@tangle-network/agent-eval 0.186.2 → 0.187.1

package/dist/benchmarks/index.d.ts

@@ -1,7 +1,7 @@
 import { s as RunSplitTag } from "../run-record-D9wS35fK.js";
 import { a as CampaignResult } from "../types-BTRcIqFr.js";
 import { qa as CampaignStorage } from "../index-BRsrfSHT.js";
-import { a as BenchmarkFamily, c as BenchmarkSource, i as BenchmarkEvaluation, l as BenchmarkTaskKind, n as BenchmarkAdapter, o as BenchmarkResponder, r as BenchmarkDatasetItem, s as BenchmarkScenario, t as BENCHMARK_SPLIT_SEED, u as deterministicSplit } from "../types-c3ZRTbw9.js";
+import { a as BenchmarkFamily, c as BenchmarkSource, i as BenchmarkEvaluation, l as BenchmarkTaskKind, n as BenchmarkAdapter, o as BenchmarkResponder, r as BenchmarkDatasetItem, s as BenchmarkScenario, t as BENCHMARK_SPLIT_SEED, u as deterministicSplit } from "../types--18oHwVF.js";
 //#region src/benchmarks/calibration.d.ts
 interface BenchmarkMetricCalibrationOptions<TPayload = unknown, TArtifact = string> {
   adapter: BenchmarkAdapter<BenchmarkDatasetItem<TPayload>, TPayload, TArtifact>;
@@ -23,6 +23,76 @@ interface BenchmarkMetricCalibrationResult {
 }
 declare function calibrateBenchmarkMetric<TPayload = unknown, TArtifact = string>(options: BenchmarkMetricCalibrationOptions<TPayload, TArtifact>): Promise<BenchmarkMetricCalibrationResult>;
 //#endregion
+//#region src/benchmarks/harbor.d.ts
+/** One Harbor task, as its `task.toml` and `instruction.md` state it. */
+interface HarborTaskPayload {
+  /** `[task].name`, such as `terminal-bench/interleaved-vigenere`; the directory name when absent. */
+  name: string;
+  /** The task's directory name inside the dataset's `tasks/`. */
+  dir: string;
+  /** Absolute path of the task directory, for the grader. */
+  taskDir: string;
+  /** `instruction.md`, verbatim. */
+  instruction: string;
+  /** `[agent].timeout_sec`; null when the task sets none. */
+  agentTimeoutSec: number | null;
+  /** `[verifier].timeout_sec`; null when the task sets none. */
+  verifierTimeoutSec: number | null;
+  /** `[verifier].environment_mode`: `separate` runs the tests in their own image. */
+  verifierMode: string | null;
+  /** `[environment]` requests; null where the task states none. */
+  resources: {
+    cpus: number | null;
+    memoryMb: number | null;
+    storageMb: number | null;
+    gpus: number;
+  };
+  /** Paths the verifier reads from the agent's environment (`artifacts`, sources only). */
+  artifacts: string[];
+  /** The environment starts more than one container (a docker-compose file). */
+  multiContainer: boolean;
+  /** `[metadata].category`. */
+  category: string | null;
+  /** `[metadata].expert_time_estimate_hours`. */
+  expertTimeEstimateHours: number | null;
+}
+type HarborTaskItem = BenchmarkDatasetItem<HarborTaskPayload>;
+/** What the verifier returned for one task, or why it did not return. */
+type HarborGradeOutcome = {
+  succeeded: true;
+  value: {
+    reward: number;
+    detail?: string;
+  };
+} | {
+  succeeded: false;
+  error: string;
+};
+interface HarborBenchmarkOptions<TArtifact> {
+  /** Stable benchmark id, such as `terminal-bench@4.0.0`. */
+  id: string;
+  /** A checkout of the dataset: the directory that holds `tasks/`. */
+  datasetDir: string;
+  /** The upstream name, URL, version, revision, rules and contamination status the checkout is. */
+  source: BenchmarkSource & {
+    revision: string;
+  };
+  /** Task directory names to load, in this order; every task under `tasks/` when absent. */
+  tasks?: readonly string[];
+  /** Run the task's own verifier on the agent's artifact. */
+  grade: (item: HarborTaskItem, artifact: TArtifact) => Promise<HarborGradeOutcome>;
+  /** Split assignment; `deterministicSplit` over the benchmark id and task directory by default. */
+  assignSplit?: (itemId: string) => RunSplitTag;
+}
+/** Harbor adapter plus a loader that ignores splits, since a public benchmark reports every task. */
+interface HarborBenchmarkAdapter<TArtifact> extends BenchmarkAdapter<HarborTaskItem, HarborTaskPayload, TArtifact> {
+  loadAll(): Promise<HarborTaskItem[]>;
+}
+/** Read one Harbor task directory. Throws when `task.toml` or `instruction.md` is missing or malformed. */
+declare function readHarborTask(taskDir: string): Promise<HarborTaskPayload>;
+/** Import a Harbor dataset checkout as a `BenchmarkAdapter` whose scorer is the dataset's own verifier. */
+declare function createHarborBenchmarkAdapter<TArtifact = string>(options: HarborBenchmarkOptions<TArtifact>): HarborBenchmarkAdapter<TArtifact>;
+//#endregion
 //#region src/benchmarks/routing/dataset.d.ts
 /**
  * Synthetic routing dataset. 16 tasks across 4 categories. Used as a
@@ -226,5 +296,5 @@ declare function evaluateStandardRetrieval(payload: StandardRetrievalPayload, ar
   };
 };
 //#endregion
-export { BENCHMARK_SPLIT_SEED, type BenchmarkAdapter, type BenchmarkDatasetItem, type BenchmarkDistribution, type BenchmarkEvaluation, type BenchmarkFamily, type BenchmarkMetricCalibrationOptions, type BenchmarkMetricCalibrationResult, type BenchmarkReport, type BenchmarkResponder, type BenchmarkRunOptions, type BenchmarkRunResult, type BenchmarkScenario, type BenchmarkSliceSummary, type BenchmarkSource, type BenchmarkTaskKind, type BuildStandardRetrievalItemsOptions, type RetrievalIdAdapterOptions, type StandardRetrievalArtifact, type StandardRetrievalDocument, type StandardRetrievalEvaluationOptions, type StandardRetrievalPayload, type StandardRetrievalQrel, type StandardRetrievalQuery, type StandardRetrievalResult, buildStandardRetrievalItems, calibrateBenchmarkMetric, createRetrievalIdBenchmarkAdapter, deterministicSplit, evaluateStandardRetrieval, parseBeirCorpusJsonl, parseBeirQueriesJsonl, parseJsonlRows, parseQrels, index_d_exports as routing, runBenchmarkAdapter };
+export { BENCHMARK_SPLIT_SEED, type BenchmarkAdapter, type BenchmarkDatasetItem, type BenchmarkDistribution, type BenchmarkEvaluation, type BenchmarkFamily, type BenchmarkMetricCalibrationOptions, type BenchmarkMetricCalibrationResult, type BenchmarkReport, type BenchmarkResponder, type BenchmarkRunOptions, type BenchmarkRunResult, type BenchmarkScenario, type BenchmarkSliceSummary, type BenchmarkSource, type BenchmarkTaskKind, type BuildStandardRetrievalItemsOptions, type HarborBenchmarkAdapter, type HarborBenchmarkOptions, type HarborGradeOutcome, type HarborTaskItem, type HarborTaskPayload, type RetrievalIdAdapterOptions, type StandardRetrievalArtifact, type StandardRetrievalDocument, type StandardRetrievalEvaluationOptions, type StandardRetrievalPayload, type StandardRetrievalQrel, type StandardRetrievalQuery, type StandardRetrievalResult, buildStandardRetrievalItems, calibrateBenchmarkMetric, createHarborBenchmarkAdapter, createRetrievalIdBenchmarkAdapter, deterministicSplit, evaluateStandardRetrieval, parseBeirCorpusJsonl, parseBeirQueriesJsonl, parseJsonlRows, parseQrels, readHarborTask, index_d_exports as routing, runBenchmarkAdapter };
 //# sourceMappingURL=index.d.ts.map

package/dist/benchmarks/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","names":[],"sources":["../../src/benchmarks/calibration.ts","../../src/benchmarks/routing/dataset.ts","../../src/benchmarks/routing/index.ts","../../src/benchmarks/runner.ts","../../src/benchmarks/standard-formats.ts"],"mappings":";;;;;UAEiB,kCAAkC,oBAAoB;EACrE,SAAS,iBAAiB,qBAAqB,WAAW,UAAU;EACpE,MAAM,qBAAqB;EAC3B,cAAc;EACd,gBAAgB;EAChB;EACA;EACA;;UAGe;EACf;EACA,MAAM;EACN,QAAQ;EACR;EACA;EACA;EACA;;iBAGoB,yBAAyB,oBAAoB,oBACjE,SAAS,kCAAkC,UAAU,aACpD,QAAQ;;;;;;;;;;;;;;;;;;;;;;;;UCFM;EACf;EACA;EACA;;EAEA;;EAEA;;EAEA;;cAGW,iBAAiB;;;;KChBlB,iBAAiB;KACjB,qBAAqB,qBAAqB;cAEhD,0BAA0B,iBAAiB,oBAAoB;WAC1D;WACA;WACA;WACA;WACA;EAEH,YAAY,OAAO,cAAc,QAAQ;EAMzC,SAAS,MAAM,oBAAoB,mBAAmB,QAAQ;EAqBpE,YAAY,iBAAiB;;;;;;;;iBAef,mBAAmB;cAOtB,cAAW,OAjDG,gBAAc,QAAQ;cAkDpC,WAAQ,MA5CE,oBAAkB,qBAAqB,QAAQ;cA6CzD,cAAW,mBAxBO;;;UClCd,oBAAoB,oBAAoB;EACvD,SAAS,iBAAiB,qBAAqB,WAAW,UAAU;EACpE,SAAS,mBAAmB,UAAU;EACtC,kBAAkB;EAClB;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA,UAAU;EACV,YAAY;;UAGG;EACf;EACA,QAAQ;EACR,UAAU;EACV,SAAS;EACT;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA,QAAQ,eAAe;EACvB,MAAM,eAAe;EACrB,YAAY,eAAe;EAC3B,OAAO;EACP,SAAS;EACT,WAAW;;UAGI;EACf;EACA;EACA;EACA,OAAO;EACP,SAAS;EACT,WAAW;;UAGI;EACf;EACA;EACA;EACA;EACA;EACA;;UAGe,mBAAmB,oBAAoB;EACtD,WAAW,MAAM,kBAAkB;EACnC,UAAU,eAAe,WAAW,kBAAkB;EACtD,QAAQ;EACR;EACA;;iBAGoB,oBAAoB,oBAAoB,oBAC5D,SAAS,oBAAoB,UAAU,aACtC,QAAQ,mBAAmB,UAAU;;;UChFvB;EACf;EACA;EACA;EACA,WAAW;;UAGI;EACf;EACA;EACA,WAAW;;UAGI;EACf;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA,gBAAgB;EAChB,SAAS,eAAe;EACxB,WAAW;;UAGI;EACf;EACA,QAAQ;EACR,kBAAkB;EAClB,gBAAgB;EAChB,kBAAkB;EAClB;EACA,SAAS;EACT;EACA,WAAW,oBAAoB;;UAGhB,kCAAkC;EACjD,oBAAoB;EACpB;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA;;KAGU,kEAGC;EAEP;EACA;EACA,mBAAmB;;UAGR;EACf,oBAAoB;EACpB;EACA;EACA;EACA;;iBAGc,eAAe,aAAa,eAAe;iBAuB3C,WAAW,eAAe;iBAgB1B,qBAAqB,eAAe;iBAgBpC,sBAAsB,eAAe;iBAerC,4BACd,SAAS,qCACR,MAAM,qBAAqB;iBAuCd,kCACd,SAAS,4BACR,iBACD,qBAAqB,2BACrB,0BACA;iBAqBc,0BACd,SAAS,0BACT,UAAU,2BACV,UAAS"}
+{"version":3,"file":"index.d.ts","names":[],"sources":["../../src/benchmarks/calibration.ts","../../src/benchmarks/harbor.ts","../../src/benchmarks/routing/dataset.ts","../../src/benchmarks/routing/index.ts","../../src/benchmarks/runner.ts","../../src/benchmarks/standard-formats.ts"],"mappings":";;;;;UAEiB,kCAAkC,oBAAoB;EACrE,SAAS,iBAAiB,qBAAqB,WAAW,UAAU;EACpE,MAAM,qBAAqB;EAC3B,cAAc;EACd,gBAAgB;EAChB;EACA;EACA;;UAGe;EACf;EACA,MAAM;EACN,QAAQ;EACR;EACA;EACA;EACA;;iBAGoB,yBAAyB,oBAAoB,oBACjE,SAAS,kCAAkC,UAAU,aACpD,QAAQ;;;;UCKM;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;IACE;IACA;IACA;IACA;;;EAGF;;EAEA;;EAEA;;EAEA;;KAGU,iBAAiB,qBAAqB;;KAGtC;EACN;EAAiB;IAAS;IAAgB;;;EAC1C;EAAkB;;UAEP,uBAAuB;;EAEtC;;EAEA;;EAEA,QAAQ;IAAoB;;;EAE5B;;EAEA,QAAQ,MAAM,gBAAgB,UAAU,cAAc,QAAQ;;EAE9D,eAAe,mBAAmB;;;UAInB,uBAAuB,mBAC9B,iBAAiB,gBAAgB,mBAAmB;EAC5D,WAAW,QAAQ;;;iBAaC,eAAe,kBAAkB,QAAQ;;iBA6C/C,6BAA6B,oBAC3C,SAAS,uBAAuB,aAC/B,uBAAuB;;;;;;;;;;;;;;;;;;;;;;;;UC5HT;EACf;EACA;EACA;;EAEA;;EAEA;;EAEA;;cAGW,iBAAiB;;;;KChBlB,iBAAiB;KACjB,qBAAqB,qBAAqB;cAEhD,0BAA0B,iBAAiB,oBAAoB;WAC1D;WACA;WACA;WACA;WACA;EAEH,YAAY,OAAO,cAAc,QAAQ;EAMzC,SAAS,MAAM,oBAAoB,mBAAmB,QAAQ;EAqBpE,YAAY,iBAAiB;;;;;;;;iBAef,mBAAmB;cAOtB,cAAW,OAjDG,gBAAc,QAAQ;cAkDpC,WAAQ,MA5CE,oBAAkB,qBAAqB,QAAQ;cA6CzD,cAAW,mBAxBO;;;UClCd,oBAAoB,oBAAoB;EACvD,SAAS,iBAAiB,qBAAqB,WAAW,UAAU;EACpE,SAAS,mBAAmB,UAAU;EACtC,kBAAkB;EAClB;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA,UAAU;EACV,YAAY;;UAGG;EACf;EACA,QAAQ;EACR,UAAU;EACV,SAAS;EACT;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA,QAAQ,eAAe;EACvB,MAAM,eAAe;EACrB,YAAY,eAAe;EAC3B,OAAO;EACP,SAAS;EACT,WAAW;;UAGI;EACf;EACA;EACA;EACA,OAAO;EACP,SAAS;EACT,WAAW;;UAGI;EACf;EACA;EACA;EACA;EACA;EACA;;UAGe,mBAAmB,oBAAoB;EACtD,WAAW,MAAM,kBAAkB;EACnC,UAAU,eAAe,WAAW,kBAAkB;EACtD,QAAQ;EACR;EACA;;iBAGoB,oBAAoB,oBAAoB,oBAC5D,SAAS,oBAAoB,UAAU,aACtC,QAAQ,mBAAmB,UAAU;;;UChFvB;EACf;EACA;EACA;EACA,WAAW;;UAGI;EACf;EACA;EACA,WAAW;;UAGI;EACf;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA,gBAAgB;EAChB,SAAS,eAAe;EACxB,WAAW;;UAGI;EACf;EACA,QAAQ;EACR,kBAAkB;EAClB,gBAAgB;EAChB,kBAAkB;EAClB;EACA,SAAS;EACT;EACA,WAAW,oBAAoB;;UAGhB,kCAAkC;EACjD,oBAAoB;EACpB;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA;;KAGU,kEAGC;EAEP;EACA;EACA,mBAAmB;;UAGR;EACf,oBAAoB;EACpB;EACA;EACA;EACA;;iBAGc,eAAe,aAAa,eAAe;iBAuB3C,WAAW,eAAe;iBAgB1B,qBAAqB,eAAe;iBAgBpC,sBAAsB,eAAe;iBAerC,4BACd,SAAS,qCACR,MAAM,qBAAqB;iBAuCd,kCACd,SAAS,4BACR,iBACD,qBAAqB,2BACrB,0BACA;iBAqBc,0BACd,SAAS,0BACT,UAAU,2BACV,UAAS"}

package/dist/benchmarks/index.js

@@ -2,7 +2,9 @@ import { t as __exportAll } from "../rolldown-runtime-8H4AJuhK.js";
 import { X as runCampaign } from "../llm-judge-BVso5x-u.js";
 import { S as fsCampaignStorage } from "../external-optimizer-subprocess-DMgmqTB6.js";
 import "../campaign-DlxdTOvw.js";
-import { join } from "node:path";
+import { join, resolve } from "node:path";
+import { readFile, readdir, stat } from "node:fs/promises";
+import { parse } from "smol-toml";
 //#region src/benchmarks/calibration.ts
 async function calibrateBenchmarkMetric(options) {
 	const weak = await options.adapter.evaluate(options.item, options.weakArtifact);
@@ -67,6 +69,108 @@ function deterministicSplit(itemId, seed = BENCHMARK_SPLIT_SEED) {
 	return "holdout";
 }
 //#endregion
+//#region src/benchmarks/harbor.ts
+/**
+* A Harbor dataset as a `BenchmarkAdapter`.
+*
+* Harbor (https://harborframework.com) lays out each task as a directory with `task.toml`,
+* `instruction.md`, the agent's `environment/`, and the `tests/` its verifier runs, usually in a
+* verifier image of their own. This importer reads a local checkout of such a dataset, at the
+* revision the caller pins in `source`, into `BenchmarkDatasetItem`s.
+*
+* Scoring stays Harbor's. `evaluate` hands the task and the agent's artifact to the caller's
+* `grade` callback, which runs the task's own verifier (for example `harbor run -p <taskDir>`
+* with an agent that installs the collected artifacts) and reports its reward. This package does
+* not start containers. A verifier that did not run is an error, never a score of zero.
+*/
+const table = (value) => value !== null && typeof value === "object" && !Array.isArray(value) ? value : {};
+const numberOrNull = (value) => typeof value === "number" && Number.isFinite(value) ? value : null;
+const textOrNull = (value) => typeof value === "string" && value.length > 0 ? value : null;
+/** Read one Harbor task directory. Throws when `task.toml` or `instruction.md` is missing or malformed. */
+async function readHarborTask(taskDir) {
+	const dir = resolve(taskDir);
+	const config = parse(await readFile(join(dir, "task.toml"), "utf8"));
+	const instruction = await readFile(join(dir, "instruction.md"), "utf8");
+	const task = table(config.task);
+	const agent = table(config.agent);
+	const verifier = table(config.verifier);
+	const environment = table(config.environment);
+	const metadata = table(config.metadata);
+	const artifacts = (Array.isArray(config.artifacts) ? config.artifacts : []).map((entry, index) => {
+		const source = typeof entry === "string" ? entry : table(entry).source;
+		if (typeof source !== "string" || source.length === 0) throw new Error(`${dir}/task.toml: artifacts[${index}] names no source path`);
+		return source;
+	});
+	const environmentFiles = await readdir(join(dir, "environment")).catch(() => []);
+	const name = dir.split("/").pop() ?? dir;
+	return {
+		name: textOrNull(task.name) ?? name,
+		dir: name,
+		taskDir: dir,
+		instruction,
+		agentTimeoutSec: numberOrNull(agent.timeout_sec),
+		verifierTimeoutSec: numberOrNull(verifier.timeout_sec),
+		verifierMode: textOrNull(verifier.environment_mode),
+		resources: {
+			cpus: numberOrNull(environment.cpus),
+			memoryMb: numberOrNull(environment.memory_mb),
+			storageMb: numberOrNull(environment.storage_mb),
+			gpus: numberOrNull(environment.gpus) ?? 0
+		},
+		artifacts,
+		multiContainer: environmentFiles.some((file) => /compose\.ya?ml$/u.test(file)),
+		category: textOrNull(metadata.category),
+		expertTimeEstimateHours: numberOrNull(metadata.expert_time_estimate_hours)
+	};
+}
+/** Import a Harbor dataset checkout as a `BenchmarkAdapter` whose scorer is the dataset's own verifier. */
+function createHarborBenchmarkAdapter(options) {
+	const split = options.assignSplit ?? ((itemId) => deterministicSplit(`${options.id}::${itemId}`));
+	const tasksDir = join(resolve(options.datasetDir), "tasks");
+	async function loadAll() {
+		const names = options.tasks ?? (await Promise.all((await readdir(tasksDir)).map(async (entry) => await stat(join(tasksDir, entry, "task.toml")).catch(() => null) === null ? null : entry))).filter((entry) => entry !== null).sort();
+		return Promise.all(names.map(async (dir) => {
+			const payload = await readHarborTask(join(tasksDir, dir));
+			return {
+				id: dir,
+				payload,
+				split: split(dir),
+				family: options.source.name ?? "harbor",
+				taskKind: "harbor-task",
+				tags: payload.category === null ? [] : [payload.category],
+				source: options.source
+			};
+		}));
+	}
+	return {
+		id: options.id,
+		family: options.source.name ?? "harbor",
+		taskKind: "harbor-task",
+		description: `${options.source.name ?? "Harbor dataset"} ${options.source.version ?? ""} at ${options.source.revision}, scored by its own verifier`.replace(/\s+/gu, " "),
+		source: options.source,
+		defaultMetric: "reward",
+		loadAll,
+		async loadDataset(tag) {
+			return (await loadAll()).filter((item) => item.split === tag);
+		},
+		async evaluate(item, artifact) {
+			const outcome = await options.grade(item, artifact);
+			if (!outcome.succeeded) throw new Error(`${options.id} ${item.id}: the verifier did not score this artifact: ${outcome.error}`);
+			const { reward, detail } = outcome.value;
+			if (!(Number.isFinite(reward) && reward >= 0 && reward <= 1)) throw new Error(`${options.id} ${item.id}: the verifier returned reward ${reward}, outside [0, 1]`);
+			return {
+				score: reward,
+				passed: reward === 1,
+				raw: {
+					reward,
+					...detail === void 0 ? {} : { detail }
+				}
+			};
+		},
+		assignSplit: split
+	};
+}
+//#endregion
 //#region src/benchmarks/routing/dataset.ts
 const ROUTING_DATASET = [
 	{
@@ -729,6 +833,6 @@ function discountedGain(relevance, zeroBasedRank) {
 	return (2 ** relevance - 1) / Math.log2(zeroBasedRank + 2);
 }
 //#endregion
-export { BENCHMARK_SPLIT_SEED, buildStandardRetrievalItems, calibrateBenchmarkMetric, createRetrievalIdBenchmarkAdapter, deterministicSplit, evaluateStandardRetrieval, parseBeirCorpusJsonl, parseBeirQueriesJsonl, parseJsonlRows, parseQrels, routing_exports as routing, runBenchmarkAdapter };
+export { BENCHMARK_SPLIT_SEED, buildStandardRetrievalItems, calibrateBenchmarkMetric, createHarborBenchmarkAdapter, createRetrievalIdBenchmarkAdapter, deterministicSplit, evaluateStandardRetrieval, parseBeirCorpusJsonl, parseBeirQueriesJsonl, parseJsonlRows, parseQrels, readHarborTask, routing_exports as routing, runBenchmarkAdapter };
 
 //# sourceMappingURL=index.js.map

package/dist/benchmarks/index.js.map

@@ -1 +1 @@
-{"version":3,"file":"index.js","names":["clamp01","entries"],"sources":["../../src/benchmarks/calibration.ts","../../src/benchmarks/types.ts","../../src/benchmarks/routing/dataset.ts","../../src/benchmarks/routing/index.ts","../../src/benchmarks/runner.ts","../../src/benchmarks/standard-formats.ts"],"sourcesContent":["import type { BenchmarkAdapter, BenchmarkDatasetItem, BenchmarkEvaluation } from './types'\n\nexport interface BenchmarkMetricCalibrationOptions<TPayload = unknown, TArtifact = string> {\n  adapter: BenchmarkAdapter<BenchmarkDatasetItem<TPayload>, TPayload, TArtifact>\n  item: BenchmarkDatasetItem<TPayload>\n  weakArtifact: TArtifact\n  strongArtifact: TArtifact\n  maxWeakScore?: number\n  minStrongScore?: number\n  minGap?: number\n}\n\nexport interface BenchmarkMetricCalibrationResult {\n  passed: boolean\n  weak: BenchmarkEvaluation\n  strong: BenchmarkEvaluation\n  weakScore: number\n  strongScore: number\n  gap: number\n  reasons: string[]\n}\n\nexport async function calibrateBenchmarkMetric<TPayload = unknown, TArtifact = string>(\n  options: BenchmarkMetricCalibrationOptions<TPayload, TArtifact>,\n): Promise<BenchmarkMetricCalibrationResult> {\n  const weak = await options.adapter.evaluate(options.item, options.weakArtifact)\n  const strong = await options.adapter.evaluate(options.item, options.strongArtifact)\n  const weakScore = clamp01(weak.score)\n  const strongScore = clamp01(strong.score)\n  const maxWeakScore = options.maxWeakScore ?? 0.3\n  const minStrongScore = options.minStrongScore ?? 0.7\n  const minGap = options.minGap ?? 0.4\n  const gap = strongScore - weakScore\n  const reasons = [\n    weakScore <= maxWeakScore\n      ? undefined\n      : `weak score ${weakScore.toFixed(3)} exceeds max ${maxWeakScore.toFixed(3)}`,\n    strongScore >= minStrongScore\n      ? undefined\n      : `strong score ${strongScore.toFixed(3)} below min ${minStrongScore.toFixed(3)}`,\n    gap >= minGap ? undefined : `gap ${gap.toFixed(3)} below min ${minGap.toFixed(3)}`,\n  ].filter((reason): reason is string => Boolean(reason))\n\n  return {\n    passed: reasons.length === 0,\n    weak,\n    strong,\n    weakScore,\n    strongScore,\n    gap,\n    reasons,\n  }\n}\n\nfunction clamp01(value: number): number {\n  if (!Number.isFinite(value)) return 0\n  if (value < 0) return 0\n  if (value > 1) return 1\n  return value\n}\n","/**\n * Shared types for the reference benchmark wrappers under\n * `src/benchmarks/`. Each wrapper exports the three functions in\n * `BenchmarkAdapter` plus its own typed `DatasetItem` shape.\n */\n\nimport type { DispatchContext, Scenario } from '../campaign/types'\nimport type { RunSplitTag } from '../run-record'\n\nexport type BenchmarkTaskKind =\n  | 'retrieval'\n  | 'rag-answer'\n  | 'hallucination'\n  | 'kb-improvement'\n  | 'routing'\n  | 'custom'\n\nexport type BenchmarkFamily =\n  | 'beir'\n  | 'mteb-retrieval'\n  | 'msmarco'\n  | 'trec-dl'\n  | 'miracl'\n  | 'lotte'\n  | 'bright'\n  | 'crag'\n  | 'hotpotqa'\n  | 'kilt'\n  | 'ragtruth'\n  | 'faithbench'\n  | 'first-party'\n  | 'custom'\n\nexport interface BenchmarkDatasetItem<TPayload = unknown> {\n  /** Stable dataset-local item id (used for split assignment + paper\n   *  references). Unique within a benchmark. */\n  id: string\n  /** Free-form payload. Each benchmark defines its own shape. */\n  payload: TPayload\n  /** Optional precomputed split. When absent, adapters use assignSplit(id). */\n  split?: RunSplitTag\n  /** Benchmark family this row came from, e.g. `beir` or `crag`. */\n  family?: BenchmarkFamily | string\n  /** Benchmark-local task kind, e.g. retrieval vs answer quality. */\n  taskKind?: BenchmarkTaskKind | string\n  /** Slice labels such as language, domain, freshness, multihop, long-tail. */\n  tags?: string[]\n  /** Dataset provenance, version, URL, or license notes. */\n  source?: BenchmarkSource\n  metadata?: Record<string, unknown>\n}\n\nexport interface BenchmarkEvaluation {\n  /** [0, 1] score for the response on this item. Exact-match\n   *  benchmarks use 0/1; partial-credit benchmarks may return\n   *  fractional values. */\n  score: number\n  /** Optional pass/fail projection. Defaults to `score > 0` when absent. */\n  passed?: boolean\n  /** Numeric sub-metrics. These become report dimensions. */\n  dimensions?: Record<string, number>\n  /** Optional bag of raw scoring signals — e.g. parsed numeric\n   *  answer, regex match, judge sub-scores. */\n  raw: Record<string, unknown>\n  notes?: string\n}\n\nexport interface BenchmarkSource {\n  name?: string\n  url?: string\n  version?: string\n  license?: string\n  citation?: string\n}\n\n/** Common signature implemented by every adapter under `src/benchmarks/*`. */\n// `TPayload` is the per-item payload type; `_TItem` is preserved for\n// downstream type-narrowing extensions (a richer `BenchmarkDatasetItem`\n// subclass that adds e.g. provenance metadata) but is intentionally\n// unused here. `noUnusedLocals` requires the leading underscore.\nexport interface BenchmarkAdapter<_TItem = unknown, TPayload = unknown, TArtifact = string> {\n  /** Stable benchmark id such as `beir/nfcorpus` or `crag/smoke`. */\n  id?: string\n  family?: BenchmarkFamily | string\n  taskKind?: BenchmarkTaskKind | string\n  description?: string\n  source?: BenchmarkSource\n  defaultMetric?: string\n  /** Load the dataset for the given split. May hit the network on\n   *  first call but should be cache-friendly. Adapters that don't\n   *  ship the dataset itself MUST throw a clearly-marked error\n   *  pointing the caller at the loader script. */\n  loadDataset(split: RunSplitTag): Promise<BenchmarkDatasetItem<TPayload>[]>\n  /** Score a single response. Pure with respect to the inputs. */\n  evaluate(item: BenchmarkDatasetItem<TPayload>, artifact: TArtifact): Promise<BenchmarkEvaluation>\n  /** Deterministic split assignment via item id hashing. The\n   *  fraction of items in each split is implementation-defined but\n   *  MUST be stable across processes and platforms. */\n  assignSplit(itemId: string): RunSplitTag\n}\n\nexport interface BenchmarkScenario<TPayload = unknown> extends Scenario {\n  kind: 'benchmark'\n  benchmarkId: string\n  family: BenchmarkFamily | string\n  taskKind: BenchmarkTaskKind | string\n  splitTag: RunSplitTag\n  item: BenchmarkDatasetItem<TPayload>\n}\n\nexport type BenchmarkResponder<TPayload = unknown, TArtifact = string> = (input: {\n  scenario: BenchmarkScenario<TPayload>\n  item: BenchmarkDatasetItem<TPayload>\n  context: DispatchContext\n}) => Promise<TArtifact> | TArtifact\n\n// ── Deterministic split assignment ───────────────────────────────────\n\n/**\n * 32-bit FNV-1a hash. Stable, allocation-free, deterministic across\n * runtimes. We use it to assign items to splits rather than depending\n * on a polyfilled crypto.subtle path.\n */\nfunction fnv1a32(input: string): number {\n  let h = 0x811c9dc5\n  for (let i = 0; i < input.length; i++) {\n    h ^= input.charCodeAt(i) & 0xff\n    h = (h + ((h << 1) + (h << 4) + (h << 7) + (h << 8) + (h << 24))) >>> 0\n  }\n  return h >>> 0\n}\n\n/** Split-assignment seed shared across all benchmarks. Bumping this\n *  value reshuffles every split — do NOT do that lightly. */\nexport const BENCHMARK_SPLIT_SEED = 'agent-eval-v1'\n\n/**\n * Assign an item id to one of `'search' | 'dev' | 'holdout'` using a\n * stable 32-bit hash of `${seed}::${id}`. Default proportions:\n *\n *   search: 60%   (optimization-readable)\n *   dev:    20%   (held-out for tuning, leak-on-purpose during dev)\n *   holdout:20%   (paper-grade held-out, gated reads)\n */\nexport function deterministicSplit(\n  itemId: string,\n  seed: string = BENCHMARK_SPLIT_SEED,\n): RunSplitTag {\n  const h = fnv1a32(`${seed}::${itemId}`)\n  const pos = h / 0x100000000\n  if (pos < 0.6) return 'search'\n  if (pos < 0.8) return 'dev'\n  return 'holdout'\n}\n","/**\n * Synthetic routing dataset. 16 tasks across 4 categories. Used as a\n * deterministic, dependency-free benchmark for any router that maps a\n * natural-language request to one of a fixed set of route labels.\n *\n * Format (see `routing/README.md` for prose):\n *\n *   {\n *     id:           stable per-task ID (matches across processes).\n *     category:     one of the four route labels.\n *     prompt:       the user-facing request the router must classify.\n *     route:        the ground-truth route the router should pick.\n *     synonyms:     other strings that count as a correct answer.\n *     hardNegatives:close-but-wrong route labels — used to detect the\n *                   \"always picks the popular route\" failure mode.\n *   }\n *\n * The four categories are intentionally cross-domain (file ops,\n * math, search, conversation) so a router that collapses to one\n * category is easy to spot.\n */\n\nexport interface RoutingItem {\n  id: string\n  category: 'file' | 'math' | 'search' | 'chat'\n  prompt: string\n  /** Canonical correct route label. */\n  route: string\n  /** Alternate route labels that also count as correct. */\n  synonyms: string[]\n  /** Wrong-but-tempting route labels (for analysis, not grading). */\n  hardNegatives: string[]\n}\n\nexport const ROUTING_DATASET: RoutingItem[] = [\n  {\n    id: 'file_001',\n    category: 'file',\n    prompt: 'Save the meeting notes to /tmp/notes-2025-04.md as markdown.',\n    route: 'fs.write',\n    synonyms: ['filesystem.write', 'write_file'],\n    hardNegatives: ['fs.read', 'chat.reply'],\n  },\n  {\n    id: 'file_002',\n    category: 'file',\n    prompt: 'Read the contents of /etc/hosts and summarize the entries.',\n    route: 'fs.read',\n    synonyms: ['filesystem.read', 'read_file'],\n    hardNegatives: ['fs.write', 'search.web'],\n  },\n  {\n    id: 'file_003',\n    category: 'file',\n    prompt: 'List every Python file under src/ recursively.',\n    route: 'fs.list',\n    synonyms: ['filesystem.list', 'list_files'],\n    hardNegatives: ['fs.read', 'search.code'],\n  },\n  {\n    id: 'file_004',\n    category: 'file',\n    prompt: 'Delete the cached build at .turbo/cache.',\n    route: 'fs.delete',\n    synonyms: ['filesystem.delete', 'remove_file'],\n    hardNegatives: ['fs.write', 'fs.list'],\n  },\n  {\n    id: 'math_001',\n    category: 'math',\n    prompt: 'What is the integral of 3x^2 + 2x from 0 to 5?',\n    route: 'math.integral',\n    synonyms: ['calculator.integral', 'math.solve'],\n    hardNegatives: ['math.derivative', 'chat.reply'],\n  },\n  {\n    id: 'math_002',\n    category: 'math',\n    prompt: 'Compute the derivative of sin(x) * cos(x).',\n    route: 'math.derivative',\n    synonyms: ['calculator.derivative', 'math.solve'],\n    hardNegatives: ['math.integral', 'math.algebra'],\n  },\n  {\n    id: 'math_003',\n    category: 'math',\n    prompt: 'Solve 2x + 7 = 19 for x.',\n    route: 'math.algebra',\n    synonyms: ['calculator.algebra', 'math.solve'],\n    hardNegatives: ['math.derivative', 'math.integral'],\n  },\n  {\n    id: 'math_004',\n    category: 'math',\n    prompt: 'What is the prime factorization of 360?',\n    route: 'math.numbertheory',\n    synonyms: ['calculator.factor', 'math.solve'],\n    hardNegatives: ['math.algebra', 'search.web'],\n  },\n  {\n    id: 'search_001',\n    category: 'search',\n    prompt: 'Find recent papers on agent prompt optimization with held-out promotion gates.',\n    route: 'search.web',\n    synonyms: ['web.search', 'search.papers'],\n    hardNegatives: ['search.code', 'chat.reply'],\n  },\n  {\n    id: 'search_002',\n    category: 'search',\n    prompt: 'Search the codebase for every call site of `runProposeReview`.',\n    route: 'search.code',\n    synonyms: ['code.search', 'grep'],\n    hardNegatives: ['search.web', 'fs.read'],\n  },\n  {\n    id: 'search_003',\n    category: 'search',\n    prompt: 'What is the latest release of the Tangle network on GitHub?',\n    route: 'search.web',\n    synonyms: ['web.search', 'github.releases'],\n    hardNegatives: ['search.code', 'chat.reply'],\n  },\n  {\n    id: 'search_004',\n    category: 'search',\n    prompt: 'Find all TODO comments in the agent-eval src tree.',\n    route: 'search.code',\n    synonyms: ['code.search', 'grep'],\n    hardNegatives: ['search.web', 'fs.list'],\n  },\n  {\n    id: 'chat_001',\n    category: 'chat',\n    prompt: 'Hi there, how are you doing today?',\n    route: 'chat.reply',\n    synonyms: ['conversation.reply'],\n    hardNegatives: ['search.web', 'fs.read'],\n  },\n  {\n    id: 'chat_002',\n    category: 'chat',\n    prompt: 'Please explain the difference between an LLM and a foundation model.',\n    route: 'chat.reply',\n    synonyms: ['conversation.reply', 'qa.answer'],\n    hardNegatives: ['search.web', 'math.algebra'],\n  },\n  {\n    id: 'chat_003',\n    category: 'chat',\n    prompt: 'Tell me a short joke about distributed systems.',\n    route: 'chat.reply',\n    synonyms: ['conversation.reply'],\n    hardNegatives: ['search.web', 'fs.read'],\n  },\n  {\n    id: 'chat_004',\n    category: 'chat',\n    prompt: 'Acknowledge my last message with a thumbs up.',\n    route: 'chat.reply',\n    synonyms: ['conversation.reply', 'react'],\n    hardNegatives: ['fs.write', 'search.web'],\n  },\n]\n","/**\n * Routing benchmark — synthetic, dependency-free, ships in the\n * package. 16 cross-category items in `dataset.ts`. See\n * `routing/README.md` for the format.\n *\n * `evaluate` does case-insensitive exact match against the canonical\n * route plus declared synonyms. The first valid route token in the\n * response wins; everything else is ignored. Wrong answers also\n * report whether they hit a hard negative — useful when triaging\n * \"always picks the popular route\" failure modes.\n */\n\nimport type { RunSplitTag } from '../../run-record'\nimport type { BenchmarkAdapter, BenchmarkDatasetItem, BenchmarkEvaluation } from '../types'\nimport { deterministicSplit } from '../types'\nimport { ROUTING_DATASET, type RoutingItem } from './dataset'\n\nexport type { RoutingItem }\nexport type RoutingPayload = RoutingItem\nexport type RoutingDatasetItem = BenchmarkDatasetItem<RoutingPayload>\n\nclass RoutingAdapter implements BenchmarkAdapter<RoutingDatasetItem, RoutingPayload> {\n  readonly id = 'first-party/routing'\n  readonly family = 'first-party'\n  readonly taskKind = 'routing'\n  readonly description = 'Synthetic fixed-route classification smoke benchmark'\n  readonly defaultMetric = 'route_exact_match'\n\n  async loadDataset(split: RunSplitTag): Promise<RoutingDatasetItem[]> {\n    return ROUTING_DATASET.map((item) => ({ id: item.id, payload: item })).filter(\n      (it) => assignSplitImpl(it.id) === split,\n    )\n  }\n\n  async evaluate(item: RoutingDatasetItem, response: string): Promise<BenchmarkEvaluation> {\n    const tokens = extractRouteTokens(response)\n    const correct = new Set<string>(\n      [item.payload.route, ...item.payload.synonyms].map((s) => s.toLowerCase()),\n    )\n    const hardNeg = new Set<string>(item.payload.hardNegatives.map((s) => s.toLowerCase()))\n    const firstMatch = tokens.find((t) => correct.has(t.toLowerCase())) ?? null\n    const firstHardNeg = tokens.find((t) => hardNeg.has(t.toLowerCase())) ?? null\n    const score = firstMatch ? 1 : 0\n    return {\n      score,\n      raw: {\n        firstToken: tokens[0] ?? null,\n        matchedRoute: firstMatch,\n        hitHardNegative: Boolean(firstHardNeg),\n        hardNegativeRoute: firstHardNeg,\n        category: item.payload.category,\n      },\n    }\n  }\n\n  assignSplit(itemId: string): RunSplitTag {\n    return assignSplitImpl(itemId)\n  }\n}\n\nfunction assignSplitImpl(itemId: string): RunSplitTag {\n  return deterministicSplit(`routing::${itemId}`)\n}\n\n/**\n * Pull route-shaped tokens out of a model response. Routes look like\n * `category.action` (`fs.write`, `chat.reply`). Bare alphanumerics\n * are not routes, but `category.action` patterns are robust to most\n * model wrappers (JSON output, prose explanations, code fences).\n */\nexport function extractRouteTokens(response: string): string[] {\n  const matches = response.match(/[a-z][a-z0-9_]*\\.[a-z][a-z0-9_]*/gi)\n  return matches ?? []\n}\n\nconst adapter = new RoutingAdapter()\n\nexport const loadDataset = adapter.loadDataset.bind(adapter)\nexport const evaluate = adapter.evaluate.bind(adapter)\nexport const assignSplit = adapter.assignSplit.bind(adapter)\nexport { ROUTING_DATASET, RoutingAdapter }\n","import { join } from 'node:path'\nimport {\n  type CampaignResult,\n  type DispatchFn,\n  type JudgeConfig,\n  type JudgeScore,\n  runCampaign,\n  type Scenario,\n} from '../campaign'\nimport { type CampaignStorage, fsCampaignStorage } from '../campaign/storage'\nimport type { RunSplitTag } from '../run-record'\nimport type {\n  BenchmarkAdapter,\n  BenchmarkDatasetItem,\n  BenchmarkEvaluation,\n  BenchmarkFamily,\n  BenchmarkResponder,\n  BenchmarkScenario,\n  BenchmarkTaskKind,\n} from './types'\n\nexport interface BenchmarkRunOptions<TPayload = unknown, TArtifact = string> {\n  adapter: BenchmarkAdapter<BenchmarkDatasetItem<TPayload>, TPayload, TArtifact>\n  respond: BenchmarkResponder<TPayload, TArtifact>\n  splits?: readonly RunSplitTag[]\n  runDir: string\n  repo?: string\n  seed?: number\n  reps?: number\n  resumable?: boolean\n  costCeiling?: number\n  maxConcurrency?: number\n  dispatchTimeoutMs?: number\n  expectUsage?: 'assert' | 'warn' | 'off'\n  storage?: CampaignStorage\n  now?: () => Date\n}\n\nexport interface BenchmarkReport {\n  benchmarkId: string\n  family: BenchmarkFamily | string\n  taskKind: BenchmarkTaskKind | string\n  source?: BenchmarkAdapter['source']\n  runDir: string\n  manifestHash: string\n  seed: number\n  startedAt: string\n  endedAt: string\n  durationMs: number\n  totalItems: number\n  totalCells: number\n  cellsFailed: number\n  cellsCached: number\n  totalCostUsd: number\n  splits: Record<string, BenchmarkSliceSummary>\n  tags: Record<string, BenchmarkSliceSummary>\n  dimensions: Record<string, BenchmarkDistribution>\n  score: BenchmarkDistribution\n  costUsd: BenchmarkDistribution\n  latencyMs: BenchmarkDistribution\n}\n\nexport interface BenchmarkSliceSummary {\n  n: number\n  meanScore: number\n  passRate: number\n  score: BenchmarkDistribution\n  costUsd: BenchmarkDistribution\n  latencyMs: BenchmarkDistribution\n}\n\nexport interface BenchmarkDistribution {\n  n: number\n  min: number\n  mean: number\n  median: number\n  p90: number\n  max: number\n}\n\nexport interface BenchmarkRunResult<TPayload = unknown, TArtifact = string> {\n  scenarios: Array<BenchmarkScenario<TPayload>>\n  campaign: CampaignResult<TArtifact, BenchmarkScenario<TPayload>>\n  report: BenchmarkReport\n  reportJsonPath: string\n  reportMarkdownPath: string\n}\n\nexport async function runBenchmarkAdapter<TPayload = unknown, TArtifact = string>(\n  options: BenchmarkRunOptions<TPayload, TArtifact>,\n): Promise<BenchmarkRunResult<TPayload, TArtifact>> {\n  const storage = options.storage ?? fsCampaignStorage()\n  const benchmarkId = benchmarkIdFor(options.adapter)\n  const scenarios = await loadBenchmarkScenarios(options.adapter, options.splits)\n  const judge = benchmarkAdapterJudge(options.adapter)\n  const dispatch: DispatchFn<BenchmarkScenario<TPayload>, TArtifact> = async (\n    scenario,\n    context,\n  ) => {\n    return options.respond({ scenario, item: scenario.item, context })\n  }\n\n  const campaign = await runCampaign<BenchmarkScenario<TPayload>, TArtifact>({\n    scenarios,\n    dispatch,\n    dispatchRef: `benchmark:${benchmarkId}`,\n    judges: [judge],\n    seed: options.seed,\n    reps: options.reps,\n    resumable: options.resumable,\n    costCeiling: options.costCeiling,\n    maxConcurrency: options.maxConcurrency,\n    dispatchTimeoutMs: options.dispatchTimeoutMs,\n    expectUsage: options.expectUsage ?? 'off',\n    runDir: options.runDir,\n    repo: options.repo,\n    storage,\n    now: options.now,\n  })\n  const report = summarizeBenchmarkCampaign({\n    adapter: options.adapter,\n    scenarios,\n    campaign,\n  })\n  storage.ensureDir(campaign.runDir)\n  const reportJsonPath = join(campaign.runDir, 'benchmark-report.json')\n  const reportMarkdownPath = join(campaign.runDir, 'benchmark-report.md')\n  storage.write(reportJsonPath, `${JSON.stringify(report, null, 2)}\\n`)\n  storage.write(reportMarkdownPath, renderBenchmarkReportMarkdown(report))\n  return { scenarios, campaign, report, reportJsonPath, reportMarkdownPath }\n}\n\nexport async function loadBenchmarkScenarios<TPayload>(\n  adapter: BenchmarkAdapter<BenchmarkDatasetItem<TPayload>, TPayload, unknown>,\n  splits: readonly RunSplitTag[] = ['search', 'dev', 'holdout'],\n): Promise<Array<BenchmarkScenario<TPayload>>> {\n  const benchmarkId = benchmarkIdFor(adapter)\n  const family = adapter.family ?? 'custom'\n  const taskKind = adapter.taskKind ?? 'custom'\n  const out: Array<BenchmarkScenario<TPayload>> = []\n  const seen = new Set<string>()\n  for (const split of splits) {\n    const items = await adapter.loadDataset(split)\n    for (const item of items) {\n      const splitTag = item.split ?? adapter.assignSplit(item.id)\n      if (splitTag !== split) continue\n      const key = `${benchmarkId}:${item.id}`\n      if (seen.has(key)) continue\n      seen.add(key)\n      out.push({\n        id: key,\n        kind: 'benchmark',\n        benchmarkId,\n        family: item.family ?? family,\n        taskKind: item.taskKind ?? taskKind,\n        splitTag,\n        tags: [...new Set([splitTag, ...(item.tags ?? [])])],\n        item,\n      })\n    }\n  }\n  return out\n}\n\nexport function benchmarkAdapterJudge<TPayload, TArtifact>(\n  adapter: BenchmarkAdapter<BenchmarkDatasetItem<TPayload>, TPayload, TArtifact>,\n): JudgeConfig<TArtifact, BenchmarkScenario<TPayload>> {\n  const benchmarkId = benchmarkIdFor(adapter)\n  return {\n    name: `${benchmarkId}:score`,\n    dimensions: [\n      { key: 'score', description: `Primary ${benchmarkId} benchmark score` },\n      { key: 'passed', description: 'Binary pass projection for aggregate reporting' },\n    ],\n    appliesTo: (scenario: Scenario): scenario is BenchmarkScenario<TPayload> => {\n      return scenario.kind === 'benchmark' && scenario.id.startsWith(`${benchmarkId}:`)\n    },\n    async score({ artifact, scenario }) {\n      const evaluation = await adapter.evaluate(scenario.item, artifact)\n      const dimensions = normalizeEvaluationDimensions(evaluation)\n      return {\n        dimensions,\n        composite: clamp01(evaluation.score),\n        notes: evaluation.notes ?? '',\n      }\n    },\n  }\n}\n\nfunction summarizeBenchmarkCampaign<TPayload, TArtifact>(input: {\n  adapter: BenchmarkAdapter<BenchmarkDatasetItem<TPayload>, TPayload, TArtifact>\n  scenarios: Array<BenchmarkScenario<TPayload>>\n  campaign: CampaignResult<TArtifact, BenchmarkScenario<TPayload>>\n}): BenchmarkReport {\n  const scenarioById = new Map(input.scenarios.map((scenario) => [scenario.id, scenario]))\n  const rows = input.campaign.cells.map((cell) => {\n    const scenario = scenarioById.get(cell.scenarioId)\n    const judge = firstJudgeScore(cell.judgeScores)\n    const score = judge?.composite ?? 0\n    return {\n      cell,\n      scenario,\n      score,\n      passed: (judge?.dimensions.passed ?? (score > 0 ? 1 : 0)) >= 1,\n      dimensions: judge?.dimensions ?? {},\n    }\n  })\n  const successful = rows.filter((row) => !row.cell.error)\n  const scoreValues = successful.map((row) => row.score)\n  return {\n    benchmarkId: benchmarkIdFor(input.adapter),\n    family: input.adapter.family ?? 'custom',\n    taskKind: input.adapter.taskKind ?? 'custom',\n    ...(input.adapter.source ? { source: input.adapter.source } : {}),\n    runDir: input.campaign.runDir,\n    manifestHash: input.campaign.manifestHash,\n    seed: input.campaign.seed,\n    startedAt: input.campaign.startedAt,\n    endedAt: input.campaign.endedAt,\n    durationMs: input.campaign.durationMs,\n    totalItems: input.scenarios.length,\n    totalCells: input.campaign.cells.length,\n    cellsFailed: input.campaign.aggregates.cellsFailed,\n    cellsCached: input.campaign.aggregates.cellsCached,\n    totalCostUsd: input.campaign.aggregates.cost.totalCostUsd,\n    splits: summarizeSlices(successful, (row) => row.scenario?.splitTag ?? 'unknown', [\n      'search',\n      'dev',\n      'holdout',\n    ]),\n    tags: summarizeSlices(successful, (row) => row.scenario?.tags ?? []),\n    dimensions: summarizeDimensions(successful.map((row) => row.dimensions)),\n    score: distribution(scoreValues),\n    costUsd: distribution(successful.map((row) => row.cell.costUsd)),\n    latencyMs: distribution(successful.map((row) => row.cell.durationMs)),\n  }\n}\n\nfunction renderBenchmarkReportMarkdown(report: BenchmarkReport): string {\n  const splitRows = Object.entries(report.splits)\n    .map(([split, summary]) => {\n      return `| ${split} | ${summary.n} | ${fmt(summary.meanScore)} | ${fmt(summary.passRate)} | ${fmt(summary.score.p90)} | ${fmt(summary.costUsd.mean)} | ${fmt(summary.latencyMs.p90)} |`\n    })\n    .join('\\n')\n  const dimRows = Object.entries(report.dimensions)\n    .sort(([a], [b]) => a.localeCompare(b))\n    .map(([key, dist]) => `| ${key} | ${dist.n} | ${fmt(dist.mean)} | ${fmt(dist.p90)} |`)\n    .join('\\n')\n  return [\n    `# Benchmark Report: ${report.benchmarkId}`,\n    '',\n    `- family: ${report.family}`,\n    `- task kind: ${report.taskKind}`,\n    `- run dir: ${report.runDir}`,\n    `- manifest: ${report.manifestHash}`,\n    `- cells: ${report.totalCells} total, ${report.cellsFailed} failed, ${report.cellsCached} cached`,\n    `- cost: $${fmt(report.totalCostUsd)}`,\n    `- score: mean ${fmt(report.score.mean)}, median ${fmt(report.score.median)}, p90 ${fmt(report.score.p90)}, n=${report.score.n}`,\n    '',\n    '## Splits',\n    '',\n    '| split | n | mean score | pass rate | score p90 | mean cost | latency p90 ms |',\n    '| --- | ---: | ---: | ---: | ---: | ---: | ---: |',\n    splitRows || '| none | 0 | 0 | 0 | 0 | 0 | 0 |',\n    '',\n    '## Dimensions',\n    '',\n    '| dimension | n | mean | p90 |',\n    '| --- | ---: | ---: | ---: |',\n    dimRows || '| none | 0 | 0 | 0 |',\n    '',\n  ].join('\\n')\n}\n\nfunction benchmarkIdFor(adapter: Pick<BenchmarkAdapter, 'id' | 'family' | 'taskKind'>): string {\n  return adapter.id ?? `${adapter.family ?? 'custom'}/${adapter.taskKind ?? 'custom'}`\n}\n\nfunction normalizeEvaluationDimensions(evaluation: BenchmarkEvaluation): Record<string, number> {\n  const dimensions: Record<string, number> = { score: clamp01(evaluation.score) }\n  for (const [key, value] of Object.entries(evaluation.dimensions ?? {})) {\n    if (Number.isFinite(value)) dimensions[key] = value\n  }\n  dimensions.passed = (evaluation.passed ?? evaluation.score > 0) ? 1 : 0\n  return dimensions\n}\n\nfunction summarizeDimensions(\n  rows: Array<Record<string, number>>,\n): Record<string, BenchmarkDistribution> {\n  const values = new Map<string, number[]>()\n  for (const row of rows) {\n    for (const [key, value] of Object.entries(row)) {\n      if (!Number.isFinite(value)) continue\n      const list = values.get(key) ?? []\n      list.push(value)\n      values.set(key, list)\n    }\n  }\n  return Object.fromEntries([...values.entries()].map(([key, vals]) => [key, distribution(vals)]))\n}\n\nfunction summarizeSlices<T>(\n  rows: T[],\n  keyOf: (row: T) => string | string[],\n  knownKeys: readonly string[] = [],\n): Record<string, BenchmarkSliceSummary> {\n  const grouped = new Map<string, T[]>()\n  for (const key of knownKeys) grouped.set(key, [])\n  for (const row of rows) {\n    const keys = keyOf(row)\n    for (const key of Array.isArray(keys) ? keys : [keys]) {\n      const list = grouped.get(key) ?? []\n      list.push(row)\n      grouped.set(key, list)\n    }\n  }\n  const out: Record<string, BenchmarkSliceSummary> = {}\n  for (const [key, list] of grouped) {\n    const withShape = list as Array<{\n      score: number\n      passed: boolean\n      cell: { costUsd: number; durationMs: number }\n    }>\n    out[key] = {\n      n: list.length,\n      meanScore: mean(withShape.map((row) => row.score)),\n      passRate: mean(withShape.map((row) => (row.passed ? 1 : 0))),\n      score: distribution(withShape.map((row) => row.score)),\n      costUsd: distribution(withShape.map((row) => row.cell.costUsd)),\n      latencyMs: distribution(withShape.map((row) => row.cell.durationMs)),\n    }\n  }\n  return out\n}\n\nfunction firstJudgeScore(\n  judgeScores: CampaignResult<unknown>['cells'][number]['judgeScores'],\n): JudgeScore | undefined {\n  return Object.values(judgeScores)[0]\n}\n\nfunction distribution(values: readonly number[]): BenchmarkDistribution {\n  const finite = [...values].filter(Number.isFinite).sort((a, b) => a - b)\n  if (finite.length === 0) return { n: 0, min: 0, mean: 0, median: 0, p90: 0, max: 0 }\n  return {\n    n: finite.length,\n    min: finite[0]!,\n    mean: mean(finite),\n    median: percentile(finite, 0.5),\n    p90: percentile(finite, 0.9),\n    max: finite[finite.length - 1]!,\n  }\n}\n\nfunction percentile(sortedValues: readonly number[], p: number): number {\n  if (sortedValues.length === 0) return 0\n  const index = Math.min(\n    sortedValues.length - 1,\n    Math.max(0, Math.ceil(p * sortedValues.length) - 1),\n  )\n  return sortedValues[index]!\n}\n\nfunction mean(values: readonly number[]): number {\n  const finite = values.filter(Number.isFinite)\n  if (finite.length === 0) return 0\n  return finite.reduce((sum, value) => sum + value, 0) / finite.length\n}\n\nfunction clamp01(value: number): number {\n  if (!Number.isFinite(value)) return 0\n  if (value < 0) return 0\n  if (value > 1) return 1\n  return value\n}\n\nfunction fmt(value: number): string {\n  if (!Number.isFinite(value)) return '0'\n  return value.toFixed(value === 0 || Math.abs(value) >= 10 ? 0 : 3)\n}\n","import type { RunSplitTag } from '../run-record'\nimport type {\n  BenchmarkAdapter,\n  BenchmarkDatasetItem,\n  BenchmarkFamily,\n  BenchmarkSource,\n  BenchmarkTaskKind,\n} from './types'\nimport { deterministicSplit } from './types'\n\nexport interface StandardRetrievalDocument {\n  id: string\n  title?: string\n  text: string\n  metadata?: Record<string, unknown>\n}\n\nexport interface StandardRetrievalQuery {\n  id: string\n  text: string\n  metadata?: Record<string, unknown>\n}\n\nexport interface StandardRetrievalQrel {\n  queryId: string\n  documentId: string\n  score: number\n}\n\nexport interface StandardRetrievalPayload {\n  queryId: string\n  query: string\n  expectedDocumentIds: string[]\n  expectedScores: Record<string, number>\n  corpus?: Record<string, StandardRetrievalDocument>\n  metadata?: Record<string, unknown>\n}\n\nexport interface BuildStandardRetrievalItemsOptions {\n  benchmarkId: string\n  family: BenchmarkFamily | string\n  queries: readonly StandardRetrievalQuery[]\n  qrels: readonly StandardRetrievalQrel[]\n  corpus?: readonly StandardRetrievalDocument[]\n  includeCorpusInPayload?: boolean\n  source?: BenchmarkSource\n  tags?: readonly string[]\n  splitOf?: (queryId: string) => RunSplitTag\n}\n\nexport interface RetrievalIdAdapterOptions extends BuildStandardRetrievalItemsOptions {\n  responseIdPattern?: RegExp\n  cutoffs?: readonly number[]\n  primaryMetric?: string\n  passMetric?: string\n  passThreshold?: number\n}\n\nexport interface StandardRetrievalResult {\n  id?: string\n  documentId?: string\n  docId?: string\n  score?: number\n}\n\nexport type StandardRetrievalArtifact =\n  | string\n  | readonly string[]\n  | readonly StandardRetrievalResult[]\n  | {\n      ids?: readonly string[]\n      documentIds?: readonly string[]\n      results?: readonly StandardRetrievalResult[]\n    }\n\nexport interface StandardRetrievalEvaluationOptions {\n  responseIdPattern?: RegExp\n  cutoffs?: readonly number[]\n  primaryMetric?: string\n  passMetric?: string\n  passThreshold?: number\n}\n\nexport function parseJsonlRows<T = unknown>(text: string): T[] {\n  return text\n    .split(/\\r?\\n/)\n    .map((line) => line.trim())\n    .filter(Boolean)\n    .map((line, index) => {\n      try {\n        return JSON.parse(line) as T\n      } catch (error) {\n        throw new Error(`invalid JSONL row ${index + 1}: ${(error as Error).message}`)\n      }\n    })\n}\n\nfunction parseTsvRows(text: string): string[][] {\n  return text\n    .split(/\\r?\\n/)\n    .map((line) => line.trim())\n    .filter(Boolean)\n    .filter((line) => !line.startsWith('#'))\n    .map((line) => line.split(/\\t|\\s+/))\n}\n\nexport function parseQrels(text: string): StandardRetrievalQrel[] {\n  return parseTsvRows(text).flatMap((parts, index) => {\n    if (parts.length < 3) return []\n    const [queryId, maybeZeroOrDocId, maybeDocIdOrScore, maybeScore] = parts\n    if (!queryId || !maybeZeroOrDocId || !maybeDocIdOrScore) return []\n    if (queryId.toLowerCase() === 'query-id' || queryId.toLowerCase() === 'qid') return []\n    const documentId = maybeScore === undefined ? maybeZeroOrDocId : maybeDocIdOrScore\n    const scoreText = maybeScore === undefined ? maybeDocIdOrScore : maybeScore\n    const score = Number(scoreText)\n    if (!documentId || !Number.isFinite(score)) {\n      throw new Error(`invalid qrels row ${index + 1}: expected query id, doc id, score`)\n    }\n    return [{ queryId, documentId, score }]\n  })\n}\n\nexport function parseBeirCorpusJsonl(text: string): StandardRetrievalDocument[] {\n  return parseJsonlRows<Record<string, unknown>>(text).map((row, index) => {\n    const id = stringField(row, '_id') ?? stringField(row, 'id')\n    const body = stringField(row, 'text') ?? stringField(row, 'contents')\n    if (!id || body === undefined) {\n      throw new Error(`invalid BEIR corpus row ${index + 1}: expected _id/id and text/contents`)\n    }\n    return {\n      id,\n      title: stringField(row, 'title'),\n      text: body,\n      metadata: stripKnown(row, ['_id', 'id', 'title', 'text', 'contents']),\n    }\n  })\n}\n\nexport function parseBeirQueriesJsonl(text: string): StandardRetrievalQuery[] {\n  return parseJsonlRows<Record<string, unknown>>(text).map((row, index) => {\n    const id = stringField(row, '_id') ?? stringField(row, 'id') ?? stringField(row, 'query_id')\n    const query = stringField(row, 'text') ?? stringField(row, 'query')\n    if (!id || !query) {\n      throw new Error(`invalid BEIR query row ${index + 1}: expected _id/id and text/query`)\n    }\n    return {\n      id,\n      text: query,\n      metadata: stripKnown(row, ['_id', 'id', 'query_id', 'text', 'query']),\n    }\n  })\n}\n\nexport function buildStandardRetrievalItems(\n  options: BuildStandardRetrievalItemsOptions,\n): Array<BenchmarkDatasetItem<StandardRetrievalPayload>> {\n  const qrelsByQuery = new Map<string, StandardRetrievalQrel[]>()\n  for (const qrel of options.qrels) {\n    if (qrel.score <= 0) continue\n    const list = qrelsByQuery.get(qrel.queryId) ?? []\n    list.push(qrel)\n    qrelsByQuery.set(qrel.queryId, list)\n  }\n  const corpus =\n    options.includeCorpusInPayload && options.corpus\n      ? Object.fromEntries(options.corpus.map((document) => [document.id, document]))\n      : undefined\n  return options.queries.flatMap((query) => {\n    const qrels = qrelsByQuery.get(query.id) ?? []\n    if (qrels.length === 0) return []\n    const split =\n      options.splitOf?.(query.id) ?? deterministicSplit(`${options.benchmarkId}:${query.id}`)\n    return [\n      {\n        id: query.id,\n        split,\n        family: options.family,\n        taskKind: 'retrieval',\n        tags: [...new Set([...(options.tags ?? []), split])],\n        ...(options.source ? { source: options.source } : {}),\n        ...(query.metadata ? { metadata: query.metadata } : {}),\n        payload: {\n          queryId: query.id,\n          query: query.text,\n          expectedDocumentIds: qrels.map((qrel) => qrel.documentId),\n          expectedScores: Object.fromEntries(qrels.map((qrel) => [qrel.documentId, qrel.score])),\n          ...(corpus ? { corpus } : {}),\n          ...(query.metadata ? { metadata: query.metadata } : {}),\n        },\n      },\n    ]\n  })\n}\n\nexport function createRetrievalIdBenchmarkAdapter(\n  options: RetrievalIdAdapterOptions,\n): BenchmarkAdapter<\n  BenchmarkDatasetItem<StandardRetrievalPayload>,\n  StandardRetrievalPayload,\n  StandardRetrievalArtifact\n> {\n  const items = buildStandardRetrievalItems(options)\n  return {\n    id: options.benchmarkId,\n    family: options.family,\n    taskKind: 'retrieval' satisfies BenchmarkTaskKind,\n    source: options.source,\n    defaultMetric: options.primaryMetric ?? 'ndcg@10',\n    async loadDataset(split) {\n      return items.filter((item) => item.split === split)\n    },\n    async evaluate(item, artifact) {\n      return evaluateStandardRetrieval(item.payload, artifact, options)\n    },\n    assignSplit(itemId) {\n      return options.splitOf?.(itemId) ?? deterministicSplit(`${options.benchmarkId}:${itemId}`)\n    },\n  }\n}\n\nexport function evaluateStandardRetrieval(\n  payload: StandardRetrievalPayload,\n  artifact: StandardRetrievalArtifact,\n  options: StandardRetrievalEvaluationOptions = {},\n) {\n  const rankedDocumentIds = normalizeRetrievedDocumentIds(\n    artifact,\n    options.responseIdPattern ?? /[A-Za-z0-9_.:/-]+/g,\n  )\n  const cutoffs = normalizeCutoffs(options.cutoffs ?? [1, 3, 5, 10])\n  const dimensions: Record<string, number> = {\n    expected_count: payload.expectedDocumentIds.length,\n    returned_count: rankedDocumentIds.length,\n  }\n  for (const cutoff of cutoffs) {\n    Object.assign(\n      dimensions,\n      retrievalMetricsAtCutoff({\n        rankedDocumentIds,\n        expectedScores: payload.expectedScores,\n        cutoff,\n      }),\n    )\n  }\n  const primaryMetric = options.primaryMetric ?? 'ndcg@10'\n  const passMetric = options.passMetric ?? 'hit@10'\n  const score = dimensions[primaryMetric] ?? dimensions[`ndcg@${cutoffs[cutoffs.length - 1]}`] ?? 0\n  const passScore =\n    dimensions[passMetric] ?? dimensions[`hit@${cutoffs[cutoffs.length - 1]}`] ?? score\n  return {\n    score,\n    passed: passScore >= (options.passThreshold ?? 1),\n    dimensions,\n    raw: {\n      rankedDocumentIds,\n      expectedDocumentIds: payload.expectedDocumentIds,\n      expectedScores: payload.expectedScores,\n    },\n  }\n}\n\nfunction normalizeRetrievedDocumentIds(\n  artifact: StandardRetrievalArtifact,\n  responseIdPattern: RegExp = /[A-Za-z0-9_.:/-]+/g,\n): string[] {\n  if (typeof artifact === 'string') {\n    return uniqueOrdered((artifact.match(responseIdPattern) ?? []).map((value) => value.trim()))\n  }\n  if (Array.isArray(artifact)) {\n    const entries = artifact as readonly (string | StandardRetrievalResult)[]\n    return uniqueOrdered(\n      entries.flatMap((entry) => {\n        if (typeof entry === 'string') return [entry]\n        return [entry.documentId ?? entry.docId ?? entry.id ?? '']\n      }),\n    )\n  }\n  const objectArtifact = artifact as Exclude<StandardRetrievalArtifact, string | readonly unknown[]>\n  return uniqueOrdered(\n    [\n      ...(objectArtifact.documentIds ?? []),\n      ...(objectArtifact.ids ?? []),\n      ...(objectArtifact.results ?? []).map(\n        (entry) => entry.documentId ?? entry.docId ?? entry.id ?? '',\n      ),\n    ].map((value) => value.trim()),\n  )\n}\n\nfunction retrievalMetricsAtCutoff(input: {\n  rankedDocumentIds: readonly string[]\n  expectedScores: Record<string, number>\n  cutoff: number\n}): Record<string, number> {\n  const top = input.rankedDocumentIds.slice(0, input.cutoff)\n  const relevantIds = Object.keys(input.expectedScores).filter(\n    (id) => input.expectedScores[id]! > 0,\n  )\n  const relevant = new Set(relevantIds)\n  const hits = top.filter((id) => relevant.has(id)).length\n  const firstRelevantRank = top.findIndex((id) => relevant.has(id))\n  const prefix = `@${input.cutoff}`\n  return {\n    [`hit${prefix}`]: hits > 0 ? 1 : 0,\n    [`recall${prefix}`]: relevant.size === 0 ? 1 : hits / relevant.size,\n    [`precision${prefix}`]: hits / input.cutoff,\n    [`mrr${prefix}`]: firstRelevantRank === -1 ? 0 : 1 / (firstRelevantRank + 1),\n    [`ndcg${prefix}`]: ndcgAt(input.rankedDocumentIds, input.expectedScores, input.cutoff),\n  }\n}\n\nfunction stringField(row: Record<string, unknown>, key: string): string | undefined {\n  const value = row[key]\n  return typeof value === 'string' ? value : undefined\n}\n\nfunction stripKnown(\n  row: Record<string, unknown>,\n  keys: readonly string[],\n): Record<string, unknown> {\n  const known = new Set(keys)\n  return Object.fromEntries(Object.entries(row).filter(([key]) => !known.has(key)))\n}\n\nfunction normalizeCutoffs(cutoffs: readonly number[]): number[] {\n  return [\n    ...new Set(cutoffs.map((cutoff) => Math.trunc(cutoff)).filter((cutoff) => cutoff > 0)),\n  ].sort((a, b) => a - b)\n}\n\nfunction uniqueOrdered(values: readonly string[]): string[] {\n  const seen = new Set<string>()\n  const out: string[] = []\n  for (const value of values) {\n    const trimmed = value.trim()\n    if (!trimmed || seen.has(trimmed)) continue\n    seen.add(trimmed)\n    out.push(trimmed)\n  }\n  return out\n}\n\nfunction ndcgAt(\n  rankedDocumentIds: readonly string[],\n  expectedScores: Record<string, number>,\n  cutoff: number,\n): number {\n  const dcg = rankedDocumentIds\n    .slice(0, cutoff)\n    .reduce(\n      (sum, documentId, index) => sum + discountedGain(expectedScores[documentId] ?? 0, index),\n      0,\n    )\n  const ideal = Object.values(expectedScores)\n    .filter((score) => score > 0)\n    .sort((a, b) => b - a)\n    .slice(0, cutoff)\n    .reduce((sum, score, index) => sum + discountedGain(score, index), 0)\n  return ideal === 0 ? 1 : dcg / ideal\n}\n\nfunction discountedGain(relevance: number, zeroBasedRank: number): number {\n  if (relevance <= 0) return 0\n  return (2 ** relevance - 1) / Math.log2(zeroBasedRank + 2)\n}\n"],"mappings":";;;;;;AAsBA,eAAsB,yBACpB,SAC2C;CAC3C,MAAM,OAAO,MAAM,QAAQ,QAAQ,SAAS,QAAQ,MAAM,QAAQ,YAAY;CAC9E,MAAM,SAAS,MAAM,QAAQ,QAAQ,SAAS,QAAQ,MAAM,QAAQ,cAAc;CAClF,MAAM,YAAYA,UAAQ,KAAK,KAAK;CACpC,MAAM,cAAcA,UAAQ,OAAO,KAAK;CACxC,MAAM,eAAe,QAAQ,gBAAgB;CAC7C,MAAM,iBAAiB,QAAQ,kBAAkB;CACjD,MAAM,SAAS,QAAQ,UAAU;CACjC,MAAM,MAAM,cAAc;CAC1B,MAAM,UAAU;EACd,aAAa,eACT,KAAA,IACA,cAAc,UAAU,QAAQ,CAAC,EAAE,eAAe,aAAa,QAAQ,CAAC;EAC5E,eAAe,iBACX,KAAA,IACA,gBAAgB,YAAY,QAAQ,CAAC,EAAE,aAAa,eAAe,QAAQ,CAAC;EAChF,OAAO,SAAS,KAAA,IAAY,OAAO,IAAI,QAAQ,CAAC,EAAE,aAAa,OAAO,QAAQ,CAAC;CACjF,CAAC,CAAC,QAAQ,WAA6B,QAAQ,MAAM,CAAC;CAEtD,OAAO;EACL,QAAQ,QAAQ,WAAW;EAC3B;EACA;EACA;EACA;EACA;EACA;CACF;AACF;AAEA,SAASA,UAAQ,OAAuB;CACtC,IAAI,CAAC,OAAO,SAAS,KAAK,GAAG,OAAO;CACpC,IAAI,QAAQ,GAAG,OAAO;CACtB,IAAI,QAAQ,GAAG,OAAO;CACtB,OAAO;AACT;;;;;;;;ACgEA,SAAS,QAAQ,OAAuB;CACtC,IAAI,IAAI;CACR,KAAK,IAAI,IAAI,GAAG,IAAI,MAAM,QAAQ,KAAK;EACrC,KAAK,MAAM,WAAW,CAAC,IAAI;EAC3B,IAAK,MAAM,KAAK,MAAM,KAAK,MAAM,KAAK,MAAM,KAAK,MAAM,KAAK,SAAU;CACxE;CACA,OAAO,MAAM;AACf;;;AAIA,MAAa,uBAAuB;;;;;;;;;AAUpC,SAAgB,mBACd,QACA,OAAe,sBACF;CAEb,MAAM,MADI,QAAQ,GAAG,KAAK,IAAI,QAClB,IAAI;CAChB,IAAI,MAAM,IAAK,OAAO;CACtB,IAAI,MAAM,IAAK,OAAO;CACtB,OAAO;AACT;;;ACvHA,MAAa,kBAAiC;CAC5C;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,oBAAoB,YAAY;EAC3C,eAAe,CAAC,WAAW,YAAY;CACzC;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,mBAAmB,WAAW;EACzC,eAAe,CAAC,YAAY,YAAY;CAC1C;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,mBAAmB,YAAY;EAC1C,eAAe,CAAC,WAAW,aAAa;CAC1C;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,qBAAqB,aAAa;EAC7C,eAAe,CAAC,YAAY,SAAS;CACvC;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,uBAAuB,YAAY;EAC9C,eAAe,CAAC,mBAAmB,YAAY;CACjD;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,yBAAyB,YAAY;EAChD,eAAe,CAAC,iBAAiB,cAAc;CACjD;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,sBAAsB,YAAY;EAC7C,eAAe,CAAC,mBAAmB,eAAe;CACpD;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,qBAAqB,YAAY;EAC5C,eAAe,CAAC,gBAAgB,YAAY;CAC9C;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,cAAc,eAAe;EACxC,eAAe,CAAC,eAAe,YAAY;CAC7C;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,eAAe,MAAM;EAChC,eAAe,CAAC,cAAc,SAAS;CACzC;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,cAAc,iBAAiB;EAC1C,eAAe,CAAC,eAAe,YAAY;CAC7C;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,eAAe,MAAM;EAChC,eAAe,CAAC,cAAc,SAAS;CACzC;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,oBAAoB;EAC/B,eAAe,CAAC,cAAc,SAAS;CACzC;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,sBAAsB,WAAW;EAC5C,eAAe,CAAC,cAAc,cAAc;CAC9C;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,oBAAoB;EAC/B,eAAe,CAAC,cAAc,SAAS;CACzC;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,sBAAsB,OAAO;EACxC,eAAe,CAAC,YAAY,YAAY;CAC1C;AACF;;;;;;;;;;;AC9IA,IAAM,iBAAN,MAAqF;CACnF,KAAc;CACd,SAAkB;CAClB,WAAoB;CACpB,cAAuB;CACvB,gBAAyB;CAEzB,MAAM,YAAY,OAAmD;EACnE,OAAO,gBAAgB,KAAK,UAAU;GAAE,IAAI,KAAK;GAAI,SAAS;EAAK,EAAE,CAAC,CAAC,QACpE,OAAO,gBAAgB,GAAG,EAAE,MAAM,KACrC;CACF;CAEA,MAAM,SAAS,MAA0B,UAAgD;EACvF,MAAM,SAAS,mBAAmB,QAAQ;EAC1C,MAAM,UAAU,IAAI,IAClB,CAAC,KAAK,QAAQ,OAAO,GAAG,KAAK,QAAQ,QAAQ,CAAC,CAAC,KAAK,MAAM,EAAE,YAAY,CAAC,CAC3E;EACA,MAAM,UAAU,IAAI,IAAY,KAAK,QAAQ,cAAc,KAAK,MAAM,EAAE,YAAY,CAAC,CAAC;EACtF,MAAM,aAAa,OAAO,MAAM,MAAM,QAAQ,IAAI,EAAE,YAAY,CAAC,CAAC,KAAK;EACvE,MAAM,eAAe,OAAO,MAAM,MAAM,QAAQ,IAAI,EAAE,YAAY,CAAC,CAAC,KAAK;EAEzE,OAAO;GACL,OAFY,aAAa,IAAI;GAG7B,KAAK;IACH,YAAY,OAAO,MAAM;IACzB,cAAc;IACd,iBAAiB,QAAQ,YAAY;IACrC,mBAAmB;IACnB,UAAU,KAAK,QAAQ;GACzB;EACF;CACF;CAEA,YAAY,QAA6B;EACvC,OAAO,gBAAgB,MAAM;CAC/B;AACF;AAEA,SAAS,gBAAgB,QAA6B;CACpD,OAAO,mBAAmB,YAAY,QAAQ;AAChD;;;;;;;AAQA,SAAgB,mBAAmB,UAA4B;CAE7D,OADgB,SAAS,MAAM,oCAClB,KAAK,CAAC;AACrB;AAEA,MAAM,UAAU,IAAI,eAAe;AAEnC,MAAa,cAAc,QAAQ,YAAY,KAAK,OAAO;AAC3D,MAAa,WAAW,QAAQ,SAAS,KAAK,OAAO;AACrD,MAAa,cAAc,QAAQ,YAAY,KAAK,OAAO;;;ACS3D,eAAsB,oBACpB,SACkD;CAClD,MAAM,UAAU,QAAQ,WAAW,kBAAkB;CACrD,MAAM,cAAc,eAAe,QAAQ,OAAO;CAClD,MAAM,YAAY,MAAM,uBAAuB,QAAQ,SAAS,QAAQ,MAAM;CAC9E,MAAM,QAAQ,sBAAsB,QAAQ,OAAO;CACnD,MAAM,WAA+D,OACnE,UACA,YACG;EACH,OAAO,QAAQ,QAAQ;GAAE;GAAU,MAAM,SAAS;GAAM;EAAQ,CAAC;CACnE;CAEA,MAAM,WAAW,MAAM,YAAoD;EACzE;EACA;EACA,aAAa,aAAa;EAC1B,QAAQ,CAAC,KAAK;EACd,MAAM,QAAQ;EACd,MAAM,QAAQ;EACd,WAAW,QAAQ;EACnB,aAAa,QAAQ;EACrB,gBAAgB,QAAQ;EACxB,mBAAmB,QAAQ;EAC3B,aAAa,QAAQ,eAAe;EACpC,QAAQ,QAAQ;EAChB,MAAM,QAAQ;EACd;EACA,KAAK,QAAQ;CACf,CAAC;CACD,MAAM,SAAS,2BAA2B;EACxC,SAAS,QAAQ;EACjB;EACA;CACF,CAAC;CACD,QAAQ,UAAU,SAAS,MAAM;CACjC,MAAM,iBAAiB,KAAK,SAAS,QAAQ,uBAAuB;CACpE,MAAM,qBAAqB,KAAK,SAAS,QAAQ,qBAAqB;CACtE,QAAQ,MAAM,gBAAgB,GAAG,KAAK,UAAU,QAAQ,MAAM,CAAC,EAAE,GAAG;CACpE,QAAQ,MAAM,oBAAoB,8BAA8B,MAAM,CAAC;CACvE,OAAO;EAAE;EAAW;EAAU;EAAQ;EAAgB;CAAmB;AAC3E;AAEA,eAAsB,uBACpB,SACA,SAAiC;CAAC;CAAU;CAAO;AAAS,GACf;CAC7C,MAAM,cAAc,eAAe,OAAO;CAC1C,MAAM,SAAS,QAAQ,UAAU;CACjC,MAAM,WAAW,QAAQ,YAAY;CACrC,MAAM,MAA0C,CAAC;CACjD,MAAM,uBAAO,IAAI,IAAY;CAC7B,KAAK,MAAM,SAAS,QAAQ;EAC1B,MAAM,QAAQ,MAAM,QAAQ,YAAY,KAAK;EAC7C,KAAK,MAAM,QAAQ,OAAO;GACxB,MAAM,WAAW,KAAK,SAAS,QAAQ,YAAY,KAAK,EAAE;GAC1D,IAAI,aAAa,OAAO;GACxB,MAAM,MAAM,GAAG,YAAY,GAAG,KAAK;GACnC,IAAI,KAAK,IAAI,GAAG,GAAG;GACnB,KAAK,IAAI,GAAG;GACZ,IAAI,KAAK;IACP,IAAI;IACJ,MAAM;IACN;IACA,QAAQ,KAAK,UAAU;IACvB,UAAU,KAAK,YAAY;IAC3B;IACA,MAAM,CAAC,mBAAG,IAAI,IAAI,CAAC,UAAU,GAAI,KAAK,QAAQ,CAAC,CAAE,CAAC,CAAC;IACnD;GACF,CAAC;EACH;CACF;CACA,OAAO;AACT;AAEA,SAAgB,sBACd,SACqD;CACrD,MAAM,cAAc,eAAe,OAAO;CAC1C,OAAO;EACL,MAAM,GAAG,YAAY;EACrB,YAAY,CACV;GAAE,KAAK;GAAS,aAAa,WAAW,YAAY;EAAkB,GACtE;GAAE,KAAK;GAAU,aAAa;EAAiD,CACjF;EACA,YAAY,aAAgE;GAC1E,OAAO,SAAS,SAAS,eAAe,SAAS,GAAG,WAAW,GAAG,YAAY,EAAE;EAClF;EACA,MAAM,MAAM,EAAE,UAAU,YAAY;GAClC,MAAM,aAAa,MAAM,QAAQ,SAAS,SAAS,MAAM,QAAQ;GAEjE,OAAO;IACL,YAFiB,8BAA8B,UAEtC;IACT,WAAW,QAAQ,WAAW,KAAK;IACnC,OAAO,WAAW,SAAS;GAC7B;EACF;CACF;AACF;AAEA,SAAS,2BAAgD,OAIrC;CAClB,MAAM,eAAe,IAAI,IAAI,MAAM,UAAU,KAAK,aAAa,CAAC,SAAS,IAAI,QAAQ,CAAC,CAAC;CAavF,MAAM,aAZO,MAAM,SAAS,MAAM,KAAK,SAAS;EAC9C,MAAM,WAAW,aAAa,IAAI,KAAK,UAAU;EACjD,MAAM,QAAQ,gBAAgB,KAAK,WAAW;EAC9C,MAAM,QAAQ,OAAO,aAAa;EAClC,OAAO;GACL;GACA;GACA;GACA,SAAS,OAAO,WAAW,WAAW,QAAQ,IAAI,IAAI,OAAO;GAC7D,YAAY,OAAO,cAAc,CAAC;EACpC;CACF,CACsB,CAAC,CAAC,QAAQ,QAAQ,CAAC,IAAI,KAAK,KAAK;CACvD,MAAM,cAAc,WAAW,KAAK,QAAQ,IAAI,KAAK;CACrD,OAAO;EACL,aAAa,eAAe,MAAM,OAAO;EACzC,QAAQ,MAAM,QAAQ,UAAU;EAChC,UAAU,MAAM,QAAQ,YAAY;EACpC,GAAI,MAAM,QAAQ,SAAS,EAAE,QAAQ,MAAM,QAAQ,OAAO,IAAI,CAAC;EAC/D,QAAQ,MAAM,SAAS;EACvB,cAAc,MAAM,SAAS;EAC7B,MAAM,MAAM,SAAS;EACrB,WAAW,MAAM,SAAS;EAC1B,SAAS,MAAM,SAAS;EACxB,YAAY,MAAM,SAAS;EAC3B,YAAY,MAAM,UAAU;EAC5B,YAAY,MAAM,SAAS,MAAM;EACjC,aAAa,MAAM,SAAS,WAAW;EACvC,aAAa,MAAM,SAAS,WAAW;EACvC,cAAc,MAAM,SAAS,WAAW,KAAK;EAC7C,QAAQ,gBAAgB,aAAa,QAAQ,IAAI,UAAU,YAAY,WAAW;GAChF;GACA;GACA;EACF,CAAC;EACD,MAAM,gBAAgB,aAAa,QAAQ,IAAI,UAAU,QAAQ,CAAC,CAAC;EACnE,YAAY,oBAAoB,WAAW,KAAK,QAAQ,IAAI,UAAU,CAAC;EACvE,OAAO,aAAa,WAAW;EAC/B,SAAS,aAAa,WAAW,KAAK,QAAQ,IAAI,KAAK,OAAO,CAAC;EAC/D,WAAW,aAAa,WAAW,KAAK,QAAQ,IAAI,KAAK,UAAU,CAAC;CACtE;AACF;AAEA,SAAS,8BAA8B,QAAiC;CACtE,MAAM,YAAY,OAAO,QAAQ,OAAO,MAAM,CAAC,CAC5C,KAAK,CAAC,OAAO,aAAa;EACzB,OAAO,KAAK,MAAM,KAAK,QAAQ,EAAE,KAAK,IAAI,QAAQ,SAAS,EAAE,KAAK,IAAI,QAAQ,QAAQ,EAAE,KAAK,IAAI,QAAQ,MAAM,GAAG,EAAE,KAAK,IAAI,QAAQ,QAAQ,IAAI,EAAE,KAAK,IAAI,QAAQ,UAAU,GAAG,EAAE;CACrL,CAAC,CAAC,CACD,KAAK,IAAI;CACZ,MAAM,UAAU,OAAO,QAAQ,OAAO,UAAU,CAAC,CAC9C,MAAM,CAAC,IAAI,CAAC,OAAO,EAAE,cAAc,CAAC,CAAC,CAAC,CACtC,KAAK,CAAC,KAAK,UAAU,KAAK,IAAI,KAAK,KAAK,EAAE,KAAK,IAAI,KAAK,IAAI,EAAE,KAAK,IAAI,KAAK,GAAG,EAAE,GAAG,CAAC,CACrF,KAAK,IAAI;CACZ,OAAO;EACL,uBAAuB,OAAO;EAC9B;EACA,aAAa,OAAO;EACpB,gBAAgB,OAAO;EACvB,cAAc,OAAO;EACrB,eAAe,OAAO;EACtB,YAAY,OAAO,WAAW,UAAU,OAAO,YAAY,WAAW,OAAO,YAAY;EACzF,YAAY,IAAI,OAAO,YAAY;EACnC,iBAAiB,IAAI,OAAO,MAAM,IAAI,EAAE,WAAW,IAAI,OAAO,MAAM,MAAM,EAAE,QAAQ,IAAI,OAAO,MAAM,GAAG,EAAE,MAAM,OAAO,MAAM;EAC7H;EACA;EACA;EACA;EACA;EACA,aAAa;EACb;EACA;EACA;EACA;EACA;EACA,WAAW;EACX;CACF,CAAC,CAAC,KAAK,IAAI;AACb;AAEA,SAAS,eAAe,SAAuE;CAC7F,OAAO,QAAQ,MAAM,GAAG,QAAQ,UAAU,SAAS,GAAG,QAAQ,YAAY;AAC5E;AAEA,SAAS,8BAA8B,YAAyD;CAC9F,MAAM,aAAqC,EAAE,OAAO,QAAQ,WAAW,KAAK,EAAE;CAC9E,KAAK,MAAM,CAAC,KAAK,UAAU,OAAO,QAAQ,WAAW,cAAc,CAAC,CAAC,GACnE,IAAI,OAAO,SAAS,KAAK,GAAG,WAAW,OAAO;CAEhD,WAAW,SAAU,WAAW,UAAU,WAAW,QAAQ,IAAK,IAAI;CACtE,OAAO;AACT;AAEA,SAAS,oBACP,MACuC;CACvC,MAAM,yBAAS,IAAI,IAAsB;CACzC,KAAK,MAAM,OAAO,MAChB,KAAK,MAAM,CAAC,KAAK,UAAU,OAAO,QAAQ,GAAG,GAAG;EAC9C,IAAI,CAAC,OAAO,SAAS,KAAK,GAAG;EAC7B,MAAM,OAAO,OAAO,IAAI,GAAG,KAAK,CAAC;EACjC,KAAK,KAAK,KAAK;EACf,OAAO,IAAI,KAAK,IAAI;CACtB;CAEF,OAAO,OAAO,YAAY,CAAC,GAAG,OAAO,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,KAAK,UAAU,CAAC,KAAK,aAAa,IAAI,CAAC,CAAC,CAAC;AACjG;AAEA,SAAS,gBACP,MACA,OACA,YAA+B,CAAC,GACO;CACvC,MAAM,0BAAU,IAAI,IAAiB;CACrC,KAAK,MAAM,OAAO,WAAW,QAAQ,IAAI,KAAK,CAAC,CAAC;CAChD,KAAK,MAAM,OAAO,MAAM;EACtB,MAAM,OAAO,MAAM,GAAG;EACtB,KAAK,MAAM,OAAO,MAAM,QAAQ,IAAI,IAAI,OAAO,CAAC,IAAI,GAAG;GACrD,MAAM,OAAO,QAAQ,IAAI,GAAG,KAAK,CAAC;GAClC,KAAK,KAAK,GAAG;GACb,QAAQ,IAAI,KAAK,IAAI;EACvB;CACF;CACA,MAAM,MAA6C,CAAC;CACpD,KAAK,MAAM,CAAC,KAAK,SAAS,SAAS;EACjC,MAAM,YAAY;EAKlB,IAAI,OAAO;GACT,GAAG,KAAK;GACR,WAAW,KAAK,UAAU,KAAK,QAAQ,IAAI,KAAK,CAAC;GACjD,UAAU,KAAK,UAAU,KAAK,QAAS,IAAI,SAAS,IAAI,CAAE,CAAC;GAC3D,OAAO,aAAa,UAAU,KAAK,QAAQ,IAAI,KAAK,CAAC;GACrD,SAAS,aAAa,UAAU,KAAK,QAAQ,IAAI,KAAK,OAAO,CAAC;GAC9D,WAAW,aAAa,UAAU,KAAK,QAAQ,IAAI,KAAK,UAAU,CAAC;EACrE;CACF;CACA,OAAO;AACT;AAEA,SAAS,gBACP,aACwB;CACxB,OAAO,OAAO,OAAO,WAAW,CAAC,CAAC;AACpC;AAEA,SAAS,aAAa,QAAkD;CACtE,MAAM,SAAS,CAAC,GAAG,MAAM,CAAC,CAAC,OAAO,OAAO,QAAQ,CAAC,CAAC,MAAM,GAAG,MAAM,IAAI,CAAC;CACvE,IAAI,OAAO,WAAW,GAAG,OAAO;EAAE,GAAG;EAAG,KAAK;EAAG,MAAM;EAAG,QAAQ;EAAG,KAAK;EAAG,KAAK;CAAE;CACnF,OAAO;EACL,GAAG,OAAO;EACV,KAAK,OAAO;EACZ,MAAM,KAAK,MAAM;EACjB,QAAQ,WAAW,QAAQ,EAAG;EAC9B,KAAK,WAAW,QAAQ,EAAG;EAC3B,KAAK,OAAO,OAAO,SAAS;CAC9B;AACF;AAEA,SAAS,WAAW,cAAiC,GAAmB;CACtE,IAAI,aAAa,WAAW,GAAG,OAAO;CAKtC,OAAO,aAJO,KAAK,IACjB,aAAa,SAAS,GACtB,KAAK,IAAI,GAAG,KAAK,KAAK,IAAI,aAAa,MAAM,IAAI,CAAC,CAE5B;AAC1B;AAEA,SAAS,KAAK,QAAmC;CAC/C,MAAM,SAAS,OAAO,OAAO,OAAO,QAAQ;CAC5C,IAAI,OAAO,WAAW,GAAG,OAAO;CAChC,OAAO,OAAO,QAAQ,KAAK,UAAU,MAAM,OAAO,CAAC,IAAI,OAAO;AAChE;AAEA,SAAS,QAAQ,OAAuB;CACtC,IAAI,CAAC,OAAO,SAAS,KAAK,GAAG,OAAO;CACpC,IAAI,QAAQ,GAAG,OAAO;CACtB,IAAI,QAAQ,GAAG,OAAO;CACtB,OAAO;AACT;AAEA,SAAS,IAAI,OAAuB;CAClC,IAAI,CAAC,OAAO,SAAS,KAAK,GAAG,OAAO;CACpC,OAAO,MAAM,QAAQ,UAAU,KAAK,KAAK,IAAI,KAAK,KAAK,KAAK,IAAI,CAAC;AACnE;;;ACzSA,SAAgB,eAA4B,MAAmB;CAC7D,OAAO,KACJ,MAAM,OAAO,CAAC,CACd,KAAK,SAAS,KAAK,KAAK,CAAC,CAAC,CAC1B,OAAO,OAAO,CAAC,CACf,KAAK,MAAM,UAAU;EACpB,IAAI;GACF,OAAO,KAAK,MAAM,IAAI;EACxB,SAAS,OAAO;GACd,MAAM,IAAI,MAAM,qBAAqB,QAAQ,EAAE,IAAK,MAAgB,SAAS;EAC/E;CACF,CAAC;AACL;AAEA,SAAS,aAAa,MAA0B;CAC9C,OAAO,KACJ,MAAM,OAAO,CAAC,CACd,KAAK,SAAS,KAAK,KAAK,CAAC,CAAC,CAC1B,OAAO,OAAO,CAAC,CACf,QAAQ,SAAS,CAAC,KAAK,WAAW,GAAG,CAAC,CAAC,CACvC,KAAK,SAAS,KAAK,MAAM,QAAQ,CAAC;AACvC;AAEA,SAAgB,WAAW,MAAuC;CAChE,OAAO,aAAa,IAAI,CAAC,CAAC,SAAS,OAAO,UAAU;EAClD,IAAI,MAAM,SAAS,GAAG,OAAO,CAAC;EAC9B,MAAM,CAAC,SAAS,kBAAkB,mBAAmB,cAAc;EACnE,IAAI,CAAC,WAAW,CAAC,oBAAoB,CAAC,mBAAmB,OAAO,CAAC;EACjE,IAAI,QAAQ,YAAY,MAAM,cAAc,QAAQ,YAAY,MAAM,OAAO,OAAO,CAAC;EACrF,MAAM,aAAa,eAAe,KAAA,IAAY,mBAAmB;EAEjE,MAAM,QAAQ,OADI,eAAe,KAAA,IAAY,oBAAoB,UACnC;EAC9B,IAAI,CAAC,cAAc,CAAC,OAAO,SAAS,KAAK,GACvC,MAAM,IAAI,MAAM,qBAAqB,QAAQ,EAAE,mCAAmC;EAEpF,OAAO,CAAC;GAAE;GAAS;GAAY;EAAM,CAAC;CACxC,CAAC;AACH;AAEA,SAAgB,qBAAqB,MAA2C;CAC9E,OAAO,eAAwC,IAAI,CAAC,CAAC,KAAK,KAAK,UAAU;EACvE,MAAM,KAAK,YAAY,KAAK,KAAK,KAAK,YAAY,KAAK,IAAI;EAC3D,MAAM,OAAO,YAAY,KAAK,MAAM,KAAK,YAAY,KAAK,UAAU;EACpE,IAAI,CAAC,MAAM,SAAS,KAAA,GAClB,MAAM,IAAI,MAAM,2BAA2B,QAAQ,EAAE,oCAAoC;EAE3F,OAAO;GACL;GACA,OAAO,YAAY,KAAK,OAAO;GAC/B,MAAM;GACN,UAAU,WAAW,KAAK;IAAC;IAAO;IAAM;IAAS;IAAQ;GAAU,CAAC;EACtE;CACF,CAAC;AACH;AAEA,SAAgB,sBAAsB,MAAwC;CAC5E,OAAO,eAAwC,IAAI,CAAC,CAAC,KAAK,KAAK,UAAU;EACvE,MAAM,KAAK,YAAY,KAAK,KAAK,KAAK,YAAY,KAAK,IAAI,KAAK,YAAY,KAAK,UAAU;EAC3F,MAAM,QAAQ,YAAY,KAAK,MAAM,KAAK,YAAY,KAAK,OAAO;EAClE,IAAI,CAAC,MAAM,CAAC,OACV,MAAM,IAAI,MAAM,0BAA0B,QAAQ,EAAE,iCAAiC;EAEvF,OAAO;GACL;GACA,MAAM;GACN,UAAU,WAAW,KAAK;IAAC;IAAO;IAAM;IAAY;IAAQ;GAAO,CAAC;EACtE;CACF,CAAC;AACH;AAEA,SAAgB,4BACd,SACuD;CACvD,MAAM,+BAAe,IAAI,IAAqC;CAC9D,KAAK,MAAM,QAAQ,QAAQ,OAAO;EAChC,IAAI,KAAK,SAAS,GAAG;EACrB,MAAM,OAAO,aAAa,IAAI,KAAK,OAAO,KAAK,CAAC;EAChD,KAAK,KAAK,IAAI;EACd,aAAa,IAAI,KAAK,SAAS,IAAI;CACrC;CACA,MAAM,SACJ,QAAQ,0BAA0B,QAAQ,SACtC,OAAO,YAAY,QAAQ,OAAO,KAAK,aAAa,CAAC,SAAS,IAAI,QAAQ,CAAC,CAAC,IAC5E,KAAA;CACN,OAAO,QAAQ,QAAQ,SAAS,UAAU;EACxC,MAAM,QAAQ,aAAa,IAAI,MAAM,EAAE,KAAK,CAAC;EAC7C,IAAI,MAAM,WAAW,GAAG,OAAO,CAAC;EAChC,MAAM,QACJ,QAAQ,UAAU,MAAM,EAAE,KAAK,mBAAmB,GAAG,QAAQ,YAAY,GAAG,MAAM,IAAI;EACxF,OAAO,CACL;GACE,IAAI,MAAM;GACV;GACA,QAAQ,QAAQ;GAChB,UAAU;GACV,MAAM,CAAC,mBAAG,IAAI,IAAI,CAAC,GAAI,QAAQ,QAAQ,CAAC,GAAI,KAAK,CAAC,CAAC;GACnD,GAAI,QAAQ,SAAS,EAAE,QAAQ,QAAQ,OAAO,IAAI,CAAC;GACnD,GAAI,MAAM,WAAW,EAAE,UAAU,MAAM,SAAS,IAAI,CAAC;GACrD,SAAS;IACP,SAAS,MAAM;IACf,OAAO,MAAM;IACb,qBAAqB,MAAM,KAAK,SAAS,KAAK,UAAU;IACxD,gBAAgB,OAAO,YAAY,MAAM,KAAK,SAAS,CAAC,KAAK,YAAY,KAAK,KAAK,CAAC,CAAC;IACrF,GAAI,SAAS,EAAE,OAAO,IAAI,CAAC;IAC3B,GAAI,MAAM,WAAW,EAAE,UAAU,MAAM,SAAS,IAAI,CAAC;GACvD;EACF,CACF;CACF,CAAC;AACH;AAEA,SAAgB,kCACd,SAKA;CACA,MAAM,QAAQ,4BAA4B,OAAO;CACjD,OAAO;EACL,IAAI,QAAQ;EACZ,QAAQ,QAAQ;EAChB,UAAU;EACV,QAAQ,QAAQ;EAChB,eAAe,QAAQ,iBAAiB;EACxC,MAAM,YAAY,OAAO;GACvB,OAAO,MAAM,QAAQ,SAAS,KAAK,UAAU,KAAK;EACpD;EACA,MAAM,SAAS,MAAM,UAAU;GAC7B,OAAO,0BAA0B,KAAK,SAAS,UAAU,OAAO;EAClE;EACA,YAAY,QAAQ;GAClB,OAAO,QAAQ,UAAU,MAAM,KAAK,mBAAmB,GAAG,QAAQ,YAAY,GAAG,QAAQ;EAC3F;CACF;AACF;AAEA,SAAgB,0BACd,SACA,UACA,UAA8C,CAAC,GAC/C;CACA,MAAM,oBAAoB,8BACxB,UACA,QAAQ,qBAAqB,oBAC/B;CACA,MAAM,UAAU,iBAAiB,QAAQ,WAAW;EAAC;EAAG;EAAG;EAAG;CAAE,CAAC;CACjE,MAAM,aAAqC;EACzC,gBAAgB,QAAQ,oBAAoB;EAC5C,gBAAgB,kBAAkB;CACpC;CACA,KAAK,MAAM,UAAU,SACnB,OAAO,OACL,YACA,yBAAyB;EACvB;EACA,gBAAgB,QAAQ;EACxB;CACF,CAAC,CACH;CAEF,MAAM,gBAAgB,QAAQ,iBAAiB;CAC/C,MAAM,aAAa,QAAQ,cAAc;CACzC,MAAM,QAAQ,WAAW,kBAAkB,WAAW,QAAQ,QAAQ,QAAQ,SAAS,SAAS;CAGhG,OAAO;EACL;EACA,SAHA,WAAW,eAAe,WAAW,OAAO,QAAQ,QAAQ,SAAS,SAAS,WAGxD,QAAQ,iBAAiB;EAC/C;EACA,KAAK;GACH;GACA,qBAAqB,QAAQ;GAC7B,gBAAgB,QAAQ;EAC1B;CACF;AACF;AAEA,SAAS,8BACP,UACA,oBAA4B,sBAClB;CACV,IAAI,OAAO,aAAa,UACtB,OAAO,eAAe,SAAS,MAAM,iBAAiB,KAAK,CAAC,EAAA,CAAG,KAAK,UAAU,MAAM,KAAK,CAAC,CAAC;CAE7F,IAAI,MAAM,QAAQ,QAAQ,GAExB,OAAO,cACLC,SAAQ,SAAS,UAAU;EACzB,IAAI,OAAO,UAAU,UAAU,OAAO,CAAC,KAAK;EAC5C,OAAO,CAAC,MAAM,cAAc,MAAM,SAAS,MAAM,MAAM,EAAE;CAC3D,CAAC,CACH;CAEF,MAAM,iBAAiB;CACvB,OAAO,cACL;EACE,GAAI,eAAe,eAAe,CAAC;EACnC,GAAI,eAAe,OAAO,CAAC;EAC3B,IAAI,eAAe,WAAW,CAAC,EAAA,CAAG,KAC/B,UAAU,MAAM,cAAc,MAAM,SAAS,MAAM,MAAM,EAC5D;CACF,CAAC,CAAC,KAAK,UAAU,MAAM,KAAK,CAAC,CAC/B;AACF;AAEA,SAAS,yBAAyB,OAIP;CACzB,MAAM,MAAM,MAAM,kBAAkB,MAAM,GAAG,MAAM,MAAM;CACzD,MAAM,cAAc,OAAO,KAAK,MAAM,cAAc,CAAC,CAAC,QACnD,OAAO,MAAM,eAAe,MAAO,CACtC;CACA,MAAM,WAAW,IAAI,IAAI,WAAW;CACpC,MAAM,OAAO,IAAI,QAAQ,OAAO,SAAS,IAAI,EAAE,CAAC,CAAC,CAAC;CAClD,MAAM,oBAAoB,IAAI,WAAW,OAAO,SAAS,IAAI,EAAE,CAAC;CAChE,MAAM,SAAS,IAAI,MAAM;CACzB,OAAO;GACJ,MAAM,WAAW,OAAO,IAAI,IAAI;GAChC,SAAS,WAAW,SAAS,SAAS,IAAI,IAAI,OAAO,SAAS;GAC9D,YAAY,WAAW,OAAO,MAAM;GACpC,MAAM,WAAW,sBAAsB,KAAK,IAAI,KAAK,oBAAoB;GACzE,OAAO,WAAW,OAAO,MAAM,mBAAmB,MAAM,gBAAgB,MAAM,MAAM;CACvF;AACF;AAEA,SAAS,YAAY,KAA8B,KAAiC;CAClF,MAAM,QAAQ,IAAI;CAClB,OAAO,OAAO,UAAU,WAAW,QAAQ,KAAA;AAC7C;AAEA,SAAS,WACP,KACA,MACyB;CACzB,MAAM,QAAQ,IAAI,IAAI,IAAI;CAC1B,OAAO,OAAO,YAAY,OAAO,QAAQ,GAAG,CAAC,CAAC,QAAQ,CAAC,SAAS,CAAC,MAAM,IAAI,GAAG,CAAC,CAAC;AAClF;AAEA,SAAS,iBAAiB,SAAsC;CAC9D,OAAO,CACL,GAAG,IAAI,IAAI,QAAQ,KAAK,WAAW,KAAK,MAAM,MAAM,CAAC,CAAC,CAAC,QAAQ,WAAW,SAAS,CAAC,CAAC,CACvF,CAAC,CAAC,MAAM,GAAG,MAAM,IAAI,CAAC;AACxB;AAEA,SAAS,cAAc,QAAqC;CAC1D,MAAM,uBAAO,IAAI,IAAY;CAC7B,MAAM,MAAgB,CAAC;CACvB,KAAK,MAAM,SAAS,QAAQ;EAC1B,MAAM,UAAU,MAAM,KAAK;EAC3B,IAAI,CAAC,WAAW,KAAK,IAAI,OAAO,GAAG;EACnC,KAAK,IAAI,OAAO;EAChB,IAAI,KAAK,OAAO;CAClB;CACA,OAAO;AACT;AAEA,SAAS,OACP,mBACA,gBACA,QACQ;CACR,MAAM,MAAM,kBACT,MAAM,GAAG,MAAM,CAAC,CAChB,QACE,KAAK,YAAY,UAAU,MAAM,eAAe,eAAe,eAAe,GAAG,KAAK,GACvF,CACF;CACF,MAAM,QAAQ,OAAO,OAAO,cAAc,CAAC,CACxC,QAAQ,UAAU,QAAQ,CAAC,CAAC,CAC5B,MAAM,GAAG,MAAM,IAAI,CAAC,CAAC,CACrB,MAAM,GAAG,MAAM,CAAC,CAChB,QAAQ,KAAK,OAAO,UAAU,MAAM,eAAe,OAAO,KAAK,GAAG,CAAC;CACtE,OAAO,UAAU,IAAI,IAAI,MAAM;AACjC;AAEA,SAAS,eAAe,WAAmB,eAA+B;CACxE,IAAI,aAAa,GAAG,OAAO;CAC3B,QAAQ,KAAK,YAAY,KAAK,KAAK,KAAK,gBAAgB,CAAC;AAC3D"}
+{"version":3,"file":"index.js","names":["clamp01","parseToml","entries"],"sources":["../../src/benchmarks/calibration.ts","../../src/benchmarks/types.ts","../../src/benchmarks/harbor.ts","../../src/benchmarks/routing/dataset.ts","../../src/benchmarks/routing/index.ts","../../src/benchmarks/runner.ts","../../src/benchmarks/standard-formats.ts"],"sourcesContent":["import type { BenchmarkAdapter, BenchmarkDatasetItem, BenchmarkEvaluation } from './types'\n\nexport interface BenchmarkMetricCalibrationOptions<TPayload = unknown, TArtifact = string> {\n  adapter: BenchmarkAdapter<BenchmarkDatasetItem<TPayload>, TPayload, TArtifact>\n  item: BenchmarkDatasetItem<TPayload>\n  weakArtifact: TArtifact\n  strongArtifact: TArtifact\n  maxWeakScore?: number\n  minStrongScore?: number\n  minGap?: number\n}\n\nexport interface BenchmarkMetricCalibrationResult {\n  passed: boolean\n  weak: BenchmarkEvaluation\n  strong: BenchmarkEvaluation\n  weakScore: number\n  strongScore: number\n  gap: number\n  reasons: string[]\n}\n\nexport async function calibrateBenchmarkMetric<TPayload = unknown, TArtifact = string>(\n  options: BenchmarkMetricCalibrationOptions<TPayload, TArtifact>,\n): Promise<BenchmarkMetricCalibrationResult> {\n  const weak = await options.adapter.evaluate(options.item, options.weakArtifact)\n  const strong = await options.adapter.evaluate(options.item, options.strongArtifact)\n  const weakScore = clamp01(weak.score)\n  const strongScore = clamp01(strong.score)\n  const maxWeakScore = options.maxWeakScore ?? 0.3\n  const minStrongScore = options.minStrongScore ?? 0.7\n  const minGap = options.minGap ?? 0.4\n  const gap = strongScore - weakScore\n  const reasons = [\n    weakScore <= maxWeakScore\n      ? undefined\n      : `weak score ${weakScore.toFixed(3)} exceeds max ${maxWeakScore.toFixed(3)}`,\n    strongScore >= minStrongScore\n      ? undefined\n      : `strong score ${strongScore.toFixed(3)} below min ${minStrongScore.toFixed(3)}`,\n    gap >= minGap ? undefined : `gap ${gap.toFixed(3)} below min ${minGap.toFixed(3)}`,\n  ].filter((reason): reason is string => Boolean(reason))\n\n  return {\n    passed: reasons.length === 0,\n    weak,\n    strong,\n    weakScore,\n    strongScore,\n    gap,\n    reasons,\n  }\n}\n\nfunction clamp01(value: number): number {\n  if (!Number.isFinite(value)) return 0\n  if (value < 0) return 0\n  if (value > 1) return 1\n  return value\n}\n","/**\n * Shared types for the reference benchmark wrappers under\n * `src/benchmarks/`. Each wrapper exports the three functions in\n * `BenchmarkAdapter` plus its own typed `DatasetItem` shape.\n */\n\nimport type { DispatchContext, Scenario } from '../campaign/types'\nimport type { RunSplitTag } from '../run-record'\n\nexport type BenchmarkTaskKind =\n  | 'retrieval'\n  | 'rag-answer'\n  | 'hallucination'\n  | 'kb-improvement'\n  | 'routing'\n  | 'custom'\n\nexport type BenchmarkFamily =\n  | 'beir'\n  | 'mteb-retrieval'\n  | 'msmarco'\n  | 'trec-dl'\n  | 'miracl'\n  | 'lotte'\n  | 'bright'\n  | 'crag'\n  | 'hotpotqa'\n  | 'kilt'\n  | 'ragtruth'\n  | 'faithbench'\n  | 'first-party'\n  | 'custom'\n\nexport interface BenchmarkDatasetItem<TPayload = unknown> {\n  /** Stable dataset-local item id (used for split assignment + paper\n   *  references). Unique within a benchmark. */\n  id: string\n  /** Free-form payload. Each benchmark defines its own shape. */\n  payload: TPayload\n  /** Optional precomputed split. When absent, adapters use assignSplit(id). */\n  split?: RunSplitTag\n  /** Benchmark family this row came from, e.g. `beir` or `crag`. */\n  family?: BenchmarkFamily | string\n  /** Benchmark-local task kind, e.g. retrieval vs answer quality. */\n  taskKind?: BenchmarkTaskKind | string\n  /** Slice labels such as language, domain, freshness, multihop, long-tail. */\n  tags?: string[]\n  /** Dataset provenance, version, URL, or license notes. */\n  source?: BenchmarkSource\n  metadata?: Record<string, unknown>\n}\n\nexport interface BenchmarkEvaluation {\n  /** [0, 1] score for the response on this item. Exact-match\n   *  benchmarks use 0/1; partial-credit benchmarks may return\n   *  fractional values. */\n  score: number\n  /** Optional pass/fail projection. Defaults to `score > 0` when absent. */\n  passed?: boolean\n  /** Numeric sub-metrics. These become report dimensions. */\n  dimensions?: Record<string, number>\n  /** Optional bag of raw scoring signals — e.g. parsed numeric\n   *  answer, regex match, judge sub-scores. */\n  raw: Record<string, unknown>\n  notes?: string\n}\n\nexport interface BenchmarkSource {\n  name?: string\n  url?: string\n  version?: string\n  /** The exact upstream revision, such as a git commit, when `version` names a release. */\n  revision?: string\n  license?: string\n  citation?: string\n  /** Where the upstream states how a result must be produced to count, such as leaderboard rules. */\n  rules?: string\n  /**\n   * Whether the tasks or their solutions may be in a model's training data: `unknown` until\n   * checked, `none-known`, `suspected`, or `confirmed` (then report results as historical).\n   */\n  contamination?: 'unknown' | 'none-known' | 'suspected' | 'confirmed'\n}\n\n/** Common signature implemented by every adapter under `src/benchmarks/*`. */\n// `TPayload` is the per-item payload type; `_TItem` is preserved for\n// downstream type-narrowing extensions (a richer `BenchmarkDatasetItem`\n// subclass that adds e.g. provenance metadata) but is intentionally\n// unused here. `noUnusedLocals` requires the leading underscore.\nexport interface BenchmarkAdapter<_TItem = unknown, TPayload = unknown, TArtifact = string> {\n  /** Stable benchmark id such as `beir/nfcorpus` or `crag/smoke`. */\n  id?: string\n  family?: BenchmarkFamily | string\n  taskKind?: BenchmarkTaskKind | string\n  description?: string\n  source?: BenchmarkSource\n  defaultMetric?: string\n  /** Load the dataset for the given split. May hit the network on\n   *  first call but should be cache-friendly. Adapters that don't\n   *  ship the dataset itself MUST throw a clearly-marked error\n   *  pointing the caller at the loader script. */\n  loadDataset(split: RunSplitTag): Promise<BenchmarkDatasetItem<TPayload>[]>\n  /** Score a single response. Pure with respect to the inputs. */\n  evaluate(item: BenchmarkDatasetItem<TPayload>, artifact: TArtifact): Promise<BenchmarkEvaluation>\n  /** Deterministic split assignment via item id hashing. The\n   *  fraction of items in each split is implementation-defined but\n   *  MUST be stable across processes and platforms. */\n  assignSplit(itemId: string): RunSplitTag\n}\n\nexport interface BenchmarkScenario<TPayload = unknown> extends Scenario {\n  kind: 'benchmark'\n  benchmarkId: string\n  family: BenchmarkFamily | string\n  taskKind: BenchmarkTaskKind | string\n  splitTag: RunSplitTag\n  item: BenchmarkDatasetItem<TPayload>\n}\n\nexport type BenchmarkResponder<TPayload = unknown, TArtifact = string> = (input: {\n  scenario: BenchmarkScenario<TPayload>\n  item: BenchmarkDatasetItem<TPayload>\n  context: DispatchContext\n}) => Promise<TArtifact> | TArtifact\n\n// ── Deterministic split assignment ───────────────────────────────────\n\n/**\n * 32-bit FNV-1a hash. Stable, allocation-free, deterministic across\n * runtimes. We use it to assign items to splits rather than depending\n * on a polyfilled crypto.subtle path.\n */\nfunction fnv1a32(input: string): number {\n  let h = 0x811c9dc5\n  for (let i = 0; i < input.length; i++) {\n    h ^= input.charCodeAt(i) & 0xff\n    h = (h + ((h << 1) + (h << 4) + (h << 7) + (h << 8) + (h << 24))) >>> 0\n  }\n  return h >>> 0\n}\n\n/** Split-assignment seed shared across all benchmarks. Bumping this\n *  value reshuffles every split — do NOT do that lightly. */\nexport const BENCHMARK_SPLIT_SEED = 'agent-eval-v1'\n\n/**\n * Assign an item id to one of `'search' | 'dev' | 'holdout'` using a\n * stable 32-bit hash of `${seed}::${id}`. Default proportions:\n *\n *   search: 60%   (optimization-readable)\n *   dev:    20%   (held-out for tuning, leak-on-purpose during dev)\n *   holdout:20%   (paper-grade held-out, gated reads)\n */\nexport function deterministicSplit(\n  itemId: string,\n  seed: string = BENCHMARK_SPLIT_SEED,\n): RunSplitTag {\n  const h = fnv1a32(`${seed}::${itemId}`)\n  const pos = h / 0x100000000\n  if (pos < 0.6) return 'search'\n  if (pos < 0.8) return 'dev'\n  return 'holdout'\n}\n","/**\n * A Harbor dataset as a `BenchmarkAdapter`.\n *\n * Harbor (https://harborframework.com) lays out each task as a directory with `task.toml`,\n * `instruction.md`, the agent's `environment/`, and the `tests/` its verifier runs, usually in a\n * verifier image of their own. This importer reads a local checkout of such a dataset, at the\n * revision the caller pins in `source`, into `BenchmarkDatasetItem`s.\n *\n * Scoring stays Harbor's. `evaluate` hands the task and the agent's artifact to the caller's\n * `grade` callback, which runs the task's own verifier (for example `harbor run -p <taskDir>`\n * with an agent that installs the collected artifacts) and reports its reward. This package does\n * not start containers. A verifier that did not run is an error, never a score of zero.\n */\n\nimport { readdir, readFile, stat } from 'node:fs/promises'\nimport { join, resolve } from 'node:path'\n\nimport { parse as parseToml } from 'smol-toml'\n\nimport type { RunSplitTag } from '../run-record'\nimport type {\n  BenchmarkAdapter,\n  BenchmarkDatasetItem,\n  BenchmarkEvaluation,\n  BenchmarkSource,\n} from './types'\nimport { deterministicSplit } from './types'\n\n/** One Harbor task, as its `task.toml` and `instruction.md` state it. */\nexport interface HarborTaskPayload {\n  /** `[task].name`, such as `terminal-bench/interleaved-vigenere`; the directory name when absent. */\n  name: string\n  /** The task's directory name inside the dataset's `tasks/`. */\n  dir: string\n  /** Absolute path of the task directory, for the grader. */\n  taskDir: string\n  /** `instruction.md`, verbatim. */\n  instruction: string\n  /** `[agent].timeout_sec`; null when the task sets none. */\n  agentTimeoutSec: number | null\n  /** `[verifier].timeout_sec`; null when the task sets none. */\n  verifierTimeoutSec: number | null\n  /** `[verifier].environment_mode`: `separate` runs the tests in their own image. */\n  verifierMode: string | null\n  /** `[environment]` requests; null where the task states none. */\n  resources: {\n    cpus: number | null\n    memoryMb: number | null\n    storageMb: number | null\n    gpus: number\n  }\n  /** Paths the verifier reads from the agent's environment (`artifacts`, sources only). */\n  artifacts: string[]\n  /** The environment starts more than one container (a docker-compose file). */\n  multiContainer: boolean\n  /** `[metadata].category`. */\n  category: string | null\n  /** `[metadata].expert_time_estimate_hours`. */\n  expertTimeEstimateHours: number | null\n}\n\nexport type HarborTaskItem = BenchmarkDatasetItem<HarborTaskPayload>\n\n/** What the verifier returned for one task, or why it did not return. */\nexport type HarborGradeOutcome =\n  | { succeeded: true; value: { reward: number; detail?: string } }\n  | { succeeded: false; error: string }\n\nexport interface HarborBenchmarkOptions<TArtifact> {\n  /** Stable benchmark id, such as `terminal-bench@4.0.0`. */\n  id: string\n  /** A checkout of the dataset: the directory that holds `tasks/`. */\n  datasetDir: string\n  /** The upstream name, URL, version, revision, rules and contamination status the checkout is. */\n  source: BenchmarkSource & { revision: string }\n  /** Task directory names to load, in this order; every task under `tasks/` when absent. */\n  tasks?: readonly string[]\n  /** Run the task's own verifier on the agent's artifact. */\n  grade: (item: HarborTaskItem, artifact: TArtifact) => Promise<HarborGradeOutcome>\n  /** Split assignment; `deterministicSplit` over the benchmark id and task directory by default. */\n  assignSplit?: (itemId: string) => RunSplitTag\n}\n\n/** Harbor adapter plus a loader that ignores splits, since a public benchmark reports every task. */\nexport interface HarborBenchmarkAdapter<TArtifact>\n  extends BenchmarkAdapter<HarborTaskItem, HarborTaskPayload, TArtifact> {\n  loadAll(): Promise<HarborTaskItem[]>\n}\n\nconst table = (value: unknown): Record<string, unknown> =>\n  value !== null && typeof value === 'object' && !Array.isArray(value)\n    ? (value as Record<string, unknown>)\n    : {}\nconst numberOrNull = (value: unknown): number | null =>\n  typeof value === 'number' && Number.isFinite(value) ? value : null\nconst textOrNull = (value: unknown): string | null =>\n  typeof value === 'string' && value.length > 0 ? value : null\n\n/** Read one Harbor task directory. Throws when `task.toml` or `instruction.md` is missing or malformed. */\nexport async function readHarborTask(taskDir: string): Promise<HarborTaskPayload> {\n  const dir = resolve(taskDir)\n  const config = parseToml(await readFile(join(dir, 'task.toml'), 'utf8')) as Record<\n    string,\n    unknown\n  >\n  const instruction = await readFile(join(dir, 'instruction.md'), 'utf8')\n  const task = table(config.task)\n  const agent = table(config.agent)\n  const verifier = table(config.verifier)\n  const environment = table(config.environment)\n  const metadata = table(config.metadata)\n  const artifacts = (Array.isArray(config.artifacts) ? config.artifacts : []).map(\n    (entry, index) => {\n      const source = typeof entry === 'string' ? entry : table(entry).source\n      if (typeof source !== 'string' || source.length === 0) {\n        throw new Error(`${dir}/task.toml: artifacts[${index}] names no source path`)\n      }\n      return source\n    },\n  )\n  const environmentFiles = await readdir(join(dir, 'environment')).catch(() => [] as string[])\n  const name = dir.split('/').pop() ?? dir\n  return {\n    name: textOrNull(task.name) ?? name,\n    dir: name,\n    taskDir: dir,\n    instruction,\n    agentTimeoutSec: numberOrNull(agent.timeout_sec),\n    verifierTimeoutSec: numberOrNull(verifier.timeout_sec),\n    verifierMode: textOrNull(verifier.environment_mode),\n    resources: {\n      cpus: numberOrNull(environment.cpus),\n      memoryMb: numberOrNull(environment.memory_mb),\n      storageMb: numberOrNull(environment.storage_mb),\n      gpus: numberOrNull(environment.gpus) ?? 0,\n    },\n    artifacts,\n    multiContainer: environmentFiles.some((file) => /compose\\.ya?ml$/u.test(file)),\n    category: textOrNull(metadata.category),\n    expertTimeEstimateHours: numberOrNull(metadata.expert_time_estimate_hours),\n  }\n}\n\n/** Import a Harbor dataset checkout as a `BenchmarkAdapter` whose scorer is the dataset's own verifier. */\nexport function createHarborBenchmarkAdapter<TArtifact = string>(\n  options: HarborBenchmarkOptions<TArtifact>,\n): HarborBenchmarkAdapter<TArtifact> {\n  const split =\n    options.assignSplit ?? ((itemId: string) => deterministicSplit(`${options.id}::${itemId}`))\n  const tasksDir = join(resolve(options.datasetDir), 'tasks')\n\n  async function loadAll(): Promise<HarborTaskItem[]> {\n    const names =\n      options.tasks ??\n      (\n        await Promise.all(\n          (\n            await readdir(tasksDir)\n          ).map(async (entry) =>\n            (await stat(join(tasksDir, entry, 'task.toml')).catch(() => null)) === null\n              ? null\n              : entry,\n          ),\n        )\n      )\n        .filter((entry): entry is string => entry !== null)\n        .sort()\n    return Promise.all(\n      names.map(async (dir) => {\n        const payload = await readHarborTask(join(tasksDir, dir))\n        return {\n          id: dir,\n          payload,\n          split: split(dir),\n          family: options.source.name ?? 'harbor',\n          taskKind: 'harbor-task',\n          tags: payload.category === null ? [] : [payload.category],\n          source: options.source,\n        }\n      }),\n    )\n  }\n\n  return {\n    id: options.id,\n    family: options.source.name ?? 'harbor',\n    taskKind: 'harbor-task',\n    description:\n      `${options.source.name ?? 'Harbor dataset'} ${options.source.version ?? ''} at ${options.source.revision}, scored by its own verifier`.replace(\n        /\\s+/gu,\n        ' ',\n      ),\n    source: options.source,\n    defaultMetric: 'reward',\n    loadAll,\n    async loadDataset(tag: RunSplitTag) {\n      return (await loadAll()).filter((item) => item.split === tag)\n    },\n    async evaluate(item: HarborTaskItem, artifact: TArtifact): Promise<BenchmarkEvaluation> {\n      const outcome = await options.grade(item, artifact)\n      if (!outcome.succeeded)\n        throw new Error(\n          `${options.id} ${item.id}: the verifier did not score this artifact: ${outcome.error}`,\n        )\n      const { reward, detail } = outcome.value\n      if (!(Number.isFinite(reward) && reward >= 0 && reward <= 1)) {\n        throw new Error(\n          `${options.id} ${item.id}: the verifier returned reward ${reward}, outside [0, 1]`,\n        )\n      }\n      return {\n        score: reward,\n        passed: reward === 1,\n        raw: { reward, ...(detail === undefined ? {} : { detail }) },\n      }\n    },\n    assignSplit: split,\n  }\n}\n","/**\n * Synthetic routing dataset. 16 tasks across 4 categories. Used as a\n * deterministic, dependency-free benchmark for any router that maps a\n * natural-language request to one of a fixed set of route labels.\n *\n * Format (see `routing/README.md` for prose):\n *\n *   {\n *     id:           stable per-task ID (matches across processes).\n *     category:     one of the four route labels.\n *     prompt:       the user-facing request the router must classify.\n *     route:        the ground-truth route the router should pick.\n *     synonyms:     other strings that count as a correct answer.\n *     hardNegatives:close-but-wrong route labels — used to detect the\n *                   \"always picks the popular route\" failure mode.\n *   }\n *\n * The four categories are intentionally cross-domain (file ops,\n * math, search, conversation) so a router that collapses to one\n * category is easy to spot.\n */\n\nexport interface RoutingItem {\n  id: string\n  category: 'file' | 'math' | 'search' | 'chat'\n  prompt: string\n  /** Canonical correct route label. */\n  route: string\n  /** Alternate route labels that also count as correct. */\n  synonyms: string[]\n  /** Wrong-but-tempting route labels (for analysis, not grading). */\n  hardNegatives: string[]\n}\n\nexport const ROUTING_DATASET: RoutingItem[] = [\n  {\n    id: 'file_001',\n    category: 'file',\n    prompt: 'Save the meeting notes to /tmp/notes-2025-04.md as markdown.',\n    route: 'fs.write',\n    synonyms: ['filesystem.write', 'write_file'],\n    hardNegatives: ['fs.read', 'chat.reply'],\n  },\n  {\n    id: 'file_002',\n    category: 'file',\n    prompt: 'Read the contents of /etc/hosts and summarize the entries.',\n    route: 'fs.read',\n    synonyms: ['filesystem.read', 'read_file'],\n    hardNegatives: ['fs.write', 'search.web'],\n  },\n  {\n    id: 'file_003',\n    category: 'file',\n    prompt: 'List every Python file under src/ recursively.',\n    route: 'fs.list',\n    synonyms: ['filesystem.list', 'list_files'],\n    hardNegatives: ['fs.read', 'search.code'],\n  },\n  {\n    id: 'file_004',\n    category: 'file',\n    prompt: 'Delete the cached build at .turbo/cache.',\n    route: 'fs.delete',\n    synonyms: ['filesystem.delete', 'remove_file'],\n    hardNegatives: ['fs.write', 'fs.list'],\n  },\n  {\n    id: 'math_001',\n    category: 'math',\n    prompt: 'What is the integral of 3x^2 + 2x from 0 to 5?',\n    route: 'math.integral',\n    synonyms: ['calculator.integral', 'math.solve'],\n    hardNegatives: ['math.derivative', 'chat.reply'],\n  },\n  {\n    id: 'math_002',\n    category: 'math',\n    prompt: 'Compute the derivative of sin(x) * cos(x).',\n    route: 'math.derivative',\n    synonyms: ['calculator.derivative', 'math.solve'],\n    hardNegatives: ['math.integral', 'math.algebra'],\n  },\n  {\n    id: 'math_003',\n    category: 'math',\n    prompt: 'Solve 2x + 7 = 19 for x.',\n    route: 'math.algebra',\n    synonyms: ['calculator.algebra', 'math.solve'],\n    hardNegatives: ['math.derivative', 'math.integral'],\n  },\n  {\n    id: 'math_004',\n    category: 'math',\n    prompt: 'What is the prime factorization of 360?',\n    route: 'math.numbertheory',\n    synonyms: ['calculator.factor', 'math.solve'],\n    hardNegatives: ['math.algebra', 'search.web'],\n  },\n  {\n    id: 'search_001',\n    category: 'search',\n    prompt: 'Find recent papers on agent prompt optimization with held-out promotion gates.',\n    route: 'search.web',\n    synonyms: ['web.search', 'search.papers'],\n    hardNegatives: ['search.code', 'chat.reply'],\n  },\n  {\n    id: 'search_002',\n    category: 'search',\n    prompt: 'Search the codebase for every call site of `runProposeReview`.',\n    route: 'search.code',\n    synonyms: ['code.search', 'grep'],\n    hardNegatives: ['search.web', 'fs.read'],\n  },\n  {\n    id: 'search_003',\n    category: 'search',\n    prompt: 'What is the latest release of the Tangle network on GitHub?',\n    route: 'search.web',\n    synonyms: ['web.search', 'github.releases'],\n    hardNegatives: ['search.code', 'chat.reply'],\n  },\n  {\n    id: 'search_004',\n    category: 'search',\n    prompt: 'Find all TODO comments in the agent-eval src tree.',\n    route: 'search.code',\n    synonyms: ['code.search', 'grep'],\n    hardNegatives: ['search.web', 'fs.list'],\n  },\n  {\n    id: 'chat_001',\n    category: 'chat',\n    prompt: 'Hi there, how are you doing today?',\n    route: 'chat.reply',\n    synonyms: ['conversation.reply'],\n    hardNegatives: ['search.web', 'fs.read'],\n  },\n  {\n    id: 'chat_002',\n    category: 'chat',\n    prompt: 'Please explain the difference between an LLM and a foundation model.',\n    route: 'chat.reply',\n    synonyms: ['conversation.reply', 'qa.answer'],\n    hardNegatives: ['search.web', 'math.algebra'],\n  },\n  {\n    id: 'chat_003',\n    category: 'chat',\n    prompt: 'Tell me a short joke about distributed systems.',\n    route: 'chat.reply',\n    synonyms: ['conversation.reply'],\n    hardNegatives: ['search.web', 'fs.read'],\n  },\n  {\n    id: 'chat_004',\n    category: 'chat',\n    prompt: 'Acknowledge my last message with a thumbs up.',\n    route: 'chat.reply',\n    synonyms: ['conversation.reply', 'react'],\n    hardNegatives: ['fs.write', 'search.web'],\n  },\n]\n","/**\n * Routing benchmark — synthetic, dependency-free, ships in the\n * package. 16 cross-category items in `dataset.ts`. See\n * `routing/README.md` for the format.\n *\n * `evaluate` does case-insensitive exact match against the canonical\n * route plus declared synonyms. The first valid route token in the\n * response wins; everything else is ignored. Wrong answers also\n * report whether they hit a hard negative — useful when triaging\n * \"always picks the popular route\" failure modes.\n */\n\nimport type { RunSplitTag } from '../../run-record'\nimport type { BenchmarkAdapter, BenchmarkDatasetItem, BenchmarkEvaluation } from '../types'\nimport { deterministicSplit } from '../types'\nimport { ROUTING_DATASET, type RoutingItem } from './dataset'\n\nexport type { RoutingItem }\nexport type RoutingPayload = RoutingItem\nexport type RoutingDatasetItem = BenchmarkDatasetItem<RoutingPayload>\n\nclass RoutingAdapter implements BenchmarkAdapter<RoutingDatasetItem, RoutingPayload> {\n  readonly id = 'first-party/routing'\n  readonly family = 'first-party'\n  readonly taskKind = 'routing'\n  readonly description = 'Synthetic fixed-route classification smoke benchmark'\n  readonly defaultMetric = 'route_exact_match'\n\n  async loadDataset(split: RunSplitTag): Promise<RoutingDatasetItem[]> {\n    return ROUTING_DATASET.map((item) => ({ id: item.id, payload: item })).filter(\n      (it) => assignSplitImpl(it.id) === split,\n    )\n  }\n\n  async evaluate(item: RoutingDatasetItem, response: string): Promise<BenchmarkEvaluation> {\n    const tokens = extractRouteTokens(response)\n    const correct = new Set<string>(\n      [item.payload.route, ...item.payload.synonyms].map((s) => s.toLowerCase()),\n    )\n    const hardNeg = new Set<string>(item.payload.hardNegatives.map((s) => s.toLowerCase()))\n    const firstMatch = tokens.find((t) => correct.has(t.toLowerCase())) ?? null\n    const firstHardNeg = tokens.find((t) => hardNeg.has(t.toLowerCase())) ?? null\n    const score = firstMatch ? 1 : 0\n    return {\n      score,\n      raw: {\n        firstToken: tokens[0] ?? null,\n        matchedRoute: firstMatch,\n        hitHardNegative: Boolean(firstHardNeg),\n        hardNegativeRoute: firstHardNeg,\n        category: item.payload.category,\n      },\n    }\n  }\n\n  assignSplit(itemId: string): RunSplitTag {\n    return assignSplitImpl(itemId)\n  }\n}\n\nfunction assignSplitImpl(itemId: string): RunSplitTag {\n  return deterministicSplit(`routing::${itemId}`)\n}\n\n/**\n * Pull route-shaped tokens out of a model response. Routes look like\n * `category.action` (`fs.write`, `chat.reply`). Bare alphanumerics\n * are not routes, but `category.action` patterns are robust to most\n * model wrappers (JSON output, prose explanations, code fences).\n */\nexport function extractRouteTokens(response: string): string[] {\n  const matches = response.match(/[a-z][a-z0-9_]*\\.[a-z][a-z0-9_]*/gi)\n  return matches ?? []\n}\n\nconst adapter = new RoutingAdapter()\n\nexport const loadDataset = adapter.loadDataset.bind(adapter)\nexport const evaluate = adapter.evaluate.bind(adapter)\nexport const assignSplit = adapter.assignSplit.bind(adapter)\nexport { ROUTING_DATASET, RoutingAdapter }\n","import { join } from 'node:path'\nimport {\n  type CampaignResult,\n  type DispatchFn,\n  type JudgeConfig,\n  type JudgeScore,\n  runCampaign,\n  type Scenario,\n} from '../campaign'\nimport { type CampaignStorage, fsCampaignStorage } from '../campaign/storage'\nimport type { RunSplitTag } from '../run-record'\nimport type {\n  BenchmarkAdapter,\n  BenchmarkDatasetItem,\n  BenchmarkEvaluation,\n  BenchmarkFamily,\n  BenchmarkResponder,\n  BenchmarkScenario,\n  BenchmarkTaskKind,\n} from './types'\n\nexport interface BenchmarkRunOptions<TPayload = unknown, TArtifact = string> {\n  adapter: BenchmarkAdapter<BenchmarkDatasetItem<TPayload>, TPayload, TArtifact>\n  respond: BenchmarkResponder<TPayload, TArtifact>\n  splits?: readonly RunSplitTag[]\n  runDir: string\n  repo?: string\n  seed?: number\n  reps?: number\n  resumable?: boolean\n  costCeiling?: number\n  maxConcurrency?: number\n  dispatchTimeoutMs?: number\n  expectUsage?: 'assert' | 'warn' | 'off'\n  storage?: CampaignStorage\n  now?: () => Date\n}\n\nexport interface BenchmarkReport {\n  benchmarkId: string\n  family: BenchmarkFamily | string\n  taskKind: BenchmarkTaskKind | string\n  source?: BenchmarkAdapter['source']\n  runDir: string\n  manifestHash: string\n  seed: number\n  startedAt: string\n  endedAt: string\n  durationMs: number\n  totalItems: number\n  totalCells: number\n  cellsFailed: number\n  cellsCached: number\n  totalCostUsd: number\n  splits: Record<string, BenchmarkSliceSummary>\n  tags: Record<string, BenchmarkSliceSummary>\n  dimensions: Record<string, BenchmarkDistribution>\n  score: BenchmarkDistribution\n  costUsd: BenchmarkDistribution\n  latencyMs: BenchmarkDistribution\n}\n\nexport interface BenchmarkSliceSummary {\n  n: number\n  meanScore: number\n  passRate: number\n  score: BenchmarkDistribution\n  costUsd: BenchmarkDistribution\n  latencyMs: BenchmarkDistribution\n}\n\nexport interface BenchmarkDistribution {\n  n: number\n  min: number\n  mean: number\n  median: number\n  p90: number\n  max: number\n}\n\nexport interface BenchmarkRunResult<TPayload = unknown, TArtifact = string> {\n  scenarios: Array<BenchmarkScenario<TPayload>>\n  campaign: CampaignResult<TArtifact, BenchmarkScenario<TPayload>>\n  report: BenchmarkReport\n  reportJsonPath: string\n  reportMarkdownPath: string\n}\n\nexport async function runBenchmarkAdapter<TPayload = unknown, TArtifact = string>(\n  options: BenchmarkRunOptions<TPayload, TArtifact>,\n): Promise<BenchmarkRunResult<TPayload, TArtifact>> {\n  const storage = options.storage ?? fsCampaignStorage()\n  const benchmarkId = benchmarkIdFor(options.adapter)\n  const scenarios = await loadBenchmarkScenarios(options.adapter, options.splits)\n  const judge = benchmarkAdapterJudge(options.adapter)\n  const dispatch: DispatchFn<BenchmarkScenario<TPayload>, TArtifact> = async (\n    scenario,\n    context,\n  ) => {\n    return options.respond({ scenario, item: scenario.item, context })\n  }\n\n  const campaign = await runCampaign<BenchmarkScenario<TPayload>, TArtifact>({\n    scenarios,\n    dispatch,\n    dispatchRef: `benchmark:${benchmarkId}`,\n    judges: [judge],\n    seed: options.seed,\n    reps: options.reps,\n    resumable: options.resumable,\n    costCeiling: options.costCeiling,\n    maxConcurrency: options.maxConcurrency,\n    dispatchTimeoutMs: options.dispatchTimeoutMs,\n    expectUsage: options.expectUsage ?? 'off',\n    runDir: options.runDir,\n    repo: options.repo,\n    storage,\n    now: options.now,\n  })\n  const report = summarizeBenchmarkCampaign({\n    adapter: options.adapter,\n    scenarios,\n    campaign,\n  })\n  storage.ensureDir(campaign.runDir)\n  const reportJsonPath = join(campaign.runDir, 'benchmark-report.json')\n  const reportMarkdownPath = join(campaign.runDir, 'benchmark-report.md')\n  storage.write(reportJsonPath, `${JSON.stringify(report, null, 2)}\\n`)\n  storage.write(reportMarkdownPath, renderBenchmarkReportMarkdown(report))\n  return { scenarios, campaign, report, reportJsonPath, reportMarkdownPath }\n}\n\nexport async function loadBenchmarkScenarios<TPayload>(\n  adapter: BenchmarkAdapter<BenchmarkDatasetItem<TPayload>, TPayload, unknown>,\n  splits: readonly RunSplitTag[] = ['search', 'dev', 'holdout'],\n): Promise<Array<BenchmarkScenario<TPayload>>> {\n  const benchmarkId = benchmarkIdFor(adapter)\n  const family = adapter.family ?? 'custom'\n  const taskKind = adapter.taskKind ?? 'custom'\n  const out: Array<BenchmarkScenario<TPayload>> = []\n  const seen = new Set<string>()\n  for (const split of splits) {\n    const items = await adapter.loadDataset(split)\n    for (const item of items) {\n      const splitTag = item.split ?? adapter.assignSplit(item.id)\n      if (splitTag !== split) continue\n      const key = `${benchmarkId}:${item.id}`\n      if (seen.has(key)) continue\n      seen.add(key)\n      out.push({\n        id: key,\n        kind: 'benchmark',\n        benchmarkId,\n        family: item.family ?? family,\n        taskKind: item.taskKind ?? taskKind,\n        splitTag,\n        tags: [...new Set([splitTag, ...(item.tags ?? [])])],\n        item,\n      })\n    }\n  }\n  return out\n}\n\nexport function benchmarkAdapterJudge<TPayload, TArtifact>(\n  adapter: BenchmarkAdapter<BenchmarkDatasetItem<TPayload>, TPayload, TArtifact>,\n): JudgeConfig<TArtifact, BenchmarkScenario<TPayload>> {\n  const benchmarkId = benchmarkIdFor(adapter)\n  return {\n    name: `${benchmarkId}:score`,\n    dimensions: [\n      { key: 'score', description: `Primary ${benchmarkId} benchmark score` },\n      { key: 'passed', description: 'Binary pass projection for aggregate reporting' },\n    ],\n    appliesTo: (scenario: Scenario): scenario is BenchmarkScenario<TPayload> => {\n      return scenario.kind === 'benchmark' && scenario.id.startsWith(`${benchmarkId}:`)\n    },\n    async score({ artifact, scenario }) {\n      const evaluation = await adapter.evaluate(scenario.item, artifact)\n      const dimensions = normalizeEvaluationDimensions(evaluation)\n      return {\n        dimensions,\n        composite: clamp01(evaluation.score),\n        notes: evaluation.notes ?? '',\n      }\n    },\n  }\n}\n\nfunction summarizeBenchmarkCampaign<TPayload, TArtifact>(input: {\n  adapter: BenchmarkAdapter<BenchmarkDatasetItem<TPayload>, TPayload, TArtifact>\n  scenarios: Array<BenchmarkScenario<TPayload>>\n  campaign: CampaignResult<TArtifact, BenchmarkScenario<TPayload>>\n}): BenchmarkReport {\n  const scenarioById = new Map(input.scenarios.map((scenario) => [scenario.id, scenario]))\n  const rows = input.campaign.cells.map((cell) => {\n    const scenario = scenarioById.get(cell.scenarioId)\n    const judge = firstJudgeScore(cell.judgeScores)\n    const score = judge?.composite ?? 0\n    return {\n      cell,\n      scenario,\n      score,\n      passed: (judge?.dimensions.passed ?? (score > 0 ? 1 : 0)) >= 1,\n      dimensions: judge?.dimensions ?? {},\n    }\n  })\n  const successful = rows.filter((row) => !row.cell.error)\n  const scoreValues = successful.map((row) => row.score)\n  return {\n    benchmarkId: benchmarkIdFor(input.adapter),\n    family: input.adapter.family ?? 'custom',\n    taskKind: input.adapter.taskKind ?? 'custom',\n    ...(input.adapter.source ? { source: input.adapter.source } : {}),\n    runDir: input.campaign.runDir,\n    manifestHash: input.campaign.manifestHash,\n    seed: input.campaign.seed,\n    startedAt: input.campaign.startedAt,\n    endedAt: input.campaign.endedAt,\n    durationMs: input.campaign.durationMs,\n    totalItems: input.scenarios.length,\n    totalCells: input.campaign.cells.length,\n    cellsFailed: input.campaign.aggregates.cellsFailed,\n    cellsCached: input.campaign.aggregates.cellsCached,\n    totalCostUsd: input.campaign.aggregates.cost.totalCostUsd,\n    splits: summarizeSlices(successful, (row) => row.scenario?.splitTag ?? 'unknown', [\n      'search',\n      'dev',\n      'holdout',\n    ]),\n    tags: summarizeSlices(successful, (row) => row.scenario?.tags ?? []),\n    dimensions: summarizeDimensions(successful.map((row) => row.dimensions)),\n    score: distribution(scoreValues),\n    costUsd: distribution(successful.map((row) => row.cell.costUsd)),\n    latencyMs: distribution(successful.map((row) => row.cell.durationMs)),\n  }\n}\n\nfunction renderBenchmarkReportMarkdown(report: BenchmarkReport): string {\n  const splitRows = Object.entries(report.splits)\n    .map(([split, summary]) => {\n      return `| ${split} | ${summary.n} | ${fmt(summary.meanScore)} | ${fmt(summary.passRate)} | ${fmt(summary.score.p90)} | ${fmt(summary.costUsd.mean)} | ${fmt(summary.latencyMs.p90)} |`\n    })\n    .join('\\n')\n  const dimRows = Object.entries(report.dimensions)\n    .sort(([a], [b]) => a.localeCompare(b))\n    .map(([key, dist]) => `| ${key} | ${dist.n} | ${fmt(dist.mean)} | ${fmt(dist.p90)} |`)\n    .join('\\n')\n  return [\n    `# Benchmark Report: ${report.benchmarkId}`,\n    '',\n    `- family: ${report.family}`,\n    `- task kind: ${report.taskKind}`,\n    `- run dir: ${report.runDir}`,\n    `- manifest: ${report.manifestHash}`,\n    `- cells: ${report.totalCells} total, ${report.cellsFailed} failed, ${report.cellsCached} cached`,\n    `- cost: $${fmt(report.totalCostUsd)}`,\n    `- score: mean ${fmt(report.score.mean)}, median ${fmt(report.score.median)}, p90 ${fmt(report.score.p90)}, n=${report.score.n}`,\n    '',\n    '## Splits',\n    '',\n    '| split | n | mean score | pass rate | score p90 | mean cost | latency p90 ms |',\n    '| --- | ---: | ---: | ---: | ---: | ---: | ---: |',\n    splitRows || '| none | 0 | 0 | 0 | 0 | 0 | 0 |',\n    '',\n    '## Dimensions',\n    '',\n    '| dimension | n | mean | p90 |',\n    '| --- | ---: | ---: | ---: |',\n    dimRows || '| none | 0 | 0 | 0 |',\n    '',\n  ].join('\\n')\n}\n\nfunction benchmarkIdFor(adapter: Pick<BenchmarkAdapter, 'id' | 'family' | 'taskKind'>): string {\n  return adapter.id ?? `${adapter.family ?? 'custom'}/${adapter.taskKind ?? 'custom'}`\n}\n\nfunction normalizeEvaluationDimensions(evaluation: BenchmarkEvaluation): Record<string, number> {\n  const dimensions: Record<string, number> = { score: clamp01(evaluation.score) }\n  for (const [key, value] of Object.entries(evaluation.dimensions ?? {})) {\n    if (Number.isFinite(value)) dimensions[key] = value\n  }\n  dimensions.passed = (evaluation.passed ?? evaluation.score > 0) ? 1 : 0\n  return dimensions\n}\n\nfunction summarizeDimensions(\n  rows: Array<Record<string, number>>,\n): Record<string, BenchmarkDistribution> {\n  const values = new Map<string, number[]>()\n  for (const row of rows) {\n    for (const [key, value] of Object.entries(row)) {\n      if (!Number.isFinite(value)) continue\n      const list = values.get(key) ?? []\n      list.push(value)\n      values.set(key, list)\n    }\n  }\n  return Object.fromEntries([...values.entries()].map(([key, vals]) => [key, distribution(vals)]))\n}\n\nfunction summarizeSlices<T>(\n  rows: T[],\n  keyOf: (row: T) => string | string[],\n  knownKeys: readonly string[] = [],\n): Record<string, BenchmarkSliceSummary> {\n  const grouped = new Map<string, T[]>()\n  for (const key of knownKeys) grouped.set(key, [])\n  for (const row of rows) {\n    const keys = keyOf(row)\n    for (const key of Array.isArray(keys) ? keys : [keys]) {\n      const list = grouped.get(key) ?? []\n      list.push(row)\n      grouped.set(key, list)\n    }\n  }\n  const out: Record<string, BenchmarkSliceSummary> = {}\n  for (const [key, list] of grouped) {\n    const withShape = list as Array<{\n      score: number\n      passed: boolean\n      cell: { costUsd: number; durationMs: number }\n    }>\n    out[key] = {\n      n: list.length,\n      meanScore: mean(withShape.map((row) => row.score)),\n      passRate: mean(withShape.map((row) => (row.passed ? 1 : 0))),\n      score: distribution(withShape.map((row) => row.score)),\n      costUsd: distribution(withShape.map((row) => row.cell.costUsd)),\n      latencyMs: distribution(withShape.map((row) => row.cell.durationMs)),\n    }\n  }\n  return out\n}\n\nfunction firstJudgeScore(\n  judgeScores: CampaignResult<unknown>['cells'][number]['judgeScores'],\n): JudgeScore | undefined {\n  return Object.values(judgeScores)[0]\n}\n\nfunction distribution(values: readonly number[]): BenchmarkDistribution {\n  const finite = [...values].filter(Number.isFinite).sort((a, b) => a - b)\n  if (finite.length === 0) return { n: 0, min: 0, mean: 0, median: 0, p90: 0, max: 0 }\n  return {\n    n: finite.length,\n    min: finite[0]!,\n    mean: mean(finite),\n    median: percentile(finite, 0.5),\n    p90: percentile(finite, 0.9),\n    max: finite[finite.length - 1]!,\n  }\n}\n\nfunction percentile(sortedValues: readonly number[], p: number): number {\n  if (sortedValues.length === 0) return 0\n  const index = Math.min(\n    sortedValues.length - 1,\n    Math.max(0, Math.ceil(p * sortedValues.length) - 1),\n  )\n  return sortedValues[index]!\n}\n\nfunction mean(values: readonly number[]): number {\n  const finite = values.filter(Number.isFinite)\n  if (finite.length === 0) return 0\n  return finite.reduce((sum, value) => sum + value, 0) / finite.length\n}\n\nfunction clamp01(value: number): number {\n  if (!Number.isFinite(value)) return 0\n  if (value < 0) return 0\n  if (value > 1) return 1\n  return value\n}\n\nfunction fmt(value: number): string {\n  if (!Number.isFinite(value)) return '0'\n  return value.toFixed(value === 0 || Math.abs(value) >= 10 ? 0 : 3)\n}\n","import type { RunSplitTag } from '../run-record'\nimport type {\n  BenchmarkAdapter,\n  BenchmarkDatasetItem,\n  BenchmarkFamily,\n  BenchmarkSource,\n  BenchmarkTaskKind,\n} from './types'\nimport { deterministicSplit } from './types'\n\nexport interface StandardRetrievalDocument {\n  id: string\n  title?: string\n  text: string\n  metadata?: Record<string, unknown>\n}\n\nexport interface StandardRetrievalQuery {\n  id: string\n  text: string\n  metadata?: Record<string, unknown>\n}\n\nexport interface StandardRetrievalQrel {\n  queryId: string\n  documentId: string\n  score: number\n}\n\nexport interface StandardRetrievalPayload {\n  queryId: string\n  query: string\n  expectedDocumentIds: string[]\n  expectedScores: Record<string, number>\n  corpus?: Record<string, StandardRetrievalDocument>\n  metadata?: Record<string, unknown>\n}\n\nexport interface BuildStandardRetrievalItemsOptions {\n  benchmarkId: string\n  family: BenchmarkFamily | string\n  queries: readonly StandardRetrievalQuery[]\n  qrels: readonly StandardRetrievalQrel[]\n  corpus?: readonly StandardRetrievalDocument[]\n  includeCorpusInPayload?: boolean\n  source?: BenchmarkSource\n  tags?: readonly string[]\n  splitOf?: (queryId: string) => RunSplitTag\n}\n\nexport interface RetrievalIdAdapterOptions extends BuildStandardRetrievalItemsOptions {\n  responseIdPattern?: RegExp\n  cutoffs?: readonly number[]\n  primaryMetric?: string\n  passMetric?: string\n  passThreshold?: number\n}\n\nexport interface StandardRetrievalResult {\n  id?: string\n  documentId?: string\n  docId?: string\n  score?: number\n}\n\nexport type StandardRetrievalArtifact =\n  | string\n  | readonly string[]\n  | readonly StandardRetrievalResult[]\n  | {\n      ids?: readonly string[]\n      documentIds?: readonly string[]\n      results?: readonly StandardRetrievalResult[]\n    }\n\nexport interface StandardRetrievalEvaluationOptions {\n  responseIdPattern?: RegExp\n  cutoffs?: readonly number[]\n  primaryMetric?: string\n  passMetric?: string\n  passThreshold?: number\n}\n\nexport function parseJsonlRows<T = unknown>(text: string): T[] {\n  return text\n    .split(/\\r?\\n/)\n    .map((line) => line.trim())\n    .filter(Boolean)\n    .map((line, index) => {\n      try {\n        return JSON.parse(line) as T\n      } catch (error) {\n        throw new Error(`invalid JSONL row ${index + 1}: ${(error as Error).message}`)\n      }\n    })\n}\n\nfunction parseTsvRows(text: string): string[][] {\n  return text\n    .split(/\\r?\\n/)\n    .map((line) => line.trim())\n    .filter(Boolean)\n    .filter((line) => !line.startsWith('#'))\n    .map((line) => line.split(/\\t|\\s+/))\n}\n\nexport function parseQrels(text: string): StandardRetrievalQrel[] {\n  return parseTsvRows(text).flatMap((parts, index) => {\n    if (parts.length < 3) return []\n    const [queryId, maybeZeroOrDocId, maybeDocIdOrScore, maybeScore] = parts\n    if (!queryId || !maybeZeroOrDocId || !maybeDocIdOrScore) return []\n    if (queryId.toLowerCase() === 'query-id' || queryId.toLowerCase() === 'qid') return []\n    const documentId = maybeScore === undefined ? maybeZeroOrDocId : maybeDocIdOrScore\n    const scoreText = maybeScore === undefined ? maybeDocIdOrScore : maybeScore\n    const score = Number(scoreText)\n    if (!documentId || !Number.isFinite(score)) {\n      throw new Error(`invalid qrels row ${index + 1}: expected query id, doc id, score`)\n    }\n    return [{ queryId, documentId, score }]\n  })\n}\n\nexport function parseBeirCorpusJsonl(text: string): StandardRetrievalDocument[] {\n  return parseJsonlRows<Record<string, unknown>>(text).map((row, index) => {\n    const id = stringField(row, '_id') ?? stringField(row, 'id')\n    const body = stringField(row, 'text') ?? stringField(row, 'contents')\n    if (!id || body === undefined) {\n      throw new Error(`invalid BEIR corpus row ${index + 1}: expected _id/id and text/contents`)\n    }\n    return {\n      id,\n      title: stringField(row, 'title'),\n      text: body,\n      metadata: stripKnown(row, ['_id', 'id', 'title', 'text', 'contents']),\n    }\n  })\n}\n\nexport function parseBeirQueriesJsonl(text: string): StandardRetrievalQuery[] {\n  return parseJsonlRows<Record<string, unknown>>(text).map((row, index) => {\n    const id = stringField(row, '_id') ?? stringField(row, 'id') ?? stringField(row, 'query_id')\n    const query = stringField(row, 'text') ?? stringField(row, 'query')\n    if (!id || !query) {\n      throw new Error(`invalid BEIR query row ${index + 1}: expected _id/id and text/query`)\n    }\n    return {\n      id,\n      text: query,\n      metadata: stripKnown(row, ['_id', 'id', 'query_id', 'text', 'query']),\n    }\n  })\n}\n\nexport function buildStandardRetrievalItems(\n  options: BuildStandardRetrievalItemsOptions,\n): Array<BenchmarkDatasetItem<StandardRetrievalPayload>> {\n  const qrelsByQuery = new Map<string, StandardRetrievalQrel[]>()\n  for (const qrel of options.qrels) {\n    if (qrel.score <= 0) continue\n    const list = qrelsByQuery.get(qrel.queryId) ?? []\n    list.push(qrel)\n    qrelsByQuery.set(qrel.queryId, list)\n  }\n  const corpus =\n    options.includeCorpusInPayload && options.corpus\n      ? Object.fromEntries(options.corpus.map((document) => [document.id, document]))\n      : undefined\n  return options.queries.flatMap((query) => {\n    const qrels = qrelsByQuery.get(query.id) ?? []\n    if (qrels.length === 0) return []\n    const split =\n      options.splitOf?.(query.id) ?? deterministicSplit(`${options.benchmarkId}:${query.id}`)\n    return [\n      {\n        id: query.id,\n        split,\n        family: options.family,\n        taskKind: 'retrieval',\n        tags: [...new Set([...(options.tags ?? []), split])],\n        ...(options.source ? { source: options.source } : {}),\n        ...(query.metadata ? { metadata: query.metadata } : {}),\n        payload: {\n          queryId: query.id,\n          query: query.text,\n          expectedDocumentIds: qrels.map((qrel) => qrel.documentId),\n          expectedScores: Object.fromEntries(qrels.map((qrel) => [qrel.documentId, qrel.score])),\n          ...(corpus ? { corpus } : {}),\n          ...(query.metadata ? { metadata: query.metadata } : {}),\n        },\n      },\n    ]\n  })\n}\n\nexport function createRetrievalIdBenchmarkAdapter(\n  options: RetrievalIdAdapterOptions,\n): BenchmarkAdapter<\n  BenchmarkDatasetItem<StandardRetrievalPayload>,\n  StandardRetrievalPayload,\n  StandardRetrievalArtifact\n> {\n  const items = buildStandardRetrievalItems(options)\n  return {\n    id: options.benchmarkId,\n    family: options.family,\n    taskKind: 'retrieval' satisfies BenchmarkTaskKind,\n    source: options.source,\n    defaultMetric: options.primaryMetric ?? 'ndcg@10',\n    async loadDataset(split) {\n      return items.filter((item) => item.split === split)\n    },\n    async evaluate(item, artifact) {\n      return evaluateStandardRetrieval(item.payload, artifact, options)\n    },\n    assignSplit(itemId) {\n      return options.splitOf?.(itemId) ?? deterministicSplit(`${options.benchmarkId}:${itemId}`)\n    },\n  }\n}\n\nexport function evaluateStandardRetrieval(\n  payload: StandardRetrievalPayload,\n  artifact: StandardRetrievalArtifact,\n  options: StandardRetrievalEvaluationOptions = {},\n) {\n  const rankedDocumentIds = normalizeRetrievedDocumentIds(\n    artifact,\n    options.responseIdPattern ?? /[A-Za-z0-9_.:/-]+/g,\n  )\n  const cutoffs = normalizeCutoffs(options.cutoffs ?? [1, 3, 5, 10])\n  const dimensions: Record<string, number> = {\n    expected_count: payload.expectedDocumentIds.length,\n    returned_count: rankedDocumentIds.length,\n  }\n  for (const cutoff of cutoffs) {\n    Object.assign(\n      dimensions,\n      retrievalMetricsAtCutoff({\n        rankedDocumentIds,\n        expectedScores: payload.expectedScores,\n        cutoff,\n      }),\n    )\n  }\n  const primaryMetric = options.primaryMetric ?? 'ndcg@10'\n  const passMetric = options.passMetric ?? 'hit@10'\n  const score = dimensions[primaryMetric] ?? dimensions[`ndcg@${cutoffs[cutoffs.length - 1]}`] ?? 0\n  const passScore =\n    dimensions[passMetric] ?? dimensions[`hit@${cutoffs[cutoffs.length - 1]}`] ?? score\n  return {\n    score,\n    passed: passScore >= (options.passThreshold ?? 1),\n    dimensions,\n    raw: {\n      rankedDocumentIds,\n      expectedDocumentIds: payload.expectedDocumentIds,\n      expectedScores: payload.expectedScores,\n    },\n  }\n}\n\nfunction normalizeRetrievedDocumentIds(\n  artifact: StandardRetrievalArtifact,\n  responseIdPattern: RegExp = /[A-Za-z0-9_.:/-]+/g,\n): string[] {\n  if (typeof artifact === 'string') {\n    return uniqueOrdered((artifact.match(responseIdPattern) ?? []).map((value) => value.trim()))\n  }\n  if (Array.isArray(artifact)) {\n    const entries = artifact as readonly (string | StandardRetrievalResult)[]\n    return uniqueOrdered(\n      entries.flatMap((entry) => {\n        if (typeof entry === 'string') return [entry]\n        return [entry.documentId ?? entry.docId ?? entry.id ?? '']\n      }),\n    )\n  }\n  const objectArtifact = artifact as Exclude<StandardRetrievalArtifact, string | readonly unknown[]>\n  return uniqueOrdered(\n    [\n      ...(objectArtifact.documentIds ?? []),\n      ...(objectArtifact.ids ?? []),\n      ...(objectArtifact.results ?? []).map(\n        (entry) => entry.documentId ?? entry.docId ?? entry.id ?? '',\n      ),\n    ].map((value) => value.trim()),\n  )\n}\n\nfunction retrievalMetricsAtCutoff(input: {\n  rankedDocumentIds: readonly string[]\n  expectedScores: Record<string, number>\n  cutoff: number\n}): Record<string, number> {\n  const top = input.rankedDocumentIds.slice(0, input.cutoff)\n  const relevantIds = Object.keys(input.expectedScores).filter(\n    (id) => input.expectedScores[id]! > 0,\n  )\n  const relevant = new Set(relevantIds)\n  const hits = top.filter((id) => relevant.has(id)).length\n  const firstRelevantRank = top.findIndex((id) => relevant.has(id))\n  const prefix = `@${input.cutoff}`\n  return {\n    [`hit${prefix}`]: hits > 0 ? 1 : 0,\n    [`recall${prefix}`]: relevant.size === 0 ? 1 : hits / relevant.size,\n    [`precision${prefix}`]: hits / input.cutoff,\n    [`mrr${prefix}`]: firstRelevantRank === -1 ? 0 : 1 / (firstRelevantRank + 1),\n    [`ndcg${prefix}`]: ndcgAt(input.rankedDocumentIds, input.expectedScores, input.cutoff),\n  }\n}\n\nfunction stringField(row: Record<string, unknown>, key: string): string | undefined {\n  const value = row[key]\n  return typeof value === 'string' ? value : undefined\n}\n\nfunction stripKnown(\n  row: Record<string, unknown>,\n  keys: readonly string[],\n): Record<string, unknown> {\n  const known = new Set(keys)\n  return Object.fromEntries(Object.entries(row).filter(([key]) => !known.has(key)))\n}\n\nfunction normalizeCutoffs(cutoffs: readonly number[]): number[] {\n  return [\n    ...new Set(cutoffs.map((cutoff) => Math.trunc(cutoff)).filter((cutoff) => cutoff > 0)),\n  ].sort((a, b) => a - b)\n}\n\nfunction uniqueOrdered(values: readonly string[]): string[] {\n  const seen = new Set<string>()\n  const out: string[] = []\n  for (const value of values) {\n    const trimmed = value.trim()\n    if (!trimmed || seen.has(trimmed)) continue\n    seen.add(trimmed)\n    out.push(trimmed)\n  }\n  return out\n}\n\nfunction ndcgAt(\n  rankedDocumentIds: readonly string[],\n  expectedScores: Record<string, number>,\n  cutoff: number,\n): number {\n  const dcg = rankedDocumentIds\n    .slice(0, cutoff)\n    .reduce(\n      (sum, documentId, index) => sum + discountedGain(expectedScores[documentId] ?? 0, index),\n      0,\n    )\n  const ideal = Object.values(expectedScores)\n    .filter((score) => score > 0)\n    .sort((a, b) => b - a)\n    .slice(0, cutoff)\n    .reduce((sum, score, index) => sum + discountedGain(score, index), 0)\n  return ideal === 0 ? 1 : dcg / ideal\n}\n\nfunction discountedGain(relevance: number, zeroBasedRank: number): number {\n  if (relevance <= 0) return 0\n  return (2 ** relevance - 1) / Math.log2(zeroBasedRank + 2)\n}\n"],"mappings":";;;;;;;;AAsBA,eAAsB,yBACpB,SAC2C;CAC3C,MAAM,OAAO,MAAM,QAAQ,QAAQ,SAAS,QAAQ,MAAM,QAAQ,YAAY;CAC9E,MAAM,SAAS,MAAM,QAAQ,QAAQ,SAAS,QAAQ,MAAM,QAAQ,cAAc;CAClF,MAAM,YAAYA,UAAQ,KAAK,KAAK;CACpC,MAAM,cAAcA,UAAQ,OAAO,KAAK;CACxC,MAAM,eAAe,QAAQ,gBAAgB;CAC7C,MAAM,iBAAiB,QAAQ,kBAAkB;CACjD,MAAM,SAAS,QAAQ,UAAU;CACjC,MAAM,MAAM,cAAc;CAC1B,MAAM,UAAU;EACd,aAAa,eACT,KAAA,IACA,cAAc,UAAU,QAAQ,CAAC,EAAE,eAAe,aAAa,QAAQ,CAAC;EAC5E,eAAe,iBACX,KAAA,IACA,gBAAgB,YAAY,QAAQ,CAAC,EAAE,aAAa,eAAe,QAAQ,CAAC;EAChF,OAAO,SAAS,KAAA,IAAY,OAAO,IAAI,QAAQ,CAAC,EAAE,aAAa,OAAO,QAAQ,CAAC;CACjF,CAAC,CAAC,QAAQ,WAA6B,QAAQ,MAAM,CAAC;CAEtD,OAAO;EACL,QAAQ,QAAQ,WAAW;EAC3B;EACA;EACA;EACA;EACA;EACA;CACF;AACF;AAEA,SAASA,UAAQ,OAAuB;CACtC,IAAI,CAAC,OAAO,SAAS,KAAK,GAAG,OAAO;CACpC,IAAI,QAAQ,GAAG,OAAO;CACtB,IAAI,QAAQ,GAAG,OAAO;CACtB,OAAO;AACT;;;;;;;;ACyEA,SAAS,QAAQ,OAAuB;CACtC,IAAI,IAAI;CACR,KAAK,IAAI,IAAI,GAAG,IAAI,MAAM,QAAQ,KAAK;EACrC,KAAK,MAAM,WAAW,CAAC,IAAI;EAC3B,IAAK,MAAM,KAAK,MAAM,KAAK,MAAM,KAAK,MAAM,KAAK,MAAM,KAAK,SAAU;CACxE;CACA,OAAO,MAAM;AACf;;;AAIA,MAAa,uBAAuB;;;;;;;;;AAUpC,SAAgB,mBACd,QACA,OAAe,sBACF;CAEb,MAAM,MADI,QAAQ,GAAG,KAAK,IAAI,QAClB,IAAI;CAChB,IAAI,MAAM,IAAK,OAAO;CACtB,IAAI,MAAM,IAAK,OAAO;CACtB,OAAO;AACT;;;;;;;;;;;;;;;;ACzEA,MAAM,SAAS,UACb,UAAU,QAAQ,OAAO,UAAU,YAAY,CAAC,MAAM,QAAQ,KAAK,IAC9D,QACD,CAAC;AACP,MAAM,gBAAgB,UACpB,OAAO,UAAU,YAAY,OAAO,SAAS,KAAK,IAAI,QAAQ;AAChE,MAAM,cAAc,UAClB,OAAO,UAAU,YAAY,MAAM,SAAS,IAAI,QAAQ;;AAG1D,eAAsB,eAAe,SAA6C;CAChF,MAAM,MAAM,QAAQ,OAAO;CAC3B,MAAM,SAASC,MAAU,MAAM,SAAS,KAAK,KAAK,WAAW,GAAG,MAAM,CAAC;CAIvE,MAAM,cAAc,MAAM,SAAS,KAAK,KAAK,gBAAgB,GAAG,MAAM;CACtE,MAAM,OAAO,MAAM,OAAO,IAAI;CAC9B,MAAM,QAAQ,MAAM,OAAO,KAAK;CAChC,MAAM,WAAW,MAAM,OAAO,QAAQ;CACtC,MAAM,cAAc,MAAM,OAAO,WAAW;CAC5C,MAAM,WAAW,MAAM,OAAO,QAAQ;CACtC,MAAM,aAAa,MAAM,QAAQ,OAAO,SAAS,IAAI,OAAO,YAAY,CAAC,EAAA,CAAG,KACzE,OAAO,UAAU;EAChB,MAAM,SAAS,OAAO,UAAU,WAAW,QAAQ,MAAM,KAAK,CAAC,CAAC;EAChE,IAAI,OAAO,WAAW,YAAY,OAAO,WAAW,GAClD,MAAM,IAAI,MAAM,GAAG,IAAI,wBAAwB,MAAM,uBAAuB;EAE9E,OAAO;CACT,CACF;CACA,MAAM,mBAAmB,MAAM,QAAQ,KAAK,KAAK,aAAa,CAAC,CAAC,CAAC,YAAY,CAAC,CAAa;CAC3F,MAAM,OAAO,IAAI,MAAM,GAAG,CAAC,CAAC,IAAI,KAAK;CACrC,OAAO;EACL,MAAM,WAAW,KAAK,IAAI,KAAK;EAC/B,KAAK;EACL,SAAS;EACT;EACA,iBAAiB,aAAa,MAAM,WAAW;EAC/C,oBAAoB,aAAa,SAAS,WAAW;EACrD,cAAc,WAAW,SAAS,gBAAgB;EAClD,WAAW;GACT,MAAM,aAAa,YAAY,IAAI;GACnC,UAAU,aAAa,YAAY,SAAS;GAC5C,WAAW,aAAa,YAAY,UAAU;GAC9C,MAAM,aAAa,YAAY,IAAI,KAAK;EAC1C;EACA;EACA,gBAAgB,iBAAiB,MAAM,SAAS,mBAAmB,KAAK,IAAI,CAAC;EAC7E,UAAU,WAAW,SAAS,QAAQ;EACtC,yBAAyB,aAAa,SAAS,0BAA0B;CAC3E;AACF;;AAGA,SAAgB,6BACd,SACmC;CACnC,MAAM,QACJ,QAAQ,iBAAiB,WAAmB,mBAAmB,GAAG,QAAQ,GAAG,IAAI,QAAQ;CAC3F,MAAM,WAAW,KAAK,QAAQ,QAAQ,UAAU,GAAG,OAAO;CAE1D,eAAe,UAAqC;EAClD,MAAM,QACJ,QAAQ,UAEN,MAAM,QAAQ,KAEV,MAAM,QAAQ,QAAQ,EAAA,CACtB,IAAI,OAAO,UACV,MAAM,KAAK,KAAK,UAAU,OAAO,WAAW,CAAC,CAAC,CAAC,YAAY,IAAI,MAAO,OACnE,OACA,KACN,CACF,EAAA,CAEC,QAAQ,UAA2B,UAAU,IAAI,CAAC,CAClD,KAAK;EACV,OAAO,QAAQ,IACb,MAAM,IAAI,OAAO,QAAQ;GACvB,MAAM,UAAU,MAAM,eAAe,KAAK,UAAU,GAAG,CAAC;GACxD,OAAO;IACL,IAAI;IACJ;IACA,OAAO,MAAM,GAAG;IAChB,QAAQ,QAAQ,OAAO,QAAQ;IAC/B,UAAU;IACV,MAAM,QAAQ,aAAa,OAAO,CAAC,IAAI,CAAC,QAAQ,QAAQ;IACxD,QAAQ,QAAQ;GAClB;EACF,CAAC,CACH;CACF;CAEA,OAAO;EACL,IAAI,QAAQ;EACZ,QAAQ,QAAQ,OAAO,QAAQ;EAC/B,UAAU;EACV,aACE,GAAG,QAAQ,OAAO,QAAQ,iBAAiB,GAAG,QAAQ,OAAO,WAAW,GAAG,MAAM,QAAQ,OAAO,SAAS,8BAA8B,QACrI,SACA,GACF;EACF,QAAQ,QAAQ;EAChB,eAAe;EACf;EACA,MAAM,YAAY,KAAkB;GAClC,QAAQ,MAAM,QAAQ,EAAA,CAAG,QAAQ,SAAS,KAAK,UAAU,GAAG;EAC9D;EACA,MAAM,SAAS,MAAsB,UAAmD;GACtF,MAAM,UAAU,MAAM,QAAQ,MAAM,MAAM,QAAQ;GAClD,IAAI,CAAC,QAAQ,WACX,MAAM,IAAI,MACR,GAAG,QAAQ,GAAG,GAAG,KAAK,GAAG,8CAA8C,QAAQ,OACjF;GACF,MAAM,EAAE,QAAQ,WAAW,QAAQ;GACnC,IAAI,EAAE,OAAO,SAAS,MAAM,KAAK,UAAU,KAAK,UAAU,IACxD,MAAM,IAAI,MACR,GAAG,QAAQ,GAAG,GAAG,KAAK,GAAG,iCAAiC,OAAO,iBACnE;GAEF,OAAO;IACL,OAAO;IACP,QAAQ,WAAW;IACnB,KAAK;KAAE;KAAQ,GAAI,WAAW,KAAA,IAAY,CAAC,IAAI,EAAE,OAAO;IAAG;GAC7D;EACF;EACA,aAAa;CACf;AACF;;;ACxLA,MAAa,kBAAiC;CAC5C;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,oBAAoB,YAAY;EAC3C,eAAe,CAAC,WAAW,YAAY;CACzC;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,mBAAmB,WAAW;EACzC,eAAe,CAAC,YAAY,YAAY;CAC1C;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,mBAAmB,YAAY;EAC1C,eAAe,CAAC,WAAW,aAAa;CAC1C;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,qBAAqB,aAAa;EAC7C,eAAe,CAAC,YAAY,SAAS;CACvC;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,uBAAuB,YAAY;EAC9C,eAAe,CAAC,mBAAmB,YAAY;CACjD;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,yBAAyB,YAAY;EAChD,eAAe,CAAC,iBAAiB,cAAc;CACjD;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,sBAAsB,YAAY;EAC7C,eAAe,CAAC,mBAAmB,eAAe;CACpD;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,qBAAqB,YAAY;EAC5C,eAAe,CAAC,gBAAgB,YAAY;CAC9C;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,cAAc,eAAe;EACxC,eAAe,CAAC,eAAe,YAAY;CAC7C;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,eAAe,MAAM;EAChC,eAAe,CAAC,cAAc,SAAS;CACzC;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,cAAc,iBAAiB;EAC1C,eAAe,CAAC,eAAe,YAAY;CAC7C;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,eAAe,MAAM;EAChC,eAAe,CAAC,cAAc,SAAS;CACzC;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,oBAAoB;EAC/B,eAAe,CAAC,cAAc,SAAS;CACzC;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,sBAAsB,WAAW;EAC5C,eAAe,CAAC,cAAc,cAAc;CAC9C;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,oBAAoB;EAC/B,eAAe,CAAC,cAAc,SAAS;CACzC;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,sBAAsB,OAAO;EACxC,eAAe,CAAC,YAAY,YAAY;CAC1C;AACF;;;;;;;;;;;AC9IA,IAAM,iBAAN,MAAqF;CACnF,KAAc;CACd,SAAkB;CAClB,WAAoB;CACpB,cAAuB;CACvB,gBAAyB;CAEzB,MAAM,YAAY,OAAmD;EACnE,OAAO,gBAAgB,KAAK,UAAU;GAAE,IAAI,KAAK;GAAI,SAAS;EAAK,EAAE,CAAC,CAAC,QACpE,OAAO,gBAAgB,GAAG,EAAE,MAAM,KACrC;CACF;CAEA,MAAM,SAAS,MAA0B,UAAgD;EACvF,MAAM,SAAS,mBAAmB,QAAQ;EAC1C,MAAM,UAAU,IAAI,IAClB,CAAC,KAAK,QAAQ,OAAO,GAAG,KAAK,QAAQ,QAAQ,CAAC,CAAC,KAAK,MAAM,EAAE,YAAY,CAAC,CAC3E;EACA,MAAM,UAAU,IAAI,IAAY,KAAK,QAAQ,cAAc,KAAK,MAAM,EAAE,YAAY,CAAC,CAAC;EACtF,MAAM,aAAa,OAAO,MAAM,MAAM,QAAQ,IAAI,EAAE,YAAY,CAAC,CAAC,KAAK;EACvE,MAAM,eAAe,OAAO,MAAM,MAAM,QAAQ,IAAI,EAAE,YAAY,CAAC,CAAC,KAAK;EAEzE,OAAO;GACL,OAFY,aAAa,IAAI;GAG7B,KAAK;IACH,YAAY,OAAO,MAAM;IACzB,cAAc;IACd,iBAAiB,QAAQ,YAAY;IACrC,mBAAmB;IACnB,UAAU,KAAK,QAAQ;GACzB;EACF;CACF;CAEA,YAAY,QAA6B;EACvC,OAAO,gBAAgB,MAAM;CAC/B;AACF;AAEA,SAAS,gBAAgB,QAA6B;CACpD,OAAO,mBAAmB,YAAY,QAAQ;AAChD;;;;;;;AAQA,SAAgB,mBAAmB,UAA4B;CAE7D,OADgB,SAAS,MAAM,oCAClB,KAAK,CAAC;AACrB;AAEA,MAAM,UAAU,IAAI,eAAe;AAEnC,MAAa,cAAc,QAAQ,YAAY,KAAK,OAAO;AAC3D,MAAa,WAAW,QAAQ,SAAS,KAAK,OAAO;AACrD,MAAa,cAAc,QAAQ,YAAY,KAAK,OAAO;;;ACS3D,eAAsB,oBACpB,SACkD;CAClD,MAAM,UAAU,QAAQ,WAAW,kBAAkB;CACrD,MAAM,cAAc,eAAe,QAAQ,OAAO;CAClD,MAAM,YAAY,MAAM,uBAAuB,QAAQ,SAAS,QAAQ,MAAM;CAC9E,MAAM,QAAQ,sBAAsB,QAAQ,OAAO;CACnD,MAAM,WAA+D,OACnE,UACA,YACG;EACH,OAAO,QAAQ,QAAQ;GAAE;GAAU,MAAM,SAAS;GAAM;EAAQ,CAAC;CACnE;CAEA,MAAM,WAAW,MAAM,YAAoD;EACzE;EACA;EACA,aAAa,aAAa;EAC1B,QAAQ,CAAC,KAAK;EACd,MAAM,QAAQ;EACd,MAAM,QAAQ;EACd,WAAW,QAAQ;EACnB,aAAa,QAAQ;EACrB,gBAAgB,QAAQ;EACxB,mBAAmB,QAAQ;EAC3B,aAAa,QAAQ,eAAe;EACpC,QAAQ,QAAQ;EAChB,MAAM,QAAQ;EACd;EACA,KAAK,QAAQ;CACf,CAAC;CACD,MAAM,SAAS,2BAA2B;EACxC,SAAS,QAAQ;EACjB;EACA;CACF,CAAC;CACD,QAAQ,UAAU,SAAS,MAAM;CACjC,MAAM,iBAAiB,KAAK,SAAS,QAAQ,uBAAuB;CACpE,MAAM,qBAAqB,KAAK,SAAS,QAAQ,qBAAqB;CACtE,QAAQ,MAAM,gBAAgB,GAAG,KAAK,UAAU,QAAQ,MAAM,CAAC,EAAE,GAAG;CACpE,QAAQ,MAAM,oBAAoB,8BAA8B,MAAM,CAAC;CACvE,OAAO;EAAE;EAAW;EAAU;EAAQ;EAAgB;CAAmB;AAC3E;AAEA,eAAsB,uBACpB,SACA,SAAiC;CAAC;CAAU;CAAO;AAAS,GACf;CAC7C,MAAM,cAAc,eAAe,OAAO;CAC1C,MAAM,SAAS,QAAQ,UAAU;CACjC,MAAM,WAAW,QAAQ,YAAY;CACrC,MAAM,MAA0C,CAAC;CACjD,MAAM,uBAAO,IAAI,IAAY;CAC7B,KAAK,MAAM,SAAS,QAAQ;EAC1B,MAAM,QAAQ,MAAM,QAAQ,YAAY,KAAK;EAC7C,KAAK,MAAM,QAAQ,OAAO;GACxB,MAAM,WAAW,KAAK,SAAS,QAAQ,YAAY,KAAK,EAAE;GAC1D,IAAI,aAAa,OAAO;GACxB,MAAM,MAAM,GAAG,YAAY,GAAG,KAAK;GACnC,IAAI,KAAK,IAAI,GAAG,GAAG;GACnB,KAAK,IAAI,GAAG;GACZ,IAAI,KAAK;IACP,IAAI;IACJ,MAAM;IACN;IACA,QAAQ,KAAK,UAAU;IACvB,UAAU,KAAK,YAAY;IAC3B;IACA,MAAM,CAAC,mBAAG,IAAI,IAAI,CAAC,UAAU,GAAI,KAAK,QAAQ,CAAC,CAAE,CAAC,CAAC;IACnD;GACF,CAAC;EACH;CACF;CACA,OAAO;AACT;AAEA,SAAgB,sBACd,SACqD;CACrD,MAAM,cAAc,eAAe,OAAO;CAC1C,OAAO;EACL,MAAM,GAAG,YAAY;EACrB,YAAY,CACV;GAAE,KAAK;GAAS,aAAa,WAAW,YAAY;EAAkB,GACtE;GAAE,KAAK;GAAU,aAAa;EAAiD,CACjF;EACA,YAAY,aAAgE;GAC1E,OAAO,SAAS,SAAS,eAAe,SAAS,GAAG,WAAW,GAAG,YAAY,EAAE;EAClF;EACA,MAAM,MAAM,EAAE,UAAU,YAAY;GAClC,MAAM,aAAa,MAAM,QAAQ,SAAS,SAAS,MAAM,QAAQ;GAEjE,OAAO;IACL,YAFiB,8BAA8B,UAEtC;IACT,WAAW,QAAQ,WAAW,KAAK;IACnC,OAAO,WAAW,SAAS;GAC7B;EACF;CACF;AACF;AAEA,SAAS,2BAAgD,OAIrC;CAClB,MAAM,eAAe,IAAI,IAAI,MAAM,UAAU,KAAK,aAAa,CAAC,SAAS,IAAI,QAAQ,CAAC,CAAC;CAavF,MAAM,aAZO,MAAM,SAAS,MAAM,KAAK,SAAS;EAC9C,MAAM,WAAW,aAAa,IAAI,KAAK,UAAU;EACjD,MAAM,QAAQ,gBAAgB,KAAK,WAAW;EAC9C,MAAM,QAAQ,OAAO,aAAa;EAClC,OAAO;GACL;GACA;GACA;GACA,SAAS,OAAO,WAAW,WAAW,QAAQ,IAAI,IAAI,OAAO;GAC7D,YAAY,OAAO,cAAc,CAAC;EACpC;CACF,CACsB,CAAC,CAAC,QAAQ,QAAQ,CAAC,IAAI,KAAK,KAAK;CACvD,MAAM,cAAc,WAAW,KAAK,QAAQ,IAAI,KAAK;CACrD,OAAO;EACL,aAAa,eAAe,MAAM,OAAO;EACzC,QAAQ,MAAM,QAAQ,UAAU;EAChC,UAAU,MAAM,QAAQ,YAAY;EACpC,GAAI,MAAM,QAAQ,SAAS,EAAE,QAAQ,MAAM,QAAQ,OAAO,IAAI,CAAC;EAC/D,QAAQ,MAAM,SAAS;EACvB,cAAc,MAAM,SAAS;EAC7B,MAAM,MAAM,SAAS;EACrB,WAAW,MAAM,SAAS;EAC1B,SAAS,MAAM,SAAS;EACxB,YAAY,MAAM,SAAS;EAC3B,YAAY,MAAM,UAAU;EAC5B,YAAY,MAAM,SAAS,MAAM;EACjC,aAAa,MAAM,SAAS,WAAW;EACvC,aAAa,MAAM,SAAS,WAAW;EACvC,cAAc,MAAM,SAAS,WAAW,KAAK;EAC7C,QAAQ,gBAAgB,aAAa,QAAQ,IAAI,UAAU,YAAY,WAAW;GAChF;GACA;GACA;EACF,CAAC;EACD,MAAM,gBAAgB,aAAa,QAAQ,IAAI,UAAU,QAAQ,CAAC,CAAC;EACnE,YAAY,oBAAoB,WAAW,KAAK,QAAQ,IAAI,UAAU,CAAC;EACvE,OAAO,aAAa,WAAW;EAC/B,SAAS,aAAa,WAAW,KAAK,QAAQ,IAAI,KAAK,OAAO,CAAC;EAC/D,WAAW,aAAa,WAAW,KAAK,QAAQ,IAAI,KAAK,UAAU,CAAC;CACtE;AACF;AAEA,SAAS,8BAA8B,QAAiC;CACtE,MAAM,YAAY,OAAO,QAAQ,OAAO,MAAM,CAAC,CAC5C,KAAK,CAAC,OAAO,aAAa;EACzB,OAAO,KAAK,MAAM,KAAK,QAAQ,EAAE,KAAK,IAAI,QAAQ,SAAS,EAAE,KAAK,IAAI,QAAQ,QAAQ,EAAE,KAAK,IAAI,QAAQ,MAAM,GAAG,EAAE,KAAK,IAAI,QAAQ,QAAQ,IAAI,EAAE,KAAK,IAAI,QAAQ,UAAU,GAAG,EAAE;CACrL,CAAC,CAAC,CACD,KAAK,IAAI;CACZ,MAAM,UAAU,OAAO,QAAQ,OAAO,UAAU,CAAC,CAC9C,MAAM,CAAC,IAAI,CAAC,OAAO,EAAE,cAAc,CAAC,CAAC,CAAC,CACtC,KAAK,CAAC,KAAK,UAAU,KAAK,IAAI,KAAK,KAAK,EAAE,KAAK,IAAI,KAAK,IAAI,EAAE,KAAK,IAAI,KAAK,GAAG,EAAE,GAAG,CAAC,CACrF,KAAK,IAAI;CACZ,OAAO;EACL,uBAAuB,OAAO;EAC9B;EACA,aAAa,OAAO;EACpB,gBAAgB,OAAO;EACvB,cAAc,OAAO;EACrB,eAAe,OAAO;EACtB,YAAY,OAAO,WAAW,UAAU,OAAO,YAAY,WAAW,OAAO,YAAY;EACzF,YAAY,IAAI,OAAO,YAAY;EACnC,iBAAiB,IAAI,OAAO,MAAM,IAAI,EAAE,WAAW,IAAI,OAAO,MAAM,MAAM,EAAE,QAAQ,IAAI,OAAO,MAAM,GAAG,EAAE,MAAM,OAAO,MAAM;EAC7H;EACA;EACA;EACA;EACA;EACA,aAAa;EACb;EACA;EACA;EACA;EACA;EACA,WAAW;EACX;CACF,CAAC,CAAC,KAAK,IAAI;AACb;AAEA,SAAS,eAAe,SAAuE;CAC7F,OAAO,QAAQ,MAAM,GAAG,QAAQ,UAAU,SAAS,GAAG,QAAQ,YAAY;AAC5E;AAEA,SAAS,8BAA8B,YAAyD;CAC9F,MAAM,aAAqC,EAAE,OAAO,QAAQ,WAAW,KAAK,EAAE;CAC9E,KAAK,MAAM,CAAC,KAAK,UAAU,OAAO,QAAQ,WAAW,cAAc,CAAC,CAAC,GACnE,IAAI,OAAO,SAAS,KAAK,GAAG,WAAW,OAAO;CAEhD,WAAW,SAAU,WAAW,UAAU,WAAW,QAAQ,IAAK,IAAI;CACtE,OAAO;AACT;AAEA,SAAS,oBACP,MACuC;CACvC,MAAM,yBAAS,IAAI,IAAsB;CACzC,KAAK,MAAM,OAAO,MAChB,KAAK,MAAM,CAAC,KAAK,UAAU,OAAO,QAAQ,GAAG,GAAG;EAC9C,IAAI,CAAC,OAAO,SAAS,KAAK,GAAG;EAC7B,MAAM,OAAO,OAAO,IAAI,GAAG,KAAK,CAAC;EACjC,KAAK,KAAK,KAAK;EACf,OAAO,IAAI,KAAK,IAAI;CACtB;CAEF,OAAO,OAAO,YAAY,CAAC,GAAG,OAAO,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,KAAK,UAAU,CAAC,KAAK,aAAa,IAAI,CAAC,CAAC,CAAC;AACjG;AAEA,SAAS,gBACP,MACA,OACA,YAA+B,CAAC,GACO;CACvC,MAAM,0BAAU,IAAI,IAAiB;CACrC,KAAK,MAAM,OAAO,WAAW,QAAQ,IAAI,KAAK,CAAC,CAAC;CAChD,KAAK,MAAM,OAAO,MAAM;EACtB,MAAM,OAAO,MAAM,GAAG;EACtB,KAAK,MAAM,OAAO,MAAM,QAAQ,IAAI,IAAI,OAAO,CAAC,IAAI,GAAG;GACrD,MAAM,OAAO,QAAQ,IAAI,GAAG,KAAK,CAAC;GAClC,KAAK,KAAK,GAAG;GACb,QAAQ,IAAI,KAAK,IAAI;EACvB;CACF;CACA,MAAM,MAA6C,CAAC;CACpD,KAAK,MAAM,CAAC,KAAK,SAAS,SAAS;EACjC,MAAM,YAAY;EAKlB,IAAI,OAAO;GACT,GAAG,KAAK;GACR,WAAW,KAAK,UAAU,KAAK,QAAQ,IAAI,KAAK,CAAC;GACjD,UAAU,KAAK,UAAU,KAAK,QAAS,IAAI,SAAS,IAAI,CAAE,CAAC;GAC3D,OAAO,aAAa,UAAU,KAAK,QAAQ,IAAI,KAAK,CAAC;GACrD,SAAS,aAAa,UAAU,KAAK,QAAQ,IAAI,KAAK,OAAO,CAAC;GAC9D,WAAW,aAAa,UAAU,KAAK,QAAQ,IAAI,KAAK,UAAU,CAAC;EACrE;CACF;CACA,OAAO;AACT;AAEA,SAAS,gBACP,aACwB;CACxB,OAAO,OAAO,OAAO,WAAW,CAAC,CAAC;AACpC;AAEA,SAAS,aAAa,QAAkD;CACtE,MAAM,SAAS,CAAC,GAAG,MAAM,CAAC,CAAC,OAAO,OAAO,QAAQ,CAAC,CAAC,MAAM,GAAG,MAAM,IAAI,CAAC;CACvE,IAAI,OAAO,WAAW,GAAG,OAAO;EAAE,GAAG;EAAG,KAAK;EAAG,MAAM;EAAG,QAAQ;EAAG,KAAK;EAAG,KAAK;CAAE;CACnF,OAAO;EACL,GAAG,OAAO;EACV,KAAK,OAAO;EACZ,MAAM,KAAK,MAAM;EACjB,QAAQ,WAAW,QAAQ,EAAG;EAC9B,KAAK,WAAW,QAAQ,EAAG;EAC3B,KAAK,OAAO,OAAO,SAAS;CAC9B;AACF;AAEA,SAAS,WAAW,cAAiC,GAAmB;CACtE,IAAI,aAAa,WAAW,GAAG,OAAO;CAKtC,OAAO,aAJO,KAAK,IACjB,aAAa,SAAS,GACtB,KAAK,IAAI,GAAG,KAAK,KAAK,IAAI,aAAa,MAAM,IAAI,CAAC,CAE5B;AAC1B;AAEA,SAAS,KAAK,QAAmC;CAC/C,MAAM,SAAS,OAAO,OAAO,OAAO,QAAQ;CAC5C,IAAI,OAAO,WAAW,GAAG,OAAO;CAChC,OAAO,OAAO,QAAQ,KAAK,UAAU,MAAM,OAAO,CAAC,IAAI,OAAO;AAChE;AAEA,SAAS,QAAQ,OAAuB;CACtC,IAAI,CAAC,OAAO,SAAS,KAAK,GAAG,OAAO;CACpC,IAAI,QAAQ,GAAG,OAAO;CACtB,IAAI,QAAQ,GAAG,OAAO;CACtB,OAAO;AACT;AAEA,SAAS,IAAI,OAAuB;CAClC,IAAI,CAAC,OAAO,SAAS,KAAK,GAAG,OAAO;CACpC,OAAO,MAAM,QAAQ,UAAU,KAAK,KAAK,IAAI,KAAK,KAAK,KAAK,IAAI,CAAC;AACnE;;;ACzSA,SAAgB,eAA4B,MAAmB;CAC7D,OAAO,KACJ,MAAM,OAAO,CAAC,CACd,KAAK,SAAS,KAAK,KAAK,CAAC,CAAC,CAC1B,OAAO,OAAO,CAAC,CACf,KAAK,MAAM,UAAU;EACpB,IAAI;GACF,OAAO,KAAK,MAAM,IAAI;EACxB,SAAS,OAAO;GACd,MAAM,IAAI,MAAM,qBAAqB,QAAQ,EAAE,IAAK,MAAgB,SAAS;EAC/E;CACF,CAAC;AACL;AAEA,SAAS,aAAa,MAA0B;CAC9C,OAAO,KACJ,MAAM,OAAO,CAAC,CACd,KAAK,SAAS,KAAK,KAAK,CAAC,CAAC,CAC1B,OAAO,OAAO,CAAC,CACf,QAAQ,SAAS,CAAC,KAAK,WAAW,GAAG,CAAC,CAAC,CACvC,KAAK,SAAS,KAAK,MAAM,QAAQ,CAAC;AACvC;AAEA,SAAgB,WAAW,MAAuC;CAChE,OAAO,aAAa,IAAI,CAAC,CAAC,SAAS,OAAO,UAAU;EAClD,IAAI,MAAM,SAAS,GAAG,OAAO,CAAC;EAC9B,MAAM,CAAC,SAAS,kBAAkB,mBAAmB,cAAc;EACnE,IAAI,CAAC,WAAW,CAAC,oBAAoB,CAAC,mBAAmB,OAAO,CAAC;EACjE,IAAI,QAAQ,YAAY,MAAM,cAAc,QAAQ,YAAY,MAAM,OAAO,OAAO,CAAC;EACrF,MAAM,aAAa,eAAe,KAAA,IAAY,mBAAmB;EAEjE,MAAM,QAAQ,OADI,eAAe,KAAA,IAAY,oBAAoB,UACnC;EAC9B,IAAI,CAAC,cAAc,CAAC,OAAO,SAAS,KAAK,GACvC,MAAM,IAAI,MAAM,qBAAqB,QAAQ,EAAE,mCAAmC;EAEpF,OAAO,CAAC;GAAE;GAAS;GAAY;EAAM,CAAC;CACxC,CAAC;AACH;AAEA,SAAgB,qBAAqB,MAA2C;CAC9E,OAAO,eAAwC,IAAI,CAAC,CAAC,KAAK,KAAK,UAAU;EACvE,MAAM,KAAK,YAAY,KAAK,KAAK,KAAK,YAAY,KAAK,IAAI;EAC3D,MAAM,OAAO,YAAY,KAAK,MAAM,KAAK,YAAY,KAAK,UAAU;EACpE,IAAI,CAAC,MAAM,SAAS,KAAA,GAClB,MAAM,IAAI,MAAM,2BAA2B,QAAQ,EAAE,oCAAoC;EAE3F,OAAO;GACL;GACA,OAAO,YAAY,KAAK,OAAO;GAC/B,MAAM;GACN,UAAU,WAAW,KAAK;IAAC;IAAO;IAAM;IAAS;IAAQ;GAAU,CAAC;EACtE;CACF,CAAC;AACH;AAEA,SAAgB,sBAAsB,MAAwC;CAC5E,OAAO,eAAwC,IAAI,CAAC,CAAC,KAAK,KAAK,UAAU;EACvE,MAAM,KAAK,YAAY,KAAK,KAAK,KAAK,YAAY,KAAK,IAAI,KAAK,YAAY,KAAK,UAAU;EAC3F,MAAM,QAAQ,YAAY,KAAK,MAAM,KAAK,YAAY,KAAK,OAAO;EAClE,IAAI,CAAC,MAAM,CAAC,OACV,MAAM,IAAI,MAAM,0BAA0B,QAAQ,EAAE,iCAAiC;EAEvF,OAAO;GACL;GACA,MAAM;GACN,UAAU,WAAW,KAAK;IAAC;IAAO;IAAM;IAAY;IAAQ;GAAO,CAAC;EACtE;CACF,CAAC;AACH;AAEA,SAAgB,4BACd,SACuD;CACvD,MAAM,+BAAe,IAAI,IAAqC;CAC9D,KAAK,MAAM,QAAQ,QAAQ,OAAO;EAChC,IAAI,KAAK,SAAS,GAAG;EACrB,MAAM,OAAO,aAAa,IAAI,KAAK,OAAO,KAAK,CAAC;EAChD,KAAK,KAAK,IAAI;EACd,aAAa,IAAI,KAAK,SAAS,IAAI;CACrC;CACA,MAAM,SACJ,QAAQ,0BAA0B,QAAQ,SACtC,OAAO,YAAY,QAAQ,OAAO,KAAK,aAAa,CAAC,SAAS,IAAI,QAAQ,CAAC,CAAC,IAC5E,KAAA;CACN,OAAO,QAAQ,QAAQ,SAAS,UAAU;EACxC,MAAM,QAAQ,aAAa,IAAI,MAAM,EAAE,KAAK,CAAC;EAC7C,IAAI,MAAM,WAAW,GAAG,OAAO,CAAC;EAChC,MAAM,QACJ,QAAQ,UAAU,MAAM,EAAE,KAAK,mBAAmB,GAAG,QAAQ,YAAY,GAAG,MAAM,IAAI;EACxF,OAAO,CACL;GACE,IAAI,MAAM;GACV;GACA,QAAQ,QAAQ;GAChB,UAAU;GACV,MAAM,CAAC,mBAAG,IAAI,IAAI,CAAC,GAAI,QAAQ,QAAQ,CAAC,GAAI,KAAK,CAAC,CAAC;GACnD,GAAI,QAAQ,SAAS,EAAE,QAAQ,QAAQ,OAAO,IAAI,CAAC;GACnD,GAAI,MAAM,WAAW,EAAE,UAAU,MAAM,SAAS,IAAI,CAAC;GACrD,SAAS;IACP,SAAS,MAAM;IACf,OAAO,MAAM;IACb,qBAAqB,MAAM,KAAK,SAAS,KAAK,UAAU;IACxD,gBAAgB,OAAO,YAAY,MAAM,KAAK,SAAS,CAAC,KAAK,YAAY,KAAK,KAAK,CAAC,CAAC;IACrF,GAAI,SAAS,EAAE,OAAO,IAAI,CAAC;IAC3B,GAAI,MAAM,WAAW,EAAE,UAAU,MAAM,SAAS,IAAI,CAAC;GACvD;EACF,CACF;CACF,CAAC;AACH;AAEA,SAAgB,kCACd,SAKA;CACA,MAAM,QAAQ,4BAA4B,OAAO;CACjD,OAAO;EACL,IAAI,QAAQ;EACZ,QAAQ,QAAQ;EAChB,UAAU;EACV,QAAQ,QAAQ;EAChB,eAAe,QAAQ,iBAAiB;EACxC,MAAM,YAAY,OAAO;GACvB,OAAO,MAAM,QAAQ,SAAS,KAAK,UAAU,KAAK;EACpD;EACA,MAAM,SAAS,MAAM,UAAU;GAC7B,OAAO,0BAA0B,KAAK,SAAS,UAAU,OAAO;EAClE;EACA,YAAY,QAAQ;GAClB,OAAO,QAAQ,UAAU,MAAM,KAAK,mBAAmB,GAAG,QAAQ,YAAY,GAAG,QAAQ;EAC3F;CACF;AACF;AAEA,SAAgB,0BACd,SACA,UACA,UAA8C,CAAC,GAC/C;CACA,MAAM,oBAAoB,8BACxB,UACA,QAAQ,qBAAqB,oBAC/B;CACA,MAAM,UAAU,iBAAiB,QAAQ,WAAW;EAAC;EAAG;EAAG;EAAG;CAAE,CAAC;CACjE,MAAM,aAAqC;EACzC,gBAAgB,QAAQ,oBAAoB;EAC5C,gBAAgB,kBAAkB;CACpC;CACA,KAAK,MAAM,UAAU,SACnB,OAAO,OACL,YACA,yBAAyB;EACvB;EACA,gBAAgB,QAAQ;EACxB;CACF,CAAC,CACH;CAEF,MAAM,gBAAgB,QAAQ,iBAAiB;CAC/C,MAAM,aAAa,QAAQ,cAAc;CACzC,MAAM,QAAQ,WAAW,kBAAkB,WAAW,QAAQ,QAAQ,QAAQ,SAAS,SAAS;CAGhG,OAAO;EACL;EACA,SAHA,WAAW,eAAe,WAAW,OAAO,QAAQ,QAAQ,SAAS,SAAS,WAGxD,QAAQ,iBAAiB;EAC/C;EACA,KAAK;GACH;GACA,qBAAqB,QAAQ;GAC7B,gBAAgB,QAAQ;EAC1B;CACF;AACF;AAEA,SAAS,8BACP,UACA,oBAA4B,sBAClB;CACV,IAAI,OAAO,aAAa,UACtB,OAAO,eAAe,SAAS,MAAM,iBAAiB,KAAK,CAAC,EAAA,CAAG,KAAK,UAAU,MAAM,KAAK,CAAC,CAAC;CAE7F,IAAI,MAAM,QAAQ,QAAQ,GAExB,OAAO,cACLC,SAAQ,SAAS,UAAU;EACzB,IAAI,OAAO,UAAU,UAAU,OAAO,CAAC,KAAK;EAC5C,OAAO,CAAC,MAAM,cAAc,MAAM,SAAS,MAAM,MAAM,EAAE;CAC3D,CAAC,CACH;CAEF,MAAM,iBAAiB;CACvB,OAAO,cACL;EACE,GAAI,eAAe,eAAe,CAAC;EACnC,GAAI,eAAe,OAAO,CAAC;EAC3B,IAAI,eAAe,WAAW,CAAC,EAAA,CAAG,KAC/B,UAAU,MAAM,cAAc,MAAM,SAAS,MAAM,MAAM,EAC5D;CACF,CAAC,CAAC,KAAK,UAAU,MAAM,KAAK,CAAC,CAC/B;AACF;AAEA,SAAS,yBAAyB,OAIP;CACzB,MAAM,MAAM,MAAM,kBAAkB,MAAM,GAAG,MAAM,MAAM;CACzD,MAAM,cAAc,OAAO,KAAK,MAAM,cAAc,CAAC,CAAC,QACnD,OAAO,MAAM,eAAe,MAAO,CACtC;CACA,MAAM,WAAW,IAAI,IAAI,WAAW;CACpC,MAAM,OAAO,IAAI,QAAQ,OAAO,SAAS,IAAI,EAAE,CAAC,CAAC,CAAC;CAClD,MAAM,oBAAoB,IAAI,WAAW,OAAO,SAAS,IAAI,EAAE,CAAC;CAChE,MAAM,SAAS,IAAI,MAAM;CACzB,OAAO;GACJ,MAAM,WAAW,OAAO,IAAI,IAAI;GAChC,SAAS,WAAW,SAAS,SAAS,IAAI,IAAI,OAAO,SAAS;GAC9D,YAAY,WAAW,OAAO,MAAM;GACpC,MAAM,WAAW,sBAAsB,KAAK,IAAI,KAAK,oBAAoB;GACzE,OAAO,WAAW,OAAO,MAAM,mBAAmB,MAAM,gBAAgB,MAAM,MAAM;CACvF;AACF;AAEA,SAAS,YAAY,KAA8B,KAAiC;CAClF,MAAM,QAAQ,IAAI;CAClB,OAAO,OAAO,UAAU,WAAW,QAAQ,KAAA;AAC7C;AAEA,SAAS,WACP,KACA,MACyB;CACzB,MAAM,QAAQ,IAAI,IAAI,IAAI;CAC1B,OAAO,OAAO,YAAY,OAAO,QAAQ,GAAG,CAAC,CAAC,QAAQ,CAAC,SAAS,CAAC,MAAM,IAAI,GAAG,CAAC,CAAC;AAClF;AAEA,SAAS,iBAAiB,SAAsC;CAC9D,OAAO,CACL,GAAG,IAAI,IAAI,QAAQ,KAAK,WAAW,KAAK,MAAM,MAAM,CAAC,CAAC,CAAC,QAAQ,WAAW,SAAS,CAAC,CAAC,CACvF,CAAC,CAAC,MAAM,GAAG,MAAM,IAAI,CAAC;AACxB;AAEA,SAAS,cAAc,QAAqC;CAC1D,MAAM,uBAAO,IAAI,IAAY;CAC7B,MAAM,MAAgB,CAAC;CACvB,KAAK,MAAM,SAAS,QAAQ;EAC1B,MAAM,UAAU,MAAM,KAAK;EAC3B,IAAI,CAAC,WAAW,KAAK,IAAI,OAAO,GAAG;EACnC,KAAK,IAAI,OAAO;EAChB,IAAI,KAAK,OAAO;CAClB;CACA,OAAO;AACT;AAEA,SAAS,OACP,mBACA,gBACA,QACQ;CACR,MAAM,MAAM,kBACT,MAAM,GAAG,MAAM,CAAC,CAChB,QACE,KAAK,YAAY,UAAU,MAAM,eAAe,eAAe,eAAe,GAAG,KAAK,GACvF,CACF;CACF,MAAM,QAAQ,OAAO,OAAO,cAAc,CAAC,CACxC,QAAQ,UAAU,QAAQ,CAAC,CAAC,CAC5B,MAAM,GAAG,MAAM,IAAI,CAAC,CAAC,CACrB,MAAM,GAAG,MAAM,CAAC,CAChB,QAAQ,KAAK,OAAO,UAAU,MAAM,eAAe,OAAO,KAAK,GAAG,CAAC;CACtE,OAAO,UAAU,IAAI,IAAI,MAAM;AACjC;AAEA,SAAS,eAAe,WAAmB,eAA+B;CACxE,IAAI,aAAa,GAAG,OAAO;CAC3B,QAAQ,KAAK,YAAY,KAAK,KAAK,KAAK,gBAAgB,CAAC;AAC3D"}

package/dist/cli.js

@@ -1,9 +1,9 @@
 #!/usr/bin/env node
 import { i as runRolloutReleaseCli } from "./hf-dataset-CRqtSjnR.js";
 import { t as createChatClient } from "./chat-client-DqPqWzVV.js";
-import { t as runAnalystBenchmarkCommand } from "./benchmark-command-NATyhXB4.js";
+import { t as runAnalystBenchmarkCommand } from "./benchmark-command-CERRH0E6.js";
 import { a as runRpcBatch, o as runRpcOnce, p as handleVersion, r as startServerAsync, s as buildOpenApi } from "./server-DmpcFiVK.js";
-import { t as runSupervisorRunCommand } from "./report-command-BvqaxlVG.js";
+import { t as runSupervisorRunCommand } from "./report-command-CSTjzFGY.js";
 import { writeFileSync } from "node:fs";
 //#region src/cli-config.ts
 /**

package/dist/contract/index.js

@@ -12,7 +12,7 @@ import { f as mapConcurrentRange } from "../journal-BuRLm6ho.js";
 import { C as inMemoryCampaignStorage, S as fsCampaignStorage } from "../external-optimizer-subprocess-DMgmqTB6.js";
 import { i as makeProposalFinding } from "../types-C8VBbiw6.js";
 import { n as SelfImproveRunError, r as selfImprove, t as defineAgentEval } from "../define-agent-eval-C0EqJ9Gz.js";
-import { t as buildDefaultAnalystRegistry } from "../default-registry-xiHrldW8.js";
+import { t as buildDefaultAnalystRegistry } from "../default-registry-Ca2IWH1X.js";
 import { t as extractUsage } from "../extract-usage-BrQ8mCLX.js";
 import { t as createChatClient } from "../chat-client-DqPqWzVV.js";
 import { n as InMemoryOutcomeStore, t as FileSystemOutcomeStore } from "../outcome-store-BXlkwMPR.js";

package/dist/{default-registry-xiHrldW8.js → default-registry-Ca2IWH1X.js}

@@ -6,7 +6,7 @@ import { i as validateUsageSettlementTimeout, t as assertValidAnalystUsageReceip
 import { LLM_CONTEXT_TOKENS, LLM_INPUT_TOKEN_ATTR_KEYS, LLM_OUTPUT_TOKEN_ATTR_KEYS, TOOL_NAME_ATTR_KEYS } from "./trace-attributes.js";
 import "./otlp-attributes-BpC_blLF.js";
 import { t as executionTrackByLane } from "./execution-tracks-CpgFPpS5.js";
-import { t as analyzeSupervisorRunIntegrity } from "./integrity-CgQce0FV.js";
+import { t as analyzeSupervisorRunIntegrity } from "./integrity-BvHANvnp.js";
 import { randomUUID } from "node:crypto";
 import { z } from "zod";
 //#region src/feedback-trajectory-review.ts
@@ -2365,4 +2365,4 @@ function buildDefaultAnalystRegistry(options = {}) {
 //#endregion
 export { completedAnalystReviewQuality as _, KNOWLEDGE_POISONING_KIND_SPEC as a, validateAnalystReviewDecisions as b, FAILURE_MODE_KIND_SPEC as c, emitControlIntegrityFindings as d, behavioralAnalyst as f, assertUniqueFindingIds as g, analystRunDigest as h, DEFAULT_TRACE_ANALYST_KINDS as i, CONTROL_INTEGRITY_ANALYST as l, analystFindingDigest as m, AnalystRegistry as n, KNOWLEDGE_GAP_KIND_SPEC as o, deriveEfficiencyFindings as p, ExactAnalystRunExecutionError as r, IMPROVEMENT_KIND_SPEC as s, buildDefaultAnalystRegistry as t, ControlIntegrityAnalyst as u, readAnalystReview as v, snapshotAnalystRun as y };
 
-//# sourceMappingURL=default-registry-xiHrldW8.js.map
+//# sourceMappingURL=default-registry-Ca2IWH1X.js.map