@tangle-network/agent-eval 0.130.1 → 0.131.0

package/docs/concepts.md

@@ -32,6 +32,17 @@ It contains score distributions, paired lift intervals, judge agreement, cost, f
 Run records, scenarios, judge scores, statistics, and release decisions belong here because they work without an agent runtime.
 Agent sessions, worker coordination, sandbox execution, and runtime-specific profiles belong in `agent-runtime`.
 
+## Measuring a complete profile
+
+Use the profile improvement functions from `/contract` when a host owns immutable agent-profile snapshots.
+`sealAgentProfileImprovementExperiment()` freezes the exact baseline, candidate diff, held-out tasks, model, limits, and policy.
+`runAgentProfileImprovementExperiment()` asks the host to execute every frozen baseline/candidate cell and requires one complete receipt per execution.
+`measuredComparisonFromAgentProfileImprovementExperiment()` recomputes scores, uncertainty, cost, latency, and the release decision from those receipts.
+
+This API never activates a candidate or runs an agent itself.
+The host owns authorization, billing, task isolation, profile materialization, execution, and durable evidence.
+The first portable profile contract accepts prompt and skill changes only; a host must add its own exact-state adapter before measuring tools, MCP servers, hooks, subagents, or external knowledge.
+
 ## Main Objects
 
 | Thing | What it is | One-line example |

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tangle-network/agent-eval",
-  "version": "0.130.1",
+  "version": "0.131.0",
   "description": "Evaluate and improve AI agents from runs, traces, judges, and feedback. Compare candidates, cluster failures, measure lift, and gate releases.",
   "homepage": "https://github.com/tangle-network/agent-eval#readme",
   "repository": {

package/dist/concurrency-DIxRZF_J.js

@@ -1,85 +0,0 @@
-//#region src/concurrency.ts
-/**
-* concurrency — small primitives the evolution loop needs.
-*
-* `Mutex` is a zero-dep async lock with FIFO fairness. The evolution loop
-* uses it to serialise checkout/build/commit sequences inside a single
-* pool slot, and to gate concurrent JSONL writers (see
-* `lockedJsonlReferenceReplayStore`).
-*
-* Deliberately minimal — no priority queue, no timeouts. If you need
-* those, swap to `async-mutex` at the call site.
-*/
-var Mutex = class {
-	locked = false;
-	waiters = [];
-	async acquire() {
-		if (!this.locked) {
-			this.locked = true;
-			return () => this.release();
-		}
-		return new Promise((resolve) => {
-			this.waiters.push(() => {
-				resolve(() => this.release());
-			});
-		});
-	}
-	release() {
-		const next = this.waiters.shift();
-		if (next) next();
-		else this.locked = false;
-	}
-	async runExclusive(fn) {
-		const release = await this.acquire();
-		try {
-			return await fn();
-		} finally {
-			release();
-		}
-	}
-	/** True iff someone holds the lock right now. Diagnostics only. */
-	get isLocked() {
-		return this.locked;
-	}
-	/** Pending waiter count. Diagnostics only. */
-	get pending() {
-		return this.waiters.length;
-	}
-};
-/**
-* Map independent work with a fixed worker count while preserving input order.
-* After the first rejection, no new items start; already-running work is allowed
-* to settle before the returned promise rejects. Partial results are discarded.
-*/
-async function mapConcurrent(items, concurrency, map) {
-	if (!Number.isInteger(concurrency) || concurrency < 1) throw new Error(`mapConcurrent: concurrency must be a positive integer, got ${concurrency}`);
-	if (items.length === 0) return [];
-	const results = new Array(items.length);
-	let nextIndex = 0;
-	let stopped = false;
-	let failed = false;
-	let failure;
-	const worker = async () => {
-		while (!stopped) {
-			const index = nextIndex;
-			nextIndex += 1;
-			if (index >= items.length) return;
-			try {
-				results[index] = await map(items[index], index);
-			} catch (error) {
-				stopped = true;
-				if (!failed) {
-					failed = true;
-					failure = error;
-				}
-			}
-		}
-	};
-	await Promise.all(Array.from({ length: Math.min(concurrency, items.length) }, () => worker()));
-	if (failed) throw failure;
-	return results;
-}
-//#endregion
-export { mapConcurrent as n, Mutex as t };
-
-//# sourceMappingURL=concurrency-DIxRZF_J.js.map

package/dist/concurrency-DIxRZF_J.js.map

@@ -1 +0,0 @@
-{"version":3,"file":"concurrency-DIxRZF_J.js","names":[],"sources":["../src/concurrency.ts"],"sourcesContent":["/**\n * concurrency — small primitives the evolution loop needs.\n *\n * `Mutex` is a zero-dep async lock with FIFO fairness. The evolution loop\n * uses it to serialise checkout/build/commit sequences inside a single\n * pool slot, and to gate concurrent JSONL writers (see\n * `lockedJsonlReferenceReplayStore`).\n *\n * Deliberately minimal — no priority queue, no timeouts. If you need\n * those, swap to `async-mutex` at the call site.\n */\n\nexport class Mutex {\n  private locked = false\n  private readonly waiters: Array<() => void> = []\n\n  async acquire(): Promise<() => void> {\n    if (!this.locked) {\n      this.locked = true\n      return () => this.release()\n    }\n    return new Promise<() => void>((resolve) => {\n      this.waiters.push(() => {\n        resolve(() => this.release())\n      })\n    })\n  }\n\n  private release(): void {\n    const next = this.waiters.shift()\n    if (next) {\n      next()\n    } else {\n      this.locked = false\n    }\n  }\n\n  async runExclusive<T>(fn: () => Promise<T> | T): Promise<T> {\n    const release = await this.acquire()\n    try {\n      return await fn()\n    } finally {\n      release()\n    }\n  }\n\n  /** True iff someone holds the lock right now. Diagnostics only. */\n  get isLocked(): boolean {\n    return this.locked\n  }\n\n  /** Pending waiter count. Diagnostics only. */\n  get pending(): number {\n    return this.waiters.length\n  }\n}\n\n/**\n * Map independent work with a fixed worker count while preserving input order.\n * After the first rejection, no new items start; already-running work is allowed\n * to settle before the returned promise rejects. Partial results are discarded.\n */\nexport async function mapConcurrent<T, R>(\n  items: readonly T[],\n  concurrency: number,\n  map: (item: T, index: number) => Promise<R>,\n): Promise<R[]> {\n  if (!Number.isInteger(concurrency) || concurrency < 1) {\n    throw new Error(`mapConcurrent: concurrency must be a positive integer, got ${concurrency}`)\n  }\n  if (items.length === 0) return []\n\n  const results = new Array<R>(items.length)\n  let nextIndex = 0\n  let stopped = false\n  let failed = false\n  let failure: unknown\n\n  const worker = async (): Promise<void> => {\n    while (!stopped) {\n      const index = nextIndex\n      nextIndex += 1\n      if (index >= items.length) return\n\n      try {\n        results[index] = await map(items[index]!, index)\n      } catch (error) {\n        stopped = true\n        if (!failed) {\n          failed = true\n          failure = error\n        }\n      }\n    }\n  }\n\n  await Promise.all(Array.from({ length: Math.min(concurrency, items.length) }, () => worker()))\n  if (failed) throw failure\n  return results\n}\n"],"mappings":";;;;;;;;;;;;AAYA,IAAa,QAAb,MAAmB;CACjB,SAAiB;CACjB,UAA8C,CAAC;CAE/C,MAAM,UAA+B;EACnC,IAAI,CAAC,KAAK,QAAQ;GAChB,KAAK,SAAS;GACd,aAAa,KAAK,QAAQ;EAC5B;EACA,OAAO,IAAI,SAAqB,YAAY;GAC1C,KAAK,QAAQ,WAAW;IACtB,cAAc,KAAK,QAAQ,CAAC;GAC9B,CAAC;EACH,CAAC;CACH;CAEA,UAAwB;EACtB,MAAM,OAAO,KAAK,QAAQ,MAAM;EAChC,IAAI,MACF,KAAK;OAEL,KAAK,SAAS;CAElB;CAEA,MAAM,aAAgB,IAAsC;EAC1D,MAAM,UAAU,MAAM,KAAK,QAAQ;EACnC,IAAI;GACF,OAAO,MAAM,GAAG;EAClB,UAAU;GACR,QAAQ;EACV;CACF;;CAGA,IAAI,WAAoB;EACtB,OAAO,KAAK;CACd;;CAGA,IAAI,UAAkB;EACpB,OAAO,KAAK,QAAQ;CACtB;AACF;;;;;;AAOA,eAAsB,cACpB,OACA,aACA,KACc;CACd,IAAI,CAAC,OAAO,UAAU,WAAW,KAAK,cAAc,GAClD,MAAM,IAAI,MAAM,8DAA8D,aAAa;CAE7F,IAAI,MAAM,WAAW,GAAG,OAAO,CAAC;CAEhC,MAAM,UAAU,IAAI,MAAS,MAAM,MAAM;CACzC,IAAI,YAAY;CAChB,IAAI,UAAU;CACd,IAAI,SAAS;CACb,IAAI;CAEJ,MAAM,SAAS,YAA2B;EACxC,OAAO,CAAC,SAAS;GACf,MAAM,QAAQ;GACd,aAAa;GACb,IAAI,SAAS,MAAM,QAAQ;GAE3B,IAAI;IACF,QAAQ,SAAS,MAAM,IAAI,MAAM,QAAS,KAAK;GACjD,SAAS,OAAO;IACd,UAAU;IACV,IAAI,CAAC,QAAQ;KACX,SAAS;KACT,UAAU;IACZ;GACF;EACF;CACF;CAEA,MAAM,QAAQ,IAAI,MAAM,KAAK,EAAE,QAAQ,KAAK,IAAI,aAAa,MAAM,MAAM,EAAE,SAAS,OAAO,CAAC,CAAC;CAC7F,IAAI,QAAQ,MAAM;CAClB,OAAO;AACT"}