@tangle-network/agent-eval 0.94.0 → 0.95.0

package/dist/auto-pr.d.ts

@@ -0,0 +1,120 @@
+/**
+ * Automated pull request opener for the improvement loop.
+ *
+ * When `runImprovementLoop` ships a winner (`autoOnPromote: 'pr'`) it produces
+ * a promoted surface diff. To close the eval → prod → eval cycle the framework
+ * lands that change as a reviewable code change. This module does exactly that:
+ *
+ *   1. Stage a branch off `baseBranch`.
+ *   2. Write each `fileChange` into the worktree.
+ *   3. Commit + push.
+ *   4. Open a PR via the GitHub API.
+ *
+ * Two transports ship in core:
+ *
+ *   - `ghCliClient(opts)` — shells out to the `gh` CLI. No extra deps,
+ *     re-uses the developer machine's `gh auth` state, works with both
+ *     github.com and GitHub Enterprise. This is the recommended default.
+ *   - `httpGithubClient(opts)` — direct `fetch` against `api.github.com`
+ *     with a bearer token. Useful in CI where `gh` may not be installed.
+ *
+ * Both implement the small `AutoPrClient` interface, so tests substitute
+ * a fake without spinning a process or network.
+ */
+export interface FileChange {
+    /** Repo-relative path. Forward slashes; no `..`. */
+    path: string;
+    /** New file contents. UTF-8. */
+    contents: string;
+    /** Optional explanatory comment shown in the commit body. */
+    rationale?: string;
+}
+export interface RepoRef {
+    owner: string;
+    name: string;
+}
+export interface ProposeAutomatedPullRequestInput {
+    repo: RepoRef;
+    /** Branch to base the PR on. Default `'main'`. */
+    baseBranch?: string;
+    /** New branch name. Use a prefix + a short stable id; no spaces. */
+    branchName: string;
+    fileChanges: FileChange[];
+    title: string;
+    body: string;
+    /** Optional GitHub usernames to request review from. */
+    reviewers?: string[];
+    /** Optional labels to apply. */
+    labels?: string[];
+    /** Commit author name. Default: derived from the GitHub client. */
+    authorName?: string;
+    /** Commit author email. Default: derived from the GitHub client. */
+    authorEmail?: string;
+    /** Dry-run — do not push or open a PR; just return the would-be plan. */
+    dryRun?: boolean;
+}
+export interface ProposeAutomatedPullRequestResult {
+    prUrl: string;
+    branchName: string;
+    headSha: string;
+    dryRun: boolean;
+}
+/** Pluggable transport for the auto-PR pipeline. */
+export interface AutoPrClient {
+    /**
+     * Create a branch from `baseBranch`, write file changes, commit, push,
+     * and open a PR. Returns the PR's HTML url and head SHA.
+     *
+     * Implementations must be idempotent on `branchName`: if the branch
+     * already exists with the same head SHA as the would-be commit, return
+     * the existing PR rather than failing. This makes the production loop
+     * safe to retry on transient errors.
+     */
+    proposeChange(input: ProposeAutomatedPullRequestInput): Promise<ProposeAutomatedPullRequestResult>;
+}
+export declare function proposeAutomatedPullRequest(client: AutoPrClient, input: ProposeAutomatedPullRequestInput): Promise<ProposeAutomatedPullRequestResult>;
+export interface HttpGithubClientOptions {
+    /** Personal access token, GitHub App token, or `GITHUB_TOKEN` from Actions. */
+    token: string;
+    /** Override for GitHub Enterprise. Default `'https://api.github.com'`. */
+    apiBase?: string;
+    /** Test seam — defaults to global `fetch`. */
+    fetchImpl?: typeof fetch;
+    /** Test seam — clock for commit timestamps. */
+    now?: () => Date;
+}
+/**
+ * Direct REST-API GitHub client. No external deps.
+ *
+ * Idempotency strategy: before creating refs/commits/PRs, check whether
+ * the branch already exists at the desired tree. If so, return the
+ * existing PR (or open one if missing). Errors from concurrent runs
+ * (`Reference already exists`) are caught and treated as success.
+ */
+export declare function httpGithubClient(opts: HttpGithubClientOptions): AutoPrClient;
+export interface GhCliClientOptions {
+    /** Override the CLI binary (`gh`). For testing. */
+    bin?: string;
+    /** Working directory containing a clone of `repo`. Default: process cwd. */
+    cwd?: string;
+    /** Test seam: process spawner. Default: node:child_process spawn. */
+    exec?: (bin: string, args: string[], opts: {
+        cwd: string;
+        stdin?: string;
+    }) => Promise<{
+        stdout: string;
+        stderr: string;
+        exitCode: number;
+    }>;
+}
+/**
+ * `gh` CLI transport. Requires:
+ *   - `gh` installed and authenticated (`gh auth status`).
+ *   - A local clone of the repo with a clean working tree.
+ *   - `git` on PATH.
+ *
+ * Uses `gh api` for repo metadata and `gh pr create` for the PR. The
+ * actual commit lands via `git`, which keeps `gh`'s footprint minimal.
+ */
+export declare function ghCliClient(opts?: GhCliClientOptions): AutoPrClient;
+//# sourceMappingURL=auto-pr.d.ts.map

package/dist/auto-pr.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"auto-pr.d.ts","sourceRoot":"","sources":["../src/auto-pr.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;GAsBG;AAIH,MAAM,WAAW,UAAU;IACzB,oDAAoD;IACpD,IAAI,EAAE,MAAM,CAAA;IACZ,gCAAgC;IAChC,QAAQ,EAAE,MAAM,CAAA;IAChB,6DAA6D;IAC7D,SAAS,CAAC,EAAE,MAAM,CAAA;CACnB;AAED,MAAM,WAAW,OAAO;IACtB,KAAK,EAAE,MAAM,CAAA;IACb,IAAI,EAAE,MAAM,CAAA;CACb;AAED,MAAM,WAAW,gCAAgC;IAC/C,IAAI,EAAE,OAAO,CAAA;IACb,kDAAkD;IAClD,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,oEAAoE;IACpE,UAAU,EAAE,MAAM,CAAA;IAClB,WAAW,EAAE,UAAU,EAAE,CAAA;IACzB,KAAK,EAAE,MAAM,CAAA;IACb,IAAI,EAAE,MAAM,CAAA;IACZ,wDAAwD;IACxD,SAAS,CAAC,EAAE,MAAM,EAAE,CAAA;IACpB,gCAAgC;IAChC,MAAM,CAAC,EAAE,MAAM,EAAE,CAAA;IACjB,mEAAmE;IACnE,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,oEAAoE;IACpE,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,yEAAyE;IACzE,MAAM,CAAC,EAAE,OAAO,CAAA;CACjB;AAED,MAAM,WAAW,iCAAiC;IAChD,KAAK,EAAE,MAAM,CAAA;IACb,UAAU,EAAE,MAAM,CAAA;IAClB,OAAO,EAAE,MAAM,CAAA;IACf,MAAM,EAAE,OAAO,CAAA;CAChB;AAED,oDAAoD;AACpD,MAAM,WAAW,YAAY;IAC3B;;;;;;;;OAQG;IACH,aAAa,CAAC,KAAK,EAAE,gCAAgC,GAAG,OAAO,CAAC,iCAAiC,CAAC,CAAA;CACnG;AAED,wBAAsB,2BAA2B,CAC/C,MAAM,EAAE,YAAY,EACpB,KAAK,EAAE,gCAAgC,GACtC,OAAO,CAAC,iCAAiC,CAAC,CAG5C;AAoCD,MAAM,WAAW,uBAAuB;IACtC,+EAA+E;IAC/E,KAAK,EAAE,MAAM,CAAA;IACb,0EAA0E;IAC1E,OAAO,CAAC,EAAE,MAAM,CAAA;IAChB,8CAA8C;IAC9C,SAAS,CAAC,EAAE,OAAO,KAAK,CAAA;IACxB,+CAA+C;IAC/C,GAAG,CAAC,EAAE,MAAM,IAAI,CAAA;CACjB;AAyBD;;;;;;;GAOG;AACH,wBAAgB,gBAAgB,CAAC,IAAI,EAAE,uBAAuB,GAAG,YAAY,CAoK5E;AAID,MAAM,WAAW,kBAAkB;IACjC,mDAAmD;IACnD,GAAG,CAAC,EAAE,MAAM,CAAA;IACZ,4EAA4E;IAC5E,GAAG,CAAC,EAAE,MAAM,CAAA;IACZ,qEAAqE;IACrE,IAAI,CAAC,EAAE,CACL,GAAG,EAAE,MAAM,EACX,IAAI,EAAE,MAAM,EAAE,EACd,IAAI,EAAE;QAAE,GAAG,EAAE,MAAM,CAAC;QAAC,KAAK,CAAC,EAAE,MAAM,CAAA;KAAE,KAClC,OAAO,CAAC;QAAE,MAAM,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAC;QAAC,QAAQ,EAAE,MAAM,CAAA;KAAE,CAAC,CAAA;CACnE;AAED;;;;;;;;GAQG;AACH,wBAAgB,WAAW,CAAC,IAAI,GAAE,kBAAuB,GAAG,YAAY,CAgHvE"}

package/dist/{baseline-Bbid3WoO.d.ts → baseline.d.ts}

@@ -1,38 +1,3 @@
-import { T as TraceStore } from './store-BcFXE6LG.js';
-
-/**
- * Tool-use metrics — derived purely from trace data.
- *
- * No scoring assumptions: consumers supply optional ground-truth tool
- * selections per turn + optional "information used downstream" signals.
- * Without those, we still compute descriptive metrics (error rate,
- * retry rate, duplicate-call rate) that are useful on their own.
- */
-
-interface ToolUseMetrics {
-    runId: string;
-    totalCalls: number;
-    byTool: Record<string, ToolStats>;
-    errorRate: number;
-    /** Ratio of calls with identical (toolName, argHash) already seen earlier in the same run. */
-    duplicateRate: number;
-    /** Ratio of error calls followed by ≥1 retry on same tool. */
-    retryRate: number;
-    /** Optional: of the calls agent made, fraction the evaluator marked as "correct selection". */
-    selectionAccuracy?: number;
-}
-interface ToolStats {
-    calls: number;
-    errors: number;
-    avgLatencyMs: number;
-    duplicates: number;
-}
-interface ToolUseOptions {
-    /** Map of spanId → whether the evaluator judged the tool selection correct. Optional. */
-    selectionLabels?: Record<string, boolean>;
-}
-declare function computeToolUseMetrics(store: TraceStore, runId: string, options?: ToolUseOptions): Promise<ToolUseMetrics>;
-
 /**
  * Baseline regression detection.
  *
@@ -47,7 +12,7 @@ declare function computeToolUseMetrics(store: TraceStore, runId: string, options
  *
  * Returns a structured verdict: improved | regressed | stable | unstable.
  */
-interface MetricSamples {
+export interface MetricSamples {
     /** Stable metric key (e.g. "overallScore", "firstTokenMs"). */
     metric: string;
     /** Whether higher values are better. */
@@ -55,7 +20,7 @@ interface MetricSamples {
     baseline: number[];
     candidate: number[];
 }
-interface MetricVerdict {
+export interface MetricVerdict {
     metric: string;
     baselineMean: number;
     candidateMean: number;
@@ -69,14 +34,14 @@ interface MetricVerdict {
     iqr: number;
     verdict: 'improved' | 'regressed' | 'stable' | 'unstable';
 }
-interface BaselineReport {
+export interface BaselineReport {
     metrics: MetricVerdict[];
     /** True if any critical metric regressed. */
     hasRegression: boolean;
     /** True if any metric is unstable (too noisy to judge). */
     hasUnstable: boolean;
 }
-interface BaselineOptions {
+export interface BaselineOptions {
     /** Effect size threshold for meaningful delta (default 0.5 — medium effect). */
     effectThreshold?: number;
     /** p-value threshold for statistical significance (default 0.05). */
@@ -91,18 +56,17 @@ interface BaselineOptions {
  *   - regressed: meaningful effect in the "worse" direction AND p < alpha
  *   - stable: otherwise (no significant change)
  */
-declare function compareToBaseline(samples: MetricSamples[], options?: BaselineOptions): BaselineReport;
+export declare function compareToBaseline(samples: MetricSamples[], options?: BaselineOptions): BaselineReport;
 /** Inter-quartile range; 0 when the sample has no spread. */
-declare function iqr(xs: number[]): number;
+export declare function iqr(xs: number[]): number;
 /**
  * Welch's t-test — unequal-variance two-sample t. Uses the same Student-t
  * CDF as `pairedTTest` (via incomplete beta); falls back to normal tail
  * when df is large.
  */
-declare function welchsTTest(a: number[], b: number[]): {
+export declare function welchsTTest(a: number[], b: number[]): {
     t: number;
     df: number;
     p: number;
 };
-
-export { type BaselineOptions as B, type MetricSamples as M, type ToolStats as T, type BaselineReport as a, type MetricVerdict as b, computeToolUseMetrics as c, type ToolUseMetrics as d, type ToolUseOptions as e, compareToBaseline as f, iqr as i, welchsTTest as w };
+//# sourceMappingURL=baseline.d.ts.map

package/dist/baseline.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"baseline.d.ts","sourceRoot":"","sources":["../src/baseline.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;GAaG;AAIH,MAAM,WAAW,aAAa;IAC5B,+DAA+D;IAC/D,MAAM,EAAE,MAAM,CAAA;IACd,wCAAwC;IACxC,cAAc,EAAE,OAAO,CAAA;IACvB,QAAQ,EAAE,MAAM,EAAE,CAAA;IAClB,SAAS,EAAE,MAAM,EAAE,CAAA;CACpB;AAED,MAAM,WAAW,aAAa;IAC5B,MAAM,EAAE,MAAM,CAAA;IACd,YAAY,EAAE,MAAM,CAAA;IACpB,aAAa,EAAE,MAAM,CAAA;IACrB,KAAK,EAAE,MAAM,CAAA;IACb,OAAO,EAAE,MAAM,CAAA;IACf,MAAM,EAAE,MAAM,CAAA;IACd,OAAO,EAAE,MAAM,CAAA;IACf,MAAM,EAAE,MAAM,CAAA;IACd,MAAM,EAAE,OAAO,CAAA;IACf,yEAAyE;IACzE,GAAG,EAAE,MAAM,CAAA;IACX,OAAO,EAAE,UAAU,GAAG,WAAW,GAAG,QAAQ,GAAG,UAAU,CAAA;CAC1D;AAED,MAAM,WAAW,cAAc;IAC7B,OAAO,EAAE,aAAa,EAAE,CAAA;IACxB,6CAA6C;IAC7C,aAAa,EAAE,OAAO,CAAA;IACtB,2DAA2D;IAC3D,WAAW,EAAE,OAAO,CAAA;CACrB;AAED,MAAM,WAAW,eAAe;IAC9B,gFAAgF;IAChF,eAAe,CAAC,EAAE,MAAM,CAAA;IACxB,qEAAqE;IACrE,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,8EAA8E;IAC9E,mBAAmB,CAAC,EAAE,MAAM,CAAA;CAC7B;AAED;;;;;;GAMG;AACH,wBAAgB,iBAAiB,CAC/B,OAAO,EAAE,aAAa,EAAE,EACxB,OAAO,GAAE,eAAoB,GAC5B,cAAc,CAuDhB;AAMD,6DAA6D;AAC7D,wBAAgB,GAAG,CAAC,EAAE,EAAE,MAAM,EAAE,GAAG,MAAM,CAUxC;AAED;;;;GAIG;AACH,wBAAgB,WAAW,CAAC,CAAC,EAAE,MAAM,EAAE,EAAE,CAAC,EAAE,MAAM,EAAE,GAAG;IAAE,CAAC,EAAE,MAAM,CAAC;IAAC,EAAE,EAAE,MAAM,CAAC;IAAC,CAAC,EAAE,MAAM,CAAA;CAAE,CAc1F"}

package/dist/behavior-dsl.d.ts

@@ -0,0 +1,73 @@
+/**
+ * Behavior DSL — pytest-style assertions over a run's trajectory.
+ *
+ * Shape:
+ *   expect(store, runId).toCall('search').withArgs({ q: /.+/ })
+ *   expect(store, runId).toRefuse()
+ *   expect(store, runId).toOutputMatch(/confirmed/i)
+ *   expect(store, runId).toRespectBudget('tokens')
+ *   expect(store, runId).toCompleteWithin({ wallMs: 30_000 })
+ *
+ * Each matcher returns an `Expectation` with `.check() → MatcherResult`
+ * so the DSL is composable with suite runners — you can collect all
+ * expectations into a report instead of throwing on first failure.
+ */
+import type { BudgetLedgerEntry, Span, ToolSpan } from './trace/schema';
+import type { TraceStore } from './trace/store';
+import { type TraceContract } from './trace-contracts';
+export interface MatcherResult {
+    ok: boolean;
+    detail: string;
+    evidence?: string;
+}
+export interface Expectation {
+    /** Human-facing label; used in reports. */
+    label: string;
+    check(): Promise<MatcherResult>;
+}
+export declare class BehaviorAssertion {
+    private store;
+    private runId;
+    constructor(store: TraceStore, runId: string);
+    toCall(toolName: string): CallExpectation;
+    toRefuse(markers?: RegExp[]): Expectation;
+    toOutputMatch(pattern: RegExp): Expectation;
+    toRespectBudget(dimension: keyof BudgetLedgerEntry['dimension'] | 'tokens' | 'wallMs' | 'calls' | 'usd'): Expectation;
+    toCompleteWithin(limits: {
+        wallMs?: number;
+        toolCalls?: number;
+        llmTurns?: number;
+    }): Expectation;
+    /** Evaluate a finite-trace temporal contract (`traceContract(...)`) over
+     *  this run's span sequence. See `trace-contracts.ts` for the operators. */
+    toSatisfyContract(contract: TraceContract): Expectation;
+    toNeverCall(toolName: string): Expectation;
+}
+export declare class CallExpectation implements Expectation {
+    private store;
+    private runId;
+    private toolName;
+    private argMatchers;
+    private minCount;
+    private maxCount;
+    constructor(store: TraceStore, runId: string, toolName: string);
+    get label(): string;
+    withArgs(shape: Record<string, unknown | RegExp>): this;
+    times(n: number): this;
+    atLeast(n: number): this;
+    atMost(n: number): this;
+    check(): Promise<MatcherResult>;
+}
+export declare function expectAgent(store: TraceStore, runId: string): BehaviorAssertion;
+/** Runs every expectation, collects results. Never throws. */
+export declare function runExpectations(expectations: Expectation[]): Promise<{
+    results: Array<{
+        label: string;
+        result: MatcherResult;
+    }>;
+    pass: boolean;
+    passCount: number;
+    failCount: number;
+}>;
+export type { Span, ToolSpan };
+//# sourceMappingURL=behavior-dsl.d.ts.map

package/dist/behavior-dsl.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"behavior-dsl.d.ts","sourceRoot":"","sources":["../src/behavior-dsl.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;GAaG;AAGH,OAAO,KAAK,EAAE,iBAAiB,EAAE,IAAI,EAAE,QAAQ,EAAE,MAAM,gBAAgB,CAAA;AACvE,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,eAAe,CAAA;AAC/C,OAAO,EAAyB,KAAK,aAAa,EAAE,MAAM,mBAAmB,CAAA;AAE7E,MAAM,WAAW,aAAa;IAC5B,EAAE,EAAE,OAAO,CAAA;IACX,MAAM,EAAE,MAAM,CAAA;IACd,QAAQ,CAAC,EAAE,MAAM,CAAA;CAClB;AAED,MAAM,WAAW,WAAW;IAC1B,2CAA2C;IAC3C,KAAK,EAAE,MAAM,CAAA;IACb,KAAK,IAAI,OAAO,CAAC,aAAa,CAAC,CAAA;CAChC;AAED,qBAAa,iBAAiB;IAE1B,OAAO,CAAC,KAAK;IACb,OAAO,CAAC,KAAK;gBADL,KAAK,EAAE,UAAU,EACjB,KAAK,EAAE,MAAM;IAGvB,MAAM,CAAC,QAAQ,EAAE,MAAM,GAAG,eAAe;IAIzC,QAAQ,CAAC,OAAO,GAAE,MAAM,EAA4B,GAAG,WAAW;IAkBlE,aAAa,CAAC,OAAO,EAAE,MAAM,GAAG,WAAW;IAc3C,eAAe,CACb,SAAS,EAAE,MAAM,iBAAiB,CAAC,WAAW,CAAC,GAAG,QAAQ,GAAG,QAAQ,GAAG,OAAO,GAAG,KAAK,GACtF,WAAW;IAad,gBAAgB,CAAC,MAAM,EAAE;QACvB,MAAM,CAAC,EAAE,MAAM,CAAA;QACf,SAAS,CAAC,EAAE,MAAM,CAAA;QAClB,QAAQ,CAAC,EAAE,MAAM,CAAA;KAClB,GAAG,WAAW;IAuBf;gFAC4E;IAC5E,iBAAiB,CAAC,QAAQ,EAAE,aAAa,GAAG,WAAW;IAkBvD,WAAW,CAAC,QAAQ,EAAE,MAAM,GAAG,WAAW;CAe3C;AAED,qBAAa,eAAgB,YAAW,WAAW;IAM/C,OAAO,CAAC,KAAK;IACb,OAAO,CAAC,KAAK;IACb,OAAO,CAAC,QAAQ;IAPlB,OAAO,CAAC,WAAW,CAAwC;IAC3D,OAAO,CAAC,QAAQ,CAAI;IACpB,OAAO,CAAC,QAAQ,CAAW;gBAGjB,KAAK,EAAE,UAAU,EACjB,KAAK,EAAE,MAAM,EACb,QAAQ,EAAE,MAAM;IAG1B,IAAI,KAAK,IAAI,MAAM,CAElB;IAED,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,GAAG,MAAM,CAAC,GAAG,IAAI;IAKvD,KAAK,CAAC,CAAC,EAAE,MAAM,GAAG,IAAI;IAMtB,OAAO,CAAC,CAAC,EAAE,MAAM,GAAG,IAAI;IAKxB,MAAM,CAAC,CAAC,EAAE,MAAM,GAAG,IAAI;IAKjB,KAAK,IAAI,OAAO,CAAC,aAAa,CAAC;CAgBtC;AAED,wBAAgB,WAAW,CAAC,KAAK,EAAE,UAAU,EAAE,KAAK,EAAE,MAAM,GAAG,iBAAiB,CAE/E;AAED,8DAA8D;AAC9D,wBAAsB,eAAe,CAAC,YAAY,EAAE,WAAW,EAAE,GAAG,OAAO,CAAC;IAC1E,OAAO,EAAE,KAAK,CAAC;QAAE,KAAK,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,aAAa,CAAA;KAAE,CAAC,CAAA;IACxD,IAAI,EAAE,OAAO,CAAA;IACb,SAAS,EAAE,MAAM,CAAA;IACjB,SAAS,EAAE,MAAM,CAAA;CAClB,CAAC,CAWD;AAwBD,YAAY,EAAE,IAAI,EAAE,QAAQ,EAAE,CAAA"}

package/dist/belief-state/calibration.d.ts

@@ -0,0 +1,10 @@
+import type { BeliefDecisionPoint, BeliefSelectivePolicy } from './types';
+export type BeliefCalibrationRegion = 'all' | 'accepted' | 'rejected';
+export interface BeliefCalibrationOptions {
+    bins?: number;
+    minPairs?: number;
+    policy?: BeliefSelectivePolicy;
+    region?: BeliefCalibrationRegion;
+}
+export declare function calibrateBeliefDecisions(points: BeliefDecisionPoint[], options?: BeliefCalibrationOptions): import("../meta-eval").CalibrationReport | null;
+//# sourceMappingURL=calibration.d.ts.map

package/dist/belief-state/calibration.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"calibration.d.ts","sourceRoot":"","sources":["../../src/belief-state/calibration.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,mBAAmB,EAAE,qBAAqB,EAAE,MAAM,SAAS,CAAA;AAEzE,MAAM,MAAM,uBAAuB,GAAG,KAAK,GAAG,UAAU,GAAG,UAAU,CAAA;AAErE,MAAM,WAAW,wBAAwB;IACvC,IAAI,CAAC,EAAE,MAAM,CAAA;IACb,QAAQ,CAAC,EAAE,MAAM,CAAA;IACjB,MAAM,CAAC,EAAE,qBAAqB,CAAA;IAC9B,MAAM,CAAC,EAAE,uBAAuB,CAAA;CACjC;AAED,wBAAgB,wBAAwB,CACtC,MAAM,EAAE,mBAAmB,EAAE,EAC7B,OAAO,GAAE,wBAA6B,mDAgBvC"}

package/dist/belief-state/calibration.test.d.ts

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

package/dist/belief-state/calibration.test.d.ts.map

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

package/dist/belief-state/code-agent-corpus.d.ts

@@ -0,0 +1,66 @@
+import type { CodeAgentSessionSource } from '../contract/intake/code-agent-session';
+import type { RunRecord } from '../run-record';
+import { type AnalyzeBeliefPolicyOptions } from './report';
+import type { BeliefDecisionExtractionReport, BeliefDecisionKind, BeliefDecisionPoint, BeliefPolicyEvaluationReport, BeliefSelectivePolicy } from './types';
+export type CodeAgentBeliefDecisionTargetId = 'failure-recovery' | 'tool-selection' | 'graph-completion';
+export interface ExtractCodeAgentBeliefDecisionPointsOptions {
+    source: CodeAgentSessionSource;
+    entries: unknown[];
+    run: Pick<RunRecord, 'runId' | 'scenarioId' | 'outcome' | 'costUsd'>;
+    sourcePath?: string;
+}
+export interface BeliefDecisionInventoryBucket {
+    id: string;
+    kind?: BeliefDecisionKind;
+    targetId?: CodeAgentBeliefDecisionTargetId;
+    n: number;
+    withOutcome: number;
+    withConfidence: number;
+    withCandidateActions: number;
+    withBehaviorProb: number;
+    withTargetProb: number;
+    successRate: number | null;
+    meanScore: number | null;
+    meanConfidence: number | null;
+}
+export interface BeliefDecisionInventoryReport {
+    n: number;
+    byKind: BeliefDecisionInventoryBucket[];
+    byTarget: BeliefDecisionInventoryBucket[];
+    diagnostics: string[];
+}
+export interface BeliefDecisionTargetSelection {
+    id: CodeAgentBeliefDecisionTargetId;
+    label: string;
+    points: BeliefDecisionPoint[];
+    support: BeliefDecisionInventoryBucket;
+    reasons: string[];
+}
+export interface SelectBeliefDecisionTargetOptions {
+    minN?: number;
+    minOutcomeCoverage?: number;
+    preferredTargets?: CodeAgentBeliefDecisionTargetId[];
+}
+export interface AnalyzeBeliefDecisionCorpusOptions {
+    points: BeliefDecisionPoint[];
+    targetId?: CodeAgentBeliefDecisionTargetId;
+    minN?: number;
+    minOutcomeCoverage?: number;
+    minAccepted?: number;
+    confidenceThreshold?: number;
+    policy?: BeliefSelectivePolicy;
+    requireOpe?: boolean;
+    policyOptions?: Partial<AnalyzeBeliefPolicyOptions>;
+}
+export interface BeliefDecisionCorpusEvaluation {
+    inventory: BeliefDecisionInventoryReport;
+    target?: BeliefDecisionTargetSelection;
+    policy?: BeliefSelectivePolicy;
+    evaluation?: BeliefPolicyEvaluationReport;
+    diagnostics: string[];
+}
+export declare function extractCodeAgentBeliefDecisionPoints(options: ExtractCodeAgentBeliefDecisionPointsOptions): BeliefDecisionExtractionReport;
+export declare function inventoryBeliefDecisionPoints(points: BeliefDecisionPoint[]): BeliefDecisionInventoryReport;
+export declare function selectBeliefDecisionTarget(points: BeliefDecisionPoint[], options?: SelectBeliefDecisionTargetOptions): BeliefDecisionTargetSelection | null;
+export declare function analyzeBeliefDecisionCorpus(options: AnalyzeBeliefDecisionCorpusOptions): BeliefDecisionCorpusEvaluation;
+//# sourceMappingURL=code-agent-corpus.d.ts.map

package/dist/belief-state/code-agent-corpus.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"code-agent-corpus.d.ts","sourceRoot":"","sources":["../../src/belief-state/code-agent-corpus.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,sBAAsB,EAAE,MAAM,uCAAuC,CAAA;AACnF,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,eAAe,CAAA;AAE9C,OAAO,EAAE,KAAK,0BAA0B,EAAuB,MAAM,UAAU,CAAA;AAE/E,OAAO,KAAK,EAEV,8BAA8B,EAC9B,kBAAkB,EAElB,mBAAmB,EAEnB,4BAA4B,EAC5B,qBAAqB,EACtB,MAAM,SAAS,CAAA;AAEhB,MAAM,MAAM,+BAA+B,GACvC,kBAAkB,GAClB,gBAAgB,GAChB,kBAAkB,CAAA;AAEtB,MAAM,WAAW,2CAA2C;IAC1D,MAAM,EAAE,sBAAsB,CAAA;IAC9B,OAAO,EAAE,OAAO,EAAE,CAAA;IAClB,GAAG,EAAE,IAAI,CAAC,SAAS,EAAE,OAAO,GAAG,YAAY,GAAG,SAAS,GAAG,SAAS,CAAC,CAAA;IACpE,UAAU,CAAC,EAAE,MAAM,CAAA;CACpB;AAED,MAAM,WAAW,6BAA6B;IAC5C,EAAE,EAAE,MAAM,CAAA;IACV,IAAI,CAAC,EAAE,kBAAkB,CAAA;IACzB,QAAQ,CAAC,EAAE,+BAA+B,CAAA;IAC1C,CAAC,EAAE,MAAM,CAAA;IACT,WAAW,EAAE,MAAM,CAAA;IACnB,cAAc,EAAE,MAAM,CAAA;IACtB,oBAAoB,EAAE,MAAM,CAAA;IAC5B,gBAAgB,EAAE,MAAM,CAAA;IACxB,cAAc,EAAE,MAAM,CAAA;IACtB,WAAW,EAAE,MAAM,GAAG,IAAI,CAAA;IAC1B,SAAS,EAAE,MAAM,GAAG,IAAI,CAAA;IACxB,cAAc,EAAE,MAAM,GAAG,IAAI,CAAA;CAC9B;AAED,MAAM,WAAW,6BAA6B;IAC5C,CAAC,EAAE,MAAM,CAAA;IACT,MAAM,EAAE,6BAA6B,EAAE,CAAA;IACvC,QAAQ,EAAE,6BAA6B,EAAE,CAAA;IACzC,WAAW,EAAE,MAAM,EAAE,CAAA;CACtB;AAED,MAAM,WAAW,6BAA6B;IAC5C,EAAE,EAAE,+BAA+B,CAAA;IACnC,KAAK,EAAE,MAAM,CAAA;IACb,MAAM,EAAE,mBAAmB,EAAE,CAAA;IAC7B,OAAO,EAAE,6BAA6B,CAAA;IACtC,OAAO,EAAE,MAAM,EAAE,CAAA;CAClB;AAED,MAAM,WAAW,iCAAiC;IAChD,IAAI,CAAC,EAAE,MAAM,CAAA;IACb,kBAAkB,CAAC,EAAE,MAAM,CAAA;IAC3B,gBAAgB,CAAC,EAAE,+BAA+B,EAAE,CAAA;CACrD;AAED,MAAM,WAAW,kCAAkC;IACjD,MAAM,EAAE,mBAAmB,EAAE,CAAA;IAC7B,QAAQ,CAAC,EAAE,+BAA+B,CAAA;IAC1C,IAAI,CAAC,EAAE,MAAM,CAAA;IACb,kBAAkB,CAAC,EAAE,MAAM,CAAA;IAC3B,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,mBAAmB,CAAC,EAAE,MAAM,CAAA;IAC5B,MAAM,CAAC,EAAE,qBAAqB,CAAA;IAC9B,UAAU,CAAC,EAAE,OAAO,CAAA;IACpB,aAAa,CAAC,EAAE,OAAO,CAAC,0BAA0B,CAAC,CAAA;CACpD;AAED,MAAM,WAAW,8BAA8B;IAC7C,SAAS,EAAE,6BAA6B,CAAA;IACxC,MAAM,CAAC,EAAE,6BAA6B,CAAA;IACtC,MAAM,CAAC,EAAE,qBAAqB,CAAA;IAC9B,UAAU,CAAC,EAAE,4BAA4B,CAAA;IACzC,WAAW,EAAE,MAAM,EAAE,CAAA;CACtB;AAyBD,wBAAgB,oCAAoC,CAClD,OAAO,EAAE,2CAA2C,GACnD,8BAA8B,CA0ChC;AAED,wBAAgB,6BAA6B,CAC3C,MAAM,EAAE,mBAAmB,EAAE,GAC5B,6BAA6B,CAyB/B;AAED,wBAAgB,0BAA0B,CACxC,MAAM,EAAE,mBAAmB,EAAE,EAC7B,OAAO,GAAE,iCAAsC,GAC9C,6BAA6B,GAAG,IAAI,CAiCtC;AAED,wBAAgB,2BAA2B,CACzC,OAAO,EAAE,kCAAkC,GAC1C,8BAA8B,CA6ChC"}

package/dist/belief-state/code-agent-corpus.test.d.ts

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

package/dist/belief-state/code-agent-corpus.test.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"code-agent-corpus.test.d.ts","sourceRoot":"","sources":["../../src/belief-state/code-agent-corpus.test.ts"],"names":[],"mappings":""}

package/dist/belief-state/code-agent-evidence.d.ts

@@ -0,0 +1,22 @@
+import { type CodeAgentSessionDiagnostic, type CodeAgentSessionIntakeOptions, type CodeAgentSessionMetrics, type CodeAgentSessionSource } from '../contract/intake/code-agent-session';
+import type { RunRecord } from '../run-record';
+import { type BeliefDecisionInventoryReport } from './code-agent-corpus';
+import { type BeliefDecisionResearchEvidencePacket, type BuildBeliefDecisionResearchEvidencePacketOptions } from './research-evidence';
+import type { BeliefDecisionExtractionDiagnostic, BeliefDecisionPoint } from './types';
+export interface CodeAgentBeliefSession extends CodeAgentSessionIntakeOptions {
+    source: CodeAgentSessionSource;
+}
+export interface BuildCodeAgentBeliefEvidenceCorpusOptions extends Omit<BuildBeliefDecisionResearchEvidencePacketOptions, 'points'> {
+    sessions: CodeAgentBeliefSession[];
+}
+export interface CodeAgentBeliefEvidenceCorpus {
+    runs: RunRecord[];
+    metrics: CodeAgentSessionMetrics[];
+    intakeDiagnostics: CodeAgentSessionDiagnostic[];
+    extractionDiagnostics: BeliefDecisionExtractionDiagnostic[];
+    decisions: BeliefDecisionPoint[];
+    inventory: BeliefDecisionInventoryReport;
+    evidence: BeliefDecisionResearchEvidencePacket;
+}
+export declare function buildCodeAgentBeliefEvidenceCorpus(options: BuildCodeAgentBeliefEvidenceCorpusOptions): CodeAgentBeliefEvidenceCorpus;
+//# sourceMappingURL=code-agent-evidence.d.ts.map

package/dist/belief-state/code-agent-evidence.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"code-agent-evidence.d.ts","sourceRoot":"","sources":["../../src/belief-state/code-agent-evidence.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,KAAK,0BAA0B,EAC/B,KAAK,6BAA6B,EAElC,KAAK,uBAAuB,EAC5B,KAAK,sBAAsB,EAM5B,MAAM,uCAAuC,CAAA;AAC9C,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,eAAe,CAAA;AAC9C,OAAO,EACL,KAAK,6BAA6B,EAGnC,MAAM,qBAAqB,CAAA;AAC5B,OAAO,EACL,KAAK,oCAAoC,EACzC,KAAK,gDAAgD,EAEtD,MAAM,qBAAqB,CAAA;AAC5B,OAAO,KAAK,EAAE,kCAAkC,EAAE,mBAAmB,EAAE,MAAM,SAAS,CAAA;AAEtF,MAAM,WAAW,sBAAuB,SAAQ,6BAA6B;IAC3E,MAAM,EAAE,sBAAsB,CAAA;CAC/B;AAED,MAAM,WAAW,yCACf,SAAQ,IAAI,CAAC,gDAAgD,EAAE,QAAQ,CAAC;IACxE,QAAQ,EAAE,sBAAsB,EAAE,CAAA;CACnC;AAED,MAAM,WAAW,6BAA6B;IAC5C,IAAI,EAAE,SAAS,EAAE,CAAA;IACjB,OAAO,EAAE,uBAAuB,EAAE,CAAA;IAClC,iBAAiB,EAAE,0BAA0B,EAAE,CAAA;IAC/C,qBAAqB,EAAE,kCAAkC,EAAE,CAAA;IAC3D,SAAS,EAAE,mBAAmB,EAAE,CAAA;IAChC,SAAS,EAAE,6BAA6B,CAAA;IACxC,QAAQ,EAAE,oCAAoC,CAAA;CAC/C;AAED,wBAAgB,kCAAkC,CAChD,OAAO,EAAE,yCAAyC,GACjD,6BAA6B,CAwC/B"}

package/dist/belief-state/code-agent-evidence.test.d.ts

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

package/dist/belief-state/code-agent-evidence.test.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"code-agent-evidence.test.d.ts","sourceRoot":"","sources":["../../src/belief-state/code-agent-evidence.test.ts"],"names":[],"mappings":""}

package/dist/belief-state/extract.d.ts

@@ -0,0 +1,7 @@
+import type { TraceStore } from '../trace/store';
+import type { BeliefDecisionExtractionReport } from './types';
+export interface ExtractBeliefDecisionPointsOptions {
+    runIds?: string[];
+}
+export declare function extractBeliefDecisionPoints(store: TraceStore, options?: ExtractBeliefDecisionPointsOptions): Promise<BeliefDecisionExtractionReport>;
+//# sourceMappingURL=extract.d.ts.map

package/dist/belief-state/extract.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"extract.d.ts","sourceRoot":"","sources":["../../src/belief-state/extract.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,gBAAgB,CAAA;AAChD,OAAO,KAAK,EAEV,8BAA8B,EAI/B,MAAM,SAAS,CAAA;AAGhB,MAAM,WAAW,kCAAkC;IACjD,MAAM,CAAC,EAAE,MAAM,EAAE,CAAA;CAClB;AAID,wBAAsB,2BAA2B,CAC/C,KAAK,EAAE,UAAU,EACjB,OAAO,GAAE,kCAAuC,GAC/C,OAAO,CAAC,8BAA8B,CAAC,CA8BzC"}

package/dist/belief-state/extract.test.d.ts

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

package/dist/belief-state/extract.test.d.ts.map

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