sensei-eval 0.1.0

package/README.md

@@ -0,0 +1,201 @@
+# edu-eval
+
+TypeScript library for evaluating AI-generated educational content using deterministic checks and LLM-as-judge scoring.
+
+## Install
+
+```bash
+npm install edu-eval
+```
+
+## Quick Start
+
+```typescript
+import { EvalRunner, createJudge, criteria } from 'edu-eval';
+
+// Full evaluation (deterministic + LLM judge)
+const judge = createJudge({ apiKey: process.env.ANTHROPIC_API_KEY! });
+const runner = new EvalRunner({
+  criteria: [
+    ...criteria.universal,
+    ...criteria.lesson,
+    ...criteria.challenge,
+    ...criteria.review,
+  ],
+  judge,
+});
+
+const result = await runner.evaluate({
+  content: lessonMarkdown,
+  contentType: 'lesson',
+  topic: 'Array methods in JavaScript',
+  difficulty: 'beginner',
+});
+
+console.log(result.passed);       // true if all criteria met their thresholds
+console.log(result.overallScore); // weighted average (0-1)
+console.log(result.scores);      // per-criterion breakdown
+
+// Quick check — deterministic only, no API calls
+const quick = await runner.quickCheck({
+  content: lessonMarkdown,
+  contentType: 'lesson',
+});
+```
+
+## How It Works
+
+Content is evaluated through a two-tier system:
+
+1. **Deterministic criteria** — fast, pure functions that check markdown formatting, length, and structure
+2. **LLM-judge criteria** — Claude API calls that score content against rubrics on a 1-5 scale, normalized to 0-1
+
+`EvalRunner.evaluate()` filters criteria by `contentType`, runs deterministic checks in parallel, then runs LLM checks in parallel, computes a weighted overall score, and returns an `EvalResult`.
+
+`EvalRunner.quickCheck()` does the same but skips all LLM criteria (no API calls).
+
+### Scoring
+
+- **Overall score**: weighted average of all criterion scores (0-1)
+- **Pass/fail**: every individual criterion must meet its threshold for `passed` to be `true`
+- **LLM normalization**: raw 1-5 scores are mapped to 0-1 via `(score - 1) / 4`
+
+### Weights
+
+| Weight | Criteria |
+|--------|----------|
+| 1.5 | `topic_accuracy`, `pedagogical_structure`, `problem_clarity` |
+| 1.0 | Most criteria (engagement, repetition_avoidance, code_quality, etc.) |
+| 0.5 | `has_code_block`, `has_structure`, `brevity` |
+
+## Criteria
+
+### Universal (all content types)
+
+| Criterion | Method | Description |
+|-----------|--------|-------------|
+| `format_compliance` | deterministic | No unclosed code blocks, bold, or italic markers |
+| `length_compliance` | deterministic | Content within min/max character limits per type |
+| `has_code_block` | deterministic | At least one fenced code block (lesson, challenge only) |
+| `has_structure` | deterministic | Headings and multiple sections (lesson, challenge only) |
+| `engagement` | LLM judge | Hook, pacing, and closure quality |
+| `repetition_avoidance` | LLM judge | Avoids repeating previous content |
+
+### Lesson
+
+| Criterion | Method | Description |
+|-----------|--------|-------------|
+| `topic_accuracy` | LLM judge | Covers stated topic correctly, no factual errors |
+| `pedagogical_structure` | LLM judge | Progression from intuition to examples to practice |
+| `code_quality` | LLM judge | Relevant, correct, well-explained code examples |
+| `progressive_difficulty` | LLM judge | Appropriate scaffolding for stated difficulty level |
+
+### Challenge
+
+| Criterion | Method | Description |
+|-----------|--------|-------------|
+| `problem_clarity` | LLM judge | Unambiguous requirements with clear inputs/outputs |
+| `difficulty_calibration` | LLM judge | Matches stated difficulty level |
+| `hint_quality` | LLM judge | Progressive hints without spoiling the solution |
+| `testability` | LLM judge | Verifiable with clear expected outputs or test cases |
+
+### Review
+
+| Criterion | Method | Description |
+|-----------|--------|-------------|
+| `brevity` | deterministic | Under character limit (default 2000, configurable via metadata) |
+| `actionability` | LLM judge | Concrete, specific next steps |
+| `honesty` | LLM judge | Honest gap assessment without sugarcoating |
+
+## API
+
+### `EvalRunner`
+
+```typescript
+const runner = new EvalRunner({ criteria: EvalCriterion[], judge?: Judge });
+
+runner.evaluate(input: EvalInput): Promise<EvalResult>    // full eval
+runner.quickCheck(input: EvalInput): Promise<EvalResult>   // deterministic only
+runner.getCriteria(contentType: string): EvalCriterion[]   // filter by type
+```
+
+### `createJudge`
+
+```typescript
+const judge = createJudge({
+  apiKey: string,
+  model?: string,      // default: 'claude-sonnet-4-20250514'
+  maxTokens?: number,
+});
+```
+
+### Types
+
+```typescript
+interface EvalInput {
+  content: string;
+  contentType: string;              // 'lesson' | 'challenge' | 'review' | 'quiz'
+  topic?: string;
+  difficulty?: string;
+  previousContent?: string[];       // for repetition checking
+  metadata?: Record<string, unknown>;
+}
+
+interface EvalResult {
+  overallScore: number;             // weighted average (0-1)
+  passed: boolean;                  // all criteria passed their thresholds
+  scores: EvalScore[];
+  contentType: string;
+  evaluatedAt: string;              // ISO timestamp
+}
+
+interface EvalScore {
+  criterion: string;
+  score: number;                    // normalized (0-1)
+  rawScore: number;
+  maxScore: number;
+  passed: boolean;
+  reasoning: string;
+  metadata?: Record<string, unknown>;
+}
+```
+
+## Adding a New Criterion
+
+1. Define a `JudgeRubric` with a 1-5 scale (if LLM) in the appropriate `src/criteria/*.ts`
+2. Export an `EvalCriterion` object with an `evaluate()` function
+3. Add it to the exported array at the bottom of the file
+4. Add tests — mock the `Judge` for LLM criteria, use fixtures for deterministic
+
+## Development
+
+```bash
+npm run build        # tsc -> dist/
+npm test             # vitest run (unit tests, no API calls)
+npm run test:watch   # vitest in watch mode
+```
+
+Tests mock all LLM calls — no Anthropic API key needed to run the test suite.
+
+## Project Structure
+
+```
+src/
+  index.ts              # public API
+  types.ts              # all interfaces
+  runner.ts             # EvalRunner orchestration
+  judge.ts              # Anthropic SDK wrapper
+  criteria/
+    index.ts            # re-exports all criterion sets
+    universal.ts        # criteria for all content types
+    lesson.ts           # lesson-specific criteria
+    challenge.ts        # challenge-specific criteria
+    review.ts           # review-specific criteria
+  utils/
+    markdown.ts         # markdown parsing helpers
+tests/
+  fixtures/             # good/bad lesson and challenge markdown
+  criteria/             # per-criterion tests
+  runner.test.ts        # orchestration tests
+  judge.test.ts         # judge tests
+```

package/dist/criteria/challenge.d.ts

@@ -0,0 +1,7 @@
+import type { EvalCriterion } from '../types.js';
+export declare const problemClarity: EvalCriterion;
+export declare const difficultyCalibration: EvalCriterion;
+export declare const hintQuality: EvalCriterion;
+export declare const testability: EvalCriterion;
+export declare const challenge: EvalCriterion[];
+//# sourceMappingURL=challenge.d.ts.map

package/dist/criteria/challenge.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"challenge.d.ts","sourceRoot":"","sources":["../../src/criteria/challenge.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAA4C,MAAM,aAAa,CAAC;AAe3F,eAAO,MAAM,cAAc,EAAE,aAoB5B,CAAC;AAeF,eAAO,MAAM,qBAAqB,EAAE,aAuBnC,CAAC;AAeF,eAAO,MAAM,WAAW,EAAE,aAoBzB,CAAC;AAeF,eAAO,MAAM,WAAW,EAAE,aAoBzB,CAAC;AAEF,eAAO,MAAM,SAAS,EAAE,aAAa,EAKpC,CAAC"}

package/dist/criteria/challenge.js

@@ -0,0 +1,142 @@
+const problemClarityRubric = {
+    criterion: 'Problem Clarity',
+    description: 'Is the problem statement unambiguous? Can the reader understand exactly what they need to build/solve?',
+    scale: [
+        { score: 1, label: 'Unclear', description: 'Vague or contradictory requirements, impossible to know what to build' },
+        { score: 2, label: 'Ambiguous', description: 'General idea is clear but key details are missing or unclear' },
+        { score: 3, label: 'Clear', description: 'Requirements are understandable, minor ambiguities acceptable' },
+        { score: 4, label: 'Precise', description: 'Clear inputs, outputs, and constraints with examples' },
+        { score: 5, label: 'Crystal clear', description: 'Unambiguous spec with examples, edge cases noted, nothing left to guess' },
+    ],
+};
+export const problemClarity = {
+    name: 'problem_clarity',
+    description: 'Problem statement is unambiguous and complete',
+    contentTypes: ['challenge'],
+    method: 'llm_judge',
+    threshold: 0.5,
+    weight: 1.5,
+    async evaluate(input, judge) {
+        if (!judge)
+            throw new Error('Judge required for problem_clarity criterion');
+        const result = await judge.score(input.content, problemClarityRubric);
+        const score = (result.score - 1) / 4;
+        return {
+            criterion: 'problem_clarity',
+            score,
+            rawScore: result.score,
+            maxScore: 5,
+            passed: score >= this.threshold,
+            reasoning: result.reasoning,
+        };
+    },
+};
+const difficultyCalibrationsRubric = {
+    criterion: 'Difficulty Calibration',
+    description: 'Does the challenge match the expected difficulty level? Consider: concept complexity, number of steps, edge cases, and expected time to solve.',
+    scale: [
+        { score: 1, label: 'Way off', description: 'Trivial problem labeled hard, or impossible problem labeled easy' },
+        { score: 2, label: 'Miscalibrated', description: 'Noticeably easier or harder than stated level' },
+        { score: 3, label: 'Reasonable', description: 'Roughly matches the stated difficulty, minor miscalibration' },
+        { score: 4, label: 'Well-calibrated', description: 'Difficulty matches expectations, appropriate scope' },
+        { score: 5, label: 'Perfectly tuned', description: 'Exactly the right level of challenge for the stated difficulty' },
+    ],
+};
+export const difficultyCalibration = {
+    name: 'difficulty_calibration',
+    description: 'Challenge matches the expected difficulty level',
+    contentTypes: ['challenge'],
+    method: 'llm_judge',
+    threshold: 0.5,
+    weight: 1.0,
+    async evaluate(input, judge) {
+        if (!judge)
+            throw new Error('Judge required for difficulty_calibration criterion');
+        const context = input.difficulty
+            ? `Expected difficulty: ${input.difficulty}`
+            : undefined;
+        const result = await judge.score(input.content, difficultyCalibrationsRubric, context);
+        const score = (result.score - 1) / 4;
+        return {
+            criterion: 'difficulty_calibration',
+            score,
+            rawScore: result.score,
+            maxScore: 5,
+            passed: score >= this.threshold,
+            reasoning: result.reasoning,
+        };
+    },
+};
+const hintQualityRubric = {
+    criterion: 'Hint Quality',
+    description: 'Are hints provided? Are they progressive (nudge → more direct)? Do they help without giving away the answer?',
+    scale: [
+        { score: 1, label: 'No hints', description: 'No hints or guidance provided at all' },
+        { score: 2, label: 'Weak hints', description: 'Hints exist but are too vague or give away the answer' },
+        { score: 3, label: 'Adequate', description: 'Hints provide useful direction without spoiling the solution' },
+        { score: 4, label: 'Good progression', description: 'Multiple hint levels from gentle nudge to more direct guidance' },
+        { score: 5, label: 'Perfect scaffolding', description: 'Progressive hints that teach problem-solving strategy, not just the answer' },
+    ],
+};
+export const hintQuality = {
+    name: 'hint_quality',
+    description: 'Hints are progressive and helpful without spoiling the solution',
+    contentTypes: ['challenge'],
+    method: 'llm_judge',
+    threshold: 0.5,
+    weight: 1.0,
+    async evaluate(input, judge) {
+        if (!judge)
+            throw new Error('Judge required for hint_quality criterion');
+        const result = await judge.score(input.content, hintQualityRubric);
+        const score = (result.score - 1) / 4;
+        return {
+            criterion: 'hint_quality',
+            score,
+            rawScore: result.score,
+            maxScore: 5,
+            passed: score >= this.threshold,
+            reasoning: result.reasoning,
+        };
+    },
+};
+const testabilityRubric = {
+    criterion: 'Testability',
+    description: 'Can the solution be verified? Are there clear expected outputs, test cases, or validation criteria?',
+    scale: [
+        { score: 1, label: 'Unverifiable', description: 'No way to know if a solution is correct' },
+        { score: 2, label: 'Vaguely testable', description: 'General idea of correctness but no concrete checks' },
+        { score: 3, label: 'Testable', description: 'Expected behavior is clear enough to verify manually' },
+        { score: 4, label: 'Well-specified', description: 'Clear expected outputs or test cases provided' },
+        { score: 5, label: 'Fully specified', description: 'Complete test cases with edge cases, ready to validate automatically' },
+    ],
+};
+export const testability = {
+    name: 'testability',
+    description: 'Solution is verifiable with clear expected outputs',
+    contentTypes: ['challenge'],
+    method: 'llm_judge',
+    threshold: 0.5,
+    weight: 1.0,
+    async evaluate(input, judge) {
+        if (!judge)
+            throw new Error('Judge required for testability criterion');
+        const result = await judge.score(input.content, testabilityRubric);
+        const score = (result.score - 1) / 4;
+        return {
+            criterion: 'testability',
+            score,
+            rawScore: result.score,
+            maxScore: 5,
+            passed: score >= this.threshold,
+            reasoning: result.reasoning,
+        };
+    },
+};
+export const challenge = [
+    problemClarity,
+    difficultyCalibration,
+    hintQuality,
+    testability,
+];
+//# sourceMappingURL=challenge.js.map

package/dist/criteria/challenge.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"challenge.js","sourceRoot":"","sources":["../../src/criteria/challenge.ts"],"names":[],"mappings":"AAEA,MAAM,oBAAoB,GAAgB;IACxC,SAAS,EAAE,iBAAiB;IAC5B,WAAW,EACT,wGAAwG;IAC1G,KAAK,EAAE;QACL,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,SAAS,EAAE,WAAW,EAAE,uEAAuE,EAAE;QACpH,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,WAAW,EAAE,WAAW,EAAE,8DAA8D,EAAE;QAC7G,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,OAAO,EAAE,WAAW,EAAE,+DAA+D,EAAE;QAC1G,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,SAAS,EAAE,WAAW,EAAE,sDAAsD,EAAE;QACnG,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,eAAe,EAAE,WAAW,EAAE,yEAAyE,EAAE;KAC7H;CACF,CAAC;AAEF,MAAM,CAAC,MAAM,cAAc,GAAkB;IAC3C,IAAI,EAAE,iBAAiB;IACvB,WAAW,EAAE,+CAA+C;IAC5D,YAAY,EAAE,CAAC,WAAW,CAAC;IAC3B,MAAM,EAAE,WAAW;IACnB,SAAS,EAAE,GAAG;IACd,MAAM,EAAE,GAAG;IACX,KAAK,CAAC,QAAQ,CAAC,KAAgB,EAAE,KAAa;QAC5C,IAAI,CAAC,KAAK;YAAE,MAAM,IAAI,KAAK,CAAC,8CAA8C,CAAC,CAAC;QAC5E,MAAM,MAAM,GAAG,MAAM,KAAK,CAAC,KAAK,CAAC,KAAK,CAAC,OAAO,EAAE,oBAAoB,CAAC,CAAC;QACtE,MAAM,KAAK,GAAG,CAAC,MAAM,CAAC,KAAK,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC;QACrC,OAAO;YACL,SAAS,EAAE,iBAAiB;YAC5B,KAAK;YACL,QAAQ,EAAE,MAAM,CAAC,KAAK;YACtB,QAAQ,EAAE,CAAC;YACX,MAAM,EAAE,KAAK,IAAI,IAAI,CAAC,SAAS;YAC/B,SAAS,EAAE,MAAM,CAAC,SAAS;SAC5B,CAAC;IACJ,CAAC;CACF,CAAC;AAEF,MAAM,4BAA4B,GAAgB;IAChD,SAAS,EAAE,wBAAwB;IACnC,WAAW,EACT,gJAAgJ;IAClJ,KAAK,EAAE;QACL,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,SAAS,EAAE,WAAW,EAAE,kEAAkE,EAAE;QAC/G,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,eAAe,EAAE,WAAW,EAAE,+CAA+C,EAAE;QAClG,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,YAAY,EAAE,WAAW,EAAE,6DAA6D,EAAE;QAC7G,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,iBAAiB,EAAE,WAAW,EAAE,oDAAoD,EAAE;QACzG,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,iBAAiB,EAAE,WAAW,EAAE,gEAAgE,EAAE;KACtH;CACF,CAAC;AAEF,MAAM,CAAC,MAAM,qBAAqB,GAAkB;IAClD,IAAI,EAAE,wBAAwB;IAC9B,WAAW,EAAE,iDAAiD;IAC9D,YAAY,EAAE,CAAC,WAAW,CAAC;IAC3B,MAAM,EAAE,WAAW;IACnB,SAAS,EAAE,GAAG;IACd,MAAM,EAAE,GAAG;IACX,KAAK,CAAC,QAAQ,CAAC,KAAgB,EAAE,KAAa;QAC5C,IAAI,CAAC,KAAK;YAAE,MAAM,IAAI,KAAK,CAAC,qDAAqD,CAAC,CAAC;QACnF,MAAM,OAAO,GAAG,KAAK,CAAC,UAAU;YAC9B,CAAC,CAAC,wBAAwB,KAAK,CAAC,UAAU,EAAE;YAC5C,CAAC,CAAC,SAAS,CAAC;QACd,MAAM,MAAM,GAAG,MAAM,KAAK,CAAC,KAAK,CAAC,KAAK,CAAC,OAAO,EAAE,4BAA4B,EAAE,OAAO,CAAC,CAAC;QACvF,MAAM,KAAK,GAAG,CAAC,MAAM,CAAC,KAAK,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC;QACrC,OAAO;YACL,SAAS,EAAE,wBAAwB;YACnC,KAAK;YACL,QAAQ,EAAE,MAAM,CAAC,KAAK;YACtB,QAAQ,EAAE,CAAC;YACX,MAAM,EAAE,KAAK,IAAI,IAAI,CAAC,SAAS;YAC/B,SAAS,EAAE,MAAM,CAAC,SAAS;SAC5B,CAAC;IACJ,CAAC;CACF,CAAC;AAEF,MAAM,iBAAiB,GAAgB;IACrC,SAAS,EAAE,cAAc;IACzB,WAAW,EACT,8GAA8G;IAChH,KAAK,EAAE;QACL,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,UAAU,EAAE,WAAW,EAAE,sCAAsC,EAAE;QACpF,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,YAAY,EAAE,WAAW,EAAE,uDAAuD,EAAE;QACvG,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,UAAU,EAAE,WAAW,EAAE,8DAA8D,EAAE;QAC5G,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,kBAAkB,EAAE,WAAW,EAAE,gEAAgE,EAAE;QACtH,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,qBAAqB,EAAE,WAAW,EAAE,4EAA4E,EAAE;KACtI;CACF,CAAC;AAEF,MAAM,CAAC,MAAM,WAAW,GAAkB;IACxC,IAAI,EAAE,cAAc;IACpB,WAAW,EAAE,iEAAiE;IAC9E,YAAY,EAAE,CAAC,WAAW,CAAC;IAC3B,MAAM,EAAE,WAAW;IACnB,SAAS,EAAE,GAAG;IACd,MAAM,EAAE,GAAG;IACX,KAAK,CAAC,QAAQ,CAAC,KAAgB,EAAE,KAAa;QAC5C,IAAI,CAAC,KAAK;YAAE,MAAM,IAAI,KAAK,CAAC,2CAA2C,CAAC,CAAC;QACzE,MAAM,MAAM,GAAG,MAAM,KAAK,CAAC,KAAK,CAAC,KAAK,CAAC,OAAO,EAAE,iBAAiB,CAAC,CAAC;QACnE,MAAM,KAAK,GAAG,CAAC,MAAM,CAAC,KAAK,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC;QACrC,OAAO;YACL,SAAS,EAAE,cAAc;YACzB,KAAK;YACL,QAAQ,EAAE,MAAM,CAAC,KAAK;YACtB,QAAQ,EAAE,CAAC;YACX,MAAM,EAAE,KAAK,IAAI,IAAI,CAAC,SAAS;YAC/B,SAAS,EAAE,MAAM,CAAC,SAAS;SAC5B,CAAC;IACJ,CAAC;CACF,CAAC;AAEF,MAAM,iBAAiB,GAAgB;IACrC,SAAS,EAAE,aAAa;IACxB,WAAW,EACT,qGAAqG;IACvG,KAAK,EAAE;QACL,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,cAAc,EAAE,WAAW,EAAE,yCAAyC,EAAE;QAC3F,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,kBAAkB,EAAE,WAAW,EAAE,oDAAoD,EAAE;QAC1G,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,UAAU,EAAE,WAAW,EAAE,sDAAsD,EAAE;QACpG,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,gBAAgB,EAAE,WAAW,EAAE,+CAA+C,EAAE;QACnG,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,iBAAiB,EAAE,WAAW,EAAE,sEAAsE,EAAE;KAC5H;CACF,CAAC;AAEF,MAAM,CAAC,MAAM,WAAW,GAAkB;IACxC,IAAI,EAAE,aAAa;IACnB,WAAW,EAAE,oDAAoD;IACjE,YAAY,EAAE,CAAC,WAAW,CAAC;IAC3B,MAAM,EAAE,WAAW;IACnB,SAAS,EAAE,GAAG;IACd,MAAM,EAAE,GAAG;IACX,KAAK,CAAC,QAAQ,CAAC,KAAgB,EAAE,KAAa;QAC5C,IAAI,CAAC,KAAK;YAAE,MAAM,IAAI,KAAK,CAAC,0CAA0C,CAAC,CAAC;QACxE,MAAM,MAAM,GAAG,MAAM,KAAK,CAAC,KAAK,CAAC,KAAK,CAAC,OAAO,EAAE,iBAAiB,CAAC,CAAC;QACnE,MAAM,KAAK,GAAG,CAAC,MAAM,CAAC,KAAK,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC;QACrC,OAAO;YACL,SAAS,EAAE,aAAa;YACxB,KAAK;YACL,QAAQ,EAAE,MAAM,CAAC,KAAK;YACtB,QAAQ,EAAE,CAAC;YACX,MAAM,EAAE,KAAK,IAAI,IAAI,CAAC,SAAS;YAC/B,SAAS,EAAE,MAAM,CAAC,SAAS;SAC5B,CAAC;IACJ,CAAC;CACF,CAAC;AAEF,MAAM,CAAC,MAAM,SAAS,GAAoB;IACxC,cAAc;IACd,qBAAqB;IACrB,WAAW;IACX,WAAW;CACZ,CAAC"}

package/dist/criteria/index.d.ts

@@ -0,0 +1,5 @@
+export { universal } from './universal.js';
+export { lesson } from './lesson.js';
+export { challenge } from './challenge.js';
+export { review } from './review.js';
+//# sourceMappingURL=index.d.ts.map

package/dist/criteria/index.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/criteria/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,SAAS,EAAE,MAAM,gBAAgB,CAAC;AAC3C,OAAO,EAAE,MAAM,EAAE,MAAM,aAAa,CAAC;AACrC,OAAO,EAAE,SAAS,EAAE,MAAM,gBAAgB,CAAC;AAC3C,OAAO,EAAE,MAAM,EAAE,MAAM,aAAa,CAAC"}

package/dist/criteria/index.js

@@ -0,0 +1,5 @@
+export { universal } from './universal.js';
+export { lesson } from './lesson.js';
+export { challenge } from './challenge.js';
+export { review } from './review.js';
+//# sourceMappingURL=index.js.map

package/dist/criteria/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../src/criteria/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,SAAS,EAAE,MAAM,gBAAgB,CAAC;AAC3C,OAAO,EAAE,MAAM,EAAE,MAAM,aAAa,CAAC;AACrC,OAAO,EAAE,SAAS,EAAE,MAAM,gBAAgB,CAAC;AAC3C,OAAO,EAAE,MAAM,EAAE,MAAM,aAAa,CAAC"}

package/dist/criteria/lesson.d.ts

@@ -0,0 +1,7 @@
+import type { EvalCriterion } from '../types.js';
+export declare const topicAccuracy: EvalCriterion;
+export declare const pedagogicalStructure: EvalCriterion;
+export declare const codeQuality: EvalCriterion;
+export declare const progressiveDifficulty: EvalCriterion;
+export declare const lesson: EvalCriterion[];
+//# sourceMappingURL=lesson.d.ts.map

package/dist/criteria/lesson.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"lesson.d.ts","sourceRoot":"","sources":["../../src/criteria/lesson.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAA4C,MAAM,aAAa,CAAC;AAc3F,eAAO,MAAM,aAAa,EAAE,aAqB3B,CAAC;AAeF,eAAO,MAAM,oBAAoB,EAAE,aAoBlC,CAAC;AAeF,eAAO,MAAM,WAAW,EAAE,aAoBzB,CAAC;AAeF,eAAO,MAAM,qBAAqB,EAAE,aAuBnC,CAAC;AAEF,eAAO,MAAM,MAAM,EAAE,aAAa,EAKjC,CAAC"}

package/dist/criteria/lesson.js

@@ -0,0 +1,143 @@
+const topicAccuracyRubric = {
+    criterion: 'Topic Accuracy',
+    description: 'Does the content accurately cover the stated topic? Is the information correct and relevant?',
+    scale: [
+        { score: 1, label: 'Off-topic', description: 'Content does not address the stated topic' },
+        { score: 2, label: 'Tangential', description: 'Loosely related but misses core concepts' },
+        { score: 3, label: 'Accurate', description: 'Covers the topic correctly with minor gaps' },
+        { score: 4, label: 'Thorough', description: 'Comprehensive coverage, accurate details' },
+        { score: 5, label: 'Expert-level', description: 'Deep, nuanced, no factual issues, adds genuine insight' },
+    ],
+};
+export const topicAccuracy = {
+    name: 'topic_accuracy',
+    description: 'Content accurately covers the stated topic',
+    contentTypes: ['lesson'],
+    method: 'llm_judge',
+    threshold: 0.5,
+    weight: 1.5,
+    async evaluate(input, judge) {
+        if (!judge)
+            throw new Error('Judge required for topic_accuracy criterion');
+        const context = input.topic ? `Expected topic: ${input.topic}` : undefined;
+        const result = await judge.score(input.content, topicAccuracyRubric, context);
+        const score = (result.score - 1) / 4;
+        return {
+            criterion: 'topic_accuracy',
+            score,
+            rawScore: result.score,
+            maxScore: 5,
+            passed: score >= this.threshold,
+            reasoning: result.reasoning,
+        };
+    },
+};
+const pedagogicalStructureRubric = {
+    criterion: 'Pedagogical Structure',
+    description: 'Does the lesson follow good teaching structure? Look for: intuition building → concrete examples → hands-on practice → synthesis/takeaways.',
+    scale: [
+        { score: 1, label: 'No structure', description: 'Random facts dumped with no progression' },
+        { score: 2, label: 'Weak structure', description: 'Some organization but missing key phases (e.g. no examples or no practice)' },
+        { score: 3, label: 'Solid structure', description: 'Clear progression from concept to practice, covers main phases' },
+        { score: 4, label: 'Strong pedagogy', description: 'Well-scaffolded with intuition, examples, practice, and synthesis' },
+        { score: 5, label: 'Masterful', description: 'Perfect scaffolding, builds mental models, practice reinforces theory' },
+    ],
+};
+export const pedagogicalStructure = {
+    name: 'pedagogical_structure',
+    description: 'Lesson follows intuition → examples → practice → synthesis',
+    contentTypes: ['lesson'],
+    method: 'llm_judge',
+    threshold: 0.5,
+    weight: 1.5,
+    async evaluate(input, judge) {
+        if (!judge)
+            throw new Error('Judge required for pedagogical_structure criterion');
+        const result = await judge.score(input.content, pedagogicalStructureRubric);
+        const score = (result.score - 1) / 4;
+        return {
+            criterion: 'pedagogical_structure',
+            score,
+            rawScore: result.score,
+            maxScore: 5,
+            passed: score >= this.threshold,
+            reasoning: result.reasoning,
+        };
+    },
+};
+const codeQualityRubric = {
+    criterion: 'Code Quality',
+    description: 'Are the code examples relevant, likely to run correctly, and well-explained? Consider: relevance to topic, correctness, comments/explanations, and practical applicability.',
+    scale: [
+        { score: 1, label: 'Broken', description: 'Code has obvious errors or is irrelevant to the topic' },
+        { score: 2, label: 'Weak', description: 'Code runs but is poorly explained or only loosely relevant' },
+        { score: 3, label: 'Good', description: 'Relevant, likely correct code with adequate explanation' },
+        { score: 4, label: 'Strong', description: 'Clean, well-commented code that directly reinforces the lesson' },
+        { score: 5, label: 'Exemplary', description: 'Production-quality code, excellent comments, teaches through the code itself' },
+    ],
+};
+export const codeQuality = {
+    name: 'code_quality',
+    description: 'Code examples are relevant, correct, and well-explained',
+    contentTypes: ['lesson'],
+    method: 'llm_judge',
+    threshold: 0.5,
+    weight: 1.0,
+    async evaluate(input, judge) {
+        if (!judge)
+            throw new Error('Judge required for code_quality criterion');
+        const result = await judge.score(input.content, codeQualityRubric);
+        const score = (result.score - 1) / 4;
+        return {
+            criterion: 'code_quality',
+            score,
+            rawScore: result.score,
+            maxScore: 5,
+            passed: score >= this.threshold,
+            reasoning: result.reasoning,
+        };
+    },
+};
+const progressiveDifficultyRubric = {
+    criterion: 'Progressive Difficulty',
+    description: 'Does the content build appropriately on prerequisites? Is the difficulty level well-calibrated for the stated level?',
+    scale: [
+        { score: 1, label: 'Mismatched', description: 'Far too easy or hard for the stated level, no scaffolding' },
+        { score: 2, label: 'Poorly calibrated', description: 'Difficulty jumps around or assumes too much/little' },
+        { score: 3, label: 'Appropriate', description: 'Matches stated difficulty, reasonable prerequisite assumptions' },
+        { score: 4, label: 'Well-calibrated', description: 'Smooth difficulty curve, clear about prerequisites' },
+        { score: 5, label: 'Perfect scaffolding', description: 'Masterfully builds from known to unknown at exactly the right pace' },
+    ],
+};
+export const progressiveDifficulty = {
+    name: 'progressive_difficulty',
+    description: 'Content builds appropriately on prerequisites',
+    contentTypes: ['lesson'],
+    method: 'llm_judge',
+    threshold: 0.5,
+    weight: 1.0,
+    async evaluate(input, judge) {
+        if (!judge)
+            throw new Error('Judge required for progressive_difficulty criterion');
+        const context = input.difficulty
+            ? `Expected difficulty level: ${input.difficulty}`
+            : undefined;
+        const result = await judge.score(input.content, progressiveDifficultyRubric, context);
+        const score = (result.score - 1) / 4;
+        return {
+            criterion: 'progressive_difficulty',
+            score,
+            rawScore: result.score,
+            maxScore: 5,
+            passed: score >= this.threshold,
+            reasoning: result.reasoning,
+        };
+    },
+};
+export const lesson = [
+    topicAccuracy,
+    pedagogicalStructure,
+    codeQuality,
+    progressiveDifficulty,
+];
+//# sourceMappingURL=lesson.js.map

package/dist/criteria/lesson.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"lesson.js","sourceRoot":"","sources":["../../src/criteria/lesson.ts"],"names":[],"mappings":"AAEA,MAAM,mBAAmB,GAAgB;IACvC,SAAS,EAAE,gBAAgB;IAC3B,WAAW,EAAE,8FAA8F;IAC3G,KAAK,EAAE;QACL,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,WAAW,EAAE,WAAW,EAAE,2CAA2C,EAAE;QAC1F,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,YAAY,EAAE,WAAW,EAAE,0CAA0C,EAAE;QAC1F,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,UAAU,EAAE,WAAW,EAAE,4CAA4C,EAAE;QAC1F,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,UAAU,EAAE,WAAW,EAAE,0CAA0C,EAAE;QACxF,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,cAAc,EAAE,WAAW,EAAE,wDAAwD,EAAE;KAC3G;CACF,CAAC;AAEF,MAAM,CAAC,MAAM,aAAa,GAAkB;IAC1C,IAAI,EAAE,gBAAgB;IACtB,WAAW,EAAE,4CAA4C;IACzD,YAAY,EAAE,CAAC,QAAQ,CAAC;IACxB,MAAM,EAAE,WAAW;IACnB,SAAS,EAAE,GAAG;IACd,MAAM,EAAE,GAAG;IACX,KAAK,CAAC,QAAQ,CAAC,KAAgB,EAAE,KAAa;QAC5C,IAAI,CAAC,KAAK;YAAE,MAAM,IAAI,KAAK,CAAC,6CAA6C,CAAC,CAAC;QAC3E,MAAM,OAAO,GAAG,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,mBAAmB,KAAK,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC;QAC3E,MAAM,MAAM,GAAG,MAAM,KAAK,CAAC,KAAK,CAAC,KAAK,CAAC,OAAO,EAAE,mBAAmB,EAAE,OAAO,CAAC,CAAC;QAC9E,MAAM,KAAK,GAAG,CAAC,MAAM,CAAC,KAAK,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC;QACrC,OAAO;YACL,SAAS,EAAE,gBAAgB;YAC3B,KAAK;YACL,QAAQ,EAAE,MAAM,CAAC,KAAK;YACtB,QAAQ,EAAE,CAAC;YACX,MAAM,EAAE,KAAK,IAAI,IAAI,CAAC,SAAS;YAC/B,SAAS,EAAE,MAAM,CAAC,SAAS;SAC5B,CAAC;IACJ,CAAC;CACF,CAAC;AAEF,MAAM,0BAA0B,GAAgB;IAC9C,SAAS,EAAE,uBAAuB;IAClC,WAAW,EACT,6IAA6I;IAC/I,KAAK,EAAE;QACL,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,cAAc,EAAE,WAAW,EAAE,yCAAyC,EAAE;QAC3F,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,gBAAgB,EAAE,WAAW,EAAE,4EAA4E,EAAE;QAChI,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,iBAAiB,EAAE,WAAW,EAAE,gEAAgE,EAAE;QACrH,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,iBAAiB,EAAE,WAAW,EAAE,mEAAmE,EAAE;QACxH,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,WAAW,EAAE,WAAW,EAAE,uEAAuE,EAAE;KACvH;CACF,CAAC;AAEF,MAAM,CAAC,MAAM,oBAAoB,GAAkB;IACjD,IAAI,EAAE,uBAAuB;IAC7B,WAAW,EAAE,4DAA4D;IACzE,YAAY,EAAE,CAAC,QAAQ,CAAC;IACxB,MAAM,EAAE,WAAW;IACnB,SAAS,EAAE,GAAG;IACd,MAAM,EAAE,GAAG;IACX,KAAK,CAAC,QAAQ,CAAC,KAAgB,EAAE,KAAa;QAC5C,IAAI,CAAC,KAAK;YAAE,MAAM,IAAI,KAAK,CAAC,oDAAoD,CAAC,CAAC;QAClF,MAAM,MAAM,GAAG,MAAM,KAAK,CAAC,KAAK,CAAC,KAAK,CAAC,OAAO,EAAE,0BAA0B,CAAC,CAAC;QAC5E,MAAM,KAAK,GAAG,CAAC,MAAM,CAAC,KAAK,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC;QACrC,OAAO;YACL,SAAS,EAAE,uBAAuB;YAClC,KAAK;YACL,QAAQ,EAAE,MAAM,CAAC,KAAK;YACtB,QAAQ,EAAE,CAAC;YACX,MAAM,EAAE,KAAK,IAAI,IAAI,CAAC,SAAS;YAC/B,SAAS,EAAE,MAAM,CAAC,SAAS;SAC5B,CAAC;IACJ,CAAC;CACF,CAAC;AAEF,MAAM,iBAAiB,GAAgB;IACrC,SAAS,EAAE,cAAc;IACzB,WAAW,EACT,6KAA6K;IAC/K,KAAK,EAAE;QACL,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,QAAQ,EAAE,WAAW,EAAE,uDAAuD,EAAE;QACnG,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,WAAW,EAAE,4DAA4D,EAAE;QACtG,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,WAAW,EAAE,yDAAyD,EAAE;QACnG,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,QAAQ,EAAE,WAAW,EAAE,gEAAgE,EAAE;QAC5G,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,WAAW,EAAE,WAAW,EAAE,8EAA8E,EAAE;KAC9H;CACF,CAAC;AAEF,MAAM,CAAC,MAAM,WAAW,GAAkB;IACxC,IAAI,EAAE,cAAc;IACpB,WAAW,EAAE,yDAAyD;IACtE,YAAY,EAAE,CAAC,QAAQ,CAAC;IACxB,MAAM,EAAE,WAAW;IACnB,SAAS,EAAE,GAAG;IACd,MAAM,EAAE,GAAG;IACX,KAAK,CAAC,QAAQ,CAAC,KAAgB,EAAE,KAAa;QAC5C,IAAI,CAAC,KAAK;YAAE,MAAM,IAAI,KAAK,CAAC,2CAA2C,CAAC,CAAC;QACzE,MAAM,MAAM,GAAG,MAAM,KAAK,CAAC,KAAK,CAAC,KAAK,CAAC,OAAO,EAAE,iBAAiB,CAAC,CAAC;QACnE,MAAM,KAAK,GAAG,CAAC,MAAM,CAAC,KAAK,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC;QACrC,OAAO;YACL,SAAS,EAAE,cAAc;YACzB,KAAK;YACL,QAAQ,EAAE,MAAM,CAAC,KAAK;YACtB,QAAQ,EAAE,CAAC;YACX,MAAM,EAAE,KAAK,IAAI,IAAI,CAAC,SAAS;YAC/B,SAAS,EAAE,MAAM,CAAC,SAAS;SAC5B,CAAC;IACJ,CAAC;CACF,CAAC;AAEF,MAAM,2BAA2B,GAAgB;IAC/C,SAAS,EAAE,wBAAwB;IACnC,WAAW,EACT,sHAAsH;IACxH,KAAK,EAAE;QACL,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,YAAY,EAAE,WAAW,EAAE,2DAA2D,EAAE;QAC3G,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,mBAAmB,EAAE,WAAW,EAAE,oDAAoD,EAAE;QAC3G,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,aAAa,EAAE,WAAW,EAAE,gEAAgE,EAAE;QACjH,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,iBAAiB,EAAE,WAAW,EAAE,oDAAoD,EAAE;QACzG,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,qBAAqB,EAAE,WAAW,EAAE,oEAAoE,EAAE;KAC9H;CACF,CAAC;AAEF,MAAM,CAAC,MAAM,qBAAqB,GAAkB;IAClD,IAAI,EAAE,wBAAwB;IAC9B,WAAW,EAAE,+CAA+C;IAC5D,YAAY,EAAE,CAAC,QAAQ,CAAC;IACxB,MAAM,EAAE,WAAW;IACnB,SAAS,EAAE,GAAG;IACd,MAAM,EAAE,GAAG;IACX,KAAK,CAAC,QAAQ,CAAC,KAAgB,EAAE,KAAa;QAC5C,IAAI,CAAC,KAAK;YAAE,MAAM,IAAI,KAAK,CAAC,qDAAqD,CAAC,CAAC;QACnF,MAAM,OAAO,GAAG,KAAK,CAAC,UAAU;YAC9B,CAAC,CAAC,8BAA8B,KAAK,CAAC,UAAU,EAAE;YAClD,CAAC,CAAC,SAAS,CAAC;QACd,MAAM,MAAM,GAAG,MAAM,KAAK,CAAC,KAAK,CAAC,KAAK,CAAC,OAAO,EAAE,2BAA2B,EAAE,OAAO,CAAC,CAAC;QACtF,MAAM,KAAK,GAAG,CAAC,MAAM,CAAC,KAAK,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC;QACrC,OAAO;YACL,SAAS,EAAE,wBAAwB;YACnC,KAAK;YACL,QAAQ,EAAE,MAAM,CAAC,KAAK;YACtB,QAAQ,EAAE,CAAC;YACX,MAAM,EAAE,KAAK,IAAI,IAAI,CAAC,SAAS;YAC/B,SAAS,EAAE,MAAM,CAAC,SAAS;SAC5B,CAAC;IACJ,CAAC;CACF,CAAC;AAEF,MAAM,CAAC,MAAM,MAAM,GAAoB;IACrC,aAAa;IACb,oBAAoB;IACpB,WAAW;IACX,qBAAqB;CACtB,CAAC"}

package/dist/criteria/review.d.ts

@@ -0,0 +1,6 @@
+import type { EvalCriterion } from '../types.js';
+export declare const brevity: EvalCriterion;
+export declare const actionability: EvalCriterion;
+export declare const honesty: EvalCriterion;
+export declare const review: EvalCriterion[];
+//# sourceMappingURL=review.d.ts.map

package/dist/criteria/review.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"review.d.ts","sourceRoot":"","sources":["../../src/criteria/review.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAA4C,MAAM,aAAa,CAAC;AAI3F,eAAO,MAAM,OAAO,EAAE,aAyBrB,CAAC;AAeF,eAAO,MAAM,aAAa,EAAE,aAoB3B,CAAC;AAeF,eAAO,MAAM,OAAO,EAAE,aAoBrB,CAAC;AAEF,eAAO,MAAM,MAAM,EAAE,aAAa,EAAsC,CAAC"}

package/dist/criteria/review.js

@@ -0,0 +1,93 @@
+const DEFAULT_REVIEW_MAX_CHARS = 2000;
+export const brevity = {
+    name: 'brevity',
+    description: 'Review content is concise and under character budget',
+    contentTypes: ['review'],
+    method: 'deterministic',
+    threshold: 1.0,
+    weight: 0.5,
+    async evaluate(input) {
+        const maxChars = input.metadata?.maxReviewChars ?? DEFAULT_REVIEW_MAX_CHARS;
+        const len = input.content.length;
+        const passed = len <= maxChars;
+        const score = passed ? 1.0 : Math.max(0, 1 - (len - maxChars) / maxChars);
+        return {
+            criterion: 'brevity',
+            score,
+            rawScore: len,
+            maxScore: maxChars,
+            passed,
+            reasoning: passed
+                ? `Review is ${len} chars, within ${maxChars} char budget`
+                : `Review is ${len} chars, exceeds ${maxChars} char budget`,
+        };
+    },
+};
+const actionabilityRubric = {
+    criterion: 'Actionability',
+    description: 'Does the review contain concrete, specific next steps the reader can take? Not vague encouragement but real actions.',
+    scale: [
+        { score: 1, label: 'No actions', description: 'Pure commentary with no next steps' },
+        { score: 2, label: 'Vague', description: 'Generic advice like "keep practicing" without specifics' },
+        { score: 3, label: 'Some actions', description: 'Contains at least one concrete next step' },
+        { score: 4, label: 'Actionable', description: 'Multiple specific, prioritized next steps' },
+        { score: 5, label: 'Highly actionable', description: 'Clear roadmap with prioritized, specific, measurable actions' },
+    ],
+};
+export const actionability = {
+    name: 'actionability',
+    description: 'Review contains concrete next steps',
+    contentTypes: ['review'],
+    method: 'llm_judge',
+    threshold: 0.5,
+    weight: 1.0,
+    async evaluate(input, judge) {
+        if (!judge)
+            throw new Error('Judge required for actionability criterion');
+        const result = await judge.score(input.content, actionabilityRubric);
+        const score = (result.score - 1) / 4;
+        return {
+            criterion: 'actionability',
+            score,
+            rawScore: result.score,
+            maxScore: 5,
+            passed: score >= this.threshold,
+            reasoning: result.reasoning,
+        };
+    },
+};
+const honestyRubric = {
+    criterion: 'Honesty',
+    description: "Does the review honestly assess gaps and weaknesses? Does it avoid sugarcoating or false encouragement?",
+    scale: [
+        { score: 1, label: 'Dishonest', description: 'Pure cheerleading, ignores obvious gaps' },
+        { score: 2, label: 'Sugarcoated', description: 'Acknowledges issues but softens them to the point of uselessness' },
+        { score: 3, label: 'Balanced', description: 'Honest about strengths and weaknesses, constructive tone' },
+        { score: 4, label: 'Direct', description: 'Clear-eyed assessment, names gaps specifically while remaining respectful' },
+        { score: 5, label: 'Brutally honest', description: 'Unflinching but constructive, prioritizes gaps, no wasted praise' },
+    ],
+};
+export const honesty = {
+    name: 'honesty',
+    description: "Review honestly assesses gaps without sugarcoating",
+    contentTypes: ['review'],
+    method: 'llm_judge',
+    threshold: 0.5,
+    weight: 1.0,
+    async evaluate(input, judge) {
+        if (!judge)
+            throw new Error('Judge required for honesty criterion');
+        const result = await judge.score(input.content, honestyRubric);
+        const score = (result.score - 1) / 4;
+        return {
+            criterion: 'honesty',
+            score,
+            rawScore: result.score,
+            maxScore: 5,
+            passed: score >= this.threshold,
+            reasoning: result.reasoning,
+        };
+    },
+};
+export const review = [brevity, actionability, honesty];
+//# sourceMappingURL=review.js.map