observability-toolkit 2.0.0 → 2.1.0

package/dist/lib/judge/llm-judge-code.js

@@ -0,0 +1,261 @@
+/**
+ * Code Evaluation Pattern
+ *
+ * Three-stage evaluation: syntax validation → runtime error detection → LLM semantic evaluation.
+ * Score aggregation: configurable per-language syntax/semantic weights (JS/TS: 30/70, Python: 20/80, Go: 35/65).
+ * Language-specific validators: TypeScript/JavaScript (built-in), Python (python3), Go (gofmt).
+ *
+ * @see llm-as-judge.ts for core types
+ */
+import { spawn } from 'node:child_process';
+import { z } from 'zod';
+import { normalizedScoreSchema } from '../core/shared-schemas.js';
+import { sanitizeForPrompt, validateTestCase, withTimeout, DEFAULT_LLM_TIMEOUT_MS, } from './llm-judge-security.js';
+import { G_EVAL_MIN_SCORE, G_EVAL_SCORE_RANGE, CODE_EVAL_SYNTAX_WEIGHT, CODE_EVAL_SYNTAX_WEIGHT_PYTHON, CODE_EVAL_SYNTAX_WEIGHT_GO, CODE_EVAL_SYNTAX_SCORE_PASS, CODE_EVAL_SYNTAX_SCORE_FAIL, CODE_EVAL_DEFAULT_CRITERIA, CODE_EVAL_DEFAULT_LANGUAGE, CODE_EVAL_RUNTIME_ERROR_SCORE_DETECTED, CODE_EVAL_RUNTIME_ERROR_SCORE_CLEAN, CODE_EVAL_RUNTIME_ERROR_WEIGHT_DEFAULT, } from './llm-judge-constants.js';
+import { extractScoreFromText } from './llm-judge-geval.js';
+// ============================================================================
+// Types
+// ============================================================================
+export const supportedLanguageSchema = z.enum(['typescript', 'javascript', 'python', 'go']);
+/**
+ * Zod schema for CodeEvalConfig.
+ * - `syntaxWeight`: 0.0–1.0; semantic weight = 1 - syntaxWeight; per-language default when omitted (Python: 0.2, Go: 0.35, JS/TS: 0.3).
+ * - `runtimeErrorPatterns`: replaces built-in patterns when defined; `[]` disables detection entirely.
+ * - `runtimeErrorWeight`: penalty weight applied as `runtimeErrorWeight * (1 - runtimeErrorScore)`; default 0 (no effect).
+ */
+export const codeEvalConfigSchema = z.object({
+    language: supportedLanguageSchema.optional(),
+    checkSyntax: z.boolean().default(true),
+    runtimeErrorDetection: z.boolean().default(false),
+    evaluationCriteria: z.array(z.string()).default(CODE_EVAL_DEFAULT_CRITERIA),
+    syntaxWeight: normalizedScoreSchema.optional(),
+    runtimeErrorPatterns: z.array(z.custom((v) => v instanceof RegExp)).optional(),
+    runtimeErrorWeight: normalizedScoreSchema.default(CODE_EVAL_RUNTIME_ERROR_WEIGHT_DEFAULT),
+});
+// ============================================================================
+// Weight Resolution
+// ============================================================================
+/** Normalize a language name to a `SupportedLanguage`, defaulting to `CODE_EVAL_DEFAULT_LANGUAGE` for unknown values. */
+export function normalizeLang(language) {
+    const lower = (language ?? '').toLowerCase();
+    const result = supportedLanguageSchema.safeParse(lower);
+    return result.success ? result.data : CODE_EVAL_DEFAULT_LANGUAGE;
+}
+/**
+ * Resolve the effective syntax weight for a given language and optional config override.
+ * Returns `{ syntaxWeight, semanticWeight }` where both sum to 1.0.
+ */
+export function resolveWeights(language, syntaxWeightOverride) {
+    if (syntaxWeightOverride !== undefined) {
+        const clamped = Math.min(1, Math.max(0, syntaxWeightOverride));
+        return { syntaxWeight: clamped, semanticWeight: 1 - clamped };
+    }
+    const lang = normalizeLang(language);
+    const syntaxWeight = lang === 'python' ? CODE_EVAL_SYNTAX_WEIGHT_PYTHON :
+        lang === 'go' ? CODE_EVAL_SYNTAX_WEIGHT_GO :
+            CODE_EVAL_SYNTAX_WEIGHT;
+    return { syntaxWeight, semanticWeight: 1 - syntaxWeight };
+}
+// ============================================================================
+// Syntax Validation
+// ============================================================================
+function checkSyntaxValidity(code) {
+    // Strip TypeScript-specific syntax for Function constructor check
+    const jsCode = code
+        .replace(/:\s*\w+(\[\])?(\s*\|[^,);{]+)*/g, '') // type annotations
+        .replace(/<[^>]*>/g, '') // generics
+        .replace(/\basync\b/g, '') // async keyword
+        .replace(/\bawait\b/g, ''); // await keyword
+    try {
+        // eslint-disable-next-line no-new-func
+        new Function(jsCode);
+        return CODE_EVAL_SYNTAX_SCORE_PASS;
+    }
+    catch {
+        return CODE_EVAL_SYNTAX_SCORE_FAIL;
+    }
+}
+/** Async wrapper: pipe `input` via stdin to an external process; resolve PASS if exit code is 0, FAIL otherwise. */
+function spawnSyntaxCheckAsync(cmd, args, input) {
+    return new Promise((resolve, reject) => {
+        let settled = false;
+        const child = spawn(cmd, args, { stdio: ['pipe', 'pipe', 'pipe'] });
+        child.stdin.write(input, 'utf8');
+        child.stdin.end();
+        child.on('close', (code) => {
+            if (!settled) {
+                settled = true;
+                resolve(code === 0 ? CODE_EVAL_SYNTAX_SCORE_PASS : CODE_EVAL_SYNTAX_SCORE_FAIL);
+            }
+        });
+        child.on('error', (err) => {
+            if (!settled) {
+                settled = true;
+                reject(err);
+            }
+        });
+    });
+}
+/**
+ * Validate Python syntax via stdin: `python3 -c "import sys,ast; ast.parse(sys.stdin.read())"`.
+ * Returns PASS if python3 is available and the code parses without errors, FAIL otherwise.
+ */
+export async function checkPythonSyntax(code) {
+    return spawnSyntaxCheckAsync('python3', ['-c', 'import sys, ast; ast.parse(sys.stdin.read())'], code);
+}
+/**
+ * Validate Go syntax via stdin to `gofmt`.
+ * Returns PASS if gofmt is available and the code parses without errors, FAIL otherwise.
+ * Note: gofmt does not require a `package` declaration for snippet validation.
+ */
+export async function checkGoSyntax(code) {
+    return spawnSyntaxCheckAsync('gofmt', [], code);
+}
+/**
+ * Dispatch to the appropriate language-specific syntax validator.
+ * Falls back to JS/TS validator for unknown or unspecified languages.
+ */
+async function runSyntaxCheck(code, language) {
+    const lang = normalizeLang(language);
+    if (lang === 'python') {
+        return checkPythonSyntax(code);
+    }
+    if (lang === 'go') {
+        return checkGoSyntax(code);
+    }
+    return checkSyntaxValidity(code);
+}
+// ============================================================================
+// Runtime Error Detection
+// ============================================================================
+/** Heuristic patterns that indicate likely runtime errors in JS/TS code */
+const RUNTIME_ERROR_PATTERNS_JS = [
+    /\bnull\s*\.\s*\w+/,
+    /\bundefined\s*\.\s*\w+/,
+    /\/\s*0\b/,
+];
+/** Heuristic patterns that indicate likely runtime errors in Python code */
+const RUNTIME_ERROR_PATTERNS_PYTHON = [
+    /\bNone\s*\.\s*\w+/,
+    /\/\s*0\b/,
+    /raise\s+\w+/, // false-positive risk: flags intentional error handling (e.g. `raise ValueError`)
+];
+/** Heuristic patterns that indicate likely runtime errors in Go code */
+const RUNTIME_ERROR_PATTERNS_GO = [
+    /\bnil\s*\.\s*\w+/,
+    /\/\s*0\b/,
+    /panic\s*\(/, // false-positive risk: flags intentional panics (e.g. `panic("unreachable")`)
+];
+/**
+ * Detect common runtime error patterns in code using language-specific heuristics.
+ * Returns CLEAN (1.0) when no patterns are detected, DETECTED (0.0) when patterns are found.
+ *
+ * **False-positive risk**: Python `raise\s+\w+` and Go `panic\s*\(` match intentional error
+ * handling. When `runtimeErrorWeight` is 0 (default) this is informational-only; set
+ * `runtimeErrorDetection: false` or pass `runtimeErrorPatterns: []` to suppress entirely.
+ *
+ * @param patterns - Optional override. When defined (even `[]`), replaces built-in patterns.
+ *   Pass `[]` to disable detection entirely (always returns CLEAN).
+ */
+export function detectRuntimeErrors(code, language, patterns) {
+    let activePatterns;
+    if (patterns !== undefined) {
+        activePatterns = patterns;
+    }
+    else {
+        const lang = normalizeLang(language);
+        if (lang === 'python') {
+            activePatterns = RUNTIME_ERROR_PATTERNS_PYTHON;
+        }
+        else if (lang === 'go') {
+            activePatterns = RUNTIME_ERROR_PATTERNS_GO;
+        }
+        else {
+            activePatterns = RUNTIME_ERROR_PATTERNS_JS;
+        }
+    }
+    const hasRuntimeError = activePatterns.some((pattern) => pattern.test(code));
+    return hasRuntimeError ? CODE_EVAL_RUNTIME_ERROR_SCORE_DETECTED : CODE_EVAL_RUNTIME_ERROR_SCORE_CLEAN;
+}
+// ============================================================================
+// Prompt Building
+// ============================================================================
+function buildCodeEvalPrompt(testCase, criteria) {
+    const sanitizedInput = sanitizeForPrompt(testCase.input);
+    const sanitizedOutput = sanitizeForPrompt(testCase.output);
+    const criteriaList = criteria.join(', ');
+    return [
+        'Evaluate the following code output against the stated intent.',
+        '',
+        `Intent: ${sanitizedInput}`,
+        '',
+        `Code Output:\n${sanitizedOutput}`,
+        '',
+        `Evaluation Criteria: ${criteriaList}`,
+        '',
+        'Base your assessment only on the code text provided. If the intent is ambiguous, state so explicitly before scoring. Do not infer runtime behavior that cannot be determined from static reading alone.',
+        '',
+        'Provide your assessment and end with "Score: N" where N is 1-5.',
+        '1 = Completely fails, 2 = Major issues, 3 = Adequate, 4 = Good, 5 = Excellent',
+    ].join('\n');
+}
+export async function codeEvaluation(llm, testCase, config, timeoutMs = DEFAULT_LLM_TIMEOUT_MS) {
+    validateTestCase(testCase);
+    if (testCase.output.trim().length === 0) {
+        throw new Error('Invalid input: output must not be empty');
+    }
+    if (testCase.input.trim().length === 0) {
+        throw new Error('Invalid input: input must not be empty');
+    }
+    const parseResult = codeEvalConfigSchema.safeParse(config ?? {});
+    if (!parseResult.success) {
+        throw new TypeError(`Invalid CodeEvalConfig: ${parseResult.error.message}`);
+    }
+    const cfg = parseResult.data;
+    const { syntaxWeight, semanticWeight } = resolveWeights(cfg.language, cfg.syntaxWeight);
+    // Stage 1: Syntax validation (language-aware)
+    let syntaxScore;
+    if (cfg.checkSyntax) {
+        syntaxScore = await runSyntaxCheck(testCase.output, cfg.language);
+    }
+    // Stage 2: Runtime error detection (optional heuristic pass before LLM)
+    let runtimeErrorScore;
+    if (cfg.runtimeErrorDetection) {
+        runtimeErrorScore = detectRuntimeErrors(testCase.output, cfg.language, cfg.runtimeErrorPatterns);
+    }
+    // Stage 3: Semantic evaluation via LLM
+    const prompt = buildCodeEvalPrompt(testCase, cfg.evaluationCriteria);
+    const response = await withTimeout((signal) => llm.generate(prompt, { signal }), timeoutMs);
+    const rawScore = extractScoreFromText(response.text);
+    const semanticScore = (rawScore - G_EVAL_MIN_SCORE) / G_EVAL_SCORE_RANGE;
+    function applyRuntimePenalty(baseScore) {
+        if (cfg.runtimeErrorWeight > 0 && runtimeErrorScore !== undefined) {
+            const penalized = baseScore - cfg.runtimeErrorWeight * (1 - runtimeErrorScore);
+            return Math.min(1, Math.max(0, penalized));
+        }
+        return baseScore;
+    }
+    if (cfg.checkSyntax && syntaxScore !== undefined) {
+        const rawAggregateScore = syntaxWeight * syntaxScore + semanticWeight * semanticScore;
+        const score = applyRuntimePenalty(rawAggregateScore);
+        return {
+            score,
+            reason: response.text,
+            rawResponse: response.text,
+            semanticScore,
+            syntaxScore,
+            ...(runtimeErrorScore !== undefined && { runtimeErrorScore }),
+            evaluatorType: 'llm',
+        };
+    }
+    const score = applyRuntimePenalty(semanticScore);
+    return {
+        score,
+        reason: response.text,
+        rawResponse: response.text,
+        semanticScore,
+        ...(runtimeErrorScore !== undefined && { runtimeErrorScore }),
+        evaluatorType: 'llm',
+    };
+}
+//# sourceMappingURL=llm-judge-code.js.map

package/dist/lib/judge/llm-judge-code.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"llm-judge-code.js","sourceRoot":"","sources":["../../../src/lib/judge/llm-judge-code.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAEH,OAAO,EAAE,KAAK,EAAE,MAAM,oBAAoB,CAAC;AAC3C,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,EAAE,qBAAqB,EAAE,MAAM,2BAA2B,CAAC;AAElE,OAAO,EACL,iBAAiB,EACjB,gBAAgB,EAChB,WAAW,EACX,sBAAsB,GACvB,MAAM,yBAAyB,CAAC;AACjC,OAAO,EACL,gBAAgB,EAChB,kBAAkB,EAClB,uBAAuB,EACvB,8BAA8B,EAC9B,0BAA0B,EAC1B,2BAA2B,EAC3B,2BAA2B,EAC3B,0BAA0B,EAC1B,0BAA0B,EAC1B,sCAAsC,EACtC,mCAAmC,EACnC,sCAAsC,GACvC,MAAM,0BAA0B,CAAC;AAClC,OAAO,EAAE,oBAAoB,EAAE,MAAM,sBAAsB,CAAC;AAE5D,+EAA+E;AAC/E,QAAQ;AACR,+EAA+E;AAE/E,MAAM,CAAC,MAAM,uBAAuB,GAAG,CAAC,CAAC,IAAI,CAAC,CAAC,YAAY,EAAE,YAAY,EAAE,QAAQ,EAAE,IAAI,CAAC,CAAC,CAAC;AAG5F;;;;;GAKG;AACH,MAAM,CAAC,MAAM,oBAAoB,GAAG,CAAC,CAAC,MAAM,CAAC;IAC3C,QAAQ,EAAE,uBAAuB,CAAC,QAAQ,EAAE;IAC5C,WAAW,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,IAAI,CAAC;IACtC,qBAAqB,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC;IACjD,kBAAkB,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,OAAO,CAAC,0BAA0B,CAAC;IAC3E,YAAY,EAAE,qBAAqB,CAAC,QAAQ,EAAE;IAC9C,oBAAoB,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAS,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,YAAY,MAAM,CAAC,CAAC,CAAC,QAAQ,EAAE;IACtF,kBAAkB,EAAE,qBAAqB,CAAC,OAAO,CAAC,sCAAsC,CAAC;CAC1F,CAAC,CAAC;AA6BH,+EAA+E;AAC/E,oBAAoB;AACpB,+EAA+E;AAE/E,yHAAyH;AACzH,MAAM,UAAU,aAAa,CAAC,QAAiB;IAC7C,MAAM,KAAK,GAAG,CAAC,QAAQ,IAAI,EAAE,CAAC,CAAC,WAAW,EAAE,CAAC;IAC7C,MAAM,MAAM,GAAG,uBAAuB,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;IACxD,OAAO,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,0BAA0B,CAAC;AACnE,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,cAAc,CAAC,QAA4B,EAAE,oBAA6B;IACxF,IAAI,oBAAoB,KAAK,SAAS,EAAE,CAAC;QACvC,MAAM,OAAO,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,oBAAoB,CAAC,CAAC,CAAC;QAC/D,OAAO,EAAE,YAAY,EAAE,OAAO,EAAE,cAAc,EAAE,CAAC,GAAG,OAAO,EAAE,CAAC;IAChE,CAAC;IACD,MAAM,IAAI,GAAG,aAAa,CAAC,QAAQ,CAAC,CAAC;IACrC,MAAM,YAAY,GAChB,IAAI,KAAK,QAAQ,CAAC,CAAC,CAAC,8BAA8B,CAAC,CAAC;QACpD,IAAI,KAAK,IAAI,CAAC,CAAC,CAAC,0BAA0B,CAAC,CAAC;YAC5C,uBAAuB,CAAC;IAC1B,OAAO,EAAE,YAAY,EAAE,cAAc,EAAE,CAAC,GAAG,YAAY,EAAE,CAAC;AAC5D,CAAC;AAED,+EAA+E;AAC/E,oBAAoB;AACpB,+EAA+E;AAE/E,SAAS,mBAAmB,CAAC,IAAY;IACvC,kEAAkE;IAClE,MAAM,MAAM,GAAG,IAAI;SAChB,OAAO,CAAC,iCAAiC,EAAE,EAAE,CAAC,CAAE,mBAAmB;SACnE,OAAO,CAAC,UAAU,EAAE,EAAE,CAAC,CAA4B,WAAW;SAC9D,OAAO,CAAC,YAAY,EAAE,EAAE,CAAC,CAA0B,gBAAgB;SACnE,OAAO,CAAC,YAAY,EAAE,EAAE,CAAC,CAAC,CAAyB,gBAAgB;IAEtE,IAAI,CAAC;QACH,uCAAuC;QACvC,IAAI,QAAQ,CAAC,MAAM,CAAC,CAAC;QACrB,OAAO,2BAA2B,CAAC;IACrC,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,2BAA2B,CAAC;IACrC,CAAC;AACH,CAAC;AAED,oHAAoH;AACpH,SAAS,qBAAqB,CAAC,GAAW,EAAE,IAAc,EAAE,KAAa;IACvE,OAAO,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE;QACrC,IAAI,OAAO,GAAG,KAAK,CAAC;QACpB,MAAM,KAAK,GAAG,KAAK,CAAC,GAAG,EAAE,IAAI,EAAE,EAAE,KAAK,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC,EAAE,CAAC,CAAC;QACpE,KAAK,CAAC,KAAK,CAAC,KAAK,CAAC,KAAK,EAAE,MAAM,CAAC,CAAC;QACjC,KAAK,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC;QAClB,KAAK,CAAC,EAAE,CAAC,OAAO,EAAE,CAAC,IAAI,EAAE,EAAE;YACzB,IAAI,CAAC,OAAO,EAAE,CAAC;gBAAC,OAAO,GAAG,IAAI,CAAC;gBAAC,OAAO,CAAC,IAAI,KAAK,CAAC,CAAC,CAAC,CAAC,2BAA2B,CAAC,CAAC,CAAC,2BAA2B,CAAC,CAAC;YAAC,CAAC;QACpH,CAAC,CAAC,CAAC;QACH,KAAK,CAAC,EAAE,CAAC,OAAO,EAAE,CAAC,GAAG,EAAE,EAAE;YACxB,IAAI,CAAC,OAAO,EAAE,CAAC;gBAAC,OAAO,GAAG,IAAI,CAAC;gBAAC,MAAM,CAAC,GAAG,CAAC,CAAC;YAAC,CAAC;QAChD,CAAC,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;AACL,CAAC;AAED;;;GAGG;AACH,MAAM,CAAC,KAAK,UAAU,iBAAiB,CAAC,IAAY;IAClD,OAAO,qBAAqB,CAAC,SAAS,EAAE,CAAC,IAAI,EAAE,8CAA8C,CAAC,EAAE,IAAI,CAAC,CAAC;AACxG,CAAC;AAED;;;;GAIG;AACH,MAAM,CAAC,KAAK,UAAU,aAAa,CAAC,IAAY;IAC9C,OAAO,qBAAqB,CAAC,OAAO,EAAE,EAAE,EAAE,IAAI,CAAC,CAAC;AAClD,CAAC;AAED;;;GAGG;AACH,KAAK,UAAU,cAAc,CAAC,IAAY,EAAE,QAA4B;IACtE,MAAM,IAAI,GAAG,aAAa,CAAC,QAAQ,CAAC,CAAC;IACrC,IAAI,IAAI,KAAK,QAAQ,EAAE,CAAC;QACtB,OAAO,iBAAiB,CAAC,IAAI,CAAC,CAAC;IACjC,CAAC;IACD,IAAI,IAAI,KAAK,IAAI,EAAE,CAAC;QAClB,OAAO,aAAa,CAAC,IAAI,CAAC,CAAC;IAC7B,CAAC;IACD,OAAO,mBAAmB,CAAC,IAAI,CAAC,CAAC;AACnC,CAAC;AAED,+EAA+E;AAC/E,0BAA0B;AAC1B,+EAA+E;AAE/E,2EAA2E;AAC3E,MAAM,yBAAyB,GAAa;IAC1C,mBAAmB;IACnB,wBAAwB;IACxB,UAAU;CACX,CAAC;AAEF,4EAA4E;AAC5E,MAAM,6BAA6B,GAAa;IAC9C,mBAAmB;IACnB,UAAU;IACV,aAAa,EAAE,kFAAkF;CAClG,CAAC;AAEF,wEAAwE;AACxE,MAAM,yBAAyB,GAAa;IAC1C,kBAAkB;IAClB,UAAU;IACV,YAAY,EAAE,8EAA8E;CAC7F,CAAC;AAEF;;;;;;;;;;GAUG;AACH,MAAM,UAAU,mBAAmB,CAAC,IAAY,EAAE,QAA4B,EAAE,QAAmB;IACjG,IAAI,cAAwB,CAAC;IAC7B,IAAI,QAAQ,KAAK,SAAS,EAAE,CAAC;QAC3B,cAAc,GAAG,QAAQ,CAAC;IAC5B,CAAC;SAAM,CAAC;QACN,MAAM,IAAI,GAAG,aAAa,CAAC,QAAQ,CAAC,CAAC;QACrC,IAAI,IAAI,KAAK,QAAQ,EAAE,CAAC;YACtB,cAAc,GAAG,6BAA6B,CAAC;QACjD,CAAC;aAAM,IAAI,IAAI,KAAK,IAAI,EAAE,CAAC;YACzB,cAAc,GAAG,yBAAyB,CAAC;QAC7C,CAAC;aAAM,CAAC;YACN,cAAc,GAAG,yBAAyB,CAAC;QAC7C,CAAC;IACH,CAAC;IACD,MAAM,eAAe,GAAG,cAAc,CAAC,IAAI,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC;IAC7E,OAAO,eAAe,CAAC,CAAC,CAAC,sCAAsC,CAAC,CAAC,CAAC,mCAAmC,CAAC;AACxG,CAAC;AAED,+EAA+E;AAC/E,kBAAkB;AAClB,+EAA+E;AAE/E,SAAS,mBAAmB,CAAC,QAAkB,EAAE,QAAkB;IACjE,MAAM,cAAc,GAAG,iBAAiB,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC;IACzD,MAAM,eAAe,GAAG,iBAAiB,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC;IAC3D,MAAM,YAAY,GAAG,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAEzC,OAAO;QACL,+DAA+D;QAC/D,EAAE;QACF,WAAW,cAAc,EAAE;QAC3B,EAAE;QACF,iBAAiB,eAAe,EAAE;QAClC,EAAE;QACF,wBAAwB,YAAY,EAAE;QACtC,EAAE;QACF,yMAAyM;QACzM,EAAE;QACF,iEAAiE;QACjE,+EAA+E;KAChF,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AACf,CAAC;AAoBD,MAAM,CAAC,KAAK,UAAU,cAAc,CAClC,GAAgB,EAChB,QAAkB,EAClB,MAAuB,EACvB,YAAoB,sBAAsB;IAE1C,gBAAgB,CAAC,QAAQ,CAAC,CAAC;IAE3B,IAAI,QAAQ,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACxC,MAAM,IAAI,KAAK,CAAC,yCAAyC,CAAC,CAAC;IAC7D,CAAC;IACD,IAAI,QAAQ,CAAC,KAAK,CAAC,IAAI,EAAE,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACvC,MAAM,IAAI,KAAK,CAAC,wCAAwC,CAAC,CAAC;IAC5D,CAAC;IAED,MAAM,WAAW,GAAG,oBAAoB,CAAC,SAAS,CAAC,MAAM,IAAI,EAAE,CAAC,CAAC;IACjE,IAAI,CAAC,WAAW,CAAC,OAAO,EAAE,CAAC;QACzB,MAAM,IAAI,SAAS,CAAC,2BAA2B,WAAW,CAAC,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC;IAC9E,CAAC;IACD,MAAM,GAAG,GAAG,WAAW,CAAC,IAAI,CAAC;IAE7B,MAAM,EAAE,YAAY,EAAE,cAAc,EAAE,GAAG,cAAc,CAAC,GAAG,CAAC,QAAQ,EAAE,GAAG,CAAC,YAAY,CAAC,CAAC;IAExF,8CAA8C;IAC9C,IAAI,WAA+B,CAAC;IACpC,IAAI,GAAG,CAAC,WAAW,EAAE,CAAC;QACpB,WAAW,GAAG,MAAM,cAAc,CAAC,QAAQ,CAAC,MAAM,EAAE,GAAG,CAAC,QAAQ,CAAC,CAAC;IACpE,CAAC;IAED,wEAAwE;IACxE,IAAI,iBAAqC,CAAC;IAC1C,IAAI,GAAG,CAAC,qBAAqB,EAAE,CAAC;QAC9B,iBAAiB,GAAG,mBAAmB,CAAC,QAAQ,CAAC,MAAM,EAAE,GAAG,CAAC,QAAQ,EAAE,GAAG,CAAC,oBAAoB,CAAC,CAAC;IACnG,CAAC;IAED,uCAAuC;IACvC,MAAM,MAAM,GAAG,mBAAmB,CAAC,QAAQ,EAAE,GAAG,CAAC,kBAAkB,CAAC,CAAC;IACrE,MAAM,QAAQ,GAAG,MAAM,WAAW,CAChC,CAAC,MAAM,EAAE,EAAE,CAAC,GAAG,CAAC,QAAQ,CAAC,MAAM,EAAE,EAAE,MAAM,EAAE,CAAC,EAC5C,SAAS,CACV,CAAC;IAEF,MAAM,QAAQ,GAAG,oBAAoB,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;IACrD,MAAM,aAAa,GAAG,CAAC,QAAQ,GAAG,gBAAgB,CAAC,GAAG,kBAAkB,CAAC;IAEzE,SAAS,mBAAmB,CAAC,SAAiB;QAC5C,IAAI,GAAG,CAAC,kBAAkB,GAAG,CAAC,IAAI,iBAAiB,KAAK,SAAS,EAAE,CAAC;YAClE,MAAM,SAAS,GAAG,SAAS,GAAG,GAAG,CAAC,kBAAkB,GAAG,CAAC,CAAC,GAAG,iBAAiB,CAAC,CAAC;YAC/E,OAAO,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,SAAS,CAAC,CAAC,CAAC;QAC7C,CAAC;QACD,OAAO,SAAS,CAAC;IACnB,CAAC;IAED,IAAI,GAAG,CAAC,WAAW,IAAI,WAAW,KAAK,SAAS,EAAE,CAAC;QACjD,MAAM,iBAAiB,GAAG,YAAY,GAAG,WAAW,GAAG,cAAc,GAAG,aAAa,CAAC;QACtF,MAAM,KAAK,GAAG,mBAAmB,CAAC,iBAAiB,CAAC,CAAC;QACrD,OAAO;YACL,KAAK;YACL,MAAM,EAAE,QAAQ,CAAC,IAAI;YACrB,WAAW,EAAE,QAAQ,CAAC,IAAI;YAC1B,aAAa;YACb,WAAW;YACX,GAAG,CAAC,iBAAiB,KAAK,SAAS,IAAI,EAAE,iBAAiB,EAAE,CAAC;YAC7D,aAAa,EAAE,KAAK;SACrB,CAAC;IACJ,CAAC;IAED,MAAM,KAAK,GAAG,mBAAmB,CAAC,aAAa,CAAC,CAAC;IACjD,OAAO;QACL,KAAK;QACL,MAAM,EAAE,QAAQ,CAAC,IAAI;QACrB,WAAW,EAAE,QAAQ,CAAC,IAAI;QAC1B,aAAa;QACb,GAAG,CAAC,iBAAiB,KAAK,SAAS,IAAI,EAAE,iBAAiB,EAAE,CAAC;QAC7D,aAAa,EAAE,KAAK;KACrB,CAAC;AACJ,CAAC"}

package/dist/lib/judge/llm-judge-code.test.d.ts

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

package/dist/lib/judge/llm-judge-code.test.d.ts.map

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