@rulvar/evals 1.0.0

package/LICENSE

@@ -0,0 +1,202 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

package/dist/index.d.ts

@@ -0,0 +1,335 @@
+import { CompiledWorkflow, Effort, Engine, EvidenceRef, Json, JsonSchema, ModelClaim, ModelKnowledgeStore, ModelRef, ModelSpec, RunOutcome, SchemaSpec, TaskClass, Usage, WireError, Workflow } from "@rulvar/core";
+
+//#region src/case.d.ts
+/**
+* One quality-measurement case (docs/09, section 7.1). The shape is the
+* documented interface verbatim; display names derive from the workflow
+* name (the suite runner disambiguates duplicates by ordinal).
+*/
+interface EvalCase {
+  workflow: Workflow | CompiledWorkflow;
+  args: Json;
+  graders: Grader[];
+}
+/** One grader's outcome for one case. */
+interface GraderVerdict {
+  /** The grader's display name. */
+  grader: string;
+  passed: boolean;
+  /** 0..1 where the family defines a fraction (rubric criteria met). */
+  score?: number;
+  /** Family-specific evidence: diffs, per-criterion verdicts, judge output. */
+  details?: Json;
+}
+/**
+* A judge invocation specification. The judge runs through the engine as
+* an ordinary journaled, budgeted invocation; model selection is subject
+* to the router quality floors (docs/04, section "Role quality floors"),
+* and @rulvar/evals ships NO default judge model: weak defaults for
+* judging are forbidden, so the model is always explicit.
+*/
+interface JudgeSpec {
+  model: ModelSpec;
+  prompt: string;
+  schema: SchemaSpec;
+}
+/** What a grader sees; judge() is the only channel back into the engine. */
+interface GraderContext {
+  /** The target run's structured output (RunOutcome.value). */
+  value: Json | undefined;
+  /** The full target outcome, for status- and cost-aware graders. */
+  outcome: RunOutcome<Json>;
+  /**
+  * Runs one judge invocation through the engine (journaled, budgeted,
+  * VCR-recordable) and returns the judge's structured output. Throws
+  * when the judge run itself does not settle ok.
+  */
+  judge(spec: JudgeSpec): Promise<Json>;
+}
+interface Grader {
+  name: string;
+  grade(context: GraderContext): Promise<GraderVerdict> | GraderVerdict;
+}
+/** The measured result of one EvalCase. */
+interface EvalCaseResult {
+  /** Workflow name, disambiguated by the suite runner on duplicates. */
+  name: string;
+  /** The target run's settle status. */
+  status: RunOutcome<Json>["status"];
+  /** status 'ok' AND every grader passed. */
+  passed: boolean;
+  verdicts: GraderVerdict[];
+  /** Target run cost plus all judge run costs (CostReport.totalUsd sums). */
+  costUsd: number;
+  /** The judge-run share of costUsd. */
+  judgeCostUsd: number;
+  /**
+  * run:start to run:end of the target run, from event timestamps; no
+  * separate measurement channel exists (docs/09, section 7.2).
+  */
+  latencyMs: number;
+  /** The target run's normalized usage. */
+  usage: Usage;
+  error?: WireError;
+}
+interface RunEvalCaseOptions {
+  /** Display-name override; defaults to the workflow name. */
+  name?: string;
+  /** Run ceiling for the target run. */
+  budgetUsd?: number;
+  /** Run ceiling for each judge run. */
+  judgeBudgetUsd?: number;
+}
+/** Thrown when a judge run does not settle ok. */
+declare class EvalJudgeError extends Error {
+  readonly judgeRun: string;
+  readonly status: RunOutcome<Json>["status"];
+  constructor(judgeRun: string, status: RunOutcome<Json>["status"], detail?: string);
+}
+/**
+* Runs one EvalCase on the given engine: the target workflow as its own
+* run, pure graders host-side over the outcome, judge graders through the
+* engine via GraderContext.judge. Grader thrown errors are not absorbed:
+* a grader that cannot grade is a defect of the suite, not a failed case.
+*/
+declare function runEvalCase(engine: Engine, evalCase: EvalCase, options?: RunEvalCaseOptions): Promise<EvalCaseResult>;
+/** Aggregate view of a suite run. */
+interface EvalSuiteResult {
+  results: EvalCaseResult[];
+  /** Fraction of cases with passed true; 0 for an empty suite. */
+  passRate: number;
+  totalCostUsd: number;
+  /** Arithmetic mean over cases; 0 for an empty suite. */
+  meanLatencyMs: number;
+}
+interface RunEvalSuiteOptions {
+  budgetUsd?: number;
+  judgeBudgetUsd?: number;
+}
+/**
+* Runs cases sequentially (deterministic journal and cassette order) and
+* aggregates. Duplicate workflow names get '#<ordinal>' suffixes so every
+* result row and judge journal is unambiguous.
+*/
+declare function runEvalSuite(engine: Engine, cases: EvalCase[], options?: RunEvalSuiteOptions): Promise<EvalSuiteResult>;
+//#endregion
+//#region src/matrix.d.ts
+/** One configuration under comparison. */
+interface MatrixCell {
+  name: string;
+  /** A fresh engine per cell run keeps cells isolated. */
+  engine: () => Engine | Promise<Engine>;
+}
+interface MatrixCellReport {
+  cell: string;
+  passRate: number;
+  totalCostUsd: number;
+  meanLatencyMs: number;
+  results: EvalCaseResult[];
+}
+interface EvalMatrixReport {
+  cells: MatrixCellReport[];
+}
+/**
+* Runs the same case list against every cell's engine, sequentially and
+* in declaration order (deterministic cassette consumption), and reports
+* per-cell aggregates for side-by-side comparison.
+*/
+declare function runEvalMatrix(cells: MatrixCell[], cases: EvalCase[], options?: RunEvalSuiteOptions): Promise<EvalMatrixReport>;
+//#endregion
+//#region src/graders/golden.d.ts
+interface GoldenGraderOptions {
+  name?: string;
+}
+declare function goldenGrader(expected: Json, options?: GoldenGraderOptions): Grader;
+//#endregion
+//#region src/graders/rubric.d.ts
+interface RubricCriterion {
+  name: string;
+  check: (value: Json | undefined) => boolean;
+}
+interface RubricGraderOptions {
+  name?: string;
+  /**
+  * Minimum fraction of criteria that must pass; default 1 (all).
+  * The fraction is also reported as the verdict score.
+  */
+  passThreshold?: number;
+}
+declare function rubricGrader(criteria: RubricCriterion[], options?: RubricGraderOptions): Grader;
+//#endregion
+//#region src/graders/judge.d.ts
+/** The default judge verdict shape. */
+declare const JUDGE_VERDICT_SCHEMA: JsonSchema;
+interface JudgeGraderOptions {
+  /** Judge model; required, never defaulted (docs/04 role quality floors). */
+  model: ModelSpec;
+  /** What to judge: the criteria prose embedded into the judge prompt. */
+  instruction: string;
+  name?: string;
+  /**
+  * Custom verdict schema; requires toVerdict. The default schema is
+  * JUDGE_VERDICT_SCHEMA with its boolean `passed`.
+  */
+  schema?: JsonSchema;
+  /** Maps the judge's structured output onto a pass/score pair. */
+  toVerdict?: (output: Json) => {
+    passed: boolean;
+    score?: number;
+  };
+}
+declare function judgeGrader(options: JudgeGraderOptions): Grader;
+//#endregion
+//#region src/committer.d.ts
+interface MeasuredClaimInput {
+  /** ULID (or any unique id); the caller mints it deterministically. */
+  id: string;
+  subject: {
+    model: ModelRef;
+    effort?: Effort;
+  };
+  taskClass: TaskClass;
+  polarity: "strength" | "weakness";
+  /** A typed template render, never a quote from tool output. */
+  statement: string;
+  metrics: {
+    passRate: number;
+    n: number;
+    graderId: string;
+    cost?: number;
+    baseline?: {
+      model: ModelRef;
+      passRate: number;
+    };
+  };
+  confidence: "high" | "medium" | "low";
+  /** ISO date of the sweep run. */
+  observedAt: string;
+  evidence: EvidenceRef[];
+  modelEpoch?: ModelClaim["modelEpoch"];
+}
+interface EvalCommitterOptions {
+  /** The dedicated identity recorded on the gate AND the author. */
+  committerId: string;
+  /** The emitting sweep report; every claim's gate references it. */
+  reportId: string;
+  /** CAS-rebase attempts (docs/05, 5.4); default 3. */
+  attempts?: number;
+}
+/** One measured claim, TTL applied per the docs/05 decay table. */
+declare function evalMeasuredClaim(input: MeasuredClaimInput, committerId: string): ModelClaim;
+/**
+* Commits measured claims through the eval-committer gate with the
+* documented rebase recipe: on a CAS rejection, re-read current() and
+* retry against the fresh version. Returns the committed version.
+*/
+declare function commitEvalMeasured(store: ModelKnowledgeStore, claims: readonly MeasuredClaimInput[], options: EvalCommitterOptions): Promise<number>;
+//#endregion
+//#region src/canary.d.ts
+interface CanaryProbeSet {
+  /** Registered agent profile the probes run under. */
+  agentType: string;
+  /** The fixed prompts; order matters and enters the fingerprint. */
+  prompts: string[];
+}
+/** The committed v1 normalization (OQ-06): NFC, trim, collapse whitespace. */
+declare function normalizeCanaryOutput(output: unknown): string;
+/**
+* Runs the fixed probe set through the ordinary engine and returns the
+* fingerprint. Probes run sequentially in declaration order, one run
+* per probe, so recordings replay deterministically.
+*/
+declare function canaryFingerprint(engine: Engine, probes: CanaryProbeSet): Promise<string>;
+interface CanaryDriftReport {
+  model: ModelRef;
+  freshFingerprint: string;
+  /** Claim ids flipped to stale by this call. */
+  flipped: string[];
+  /** The committed store version when anything flipped. */
+  version?: number;
+}
+/**
+* Flips the model's ACTIVE eval-measured claims to stale when their
+* recorded canary fingerprint differs from the fresh one (docs/05:
+* "a fingerprint change immediately flips the model's eval claims to
+* stale"). Claims without a recorded fingerprint have no baseline and
+* stay untouched (the documented no-probe posture); a second run is
+* an idempotent noop. CAS-rebased like every maintenance commit.
+*/
+declare function flipStaleOnCanaryDrift(store: ModelKnowledgeStore, model: ModelRef, freshFingerprint: string, options?: {
+  attempts?: number;
+}): Promise<CanaryDriftReport>;
+//#endregion
+//#region src/sweeps.d.ts
+/** One fixed pool member; effort is part of the claim subject identity. */
+interface SweepModel {
+  model: ModelRef;
+  effort?: Effort;
+}
+/** An eval case bound to the taskClass axis of the matrix. */
+interface SweepCase {
+  taskClass: TaskClass;
+  case: EvalCase;
+}
+interface SweepPool {
+  models: SweepModel[];
+  cases: SweepCase[];
+}
+interface SweepThresholds {
+  /** passRate at or above emits a strength claim; default 0.9. */
+  strength: number;
+  /** passRate at or below emits a weakness claim; default 0.5. */
+  weakness: number;
+}
+interface RunSweepOptions {
+  /** Deterministic, caller-minted; every claim's evidence and gate reference it. */
+  reportId: string;
+  /** The dedicated identity (docs/05, 5.4). */
+  committerId: string;
+  /** ISO date of the sweep; the TTL table applies from it (no wall clock inside). */
+  observedAt: string;
+  /**
+  * A fresh engine per model cell, routed at that member: the caller
+  * owns adapters, budgets, and the VCR posture, so a sweep records
+  * and replays like any engine run.
+  */
+  engineFor: (member: SweepModel) => Engine | Promise<Engine>;
+  /** Mid-band pass rates emit NO claim (uninformative); see defaults. */
+  thresholds?: Partial<SweepThresholds>;
+  /** Passed through to every suite run (budget, VCR hooks ride the engine). */
+  suite?: RunEvalSuiteOptions;
+  /** When given, emitted claims commit through the committer identity. */
+  store?: ModelKnowledgeStore;
+  /**
+  * Optional epoch stamp per pool member (capture via the core
+  * modelEpochOf; the canary fingerprint rides it when probes ran).
+  */
+  modelEpochFor?: (member: SweepModel) => ModelClaim["modelEpoch"];
+}
+interface SweepCellReport {
+  model: ModelRef;
+  effort?: Effort;
+  taskClass: TaskClass;
+  passRate: number;
+  n: number;
+  totalCostUsd: number;
+  caseNames: string[];
+}
+interface SweepReport {
+  reportId: string;
+  observedAt: string;
+  cells: SweepCellReport[];
+  /** Emitted per the thresholds; committed when a store was given. */
+  claims: MeasuredClaimInput[];
+  committedVersion?: number;
+}
+declare const SWEEP_THRESHOLD_DEFAULTS: SweepThresholds;
+/**
+* Runs the fixed matrix sequentially in declaration order
+* (deterministic cassette consumption), aggregates per (model,
+* taskClass) cell, emits threshold-crossing claims, and commits them
+* through the eval-committer identity when a store is given.
+*/
+declare function runSweepMatrix(pool: SweepPool, options: RunSweepOptions): Promise<SweepReport>;
+//#endregion
+export { type CanaryDriftReport, type CanaryProbeSet, type EvalCase, type EvalCaseResult, type EvalCommitterOptions, EvalJudgeError, type EvalMatrixReport, type EvalSuiteResult, type GoldenGraderOptions, type Grader, type GraderContext, type GraderVerdict, JUDGE_VERDICT_SCHEMA, type JudgeGraderOptions, type JudgeSpec, type MatrixCell, type MatrixCellReport, type MeasuredClaimInput, type RubricCriterion, type RubricGraderOptions, type RunEvalCaseOptions, type RunEvalSuiteOptions, type RunSweepOptions, SWEEP_THRESHOLD_DEFAULTS, type SweepCase, type SweepCellReport, type SweepModel, type SweepPool, type SweepReport, type SweepThresholds, canaryFingerprint, commitEvalMeasured, evalMeasuredClaim, flipStaleOnCanaryDrift, goldenGrader, judgeGrader, normalizeCanaryOutput, rubricGrader, runEvalCase, runEvalMatrix, runEvalSuite, runSweepMatrix };

package/dist/index.js

@@ -0,0 +1,488 @@
+import { KnowledgeCasError, claimExpiry, defineWorkflow } from "@rulvar/core";
+import { createHash } from "node:crypto";
+//#region src/case.ts
+/**
+* @rulvar/evals (M9-T02): EvalCase, the grader contract, and the case and
+* suite runners. A separate quality-measurement package built strictly on
+* the public APIs (L6).
+*
+* Owning spec: docs/09-observability-testing-spec.md, section
+* "@rulvar/evals"; determinism rules in docs/11-testing-strategy.md,
+* section "Eval CI": judge graders run THROUGH the engine, so judge calls
+* are journaled, budgeted, and VCR-recordable, and an eval suite replays
+* deterministically from cassettes with zero live calls.
+*/
+/** Thrown when a judge run does not settle ok. */
+var EvalJudgeError = class extends Error {
+	judgeRun;
+	status;
+	constructor(judgeRun, status, detail) {
+		super(`eval judge run '${judgeRun}' settled '${status}'${detail === void 0 ? "" : `: ${detail}`}`);
+		this.name = "EvalJudgeError";
+		this.judgeRun = judgeRun;
+		this.status = status;
+	}
+};
+/**
+* Runs one EvalCase on the given engine: the target workflow as its own
+* run, pure graders host-side over the outcome, judge graders through the
+* engine via GraderContext.judge. Grader thrown errors are not absorbed:
+* a grader that cannot grade is a defect of the suite, not a failed case.
+*/
+async function runEvalCase(engine, evalCase, options = {}) {
+	const name = options.name ?? evalCase.workflow.name;
+	const timing = {};
+	const handle = engine.run(evalCase.workflow, evalCase.args, {
+		name: `eval:${name}`,
+		...options.budgetUsd === void 0 ? {} : { budgetUsd: options.budgetUsd }
+	});
+	const offStart = handle.on("run:start", (event) => {
+		timing.start ??= event.ts;
+	});
+	const offEnd = handle.on("run:end", (event) => {
+		timing.end ??= event.ts;
+	});
+	const outcome = await handle.result;
+	offStart();
+	offEnd();
+	let judgeCostUsd = 0;
+	let judgeOrdinal = 0;
+	const context = {
+		value: outcome.value,
+		outcome,
+		async judge(spec) {
+			const ordinal = judgeOrdinal;
+			judgeOrdinal += 1;
+			const judged = await runJudge(engine, `${name}:${ordinal}`, spec, options.judgeBudgetUsd);
+			judgeCostUsd += judged.costUsd;
+			return judged.output;
+		}
+	};
+	const verdicts = [];
+	for (const grader of evalCase.graders) verdicts.push(await grader.grade(context));
+	const latencyMs = timing.start !== void 0 && timing.end !== void 0 ? Math.max(0, Date.parse(timing.end) - Date.parse(timing.start)) : 0;
+	return {
+		name,
+		status: outcome.status,
+		passed: outcome.status === "ok" && verdicts.every((verdict) => verdict.passed),
+		verdicts,
+		costUsd: outcome.cost.totalUsd + judgeCostUsd,
+		judgeCostUsd,
+		latencyMs,
+		usage: outcome.usage,
+		...outcome.error === void 0 ? {} : { error: outcome.error }
+	};
+}
+async function runJudge(engine, judgeName, spec, budgetUsd) {
+	const workflowName = `eval-judge:${judgeName}`;
+	const judgeWorkflow = defineWorkflow({ name: workflowName }, async (ctx) => {
+		return await ctx.agent(spec.prompt, {
+			model: spec.model,
+			schema: spec.schema,
+			label: "eval-judge",
+			onError: "throw"
+		});
+	});
+	const outcome = await engine.run(judgeWorkflow, null, {
+		name: workflowName,
+		...budgetUsd === void 0 ? {} : { budgetUsd }
+	}).result;
+	if (outcome.status !== "ok") throw new EvalJudgeError(workflowName, outcome.status, outcome.error?.message);
+	return {
+		output: outcome.value ?? null,
+		costUsd: outcome.cost.totalUsd
+	};
+}
+/**
+* Runs cases sequentially (deterministic journal and cassette order) and
+* aggregates. Duplicate workflow names get '#<ordinal>' suffixes so every
+* result row and judge journal is unambiguous.
+*/
+async function runEvalSuite(engine, cases, options = {}) {
+	const seen = /* @__PURE__ */ new Map();
+	const results = [];
+	for (const evalCase of cases) {
+		const base = evalCase.workflow.name;
+		const ordinal = seen.get(base) ?? 0;
+		seen.set(base, ordinal + 1);
+		const name = ordinal === 0 ? base : `${base}#${ordinal}`;
+		results.push(await runEvalCase(engine, evalCase, {
+			name,
+			...options.budgetUsd === void 0 ? {} : { budgetUsd: options.budgetUsd },
+			...options.judgeBudgetUsd === void 0 ? {} : { judgeBudgetUsd: options.judgeBudgetUsd }
+		}));
+	}
+	return {
+		results,
+		passRate: results.length === 0 ? 0 : results.filter((r) => r.passed).length / results.length,
+		totalCostUsd: results.reduce((sum, r) => sum + r.costUsd, 0),
+		meanLatencyMs: results.length === 0 ? 0 : results.reduce((sum, r) => sum + r.latencyMs, 0) / results.length
+	};
+}
+//#endregion
+//#region src/matrix.ts
+/**
+* Runs the same case list against every cell's engine, sequentially and
+* in declaration order (deterministic cassette consumption), and reports
+* per-cell aggregates for side-by-side comparison.
+*/
+async function runEvalMatrix(cells, cases, options = {}) {
+	const reports = [];
+	for (const cell of cells) {
+		const suite = await runEvalSuite(await cell.engine(), cases, options);
+		reports.push({
+			cell: cell.name,
+			passRate: suite.passRate,
+			totalCostUsd: suite.totalCostUsd,
+			meanLatencyMs: suite.meanLatencyMs,
+			results: suite.results
+		});
+	}
+	return { cells: reports };
+}
+//#endregion
+//#region src/graders/golden.ts
+function deepEqual(a, b) {
+	if (a === b) return true;
+	if (a === null || b === null || a === void 0 || b === void 0) return false;
+	if (Array.isArray(a) || Array.isArray(b)) {
+		if (!Array.isArray(a) || !Array.isArray(b) || a.length !== b.length) return false;
+		return a.every((item, index) => deepEqual(item, b[index]));
+	}
+	if (typeof a === "object" && typeof b === "object") {
+		const aKeys = Object.keys(a).sort();
+		const bKeys = Object.keys(b).sort();
+		if (aKeys.length !== bKeys.length || aKeys.some((key, index) => key !== bKeys[index])) return false;
+		return aKeys.every((key) => deepEqual(a[key], b[key]));
+	}
+	return false;
+}
+function goldenGrader(expected, options = {}) {
+	const name = options.name ?? "golden";
+	return {
+		name,
+		grade(context) {
+			const passed = deepEqual(context.value, expected);
+			return {
+				grader: name,
+				passed,
+				...passed ? {} : { details: {
+					expected,
+					actual: context.value ?? null
+				} }
+			};
+		}
+	};
+}
+//#endregion
+//#region src/graders/rubric.ts
+function rubricGrader(criteria, options = {}) {
+	const name = options.name ?? "rubric";
+	const threshold = options.passThreshold ?? 1;
+	return {
+		name,
+		grade(context) {
+			const rows = criteria.map((criterion) => ({
+				name: criterion.name,
+				passed: criterion.check(context.value)
+			}));
+			const score = criteria.length === 0 ? 1 : rows.filter((row) => row.passed).length / rows.length;
+			return {
+				grader: name,
+				passed: score >= threshold,
+				score,
+				details: { criteria: rows }
+			};
+		}
+	};
+}
+//#endregion
+//#region src/graders/judge.ts
+/** The default judge verdict shape. */
+const JUDGE_VERDICT_SCHEMA = {
+	type: "object",
+	properties: {
+		passed: { type: "boolean" },
+		reasoning: { type: "string" }
+	},
+	required: ["passed"],
+	additionalProperties: false
+};
+function defaultToVerdict(output) {
+	return { passed: typeof output === "object" && output !== null && !Array.isArray(output) && output.passed === true };
+}
+function judgePrompt(instruction, value) {
+	return [
+		"You are an evaluation judge. Judge the candidate output below against the instruction.",
+		"",
+		`Instruction: ${instruction}`,
+		"",
+		"Candidate output (JSON):",
+		JSON.stringify(value ?? null),
+		"",
+		"Return a verdict object matching the response schema."
+	].join("\n");
+}
+function judgeGrader(options) {
+	const name = options.name ?? "judge";
+	if (options.schema !== void 0 && options.toVerdict === void 0) throw new Error(`judgeGrader '${name}': a custom schema requires toVerdict`);
+	const toVerdict = options.toVerdict ?? defaultToVerdict;
+	const schema = options.schema ?? JUDGE_VERDICT_SCHEMA;
+	return {
+		name,
+		async grade(context) {
+			if (context.outcome.status !== "ok") return {
+				grader: name,
+				passed: false,
+				details: { skipped: `target run settled '${context.outcome.status}'` }
+			};
+			const output = await context.judge({
+				model: options.model,
+				prompt: judgePrompt(options.instruction, context.value),
+				schema
+			});
+			const verdict = toVerdict(output);
+			return {
+				grader: name,
+				passed: verdict.passed,
+				...verdict.score === void 0 ? {} : { score: verdict.score },
+				details: { output }
+			};
+		}
+	};
+}
+//#endregion
+//#region src/committer.ts
+/**
+* The eval-committer identity (M11-T01; docs/05, sections "Data model"
+* and "Commit discipline"). The pipeline-side commit path: builds
+* eval-committer-gated ops (the coherence square: class eval-measured,
+* author eval-pipeline, metrics present) and commits them with the
+* documented CAS-rebase recipe. Humans never call this; their path is
+* the human gate and it structurally cannot carry metrics.
+*/
+/** One measured claim, TTL applied per the docs/05 decay table. */
+function evalMeasuredClaim(input, committerId) {
+	return {
+		id: input.id,
+		subject: input.subject,
+		taskClass: input.taskClass,
+		polarity: input.polarity,
+		statement: input.statement,
+		class: "eval-measured",
+		status: "active",
+		evidence: input.evidence,
+		metrics: input.metrics,
+		confidence: input.confidence,
+		observedAt: input.observedAt,
+		expiresAt: claimExpiry("eval-measured", input.polarity, input.observedAt),
+		...input.modelEpoch === void 0 ? {} : { modelEpoch: input.modelEpoch },
+		author: {
+			kind: "eval-pipeline",
+			id: committerId
+		}
+	};
+}
+/**
+* Commits measured claims through the eval-committer gate with the
+* documented rebase recipe: on a CAS rejection, re-read current() and
+* retry against the fresh version. Returns the committed version.
+*/
+async function commitEvalMeasured(store, claims, options) {
+	const gate = {
+		kind: "eval-committer",
+		committerId: options.committerId,
+		reportId: options.reportId
+	};
+	const ops = claims.map((input) => ({
+		op: "add",
+		claim: evalMeasuredClaim(input, options.committerId),
+		gate
+	}));
+	const attempts = options.attempts ?? 3;
+	let lastCas;
+	for (let attempt = 0; attempt < attempts; attempt += 1) {
+		const snapshot = await store.current();
+		try {
+			return await store.commit(ops, snapshot.version);
+		} catch (thrown) {
+			if (thrown instanceof KnowledgeCasError) {
+				lastCas = thrown;
+				continue;
+			}
+			throw thrown;
+		}
+	}
+	throw lastCas ?? /* @__PURE__ */ new Error("commitEvalMeasured: unreachable");
+}
+//#endregion
+//#region src/canary.ts
+/**
+* The canary fingerprint (M11-T04; docs/05, section "Grounding and
+* decay"; OQ-06). The optional compensation for silent alias
+* re-pointing that modelEpoch honestly cannot catch: a FIXED probe set
+* at temperature 0, run through the ordinary engine (journaled,
+* budgeted, VCR-recordable), hashed over normalized outputs. A
+* fingerprint change flips the model's eval claims to stale in one
+* command.
+*
+* The committed v1 design (closing OQ-06): the probe set is CALLER
+* data (fixed, versioned alongside the store); normalization is NFC,
+* trim, and whitespace collapse per output; the fingerprint is the
+* sha256 of the JCS-serialized normalized output array, prefixed with
+* the probe count so a probe-set edit never collides with drift.
+*/
+/** The committed v1 normalization (OQ-06): NFC, trim, collapse whitespace. */
+function normalizeCanaryOutput(output) {
+	return (typeof output === "string" ? output : JSON.stringify(output ?? null)).normalize("NFC").trim().replace(/\s+/gu, " ");
+}
+/**
+* Runs the fixed probe set through the ordinary engine and returns the
+* fingerprint. Probes run sequentially in declaration order, one run
+* per probe, so recordings replay deterministically.
+*/
+async function canaryFingerprint(engine, probes) {
+	const outputs = [];
+	for (const [index, prompt] of probes.prompts.entries()) {
+		const workflow = defineWorkflow({ name: `kb-canary:${String(index)}` }, async (ctx) => await ctx.agent(prompt, { agentType: probes.agentType }));
+		const outcome = await engine.run(workflow, null).result;
+		outputs.push(outcome.status === "ok" ? normalizeCanaryOutput(outcome.value) : `!${outcome.status}`);
+	}
+	const body = JSON.stringify([probes.prompts.length, outputs]);
+	return createHash("sha256").update(body, "utf8").digest("hex");
+}
+/**
+* Flips the model's ACTIVE eval-measured claims to stale when their
+* recorded canary fingerprint differs from the fresh one (docs/05:
+* "a fingerprint change immediately flips the model's eval claims to
+* stale"). Claims without a recorded fingerprint have no baseline and
+* stay untouched (the documented no-probe posture); a second run is
+* an idempotent noop. CAS-rebased like every maintenance commit.
+*/
+async function flipStaleOnCanaryDrift(store, model, freshFingerprint, options) {
+	const attempts = options?.attempts ?? 3;
+	let lastCas;
+	for (let attempt = 0; attempt < attempts; attempt += 1) {
+		const snapshot = await store.current();
+		const drifted = snapshot.claims.filter((claim) => claim.status === "active" && claim.class === "eval-measured" && claim.subject.model === model && claim.modelEpoch?.canaryFingerprint !== void 0 && claim.modelEpoch.canaryFingerprint !== freshFingerprint);
+		if (drifted.length === 0) return {
+			model,
+			freshFingerprint,
+			flipped: []
+		};
+		const ops = drifted.map((claim) => ({
+			op: "mark_stale",
+			claimId: claim.id,
+			reason: "canary-drift"
+		}));
+		try {
+			const version = await store.commit(ops, snapshot.version);
+			return {
+				model,
+				freshFingerprint,
+				flipped: drifted.map((claim) => claim.id),
+				version
+			};
+		} catch (thrown) {
+			if (thrown instanceof KnowledgeCasError) {
+				lastCas = thrown;
+				continue;
+			}
+			throw thrown;
+		}
+	}
+	throw lastCas ?? /* @__PURE__ */ new Error("flipStaleOnCanaryDrift: unreachable");
+}
+//#endregion
+//#region src/sweeps.ts
+const SWEEP_THRESHOLD_DEFAULTS = {
+	strength: .9,
+	weakness: .5
+};
+/** Deterministic claim id: report-scoped, readable, collision-free. */
+function claimIdOf(reportId, member, taskClass) {
+	const effort = member.effort === void 0 ? "" : `@${member.effort}`;
+	return `${reportId}/${member.model}${effort}/${taskClass}`;
+}
+/** The typed statement template: never a quote from tool output. */
+function statementOf(cell, polarity) {
+	const rate = cell.passRate.toFixed(2);
+	const band = polarity === "strength" ? "at or above the strength band" : "in the weakness band";
+	return `sweep passRate ${rate} over ${String(cell.n)} ${cell.taskClass} case${cell.n === 1 ? "" : "s"}: ${band}`;
+}
+/**
+* Runs the fixed matrix sequentially in declaration order
+* (deterministic cassette consumption), aggregates per (model,
+* taskClass) cell, emits threshold-crossing claims, and commits them
+* through the eval-committer identity when a store is given.
+*/
+async function runSweepMatrix(pool, options) {
+	const thresholds = {
+		...SWEEP_THRESHOLD_DEFAULTS,
+		...options.thresholds
+	};
+	const byTaskClass = /* @__PURE__ */ new Map();
+	for (const entry of pool.cases) {
+		const bucket = byTaskClass.get(entry.taskClass) ?? [];
+		bucket.push(entry);
+		byTaskClass.set(entry.taskClass, bucket);
+	}
+	const cells = [];
+	const claims = [];
+	for (const member of pool.models) {
+		const engine = await options.engineFor(member);
+		for (const [taskClass, bucket] of byTaskClass) {
+			const suite = await runEvalSuite(engine, bucket.map((entry) => entry.case), options.suite ?? {});
+			const cell = {
+				model: member.model,
+				...member.effort === void 0 ? {} : { effort: member.effort },
+				taskClass,
+				passRate: suite.passRate,
+				n: suite.results.length,
+				totalCostUsd: suite.totalCostUsd,
+				caseNames: suite.results.map((result) => result.name)
+			};
+			cells.push(cell);
+			const polarity = cell.passRate >= thresholds.strength ? "strength" : cell.passRate <= thresholds.weakness ? "weakness" : void 0;
+			if (polarity !== void 0 && cell.n > 0) {
+				const epoch = options.modelEpochFor?.(member);
+				claims.push({
+					id: claimIdOf(options.reportId, member, taskClass),
+					subject: {
+						model: member.model,
+						...member.effort === void 0 ? {} : { effort: member.effort }
+					},
+					taskClass,
+					polarity,
+					statement: statementOf(cell, polarity),
+					metrics: {
+						passRate: cell.passRate,
+						n: cell.n,
+						graderId: "eval-suite"
+					},
+					confidence: cell.n >= 20 ? "high" : cell.n >= 5 ? "medium" : "low",
+					observedAt: options.observedAt,
+					evidence: [{
+						kind: "eval",
+						reportId: options.reportId,
+						caseIds: cell.caseNames
+					}],
+					...epoch === void 0 ? {} : { modelEpoch: epoch }
+				});
+			}
+		}
+	}
+	const report = {
+		reportId: options.reportId,
+		observedAt: options.observedAt,
+		cells,
+		claims
+	};
+	if (options.store !== void 0 && claims.length > 0) report.committedVersion = await commitEvalMeasured(options.store, claims, {
+		committerId: options.committerId,
+		reportId: options.reportId
+	});
+	return report;
+}
+//#endregion
+export { EvalJudgeError, JUDGE_VERDICT_SCHEMA, SWEEP_THRESHOLD_DEFAULTS, canaryFingerprint, commitEvalMeasured, evalMeasuredClaim, flipStaleOnCanaryDrift, goldenGrader, judgeGrader, normalizeCanaryOutput, rubricGrader, runEvalCase, runEvalMatrix, runEvalSuite, runSweepMatrix };

package/package.json

@@ -0,0 +1,48 @@
+{
+  "name": "@rulvar/evals",
+  "version": "1.0.0",
+  "description": "rulvar evals: eval cases, golden outputs, rubric and judge graders, matrix sweeps, canary fingerprint.",
+  "type": "module",
+  "license": "Apache-2.0",
+  "engines": {
+    "node": ">=22.12.0"
+  },
+  "exports": {
+    ".": {
+      "types": "./dist/index.d.ts",
+      "default": "./dist/index.js"
+    },
+    "./package.json": "./package.json"
+  },
+  "files": [
+    "dist"
+  ],
+  "sideEffects": false,
+  "publishConfig": {
+    "access": "public"
+  },
+  "dependencies": {
+    "@rulvar/testing": "1.0.0",
+    "@rulvar/core": "1.0.0"
+  },
+  "devDependencies": {
+    "@types/node": "^22.20.0",
+    "tsdown": "^0.22.3",
+    "typescript": "~6.0.3"
+  },
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/o-stepper/rulvar.git",
+    "directory": "packages/evals"
+  },
+  "homepage": "https://rulvar.com",
+  "bugs": {
+    "url": "https://github.com/o-stepper/rulvar/issues"
+  },
+  "scripts": {
+    "build": "tsdown",
+    "typecheck": "tsc --noEmit",
+    "lint": "eslint .",
+    "pack-check": "publint --pack pnpm && attw --pack . --profile esm-only"
+  }
+}