@tangle-network/agent-eval 0.95.0 → 0.96.0

package/dist/diagnose/repair.d.ts

@@ -1,77 +0,0 @@
-/**
- * Replay-validated repair — WHAT SHOULD HAVE HAPPENED?
- *
- * Takes the blamed steps from a `CausalResponsibilityReport`, asks a
- * consumer-supplied `proposeFix` (LLM-backed in live use) for candidate
- * mutations, and machine-verifies each candidate by replaying the run
- * WITH the mutation applied (through the same `runCounterfactual` seam
- * the sweep uses).
- *
- * A repair is "what should have happened" ONLY when every validation
- * replay crosses `flipThreshold` — a prescription is never speculated,
- * it is demonstrated. Candidates that don't flip, or whose replay
- * errors, land in `rejected` with a typed reason; nothing is dropped
- * silently.
- */
-import { type CounterfactualMutation, type CounterfactualRunner } from '../counterfactual';
-import type { TraceStore } from '../trace/store';
-import { type Trajectory, type TrajectoryStep } from '../trajectory';
-import type { StepRef, StepResponsibility } from './causal-sweep';
-/** Context handed to `proposeFix` so an LLM-backed proposer can see the
- *  full trajectory plus the responsibility evidence for the blamed step. */
-export interface RepairContext {
-    runId: string;
-    trajectory: Trajectory;
-    originalScore: number;
-    responsibility: StepResponsibility;
-}
-export interface PrescribeRepairOptions {
-    store: TraceStore;
-    /** The failed run the sweep diagnosed. */
-    runId: string;
-    /** Execution seam — same `CounterfactualRunner` contract as the sweep. */
-    runner: CounterfactualRunner;
-    /** Blamed steps from `causalSweep` — typically `report.steps.slice(0, k)`. */
-    blamed: StepResponsibility[];
-    /** Candidate-fix generator. Consumer-supplied; LLM-backed in live use.
-     *  Returned mutations MUST target the blamed step's index. */
-    proposeFix: (step: TrajectoryStep, context: RepairContext) => Promise<CounterfactualMutation[]>;
-    /** Score every validation replay must reach for the repair to count. Default 0.5. */
-    flipThreshold?: number;
-    /** Validation replays per candidate mutation. Default 3. */
-    repsToValidate?: number;
-    /** Max candidate mutations tried per step. Default: all proposed. */
-    maxAttemptsPerStep?: number;
-}
-export interface ValidatedRepair {
-    stepRef: StepRef;
-    mutation: CounterfactualMutation;
-    /** Always true — presence in `repairs` IS the machine-verified claim. */
-    validated: true;
-    /** Mean counterfactual score across the validation reps. */
-    meanScore: number;
-    /** meanScore − originalScore. */
-    deltaScore: number;
-    reps: number;
-    /** Replay run ids backing the validation — audit trail. */
-    counterfactualRunIds: string[];
-}
-export interface RejectedRepair {
-    stepRef: StepRef;
-    mutation: CounterfactualMutation;
-    reason: 'did-not-flip' | 'error';
-    /** Present for 'did-not-flip': mean delta over the reps that ran. */
-    deltaScore?: number;
-    /** Present for 'error': the message, preserved for diagnosis. */
-    error?: string;
-}
-export interface RepairReport {
-    runId: string;
-    originalScore: number;
-    flipThreshold: number;
-    repairs: ValidatedRepair[];
-    rejected: RejectedRepair[];
-    replaysUsed: number;
-}
-export declare function prescribeRepair(opts: PrescribeRepairOptions): Promise<RepairReport>;
-//# sourceMappingURL=repair.d.ts.map

package/dist/diagnose/repair.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"repair.d.ts","sourceRoot":"","sources":["../../src/diagnose/repair.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;GAcG;AAEH,OAAO,EACL,KAAK,sBAAsB,EAC3B,KAAK,oBAAoB,EAE1B,MAAM,mBAAmB,CAAA;AAE1B,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,gBAAgB,CAAA;AAChD,OAAO,EAAmB,KAAK,UAAU,EAAE,KAAK,cAAc,EAAE,MAAM,eAAe,CAAA;AACrF,OAAO,KAAK,EAAE,OAAO,EAAE,kBAAkB,EAAE,MAAM,gBAAgB,CAAA;AAEjE;4EAC4E;AAC5E,MAAM,WAAW,aAAa;IAC5B,KAAK,EAAE,MAAM,CAAA;IACb,UAAU,EAAE,UAAU,CAAA;IACtB,aAAa,EAAE,MAAM,CAAA;IACrB,cAAc,EAAE,kBAAkB,CAAA;CACnC;AAED,MAAM,WAAW,sBAAsB;IACrC,KAAK,EAAE,UAAU,CAAA;IACjB,0CAA0C;IAC1C,KAAK,EAAE,MAAM,CAAA;IACb,0EAA0E;IAC1E,MAAM,EAAE,oBAAoB,CAAA;IAC5B,8EAA8E;IAC9E,MAAM,EAAE,kBAAkB,EAAE,CAAA;IAC5B;kEAC8D;IAC9D,UAAU,EAAE,CAAC,IAAI,EAAE,cAAc,EAAE,OAAO,EAAE,aAAa,KAAK,OAAO,CAAC,sBAAsB,EAAE,CAAC,CAAA;IAC/F,qFAAqF;IACrF,aAAa,CAAC,EAAE,MAAM,CAAA;IACtB,4DAA4D;IAC5D,cAAc,CAAC,EAAE,MAAM,CAAA;IACvB,qEAAqE;IACrE,kBAAkB,CAAC,EAAE,MAAM,CAAA;CAC5B;AAED,MAAM,WAAW,eAAe;IAC9B,OAAO,EAAE,OAAO,CAAA;IAChB,QAAQ,EAAE,sBAAsB,CAAA;IAChC,yEAAyE;IACzE,SAAS,EAAE,IAAI,CAAA;IACf,4DAA4D;IAC5D,SAAS,EAAE,MAAM,CAAA;IACjB,iCAAiC;IACjC,UAAU,EAAE,MAAM,CAAA;IAClB,IAAI,EAAE,MAAM,CAAA;IACZ,2DAA2D;IAC3D,oBAAoB,EAAE,MAAM,EAAE,CAAA;CAC/B;AAED,MAAM,WAAW,cAAc;IAC7B,OAAO,EAAE,OAAO,CAAA;IAChB,QAAQ,EAAE,sBAAsB,CAAA;IAChC,MAAM,EAAE,cAAc,GAAG,OAAO,CAAA;IAChC,qEAAqE;IACrE,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,iEAAiE;IACjE,KAAK,CAAC,EAAE,MAAM,CAAA;CACf;AAED,MAAM,WAAW,YAAY;IAC3B,KAAK,EAAE,MAAM,CAAA;IACb,aAAa,EAAE,MAAM,CAAA;IACrB,aAAa,EAAE,MAAM,CAAA;IACrB,OAAO,EAAE,eAAe,EAAE,CAAA;IAC1B,QAAQ,EAAE,cAAc,EAAE,CAAA;IAC1B,WAAW,EAAE,MAAM,CAAA;CACpB;AAED,wBAAsB,eAAe,CAAC,IAAI,EAAE,sBAAsB,GAAG,OAAO,CAAC,YAAY,CAAC,CAgHzF"}

package/dist/discover-personas.d.ts

@@ -1,35 +0,0 @@
-/**
- * Walk a personas directory and return every file matching the convention
- * `NN-slug.{yaml,yml,json,md}`. Sorted by filename so the numeric prefix
- * gives stable persona ordering for reproducibility. Consumers filter
- * through `include` / `exclude`.
- */
-export interface DiscoverPersonasOptions {
-    /**
-     * Regex applied to filenames. Files that don't match are skipped.
-     * Default: `^[0-9]{2}-.+\.(yaml|yml|json|md)$`.
-     */
-    pattern?: RegExp;
-    /**
-     * Filenames (or basenames) to skip. Use this to exclude WIP / archived
-     * personas without removing the file.
-     */
-    exclude?: readonly string[];
-    /**
-     * If set, return only personas whose basename contains one of these
-     * substrings (post-pattern filter).
-     */
-    include?: readonly string[];
-    /** Recurse into subdirectories. Default false. */
-    recursive?: boolean;
-}
-export interface DiscoveredPersona {
-    /** Absolute file path. */
-    path: string;
-    /** Filename without directory. */
-    filename: string;
-    /** Filename without extension — the conventional persona id. */
-    id: string;
-}
-export declare function discoverPersonas(dir: string, opts?: DiscoverPersonasOptions): Promise<DiscoveredPersona[]>;
-//# sourceMappingURL=discover-personas.d.ts.map

package/dist/discover-personas.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"discover-personas.d.ts","sourceRoot":"","sources":["../src/discover-personas.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAKH,MAAM,WAAW,uBAAuB;IACtC;;;OAGG;IACH,OAAO,CAAC,EAAE,MAAM,CAAA;IAChB;;;OAGG;IACH,OAAO,CAAC,EAAE,SAAS,MAAM,EAAE,CAAA;IAC3B;;;OAGG;IACH,OAAO,CAAC,EAAE,SAAS,MAAM,EAAE,CAAA;IAC3B,kDAAkD;IAClD,SAAS,CAAC,EAAE,OAAO,CAAA;CACpB;AAED,MAAM,WAAW,iBAAiB;IAChC,0BAA0B;IAC1B,IAAI,EAAE,MAAM,CAAA;IACZ,kCAAkC;IAClC,QAAQ,EAAE,MAAM,CAAA;IAChB,gEAAgE;IAChE,EAAE,EAAE,MAAM,CAAA;CACX;AAID,wBAAsB,gBAAgB,CACpC,GAAG,EAAE,MAAM,EACX,IAAI,GAAE,uBAA4B,GACjC,OAAO,CAAC,iBAAiB,EAAE,CAAC,CA0C9B"}

package/dist/driver.d.ts

@@ -1,95 +0,0 @@
-import type { TCloud } from '@tangle-network/tcloud';
-import type { ProductClient } from './client';
-import type { DriverResult, DriverState, PersonaConfig } from './types';
-export interface AgentDriverConfig {
-    client: ProductClient;
-    driverModel?: string;
-    /** System prompt context for the driver LLM to understand the product */
-    productContext?: string;
-}
-/**
- * AgentDriver — meta-agent that plays a persona against the real product.
- *
- * Uses a driver LLM (Claude/GPT-4o) to decide what to say each turn.
- * Not scripted — the driver gets the current product state and decides
- * the next realistic user message.
- */
-export declare class AgentDriver {
-    private tc;
-    private client;
-    private driverModel;
-    private productContext;
-    constructor(tc: TCloud, config: AgentDriverConfig);
-    /**
-     * Run a persona through the product.
-     *
-     * Returns metrics on how many turns to completion, cost curve,
-     * quality curve, and convergence curve.
-     */
-    run(persona: PersonaConfig): Promise<DriverResult>;
-    /** Use the driver LLM to decide what the "user" says next */
-    private decideNextMessage;
-    /** Handle pending approvals based on persona feedback patterns */
-    private handleApprovals;
-}
-/**
- * Build the driver LLM's system prompt. The simulated user is an
- * adversarial senior professional: it judges the agent's last response by a
- * professional standard, refuses vague answers, challenges undefended
- * claims, probes the persona's pressure points without revealing them, and
- * signs off (DONE) only when a real practitioner would act on the work
- * unmodified. Pure function of persona, product state, and product context
- * — exported so harness authors can inspect and regression-test it.
- */
-export declare function buildDriverSystemPrompt(persona: PersonaConfig, state: DriverState, productContext?: string): string;
-export interface WorkerDriverContext {
-    /** The goal (or sub-goal) the driven worker must actually accomplish. */
-    goal: string;
-    /** The worker's harness — e.g. 'claude-code' | 'codex' | 'opencode' | 'router-tools'.
-     *  Names which capability profile the driver should exploit. */
-    harness?: string;
-    /** A capability + caveat brief for THIS harness (parallel tool calls, sub-agents and
-     *  their depth/concurrency limits, autonomy/runaway profile, MCP, native tool-isolation)
-     *  — sourced from the harness-compat matrix. Free text so the substrate stays decoupled
-     *  from any runtime harness type. */
-    harnessBrief?: string;
-    /** What the worker has done so far — the trace/state summary the driver reasons over to
-     *  write its next instruction. Empty on the first turn. */
-    progress?: string;
-    /** Optional extra context (repo, constraints, the deliverable's acceptance check). */
-    context?: string;
-}
-/**
- * Build the WORKER-DRIVER system prompt — the harness-aware sibling of
- * `buildDriverSystemPrompt`. Where that one role-plays a demanding *user* of a
- * product, this one is a meta-agent that DRIVES a capable coding *worker* to
- * complete a goal: it writes rich, high-signal instructions that direct the
- * worker to exploit its harness's full power (parallelize, sub-agents, run-to-
- * completion, tools/MCP), matched to what THAT harness can actually do. The
- * load-bearing contract: the driver never writes a thin steer — every message
- * is the dense, specific directive a world-class engineering lead would write.
- * Pure function; exported so harness authors can inspect and regression-test it.
- */
-export declare function buildWorkerDriverSystemPrompt(ctx: WorkerDriverContext): string;
-export interface DecideNextUserTurnOpts {
-    persona: PersonaConfig;
-    state: DriverState;
-    /** Conversation so far — alternating user/assistant messages, oldest first. */
-    history: {
-        role: string;
-        content: string;
-    }[];
-    /** Optional product context woven into the driver prompt. */
-    productContext?: string;
-    /** Driver LLM model. Defaults to claude-sonnet-4-6. */
-    model?: string;
-}
-/**
- * Decide the simulated user's next turn — the reactive, adversarial
- * turn-generation core of `AgentDriver`, exposed standalone so an in-process
- * eval harness can drive multi-shot conversations without the `ProductClient`
- * workspace machinery. Returns the next user message, or the literal "DONE"
- * when the simulated professional would sign off.
- */
-export declare function decideNextUserTurn(tc: TCloud, opts: DecideNextUserTurnOpts): Promise<string>;
-//# sourceMappingURL=driver.d.ts.map

package/dist/driver.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"driver.d.ts","sourceRoot":"","sources":["../src/driver.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,MAAM,EAAE,MAAM,wBAAwB,CAAA;AACpD,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,UAAU,CAAA;AAG7C,OAAO,KAAK,EAAE,YAAY,EAAE,WAAW,EAAE,aAAa,EAA6B,MAAM,SAAS,CAAA;AAElG,MAAM,WAAW,iBAAiB;IAChC,MAAM,EAAE,aAAa,CAAA;IACrB,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,yEAAyE;IACzE,cAAc,CAAC,EAAE,MAAM,CAAA;CACxB;AAeD;;;;;;GAMG;AACH,qBAAa,WAAW;IACtB,OAAO,CAAC,EAAE,CAAQ;IAClB,OAAO,CAAC,MAAM,CAAe;IAC7B,OAAO,CAAC,WAAW,CAAQ;IAC3B,OAAO,CAAC,cAAc,CAAQ;gBAElB,EAAE,EAAE,MAAM,EAAE,MAAM,EAAE,iBAAiB;IAOjD;;;;;OAKG;IACG,GAAG,CAAC,OAAO,EAAE,aAAa,GAAG,OAAO,CAAC,YAAY,CAAC;IAiGxD,6DAA6D;YAC/C,iBAAiB;IAc/B,kEAAkE;YACpD,eAAe;CAwB9B;AAYD;;;;;;;;GAQG;AACH,wBAAgB,uBAAuB,CACrC,OAAO,EAAE,aAAa,EACtB,KAAK,EAAE,WAAW,EAClB,cAAc,SAAK,GAClB,MAAM,CA0CR;AAED,MAAM,WAAW,mBAAmB;IAClC,yEAAyE;IACzE,IAAI,EAAE,MAAM,CAAA;IACZ;oEACgE;IAChE,OAAO,CAAC,EAAE,MAAM,CAAA;IAChB;;;yCAGqC;IACrC,YAAY,CAAC,EAAE,MAAM,CAAA;IACrB;+DAC2D;IAC3D,QAAQ,CAAC,EAAE,MAAM,CAAA;IACjB,sFAAsF;IACtF,OAAO,CAAC,EAAE,MAAM,CAAA;CACjB;AAED;;;;;;;;;;GAUG;AACH,wBAAgB,6BAA6B,CAAC,GAAG,EAAE,mBAAmB,GAAG,MAAM,CA4B9E;AAED,MAAM,WAAW,sBAAsB;IACrC,OAAO,EAAE,aAAa,CAAA;IACtB,KAAK,EAAE,WAAW,CAAA;IAClB,+EAA+E;IAC/E,OAAO,EAAE;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,OAAO,EAAE,MAAM,CAAA;KAAE,EAAE,CAAA;IAC5C,6DAA6D;IAC7D,cAAc,CAAC,EAAE,MAAM,CAAA;IACvB,uDAAuD;IACvD,KAAK,CAAC,EAAE,MAAM,CAAA;CACf;AAED;;;;;;GAMG;AACH,wBAAsB,kBAAkB,CACtC,EAAE,EAAE,MAAM,EACV,IAAI,EAAE,sBAAsB,GAC3B,OAAO,CAAC,MAAM,CAAC,CAiCjB"}

package/dist/driver.test.d.ts

@@ -1,8 +0,0 @@
-/**
- * The simulated user is the eval's difficulty knob. These tests pin the
- * adversarial contract of the driver system prompt so a future edit cannot
- * quietly soften it back into a passive script-reader — a regression that
- * would inflate every downstream agent score with no real improvement.
- */
-export {};
-//# sourceMappingURL=driver.test.d.ts.map

package/dist/driver.test.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"driver.test.d.ts","sourceRoot":"","sources":["../src/driver.test.ts"],"names":[],"mappings":"AAAA;;;;;GAKG"}

package/dist/dual-agent-bench.d.ts

@@ -1,81 +0,0 @@
-/**
- * Dual-agent convergence bench.
- *
- * Pattern lifted from dual-worker review loops: two agents take turns until
- * they converge on a consensus artifact. One proposes, the other critiques;
- * the proposer revises; repeat until a score threshold is hit or max rounds.
- *
- * Generalized so any two "agents" (gateways, local functions, anything with
- * `propose` + `critique`) compose in. Returns convergence rounds per
- * scenario + whether convergence happened.
- */
-export interface DualAgentScenario {
-    id: string;
-    initialPrompt: string;
-    /** Optional context the agents can read (e.g. source documents). */
-    context?: Record<string, unknown>;
-}
-export interface DualAgentRound {
-    roundIndex: number;
-    proposal: string;
-    critique: string;
-    convergenceScore: number;
-}
-export interface DualAgentScenarioResult {
-    scenarioId: string;
-    converged: boolean;
-    roundsToConverge: number | null;
-    finalProposal: string;
-    history: DualAgentRound[];
-    finalScore: number;
-}
-export interface DualAgentBenchConfig {
-    scenarios: DualAgentScenario[];
-    maxRounds?: number;
-    /** Convergence threshold in 0..1 (default 0.85). */
-    convergenceThreshold?: number;
-    /**
-     * Propose an answer given the scenario + the critic's prior critique (if any).
-     * Returns the proposal string.
-     */
-    propose: (args: {
-        scenario: DualAgentScenario;
-        roundIndex: number;
-        priorProposal?: string;
-        priorCritique?: string;
-    }) => Promise<string>;
-    /**
-     * Critique the proposer's current output. Returns a structured critique
-     * (free text) plus a convergence score: how close the proposal is to
-     * acceptable. 1.0 = accept, 0.0 = totally off.
-     */
-    critique: (args: {
-        scenario: DualAgentScenario;
-        roundIndex: number;
-        proposal: string;
-    }) => Promise<{
-        critique: string;
-        convergenceScore: number;
-    }>;
-    /** Optional per-round hook for progress + tracing. */
-    onRoundComplete?: (info: {
-        scenarioId: string;
-        round: DualAgentRound;
-    }) => void;
-}
-export interface DualAgentReport {
-    scenarios: DualAgentScenarioResult[];
-    aggregate: {
-        convergenceRate: number;
-        avgRoundsToConverge: number | null;
-        avgFinalScore: number;
-    };
-    config: {
-        maxRounds: number;
-        convergenceThreshold: number;
-    };
-}
-export declare class DualAgentBench {
-    run(config: DualAgentBenchConfig): Promise<DualAgentReport>;
-}
-//# sourceMappingURL=dual-agent-bench.d.ts.map

package/dist/dual-agent-bench.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"dual-agent-bench.d.ts","sourceRoot":"","sources":["../src/dual-agent-bench.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAEH,MAAM,WAAW,iBAAiB;IAChC,EAAE,EAAE,MAAM,CAAA;IACV,aAAa,EAAE,MAAM,CAAA;IACrB,oEAAoE;IACpE,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CAClC;AAED,MAAM,WAAW,cAAc;IAC7B,UAAU,EAAE,MAAM,CAAA;IAClB,QAAQ,EAAE,MAAM,CAAA;IAChB,QAAQ,EAAE,MAAM,CAAA;IAChB,gBAAgB,EAAE,MAAM,CAAA;CACzB;AAED,MAAM,WAAW,uBAAuB;IACtC,UAAU,EAAE,MAAM,CAAA;IAClB,SAAS,EAAE,OAAO,CAAA;IAClB,gBAAgB,EAAE,MAAM,GAAG,IAAI,CAAA;IAC/B,aAAa,EAAE,MAAM,CAAA;IACrB,OAAO,EAAE,cAAc,EAAE,CAAA;IACzB,UAAU,EAAE,MAAM,CAAA;CACnB;AAED,MAAM,WAAW,oBAAoB;IACnC,SAAS,EAAE,iBAAiB,EAAE,CAAA;IAC9B,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB,oDAAoD;IACpD,oBAAoB,CAAC,EAAE,MAAM,CAAA;IAC7B;;;OAGG;IACH,OAAO,EAAE,CAAC,IAAI,EAAE;QACd,QAAQ,EAAE,iBAAiB,CAAA;QAC3B,UAAU,EAAE,MAAM,CAAA;QAClB,aAAa,CAAC,EAAE,MAAM,CAAA;QACtB,aAAa,CAAC,EAAE,MAAM,CAAA;KACvB,KAAK,OAAO,CAAC,MAAM,CAAC,CAAA;IACrB;;;;OAIG;IACH,QAAQ,EAAE,CAAC,IAAI,EAAE;QACf,QAAQ,EAAE,iBAAiB,CAAA;QAC3B,UAAU,EAAE,MAAM,CAAA;QAClB,QAAQ,EAAE,MAAM,CAAA;KACjB,KAAK,OAAO,CAAC;QAAE,QAAQ,EAAE,MAAM,CAAC;QAAC,gBAAgB,EAAE,MAAM,CAAA;KAAE,CAAC,CAAA;IAC7D,sDAAsD;IACtD,eAAe,CAAC,EAAE,CAAC,IAAI,EAAE;QAAE,UAAU,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,cAAc,CAAA;KAAE,KAAK,IAAI,CAAA;CAChF;AAED,MAAM,WAAW,eAAe;IAC9B,SAAS,EAAE,uBAAuB,EAAE,CAAA;IACpC,SAAS,EAAE;QACT,eAAe,EAAE,MAAM,CAAA;QACvB,mBAAmB,EAAE,MAAM,GAAG,IAAI,CAAA;QAClC,aAAa,EAAE,MAAM,CAAA;KACtB,CAAA;IACD,MAAM,EAAE;QACN,SAAS,EAAE,MAAM,CAAA;QACjB,oBAAoB,EAAE,MAAM,CAAA;KAC7B,CAAA;CACF;AAED,qBAAa,cAAc;IACnB,GAAG,CAAC,MAAM,EAAE,oBAAoB,GAAG,OAAO,CAAC,eAAe,CAAC;CAoFlE"}

package/dist/error-count-extractor.d.ts

@@ -1,47 +0,0 @@
-/**
- * Toolchain error-count extractor.
- *
- * Given stderr/stdout from a compiler or test runner, count the number
- * of reported errors/failures. Patterns are deliberately narrow —
- * unknown stderr returns `null` rather than zero so callers can
- * distinguish "no errors" from "different toolchain, couldn't parse".
- *
- * All patterns are anchored to the start of a line and use bounded
- * character classes to avoid catastrophic backtracking on pathological
- * inputs.
- *
- * Add new toolchains by appending to {@link ERROR_COUNT_PATTERNS};
- * order matters only in the sense that the first matching pattern wins.
- */
-export interface ErrorCountPattern {
-    /** Stable identifier for logging + tests. */
-    name: string;
-    /** Must be global (`g` flag) — the extractor counts matches. */
-    regex: RegExp;
-    /** Optional post-processing to extract a count from a single captured match. */
-    transform?: (match: RegExpMatchArray) => number;
-}
-export declare const ERROR_COUNT_PATTERNS: ErrorCountPattern[];
-export interface ExtractOptions {
-    /** Restrict to named patterns — default: all patterns. */
-    only?: string[];
-    /** Additional patterns to consider BEFORE the built-in list. */
-    extra?: ErrorCountPattern[];
-}
-export interface ExtractResult {
-    /** Total count of matched errors, or null when no pattern matched. */
-    count: number | null;
-    /** Name of the pattern that matched, or null. */
-    matched: string | null;
-    /** Original matches for callers that want to surface specifics. */
-    samples: string[];
-}
-/**
- * Try each pattern in order; return the first with matches.
- *
- * Returning `null` (instead of zero) on no-match is deliberate — a
- * callsite that greps for "typescript errors" on cargo output should
- * NOT treat that as "zero TS errors" because the toolchain is wrong.
- */
-export declare function extractErrorCount(text: string, opts?: ExtractOptions): ExtractResult;
-//# sourceMappingURL=error-count-extractor.d.ts.map

package/dist/error-count-extractor.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"error-count-extractor.d.ts","sourceRoot":"","sources":["../src/error-count-extractor.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;GAcG;AAEH,MAAM,WAAW,iBAAiB;IAChC,6CAA6C;IAC7C,IAAI,EAAE,MAAM,CAAA;IACZ,gEAAgE;IAChE,KAAK,EAAE,MAAM,CAAA;IACb,gFAAgF;IAChF,SAAS,CAAC,EAAE,CAAC,KAAK,EAAE,gBAAgB,KAAK,MAAM,CAAA;CAChD;AAED,eAAO,MAAM,oBAAoB,EAAE,iBAAiB,EAkCnD,CAAA;AAED,MAAM,WAAW,cAAc;IAC7B,0DAA0D;IAC1D,IAAI,CAAC,EAAE,MAAM,EAAE,CAAA;IACf,gEAAgE;IAChE,KAAK,CAAC,EAAE,iBAAiB,EAAE,CAAA;CAC5B;AAED,MAAM,WAAW,aAAa;IAC5B,sEAAsE;IACtE,KAAK,EAAE,MAAM,GAAG,IAAI,CAAA;IACpB,iDAAiD;IACjD,OAAO,EAAE,MAAM,GAAG,IAAI,CAAA;IACtB,mEAAmE;IACnE,OAAO,EAAE,MAAM,EAAE,CAAA;CAClB;AAED;;;;;;GAMG;AACH,wBAAgB,iBAAiB,CAAC,IAAI,EAAE,MAAM,EAAE,IAAI,GAAE,cAAmB,GAAG,aAAa,CAuBxF"}

package/dist/error-count-extractor.test.d.ts

@@ -1,2 +0,0 @@
-export {};
-//# sourceMappingURL=error-count-extractor.test.d.ts.map

package/dist/error-count-extractor.test.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"error-count-extractor.test.d.ts","sourceRoot":"","sources":["../src/error-count-extractor.test.ts"],"names":[],"mappings":""}

package/dist/errors.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"errors.d.ts","sourceRoot":"","sources":["../src/errors.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;GAaG;AAEH,MAAM,MAAM,kBAAkB,GAC1B,YAAY,GACZ,WAAW,GACX,QAAQ,GACR,mBAAmB,GACnB,OAAO,GACP,cAAc,GACd,QAAQ,GACR,mBAAmB,GACnB,gBAAgB,CAAA;AAEpB,qBAAa,cAAe,SAAQ,KAAK;IACvC,oEAAoE;IACpE,QAAQ,CAAC,IAAI,EAAE,kBAAkB,CAAA;gBAErB,IAAI,EAAE,kBAAkB,EAAE,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE;QAAE,KAAK,CAAC,EAAE,OAAO,CAAA;KAAE;CAKrF;AAED,6FAA6F;AAC7F,qBAAa,eAAgB,SAAQ,cAAc;gBACrC,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE;QAAE,KAAK,CAAC,EAAE,OAAO,CAAA;KAAE;CAG3D;AAED,yFAAyF;AACzF,qBAAa,aAAc,SAAQ,cAAc;gBACnC,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE;QAAE,KAAK,CAAC,EAAE,OAAO,CAAA;KAAE;CAG3D;AAED,sGAAsG;AACtG,qBAAa,WAAY,SAAQ,cAAc;gBACjC,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE;QAAE,KAAK,CAAC,EAAE,OAAO,CAAA;KAAE;CAG3D;AAED;;;;GAIG;AACH,qBAAa,qBAAsB,SAAQ,cAAc;gBAC3C,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE;QAAE,KAAK,CAAC,EAAE,OAAO,CAAA;KAAE;CAG3D;AAED,mHAAmH;AACnH,qBAAa,UAAW,SAAQ,cAAc;gBAChC,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE;QAAE,KAAK,CAAC,EAAE,OAAO,CAAA;KAAE;CAG3D;AAED,qGAAqG;AACrG,qBAAa,iBAAkB,SAAQ,cAAc;gBACvC,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE;QAAE,KAAK,CAAC,EAAE,OAAO,CAAA;KAAE;CAG3D;AAED,wGAAwG;AACxG,qBAAa,WAAY,SAAQ,cAAc;gBACjC,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE;QAAE,KAAK,CAAC,EAAE,OAAO,CAAA;KAAE;CAG3D"}

package/dist/eval-campaign.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"eval-campaign.d.ts","sourceRoot":"","sources":["../src/eval-campaign.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAsCG;AAEH,OAAO,EACL,KAAK,gBAAgB,EACrB,KAAK,qBAAqB,EAG3B,MAAM,sBAAsB,CAAA;AAC7B,OAAO,EAAkB,KAAK,gBAAgB,EAAE,KAAK,oBAAoB,EAAE,MAAM,cAAc,CAAA;AAE/F,OAAO,KAAK,EACV,iBAAiB,EACjB,gBAAgB,EAEhB,SAAS,EACT,WAAW,EACX,aAAa,EACd,MAAM,cAAc,CAAA;AAErB,OAAO,EAAE,KAAK,cAAc,EAAE,KAAK,qBAAqB,EAAkB,MAAM,kBAAkB,CAAA;AAClG,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,iBAAiB,CAAA;AACtD,OAAO,EAAE,YAAY,EAAE,MAAM,iBAAiB,CAAA;AAC9C,OAAO,EAGL,KAAK,wBAAwB,EAC7B,KAAK,kBAAkB,EACxB,MAAM,mBAAmB,CAAA;AAC1B,OAAO,EAA6B,KAAK,eAAe,EAAE,MAAM,2BAA2B,CAAA;AAC3F,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,gBAAgB,CAAA;AAClD,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,eAAe,CAAA;AAI/C,MAAM,WAAW,eAAe,CAAC,CAAC;IAChC,EAAE,EAAE,MAAM,CAAA;IACV,OAAO,EAAE,CAAC,CAAA;CACX;AAED,MAAM,WAAW,gBAAgB;IAC/B,UAAU,EAAE,MAAM,CAAA;IAClB,yDAAyD;IACzD,IAAI,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;CAC9B;AAED,MAAM,WAAW,kBAAkB,CAAC,CAAC;IACnC,uEAAuE;IACvE,KAAK,EAAE,MAAM,CAAA;IACb,mFAAmF;IACnF,YAAY,EAAE,MAAM,CAAA;IACpB,OAAO,EAAE,CAAC,CAAA;IACV,SAAS,EAAE,MAAM,CAAA;IACjB,UAAU,EAAE,MAAM,CAAA;IAClB,YAAY,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IACpC,IAAI,EAAE,MAAM,CAAA;IACZ,QAAQ,EAAE,WAAW,CAAA;IACrB;;;;;OAKG;IACH,OAAO,EAAE,YAAY,CAAA;IACrB,KAAK,EAAE,UAAU,CAAA;IACjB,OAAO,EAAE,eAAe,CAAA;IACxB;;;;OAIG;IACH,OAAO,EAAE,gBAAgB,CAAA;CAC1B;AAED,MAAM,WAAW,kBAAkB;IACjC,6DAA6D;IAC7D,IAAI,EAAE,OAAO,CAAA;IACb,+EAA+E;IAC/E,KAAK,EAAE,MAAM,CAAA;IACb,+EAA+E;IAC/E,OAAO,EAAE,MAAM,CAAA;IACf,UAAU,EAAE,aAAa,CAAA;IACzB,+DAA+D;IAC/D,KAAK,EAAE,MAAM,CAAA;IACb,wDAAwD;IACxD,UAAU,EAAE,MAAM,CAAA;IAClB,kFAAkF;IAClF,UAAU,EAAE,MAAM,CAAA;IAClB,+DAA+D;IAC/D,GAAG,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IAC5B;;6EAEyE;IACzE,YAAY,CAAC,EAAE,YAAY,CAAA;IAC3B,sEAAsE;IACtE,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,qDAAqD;IACrD,aAAa,CAAC,EAAE,gBAAgB,CAAA;IAChC;;;;OAIG;IACH,WAAW,CAAC,EAAE,iBAAiB,CAAA;IAC/B;;;;OAIG;IACH,YAAY,CAAC,EAAE,gBAAgB,GAAG,qBAAqB,CAAA;CACxD;AAED,MAAM,MAAM,cAAc,CAAC,CAAC,IAAI,CAAC,GAAG,EAAE,kBAAkB,CAAC,CAAC,CAAC,KAAK,OAAO,CAAC,kBAAkB,CAAC,CAAA;AAE3F,MAAM,MAAM,uBAAuB,GAAG,OAAO,GAAG,aAAa,GAAG,KAAK,CAAA;AAErE,MAAM,WAAW,mBAAmB,CAAC,CAAC;IACpC;;;OAGG;IACH,UAAU,EAAE,MAAM,CAAA;IAClB,QAAQ,EAAE,eAAe,CAAC,CAAC,CAAC,EAAE,CAAA;IAC9B,SAAS,EAAE,gBAAgB,EAAE,CAAA;IAC7B,2BAA2B;IAC3B,KAAK,CAAC,EAAE,MAAM,EAAE,CAAA;IAChB,sEAAsE;IACtE,QAAQ,CAAC,EAAE,WAAW,CAAA;IACtB,iFAAiF;IACjF,SAAS,EAAE,MAAM,CAAA;IACjB;;;;OAIG;IACH,OAAO,EAAE,gBAAgB,CAAA;IACzB;;;;OAIG;IACH,iBAAiB,CAAC,EAAE,oBAAoB,CAAA;IACxC;;;;OAIG;IACH,YAAY,EAAE,CAAC,MAAM,EAAE,qBAAqB,KAAK,UAAU,CAAA;IAC3D;;;;;OAKG;IACH,cAAc,CAAC,EAAE,CAAC,MAAM,EAAE,qBAAqB,KAAK,eAAe,CAAA;IACnE;;;OAGG;IACH,OAAO,CAAC,EAAE,MAAM,CAAA;IAChB;;;OAGG;IACH,aAAa,CAAC,EAAE,eAAe,EAAE,CAAA;IACjC;;;;OAIG;IACH,SAAS,CAAC,EAAE,wBAAwB,CAAA;IACpC,+DAA+D;IAC/D,kBAAkB,CAAC,EAAE,uBAAuB,CAAA;IAC5C;;;OAGG;IACH,MAAM,EAAE,cAAc,CAAC,CAAC,CAAC,CAAA;IACzB;;;OAGG;IACH,MAAM,CAAC,EAAE;QAAE,UAAU,CAAC,EAAE,MAAM,CAAA;KAAE,GAAG,IAAI,CACrC,qBAAqB,EACrB,YAAY,GAAG,qBAAqB,GAAG,aAAa,CACrD,CAAA;IACD;;;OAGG;IACH,mBAAmB,CAAC,EAAE,MAAM,CAAA;IAC5B,mDAAmD;IACnD,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB;;OAEG;IACH,GAAG,CAAC,EAAE,MAAM,MAAM,CAAA;IAClB,oDAAoD;IACpD,KAAK,CAAC,EAAE,CAAC,MAAM,EAAE,qBAAqB,KAAK,MAAM,CAAA;IACjD;;;;;OAKG;IACH,YAAY,CAAC,EACT,gBAAgB,GAChB,qBAAqB,GACrB,CAAC,CACC,MAAM,EAAE,qBAAqB,GAAG;QAC9B,OAAO,EAAE,CAAC,CAAA;QACV,YAAY,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;KACrC,KAEC,gBAAgB,GAChB,qBAAqB,GACrB,OAAO,CAAC,gBAAgB,GAAG,qBAAqB,CAAC,CAAC,CAAA;CAC3D;AAED,MAAM,WAAW,qBAAqB;IACpC,UAAU,EAAE,MAAM,CAAA;IAClB,KAAK,EAAE,MAAM,CAAA;IACb,SAAS,EAAE,MAAM,CAAA;IACjB,UAAU,EAAE,MAAM,CAAA;IAClB,IAAI,EAAE,MAAM,CAAA;CACb;AAED,MAAM,WAAW,SAAS;IACxB,KAAK,EAAE,MAAM,CAAA;IACb,SAAS,EAAE,MAAM,CAAA;IACjB,UAAU,EAAE,MAAM,CAAA;IAClB,IAAI,EAAE,MAAM,CAAA;IACZ,MAAM,EAAE,MAAM,CAAA;IACd,KAAK,CAAC,EAAE,MAAM,CAAA;CACf;AAED,MAAM,WAAW,kBAAkB;IACjC,UAAU,EAAE,MAAM,CAAA;IAClB,mIAAmI;IACnI,mBAAmB,EAAE,MAAM,CAAA;IAC3B,mBAAmB,EAAE,MAAM,GAAG,IAAI,CAAA;IAClC,8DAA8D;IAC9D,IAAI,EAAE,SAAS,EAAE,CAAA;IACjB,kDAAkD;IAClD,gBAAgB,EAAE,kBAAkB,EAAE,CAAA;IACtC,UAAU,EAAE,SAAS,EAAE,CAAA;IACvB,gDAAgD;IAChD,MAAM,CAAC,EAAE,cAAc,CAAA;IACvB,SAAS,EAAE,MAAM,CAAA;IACjB,OAAO,EAAE,MAAM,CAAA;CAChB;AAeD,wBAAsB,eAAe,CAAC,CAAC,EACrC,IAAI,EAAE,mBAAmB,CAAC,CAAC,CAAC,GAC3B,OAAO,CAAC,kBAAkB,CAAC,CA8T7B;AAcD;;;;;;;;;;;;;;GAcG;AACH,wBAAsB,aAAa,CACjC,OAAO,EAAE,YAAY,EACrB,KAAK,EAAE,MAAM,EACb,MAAM,EAAE,MAAM,GACb,OAAO,CAAC,IAAI,CAAC,CAKf"}

package/dist/eval-campaign.test.d.ts

@@ -1,2 +0,0 @@
-export {};
-//# sourceMappingURL=eval-campaign.test.d.ts.map

package/dist/eval-campaign.test.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"eval-campaign.test.d.ts","sourceRoot":"","sources":["../src/eval-campaign.test.ts"],"names":[],"mappings":""}

package/dist/eval-tools.d.ts

@@ -1,55 +0,0 @@
-/**
- * Eval primitives as agent tools — `makeEvalTools` packages the substrate's
- * judge / completion / analysis entry points as JSON-Schema tool definitions
- * an LLM agent loop can call. The host closes over the live config (judge
- * panels, the correctness checker, analyst registry); the agent passes only
- * data over the wire.
- *
- * Three tools, each present only when its config section is supplied:
- *   - `run_judges`        — score an artifact with the configured `JudgeConfig`s
- *   - `verify_completion` — gold-spec requirement check → `CompletionVerdict`
- *   - `analyze_runs`      — `RunRecord[]` (inline or from a file) → `InsightReport`
- *
- * `toOpenAiTool` converts a definition to the OpenAI function-tool wire shape.
- */
-import type { JudgeConfig } from './campaign/types';
-import { type CorrectnessChecker } from './completion-verifier';
-import { type AnalyzeRunsOptions } from './contract/analyze-runs';
-/** One agent-callable tool. `parameters` is a JSON Schema (draft-07+) object. */
-export interface EvalToolDef {
-    name: string;
-    description: string;
-    parameters: Record<string, unknown>;
-    handler: (args: unknown, ctx?: {
-        signal?: AbortSignal;
-    }) => Promise<unknown>;
-}
-export interface MakeEvalToolsConfig {
-    /** Judges available to `run_judges`. Omit to exclude the tool. */
-    judges?: Array<JudgeConfig<unknown>>;
-    /** Host-side correctness checker for `verify_completion` (the third
-     *  `verifyCompletion` argument — a function, so it cannot cross the wire).
-     *  Omit to exclude the tool. */
-    completion?: {
-        checkCorrectness: CorrectnessChecker;
-    };
-    /** `analyzeRuns` options minus `runs` (runs arrive as tool args, inline or
-     *  via `path`). Omit to exclude the tool. */
-    analyze?: Omit<AnalyzeRunsOptions, 'runs'>;
-}
-/** OpenAI function-tool wire shape for an `EvalToolDef`. */
-export declare function toOpenAiTool(def: EvalToolDef): {
-    type: 'function';
-    function: {
-        name: string;
-        description: string;
-        parameters: Record<string, unknown>;
-    };
-};
-/**
- * Build the eval toolset for the supplied config. Only sections present in
- * `cfg` produce tools, so the agent's tool list mirrors what the host
- * actually wired. Handlers fail loud on malformed args — no silent defaults.
- */
-export declare function makeEvalTools(cfg: MakeEvalToolsConfig): EvalToolDef[];
-//# sourceMappingURL=eval-tools.d.ts.map

package/dist/eval-tools.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"eval-tools.d.ts","sourceRoot":"","sources":["../src/eval-tools.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;GAaG;AAGH,OAAO,KAAK,EAAE,WAAW,EAAwB,MAAM,kBAAkB,CAAA;AACzE,OAAO,EAEL,KAAK,kBAAkB,EAIxB,MAAM,uBAAuB,CAAA;AAC9B,OAAO,EAAE,KAAK,kBAAkB,EAAe,MAAM,yBAAyB,CAAA;AAG9E,iFAAiF;AACjF,MAAM,WAAW,WAAW;IAC1B,IAAI,EAAE,MAAM,CAAA;IACZ,WAAW,EAAE,MAAM,CAAA;IACnB,UAAU,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;IACnC,OAAO,EAAE,CAAC,IAAI,EAAE,OAAO,EAAE,GAAG,CAAC,EAAE;QAAE,MAAM,CAAC,EAAE,WAAW,CAAA;KAAE,KAAK,OAAO,CAAC,OAAO,CAAC,CAAA;CAC7E;AAED,MAAM,WAAW,mBAAmB;IAClC,kEAAkE;IAClE,MAAM,CAAC,EAAE,KAAK,CAAC,WAAW,CAAC,OAAO,CAAC,CAAC,CAAA;IACpC;;oCAEgC;IAChC,UAAU,CAAC,EAAE;QAAE,gBAAgB,EAAE,kBAAkB,CAAA;KAAE,CAAA;IACrD;iDAC6C;IAC7C,OAAO,CAAC,EAAE,IAAI,CAAC,kBAAkB,EAAE,MAAM,CAAC,CAAA;CAC3C;AAED,4DAA4D;AAC5D,wBAAgB,YAAY,CAAC,GAAG,EAAE,WAAW,GAAG;IAC9C,IAAI,EAAE,UAAU,CAAA;IAChB,QAAQ,EAAE;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,MAAM,CAAC;QAAC,UAAU,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;KAAE,CAAA;CACrF,CAKA;AAED;;;;GAIG;AACH,wBAAgB,aAAa,CAAC,GAAG,EAAE,mBAAmB,GAAG,WAAW,EAAE,CAMrE"}

package/dist/eval-trace-store.d.ts

@@ -1,107 +0,0 @@
-/**
- * EvalTraceStore — JSONL save / query / compare over the analysis-time
- * `RunRecord` row.
- *
- * `FileSystemTraceStore` (in `./trace/store`) persists the rich TraceSchema-v1
- * span corpus — runs, spans, events, artifacts — the runtime emits live. That
- * is NOT the thing three consumers hand-roll in `tests/eval/lib/trace-store.ts`.
- * What they hand-roll is the *analysis* layer: append each finished run as one
- * JSONL line keyed by scenario/candidate, then query it ("every run where score
- * < 50"), pick the best run for a scenario (few-shot seeding), and compare two
- * candidates on matched scenarios.
- *
- * The substrate already has the canonical analysis-time row — `RunRecord` — so
- * this is that query/compare API expressed over `RunRecord[]`, with a
- * JSONL-backed store. It does NOT fork `FileSystemTraceStore`; it sits beside it
- * for the analysis projection.
- */
-import { type RunRecord, type RunSplitTag } from './run-record';
-/** The score the query/compare layer ranks on: holdout when present (the
- *  gated number), else search. Throws when a record carries neither — a
- *  RunRecord is invalid without at least one, but a hand-built object might. */
-export declare function runScore(record: RunRecord): number;
-export interface RunRecordFilter {
-    experimentId?: string;
-    candidateId?: string;
-    scenarioId?: string;
-    model?: string;
-    splitTag?: RunSplitTag;
-    /** Inclusive lower bound on `runScore`. */
-    minScore?: number;
-    /** Inclusive upper bound on `runScore`. */
-    maxScore?: number;
-    /** Match a single tag in `outcome.raw` by exact numeric value. */
-    rawEquals?: {
-        key: string;
-        value: number;
-    };
-    /** Custom predicate, ANDed with the structured filters. */
-    where?: (record: RunRecord) => boolean;
-}
-export interface CandidateComparison {
-    a: string;
-    b: string;
-    /** Scenario ids present for BOTH candidates — the paired comparison set. */
-    pairedScenarioIds: string[];
-    /** Mean `runScore` for candidate a over the paired scenarios. */
-    meanA: number;
-    /** Mean `runScore` for candidate b over the paired scenarios. */
-    meanB: number;
-    /** meanB − meanA. Positive ⇒ b scored higher on the matched scenarios. */
-    meanDelta: number;
-    /** Scenarios where b beat a (strictly), tied, and a beat b. */
-    bWins: number;
-    ties: number;
-    aWins: number;
-}
-/**
- * Backing persistence for `EvalTraceStore`. The in-memory store is the default;
- * the JSONL file store appends one validated `RunRecord` per line. Both keep an
- * append order so `getBest` / `compareRuns` are reproducible.
- */
-export interface RunRecordBackend {
-    append(record: RunRecord): Promise<void>;
-    load(): Promise<RunRecord[]>;
-}
-export declare function inMemoryRunRecordBackend(initial?: RunRecord[]): RunRecordBackend;
-/**
- * JSONL-backed store at `path`, one `RunRecord` per line. Malformed lines fail
- * loud on load (a corrupt corpus must not silently shrink the analysis set);
- * pass `skipInvalid` only for forensics on a known-bad file.
- */
-export declare function jsonlRunRecordBackend(path: string, opts?: {
-    skipInvalid?: boolean;
-}): RunRecordBackend;
-/**
- * Query / compare layer over a `RunRecord` corpus. Append finished runs, query
- * with a structured filter, take the best run for a scenario, and compare two
- * candidates on their matched scenarios. Persistence is injected via
- * `RunRecordBackend` (in-memory by default, JSONL file via
- * `jsonlRunRecordBackend`).
- */
-export declare class EvalTraceStore {
-    private readonly backend;
-    constructor(backend?: RunRecordBackend);
-    /** Validate and append one run. Throws on an invalid record — the corpus
-     *  stays paper-grade. */
-    append(record: RunRecord): Promise<void>;
-    all(): Promise<RunRecord[]>;
-    query(filter?: RunRecordFilter): Promise<RunRecord[]>;
-    /**
-     * Highest-scoring run for a scenario (optionally restricted to a candidate).
-     * Returns null when no run matches. Ties resolve to the earliest-appended run
-     * so the result is stable.
-     */
-    getBest(scenarioId: string, opts?: {
-        candidateId?: string;
-        splitTag?: RunSplitTag;
-    }): Promise<RunRecord | null>;
-    /**
-     * Compare two candidates on the scenarios they BOTH ran. When a candidate
-     * ran a scenario more than once, its best `runScore` for that scenario is
-     * used. Throws when there is no paired scenario — an unpaired "comparison" is
-     * not one.
-     */
-    compareRuns(candidateA: string, candidateB: string): Promise<CandidateComparison>;
-}
-//# sourceMappingURL=eval-trace-store.d.ts.map

package/dist/eval-trace-store.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"eval-trace-store.d.ts","sourceRoot":"","sources":["../src/eval-trace-store.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;GAgBG;AAGH,OAAO,EAAe,KAAK,SAAS,EAAE,KAAK,WAAW,EAAqB,MAAM,cAAc,CAAA;AAE/F;;gFAEgF;AAChF,wBAAgB,QAAQ,CAAC,MAAM,EAAE,SAAS,GAAG,MAAM,CAOlD;AAED,MAAM,WAAW,eAAe;IAC9B,YAAY,CAAC,EAAE,MAAM,CAAA;IACrB,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,QAAQ,CAAC,EAAE,WAAW,CAAA;IACtB,2CAA2C;IAC3C,QAAQ,CAAC,EAAE,MAAM,CAAA;IACjB,2CAA2C;IAC3C,QAAQ,CAAC,EAAE,MAAM,CAAA;IACjB,kEAAkE;IAClE,SAAS,CAAC,EAAE;QAAE,GAAG,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,CAAA;IAC1C,2DAA2D;IAC3D,KAAK,CAAC,EAAE,CAAC,MAAM,EAAE,SAAS,KAAK,OAAO,CAAA;CACvC;AAeD,MAAM,WAAW,mBAAmB;IAClC,CAAC,EAAE,MAAM,CAAA;IACT,CAAC,EAAE,MAAM,CAAA;IACT,4EAA4E;IAC5E,iBAAiB,EAAE,MAAM,EAAE,CAAA;IAC3B,iEAAiE;IACjE,KAAK,EAAE,MAAM,CAAA;IACb,iEAAiE;IACjE,KAAK,EAAE,MAAM,CAAA;IACb,0EAA0E;IAC1E,SAAS,EAAE,MAAM,CAAA;IACjB,+DAA+D;IAC/D,KAAK,EAAE,MAAM,CAAA;IACb,IAAI,EAAE,MAAM,CAAA;IACZ,KAAK,EAAE,MAAM,CAAA;CACd;AAED;;;;GAIG;AACH,MAAM,WAAW,gBAAgB;IAC/B,MAAM,CAAC,MAAM,EAAE,SAAS,GAAG,OAAO,CAAC,IAAI,CAAC,CAAA;IACxC,IAAI,IAAI,OAAO,CAAC,SAAS,EAAE,CAAC,CAAA;CAC7B;AAED,wBAAgB,wBAAwB,CAAC,OAAO,GAAE,SAAS,EAAO,GAAG,gBAAgB,CAUpF;AAED;;;;GAIG;AACH,wBAAgB,qBAAqB,CACnC,IAAI,EAAE,MAAM,EACZ,IAAI,GAAE;IAAE,WAAW,CAAC,EAAE,OAAO,CAAA;CAAO,GACnC,gBAAgB,CA4ClB;AAED;;;;;;GAMG;AACH,qBAAa,cAAc;IACzB,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAkB;gBAE9B,OAAO,GAAE,gBAA6C;IAIlE;6BACyB;IACnB,MAAM,CAAC,MAAM,EAAE,SAAS,GAAG,OAAO,CAAC,IAAI,CAAC;IAIxC,GAAG,IAAI,OAAO,CAAC,SAAS,EAAE,CAAC;IAI3B,KAAK,CAAC,MAAM,GAAE,eAAoB,GAAG,OAAO,CAAC,SAAS,EAAE,CAAC;IAK/D;;;;OAIG;IACG,OAAO,CACX,UAAU,EAAE,MAAM,EAClB,IAAI,GAAE;QAAE,WAAW,CAAC,EAAE,MAAM,CAAC;QAAC,QAAQ,CAAC,EAAE,WAAW,CAAA;KAAO,GAC1D,OAAO,CAAC,SAAS,GAAG,IAAI,CAAC;IAmB5B;;;;;OAKG;IACG,WAAW,CAAC,UAAU,EAAE,MAAM,EAAE,UAAU,EAAE,MAAM,GAAG,OAAO,CAAC,mBAAmB,CAAC;CAuDxF"}

package/dist/eval-trace-store.test.d.ts

@@ -1,2 +0,0 @@
-export {};
-//# sourceMappingURL=eval-trace-store.test.d.ts.map

package/dist/eval-trace-store.test.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"eval-trace-store.test.d.ts","sourceRoot":"","sources":["../src/eval-trace-store.test.ts"],"names":[],"mappings":""}

package/dist/executor.d.ts

@@ -1,38 +0,0 @@
-import type { TCloud } from '@tangle-network/tcloud';
-import type { CollectedArtifacts, JudgeFn, Scenario, ScenarioResult } from './types';
-export interface ExecutorConfig {
-    /** System prompt for the agent under test */
-    systemPrompt: string;
-    /** Model to use for the agent */
-    model?: string;
-    /** Judges to run after execution */
-    judges: JudgeFn[];
-    /** Regex patterns for detecting tool/API calls in responses */
-    toolCallPatterns?: RegExp[];
-    /** Block delimiter pattern (default: :::type\n...\n:::) */
-    blockPattern?: RegExp;
-    /** Custom artifact checker for domain-specific checks */
-    artifactChecker?: (check: Scenario['artifactChecks'][0], artifacts: CollectedArtifacts) => {
-        passed: boolean;
-        detail: string;
-    } | null;
-    /**
-     * Sleep used between judge retries and after a judge succeeds. Defaults to
-     * real `setTimeout`. Injectable so tests exercise the retry policy without
-     * the real multi-second backoff.
-     */
-    sleep?: (ms: number) => Promise<void>;
-}
-/** Per-judge failure diagnostic, recorded additively on `ScenarioResult`. */
-export interface JudgeFailure {
-    judge: string;
-    attempts: number;
-    reason: string;
-}
-/**
- * Execute a scenario against an LLM via tcloud.
- *
- * Runs multi-turn conversation, extracts artifacts, runs judges.
- */
-export declare function executeScenario(tc: TCloud, scenario: Scenario, config: ExecutorConfig): Promise<ScenarioResult>;
-//# sourceMappingURL=executor.d.ts.map

package/dist/executor.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"executor.d.ts","sourceRoot":"","sources":["../src/executor.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,MAAM,EAAE,MAAM,wBAAwB,CAAA;AAKpD,OAAO,KAAK,EACV,kBAAkB,EAClB,OAAO,EAEP,QAAQ,EACR,cAAc,EAEf,MAAM,SAAS,CAAA;AAOhB,MAAM,WAAW,cAAc;IAC7B,6CAA6C;IAC7C,YAAY,EAAE,MAAM,CAAA;IACpB,iCAAiC;IACjC,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,oCAAoC;IACpC,MAAM,EAAE,OAAO,EAAE,CAAA;IACjB,+DAA+D;IAC/D,gBAAgB,CAAC,EAAE,MAAM,EAAE,CAAA;IAC3B,2DAA2D;IAC3D,YAAY,CAAC,EAAE,MAAM,CAAA;IACrB,yDAAyD;IACzD,eAAe,CAAC,EAAE,CAChB,KAAK,EAAE,QAAQ,CAAC,gBAAgB,CAAC,CAAC,CAAC,CAAC,EACpC,SAAS,EAAE,kBAAkB,KAC1B;QAAE,MAAM,EAAE,OAAO,CAAC;QAAC,MAAM,EAAE,MAAM,CAAA;KAAE,GAAG,IAAI,CAAA;IAC/C;;;;OAIG;IACH,KAAK,CAAC,EAAE,CAAC,EAAE,EAAE,MAAM,KAAK,OAAO,CAAC,IAAI,CAAC,CAAA;CACtC;AAYD,6EAA6E;AAC7E,MAAM,WAAW,YAAY;IAC3B,KAAK,EAAE,MAAM,CAAA;IACb,QAAQ,EAAE,MAAM,CAAA;IAChB,MAAM,EAAE,MAAM,CAAA;CACf;AAED;;;;GAIG;AACH,wBAAsB,eAAe,CACnC,EAAE,EAAE,MAAM,EACV,QAAQ,EAAE,QAAQ,EAClB,MAAM,EAAE,cAAc,GACrB,OAAO,CAAC,cAAc,CAAC,CAoOzB"}

package/dist/executor.test.d.ts

@@ -1,10 +0,0 @@
-/**
- * Capture-integrity contract for the scenario executor. Two failure classes
- * that a sloppy fallback erases must stay loud:
- *   1. a malformed/hung chat response must NOT be recorded as a real empty turn
- *      (capture defect, indistinguishable from a model that said nothing);
- *   2. a judge that errors must record WHY, retry only transient faults, and
- *      fail deterministic faults immediately.
- */
-export {};
-//# sourceMappingURL=executor.test.d.ts.map