,cAAc,OAAO,UAAU,KAAA,KAAa,OAAO,MAAM,aAAa;CAC5E,MAAM,YAAY,OAAO,QAAQ,KAAA,KAAa,OAAO,IAAI,aAAa;CACtE,MAAM,aAAa,OAAO,SAAS,KAAA,KAAa,OAAO,KAAK,aAAa;CACzE,MAAM,eAAyC,OAC3C,YACA,eAAe,YACb,oBACA,aACE,iBACA;CACR,MAAM,QAAQ,OAAO;EACnB;EACA,OAAO,OAAO;EACd;EACA,OAAO,OAAO,KAAA,IAAY,iBAAiB,MAAM;CACnD,CAAC;CACD,OAAO;EACL,OAAO,QAAQ;EACf;EACA,SAAS;EACT;EACA,OAAO,OAAO;EACd;CACF;AACF;AAEA,SAAS,iBAAiB,QAAsC;CAC9D,IAAI,OAAO,SAAS,OAAO,MAAM,aAAa,GAC5C,OAAO,sBAAsB,OAAO,MAAM;CAC5C,IAAI,OAAO,OAAO,OAAO,IAAI,aAAa,GAAG,OAAO,oBAAoB,OAAO,IAAI;CACnF,IAAI,OAAO,MAAM;EACf,IAAI,OAAO,KAAK,eAAe,KAAA,GAC7B,OAAO,UAAU,OAAO,KAAK,eAAe,EAAE,GAAG,OAAO,KAAK;EAE/D,OAAO,aAAa,OAAO,KAAK;CAClC;CACA,OAAO;AACT;;;;;;;;;;;;;;;;;;;ACtDA,IAAa,iBAAb,MAA4B;CAC1B;CACA;CACA;CACA;CACA;CACA;CACA;CAEA,YAAY,OAAmB,MAA0B,QAAwB;EAC/E,KAAK,QAAQ;EACb,KAAK,YAAY,KAAK;EACtB,KAAK,SAAS,KAAK,UAAU,YAAY;EACzC,KAAK,gBAAgB;EACrB,KAAK,iBAAiB,IAAI,aAAa,KAAK;CAC9C;;CAGA,MAAM,UAAU,aAAa,GAAG,KAAK,UAAU,QAAyB;EACtE,MAAM,KAAK,eAAe,SAAS;GACjC;GACA,WAAW,KAAK;GAChB,QAAQ,KAAK;GACb,OAAO;EACT,CAAC;EACD,KAAK,eAAe,KAAK,eAAe;EACxC,OAAO,KAAK;CACd;;CAGA,IAAI,UAAwB;EAC1B,IAAI,CAAC,KAAK,cAAc,MAAM,IAAI,MAAM,gDAAgD;EACxF,OAAO,KAAK;CACd;;;;;CAMA,MAAM,KAAK,SAAgF;EACzF,IAAI,CAAC,KAAK,cAAc,MAAM,IAAI,MAAM,6CAA6C;EACrF,MAAM,eAAe,IAAI,aAAa,KAAK,KAAK;EAChD,MAAM,aAAa,SAAS;GAC1B,YAAY,QAAQ,cAAc,GAAG,KAAK,UAAU;GACpD,WAAW,KAAK;GAChB,QAAQ,KAAK;GACb,aAAa,KAAK;GAClB,OAAO;EACT,CAAC;EAED,MAAM,SAAS,MAAM,IADD,eAAe,QAAQ,UAAU,KAAK,aAC/B,CAAC,CAAC,IAAI,QAAQ,SAAS,YAAY;EAC9D,MAAM,aAAa,OAAO;GACxB,MAAM,OAAO;GACb,OAAO,OAAO;GACd,cAAc,OAAO,SAAS,YAAY;EAC5C,CAAC;EACD,KAAK,iBAAiB,aAAa;EACnC,OAAO;GAAE,OAAO,aAAa;GAAO;EAAO;CAC7C;;;;;;;CAQA,MAAM,eAAe,SAA8D;EACjF,MAAM,cAAc,KAAK,kBAAkB,KAAK;EAChD,IAAI,CAAC,aACH,MAAM,IAAI,MAAM,gEAAgE;EAClF,MAAM,EAAE,UAAU,WAAW;EAC7B,MAAM,SAAS,MAAM,sBAAsB,UAAU,KAAK,OAAO;GAC/D,QAAQ,UAAU,KAAK;GACvB,YAAY;IAAE,SAAS,KAAA;IAAW,WAAW,KAAA;IAAW,kBAAkB,KAAA;GAAU;EACtF,CAAC;EAED,MAAM,KAAK,MAAM,UAAU,OAAO,OAAO;GACvC;GACA,WAAW,KAAK;GAChB,QAAQ,KAAK;GACb,OAAO;EACT,CAAC;EACD,OAAO;CACT;;;CAIA,MAAM,gBAAgB,OAAe,WAAmC;EACtE,IAAI,CAAC,KAAK,cACR,MAAM,IAAI,MAAM,wDAAwD;EAC1E,MAAM,KAAK,eAAe,YAAY;GACpC,SAAS;GACT,cAAc,KAAK;GACnB,WAAW;GACX;GACA;GACA,MAAM;EACR,CAAC;CACH;;CAGA,MAAM,QAAQ,SAA2E;EACvF,MAAM,KAAK,eAAe,OAAO;GAC/B,MAAM,QAAQ;GACd,OAAO,QAAQ;GACf,OAAO,QAAQ;EACjB,CAAC;CACH;;;;;;;;CASA,MAAM,gBAAgB,YAA2C;EAC/D,MAAM,cAAc,KAAK,kBAAkB,KAAK;EAChD,IAAI,CAAC,aACH,MAAM,IAAI,MACR,8EACF;EACF,MAAM,UAAU,IAAI,aAAa,KAAK,KAAK;EAC3C,MAAM,QAAQ,SAAS;GACrB;GACA,WAAW,KAAK;GAChB,QAAQ,KAAK;GACb;GACA,OAAO;EACT,CAAC;EACD,OAAO;CACT;;;;;;CAOA,MAAM,iBAAiB,MAKH;EAClB,IAAI,CAAC,KAAK,cACR,MAAM,IAAI,MAAM,yDAAyD;EAC3E,MAAM,UAAU,IAAI,aAAa,KAAK,KAAK;EAC3C,MAAM,QAAQ,SAAS;GACrB,YAAY,KAAK,cAAc,GAAG,KAAK,UAAU;GACjD,WAAW,KAAK;GAChB,QAAQ,KAAK;GACb,aAAa,KAAK;GAClB,OAAO;EACT,CAAC;EACD,MAAM,QAAQ,OAAO;GACnB,MAAM,KAAK;GACX,OAAO,KAAK;GACZ,cAAc,KAAK,OAAO,YAAY;GACtC,OAAO,KAAK;EACd,CAAC;EACD,KAAK,iBAAiB,QAAQ;EAC9B,OAAO,QAAQ;CACjB;CAEA,IAAI,sBAA0C;EAC5C,OAAO,KAAK;CACd;CACA,IAAI,oBAAwC;EAC1C,OAAO,KAAK;CACd;AACF;;;;;;AAOA,eAAsB,qBACpB,OACA,WAOC;CACD,MAAM,OAAO,MAAM,MAAM,SAAS,EAAE,UAAU,CAAC;CAC/C,MAAM,WAAW,KACd,QAAQ,MAAM,EAAE,UAAU,SAAS,CAAC,CACpC,MAAM,GAAG,MAAM,EAAE,YAAY,EAAE,SAAS;CAC3C,MAAM,YAAY,KACf,QAAQ,MAAM,EAAE,UAAU,WAAW,CAAC,CACtC,MAAM,GAAG,MAAM,EAAE,YAAY,EAAE,SAAS;CAC3C,MAAM,iBAAiB,KACpB,QAAQ,MAAM,EAAE,UAAU,aAAa,CAAC,CACxC,MAAM,GAAG,MAAM,EAAE,YAAY,EAAE,SAAS;CAC3C,OAAO;EACL;EACA;EACA,gBAAgB,SAAS;EACzB,cAAc,UAAU;EACxB,oBAAoB;CACtB;AACF;;;;;;;;;;;;;;;;;;ACxNA,SAAgB,gBAAgB,SAAuD;CACrF,MAAM,mBAAmB,QAAQ,QAAQ,MAAM,EAAE,QAAQ,CAAC,CAAC;CAC3D,OAAO;EACL,aAAa,SACX,UACC,MAAM,EAAE,YACR,MAAM,EAAE,UACX;EACA,eAAe,SACb,UACC,MAAM,EAAE,YACR,MAAM,EAAE,YACX;EACA,gBAAgB,SACd,UACC,MAAM,EAAE,aACR,MAAM,EAAE,YACX;EACA;CACF;AACF;AAEA,SAAS,SACP,SACA,GACA,GAC8B;CAC9B,MAAM,KAAe,CAAC;CACtB,MAAM,KAAe,CAAC;CACtB,KAAK,MAAM,KAAK,SAAS;EACvB,MAAM,IAAI,EAAE,CAAC;EACb,MAAM,IAAI,EAAE,CAAC;EACb,IAAI,MAAM,QAAQ,MAAM,QAAQ,OAAO,SAAS,CAAC,KAAK,OAAO,SAAS,CAAC,GAAG;GACxE,GAAG,KAAK,CAAC;GACT,GAAG,KAAK,CAAC;EACX;CACF;CACA,IAAI,GAAG,SAAS,GAAG,OAAO,KAAA;CAC1B,OAAO;EACL,GAAG,GAAG;EACN,SAAS,SAAS,IAAI,EAAE;EACxB,UAAU,UAAU,IAAI,EAAE;CAC5B;AACF;;ACjCA,IAAa,kBAAb,MAA6B;CACP;CAApB,YAAY,OAA2B;EAAnB,KAAA,QAAA;CAAoB;CAExC,MAAM,eAA0C;EAC9C,MAAM,OAAO,MAAM,KAAK,MAAM,SAAS;EACvC,MAAM,4BAAY,IAAI,IAAmB;EACzC,KAAK,MAAM,KAAK,MAAM;GACpB,IAAI,CAAC,EAAE,WAAW;GAClB,MAAM,MAAM,UAAU,IAAI,EAAE,SAAS,KAAK,CAAC;GAC3C,IAAI,KAAK,CAAC;GACV,UAAU,IAAI,EAAE,WAAW,GAAG;EAChC;EACA,MAAM,YAA8B,CAAC;EACrC,KAAK,MAAM,CAAC,WAAW,gBAAgB,WAAW;GAChD,MAAM,SAAS,YAAY,MAAM,CAAC,CAAC,MAAM,GAAG,MAAM,EAAE,YAAY,EAAE,SAAS;GAC3E,MAAM,QAAQ,YAAY,QAAQ,MAAM,EAAE,UAAU,SAAS;GAC7D,MAAM,SAAS,YAAY,QAAQ,MAAM,EAAE,UAAU,WAAW;GAChE,MAAM,WAAW,YAAY,QAAQ,MAAM,EAAE,UAAU,aAAa;GACpE,MAAM,SAAS,OAAO;GACtB,IAAI,CAAC,QAAQ;GACb,UAAU,KAAK;IACb;IACA,WAAW,MAAM;IACjB,YAAY,OAAO;IACnB,iBAAiB,SAAS;IAC1B,gBAAgB,OAAO;IACvB,cAAc,MAAM,EAAE,EAAE;IACxB,eAAe,OAAO,UAClB;KAAE,MAAM,OAAO,QAAQ,QAAQ;KAAO,OAAO,OAAO,QAAQ;IAAM,IAClE,KAAA;GACN,CAAC;EACH;EACA,OAAO,UAAU,MAAM,GAAG,MAAM,EAAE,iBAAiB,EAAE,cAAc;CACrE;CAEA,MAAM,gBAAgB,WAAoD;EAGxE,QADgB,MADG,KAAK,MAAM,SAAS,EAAE,UAAU,CAAC,EAAA,CAC/B,MAAM,CAAC,CAAC,MAAM,GAAG,MAAM,EAAE,YAAY,EAAE,SAC/C,CAAC,CAAC,KAAK,SAAS;GAC3B;GACA,aACE,IAAI,UAAU,YACV,SACA,IAAI,UAAU,cACZ,UACA,IAAI,UAAU,gBACZ,YACA;EACZ,EAAE;CACJ;CAEA,MAAM,aAAa,WAA2C;EAC5D,MAAM,eAAe,MAAM,KAAK,MAAM,SAAS;GAAE;GAAW,OAAO;EAAU,CAAC,EAAA,CAAG,MAC9E,GAAG,MAAM,EAAE,YAAY,EAAE,SAC5B;EACA,MAAM,cAAc,OAAO,UAAkB,KAAK,MAAM,SAAS,EAAE,aAAa,MAAM,CAAC;EACvF,MAAM,MAAqB,CAAC;EAC5B,KAAK,MAAM,OAAO,aAAa;GAC7B,MAAM,QAAQ,MAAM,KAAK,MAAM,MAAM,EAAE,OAAO,IAAI,MAAM,CAAC;GACzD,MAAM,WAAW,MAAM,YAAY,IAAI,KAAK;GAC5C,MAAM,QAAQ,SAAS,MAAM,MAAM,EAAE,UAAU,WAAW;GAC1D,MAAM,UAAoB,CAAC;GAG3B,IAAI,OAAO;IACT,MAAM,SAAS,MAAM,YAAY,MAAM,KAAK;IAC5C,KAAK,MAAM,KAAK,QAAQ,IAAI,EAAE,UAAU,eAAe,QAAQ,KAAK,EAAE,KAAK;GAC7E;GACA,KAAK,MAAM,KAAK,UAAU,IAAI,EAAE,UAAU,eAAe,QAAQ,KAAK,EAAE,KAAK;GAC7E,IAAI,KAAK;IACP,QAAQ,IAAI,UAAU,IAAI;IAC1B;IACA,cAAc,IAAI;IAClB,WAAW,IAAI;IACf,SAAS,IAAI;IACb,QAAQ,IAAI;IACZ,SAAS,IAAI;IACb,UAAU,MAAM,QAAQ,MAAM,EAAE,SAAS,KAAK,CAAC,CAAC;IAChD,WAAW,MAAM,QAAQ,MAAM,EAAE,SAAS,MAAM,CAAC,CAAC;IAClD,YAAY,OAAO;IACnB,kBAAkB;GACpB,CAAC;EACH;EACA,OAAO;CACT;AACF;;;;;;;;;;;;;;;;;;;;;;;;;ACzEA,eAAsB,aACpB,OACA,WACkC;CAClC,MAAM,UAAU,MAAM,MAAM,SAAS,EAAE,UAAU,CAAC;CAClD,MAAM,UAAU,cAAc,SAAS,SAAS;CAChD,MAAM,QAAQ,cAAc,SAAS,WAAW;CAChD,MAAM,UAAU,QAAQ,QAAQ,MAAM,EAAE,UAAU,aAAa;CAE/D,MAAM,YAAY,UAAU,MAAM,iBAAiB,OAAO,QAAQ,KAAK,IAAI;CAC3E,MAAM,aAAa,OAAO,SAAS,SAAS;CAC5C,MAAM,gBAAgB,QACnB,KAAK,MAAM,EAAE,SAAS,KAAK,CAAC,CAC5B,QAAQ,MAAmB,OAAO,MAAM,QAAQ;CACnD,MAAM,eACJ,cAAc,SAAS,IACnB,cAAc,QAAQ,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,cAAc,SACzD;CACN,MAAM,gBAAgB,QAAQ,QAAQ,MAAM,EAAE,SAAS,SAAS,IAAI,CAAC,CAAC;CACtE,MAAM,kBAAkB,QAAQ,SAAS,IAAI,gBAAgB,QAAQ,SAAS;CAE9E,MAAM,OAAoB,QAAQ,WAAW,IAAI,kBAAkB;CACnE,MAAM,WACJ,SAAS,kBACL,cAAc,QAAQ,eAAe,OACrC,cAAc,QAAQ,eAAe,QAAQ,iBAAiB;CAEpE,OAAO;EACL;EACA;EACA,cAAc,SAAS;EACvB;EACA,YAAY,OAAO;EACnB;EACA,kBAAkB,QAAQ,KAAK,MAAM,EAAE,KAAK;EAC5C;EACA;EACA;CACF;AACF;;AAGA,eAAsB,iBAAiB,OAAuD;CAC5F,MAAM,OAAO,MAAM,MAAM,SAAS;CAClC,MAAM,aAAa,CAAC,GAAG,IAAI,IAAI,KAAK,KAAK,MAAM,EAAE,SAAS,CAAC,CAAC,QAAQ,MAAmB,CAAC,CAAC,CAAC,CAAC,CAAC;CAC5F,OAAO,QAAQ,IAAI,WAAW,KAAK,MAAM,aAAa,OAAO,CAAC,CAAC,CAAC;AAClE;AAEA,SAAS,cAAc,MAAa,OAAsC;CAExE,OADiB,KAAK,QAAQ,MAAM,EAAE,UAAU,KAAK,CAAC,CAAC,MAAM,GAAG,MAAM,EAAE,YAAY,EAAE,SACxE,CAAC,CAAC;AAClB;AAEA,eAAe,iBAAiB,OAAmB,cAA8C;CAE/F,MAAM,QAAO,MADI,WAAW,OAAO,YAAY,EAAA,CAC/B,MACb,MAAM,EAAE,YAAY,kBAAkB,EAAE,cAAc,0BACzD;CACA,IAAI,CAAC,MAAM,OAAO;CAElB,IAAI,KAAK,SAAS,KAAK,KAAK,SAAS,GAAG,OAAO,KAAK;CACpD,IAAI,KAAK,SAAS,KAAK,KAAK,SAAS,IAAI,OAAO,KAAK,QAAQ;CAC7D,OAAO;AACT"}
+{"version":3,"file":"index.js","names":[],"sources":["../../src/test-graded-scenario.ts","../../src/builder-eval/builder-session.ts","../../src/builder-eval/correlation.ts","../../src/builder-eval/project-registry.ts","../../src/builder-eval/three-layer-eval.ts"],"sourcesContent":["import type { HarnessConfig, SandboxDriver, SandboxHarnessResult } from './sandbox-harness'\nimport { SandboxHarness } from './sandbox-harness'\nimport { TraceEmitter } from './trace/emitter'\nimport type { FailureClass, Run } from './trace/schema'\nimport type { TraceStore } from './trace/store'\n\nexport interface TestGradedScenario {\n  id: string\n  description?: string\n  harness: HarnessConfig\n  /** Optional pass threshold in 0..1 (default 1.0 = all tests must pass). */\n  passThreshold?: number\n  /** Provenance for dataset tracking. */\n  datasetVersion?: string\n  /** Free-form tags (difficulty, category, etc.). */\n  tags?: Record<string, string>\n}\n\ninterface TestGradedRunOptions {\n  variantId?: string\n  driver?: SandboxDriver\n  /** Metadata recorded on the Run (codeSha, promptSha, modelFingerprint, seed). */\n  provenance?: Pick<Run, 'codeSha' | 'promptSha' | 'modelFingerprint' | 'seed' | 'envFingerprint'>\n}\n\nexport interface TestGradedRunResult {\n  runId: string\n  scenario: TestGradedScenario\n  harness: SandboxHarnessResult\n  pass: boolean\n  score: number\n  failureClass?: FailureClass\n}\n\nexport async function runTestGradedScenario(\n  scenario: TestGradedScenario,\n  store: TraceStore,\n  options: TestGradedRunOptions = {},\n): Promise<TestGradedRunResult> {\n  const emitter = new TraceEmitter(store)\n  await emitter.startRun({\n    scenarioId: scenario.id,\n    variantId: options.variantId,\n    datasetVersion: scenario.datasetVersion,\n    tags: scenario.tags,\n    ...options.provenance,\n  })\n  const result = await new SandboxHarness(options.driver).run(scenario.harness, emitter)\n  const pass = result.passed && result.score >= (scenario.passThreshold ?? 1)\n  const failure = (['setup', 'run', 'test'] as const).find(\n    (phase) => result[phase] && result[phase]!.exitCode !== 0,\n  )\n  const failureClass: FailureClass = pass\n    ? 'success'\n    : failure === 'test'\n      ? 'format_drift'\n      : failure\n        ? 'sandbox_failure'\n        : 'unknown'\n  await emitter.endRun({\n    pass,\n    score: result.score,\n    failureClass,\n    notes: pass ? undefined : reasonForFailure(result),\n  })\n  return {\n    runId: emitter.runId,\n    scenario,\n    harness: result,\n    pass,\n    score: result.score,\n    failureClass,\n  }\n}\n\nfunction reasonForFailure(result: SandboxHarnessResult): string {\n  const failed = (['setup', 'run'] as const).find(\n    (phase) => result[phase] && result[phase]!.exitCode !== 0,\n  )\n  if (failed) return `${failed} failed: exit ${result[failed]!.exitCode}`\n  const test = result.test\n  return !test\n    ? 'no test command'\n    : test.testsTotal !== undefined\n      ? `tests: ${test.testsPassed ?? 0}/${test.testsTotal}`\n      : `test exit ${test.exitCode}`\n}\n","/**\n * BuilderSession — ties a builder-of-builders workflow together.\n *\n * Models agent-builder's shape: Project → Chat → Edit → Ship → App →\n * AppAgent. Each layer is a Run (linked via parentRunId). The\n * framework-enforced invariants:\n *\n *   - One Project → many Chats; chatId scopes runs within a project.\n *   - One Chat = one builder Run with `layer='builder'`.\n *   - One Ship = one child Run with `layer='app-build'` + SandboxHarness.\n *   - One AppScenario = one grandchild Run with `layer='app-runtime'`.\n *\n * Consumers obtain a BuilderSession, call `startChat`, drive the\n * builder agent (emitting spans), and call `ship` / `runAppScenario`\n * as the workflow progresses. The session reconstructs itself from\n * trace data via `resume(store, projectId)`.\n */\n\nimport { newRecordId } from '../record-id'\nimport type { HarnessConfig, SandboxDriver, SandboxHarnessResult } from '../sandbox-harness'\nimport { SandboxHarness } from '../sandbox-harness'\nimport type { TestGradedRunResult, TestGradedScenario } from '../test-graded-scenario'\nimport { runTestGradedScenario } from '../test-graded-scenario'\nimport { TraceEmitter } from '../trace/emitter'\nimport type { Run } from '../trace/schema'\nimport type { TraceStore } from '../trace/store'\n\nexport interface BuilderSessionInit {\n  projectId: string\n  chatId?: string\n  /** Free-form: user's task description, project name, etc. Stored on the builder Run. */\n  tags?: Record<string, string>\n}\n\nexport interface ShipOptions {\n  harness: HarnessConfig\n  driver?: SandboxDriver\n  /** scenarioId of this app-build run. Defaults to `${projectId}/build`. */\n  scenarioId?: string\n}\n\nexport interface RunAppScenarioOptions {\n  scenario: TestGradedScenario\n  /** Harness driver override; defaults to the one the session was created with. */\n  driver?: SandboxDriver\n}\n\nexport class BuilderSession {\n  private store: TraceStore\n  private builderEmitter: TraceEmitter\n  readonly projectId: string\n  readonly chatId: string\n  private builderRunId?: string\n  private lastBuildRunId?: string\n  private defaultDriver?: SandboxDriver\n\n  constructor(store: TraceStore, init: BuilderSessionInit, driver?: SandboxDriver) {\n    this.store = store\n    this.projectId = init.projectId\n    this.chatId = init.chatId ?? newRecordId()\n    this.defaultDriver = driver\n    this.builderEmitter = new TraceEmitter(store)\n  }\n\n  /** Start the builder (L0) run for this chat. Returns the runId. */\n  async startChat(scenarioId = `${this.projectId}/chat`): Promise<string> {\n    await this.builderEmitter.startRun({\n      scenarioId,\n      projectId: this.projectId,\n      chatId: this.chatId,\n      layer: 'builder',\n    })\n    this.builderRunId = this.builderEmitter.runId\n    return this.builderRunId\n  }\n\n  /** The emitter for builder-level spans (edits, LLM calls, tool invocations). */\n  get emitter(): TraceEmitter {\n    if (!this.builderRunId) throw new Error('BuilderSession.emitter: call startChat() first')\n    return this.builderEmitter\n  }\n\n  /**\n   * Ship the project's generated app: run the sandbox harness as a child\n   * Run (`layer='app-build'`). Returns the build result + runId.\n   */\n  async ship(options: ShipOptions): Promise<{ runId: string; result: SandboxHarnessResult }> {\n    if (!this.builderRunId) throw new Error('BuilderSession.ship: call startChat() first')\n    const buildEmitter = new TraceEmitter(this.store)\n    await buildEmitter.startRun({\n      scenarioId: options.scenarioId ?? `${this.projectId}/build`,\n      projectId: this.projectId,\n      chatId: this.chatId,\n      parentRunId: this.builderRunId,\n      layer: 'app-build',\n    })\n    const harness = new SandboxHarness(options.driver ?? this.defaultDriver)\n    const result = await harness.run(options.harness, buildEmitter)\n    await buildEmitter.endRun({\n      pass: result.passed,\n      score: result.score,\n      failureClass: result.passed ? 'success' : 'sandbox_failure',\n    })\n    this.lastBuildRunId = buildEmitter.runId\n    return { runId: buildEmitter.runId, result }\n  }\n\n  /**\n   * Run a domain scenario against the just-built app as a grandchild Run\n   * (`layer='app-runtime'`). The `ship` call must precede this so the\n   * parent is set correctly; if no build exists yet the session attaches\n   * directly to the builder run (useful for prototypes).\n   */\n  async runAppScenario(options: RunAppScenarioOptions): Promise<TestGradedRunResult> {\n    const parentRunId = this.lastBuildRunId ?? this.builderRunId\n    if (!parentRunId)\n      throw new Error('BuilderSession.runAppScenario: call startChat() + ship() first')\n    const { scenario, driver } = options\n    const result = await runTestGradedScenario(scenario, this.store, {\n      driver: driver ?? this.defaultDriver,\n      provenance: { codeSha: undefined, promptSha: undefined, modelFingerprint: undefined },\n    })\n    // Attach to the parent chain by updating the stored Run in place.\n    await this.store.updateRun(result.runId, {\n      parentRunId,\n      projectId: this.projectId,\n      chatId: this.chatId,\n      layer: 'app-runtime',\n    })\n    return result\n  }\n\n  /** Record an end-of-chat meta score (judge verdict on whether the builder\n   *  served the user's intent). Accepts a numeric score + optional rationale. */\n  async recordMetaScore(score: number, rationale?: string): Promise<void> {\n    if (!this.builderRunId)\n      throw new Error('BuilderSession.recordMetaScore: call startChat() first')\n    await this.builderEmitter.recordJudge({\n      judgeId: 'builder-meta',\n      targetSpanId: this.builderRunId, // attach to the builder run itself\n      dimension: 'user_intent_satisfaction',\n      score,\n      rationale,\n      name: 'builder-meta',\n    })\n  }\n\n  /** Close the builder Run with a final outcome. */\n  async endChat(outcome: { pass: boolean; score?: number; notes?: string }): Promise<void> {\n    await this.builderEmitter.endRun({\n      pass: outcome.pass,\n      score: outcome.score,\n      notes: outcome.notes,\n    })\n  }\n\n  /**\n   * Inline app-runtime run — for cases where the \"scenario\" isn't a\n   * SWE-bench-style test suite but a live agent interaction (LLM chat,\n   * domain flow). Returns an emitter bound to a fresh Run in the\n   * `app-runtime` layer; caller emits spans inside and calls\n   * `.endRun()` with the final verdict.\n   */\n  async startAppRuntime(scenarioId: string): Promise<TraceEmitter> {\n    const parentRunId = this.lastBuildRunId ?? this.builderRunId\n    if (!parentRunId)\n      throw new Error(\n        'BuilderSession.startAppRuntime: call startChat() + (optionally) ship() first',\n      )\n    const emitter = new TraceEmitter(this.store)\n    await emitter.startRun({\n      scenarioId,\n      projectId: this.projectId,\n      chatId: this.chatId,\n      parentRunId,\n      layer: 'app-runtime',\n    })\n    return emitter\n  }\n\n  /**\n   * Lightweight \"ship marker\" — record an app-build Run with a caller-\n   * provided verdict. Use when there isn't a sandbox harness to run but\n   * you still want to mark the build state at publish time.\n   */\n  async recordShipMarker(args: {\n    pass: boolean\n    score: number\n    scenarioId?: string\n    notes?: string\n  }): Promise<string> {\n    if (!this.builderRunId)\n      throw new Error('BuilderSession.recordShipMarker: call startChat() first')\n    const emitter = new TraceEmitter(this.store)\n    await emitter.startRun({\n      scenarioId: args.scenarioId ?? `${this.projectId}/ship`,\n      projectId: this.projectId,\n      chatId: this.chatId,\n      parentRunId: this.builderRunId,\n      layer: 'app-build',\n    })\n    await emitter.endRun({\n      pass: args.pass,\n      score: args.score,\n      failureClass: args.pass ? 'success' : 'sandbox_failure',\n      notes: args.notes,\n    })\n    this.lastBuildRunId = emitter.runId\n    return emitter.runId\n  }\n\n  get lastBuildRunIdValue(): string | undefined {\n    return this.lastBuildRunId\n  }\n  get builderRunIdValue(): string | undefined {\n    return this.builderRunId\n  }\n}\n\n/**\n * Reconstruct the most recent BuilderSession state for a given project —\n * returns { builderRunId, lastBuildRunId, chatRuns }. For chat-first UIs\n * this is how a resumed session finds its place in the edit history.\n */\nexport async function resumeBuilderSession(\n  store: TraceStore,\n  projectId: string,\n): Promise<{\n  projectId: string\n  chatRuns: Run[]\n  lastBuilderRun?: Run\n  lastBuildRun?: Run\n  lastAppRuntimeRuns: Run[]\n}> {\n  const runs = await store.listRuns({ projectId })\n  const chatRuns = runs\n    .filter((r) => r.layer === 'builder')\n    .sort((a, b) => b.startedAt - a.startedAt)\n  const buildRuns = runs\n    .filter((r) => r.layer === 'app-build')\n    .sort((a, b) => b.startedAt - a.startedAt)\n  const appRuntimeRuns = runs\n    .filter((r) => r.layer === 'app-runtime')\n    .sort((a, b) => b.startedAt - a.startedAt)\n  return {\n    projectId,\n    chatRuns,\n    lastBuilderRun: chatRuns[0],\n    lastBuildRun: buildRuns[0],\n    lastAppRuntimeRuns: appRuntimeRuns,\n  }\n}\n","/**\n * Meta-eval correlation — the highest-leverage signal in the framework.\n *\n * Given a corpus of three-layer project reports, compute how well each\n * pair of layers correlates. The question we care about most:\n *\n *   Does `metaScore` (what the builder thinks it did) predict\n *   `runtimeScore` (what the user actually gets)?\n *\n * If r < ~0.4, the builder's self-scoring is broken — it's optimizing\n * for something other than real-world success. If r > 0.7, meta_score\n * is a usable proxy and can drive CI gates cheaply.\n *\n * Non-parametric rank correlation (Spearman) is also reported because\n * meta scores are often ordinal-ish.\n */\n\nimport { pearsonR, spearmanR } from '../statistics'\nimport type { ThreeLayerProjectReport } from './three-layer-eval'\n\nexport interface LayerCorrelation {\n  n: number\n  pearson: number\n  spearman: number\n}\n\nexport interface CorrelationReport {\n  /** Pairs present in the corpus (layers with ≥ 2 matched data points). */\n  metaVsBuild?: LayerCorrelation\n  metaVsRuntime?: LayerCorrelation\n  buildVsRuntime?: LayerCorrelation\n  /** Number of complete projects (all 3 scores present). */\n  completeProjects: number\n}\n\nexport function correlateLayers(reports: ThreeLayerProjectReport[]): CorrelationReport {\n  const completeProjects = reports.filter((r) => r.complete).length\n  return {\n    metaVsBuild: pairwise(\n      reports,\n      (r) => r.metaScore,\n      (r) => r.buildScore,\n    ),\n    metaVsRuntime: pairwise(\n      reports,\n      (r) => r.metaScore,\n      (r) => r.runtimeScore,\n    ),\n    buildVsRuntime: pairwise(\n      reports,\n      (r) => r.buildScore,\n      (r) => r.runtimeScore,\n    ),\n    completeProjects,\n  }\n}\n\nfunction pairwise(\n  reports: ThreeLayerProjectReport[],\n  a: (r: ThreeLayerProjectReport) => number | null,\n  b: (r: ThreeLayerProjectReport) => number | null,\n): LayerCorrelation | undefined {\n  const xs: number[] = []\n  const ys: number[] = []\n  for (const r of reports) {\n    const x = a(r)\n    const y = b(r)\n    if (x !== null && y !== null && Number.isFinite(x) && Number.isFinite(y)) {\n      xs.push(x)\n      ys.push(y)\n    }\n  }\n  if (xs.length < 2) return undefined\n  return {\n    n: xs.length,\n    pearson: pearsonR(xs, ys),\n    spearman: spearmanR(xs, ys),\n  }\n}\n","/**\n * ProjectRegistry — project-level aggregation over the trace corpus.\n *\n * Thin reader over TraceStore that answers the questions a chat-first,\n * resumable UI needs:\n *   - listProjects() → project IDs with latest activity\n *   - projectTimeline(id) → chats + builds + runtime runs, chronological\n *   - projectChats(id) → chat-level summaries (turn count, outcome)\n *\n * All queries are pure reads; no state duplication.\n */\n\nimport type { Run } from '../trace/schema'\nimport type { TraceStore } from '../trace/store'\n\nexport interface ProjectSummary {\n  projectId: string\n  chatCount: number\n  buildCount: number\n  appRuntimeCount: number\n  lastActivityAt: number\n  latestChatId?: string\n  latestOutcome?: { pass: boolean; score?: number }\n}\n\nexport interface ChatSummary {\n  chatId: string\n  projectId: string\n  builderRunId: string\n  startedAt: number\n  endedAt?: number\n  status: Run['status']\n  outcome?: Run['outcome']\n  /** Counts of spans emitted during the chat. */\n  llmTurns?: number\n  toolCalls?: number\n  buildRunId?: string\n  appRuntimeRunIds: string[]\n}\n\nexport interface ProjectTimelineEntry {\n  run: Run\n  layerBucket: 'chat' | 'build' | 'runtime' | 'other'\n}\n\nexport class ProjectRegistry {\n  constructor(private store: TraceStore) {}\n\n  async listProjects(): Promise<ProjectSummary[]> {\n    const runs = await this.store.listRuns()\n    const byProject = new Map<string, Run[]>()\n    for (const r of runs) {\n      if (!r.projectId) continue\n      const arr = byProject.get(r.projectId) ?? []\n      arr.push(r)\n      byProject.set(r.projectId, arr)\n    }\n    const summaries: ProjectSummary[] = []\n    for (const [projectId, projectRuns] of byProject) {\n      const sorted = projectRuns.slice().sort((a, b) => b.startedAt - a.startedAt)\n      const chats = projectRuns.filter((r) => r.layer === 'builder')\n      const builds = projectRuns.filter((r) => r.layer === 'app-build')\n      const runtimes = projectRuns.filter((r) => r.layer === 'app-runtime')\n      const latest = sorted[0]\n      if (!latest) continue\n      summaries.push({\n        projectId,\n        chatCount: chats.length,\n        buildCount: builds.length,\n        appRuntimeCount: runtimes.length,\n        lastActivityAt: latest.startedAt,\n        latestChatId: chats[0]?.chatId,\n        latestOutcome: latest.outcome\n          ? { pass: latest.outcome.pass ?? false, score: latest.outcome.score }\n          : undefined,\n      })\n    }\n    return summaries.sort((a, b) => b.lastActivityAt - a.lastActivityAt)\n  }\n\n  async projectTimeline(projectId: string): Promise<ProjectTimelineEntry[]> {\n    const runs = await this.store.listRuns({ projectId })\n    const ordered = runs.slice().sort((a, b) => a.startedAt - b.startedAt)\n    return ordered.map((run) => ({\n      run,\n      layerBucket:\n        run.layer === 'builder'\n          ? 'chat'\n          : run.layer === 'app-build'\n            ? 'build'\n            : run.layer === 'app-runtime'\n              ? 'runtime'\n              : 'other',\n    }))\n  }\n\n  async projectChats(projectId: string): Promise<ChatSummary[]> {\n    const builderRuns = (await this.store.listRuns({ projectId, layer: 'builder' })).sort(\n      (a, b) => b.startedAt - a.startedAt,\n    )\n    const childrenFor = async (runId: string) => this.store.listRuns({ parentRunId: runId })\n    const out: ChatSummary[] = []\n    for (const run of builderRuns) {\n      const spans = await this.store.spans({ runId: run.runId })\n      const children = await childrenFor(run.runId)\n      const build = children.find((c) => c.layer === 'app-build')\n      const runtime: string[] = []\n      // Runtime runs can be grandchildren (attached to the build) or siblings\n      // when shipped skipped.\n      if (build) {\n        const grands = await childrenFor(build.runId)\n        for (const g of grands) if (g.layer === 'app-runtime') runtime.push(g.runId)\n      }\n      for (const c of children) if (c.layer === 'app-runtime') runtime.push(c.runId)\n      out.push({\n        chatId: run.chatId ?? run.runId,\n        projectId,\n        builderRunId: run.runId,\n        startedAt: run.startedAt,\n        endedAt: run.endedAt,\n        status: run.status,\n        outcome: run.outcome,\n        llmTurns: spans.filter((s) => s.kind === 'llm').length,\n        toolCalls: spans.filter((s) => s.kind === 'tool').length,\n        buildRunId: build?.runId,\n        appRuntimeRunIds: runtime,\n      })\n    }\n    return out\n  }\n}\n","/**\n * Three-layer evaluation — the canonical scoring breakdown for\n * builder-of-builders workflows.\n *\n *   meta_score:    did the builder understand + satisfy user intent?\n *                  (judge verdict attached to the builder run)\n *   build_score:   did the generated scaffold build + pass its own tests?\n *                  (outcome.score on the app-build child run)\n *   runtime_score: did the generated agent pass its domain scenarios?\n *                  (mean outcome.score over app-runtime grandchild runs)\n *\n * Returns a structured report per project. The cross-layer correlation\n * is the highest-leverage signal the framework computes — if\n * meta_score doesn't predict runtime_score, the builder's self-scoring\n * is broken.\n *\n * Scaffold-only mode: when a project has no `app-runtime` runs (e.g. a\n * scaffold-builder eval that grades compose + build without driving a\n * runtime scenario), `kind` is `'scaffold-only'` and `complete` measures\n * meta + build only. Consumers can tell the two apart without having to\n * interpret null-runtime as either \"not yet computed\" or \"N/A for this\n * project shape\".\n */\n\nimport { judgeSpans } from '../trace/query'\nimport type { Run } from '../trace/schema'\nimport type { TraceStore } from '../trace/store'\n\nexport type ProjectKind = 'full' | 'scaffold-only'\n\nexport interface ThreeLayerProjectReport {\n  projectId: string\n  /**\n   * `'full'` when the project has at least one `app-runtime` run;\n   * `'scaffold-only'` when it only has meta + build layers. Lets\n   * downstream consumers treat a null runtime score as expected\n   * (scaffold-only) vs. missing (full, pipeline broke).\n   */\n  kind: ProjectKind\n  builderRunId?: string\n  /** Judge-verdict score on the builder run (0..1 after normalization). */\n  metaScore: number | null\n  buildRunId?: string\n  /** 0..1 from the sandbox harness (testsPassed / testsTotal). */\n  buildScore: number | null\n  appRuntimeRunIds: string[]\n  /** Mean of outcome.score over app-runtime runs, 0..1. Always null in scaffold-only mode. */\n  runtimeScore: number | null\n  runtimePassRate: number | null\n  /**\n   * Layer-aware completeness:\n   *   - `kind='full'`: all three layers scored\n   *   - `kind='scaffold-only'`: meta + build scored (runtime not applicable)\n   */\n  complete: boolean\n}\n\nexport async function scoreProject(\n  store: TraceStore,\n  projectId: string,\n): Promise<ThreeLayerProjectReport> {\n  const allRuns = await store.listRuns({ projectId })\n  const builder = latestByLayer(allRuns, 'builder')\n  const build = latestByLayer(allRuns, 'app-build')\n  const runtime = allRuns.filter((r) => r.layer === 'app-runtime')\n\n  const metaScore = builder ? await extractMetaScore(store, builder.runId) : null\n  const buildScore = build?.outcome?.score ?? null\n  const runtimeScores = runtime\n    .map((r) => r.outcome?.score)\n    .filter((s): s is number => typeof s === 'number')\n  const runtimeScore =\n    runtimeScores.length > 0\n      ? runtimeScores.reduce((a, b) => a + b, 0) / runtimeScores.length\n      : null\n  const runtimePassed = runtime.filter((r) => r.outcome?.pass === true).length\n  const runtimePassRate = runtime.length > 0 ? runtimePassed / runtime.length : null\n\n  const kind: ProjectKind = runtime.length === 0 ? 'scaffold-only' : 'full'\n  const complete =\n    kind === 'scaffold-only'\n      ? metaScore !== null && buildScore !== null\n      : metaScore !== null && buildScore !== null && runtimeScore !== null\n\n  return {\n    projectId,\n    kind,\n    builderRunId: builder?.runId,\n    metaScore,\n    buildRunId: build?.runId,\n    buildScore,\n    appRuntimeRunIds: runtime.map((r) => r.runId),\n    runtimeScore,\n    runtimePassRate,\n    complete,\n  }\n}\n\n/** Aggregate scoring across every project in a corpus. */\nexport async function scoreAllProjects(store: TraceStore): Promise<ThreeLayerProjectReport[]> {\n  const runs = await store.listRuns()\n  const projectIds = [...new Set(runs.map((r) => r.projectId).filter((p): p is string => !!p))]\n  return Promise.all(projectIds.map((p) => scoreProject(store, p)))\n}\n\nfunction latestByLayer(runs: Run[], layer: Run['layer']): Run | undefined {\n  const filtered = runs.filter((r) => r.layer === layer).sort((a, b) => b.startedAt - a.startedAt)\n  return filtered[0]\n}\n\nasync function extractMetaScore(store: TraceStore, builderRunId: string): Promise<number | null> {\n  const js = await judgeSpans(store, builderRunId)\n  const meta = js.find(\n    (s) => s.judgeId === 'builder-meta' && s.dimension === 'user_intent_satisfaction',\n  )\n  if (!meta) return null\n  // Normalize score to 0..1. Accept 0-1 natively; 0-10 scale is also common.\n  if (meta.score >= 0 && meta.score <= 1) return meta.score\n  if (meta.score >= 0 && meta.score <= 10) return meta.score / 10\n  return null\n}\n"],"mappings":";;;;;;AAkCA,eAAsB,sBACpB,UACA,OACA,UAAgC,CAAC,GACH;CAC9B,MAAM,UAAU,IAAI,aAAa,KAAK;CACtC,MAAM,QAAQ,SAAS;EACrB,YAAY,SAAS;EACrB,WAAW,QAAQ;EACnB,gBAAgB,SAAS;EACzB,MAAM,SAAS;EACf,GAAG,QAAQ;CACb,CAAC;CACD,MAAM,SAAS,MAAM,IAAI,eAAe,QAAQ,MAAM,CAAC,CAAC,IAAI,SAAS,SAAS,OAAO;CACrF,MAAM,OAAO,OAAO,UAAU,OAAO,UAAU,SAAS,iBAAiB;CACzE,MAAM,UAAW;EAAC;EAAS;EAAO;CAAM,CAAC,CAAW,MACjD,UAAU,OAAO,UAAU,OAAO,MAAM,CAAE,aAAa,CAC1D;CACA,MAAM,eAA6B,OAC/B,YACA,YAAY,SACV,iBACA,UACE,oBACA;CACR,MAAM,QAAQ,OAAO;EACnB;EACA,OAAO,OAAO;EACd;EACA,OAAO,OAAO,KAAA,IAAY,iBAAiB,MAAM;CACnD,CAAC;CACD,OAAO;EACL,OAAO,QAAQ;EACf;EACA,SAAS;EACT;EACA,OAAO,OAAO;EACd;CACF;AACF;AAEA,SAAS,iBAAiB,QAAsC;CAC9D,MAAM,SAAU,CAAC,SAAS,KAAK,CAAC,CAAW,MACxC,UAAU,OAAO,UAAU,OAAO,MAAM,CAAE,aAAa,CAC1D;CACA,IAAI,QAAQ,OAAO,GAAG,OAAO,gBAAgB,OAAO,OAAO,CAAE;CAC7D,MAAM,OAAO,OAAO;CACpB,OAAO,CAAC,OACJ,oBACA,KAAK,eAAe,KAAA,IAClB,UAAU,KAAK,eAAe,EAAE,GAAG,KAAK,eACxC,aAAa,KAAK;AAC1B;;;;;;;;;;;;;;;;;;;ACvCA,IAAa,iBAAb,MAA4B;CAC1B;CACA;CACA;CACA;CACA;CACA;CACA;CAEA,YAAY,OAAmB,MAA0B,QAAwB;EAC/E,KAAK,QAAQ;EACb,KAAK,YAAY,KAAK;EACtB,KAAK,SAAS,KAAK,UAAU,YAAY;EACzC,KAAK,gBAAgB;EACrB,KAAK,iBAAiB,IAAI,aAAa,KAAK;CAC9C;;CAGA,MAAM,UAAU,aAAa,GAAG,KAAK,UAAU,QAAyB;EACtE,MAAM,KAAK,eAAe,SAAS;GACjC;GACA,WAAW,KAAK;GAChB,QAAQ,KAAK;GACb,OAAO;EACT,CAAC;EACD,KAAK,eAAe,KAAK,eAAe;EACxC,OAAO,KAAK;CACd;;CAGA,IAAI,UAAwB;EAC1B,IAAI,CAAC,KAAK,cAAc,MAAM,IAAI,MAAM,gDAAgD;EACxF,OAAO,KAAK;CACd;;;;;CAMA,MAAM,KAAK,SAAgF;EACzF,IAAI,CAAC,KAAK,cAAc,MAAM,IAAI,MAAM,6CAA6C;EACrF,MAAM,eAAe,IAAI,aAAa,KAAK,KAAK;EAChD,MAAM,aAAa,SAAS;GAC1B,YAAY,QAAQ,cAAc,GAAG,KAAK,UAAU;GACpD,WAAW,KAAK;GAChB,QAAQ,KAAK;GACb,aAAa,KAAK;GAClB,OAAO;EACT,CAAC;EAED,MAAM,SAAS,MAAM,IADD,eAAe,QAAQ,UAAU,KAAK,aAC/B,CAAC,CAAC,IAAI,QAAQ,SAAS,YAAY;EAC9D,MAAM,aAAa,OAAO;GACxB,MAAM,OAAO;GACb,OAAO,OAAO;GACd,cAAc,OAAO,SAAS,YAAY;EAC5C,CAAC;EACD,KAAK,iBAAiB,aAAa;EACnC,OAAO;GAAE,OAAO,aAAa;GAAO;EAAO;CAC7C;;;;;;;CAQA,MAAM,eAAe,SAA8D;EACjF,MAAM,cAAc,KAAK,kBAAkB,KAAK;EAChD,IAAI,CAAC,aACH,MAAM,IAAI,MAAM,gEAAgE;EAClF,MAAM,EAAE,UAAU,WAAW;EAC7B,MAAM,SAAS,MAAM,sBAAsB,UAAU,KAAK,OAAO;GAC/D,QAAQ,UAAU,KAAK;GACvB,YAAY;IAAE,SAAS,KAAA;IAAW,WAAW,KAAA;IAAW,kBAAkB,KAAA;GAAU;EACtF,CAAC;EAED,MAAM,KAAK,MAAM,UAAU,OAAO,OAAO;GACvC;GACA,WAAW,KAAK;GAChB,QAAQ,KAAK;GACb,OAAO;EACT,CAAC;EACD,OAAO;CACT;;;CAIA,MAAM,gBAAgB,OAAe,WAAmC;EACtE,IAAI,CAAC,KAAK,cACR,MAAM,IAAI,MAAM,wDAAwD;EAC1E,MAAM,KAAK,eAAe,YAAY;GACpC,SAAS;GACT,cAAc,KAAK;GACnB,WAAW;GACX;GACA;GACA,MAAM;EACR,CAAC;CACH;;CAGA,MAAM,QAAQ,SAA2E;EACvF,MAAM,KAAK,eAAe,OAAO;GAC/B,MAAM,QAAQ;GACd,OAAO,QAAQ;GACf,OAAO,QAAQ;EACjB,CAAC;CACH;;;;;;;;CASA,MAAM,gBAAgB,YAA2C;EAC/D,MAAM,cAAc,KAAK,kBAAkB,KAAK;EAChD,IAAI,CAAC,aACH,MAAM,IAAI,MACR,8EACF;EACF,MAAM,UAAU,IAAI,aAAa,KAAK,KAAK;EAC3C,MAAM,QAAQ,SAAS;GACrB;GACA,WAAW,KAAK;GAChB,QAAQ,KAAK;GACb;GACA,OAAO;EACT,CAAC;EACD,OAAO;CACT;;;;;;CAOA,MAAM,iBAAiB,MAKH;EAClB,IAAI,CAAC,KAAK,cACR,MAAM,IAAI,MAAM,yDAAyD;EAC3E,MAAM,UAAU,IAAI,aAAa,KAAK,KAAK;EAC3C,MAAM,QAAQ,SAAS;GACrB,YAAY,KAAK,cAAc,GAAG,KAAK,UAAU;GACjD,WAAW,KAAK;GAChB,QAAQ,KAAK;GACb,aAAa,KAAK;GAClB,OAAO;EACT,CAAC;EACD,MAAM,QAAQ,OAAO;GACnB,MAAM,KAAK;GACX,OAAO,KAAK;GACZ,cAAc,KAAK,OAAO,YAAY;GACtC,OAAO,KAAK;EACd,CAAC;EACD,KAAK,iBAAiB,QAAQ;EAC9B,OAAO,QAAQ;CACjB;CAEA,IAAI,sBAA0C;EAC5C,OAAO,KAAK;CACd;CACA,IAAI,oBAAwC;EAC1C,OAAO,KAAK;CACd;AACF;;;;;;AAOA,eAAsB,qBACpB,OACA,WAOC;CACD,MAAM,OAAO,MAAM,MAAM,SAAS,EAAE,UAAU,CAAC;CAC/C,MAAM,WAAW,KACd,QAAQ,MAAM,EAAE,UAAU,SAAS,CAAC,CACpC,MAAM,GAAG,MAAM,EAAE,YAAY,EAAE,SAAS;CAC3C,MAAM,YAAY,KACf,QAAQ,MAAM,EAAE,UAAU,WAAW,CAAC,CACtC,MAAM,GAAG,MAAM,EAAE,YAAY,EAAE,SAAS;CAC3C,MAAM,iBAAiB,KACpB,QAAQ,MAAM,EAAE,UAAU,aAAa,CAAC,CACxC,MAAM,GAAG,MAAM,EAAE,YAAY,EAAE,SAAS;CAC3C,OAAO;EACL;EACA;EACA,gBAAgB,SAAS;EACzB,cAAc,UAAU;EACxB,oBAAoB;CACtB;AACF;;;;;;;;;;;;;;;;;;ACxNA,SAAgB,gBAAgB,SAAuD;CACrF,MAAM,mBAAmB,QAAQ,QAAQ,MAAM,EAAE,QAAQ,CAAC,CAAC;CAC3D,OAAO;EACL,aAAa,SACX,UACC,MAAM,EAAE,YACR,MAAM,EAAE,UACX;EACA,eAAe,SACb,UACC,MAAM,EAAE,YACR,MAAM,EAAE,YACX;EACA,gBAAgB,SACd,UACC,MAAM,EAAE,aACR,MAAM,EAAE,YACX;EACA;CACF;AACF;AAEA,SAAS,SACP,SACA,GACA,GAC8B;CAC9B,MAAM,KAAe,CAAC;CACtB,MAAM,KAAe,CAAC;CACtB,KAAK,MAAM,KAAK,SAAS;EACvB,MAAM,IAAI,EAAE,CAAC;EACb,MAAM,IAAI,EAAE,CAAC;EACb,IAAI,MAAM,QAAQ,MAAM,QAAQ,OAAO,SAAS,CAAC,KAAK,OAAO,SAAS,CAAC,GAAG;GACxE,GAAG,KAAK,CAAC;GACT,GAAG,KAAK,CAAC;EACX;CACF;CACA,IAAI,GAAG,SAAS,GAAG,OAAO,KAAA;CAC1B,OAAO;EACL,GAAG,GAAG;EACN,SAAS,SAAS,IAAI,EAAE;EACxB,UAAU,UAAU,IAAI,EAAE;CAC5B;AACF;;ACjCA,IAAa,kBAAb,MAA6B;CACP;CAApB,YAAY,OAA2B;EAAnB,KAAA,QAAA;CAAoB;CAExC,MAAM,eAA0C;EAC9C,MAAM,OAAO,MAAM,KAAK,MAAM,SAAS;EACvC,MAAM,4BAAY,IAAI,IAAmB;EACzC,KAAK,MAAM,KAAK,MAAM;GACpB,IAAI,CAAC,EAAE,WAAW;GAClB,MAAM,MAAM,UAAU,IAAI,EAAE,SAAS,KAAK,CAAC;GAC3C,IAAI,KAAK,CAAC;GACV,UAAU,IAAI,EAAE,WAAW,GAAG;EAChC;EACA,MAAM,YAA8B,CAAC;EACrC,KAAK,MAAM,CAAC,WAAW,gBAAgB,WAAW;GAChD,MAAM,SAAS,YAAY,MAAM,CAAC,CAAC,MAAM,GAAG,MAAM,EAAE,YAAY,EAAE,SAAS;GAC3E,MAAM,QAAQ,YAAY,QAAQ,MAAM,EAAE,UAAU,SAAS;GAC7D,MAAM,SAAS,YAAY,QAAQ,MAAM,EAAE,UAAU,WAAW;GAChE,MAAM,WAAW,YAAY,QAAQ,MAAM,EAAE,UAAU,aAAa;GACpE,MAAM,SAAS,OAAO;GACtB,IAAI,CAAC,QAAQ;GACb,UAAU,KAAK;IACb;IACA,WAAW,MAAM;IACjB,YAAY,OAAO;IACnB,iBAAiB,SAAS;IAC1B,gBAAgB,OAAO;IACvB,cAAc,MAAM,EAAE,EAAE;IACxB,eAAe,OAAO,UAClB;KAAE,MAAM,OAAO,QAAQ,QAAQ;KAAO,OAAO,OAAO,QAAQ;IAAM,IAClE,KAAA;GACN,CAAC;EACH;EACA,OAAO,UAAU,MAAM,GAAG,MAAM,EAAE,iBAAiB,EAAE,cAAc;CACrE;CAEA,MAAM,gBAAgB,WAAoD;EAGxE,QADgB,MADG,KAAK,MAAM,SAAS,EAAE,UAAU,CAAC,EAAA,CAC/B,MAAM,CAAC,CAAC,MAAM,GAAG,MAAM,EAAE,YAAY,EAAE,SAC/C,CAAC,CAAC,KAAK,SAAS;GAC3B;GACA,aACE,IAAI,UAAU,YACV,SACA,IAAI,UAAU,cACZ,UACA,IAAI,UAAU,gBACZ,YACA;EACZ,EAAE;CACJ;CAEA,MAAM,aAAa,WAA2C;EAC5D,MAAM,eAAe,MAAM,KAAK,MAAM,SAAS;GAAE;GAAW,OAAO;EAAU,CAAC,EAAA,CAAG,MAC9E,GAAG,MAAM,EAAE,YAAY,EAAE,SAC5B;EACA,MAAM,cAAc,OAAO,UAAkB,KAAK,MAAM,SAAS,EAAE,aAAa,MAAM,CAAC;EACvF,MAAM,MAAqB,CAAC;EAC5B,KAAK,MAAM,OAAO,aAAa;GAC7B,MAAM,QAAQ,MAAM,KAAK,MAAM,MAAM,EAAE,OAAO,IAAI,MAAM,CAAC;GACzD,MAAM,WAAW,MAAM,YAAY,IAAI,KAAK;GAC5C,MAAM,QAAQ,SAAS,MAAM,MAAM,EAAE,UAAU,WAAW;GAC1D,MAAM,UAAoB,CAAC;GAG3B,IAAI,OAAO;IACT,MAAM,SAAS,MAAM,YAAY,MAAM,KAAK;IAC5C,KAAK,MAAM,KAAK,QAAQ,IAAI,EAAE,UAAU,eAAe,QAAQ,KAAK,EAAE,KAAK;GAC7E;GACA,KAAK,MAAM,KAAK,UAAU,IAAI,EAAE,UAAU,eAAe,QAAQ,KAAK,EAAE,KAAK;GAC7E,IAAI,KAAK;IACP,QAAQ,IAAI,UAAU,IAAI;IAC1B;IACA,cAAc,IAAI;IAClB,WAAW,IAAI;IACf,SAAS,IAAI;IACb,QAAQ,IAAI;IACZ,SAAS,IAAI;IACb,UAAU,MAAM,QAAQ,MAAM,EAAE,SAAS,KAAK,CAAC,CAAC;IAChD,WAAW,MAAM,QAAQ,MAAM,EAAE,SAAS,MAAM,CAAC,CAAC;IAClD,YAAY,OAAO;IACnB,kBAAkB;GACpB,CAAC;EACH;EACA,OAAO;CACT;AACF;;;;;;;;;;;;;;;;;;;;;;;;;ACzEA,eAAsB,aACpB,OACA,WACkC;CAClC,MAAM,UAAU,MAAM,MAAM,SAAS,EAAE,UAAU,CAAC;CAClD,MAAM,UAAU,cAAc,SAAS,SAAS;CAChD,MAAM,QAAQ,cAAc,SAAS,WAAW;CAChD,MAAM,UAAU,QAAQ,QAAQ,MAAM,EAAE,UAAU,aAAa;CAE/D,MAAM,YAAY,UAAU,MAAM,iBAAiB,OAAO,QAAQ,KAAK,IAAI;CAC3E,MAAM,aAAa,OAAO,SAAS,SAAS;CAC5C,MAAM,gBAAgB,QACnB,KAAK,MAAM,EAAE,SAAS,KAAK,CAAC,CAC5B,QAAQ,MAAmB,OAAO,MAAM,QAAQ;CACnD,MAAM,eACJ,cAAc,SAAS,IACnB,cAAc,QAAQ,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,cAAc,SACzD;CACN,MAAM,gBAAgB,QAAQ,QAAQ,MAAM,EAAE,SAAS,SAAS,IAAI,CAAC,CAAC;CACtE,MAAM,kBAAkB,QAAQ,SAAS,IAAI,gBAAgB,QAAQ,SAAS;CAE9E,MAAM,OAAoB,QAAQ,WAAW,IAAI,kBAAkB;CACnE,MAAM,WACJ,SAAS,kBACL,cAAc,QAAQ,eAAe,OACrC,cAAc,QAAQ,eAAe,QAAQ,iBAAiB;CAEpE,OAAO;EACL;EACA;EACA,cAAc,SAAS;EACvB;EACA,YAAY,OAAO;EACnB;EACA,kBAAkB,QAAQ,KAAK,MAAM,EAAE,KAAK;EAC5C;EACA;EACA;CACF;AACF;;AAGA,eAAsB,iBAAiB,OAAuD;CAC5F,MAAM,OAAO,MAAM,MAAM,SAAS;CAClC,MAAM,aAAa,CAAC,GAAG,IAAI,IAAI,KAAK,KAAK,MAAM,EAAE,SAAS,CAAC,CAAC,QAAQ,MAAmB,CAAC,CAAC,CAAC,CAAC,CAAC;CAC5F,OAAO,QAAQ,IAAI,WAAW,KAAK,MAAM,aAAa,OAAO,CAAC,CAAC,CAAC;AAClE;AAEA,SAAS,cAAc,MAAa,OAAsC;CAExE,OADiB,KAAK,QAAQ,MAAM,EAAE,UAAU,KAAK,CAAC,CAAC,MAAM,GAAG,MAAM,EAAE,YAAY,EAAE,SACxE,CAAC,CAAC;AAClB;AAEA,eAAe,iBAAiB,OAAmB,cAA8C;CAE/F,MAAM,QAAO,MADI,WAAW,OAAO,YAAY,EAAA,CAC/B,MACb,MAAM,EAAE,YAAY,kBAAkB,EAAE,cAAc,0BACzD;CACA,IAAI,CAAC,MAAM,OAAO;CAElB,IAAI,KAAK,SAAS,KAAK,KAAK,SAAS,GAAG,OAAO,KAAK;CACpD,IAAI,KAAK,SAAS,KAAK,KAAK,SAAS,IAAI,OAAO,KAAK,QAAQ;CAC7D,OAAO;AACT"}

package/dist/campaign/index.d.ts

@@ -7,4 +7,5 @@ import { $r as ExternalOptimizerObservationArtifact, $t as CrossSurfaceCandidate
 import { A as LabeledScenarioWrite, B as ScoredSurfaceOutcome, C as JudgeDimension, D as LabeledScenarioSampleArgs, E as LabeledScenarioRecord, F as ProposeContext, G as SearchProposerView, H as SurfaceProposer, I as ProposedCandidate, J as searchProposerView, K as SearchSummaryOptions, L as RedactionStatus, M as OptimizerConfig, N as ParetoParent, O as LabeledScenarioSource, P as ProposalTrackContext, R as Scenario, S as JudgeConfig, T as LabelTrust, U as TraceSpan, V as SessionScript, W as isProposedCandidate, X as SearchStateView, Y as SearchState, Z as searchUnitScores, _ as GateDecision, a as CampaignResult, at as SearchCompletion, b as GenerationRecord, c as CampaignTraceWriter, ct as SearchOperation, d as DispatchContext, dt as SearchUnitScore, f as DispatchFn, ft as searchCellId, g as GateContribution, h as GateContext, i as CampaignCostMeter, it as SearchCell, j as MutableSurface, k as LabeledScenarioStore, l as CodeSurface, lt as SearchScoredCell, m as GateCheckStatus, mt as searchTaskSetDigest, n as CampaignArtifactWriter, o as CampaignScenarioIdentity, ot as SearchDecisionRecord, p as Gate, pt as searchNodeId, q as renderSearchSummary, r as CampaignCellResult, s as CampaignTokenUsage, st as SearchNode, t as CampaignAggregates, u as ComponentSurface, ut as SearchSpend, v as GateResult, w as JudgeScore, x as JudgeAggregate, y as GenerationCandidate, z as ScenarioAggregate } from "../types-BF-xQGQn.js";
 import { A as SearchRungDecision, At as SearchHistoryReceipt, B as beam, Bt as openSearchLedger, C as searchClaimDecision, Ct as surfaceNode, D as SearchAllocationView, Dt as SearchHistoryCoverage, E as AshaOptions, Et as SearchHistoryAuditSummary, F as SearchExpansion, Ft as FileSearchLedger, G as NodePosterior, Gt as createRunCostLedger, H as incumbent, Ht as replaySearchLedgerText, I as SearchPolicy, It as OpenSearchLedgerOptions, J as estimateNode, K as SEARCH_ESTIMATOR, Kt as fsCampaignStorage, L as SearchPolicyNode, Lt as SearchLedger, M as asha, Mt as createSearchHistoryReceipt, N as uniform, Nt as verifySearchHistoryArtifact, O as SearchAllocator, Ot as SearchHistoryCoverageRow, P as AideOptions, Pt as verifySearchHistoryReceipt, Q as searchPosterior, R as SearchPolicyView, Rt as SearchLedgerAppendResult, S as planSearchClaim, St as surfaceDiff, T as verifySearchClaim, Tt as SearchHistoryAdmissionOptions, U as incumbentWithOperatorBandit, Ut as validateSearchLedgerEvent, V as crowdedFrontierParent, Vt as parseSearchLedgerLine, W as EstimateNodeCellsInput, Wt as CampaignStorage, X as searchCellSetDigest, Y as estimateNodeFromCells, Z as searchEstimateMethod, _ as SEARCH_CLAIM_RULE, _t as SearchRecorderOptions, a as SearchArtifactCodec, b as SearchClaimVerification, bt as developmentClaim, c as SearchExecutor, d as SearchProposalRequest, dt as AllocateSearchCellInput, f as SearchProposalResult, ft as RecordSearchEdgeInput, g as PlanSearchClaimInput, gt as SearchRecorder, h as SearchRunResult, ht as SearchOpening, i as RunSearchOptions, j as SearchUnitExtension, jt as assertSearchHistoryMatchesState, k as SearchCellPlan, kt as SearchHistoryPolicy, l as SearchLane, m as SearchProposerPort, mt as SearchLedgerBinding, n as searchDivergence, o as SearchCellResult, p as SearchProposedChild, pt as RegisterSearchNodeInput, q as SearchPosterior, qt as inMemoryCampaignStorage, r as searchPolicyView, s as SearchCellWork, t as runSearch, u as SearchProposalBlob, v as SearchClaimDecision, vt as SearchRunIdentity, w as searchClaimReserveUsd, wt as CreateSearchHistoryReceiptInput, x as decideSearchClaim, xt as searchEdgeId, y as SearchClaimPlan, yt as SettleSearchCellInput, z as aide, zt as SearchLedgerTrustedHeadMode } from "../search-kernel-BjR2Sr7w.js";
 import { B as ExternalOptimizerModelCallRequest, F as ExternalOptimizerEndpointFormat, G as ExternalOptimizerRunnerCommand, H as ExternalOptimizerModelExecutionObservation, I as ExternalOptimizerEvaluationObservation, L as ExternalOptimizerEvaluationRefusalReason, M as DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS, N as ExternalOptimizerCallbackLimits, P as ExternalOptimizerChatRequest, R as ExternalOptimizerModelBudget, U as ExternalOptimizerProcessLimits, V as ExternalOptimizerModelCallResult, j as DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, q as ExternalTextCandidate, z as ExternalOptimizerModelCall } from "../agent-profile-CoD1Dotf.js";
-export { type AideOptions, type AllocateSearchCellInput, type AnalyzeCrossSurfaceInteractionsInput, type AshaOptions, type AutoevalsScoreLike, type AutoevalsScorerLike, type AxisEvidence, type AxisVerdict, type BuildEvidenceVectorOptions, type BuildLoopProvenanceArgs, type BuildTraceAnalystSurfaceDispatchOptions, type CacheIssueReason, type CacheRead, type CampaignAggregates, type CampaignArtifactWriter, type CampaignBreakdown, type CampaignCellFailureReceipt, type CampaignCellResult, type CampaignCellRetryPolicy, type CampaignCostMeter, type CampaignCoverage, type CampaignResult, type CampaignRunPlan, type CampaignRunPlanCell, type CampaignScenarioIdentity, type CampaignStorage, type CampaignTokenUsage, type CampaignTraceWriter, type CellScheduleSlot, type CodeSurface, type CodeSurfaceVerification, type CompareOptimizationMethodsOptions, type ComparisonCost, type ComponentSurface, type CostBreakdown, type CostLedgerHandle, type CreateProfileMatrixPlanOptions, type CreateSearchHistoryReceiptInput, type CrossSurfaceAdditionDecision, type CrossSurfaceAdditionRejectionReason, type CrossSurfaceAttemptCompleteness, type CrossSurfaceBestSingleSelection, type CrossSurfaceBootstrapPolicy, type CrossSurfaceCandidate, type CrossSurfaceCandidateComparison, type CrossSurfaceCandidateEvidence, type CrossSurfaceCandidateOutcome, type CrossSurfaceCandidateSummary, type CrossSurfaceComponent, type CrossSurfaceComponentEvidence, type CrossSurfaceCompositionStep, type CrossSurfaceDistribution, type CrossSurfaceEligibility, type CrossSurfaceEvidenceBreakdown, type CrossSurfaceIneligibilityReason, type CrossSurfaceInteractionAwareSelection, type CrossSurfaceInteractionEffect, type CrossSurfaceInteractionPath, type CrossSurfaceInteractionReport, type CrossSurfaceInteractionTask, type CrossSurfaceNaiveStackSelection, type CrossSurfacePairCompatibility, type CrossSurfacePairEvidence, type CrossSurfacePairIncompatibilityReason, type CrossSurfacePairwiseEntry, type CrossSurfaceRankedSingle, type CrossSurfaceRelativeCost, type CrossSurfaceSelectionPolicy, type CrossSurfaceSelections, type CrossSurfaceTaskRow, DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS, type DefaultProductionGateCheck, type DefaultProductionGateOptions, type DefaultProductionRewardHackingOptions, type DimensionRegression, type DiscriminationScore, type DispatchContext, type DispatchFn, type EmitLoopProvenanceArgs, type EmitLoopProvenanceResult, type EstimateNodeCellsInput, type EvalFixture, type EvalFixtureFile, type EvalFixtureLoadOptions, type EvalFixtureRunPlan, type EvalFixtureScenario, type EvalFixtureValidationMode, type EvidenceVector, type ExternalOptimizationExample, type ExternalOptimizerCallbackLimits, type ExternalOptimizerChatRequest, type ExternalOptimizerEndpointFormat, type ExternalOptimizerEvaluationObservation, type ExternalOptimizerEvaluationRefusalReason, type ExternalOptimizerExecutionSummary, type ExternalOptimizerModelBudget, type ExternalOptimizerModelCall, type ExternalOptimizerModelCallRequest, type ExternalOptimizerModelCallResult, type ExternalOptimizerModelExecutionObservation, type ExternalOptimizerObservationArtifact, type ExternalOptimizerObservationSummary, type ExternalOptimizerProcessLimits, type ExternalOptimizerRunnerCommand, type ExternalOptimizerSubmittedCandidate, type ExternalSearchOptimizer, type ExternalTextCandidate, type ExternalTextEvaluationResponse, type ExternalTextOptimizationMethodConfig, type ExternalTextOptimizerContext, type ExternalTextOptimizerResult, FileSearchLedger, type FinalEvidencePolicy, type FinalEvidenceUse, type FinalizeProfileMatrixOptions, type FinalizedProfileMatrixResult, type FsLabeledScenarioStoreOptions, type Gate, type GateCheckStatus, type GateContext, type GateContribution, type GateDecision, type GateResult, type GenerationCandidate, type GenerationRecord, type GepaAdaptiveEngineRun, type GepaCandidatePopulationArtifact, type GepaCandidatePopulationCandidate, type GepaCandidatePopulationSummary, type GepaCandidateSelectionScore, type GepaEngineOptions, type GepaEngineRun, type GepaEvaluationImport, type GepaOptimizationMethodConfig, type GepaOptimizationRecipe, type GepaPopulationImport, type GepaRunnerCommand, type GitWorktreeAdapterOptions, type HeldOutGateOptions, type HeldoutSignificance, type HeldoutSignificanceOptions, type JudgeAggregate, type JudgeConfig, type JudgeDimension, type JudgeScore, type LabelTrust, type LabeledScenarioRecord, type LabeledScenarioSampleArgs, type LabeledScenarioSource, type LabeledScenarioStore, type LabeledScenarioWrite, type LlmJudgeDimension, type LlmJudgeOptions, type LoadEvalFixtureScenariosOptions, type LoopProvenanceArgsFromResult, type LoopProvenanceBackend, type LoopProvenanceCandidate, type LoopProvenanceEvidence, type LoopProvenanceOptimizationMethod, type LoopProvenanceRecord, type MutableSurface, type NeutralizationGateOptions, type NodeEstimate, type NodePosterior, type ObjectiveSource, type OpenAICompatibleOptimizerModel, type OpenAutoPrOptions, type OpenAutoPrResult, type OpenSearchLedgerOptions, type OptimizationMethod, type OptimizationMethodComparison, type OptimizationMethodComposition, type OptimizationMethodInput, type OptimizationMethodPairwise, type OptimizationMethodProvenance, type OptimizationMethodResult, type OptimizationMethodRunOptions, type OptimizationMethodScore, type OptimizationPackageSource, type OptimizationTokenUsage, type OptimizerConfig, type OptimizerModelBudget, type PairedHoldout, type ParetoParent, type ParetoSignificanceGateOptions, type PendingCostCallView, type PhoenixEvaluationResultLike, type PhoenixEvaluatorLike, type PlanCampaignRunOptions, type PlanEvalFixtureRunOptions, type PlanSearchClaimInput, type PlaybackContext, type PlaybackDriver, type PlaybackStep, type PowerPreflight, type PowerPreflightOptions, type PremeasuredOptimizationBaseline, type ProfileDispatchFn, type ProfileMatrixCoverage, type ProfileMatrixPlan, type ProfileMatrixRow, type ProfileMatrixSegmentResult, type ProfileSummary, type PromotionObjective, type PromotionPolicy, type ProposalFinding, type ProposalFindingOrigin, type ProposalTrackContext, type ProposeContext, type ProposedCandidate, type RecordSearchEdgeInput, type RedactionStatus, type ReferenceEquivalenceJudgeOptions, type ReferenceEquivalenceScenario, type RegisterSearchNodeInput, type RetainedDispatchOptions, type RetainedDispatchOutcome, type RolloutArgumentDiff, type RolloutArgumentDiffOptions, type RolloutCall, type RunCampaignOptions, type RunEvalOptions, type RunImprovementLoopOptions, type RunImprovementLoopResult, type RunOptimizationOptions, type RunOptimizationResult, type RunProfileMatrixOptions, type RunProfileMatrixResult, type RunProfileMatrixSegmentOptions, type RunSearchOptions, SEARCH_CLAIM_RULE, SEARCH_ESTIMATOR, SEARCH_LEDGER_SCHEMA, type Scenario, type ScenarioAggregate, type ScenarioRollup, type ScenarioSignal, type ScoreboardRenderOptions, type ScoreboardRow, type ScoreboardSummary, type ScoredRollout, type ScoredSurfaceOutcome, type SearchAllocationView, type SearchAllocator, type SearchArtifactCodec, type SearchArtifactKind, type SearchArtifactRef, type SearchAttemptAccounting, type SearchAudit, type SearchBudget, type SearchCancelReason, type SearchCandidateSurface, type SearchCell, type SearchCellAllocatedEvent, type SearchCellCancelledEvent, type SearchCellPlan, type SearchCellResult, type SearchCellSettledEvent, type SearchCellStage, type SearchCellWork, type SearchClaim, type SearchClaimDecision, type SearchClaimEstimator, type SearchClaimFinalist, type SearchClaimPlan, type SearchClaimPower, type SearchClaimTest, type SearchClaimVerification, type SearchCloseReason, type SearchClosedEvent, type SearchCompletion, type SearchCostAccounting, type SearchDecisionRecord, type SearchEdgeAttribution, type SearchEdgeOperator, type SearchEdgeRecordedEvent, type SearchEstimateMethod, type SearchExecutionIdentity, type SearchExecutor, type SearchExpansion, type SearchFailureReason, type SearchHistoryAdmissionOptions, type SearchHistoryAuditSummary, type SearchHistoryCoverage, type SearchHistoryCoverageRow, type SearchHistoryPolicy, type SearchHistoryReceipt, type SearchLane, type SearchLedger, type SearchLedgerAppendResult, type SearchLedgerBinding, type SearchLedgerEntry, SearchLedgerError, type SearchLedgerEvent, type SearchLedgerHash, SearchLedgerIntegrityError, type SearchLedgerTrustedHeadMode, type SearchModelIdentity, type SearchNode, type SearchNodeDecidedEvent, type SearchNodeDecision, type SearchNodeRef, type SearchNodeRegisteredEvent, type SearchNodeStatus, type SearchOpenedEvent, type SearchOpening, type SearchOperation, type SearchOperationKind, type SearchOperationRecordedEvent, type SearchOperationStartedEvent, type SearchPolicy, type SearchPolicyNode, type SearchPolicyView, type SearchPosterior, type SearchProposalBlob, type SearchProposalRequest, type SearchProposalResult, type SearchProposedChild, type SearchProposer, type SearchProposerKind, type SearchProposerPort, type SearchProposerView, SearchRecorder, type SearchRecorderOptions, type SearchReservation, type SearchRunIdentity, type SearchRunResult, type SearchRungDecision, type SearchScoredCell, type SearchSourceRef, type SearchSpend, type SearchSplit, type SearchSplitTasks, type SearchSplits, SearchState, SearchStateView, type SearchSummaryOptions, type SearchSurfaceEffect, type SearchSurfaceEvidence, type SearchSurfaceKind, type SearchTask, type SearchTaskOutcome, type SearchTokenAccounting, type SearchTraceRef, type SearchUnitExtension, type SearchUnitScore, type SearchUnknown, type SequentialDecideFn, type SequentialDecideOptions, type SequentialGateDecision, type SequentialObservation, type SequentialPairedGate, type SequentialPairedGateOptions, type SequentialStreamState, type SeriesDistribution, type SessionScript, type SettleSearchCellInput, type SingleRunLock, type SingleRunLockOptions, type SkillOptOptimizationMethodConfig, type SkillOptRunnerCommand, type SkillOptTrainerConfig, type SurfaceProposer, type TraceAnalystArtifact, type TraceAnalystScenario, type TraceSpan, type TransientFailureOptions, type UngroundedLiteralReport, type UserStory, type UserStoryVerdict, type Worktree, type WorktreeAdapter, acquireSingleRunLock, aide, analyzeCrossSurfaceInteractions, asha, assertCampaignDesign, assertCodeSurfaceIdentity, assertSearchHistoryMatchesState, beam, buildCellSchedule, buildEvidenceVector, campaignBreakdown, campaignCellSearchResult, campaignCoverage, campaignMeanComposite, campaignScenarioIdentity, campaignSplitDigest, campaignSplitDigestFromIdentities, canonicalDigest, cellCachePath, codeSurfaceIdentityMaterial, compareOptimizationMethods, componentSurfaceIdentityMaterial, composeGate, costFromLedgerSummary, createProfileMatrixPlan, createRetainedDispatch, createRunCostLedger, createSearchHistoryReceipt, crowdedFrontierParent, decideSearchClaim, decodeExternalTextCandidate, defaultProductionGate, developmentClaim, dimensionRegressions, discoverEvalFixtures, emitLoopProvenance, estimateNode, estimateNodeFromCells, externalSurface, externalTextOptimizationMethod, finalizeProfileMatrix, fsCampaignStorage, gepaOptimizationMethod, gitWorktreeAdapter, heldOutGate, heldoutSignificance, importExternalEvaluations, importGepaPopulation, inMemoryCampaignStorage, incumbent, incumbentWithOperatorBandit, isProposedCandidate, isTransientTransportFailure, llmJudge, loadEvalFixture, loadEvalFixtureScenarios, makeProposalFinding, measureCampaignCost, neutralizationGate, neutralizeText, openSearchLedger, pairHoldout, paretoPolicy, paretoSignificanceGate, parseSearchLedgerLine, planCampaignRun, planEvalFixtureRun, planSearchClaim, powerPreflight, readExternalOptimizerObservationArtifact, readGepaCandidatePopulationArtifact, renderScoreboardMarkdown, renderSearchSummary, renderSurfaceDiff, replaySearchLedgerText, resolveRunDir, resolveWorktreePath, runCampaign, runEval, runImprovementLoop, runOptimization, runProfileMatrix, runProfileMatrixSegment, runSearch, scopedOptimizationMethod, scoreDiscrimination, scoreUserStory, scoreboardSummary, searchCellId, searchCellSetDigest, searchClaimDecision, searchClaimReserveUsd, searchDivergence, searchEdgeId, searchEstimateMethod, searchModelIdentity, searchNodeId, searchPolicyView, searchPosterior, searchProposalExecution, searchProposerView, searchReceiptAccounting, searchTaskSetDigest, searchUnitScores, sequentialDecide, sequentialOptimizationMethod, sequentialPairedGate, skillOptOptimizationMethod, surfaceContentHash, surfaceDiff, surfaceDispatchRef, surfaceHash, surfaceNode, transientDispatchFailure, uniform, userStoryScoreboard, validateSearchLedgerEvent, verifyCodeSurface, verifySearchClaim, verifySearchHistoryArtifact, verifySearchHistoryReceipt };
+import { a as HarnessScope, i as HarnessScenario, l as runSandboxScenario, o as HarnessTiming, r as HarnessReport, t as HarnessAssertion } from "../sandbox-harness--VUvMji7.js";
+export { type AideOptions, type AllocateSearchCellInput, type AnalyzeCrossSurfaceInteractionsInput, type AshaOptions, type AutoevalsScoreLike, type AutoevalsScorerLike, type AxisEvidence, type AxisVerdict, type BuildEvidenceVectorOptions, type BuildLoopProvenanceArgs, type BuildTraceAnalystSurfaceDispatchOptions, type CacheIssueReason, type CacheRead, type CampaignAggregates, type CampaignArtifactWriter, type CampaignBreakdown, type CampaignCellFailureReceipt, type CampaignCellResult, type CampaignCellRetryPolicy, type CampaignCostMeter, type CampaignCoverage, type CampaignResult, type CampaignRunPlan, type CampaignRunPlanCell, type CampaignScenarioIdentity, type CampaignStorage, type CampaignTokenUsage, type CampaignTraceWriter, type CellScheduleSlot, type CodeSurface, type CodeSurfaceVerification, type CompareOptimizationMethodsOptions, type ComparisonCost, type ComponentSurface, type CostBreakdown, type CostLedgerHandle, type CreateProfileMatrixPlanOptions, type CreateSearchHistoryReceiptInput, type CrossSurfaceAdditionDecision, type CrossSurfaceAdditionRejectionReason, type CrossSurfaceAttemptCompleteness, type CrossSurfaceBestSingleSelection, type CrossSurfaceBootstrapPolicy, type CrossSurfaceCandidate, type CrossSurfaceCandidateComparison, type CrossSurfaceCandidateEvidence, type CrossSurfaceCandidateOutcome, type CrossSurfaceCandidateSummary, type CrossSurfaceComponent, type CrossSurfaceComponentEvidence, type CrossSurfaceCompositionStep, type CrossSurfaceDistribution, type CrossSurfaceEligibility, type CrossSurfaceEvidenceBreakdown, type CrossSurfaceIneligibilityReason, type CrossSurfaceInteractionAwareSelection, type CrossSurfaceInteractionEffect, type CrossSurfaceInteractionPath, type CrossSurfaceInteractionReport, type CrossSurfaceInteractionTask, type CrossSurfaceNaiveStackSelection, type CrossSurfacePairCompatibility, type CrossSurfacePairEvidence, type CrossSurfacePairIncompatibilityReason, type CrossSurfacePairwiseEntry, type CrossSurfaceRankedSingle, type CrossSurfaceRelativeCost, type CrossSurfaceSelectionPolicy, type CrossSurfaceSelections, type CrossSurfaceTaskRow, DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS, type DefaultProductionGateCheck, type DefaultProductionGateOptions, type DefaultProductionRewardHackingOptions, type DimensionRegression, type DiscriminationScore, type DispatchContext, type DispatchFn, type EmitLoopProvenanceArgs, type EmitLoopProvenanceResult, type EstimateNodeCellsInput, type EvalFixture, type EvalFixtureFile, type EvalFixtureLoadOptions, type EvalFixtureRunPlan, type EvalFixtureScenario, type EvalFixtureValidationMode, type EvidenceVector, type ExternalOptimizationExample, type ExternalOptimizerCallbackLimits, type ExternalOptimizerChatRequest, type ExternalOptimizerEndpointFormat, type ExternalOptimizerEvaluationObservation, type ExternalOptimizerEvaluationRefusalReason, type ExternalOptimizerExecutionSummary, type ExternalOptimizerModelBudget, type ExternalOptimizerModelCall, type ExternalOptimizerModelCallRequest, type ExternalOptimizerModelCallResult, type ExternalOptimizerModelExecutionObservation, type ExternalOptimizerObservationArtifact, type ExternalOptimizerObservationSummary, type ExternalOptimizerProcessLimits, type ExternalOptimizerRunnerCommand, type ExternalOptimizerSubmittedCandidate, type ExternalSearchOptimizer, type ExternalTextCandidate, type ExternalTextEvaluationResponse, type ExternalTextOptimizationMethodConfig, type ExternalTextOptimizerContext, type ExternalTextOptimizerResult, FileSearchLedger, type FinalEvidencePolicy, type FinalEvidenceUse, type FinalizeProfileMatrixOptions, type FinalizedProfileMatrixResult, type FsLabeledScenarioStoreOptions, type Gate, type GateCheckStatus, type GateContext, type GateContribution, type GateDecision, type GateResult, type GenerationCandidate, type GenerationRecord, type GepaAdaptiveEngineRun, type GepaCandidatePopulationArtifact, type GepaCandidatePopulationCandidate, type GepaCandidatePopulationSummary, type GepaCandidateSelectionScore, type GepaEngineOptions, type GepaEngineRun, type GepaEvaluationImport, type GepaOptimizationMethodConfig, type GepaOptimizationRecipe, type GepaPopulationImport, type GepaRunnerCommand, type GitWorktreeAdapterOptions, type HarnessAssertion, type HarnessReport, type HarnessScenario, type HarnessScope, type HarnessTiming, type HeldOutGateOptions, type HeldoutSignificance, type HeldoutSignificanceOptions, type JudgeAggregate, type JudgeConfig, type JudgeDimension, type JudgeScore, type LabelTrust, type LabeledScenarioRecord, type LabeledScenarioSampleArgs, type LabeledScenarioSource, type LabeledScenarioStore, type LabeledScenarioWrite, type LlmJudgeDimension, type LlmJudgeOptions, type LoadEvalFixtureScenariosOptions, type LoopProvenanceArgsFromResult, type LoopProvenanceBackend, type LoopProvenanceCandidate, type LoopProvenanceEvidence, type LoopProvenanceOptimizationMethod, type LoopProvenanceRecord, type MutableSurface, type NeutralizationGateOptions, type NodeEstimate, type NodePosterior, type ObjectiveSource, type OpenAICompatibleOptimizerModel, type OpenAutoPrOptions, type OpenAutoPrResult, type OpenSearchLedgerOptions, type OptimizationMethod, type OptimizationMethodComparison, type OptimizationMethodComposition, type OptimizationMethodInput, type OptimizationMethodPairwise, type OptimizationMethodProvenance, type OptimizationMethodResult, type OptimizationMethodRunOptions, type OptimizationMethodScore, type OptimizationPackageSource, type OptimizationTokenUsage, type OptimizerConfig, type OptimizerModelBudget, type PairedHoldout, type ParetoParent, type ParetoSignificanceGateOptions, type PendingCostCallView, type PhoenixEvaluationResultLike, type PhoenixEvaluatorLike, type PlanCampaignRunOptions, type PlanEvalFixtureRunOptions, type PlanSearchClaimInput, type PlaybackContext, type PlaybackDriver, type PlaybackStep, type PowerPreflight, type PowerPreflightOptions, type PremeasuredOptimizationBaseline, type ProfileDispatchFn, type ProfileMatrixCoverage, type ProfileMatrixPlan, type ProfileMatrixRow, type ProfileMatrixSegmentResult, type ProfileSummary, type PromotionObjective, type PromotionPolicy, type ProposalFinding, type ProposalFindingOrigin, type ProposalTrackContext, type ProposeContext, type ProposedCandidate, type RecordSearchEdgeInput, type RedactionStatus, type ReferenceEquivalenceJudgeOptions, type ReferenceEquivalenceScenario, type RegisterSearchNodeInput, type RetainedDispatchOptions, type RetainedDispatchOutcome, type RolloutArgumentDiff, type RolloutArgumentDiffOptions, type RolloutCall, type RunCampaignOptions, type RunEvalOptions, type RunImprovementLoopOptions, type RunImprovementLoopResult, type RunOptimizationOptions, type RunOptimizationResult, type RunProfileMatrixOptions, type RunProfileMatrixResult, type RunProfileMatrixSegmentOptions, type RunSearchOptions, SEARCH_CLAIM_RULE, SEARCH_ESTIMATOR, SEARCH_LEDGER_SCHEMA, type Scenario, type ScenarioAggregate, type ScenarioRollup, type ScenarioSignal, type ScoreboardRenderOptions, type ScoreboardRow, type ScoreboardSummary, type ScoredRollout, type ScoredSurfaceOutcome, type SearchAllocationView, type SearchAllocator, type SearchArtifactCodec, type SearchArtifactKind, type SearchArtifactRef, type SearchAttemptAccounting, type SearchAudit, type SearchBudget, type SearchCancelReason, type SearchCandidateSurface, type SearchCell, type SearchCellAllocatedEvent, type SearchCellCancelledEvent, type SearchCellPlan, type SearchCellResult, type SearchCellSettledEvent, type SearchCellStage, type SearchCellWork, type SearchClaim, type SearchClaimDecision, type SearchClaimEstimator, type SearchClaimFinalist, type SearchClaimPlan, type SearchClaimPower, type SearchClaimTest, type SearchClaimVerification, type SearchCloseReason, type SearchClosedEvent, type SearchCompletion, type SearchCostAccounting, type SearchDecisionRecord, type SearchEdgeAttribution, type SearchEdgeOperator, type SearchEdgeRecordedEvent, type SearchEstimateMethod, type SearchExecutionIdentity, type SearchExecutor, type SearchExpansion, type SearchFailureReason, type SearchHistoryAdmissionOptions, type SearchHistoryAuditSummary, type SearchHistoryCoverage, type SearchHistoryCoverageRow, type SearchHistoryPolicy, type SearchHistoryReceipt, type SearchLane, type SearchLedger, type SearchLedgerAppendResult, type SearchLedgerBinding, type SearchLedgerEntry, SearchLedgerError, type SearchLedgerEvent, type SearchLedgerHash, SearchLedgerIntegrityError, type SearchLedgerTrustedHeadMode, type SearchModelIdentity, type SearchNode, type SearchNodeDecidedEvent, type SearchNodeDecision, type SearchNodeRef, type SearchNodeRegisteredEvent, type SearchNodeStatus, type SearchOpenedEvent, type SearchOpening, type SearchOperation, type SearchOperationKind, type SearchOperationRecordedEvent, type SearchOperationStartedEvent, type SearchPolicy, type SearchPolicyNode, type SearchPolicyView, type SearchPosterior, type SearchProposalBlob, type SearchProposalRequest, type SearchProposalResult, type SearchProposedChild, type SearchProposer, type SearchProposerKind, type SearchProposerPort, type SearchProposerView, SearchRecorder, type SearchRecorderOptions, type SearchReservation, type SearchRunIdentity, type SearchRunResult, type SearchRungDecision, type SearchScoredCell, type SearchSourceRef, type SearchSpend, type SearchSplit, type SearchSplitTasks, type SearchSplits, SearchState, SearchStateView, type SearchSummaryOptions, type SearchSurfaceEffect, type SearchSurfaceEvidence, type SearchSurfaceKind, type SearchTask, type SearchTaskOutcome, type SearchTokenAccounting, type SearchTraceRef, type SearchUnitExtension, type SearchUnitScore, type SearchUnknown, type SequentialDecideFn, type SequentialDecideOptions, type SequentialGateDecision, type SequentialObservation, type SequentialPairedGate, type SequentialPairedGateOptions, type SequentialStreamState, type SeriesDistribution, type SessionScript, type SettleSearchCellInput, type SingleRunLock, type SingleRunLockOptions, type SkillOptOptimizationMethodConfig, type SkillOptRunnerCommand, type SkillOptTrainerConfig, type SurfaceProposer, type TraceAnalystArtifact, type TraceAnalystScenario, type TraceSpan, type TransientFailureOptions, type UngroundedLiteralReport, type UserStory, type UserStoryVerdict, type Worktree, type WorktreeAdapter, acquireSingleRunLock, aide, analyzeCrossSurfaceInteractions, asha, assertCampaignDesign, assertCodeSurfaceIdentity, assertSearchHistoryMatchesState, beam, buildCellSchedule, buildEvidenceVector, campaignBreakdown, campaignCellSearchResult, campaignCoverage, campaignMeanComposite, campaignScenarioIdentity, campaignSplitDigest, campaignSplitDigestFromIdentities, canonicalDigest, cellCachePath, codeSurfaceIdentityMaterial, compareOptimizationMethods, componentSurfaceIdentityMaterial, composeGate, costFromLedgerSummary, createProfileMatrixPlan, createRetainedDispatch, createRunCostLedger, createSearchHistoryReceipt, crowdedFrontierParent, decideSearchClaim, decodeExternalTextCandidate, defaultProductionGate, developmentClaim, dimensionRegressions, discoverEvalFixtures, emitLoopProvenance, estimateNode, estimateNodeFromCells, externalSurface, externalTextOptimizationMethod, finalizeProfileMatrix, fsCampaignStorage, gepaOptimizationMethod, gitWorktreeAdapter, heldOutGate, heldoutSignificance, importExternalEvaluations, importGepaPopulation, inMemoryCampaignStorage, incumbent, incumbentWithOperatorBandit, isProposedCandidate, isTransientTransportFailure, llmJudge, loadEvalFixture, loadEvalFixtureScenarios, makeProposalFinding, measureCampaignCost, neutralizationGate, neutralizeText, openSearchLedger, pairHoldout, paretoPolicy, paretoSignificanceGate, parseSearchLedgerLine, planCampaignRun, planEvalFixtureRun, planSearchClaim, powerPreflight, readExternalOptimizerObservationArtifact, readGepaCandidatePopulationArtifact, renderScoreboardMarkdown, renderSearchSummary, renderSurfaceDiff, replaySearchLedgerText, resolveRunDir, resolveWorktreePath, runCampaign, runEval, runImprovementLoop, runOptimization, runProfileMatrix, runProfileMatrixSegment, runSandboxScenario, runSearch, scopedOptimizationMethod, scoreDiscrimination, scoreUserStory, scoreboardSummary, searchCellId, searchCellSetDigest, searchClaimDecision, searchClaimReserveUsd, searchDivergence, searchEdgeId, searchEstimateMethod, searchModelIdentity, searchNodeId, searchPolicyView, searchPosterior, searchProposalExecution, searchProposerView, searchReceiptAccounting, searchTaskSetDigest, searchUnitScores, sequentialDecide, sequentialOptimizationMethod, sequentialPairedGate, skillOptOptimizationMethod, surfaceContentHash, surfaceDiff, surfaceDispatchRef, surfaceHash, surfaceNode, transientDispatchFailure, uniform, userStoryScoreboard, validateSearchLedgerEvent, verifyCodeSurface, verifySearchClaim, verifySearchHistoryArtifact, verifySearchHistoryReceipt };

package/dist/campaign/index.js

@@ -9,8 +9,9 @@ import { C as SEARCH_ESTIMATOR, D as searchEstimateMethod, E as searchCellSetDig
 import { a as validateSearchLedgerEvent, c as SearchStateView, d as searchNodeId, f as searchTaskSetDigest, i as replaySearchLedgerText, l as searchUnitScores, n as openSearchLedger, o as SEARCH_LEDGER_SCHEMA, r as parseSearchLedgerLine, s as SearchState, t as FileSearchLedger, u as searchCellId } from "../search-ledger-BsVi-Ngv.js";
 import { t as acquireSingleRunLock } from "../single-run-lock-BGlW3OSk.js";
 import { n as searchProposerView, t as renderSearchSummary } from "../search-summary-DedyTdiM.js";
+import { n as runSandboxScenario } from "../sandbox-harness-BuacGaVn.js";
 import { C as measureCampaignCost, S as analyzeCrossSurfaceInteractions, _ as neutralizationGate, a as createRetainedDispatch, b as loadEvalFixtureScenarios, c as runProfileMatrixSegment, d as scoreUserStory, f as scoreboardSummary, g as neutralizeText, h as sequentialOptimizationMethod, i as scoreDiscrimination, l as runProfileMatrix, m as scopedOptimizationMethod, n as resolveWorktreePath, o as createProfileMatrixPlan, p as userStoryScoreboard, r as verifyCodeSurface, s as finalizeProfileMatrix, t as gitWorktreeAdapter, u as renderScoreboardMarkdown, v as discoverEvalFixtures, x as planEvalFixtureRun, y as loadEvalFixture } from "../campaign-DKZiYHUq.js";
 import { a as importExternalEvaluations, c as composeGate, d as readExternalOptimizerObservationArtifact, i as externalSurface, l as externalTextOptimizationMethod, n as compareOptimizationMethods, o as importGepaPopulation, r as gepaOptimizationMethod, s as heldOutGate, t as skillOptOptimizationMethod, u as decodeExternalTextCandidate } from "../skillopt-optimization-method-DmfZz-2d.js";
 import { n as paretoPolicy, r as paretoSignificanceGate, t as buildEvidenceVector } from "../promotion-policy-f79oklG_.js";
 import { n as sequentialPairedGate, t as sequentialDecide } from "../sequential-Zp1tsxqI.js";
-export { DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS, FileSearchLedger, SEARCH_CLAIM_RULE, SEARCH_ESTIMATOR, SEARCH_LEDGER_SCHEMA, SearchLedgerError, SearchLedgerIntegrityError, SearchRecorder, SearchState, SearchStateView, acquireSingleRunLock, aide, analyzeCrossSurfaceInteractions, asha, assertCampaignDesign, assertCodeSurfaceIdentity, assertSearchHistoryMatchesState, beam, buildCellSchedule, buildEvidenceVector, campaignBreakdown, campaignCellSearchResult, campaignCoverage, campaignMeanComposite, campaignScenarioIdentity, campaignSplitDigest, campaignSplitDigestFromIdentities, canonicalDigest, cellCachePath, codeSurfaceIdentityMaterial, compareOptimizationMethods, componentSurfaceIdentityMaterial, composeGate, costFromLedgerSummary, createProfileMatrixPlan, createRetainedDispatch, createRunCostLedger, createSearchHistoryReceipt, crowdedFrontierParent, decideSearchClaim, decodeExternalTextCandidate, defaultProductionGate, developmentClaim, dimensionRegressions, discoverEvalFixtures, emitLoopProvenance, estimateNode, estimateNodeFromCells, externalSurface, externalTextOptimizationMethod, finalizeProfileMatrix, fsCampaignStorage, gepaOptimizationMethod, gitWorktreeAdapter, heldOutGate, heldoutSignificance, importExternalEvaluations, importGepaPopulation, inMemoryCampaignStorage, incumbent, incumbentWithOperatorBandit, isProposedCandidate, isTransientTransportFailure, llmJudge, loadEvalFixture, loadEvalFixtureScenarios, makeProposalFinding, measureCampaignCost, neutralizationGate, neutralizeText, openSearchLedger, pairHoldout, paretoPolicy, paretoSignificanceGate, parseSearchLedgerLine, planCampaignRun, planEvalFixtureRun, planSearchClaim, powerPreflight, readExternalOptimizerObservationArtifact, readGepaCandidatePopulationArtifact, renderScoreboardMarkdown, renderSearchSummary, renderSurfaceDiff, replaySearchLedgerText, resolveRunDir, resolveWorktreePath, runCampaign, runEval, runImprovementLoop, runOptimization, runProfileMatrix, runProfileMatrixSegment, runSearch, scopedOptimizationMethod, scoreDiscrimination, scoreUserStory, scoreboardSummary, searchCellId, searchCellSetDigest, searchClaimDecision, searchClaimReserveUsd, searchDivergence, searchEdgeId, searchEstimateMethod, searchModelIdentity, searchNodeId, searchPolicyView, searchPosterior, searchProposalExecution, searchProposerView, searchReceiptAccounting, searchTaskSetDigest, searchUnitScores, sequentialDecide, sequentialOptimizationMethod, sequentialPairedGate, skillOptOptimizationMethod, surfaceContentHash, surfaceDiff, surfaceDispatchRef, surfaceHash, surfaceNode, transientDispatchFailure, uniform, userStoryScoreboard, validateSearchLedgerEvent, verifyCodeSurface, verifySearchClaim, verifySearchHistoryArtifact, verifySearchHistoryReceipt };
+export { DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS, FileSearchLedger, SEARCH_CLAIM_RULE, SEARCH_ESTIMATOR, SEARCH_LEDGER_SCHEMA, SearchLedgerError, SearchLedgerIntegrityError, SearchRecorder, SearchState, SearchStateView, acquireSingleRunLock, aide, analyzeCrossSurfaceInteractions, asha, assertCampaignDesign, assertCodeSurfaceIdentity, assertSearchHistoryMatchesState, beam, buildCellSchedule, buildEvidenceVector, campaignBreakdown, campaignCellSearchResult, campaignCoverage, campaignMeanComposite, campaignScenarioIdentity, campaignSplitDigest, campaignSplitDigestFromIdentities, canonicalDigest, cellCachePath, codeSurfaceIdentityMaterial, compareOptimizationMethods, componentSurfaceIdentityMaterial, composeGate, costFromLedgerSummary, createProfileMatrixPlan, createRetainedDispatch, createRunCostLedger, createSearchHistoryReceipt, crowdedFrontierParent, decideSearchClaim, decodeExternalTextCandidate, defaultProductionGate, developmentClaim, dimensionRegressions, discoverEvalFixtures, emitLoopProvenance, estimateNode, estimateNodeFromCells, externalSurface, externalTextOptimizationMethod, finalizeProfileMatrix, fsCampaignStorage, gepaOptimizationMethod, gitWorktreeAdapter, heldOutGate, heldoutSignificance, importExternalEvaluations, importGepaPopulation, inMemoryCampaignStorage, incumbent, incumbentWithOperatorBandit, isProposedCandidate, isTransientTransportFailure, llmJudge, loadEvalFixture, loadEvalFixtureScenarios, makeProposalFinding, measureCampaignCost, neutralizationGate, neutralizeText, openSearchLedger, pairHoldout, paretoPolicy, paretoSignificanceGate, parseSearchLedgerLine, planCampaignRun, planEvalFixtureRun, planSearchClaim, powerPreflight, readExternalOptimizerObservationArtifact, readGepaCandidatePopulationArtifact, renderScoreboardMarkdown, renderSearchSummary, renderSurfaceDiff, replaySearchLedgerText, resolveRunDir, resolveWorktreePath, runCampaign, runEval, runImprovementLoop, runOptimization, runProfileMatrix, runProfileMatrixSegment, runSandboxScenario, runSearch, scopedOptimizationMethod, scoreDiscrimination, scoreUserStory, scoreboardSummary, searchCellId, searchCellSetDigest, searchClaimDecision, searchClaimReserveUsd, searchDivergence, searchEdgeId, searchEstimateMethod, searchModelIdentity, searchNodeId, searchPolicyView, searchPosterior, searchProposalExecution, searchProposerView, searchReceiptAccounting, searchTaskSetDigest, searchUnitScores, sequentialDecide, sequentialOptimizationMethod, sequentialPairedGate, skillOptOptimizationMethod, surfaceContentHash, surfaceDiff, surfaceDispatchRef, surfaceHash, surfaceNode, transientDispatchFailure, uniform, userStoryScoreboard, validateSearchLedgerEvent, verifyCodeSurface, verifySearchClaim, verifySearchHistoryArtifact, verifySearchHistoryReceipt };

package/dist/{feedback-trajectory-Ck7K7AhK.d.ts → feedback-trajectory-BgomnFOe.d.ts}

@@ -2,8 +2,8 @@ import { o as FailureClass } from "./schema-CkIvnmLU.js";
 import { i as AnalystFinding, m as EvidenceRef, u as AnalystRunResult } from "./types-DzCO-S1m.js";
 import { s as TraceStore } from "./store-CMs5wGXP.js";
 import { i as DatasetSplit, r as DatasetScenario } from "./dataset-DThC1S6G.js";
-import "./exact-types-CIViGj98.js";
 import { o as TraceEmitter } from "./emitter-mAXC5PSN.js";
+import "./exact-types-CIViGj98.js";
 type ControlSeverity = 'info' | 'warning' | 'error' | 'critical';
 type ControlActionFailureMode = 'continue' | 'stop';
 interface ControlEvalResult {
@@ -383,4 +383,4 @@ declare function controlRunToFeedbackTrajectory<TState, TAction, TActionResult>(
   createdAt?: string;
 }): FeedbackTrajectory;
 export { AnalystFindingDigest as A, ControlRuntimeConfig as B, createFeedbackTrajectory as C, renderPreferenceMemoryMarkdown as D, feedbackTrajectoryToOptimizerRow as E, ControlBudget as F, ControlSeverity as H, ControlContext as I, ControlDecision as L, AnalystRunDigest as M, analystRunDigest as N, summarizePreferenceMemory as O, ControlActionOutcome as P, ControlEvalResult as R, controlRunToFeedbackTrajectory as S, feedbackTrajectoriesToOptimizerRows as T, ControlStep as U, ControlRuntimeError as V, StopDecision as W, PreferenceMemoryEntry as _, FeedbackLabel as a, analystRunToReviewRequests as b, FeedbackOptimizerRow as c, FeedbackTask as d, FeedbackTrajectory as f, InMemoryFeedbackTrajectoryStore as g, FileSystemFeedbackTrajectoryStore as h, FeedbackAttempt as i, AnalystReviewDecision as j, withAssignedFeedbackSplit as k, FeedbackOutcome as l, FeedbackTrajectoryStore as m, AnalystReviewRequest as n, FeedbackLabelKind as o, FeedbackTrajectoryFilter as p, FeedbackArtifactType as r, FeedbackLabelSource as s, AnalystFeedbackTrajectoryOptions as t, FeedbackSplitPolicy as u, ProposedSideEffect as v, feedbackTrajectoriesToDatasetScenarios as w, assignFeedbackSplit as x, analystRunToFeedbackTrajectory as y, ControlRunResult as z };
-//# sourceMappingURL=feedback-trajectory-Ck7K7AhK.d.ts.map
+//# sourceMappingURL=feedback-trajectory-BgomnFOe.d.ts.map

package/dist/{feedback-trajectory-Ck7K7AhK.d.ts.map → feedback-trajectory-BgomnFOe.d.ts.map}

@@ -1 +1 @@
-{"version":3,"file":"feedback-trajectory-Ck7K7AhK.d.ts","names":[],"sources":["../src/control-runtime/contract.ts","../src/feedback-trajectory-review.ts","../src/feedback-trajectory.ts"],"mappings":";;;;;;KAIY;KAEA;UAEK;;EAEf;;EAEA;;EAEA;;EAEA,WAAW;;EAEX;;EAEA;;EAEA;;EAEA,WAAW;;UAGI;EACf;EACA;EACA;;UAGe,oBAAoB,QAAQ;;;;;EAK3C;;;;;EAKA;;EAEA;;EAEA,oBAAoB,OAAO;;EAE3B,qBAAqB,QAAQ;;UAGd,eACf,QACA,SACA,eACA,cAAc,oBAAoB;EAElC;EACA,OAAO;EACP,OAAO;EACP,SAAS,YAAY,QAAQ,SAAS,eAAe;EACrD,QAAQ;EACR;EACA;EACA;EACA;EACA,aAAa;EACb,UAAU;;KAGA,gBAAgB;EAEtB;EACA,QAAQ;EACR;;EAGA;EACA;EACA;EACA;;EAEA,eAAe;;UAGJ;EACf;EACA;EACA;EACA;EACA,eAAe;;UAGA,qBAAqB;EACpC;EACA,SAAS;EACT;EACA;EACA;;UAGe;EACf;EACA;EACA;;UAGe,YACf,QACA,SACA,eACA,cAAc,oBAAoB;EAElC;EACA,UAAU,gBAAgB;EAC1B,aAAa;EACb,YAAY;EACZ,aAAa;EACb,YAAY;EACZ,gBAAgB,qBAAqB;EACrC;EACA;;UAGe,iBACf,QACA,SACA,eACA,cAAc,oBAAoB;EAElC;EACA;EACA;EACA;EACA;EACA,OAAO,YAAY,QAAQ,SAAS,eAAe;EACnD,YAAY;EACZ,YAAY;EACZ;EACA;;EAEA;EACA,eAAe;EACf,eAAe;EACf;;UAGe,qBACf,QACA,SACA,eACA,cAAc,oBAAoB;EAElC;EACA,SAAS,QAAQ;EACjB,SAAS;;EAET,gBAAgB;;;;;EAKhB,oBAAoB;IAClB,QAAQ;IACR,QAAQ;IACR,OAAO;IACP,OAAO;IACP,SAAS,YAAY,QAAQ,SAAS,eAAe;;;EAIvD,UAAU;IACR,SAAS,YAAY,QAAQ,SAAS,eAAe;IACrD,aAAa;QACT,QAAQ,UAAU;;EAGxB,WAAW;IACT;IACA,OAAO;IACP,SAAS,YAAY,QAAQ,SAAS,eAAe;IACrD,aAAa;QACT,QAAQ,WAAW;;EAGzB,SACE,KAAK,eAAe,QAAQ,SAAS,eAAe,WACjD,QAAQ,gBAAgB,YAAY,gBAAgB;;EAGzD,MACE,QAAQ,SACR,KAAK,eAAe,QAAQ,SAAS,eAAe,WACjD,QAAQ,iBAAiB;;EAG9B,cACE,KAAK,eAAe,QAAQ,SAAS,eAAe,WACjD,QAAQ,gBAAgB;;EAG7B,UAAU,MAAM,YAAY,QAAQ,SAAS,eAAe,WAAW;;EAGvE,eAAe,oBAAoB,QAAQ;;EAG3C,QAAQ;EACR;EACA;EACA;;KCtMG,sBAAsB,QAAQ;KACvB;KACA;UAEK;EACf;EACA;EACA,WAAW;;UAGH;EACR,WAAW;EACX,QAAQ;EACR;EACA;EACA;EACA;;KAGU,yBACP;EACC;EACA,eAAe;EACf;EACA;MAED;EACC;EACA,cAAc;EACd;EACA;;;iBAgCU,iBAAiB,KAAK,mBAAmB;KCrD7C;KAWA;KAEA;KAYA;UAEK;EACf;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA,WAAW;;UAGI;EACf;EACA,QAAQ;EACR,MAAM;EACN;EACA;EACA,WAAW;EACX;EACA,WAAW;;UAGI;EACf;EACA;EACA,cAAc;EACd;EACA;EACA,iBAAiB;EACjB,QAAQ;EACR,WAAW;EACX;EACA,WAAW;;UAGI;EACf;EACA;EACA,UAAU;EACV;EACA;EACA;EACA,WAAW;;UAGI;EACf;EACA;EACA;EACA,MAAM;EACN,UAAU;EACV,QAAQ;EACR,UAAU;EACV,QAAQ;EACR,OAAO;EACP;EACA;EACA,WAAW;;UAGI;EACf,KAAK,YAAY,qBAAqB;EACtC,IAAI,aAAa,QAAQ;EACzB,KAAK,SAAS,2BAA2B,QAAQ;EACjD,cAAc,YAAY,SAAS,kBAAkB,QAAQ;EAC7D,YAAY,YAAY,OAAO,eAAe,qBAAqB,QAAQ;;UAG5D;EACf;EACA;EACA,QAAQ;EACR;;UAGe;EACf;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA,YAAY;EACZ;EACA,WAAW;;UAoBI;EACf,MAAM;EACN,kBAAkB;EAClB,0BAA0B;EAC1B,2BAA2B;EAC3B,UAAU;EACV;EACA;EACA;EACA,QAAQ;EACR,OAAO;EACP;EACA;IACE;IACA;IACA;;EAEF,WAAW;;UAGI;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA,UAAU;EACV;EACA;EACA,UAAU;EACV;EACA;;cAUW,2CAA2C;mBACrC;EAEX,KAAK,YAAY,qBAAqB;EAItC,IAAI,aAAa,QAAQ;EAKzB,KAAK,SAAQ,2BAAgC,QAAQ;EAMrD,cAAc,YAAY,SAAS,kBAAkB,QAAQ;EAa7D,YACJ,YACA,OAAO,eACP,qBACC,QAAQ;;cAsBA,6CAA6C;mBACvC;mBACA;UACT;EAEI,YAAA;IAAW;;EAIjB,KAAK,YAAY,qBAAqB;EAMtC,IAAI,aAAa,QAAQ;EAKzB,KAAK,SAAQ,2BAAgC,QAAQ;EAKrD,cAAc,YAAY,SAAS,kBAAkB,QAAQ;EAO7D,YACJ,YACA,OAAO,eACP,qBACC,QAAQ;UAOG;UAWA;;iBA8BA,yBAAyB;EACvC;EACA;EACA;EACA,MAAM;EACN,WAAW;EACX,SAAS;EACT,UAAU;EACV,QAAQ;EACR,OAAO;EACP;EACA,WAAW;IACT;;iBAqBY,+BACd,KAAK,kBACL,SAAS,mCACR;;iBAsFa,2BACd,KAAK,kBACL;EAAW;IACV;iBA4Da,oBACd,YAAY,KAAK,iEACjB,SAAQ,sBACP;iBAca,0BACd,YAAY,oBACZ,SAAS,sBACR;iBAuBa,uCACd,cAAc,uBACb;iBAIa,iCACd,YAAY,qBACX;iBA6Ca,oCACd,cAAc,uBACb;iBAyCa,0BACd,cAAc,sBACd;EAAW;IACV;iBA2Ba,+BAA+B,SAAS;iBAWxC,+BAA+B,QAAQ,SAAS,eAC9D,KAAK,iBAAiB,QAAQ,SAAS,gBACvC;EACE;EACA;EACA,eAAe;EACf,oBAAoB,MAAM,YAAY,QAAQ,SAAS;EACvD,0BACE,MAAM,YAAY,QAAQ,SAAS,mBAChC;EACL;IAED"}
+{"version":3,"file":"feedback-trajectory-BgomnFOe.d.ts","names":[],"sources":["../src/control-runtime/contract.ts","../src/feedback-trajectory-review.ts","../src/feedback-trajectory.ts"],"mappings":";;;;;;KAIY;KAEA;UAEK;;EAEf;;EAEA;;EAEA;;EAEA,WAAW;;EAEX;;EAEA;;EAEA;;EAEA,WAAW;;UAGI;EACf;EACA;EACA;;UAGe,oBAAoB,QAAQ;;;;;EAK3C;;;;;EAKA;;EAEA;;EAEA,oBAAoB,OAAO;;EAE3B,qBAAqB,QAAQ;;UAGd,eACf,QACA,SACA,eACA,cAAc,oBAAoB;EAElC;EACA,OAAO;EACP,OAAO;EACP,SAAS,YAAY,QAAQ,SAAS,eAAe;EACrD,QAAQ;EACR;EACA;EACA;EACA;EACA,aAAa;EACb,UAAU;;KAGA,gBAAgB;EAEtB;EACA,QAAQ;EACR;;EAGA;EACA;EACA;EACA;;EAEA,eAAe;;UAGJ;EACf;EACA;EACA;EACA;EACA,eAAe;;UAGA,qBAAqB;EACpC;EACA,SAAS;EACT;EACA;EACA;;UAGe;EACf;EACA;EACA;;UAGe,YACf,QACA,SACA,eACA,cAAc,oBAAoB;EAElC;EACA,UAAU,gBAAgB;EAC1B,aAAa;EACb,YAAY;EACZ,aAAa;EACb,YAAY;EACZ,gBAAgB,qBAAqB;EACrC;EACA;;UAGe,iBACf,QACA,SACA,eACA,cAAc,oBAAoB;EAElC;EACA;EACA;EACA;EACA;EACA,OAAO,YAAY,QAAQ,SAAS,eAAe;EACnD,YAAY;EACZ,YAAY;EACZ;EACA;;EAEA;EACA,eAAe;EACf,eAAe;EACf;;UAGe,qBACf,QACA,SACA,eACA,cAAc,oBAAoB;EAElC;EACA,SAAS,QAAQ;EACjB,SAAS;;EAET,gBAAgB;;;;;EAKhB,oBAAoB;IAClB,QAAQ;IACR,QAAQ;IACR,OAAO;IACP,OAAO;IACP,SAAS,YAAY,QAAQ,SAAS,eAAe;;;EAIvD,UAAU;IACR,SAAS,YAAY,QAAQ,SAAS,eAAe;IACrD,aAAa;QACT,QAAQ,UAAU;;EAGxB,WAAW;IACT;IACA,OAAO;IACP,SAAS,YAAY,QAAQ,SAAS,eAAe;IACrD,aAAa;QACT,QAAQ,WAAW;;EAGzB,SACE,KAAK,eAAe,QAAQ,SAAS,eAAe,WACjD,QAAQ,gBAAgB,YAAY,gBAAgB;;EAGzD,MACE,QAAQ,SACR,KAAK,eAAe,QAAQ,SAAS,eAAe,WACjD,QAAQ,iBAAiB;;EAG9B,cACE,KAAK,eAAe,QAAQ,SAAS,eAAe,WACjD,QAAQ,gBAAgB;;EAG7B,UAAU,MAAM,YAAY,QAAQ,SAAS,eAAe,WAAW;;EAGvE,eAAe,oBAAoB,QAAQ;;EAG3C,QAAQ;EACR;EACA;EACA;;KCtMG,sBAAsB,QAAQ;KACvB;KACA;UAEK;EACf;EACA;EACA,WAAW;;UAGH;EACR,WAAW;EACX,QAAQ;EACR;EACA;EACA;EACA;;KAGU,yBACP;EACC;EACA,eAAe;EACf;EACA;MAED;EACC;EACA,cAAc;EACd;EACA;;;iBAgCU,iBAAiB,KAAK,mBAAmB;KCrD7C;KAWA;KAEA;KAYA;UAEK;EACf;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA,WAAW;;UAGI;EACf;EACA,QAAQ;EACR,MAAM;EACN;EACA;EACA,WAAW;EACX;EACA,WAAW;;UAGI;EACf;EACA;EACA,cAAc;EACd;EACA;EACA,iBAAiB;EACjB,QAAQ;EACR,WAAW;EACX;EACA,WAAW;;UAGI;EACf;EACA;EACA,UAAU;EACV;EACA;EACA;EACA,WAAW;;UAGI;EACf;EACA;EACA;EACA,MAAM;EACN,UAAU;EACV,QAAQ;EACR,UAAU;EACV,QAAQ;EACR,OAAO;EACP;EACA;EACA,WAAW;;UAGI;EACf,KAAK,YAAY,qBAAqB;EACtC,IAAI,aAAa,QAAQ;EACzB,KAAK,SAAS,2BAA2B,QAAQ;EACjD,cAAc,YAAY,SAAS,kBAAkB,QAAQ;EAC7D,YAAY,YAAY,OAAO,eAAe,qBAAqB,QAAQ;;UAG5D;EACf;EACA;EACA,QAAQ;EACR;;UAGe;EACf;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA,YAAY;EACZ;EACA,WAAW;;UAoBI;EACf,MAAM;EACN,kBAAkB;EAClB,0BAA0B;EAC1B,2BAA2B;EAC3B,UAAU;EACV;EACA;EACA;EACA,QAAQ;EACR,OAAO;EACP;EACA;IACE;IACA;IACA;;EAEF,WAAW;;UAGI;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA,UAAU;EACV;EACA;EACA,UAAU;EACV;EACA;;cAUW,2CAA2C;mBACrC;EAEX,KAAK,YAAY,qBAAqB;EAItC,IAAI,aAAa,QAAQ;EAKzB,KAAK,SAAQ,2BAAgC,QAAQ;EAMrD,cAAc,YAAY,SAAS,kBAAkB,QAAQ;EAa7D,YACJ,YACA,OAAO,eACP,qBACC,QAAQ;;cAsBA,6CAA6C;mBACvC;mBACA;UACT;EAEI,YAAA;IAAW;;EAIjB,KAAK,YAAY,qBAAqB;EAMtC,IAAI,aAAa,QAAQ;EAKzB,KAAK,SAAQ,2BAAgC,QAAQ;EAKrD,cAAc,YAAY,SAAS,kBAAkB,QAAQ;EAO7D,YACJ,YACA,OAAO,eACP,qBACC,QAAQ;UAOG;UAWA;;iBA8BA,yBAAyB;EACvC;EACA;EACA;EACA,MAAM;EACN,WAAW;EACX,SAAS;EACT,UAAU;EACV,QAAQ;EACR,OAAO;EACP;EACA,WAAW;IACT;;iBAqBY,+BACd,KAAK,kBACL,SAAS,mCACR;;iBAsFa,2BACd,KAAK,kBACL;EAAW;IACV;iBA4Da,oBACd,YAAY,KAAK,iEACjB,SAAQ,sBACP;iBAca,0BACd,YAAY,oBACZ,SAAS,sBACR;iBAuBa,uCACd,cAAc,uBACb;iBAIa,iCACd,YAAY,qBACX;iBA6Ca,oCACd,cAAc,uBACb;iBAyCa,0BACd,cAAc,sBACd;EAAW;IACV;iBA2Ba,+BAA+B,SAAS;iBAWxC,+BAA+B,QAAQ,SAAS,eAC9D,KAAK,iBAAiB,QAAQ,SAAS,gBACvC;EACE;EACA;EACA,eAAe;EACf,oBAAoB,MAAM,YAAY,QAAQ,SAAS;EACvD,0BACE,MAAM,YAAY,QAAQ,SAAS,mBAChC;EACL;IAED"}

package/dist/index.d.ts

@@ -24,6 +24,8 @@ import { a as MultiLayerVerifier, c as VerifyOptions, i as LayerStatus, l as gra
 import { C as HeldOutGateRejectionCode, S as HeldOutGateDecision, _ as gainHistogram, a as ResearchReportOptions, b as HeldOutGate, d as GainDistributionOptions, g as SummaryTableRow, h as SummaryTableOptions, l as GainDistributionBin, m as SummaryTable, p as ParetoPoint, t as ResearchReport, u as GainDistributionFigureSpec, v as summaryTable, w as SplitCoverage, x as HeldOutGateConfig, y as GateEvidence } from "./summary-report-0La5vejY.js";
 import { a as FailureClassification, c as FailureRule, d as failureBlame, i as FailureBlame, l as classifyFailure, o as FailureContext, r as failureClusterView, s as FailureReasonRule, u as classifyFailureReason } from "./failure-cluster-DOoa7CBg.js";
 import { d as InsightReport, n as analyzeRuns, t as AnalyzeRunsOptions } from "./analyze-runs-BYyMShsj.js";
+import { a as SpanHandle, n as CaptureWrite, o as TraceEmitter, s as TraceEmitterOptions, t as CaptureErrorHandler } from "./emitter-mAXC5PSN.js";
+import { s as SandboxDriver } from "./sandbox-harness--VUvMji7.js";
 import { a as ProducedProposal, c as SatisfiedBy, d as createLlmCorrectnessChecker, f as verifyCompletion, i as LlmCorrectnessCheckerOpts, l as TaskGold, n as CompletionVerdict, o as ProducedState, r as CorrectnessChecker, s as RequirementCheck, t as CompletionRequirement, u as completionVerdict } from "./completion-verifier-7KEfroqx.js";
 import { _ as SelfImproveMethodResult, a as DefinedAgentEval, c as SelfImproveMethodOptions, d as SelfImproveProposerOptions, f as SelfImproveProposerResult, g as SelfImproveMethodProvenance, h as selfImprove, i as DefineAgentEvalOptions, l as SelfImproveOptions, n as AgentEvalEvaluateOptions, o as defineAgentEval, p as SelfImproveResult, r as AgentEvalImproveOptions, s as SelfImproveBudget, t as AgentEvalAgent, u as SelfImproveProgressEvent } from "./define-agent-eval-B9zPWFSK.js";
 import { a as createTraceAnalyst, i as TraceAnalystDefinition, n as buildDefaultAnalystRegistry, r as CreateTraceAnalystOptions, t as DefaultAnalystRegistryOptions } from "./default-registry-DKlKVXMx.js";
@@ -33,12 +35,10 @@ import { r as TraceAnalysisEngineResult, s as RawAnalystFinding, t as TraceAnaly
 import { a as BudgetPolicy, c as RegistryRunOpts, i as AnalystRegistryOptions, s as ExactRegistryRunOpts, t as AnalystRegistry } from "./registry-CMtGAxOw.js";
 import { C as SEMANTIC_CONCEPT_JUDGE_VERSION, D as runSemanticConceptJudge, E as SemanticConceptJudgeResult, S as ConceptSpec, T as SemanticConceptJudgeOptions, _ as DspyRlmTraceEngineOptions, a as FAILURE_MODE_KIND_SPEC, b as createChatTraceEngine, d as FindingSubject, f as FindingSubjectKind, t as DEFAULT_TRACE_ANALYST_KINDS, v as createDspyRlmTraceEngine, w as SemanticConceptJudgeInput, x as ConceptFinding, y as ChatTraceEngineOptions } from "./index-Bn-P_XAl.js";
 import { i as BenchmarkEvaluation } from "./types-yofaDmQa.js";
-import { a as SpanHandle, n as CaptureWrite, o as TraceEmitter, s as TraceEmitterOptions, t as CaptureErrorHandler } from "./emitter-mAXC5PSN.js";
-import { n as SandboxDriver } from "./sandbox-harness-B3Kx1fOA.js";
 import { _ as redactText, d as ShareSafetyVerdict, f as assessShareSafety, g as redactForShare, h as redact, i as RedactOptions, o as RedactionProfile, r as REDACTION_VERSION, s as RedactionReport, u as ShareSafetyStatus, v as shareAllowed } from "./redact-jrT0APTU.js";
 import { a as PairedEvalueStep, c as SequentialDecision, i as PairedEvalueSequence, l as evaluateInterimReleaseConfidence, n as InterimReleaseConfidenceInput, r as PairedEvalueOptions, t as InterimReleaseConfidence, u as pairedEvalueSequence } from "./sequential-Bb_pdTqG.js";
 import { S as welchsTTest, _ as computeToolUseMetrics, b as WelchTestStatus, d as Trajectory, f as TrajectoryStep, g as ToolUseOptions, h as ToolUseMetrics, i as toolWasteView, m as ToolStats, p as buildTrajectory, s as judgeAgreementView, u as budgetBreachView, x as iqr, y as WelchTestResult } from "./tool-waste-BFG1vwTV.js";
-import { A as AnalystFindingDigest, B as ControlRuntimeConfig, C as createFeedbackTrajectory, D as renderPreferenceMemoryMarkdown, E as feedbackTrajectoryToOptimizerRow, F as ControlBudget, H as ControlSeverity, I as ControlContext, L as ControlDecision, M as AnalystRunDigest, N as analystRunDigest, O as summarizePreferenceMemory, P as ControlActionOutcome, R as ControlEvalResult, S as controlRunToFeedbackTrajectory, T as feedbackTrajectoriesToOptimizerRows, U as ControlStep, V as ControlRuntimeError, W as StopDecision, _ as PreferenceMemoryEntry, a as FeedbackLabel, b as analystRunToReviewRequests, c as FeedbackOptimizerRow, d as FeedbackTask, f as FeedbackTrajectory, g as InMemoryFeedbackTrajectoryStore, h as FileSystemFeedbackTrajectoryStore, i as FeedbackAttempt, j as AnalystReviewDecision, k as withAssignedFeedbackSplit, l as FeedbackOutcome, m as FeedbackTrajectoryStore, n as AnalystReviewRequest, o as FeedbackLabelKind, p as FeedbackTrajectoryFilter, r as FeedbackArtifactType, s as FeedbackLabelSource, t as AnalystFeedbackTrajectoryOptions, u as FeedbackSplitPolicy, v as ProposedSideEffect, w as feedbackTrajectoriesToDatasetScenarios, x as assignFeedbackSplit, y as analystRunToFeedbackTrajectory, z as ControlRunResult } from "./feedback-trajectory-Ck7K7AhK.js";
+import { A as AnalystFindingDigest, B as ControlRuntimeConfig, C as createFeedbackTrajectory, D as renderPreferenceMemoryMarkdown, E as feedbackTrajectoryToOptimizerRow, F as ControlBudget, H as ControlSeverity, I as ControlContext, L as ControlDecision, M as AnalystRunDigest, N as analystRunDigest, O as summarizePreferenceMemory, P as ControlActionOutcome, R as ControlEvalResult, S as controlRunToFeedbackTrajectory, T as feedbackTrajectoriesToOptimizerRows, U as ControlStep, V as ControlRuntimeError, W as StopDecision, _ as PreferenceMemoryEntry, a as FeedbackLabel, b as analystRunToReviewRequests, c as FeedbackOptimizerRow, d as FeedbackTask, f as FeedbackTrajectory, g as InMemoryFeedbackTrajectoryStore, h as FileSystemFeedbackTrajectoryStore, i as FeedbackAttempt, j as AnalystReviewDecision, k as withAssignedFeedbackSplit, l as FeedbackOutcome, m as FeedbackTrajectoryStore, n as AnalystReviewRequest, o as FeedbackLabelKind, p as FeedbackTrajectoryFilter, r as FeedbackArtifactType, s as FeedbackLabelSource, t as AnalystFeedbackTrajectoryOptions, u as FeedbackSplitPolicy, v as ProposedSideEffect, w as feedbackTrajectoriesToDatasetScenarios, x as assignFeedbackSplit, y as analystRunToFeedbackTrajectory, z as ControlRunResult } from "./feedback-trajectory-BgomnFOe.js";
 import { A as TraceInsightReadiness, F as describeTraceInsightScope, G as AnalyzeTracesResult, H as tokenizeDomainWords, I as domainEvidencePattern, K as analyzeTraces, L as inferDomainKeywords, N as buildTraceInsightContext, P as buildTraceInsightPrompt, V as scoreTraceInsightReadiness, a as captureFetchToRawSink, b as OtlpSpan, c as ExtractedUsage, d as extractUsageFromSse, f as OtlpFileTraceStore, g as spanRecordsToTraceAnalysisStore, h as otlpTextToTraceAnalysisStore, j as TraceInsightSuite, m as SpanRecordsToTraceAnalysisStoreOptions, n as toolSpansToTraceAnalysisStore, o as OtlpFlatLine, u as extractUsage, v as OtlpExport, x as exportRunAsOtlp } from "./store-tool-spans-lWf6INTH.js";
 import { a as RunIntegrityReport, o as assertRunCaptured, t as RunIntegrityError } from "./integrity-BAdVzT8u.js";
 import { a as SteeringChange, c as CampaignRunContext, d as CampaignVariant, f as EvalCampaignOptions, h as runEvalCampaign, i as Researcher, l as CampaignRunOutcome, m as FailedRun, n as ExperimentResult, o as CampaignFactoryParams, p as EvalCampaignResult, r as FailureMode, s as CampaignIntegrityPolicy, t as ExperimentPlan, u as CampaignScenario } from "./researcher-CLnNJ7Or.js";

package/dist/openapi.json

@@ -2,7 +2,7 @@
   "openapi": "3.1.0",
   "info": {
     "title": "@tangle-network/agent-eval — wire protocol",
-    "version": "1.0.2",
+    "version": "1.0.3",
     "description": "HTTP and stdio RPC interface to agent-eval. The TypeScript runtime is the source of truth; this spec is the contract that cross-language clients (Python, Rust, Go) generate from.\n\nWire-protocol version: 1.1.0. Bumps on breaking changes to request/response schemas.",
     "contact": {
       "name": "Tangle Network",

package/dist/sandbox-harness--VUvMji7.d.ts

@@ -0,0 +1,106 @@
+interface HarnessConfig {
+  /** Setup command (e.g. "pnpm install"). Non-zero exit fails the run. */
+  setupCommand?: string;
+  /** Run command (e.g. "pnpm build"). */
+  runCommand?: string;
+  /** Test command (e.g. "pnpm test --run"). Drives the test count + pass count. */
+  testCommand?: string;
+  /** Absolute cwd for the subprocess driver. Ignored by docker driver. */
+  cwd?: string;
+  /** Max wall-clock per phase in ms. Default 10 minutes. */
+  timeoutMs?: number;
+  /** Captured stdout + stderr cap, default 16 MiB; excess is dropped. */
+  maxOutputBytes?: number;
+  /** Image for the docker driver. */
+  image?: string;
+  /** Extra env vars (validated; shell-escaped). */
+  env?: Record<string, string>;
+  /** Parser for the test output — maps stdout/stderr/exit code → pass count. */
+  testParser?: TestOutputParser;
+}
+interface TestOutputParser {
+  id: string;
+  parse(stdout: string, stderr: string, exitCode: number): {
+    testsTotal: number;
+    testsPassed: number;
+  } | undefined;
+}
+interface SandboxResult {
+  phase: 'setup' | 'run' | 'test';
+  exitCode: number;
+  stdout: string;
+  stderr: string;
+  wallMs: number;
+  testsTotal?: number;
+  testsPassed?: number;
+  /** A timed-out phase always fails, even with exitCode 0. Absent/false means natural exit. */
+  killedByTimeout?: boolean;
+  /** Captured output exceeded maxOutputBytes. */
+  outputTruncated?: boolean;
+}
+interface SandboxDriver {
+  id: string;
+  exec(phase: SandboxResult['phase'], command: string, config: HarnessConfig): Promise<SandboxResult>;
+}
+interface SandboxHarnessResult {
+  passed: boolean;
+  setup?: SandboxResult;
+  run?: SandboxResult;
+  test?: SandboxResult;
+  totalWallMs: number;
+  /** Final score — 0 when no tests; otherwise testsPassed/testsTotal. */
+  score: number;
+}
+/** Portable integration report; products may retain extra timing and assertion fields. */
+interface HarnessAssertion {
+  name: string;
+  status: 'pass' | 'fail' | 'skip';
+  expected?: string;
+  actual?: string;
+  message?: string;
+  soft?: boolean;
+}
+interface HarnessTiming {
+  name: string;
+  durationMs: number;
+  success: boolean;
+  error?: string;
+}
+interface HarnessReport {
+  pass: boolean;
+  assertions: HarnessAssertion[];
+  timings: HarnessTiming[];
+  error?: string;
+  metadata?: Record<string, unknown>;
+}
+interface HarnessScope {
+  signal: AbortSignal;
+  check(...assertions: HarnessAssertion[]): void;
+  measure(name: string, run: () => Promise<unknown>): Promise<HarnessTiming>;
+  /** Register retirement immediately after allocation; cleanup ignores workload cancellation. */
+  defer(retire: () => Promise<unknown>): void;
+  poll<T>(read: () => Promise<T>, ready: (value: T) => boolean, budget: {
+    timeoutMs: number;
+    intervalMs: number;
+  }): Promise<T | undefined>;
+}
+interface HarnessScenario<T> {
+  setup(scope: HarnessScope): Promise<T>;
+  steps: ReadonlyArray<{
+    name: string;
+    when?: (state: T) => boolean;
+    run(state: T, scope: HarnessScope): Promise<unknown>;
+    expect?: (state: T, timing: HarnessTiming) => HarnessAssertion[];
+  }>;
+  expectations?: (state: T) => HarnessAssertion[];
+  metadata?: (state: T) => Record<string, unknown>;
+  /** Includes setup and steps. Transports must honor scope.signal. */
+  budgetMs: number;
+}
+/** Campaigns own whole-scenario retries; polling only repeats the declared read. */
+declare function runSandboxScenario<T>(spec: HarnessScenario<T>, options?: {
+  signal?: AbortSignal;
+  measure?: HarnessScope['measure'];
+}): Promise<HarnessReport>;
+export { HarnessScope as a, SandboxHarnessResult as c, HarnessScenario as i, runSandboxScenario as l, HarnessConfig as n, HarnessTiming as o, HarnessReport as r, SandboxDriver as s, HarnessAssertion as t };
+//# sourceMappingURL=sandbox-harness--VUvMji7.d.ts.map

package/dist/sandbox-harness--VUvMji7.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"sandbox-harness--VUvMji7.d.ts","names":[],"sources":["../src/sandbox-harness.ts"],"mappings":"A,UAMiB;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA,MAAM;;EAEN,aAAa;;UAGE;EACf;EACA,MACE,gBACA,gBACA;IACG;IAAoB;;;UAGV;EACf;EACA;EACA;EACA;EACA;EACA;EACA;;EAEA;;EAEA;;UAGe;EACf;EACA,KACE,OAAO,wBACP,iBACA,QAAQ,gBACP,QAAQ;;UAkDI;EACf;EACA,QAAQ;EACR,MAAM;EACN,OAAO;EACP;;EAEA;;;UAgEe;EACf;EACA;EACA;EACA;EACA;EACA;;UAEe;EACf;EACA;EACA;EACA;;UAEe;EACf;EACA,YAAY;EACZ,SAAS;EACT;EACA,WAAW;;UAEI;EACf,QAAQ;EACR,SAAS,YAAY;EACrB,QAAQ,cAAc,WAAW,mBAAmB,QAAQ;;EAE5D,MAAM,cAAc;EACpB,KAAK,GACH,YAAY,QAAQ,IACpB,QAAQ,OAAO,eACf;IAAU;IAAmB;MAC5B,QAAQ;;UAEI,gBAAgB;EAC/B,MAAM,OAAO,eAAe,QAAQ;EACpC,OAAO;IACL;IACA,QAAQ,OAAO;IACf,IAAI,OAAO,GAAG,OAAO,eAAe;IACpC,UAAU,OAAO,GAAG,QAAQ,kBAAkB;;EAEhD,gBAAgB,OAAO,MAAM;EAC7B,YAAY,OAAO,MAAM;;EAEzB;;;iBAIoB,mBAAmB,GACvC,MAAM,gBAAgB,IACtB;EACE,SAAS;EACT,UAAU;IAEX,QAAQ"}

package/dist/sandbox-harness-BuacGaVn.js

@@ -0,0 +1,181 @@
+import { n as runBoundedProcess } from "./bounded-process-BZrWXIKW.js";
+import { setTimeout } from "node:timers/promises";
+// src/sandbox-harness.ts
+var SubprocessSandboxDriver = class {
+	options;
+	id = "subprocess";
+	constructor(options = {}) {
+		this.options = options;
+	}
+	async exec(phase, command, config) {
+		const effectiveEnv = {
+			...process.env,
+			...this.options.env ?? {},
+			...config.env ?? {}
+		};
+		const result = await runBoundedProcess({
+			command,
+			shell: true,
+			cwd: config.cwd ?? this.options.cwd,
+			env: effectiveEnv,
+			envMode: "replace",
+			timeoutMs: config.timeoutMs ?? 6e5,
+			maxOutputBytes: config.maxOutputBytes ?? 16777216
+		});
+		const parsed = !result.killedByTimeout && phase === "test" && config.testParser ? config.testParser.parse(result.stdout, result.stderr, result.exitCode) : void 0;
+		return {
+			phase,
+			exitCode: result.exitCode,
+			stdout: result.stdout,
+			stderr: result.runnerError ? result.stderr + result.runnerError : result.stderr,
+			wallMs: result.wallMs,
+			testsTotal: parsed?.testsTotal,
+			testsPassed: parsed?.testsPassed,
+			killedByTimeout: result.killedByTimeout || void 0,
+			outputTruncated: result.outputTruncated || void 0
+		};
+	}
+};
+var SandboxHarness = class {
+	driver;
+	constructor(driver = new SubprocessSandboxDriver()) {
+		this.driver = driver;
+	}
+	async run(config, emitter) {
+		const handle = await emitter.sandbox({
+			name: `sandbox(${this.driver.id})`,
+			image: config.image,
+			command: [
+				config.setupCommand,
+				config.runCommand,
+				config.testCommand
+			].filter(Boolean).join(" && ")
+		});
+		const result = {
+			passed: false,
+			totalWallMs: 0,
+			score: 0
+		};
+		try {
+			for (const phase of [
+				"setup",
+				"run",
+				"test"
+			]) {
+				const command = config[`${phase}Command`];
+				if (!command) continue;
+				const measured = await this.driver.exec(phase, command, config);
+				result[phase] = measured;
+				result.totalWallMs += measured.wallMs;
+				const passed = !measured.killedByTimeout && measured.exitCode === 0;
+				if (phase !== "test" && !passed) {
+					await handle.fail(measured.killedByTimeout ? `${phase} timed out after ${measured.wallMs}ms` : `${phase} failed (exit ${measured.exitCode})`, {
+						exitCode: measured.exitCode,
+						wallMs: result.totalWallMs
+					});
+					return result;
+				}
+			}
+			const test = result.test;
+			result.passed = !test || !test.killedByTimeout && test.exitCode === 0;
+			result.score = test?.testsTotal !== void 0 && test.testsTotal > 0 ? (test.testsPassed ?? 0) / test.testsTotal : result.passed ? 1 : 0;
+			await handle.end({
+				...test ? {
+					exitCode: test.exitCode,
+					testsTotal: test.testsTotal,
+					testsPassed: test.testsPassed,
+					status: result.passed ? "ok" : "error"
+				} : {},
+				wallMs: result.totalWallMs
+			});
+		} catch (err) {
+			await handle.fail(err instanceof Error ? err : String(err));
+			throw err;
+		}
+		return result;
+	}
+};
+/** Campaigns own whole-scenario retries; polling only repeats the declared read. */
+async function runSandboxScenario(spec, options = {}) {
+	if (!(spec.budgetMs > 0 && Number.isSafeInteger(spec.budgetMs))) throw new Error("Scenario budgetMs must be positive and finite");
+	const signal = AbortSignal.any([AbortSignal.timeout(spec.budgetMs), ...options.signal ? [options.signal] : []]);
+	const report = {
+		pass: false,
+		assertions: [],
+		timings: []
+	};
+	const retirements = [];
+	const scope = {
+		signal,
+		check: (...assertions) => {
+			report.assertions.push(...assertions);
+		},
+		defer: (retire) => {
+			retirements.push(retire);
+		},
+		async measure(name, run) {
+			const start = performance.now();
+			let measured = {
+				name,
+				durationMs: 0,
+				success: true
+			};
+			try {
+				if (options.measure) measured = await options.measure(name, run);
+				else {
+					const result = await run();
+					if (result?.success === false) throw new Error(result.error ?? "measured operation returned success=false");
+					measured.durationMs = Math.round(performance.now() - start);
+				}
+			} catch (error) {
+				measured = {
+					name,
+					durationMs: Math.round(performance.now() - start),
+					success: false,
+					error: error instanceof Error ? error.message : String(error)
+				};
+			}
+			report.timings.push(measured);
+			return measured;
+		},
+		async poll(read, ready, budget) {
+			if (!(budget.timeoutMs > 0 && budget.intervalMs > 0)) throw new Error("Polling timeout and interval must be positive");
+			const deadline = performance.now() + budget.timeoutMs;
+			while (performance.now() < deadline) {
+				signal.throwIfAborted();
+				const value = await read();
+				if (ready(value)) return value;
+				await setTimeout(Math.min(budget.intervalMs, Math.max(0, deadline - performance.now())), void 0, { signal });
+			}
+		}
+	};
+	try {
+		signal.throwIfAborted();
+		const state = await spec.setup(scope);
+		for (const step of spec.steps) {
+			signal.throwIfAborted();
+			if (step.when && !step.when(state)) continue;
+			const timing = await scope.measure(step.name, () => step.run(state, scope));
+			scope.check(...step.expect?.(state, timing) ?? []);
+		}
+		signal.throwIfAborted();
+		scope.check(...spec.expectations?.(state) ?? []);
+		report.metadata = spec.metadata?.(state);
+	} catch (error) {
+		report.error = error instanceof Error ? error.message : String(error);
+	} finally {
+		for (const retire of retirements.reverse()) {
+			const measured = await scope.measure("cleanup", retire);
+			if (!measured.success) scope.check({
+				name: "cleanup",
+				status: "fail",
+				message: measured.error
+			});
+		}
+	}
+	report.pass = !report.error && report.timings.every((t) => t.success) && report.assertions.every((a) => a.soft || a.status === "pass");
+	return report;
+}
+export { runSandboxScenario as n, SandboxHarness as t };
+
+//# sourceMappingURL=sandbox-harness-BuacGaVn.js.map

package/dist/sandbox-harness-BuacGaVn.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"sandbox-harness-BuacGaVn.js","names":["delay"],"sources":["../src/sandbox-harness.ts"],"sourcesContent":["import { setTimeout as delay } from 'node:timers/promises'\nimport { runBoundedProcess } from './bounded-process'\n\nimport type { TraceEmitter } from './trace/emitter'\nimport type { SandboxSpan } from './trace/schema'\n\nexport interface HarnessConfig {\n  /** Setup command (e.g. \"pnpm install\"). Non-zero exit fails the run. */\n  setupCommand?: string\n  /** Run command (e.g. \"pnpm build\"). */\n  runCommand?: string\n  /** Test command (e.g. \"pnpm test --run\"). Drives the test count + pass count. */\n  testCommand?: string\n  /** Absolute cwd for the subprocess driver. Ignored by docker driver. */\n  cwd?: string\n  /** Max wall-clock per phase in ms. Default 10 minutes. */\n  timeoutMs?: number\n  /** Captured stdout + stderr cap, default 16 MiB; excess is dropped. */\n  maxOutputBytes?: number\n  /** Image for the docker driver. */\n  image?: string\n  /** Extra env vars (validated; shell-escaped). */\n  env?: Record<string, string>\n  /** Parser for the test output — maps stdout/stderr/exit code → pass count. */\n  testParser?: TestOutputParser\n}\n\nexport interface TestOutputParser {\n  id: string\n  parse(\n    stdout: string,\n    stderr: string,\n    exitCode: number,\n  ): { testsTotal: number; testsPassed: number } | undefined\n}\n\nexport interface SandboxResult {\n  phase: 'setup' | 'run' | 'test'\n  exitCode: number\n  stdout: string\n  stderr: string\n  wallMs: number\n  testsTotal?: number\n  testsPassed?: number\n  /** A timed-out phase always fails, even with exitCode 0. Absent/false means natural exit. */\n  killedByTimeout?: boolean\n  /** Captured output exceeded maxOutputBytes. */\n  outputTruncated?: boolean\n}\n\nexport interface SandboxDriver {\n  id: string\n  exec(\n    phase: SandboxResult['phase'],\n    command: string,\n    config: HarnessConfig,\n  ): Promise<SandboxResult>\n}\n\ninterface SubprocessSandboxDriverOptions {\n  cwd?: string\n  env?: Record<string, string>\n}\n\nexport class SubprocessSandboxDriver implements SandboxDriver {\n  id = 'subprocess'\n  constructor(private options: SubprocessSandboxDriverOptions = {}) {}\n\n  async exec(\n    phase: SandboxResult['phase'],\n    command: string,\n    config: HarnessConfig,\n  ): Promise<SandboxResult> {\n    const effectiveEnv = { ...process.env, ...(this.options.env ?? {}), ...(config.env ?? {}) }\n    const result = await runBoundedProcess({\n      command,\n      shell: true,\n      cwd: config.cwd ?? this.options.cwd,\n      env: effectiveEnv,\n      envMode: 'replace',\n      timeoutMs: config.timeoutMs ?? 10 * 60_000,\n      maxOutputBytes: config.maxOutputBytes ?? 16 * 1024 * 1024,\n    })\n    // A timed-out phase never reaches the parser: the output of a killed\n    // process is a fragment, and a fragment that happens to parse would\n    // report a pass for a run that never finished.\n    const parsed =\n      !result.killedByTimeout && phase === 'test' && config.testParser\n        ? config.testParser.parse(result.stdout, result.stderr, result.exitCode)\n        : undefined\n    return {\n      phase,\n      exitCode: result.exitCode,\n      stdout: result.stdout,\n      stderr: result.runnerError ? result.stderr + result.runnerError : result.stderr,\n      wallMs: result.wallMs,\n      testsTotal: parsed?.testsTotal,\n      testsPassed: parsed?.testsPassed,\n      killedByTimeout: result.killedByTimeout || undefined,\n      outputTruncated: result.outputTruncated || undefined,\n    }\n  }\n}\n\n// ── Harness orchestration ────────────────────────────────────────────\n\nexport interface SandboxHarnessResult {\n  passed: boolean\n  setup?: SandboxResult\n  run?: SandboxResult\n  test?: SandboxResult\n  totalWallMs: number\n  /** Final score — 0 when no tests; otherwise testsPassed/testsTotal. */\n  score: number\n}\n\nexport class SandboxHarness {\n  constructor(private driver: SandboxDriver = new SubprocessSandboxDriver()) {}\n\n  async run(config: HarnessConfig, emitter: TraceEmitter): Promise<SandboxHarnessResult> {\n    const handle = await emitter.sandbox({\n      name: `sandbox(${this.driver.id})`,\n      image: config.image,\n      command: [config.setupCommand, config.runCommand, config.testCommand]\n        .filter(Boolean)\n        .join(' && '),\n    })\n    const result: SandboxHarnessResult = { passed: false, totalWallMs: 0, score: 0 }\n    try {\n      for (const phase of ['setup', 'run', 'test'] as const) {\n        const command = config[`${phase}Command`]\n        if (!command) continue\n        const measured = await this.driver.exec(phase, command, config)\n        result[phase] = measured\n        result.totalWallMs += measured.wallMs\n        const passed = !measured.killedByTimeout && measured.exitCode === 0\n        if (phase !== 'test' && !passed) {\n          await handle.fail(\n            measured.killedByTimeout\n              ? `${phase} timed out after ${measured.wallMs}ms`\n              : `${phase} failed (exit ${measured.exitCode})`,\n            {\n              exitCode: measured.exitCode,\n              wallMs: result.totalWallMs,\n            } as Partial<SandboxSpan>,\n          )\n          return result\n        }\n      }\n      const test = result.test\n      result.passed = !test || (!test.killedByTimeout && test.exitCode === 0)\n      result.score =\n        test?.testsTotal !== undefined && test.testsTotal > 0\n          ? (test.testsPassed ?? 0) / test.testsTotal\n          : result.passed\n            ? 1\n            : 0\n      await handle.end({\n        ...(test\n          ? {\n              exitCode: test.exitCode,\n              testsTotal: test.testsTotal,\n              testsPassed: test.testsPassed,\n              status: result.passed ? 'ok' : 'error',\n            }\n          : {}),\n        wallMs: result.totalWallMs,\n      } as Partial<SandboxSpan>)\n    } catch (err) {\n      await handle.fail(err instanceof Error ? err : String(err))\n      throw err\n    }\n    return result\n  }\n}\n\n/** Portable integration report; products may retain extra timing and assertion fields. */\nexport interface HarnessAssertion {\n  name: string\n  status: 'pass' | 'fail' | 'skip'\n  expected?: string\n  actual?: string\n  message?: string\n  soft?: boolean\n}\nexport interface HarnessTiming {\n  name: string\n  durationMs: number\n  success: boolean\n  error?: string\n}\nexport interface HarnessReport {\n  pass: boolean\n  assertions: HarnessAssertion[]\n  timings: HarnessTiming[]\n  error?: string\n  metadata?: Record<string, unknown>\n}\nexport interface HarnessScope {\n  signal: AbortSignal\n  check(...assertions: HarnessAssertion[]): void\n  measure(name: string, run: () => Promise<unknown>): Promise<HarnessTiming>\n  /** Register retirement immediately after allocation; cleanup ignores workload cancellation. */\n  defer(retire: () => Promise<unknown>): void\n  poll<T>(\n    read: () => Promise<T>,\n    ready: (value: T) => boolean,\n    budget: { timeoutMs: number; intervalMs: number },\n  ): Promise<T | undefined>\n}\nexport interface HarnessScenario<T> {\n  setup(scope: HarnessScope): Promise<T>\n  steps: ReadonlyArray<{\n    name: string\n    when?: (state: T) => boolean\n    run(state: T, scope: HarnessScope): Promise<unknown>\n    expect?: (state: T, timing: HarnessTiming) => HarnessAssertion[]\n  }>\n  expectations?: (state: T) => HarnessAssertion[]\n  metadata?: (state: T) => Record<string, unknown>\n  /** Includes setup and steps. Transports must honor scope.signal. */\n  budgetMs: number\n}\n\n/** Campaigns own whole-scenario retries; polling only repeats the declared read. */\nexport async function runSandboxScenario<T>(\n  spec: HarnessScenario<T>,\n  options: {\n    signal?: AbortSignal\n    measure?: HarnessScope['measure']\n  } = {},\n): Promise<HarnessReport> {\n  if (!(spec.budgetMs > 0 && Number.isSafeInteger(spec.budgetMs)))\n    throw new Error('Scenario budgetMs must be positive and finite')\n  const signal = AbortSignal.any([\n    AbortSignal.timeout(spec.budgetMs),\n    ...(options.signal ? [options.signal] : []),\n  ])\n  const report: HarnessReport = { pass: false, assertions: [], timings: [] }\n  const retirements: Array<() => Promise<unknown>> = []\n  const scope: HarnessScope = {\n    signal,\n    check: (...assertions) => {\n      report.assertions.push(...assertions)\n    },\n    defer: (retire) => {\n      retirements.push(retire)\n    },\n    async measure(name, run) {\n      const start = performance.now()\n      let measured: HarnessTiming = { name, durationMs: 0, success: true }\n      try {\n        if (options.measure) measured = await options.measure(name, run)\n        else {\n          const result = (await run()) as { success?: boolean; error?: string } | undefined\n          if (result?.success === false)\n            throw new Error(result.error ?? 'measured operation returned success=false')\n          measured.durationMs = Math.round(performance.now() - start)\n        }\n      } catch (error) {\n        measured = {\n          name,\n          durationMs: Math.round(performance.now() - start),\n          success: false,\n          error: error instanceof Error ? error.message : String(error),\n        }\n      }\n      report.timings.push(measured)\n      return measured\n    },\n    async poll(read, ready, budget) {\n      if (!(budget.timeoutMs > 0 && budget.intervalMs > 0))\n        throw new Error('Polling timeout and interval must be positive')\n      const deadline = performance.now() + budget.timeoutMs\n      while (performance.now() < deadline) {\n        signal.throwIfAborted()\n        const value = await read()\n        if (ready(value)) return value\n        await delay(\n          Math.min(budget.intervalMs, Math.max(0, deadline - performance.now())),\n          undefined,\n          { signal },\n        )\n      }\n      return undefined\n    },\n  }\n  try {\n    signal.throwIfAborted()\n    const state = await spec.setup(scope)\n    for (const step of spec.steps) {\n      signal.throwIfAborted()\n      if (step.when && !step.when(state)) continue\n      const timing = await scope.measure(step.name, () => step.run(state, scope))\n      scope.check(...(step.expect?.(state, timing) ?? []))\n    }\n    signal.throwIfAborted()\n    scope.check(...(spec.expectations?.(state) ?? []))\n    report.metadata = spec.metadata?.(state)\n  } catch (error) {\n    report.error = error instanceof Error ? error.message : String(error)\n  } finally {\n    for (const retire of retirements.reverse()) {\n      const measured = await scope.measure('cleanup', retire)\n      if (!measured.success)\n        scope.check({ name: 'cleanup', status: 'fail', message: measured.error })\n    }\n  }\n  report.pass =\n    !report.error &&\n    report.timings.every((t) => t.success) &&\n    report.assertions.every((a) => a.soft || a.status === 'pass')\n  return report\n}\n"],"mappings":";;;AAgEA,IAAa,0BAAb,MAA8D;CAExC;CADpB,KAAK;CACL,YAAY,UAAkD,CAAC,GAAG;EAA9C,KAAA,UAAA;CAA+C;CAEnE,MAAM,KACJ,OACA,SACA,QACwB;EACxB,MAAM,eAAe;GAAE,GAAG,QAAQ;GAAK,GAAI,KAAK,QAAQ,OAAO,CAAC;GAAI,GAAI,OAAO,OAAO,CAAC;EAAG;EAC1F,MAAM,SAAS,MAAM,kBAAkB;GACrC;GACA,OAAO;GACP,KAAK,OAAO,OAAO,KAAK,QAAQ;GAChC,KAAK;GACL,SAAS;GACT,WAAW,OAAO,aAAa;GAC/B,gBAAgB,OAAO,kBAAkB;EAC3C,CAAC;EAID,MAAM,SACJ,CAAC,OAAO,mBAAmB,UAAU,UAAU,OAAO,aAClD,OAAO,WAAW,MAAM,OAAO,QAAQ,OAAO,QAAQ,OAAO,QAAQ,IACrE,KAAA;EACN,OAAO;GACL;GACA,UAAU,OAAO;GACjB,QAAQ,OAAO;GACf,QAAQ,OAAO,cAAc,OAAO,SAAS,OAAO,cAAc,OAAO;GACzE,QAAQ,OAAO;GACf,YAAY,QAAQ;GACpB,aAAa,QAAQ;GACrB,iBAAiB,OAAO,mBAAmB,KAAA;GAC3C,iBAAiB,OAAO,mBAAmB,KAAA;EAC7C;CACF;AACF;AAcA,IAAa,iBAAb,MAA4B;CACN;CAApB,YAAY,SAAgC,IAAI,wBAAwB,GAAG;EAAvD,KAAA,SAAA;CAAwD;CAE5E,MAAM,IAAI,QAAuB,SAAsD;EACrF,MAAM,SAAS,MAAM,QAAQ,QAAQ;GACnC,MAAM,WAAW,KAAK,OAAO,GAAG;GAChC,OAAO,OAAO;GACd,SAAS;IAAC,OAAO;IAAc,OAAO;IAAY,OAAO;GAAW,CAAC,CAClE,OAAO,OAAO,CAAC,CACf,KAAK,MAAM;EAChB,CAAC;EACD,MAAM,SAA+B;GAAE,QAAQ;GAAO,aAAa;GAAG,OAAO;EAAE;EAC/E,IAAI;GACF,KAAK,MAAM,SAAS;IAAC;IAAS;IAAO;GAAM,GAAY;IACrD,MAAM,UAAU,OAAO,GAAG,MAAM;IAChC,IAAI,CAAC,SAAS;IACd,MAAM,WAAW,MAAM,KAAK,OAAO,KAAK,OAAO,SAAS,MAAM;IAC9D,OAAO,SAAS;IAChB,OAAO,eAAe,SAAS;IAC/B,MAAM,SAAS,CAAC,SAAS,mBAAmB,SAAS,aAAa;IAClE,IAAI,UAAU,UAAU,CAAC,QAAQ;KAC/B,MAAM,OAAO,KACX,SAAS,kBACL,GAAG,MAAM,mBAAmB,SAAS,OAAO,MAC5C,GAAG,MAAM,gBAAgB,SAAS,SAAS,IAC/C;MACE,UAAU,SAAS;MACnB,QAAQ,OAAO;KACjB,CACF;KACA,OAAO;IACT;GACF;GACA,MAAM,OAAO,OAAO;GACpB,OAAO,SAAS,CAAC,QAAS,CAAC,KAAK,mBAAmB,KAAK,aAAa;GACrE,OAAO,QACL,MAAM,eAAe,KAAA,KAAa,KAAK,aAAa,KAC/C,KAAK,eAAe,KAAK,KAAK,aAC/B,OAAO,SACL,IACA;GACR,MAAM,OAAO,IAAI;IACf,GAAI,OACA;KACE,UAAU,KAAK;KACf,YAAY,KAAK;KACjB,aAAa,KAAK;KAClB,QAAQ,OAAO,SAAS,OAAO;IACjC,IACA,CAAC;IACL,QAAQ,OAAO;GACjB,CAAyB;EAC3B,SAAS,KAAK;GACZ,MAAM,OAAO,KAAK,eAAe,QAAQ,MAAM,OAAO,GAAG,CAAC;GAC1D,MAAM;EACR;EACA,OAAO;CACT;AACF;;AAmDA,eAAsB,mBACpB,MACA,UAGI,CAAC,GACmB;CACxB,IAAI,EAAE,KAAK,WAAW,KAAK,OAAO,cAAc,KAAK,QAAQ,IAC3D,MAAM,IAAI,MAAM,+CAA+C;CACjE,MAAM,SAAS,YAAY,IAAI,CAC7B,YAAY,QAAQ,KAAK,QAAQ,GACjC,GAAI,QAAQ,SAAS,CAAC,QAAQ,MAAM,IAAI,CAAC,CAC3C,CAAC;CACD,MAAM,SAAwB;EAAE,MAAM;EAAO,YAAY,CAAC;EAAG,SAAS,CAAC;CAAE;CACzE,MAAM,cAA6C,CAAC;CACpD,MAAM,QAAsB;EAC1B;EACA,QAAQ,GAAG,eAAe;GACxB,OAAO,WAAW,KAAK,GAAG,UAAU;EACtC;EACA,QAAQ,WAAW;GACjB,YAAY,KAAK,MAAM;EACzB;EACA,MAAM,QAAQ,MAAM,KAAK;GACvB,MAAM,QAAQ,YAAY,IAAI;GAC9B,IAAI,WAA0B;IAAE;IAAM,YAAY;IAAG,SAAS;GAAK;GACnE,IAAI;IACF,IAAI,QAAQ,SAAS,WAAW,MAAM,QAAQ,QAAQ,MAAM,GAAG;SAC1D;KACH,MAAM,SAAU,MAAM,IAAI;KAC1B,IAAI,QAAQ,YAAY,OACtB,MAAM,IAAI,MAAM,OAAO,SAAS,2CAA2C;KAC7E,SAAS,aAAa,KAAK,MAAM,YAAY,IAAI,IAAI,KAAK;IAC5D;GACF,SAAS,OAAO;IACd,WAAW;KACT;KACA,YAAY,KAAK,MAAM,YAAY,IAAI,IAAI,KAAK;KAChD,SAAS;KACT,OAAO,iBAAiB,QAAQ,MAAM,UAAU,OAAO,KAAK;IAC9D;GACF;GACA,OAAO,QAAQ,KAAK,QAAQ;GAC5B,OAAO;EACT;EACA,MAAM,KAAK,MAAM,OAAO,QAAQ;GAC9B,IAAI,EAAE,OAAO,YAAY,KAAK,OAAO,aAAa,IAChD,MAAM,IAAI,MAAM,+CAA+C;GACjE,MAAM,WAAW,YAAY,IAAI,IAAI,OAAO;GAC5C,OAAO,YAAY,IAAI,IAAI,UAAU;IACnC,OAAO,eAAe;IACtB,MAAM,QAAQ,MAAM,KAAK;IACzB,IAAI,MAAM,KAAK,GAAG,OAAO;IACzB,MAAMA,WACJ,KAAK,IAAI,OAAO,YAAY,KAAK,IAAI,GAAG,WAAW,YAAY,IAAI,CAAC,CAAC,GACrE,KAAA,GACA,EAAE,OAAO,CACX;GACF;EAEF;CACF;CACA,IAAI;EACF,OAAO,eAAe;EACtB,MAAM,QAAQ,MAAM,KAAK,MAAM,KAAK;EACpC,KAAK,MAAM,QAAQ,KAAK,OAAO;GAC7B,OAAO,eAAe;GACtB,IAAI,KAAK,QAAQ,CAAC,KAAK,KAAK,KAAK,GAAG;GACpC,MAAM,SAAS,MAAM,MAAM,QAAQ,KAAK,YAAY,KAAK,IAAI,OAAO,KAAK,CAAC;GAC1E,MAAM,MAAM,GAAI,KAAK,SAAS,OAAO,MAAM,KAAK,CAAC,CAAE;EACrD;EACA,OAAO,eAAe;EACtB,MAAM,MAAM,GAAI,KAAK,eAAe,KAAK,KAAK,CAAC,CAAE;EACjD,OAAO,WAAW,KAAK,WAAW,KAAK;CACzC,SAAS,OAAO;EACd,OAAO,QAAQ,iBAAiB,QAAQ,MAAM,UAAU,OAAO,KAAK;CACtE,UAAU;EACR,KAAK,MAAM,UAAU,YAAY,QAAQ,GAAG;GAC1C,MAAM,WAAW,MAAM,MAAM,QAAQ,WAAW,MAAM;GACtD,IAAI,CAAC,SAAS,SACZ,MAAM,MAAM;IAAE,MAAM;IAAW,QAAQ;IAAQ,SAAS,SAAS;GAAM,CAAC;EAC5E;CACF;CACA,OAAO,OACL,CAAC,OAAO,SACR,OAAO,QAAQ,OAAO,MAAM,EAAE,OAAO,KACrC,OAAO,WAAW,OAAO,MAAM,EAAE,QAAQ,EAAE,WAAW,MAAM;CAC9D,OAAO;AACT"}

package/dist/traces.d.ts

@@ -4,8 +4,8 @@ import { a as RunRecord, c as RunSplitTag, d as RunTokenUsage, u as RunTerminalO
 import { $ as QueryTracesPage, G as ReadSpanSourceResult, H as createBoundedTraceAnalysisStore, J as TraceAnalysisStore, K as SpanSourceReader, Q as ErrorCluster, U as BoundedTraceAnalysisStoreOptions, W as ReadSpanSourceInput, X as DEFAULT_TRACE_ANALYST_BUDGETS, Y as TraceAnalysisStoreContext, Z as DatasetOverview, at as TraceAnalystSpan, ct as TraceAnalystTraceSummary, dt as ViewTraceResult, et as SearchSpanResult, it as TraceAnalystFilters, lt as ViewSpansResult, nt as SpanMatchRecord, ot as TraceAnalystSpanKind, q as TRACE_ANALYSIS_LIMITS, rt as TraceAnalystByteBudgets, st as TraceAnalystSpanStatus, tt as SearchTraceResult, ut as ViewTraceOversized } from "./types-DzCO-S1m.js";
 import { G as InMemoryRawProviderSink, J as ProviderRedactor, K as InMemoryRawProviderSinkOptions, Q as RawProviderSinkFilter, U as FileSystemRawProviderSink, W as FileSystemRawProviderSinkOptions, X as RawProviderEvent, Y as RawProviderDirection, Z as RawProviderSink, q as NoopRawProviderSink } from "./types-0KglnQ9o.js";
 import { a as RunFilter, i as InMemoryTraceStore, n as FileSystemTraceStore, o as SpanFilter, r as FileSystemTraceStoreOptions, s as TraceStore, t as EventFilter } from "./store-CMs5wGXP.js";
-import { d as BuildTraceAnalysisToolsOptions, f as TRACE_ANALYST_TOOL_NAMESPACE, h as buildTraceAnalysisToolDescriptors, m as UNTRUSTED_TRACE_TEXT, p as TraceAnalysisToolDescriptor } from "./engine-CaHaCc3W.js";
 import { a as SpanHandle, i as RunCompleteHookContext, n as CaptureWrite, o as TraceEmitter, r as RunCompleteHook, s as TraceEmitterOptions, t as CaptureErrorHandler } from "./emitter-mAXC5PSN.js";
+import { d as BuildTraceAnalysisToolsOptions, f as TRACE_ANALYST_TOOL_NAMESPACE, h as buildTraceAnalysisToolDescriptors, m as UNTRUSTED_TRACE_TEXT, p as TraceAnalysisToolDescriptor } from "./engine-CaHaCc3W.js";
 import { C as classifyKey, E as normalizeKey, _ as redactText, a as RedactionFinding, c as SafetyCategory, d as ShareSafetyVerdict, f as assessShareSafety, g as redactForShare, h as redact, i as RedactOptions, l as ShareSafetyFinding, m as emptyRedactionReport, o as RedactionProfile, p as combineVerdicts, r as REDACTION_VERSION, s as RedactionReport, t as AssessOptions, u as ShareSafetyStatus, v as shareAllowed, w as detectCredential } from "./redact-jrT0APTU.js";
 import { A as TraceInsightReadiness, B as isTraceInsightEvidence, C as TraceInsightContext, D as TraceInsightPromptInput, E as TraceInsightPanelRole, F as describeTraceInsightScope, G as AnalyzeTracesResult, H as tokenizeDomainWords, I as domainEvidencePattern, K as analyzeTraces, L as inferDomainKeywords, M as TraceInsightTask, N as buildTraceInsightContext, O as TraceInsightQualityGate, P as buildTraceInsightPrompt, R as isTraceInsightArtifact, S as TraceInsightAttribution, T as TraceInsightFinding, U as AnalyzeTracesInput, V as scoreTraceInsightReadiness, W as AnalyzeTracesOptions, _ as ToolSpansToTraceAnalysisStoreOptions, a as captureFetchToRawSink, b as OtlpSpan, c as ExtractedUsage, d as extractUsageFromSse, f as OtlpFileTraceStore, g as spanRecordsToTraceAnalysisStore, h as otlpTextToTraceAnalysisStore, i as CaptureFetchOptions, j as TraceInsightSuite, k as TraceInsightQuestion, l as SseUsageMode, m as SpanRecordsToTraceAnalysisStoreOptions, n as toolSpansToTraceAnalysisStore, o as OtlpFlatLine, p as OtlpFileTraceStoreOptions, r as CaptureFetchContext, s as ExtractUsageFromSseOptions, t as ToolTraceMissingError, u as extractUsage, v as OtlpExport, w as TraceInsightEvidence, x as exportRunAsOtlp, y as OtlpResourceSpans, z as isTraceInsightAttribution } from "./store-tool-spans-lWf6INTH.js";
 import { a as RunIntegrityReport, i as RunIntegrityIssueCode, n as RunIntegrityExpectations, o as assertRunCaptured, r as RunIntegrityIssue, t as RunIntegrityError } from "./integrity-BAdVzT8u.js";

package/dist/wire/index.d.ts

@@ -1,7 +1,7 @@
 import { m as CustomTokenPricing, o as CostLedgerHandle } from "../cost-ledger-Czro9wmo.js";
 import { x as ChatClient } from "../types-0KglnQ9o.js";
 import { s as TraceStore } from "../store-CMs5wGXP.js";
-import { m as FeedbackTrajectoryStore } from "../feedback-trajectory-Ck7K7AhK.js";
+import { m as FeedbackTrajectoryStore } from "../feedback-trajectory-BgomnFOe.js";
 import { z } from "zod";
 import { ServerType } from "@hono/node-server";
 import { Hono } from "hono";
@@ -267,10 +267,10 @@ export declare const FeedbackTrajectorySchema: z.ZodObject<{
     metadata: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
   }, z.core.$strip>>;
   split: z.ZodOptional<z.ZodEnum<{
-    dev: "dev";
-    holdout: "holdout";
     train: "train";
+    dev: "dev";
     test: "test";
+    holdout: "holdout";
   }>>;
   tags: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodString>>;
   createdAt: z.ZodString;

package/docs/concepts.md

@@ -56,6 +56,23 @@ Paired lift, failure clusters, contamination checks, and outcome associations re
 Run records, scenarios, judge scores, statistics, and release decisions belong here because they work without an agent runtime.
 Agent sessions, worker coordination, sandbox execution, and runtime-specific profiles belong in `agent-runtime`.
 
+## Product integration scenarios
+
+`runSandboxScenario()` from `@tangle-network/agent-eval/campaign` runs a
+`HarnessScenario`: setup, named steps, expectations and a time budget.
+The product supplies transport callbacks and keeps its SDK, credentials and
+resource authorization. Steps receive one scope for measurement, assertions,
+condition polling and retirement registration. Register retirement immediately
+after allocation; callbacks use the scope signal for workload cancellation.
+Retirements run in reverse order even after setup or workload failure, and a
+failed retirement remains a failed assertion in the report.
+
+Pass the product's existing `measure` adapter to retain its server timing
+ledgers. A poll returns `undefined` when its budget expires; the scenario must
+assert the condition when exhaustion is a failure. The scope's budget signal
+is cooperative: callbacks must pass it through to their transport. Campaigns
+remain the owner of scenario scheduling, retries, artifacts and run reports.
+
 ## Measuring a complete profile
 
 Use the profile improvement functions from `/contract` when a host owns immutable agent-profile snapshots.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tangle-network/agent-eval",
-  "version": "1.0.2",
+  "version": "1.0.3",
   "description": "Evaluate and improve AI agents from runs, traces, judges, and feedback. Compare candidates, cluster failures, measure lift, and gate releases.",
   "homepage": "https://github.com/tangle-network/agent-eval#readme",
   "repository": {

package/dist/sandbox-harness-B3Kx1fOA.d.ts

@@ -1,68 +0,0 @@
-interface HarnessConfig {
-  /** Setup command (e.g. "pnpm install"). Non-zero exit fails the run. */
-  setupCommand?: string;
-  /** Run command (e.g. "pnpm build"). */
-  runCommand?: string;
-  /** Test command (e.g. "pnpm test --run"). Drives the test count + pass count. */
-  testCommand?: string;
-  /** Absolute cwd for the subprocess driver. Ignored by docker driver. */
-  cwd?: string;
-  /** Max wall-clock per phase in ms. Default 10 minutes. */
-  timeoutMs?: number;
-  /**
-   * Cap on captured stdout+stderr bytes per phase. A runaway process can
-   * otherwise grow the in-memory buffer without bound. Once hit, further
-   * output is dropped and `outputTruncated` is set. Default 16 MiB.
-   */
-  maxOutputBytes?: number;
-  /** Image for the docker driver. */
-  image?: string;
-  /** Extra env vars (validated; shell-escaped). */
-  env?: Record<string, string>;
-  /** Parser for the test output — maps stdout/stderr/exit code → pass count. */
-  testParser?: TestOutputParser;
-}
-interface TestOutputParser {
-  id: string;
-  parse(stdout: string, stderr: string, exitCode: number): {
-    testsTotal: number;
-    testsPassed: number;
-  } | undefined;
-}
-interface SandboxResult {
-  phase: 'setup' | 'run' | 'test';
-  exitCode: number;
-  stdout: string;
-  stderr: string;
-  wallMs: number;
-  testsTotal?: number;
-  testsPassed?: number;
-  /**
-   * True when the process was killed because it exceeded `timeoutMs`. A
-   * SIGKILLed child can still close with exit code 0; callers MUST treat
-   * a timed-out phase as a hard failure regardless of `exitCode`, never
-   * as a pass. `undefined`/`false` means the process completed on its own.
-   */
-  killedByTimeout?: boolean;
-  /**
-   * True when captured stdout/stderr hit `maxOutputBytes` and further
-   * output was dropped. The result is still returned (the process was
-   * not killed for this), but downstream parsers see truncated text.
-   */
-  outputTruncated?: boolean;
-}
-interface SandboxDriver {
-  id: string;
-  exec(phase: SandboxResult['phase'], command: string, config: HarnessConfig): Promise<SandboxResult>;
-}
-interface SandboxHarnessResult {
-  passed: boolean;
-  setup?: SandboxResult;
-  run?: SandboxResult;
-  test?: SandboxResult;
-  totalWallMs: number;
-  /** Final score — 0 when no tests; otherwise testsPassed/testsTotal. */
-  score: number;
-}
-export { SandboxDriver as n, SandboxHarnessResult as r, HarnessConfig as t };
-//# sourceMappingURL=sandbox-harness-B3Kx1fOA.d.ts.map

package/dist/sandbox-harness-B3Kx1fOA.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"sandbox-harness-B3Kx1fOA.d.ts","names":[],"sources":["../src/sandbox-harness.ts"],"mappings":"A,UAoBiB;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;EAMA;;EAEA;;EAEA,MAAM;;EAEN,aAAa;;UAGE;EACf;EACA,MACE,gBACA,gBACA;IACG;IAAoB;;;UAGV;EACf;EACA;EACA;EACA;EACA;EACA;EACA;;;;;;;EAOA;;;;;;EAMA;;UAGe;EACf;EACA,KACE,OAAO,wBACP,iBACA,QAAQ,gBACP,QAAQ;;UA0II;EACf;EACA,QAAQ;EACR,MAAM;EACN,OAAO;EACP;;EAEA"}