@tangle-network/agent-eval 0.117.1 → 0.118.0

package/dist/control.js

@@ -10,9 +10,10 @@ import {
   stopOnNoProgress,
   stopOnRepeatedAction,
   subjectiveEval
-} from "./chunk-DXZRATT5.js";
-import "./chunk-TVVP3ZZQ.js";
-import "./chunk-5UF54T55.js";
+} from "./chunk-OYZAPX5G.js";
+import "./chunk-VQMK5FMP.js";
+import "./chunk-S3UZOQ5Y.js";
+import "./chunk-MA6HLL3S.js";
 import "./chunk-XJYR7XFV.js";
 import "./chunk-VSMTAMNK.js";
 import "./chunk-ONWEPEDO.js";

package/dist/fuzz.d.ts

@@ -1,7 +1,197 @@
-import { M as MaximumCharge, C as CostLedger, a as CostChannel } from './cost-ledger-DWy3XdJc.js';
-import { D as DefaultVerdict } from './verdict-C9MlYujm.js';
-import { A as AdversarialMutation } from './adversarial-B7loGVVX.js';
-import './errors-oeQrLqXC.js';
+type CostChannel = 'agent' | 'judge' | 'verifier' | 'analyst' | 'driver' | (string & {});
+interface CostUsage {
+    inputTokens: number;
+    outputTokens: number;
+    cachedTokens?: number;
+}
+interface CostCallBase {
+    callId: string;
+    channel: CostChannel;
+    phase: string;
+    actor: string;
+    model: string;
+    maximumCostUsd?: number;
+    tags?: Record<string, string>;
+    timestamp: number;
+}
+interface CostReceipt extends CostCallBase, CostUsage {
+    status: 'settled';
+    costUsd: number;
+    costUnknown: boolean;
+    usageUnknown?: boolean;
+    pricing?: {
+        inputUsdPerThousand: number;
+        outputUsdPerThousand: number;
+    };
+    actualCostUsd?: number;
+    error?: string;
+}
+interface CostReceiptInput extends CostUsage {
+    model: string;
+    actualCostUsd?: number;
+    costUnknown?: boolean;
+    usageUnknown?: boolean;
+}
+type MaximumCharge = {
+    externallyEnforcedMaximumUsd: number;
+} | ({
+    model: string;
+} & CostUsage);
+interface RunPaidCallInput<T> {
+    callId?: string;
+    channel: CostChannel;
+    phase: string;
+    actor: string;
+    /** Used before a provider receipt exists and on failures without one. */
+    model?: string;
+    tags?: Record<string, string>;
+    signal?: AbortSignal;
+    /** Provider-enforced dollar maximum, or maximum priced token usage. Required when capped. */
+    maximumCharge?: MaximumCharge;
+    /** `callId` can be forwarded as the provider's idempotency key. */
+    execute(signal: AbortSignal, callId: string): Promise<T>;
+    receipt(value: T): CostReceiptInput;
+    receiptFromError?(error: Error): CostReceiptInput | undefined;
+}
+type PaidCallResult<T> = {
+    succeeded: true;
+    callId: string;
+    value: T;
+    receipt: CostReceipt;
+} | {
+    succeeded: false;
+    callId?: string;
+    error: Error;
+    receipt?: CostReceipt;
+};
+interface ChannelRollup {
+    channel: CostChannel;
+    calls: number;
+    inputTokens: number;
+    outputTokens: number;
+    cachedTokens: number;
+    costUsd: number;
+    unpricedCalls: number;
+    unknownUsageCalls: number;
+}
+interface CostLedgerSummary {
+    totalCalls: number;
+    pendingCalls: number;
+    unresolvedCalls: number;
+    reservedCostUsd: number;
+    inputTokens: number;
+    outputTokens: number;
+    cachedTokens: number;
+    totalCostUsd: number;
+    byChannel: ChannelRollup[];
+    unpricedModels: string[];
+    fullyPriced: boolean;
+    usageComplete: boolean;
+    accountingComplete: boolean;
+    incompleteReasons: string[];
+}
+interface CostLedgerFilter {
+    channel?: CostChannel;
+    phase?: string;
+    tags?: Record<string, string>;
+}
+/** Append-only storage. `append` must atomically reject stale revisions. */
+interface CostLedgerPersistence {
+    read(): {
+        revision: string;
+        events: string;
+    };
+    append(expectedRevision: string, event: string): string | undefined;
+}
+interface CostLedgerOptions {
+    costCeilingUsd?: number;
+    persistence?: CostLedgerPersistence;
+    /** Import already-settled receipts without admitting new paid work. */
+    receipts?: readonly CostReceipt[];
+}
+/** Run-wide paid-call admission, durable call state, receipts, and summaries. */
+declare class CostLedger {
+    private readonly records;
+    private readonly activeCallIds;
+    private readonly lateCallIds;
+    private completedTasks;
+    private revision;
+    private costLimitPersisted;
+    readonly costCeilingUsd?: number;
+    private readonly persistence?;
+    constructor(input?: number | CostLedgerOptions);
+    runPaidCall<T>(input: RunPaidCallInput<T>): Promise<PaidCallResult<T>>;
+    /** Settle a call left pending by a crashed process after reconciling with the provider. */
+    reconcile(callId: string, observed: CostReceiptInput, options?: {
+        error?: string;
+    }): CostReceipt;
+    list(filter?: CostLedgerFilter): CostReceipt[];
+    summary(filter?: CostLedgerFilter): CostLedgerSummary;
+    markCompleted(count?: number): void;
+    costPerCompletedTask(): number | null;
+    private execute;
+    private captureLateOutcome;
+    private commitOutcome;
+    private captureFailure;
+    private commitReceipt;
+    private resolveMaximum;
+    private hasIncompleteSettledCall;
+    private appendRecord;
+    private ensureCostLimitPersisted;
+    private appendEvent;
+}
+
+/**
+ * Adversarial mutation contract.
+ *
+ * `AdversarialMutation<S>` is the scenario-mutation strategy the fuzz harness
+ * (`fuzzAgent`, src/fuzz) drives: paraphrase, edge-case substitution, or
+ * compositional combination of a scenario the policy currently passes, looking
+ * for the tail inputs that break it. The harness supplies the loop; consumers
+ * supply the mutations and the failure detector.
+ */
+interface AdversarialMutation<S> {
+    id: string;
+    /**
+     * Mutate one scenario. Return null to skip; return one or more new
+     * scenarios. The harness deduplicates by `mutateScenarioId(scenario)`.
+     */
+    mutate(parent: S, rng: () => number): Promise<S[]> | S[];
+}
+
+/**
+ * Validator-output verdict — substrate primitive for "did this output pass,
+ * and how well?"
+ *
+ * Used by:
+ *   - `@tangle-network/agent-eval/matrix` — verdict per cell in the cartesian.
+ *   - `@tangle-network/agent-runtime` — Validator<Output, Verdict = DefaultVerdict>.
+ *     Runtime keeps `Validator` because it's coupled to runtime-shaped
+ *     `ValidationCtx` (iteration, signal, traceEmitter); the verdict TYPE
+ *     itself is a substrate concept and lives here.
+ *
+ * Repo layering: agent-eval is the substrate (no upward deps). Both
+ * agent-runtime and agent-knowledge consume this type FROM agent-eval —
+ * never the other way around. See CLAUDE.md "Repo layering" for the rule.
+ */
+/**
+ * Minimal verdict shape — `valid` + `score` are required; `scores` +
+ * `notes` are optional surface. Validators that need richer shapes
+ * parameterise `Validator<Output, MyVerdict>` with their own type.
+ *
+ * Need structured extras? Extend DefaultVerdict with typed fields — never
+ * serialize extras into `notes`.
+ */
+interface DefaultVerdict {
+    /** Whether the output meets the validator's pass criteria. */
+    valid: boolean;
+    /** Aggregate score in [0, 1]. Drivers use this for winner selection. */
+    score: number;
+    /** Per-dimension scores. Free-form; weighted into `score` by the validator. */
+    scores?: Record<string, number>;
+    /** Human-readable rationale; surfaces in trace + final-result `winner.verdict`. */
+    notes?: string;
+}
 
 /**
  * Behavior-space exploration — types.

package/dist/fuzz.js

@@ -1,11 +1,11 @@
-import {
-  varianceBasedCurriculum
-} from "./chunk-VZSRQ272.js";
 import {
   CostLedger,
   CostReceiptCaptureError
 } from "./chunk-VCTY3W6J.js";
 import "./chunk-VI2UW6B6.js";
+import {
+  varianceBasedCurriculum
+} from "./chunk-VZSRQ272.js";
 import {
   ValidationError
 } from "./chunk-ONWEPEDO.js";