observability-toolkit 1.8.5 → 2.1.0

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

@@ -0,0 +1,150 @@
+/**
+ * LLM-as-Judge Module Constants
+ *
+ * Centralized constants for all LLM evaluation utilities.
+ * Import from here rather than defining magic numbers inline.
+ */
+import { BYTES, TIME_MS, PERCENT_MULTIPLIER, SECONDS, } from '../core/units.js';
+// ============================================================================
+// Security / Input Validation
+// ============================================================================
+/** Maximum text length per field (10KB) */
+export const MAX_TEXT_LENGTH = 10000;
+/** Maximum context array length */
+export const MAX_CONTEXT_ITEMS = 20;
+/** Maximum statements to process in QAG pattern */
+export const MAX_STATEMENTS = 20;
+/** Maximum logprob entries to process — prevents memory exhaustion from unbounded arrays */
+export const MAX_LOGPROBS = 1000;
+/** Minimum total probability for logprobs-based score normalization.
+ * Below this, text extraction is more reliable than probability weighting. */
+export const LOGPROBS_CONFIDENCE_THRESHOLD = 0.1;
+/** Default timeout for LLM calls (30 seconds) */
+export const DEFAULT_LLM_TIMEOUT_MS = 30000;
+/** Maximum JSON nesting depth */
+export const MAX_JSON_DEPTH = 5;
+// ============================================================================
+// G-Eval
+// ============================================================================
+/** G-Eval minimum score (inclusive) */
+export const G_EVAL_MIN_SCORE = 1;
+/** G-Eval maximum score (inclusive) */
+export const G_EVAL_MAX_SCORE = 5;
+/** G-Eval default/middle score for fallback scenarios */
+export const G_EVAL_DEFAULT_SCORE = 3;
+const G_EVAL_SECOND_SCORE = 2;
+const G_EVAL_FOURTH_SCORE = 4;
+/** G-Eval valid score values array */
+export const G_EVAL_VALID_SCORES = [
+    G_EVAL_MIN_SCORE,
+    G_EVAL_SECOND_SCORE,
+    G_EVAL_DEFAULT_SCORE,
+    G_EVAL_FOURTH_SCORE,
+    G_EVAL_MAX_SCORE,
+];
+/** G-Eval score range for normalization (max - min) */
+export const G_EVAL_SCORE_RANGE = G_EVAL_MAX_SCORE - G_EVAL_MIN_SCORE;
+/** Default temperature for evaluation LLM calls (slight variation allowed) */
+export const LLM_TEMPERATURE_EVALUATION = 0.1;
+/** Minimum evaluation steps to generate in G-Eval */
+export const G_EVAL_MIN_STEPS = 3;
+/** Maximum evaluation steps to generate in G-Eval */
+export const G_EVAL_MAX_STEPS = 5;
+/** Characters of tail text to search for contextual score match in extractScoreFromText */
+export const SCORE_FALLBACK_WINDOW = 100;
+// ============================================================================
+// QAG
+// ============================================================================
+/** Default temperature for deterministic LLM calls (e.g., extraction, answering) */
+export const LLM_TEMPERATURE_DETERMINISTIC = 0;
+/** Minimum statement length to be considered valid for QAG extraction */
+export const MIN_STATEMENT_LENGTH = 10;
+/** Maximum characters of LLM response to include in warning preview logs */
+export const QAG_RESPONSE_PREVIEW_LENGTH = 200;
+// ============================================================================
+// Resilience (Retry / Circuit Breaker)
+// ============================================================================
+/** Minimum valid normalized score */
+export const NORMALIZED_SCORE_MIN = 0;
+/** Maximum valid normalized score */
+export const NORMALIZED_SCORE_MAX = 1;
+/** Default circuit breaker failure threshold */
+export const DEFAULT_CIRCUIT_BREAKER_THRESHOLD = 5;
+/** Default circuit breaker reset timeout in milliseconds */
+export const DEFAULT_CIRCUIT_BREAKER_RESET_MS = 30000;
+/** Maximum exponent for backoff calculation */
+export const MAX_BACKOFF_EXPONENT = 5;
+// Max backoff for LLM API retries (rate limit windows can span up to 60s).
+// Intentionally higher than export-utils.ts BACKOFF_MAX_MS (10s) — LLM judge
+// retries tolerate longer waits to recover from per-minute rate limits.
+export const MAX_BACKOFF_MS = SECONDS.MINUTE * TIME_MS.SECOND;
+// ============================================================================
+// Canary Evaluation
+// ============================================================================
+/** Maximum canary score for poor/hallucination detection cases */
+export const CANARY_SCORE_LOW_MAX = 0.3;
+/** Maximum canary score for very low/off-topic detection cases */
+export const CANARY_SCORE_VERY_LOW_MAX = 0.2;
+// ============================================================================
+// Code Evaluation
+// ============================================================================
+/** Default weight applied to the syntax stage in the aggregate code evaluation score */
+export const CODE_EVAL_SYNTAX_WEIGHT = 0.3;
+/** Default weight applied to the semantic stage in the aggregate code evaluation score */
+export const CODE_EVAL_SEMANTIC_WEIGHT = 0.7;
+/** Syntax weight for Python — lower weight since ast.parse is comprehensive but semantics dominate */
+export const CODE_EVAL_SYNTAX_WEIGHT_PYTHON = 0.2;
+/** Syntax weight for Go — slightly higher weight since gofmt validates both format and parse */
+export const CODE_EVAL_SYNTAX_WEIGHT_GO = 0.35;
+/** Syntax score when code passes syntax validation */
+export const CODE_EVAL_SYNTAX_SCORE_PASS = 1.0;
+/** Syntax score when code fails syntax validation */
+export const CODE_EVAL_SYNTAX_SCORE_FAIL = 0.0;
+/** Default programming language for code evaluation */
+export const CODE_EVAL_DEFAULT_LANGUAGE = 'typescript';
+/** Default evaluation criteria for code evaluation */
+export const CODE_EVAL_DEFAULT_CRITERIA = ['correctness', 'completeness', 'error handling'];
+/** Score when runtime error patterns are detected in code */
+export const CODE_EVAL_RUNTIME_ERROR_SCORE_DETECTED = 0.0;
+/** Score when no runtime error patterns are detected in code */
+export const CODE_EVAL_RUNTIME_ERROR_SCORE_CLEAN = 1.0;
+/** Default weight applied to runtimeErrorScore in aggregate score (0 = no effect, backward compat) */
+export const CODE_EVAL_RUNTIME_ERROR_WEIGHT_DEFAULT = 0;
+// ============================================================================
+// Bias
+// ============================================================================
+/** Maximum possible population stdDev for a bounded [0, 1] score distribution */
+export const PANEL_MAX_STD_DEV = 0.5;
+// ============================================================================
+// Config
+// ============================================================================
+/** Default hallucination guardrail threshold (blocks when score >= this value) */
+export const DEFAULT_HALLUCINATION_THRESHOLD = 0.15;
+/** Sample rate for meta-evaluation (10% of evaluations) */
+export const META_EVAL_SAMPLE_RATE = 0.1;
+/** Maximum allowed timeout */
+export const MAX_TIMEOUT_MS = 300_000;
+/** Maximum allowed retries for LLM judge calls */
+export const MAX_RETRIES = 5;
+/** Default timeout for production evaluators (1 minute) */
+export const PRODUCTION_TIMEOUT_MS = SECONDS.MINUTE * TIME_MS.SECOND;
+/** Default retries for production evaluators */
+export const PRODUCTION_MAX_RETRIES = 2;
+// ============================================================================
+// Evaluation Hooks (Webhook Processing)
+// ============================================================================
+/** Maximum bytes per batch (1MB) to prevent memory exhaustion */
+export const MAX_BATCH_BYTES = BYTES.KB * BYTES.KB;
+/** Maximum bytes per single evaluation (100KB) */
+export const MAX_SINGLE_EVALUATION_BYTES = PERCENT_MULTIPLIER * BYTES.KB;
+/** Maximum dedup cache entries before LRU eviction */
+export const MAX_DEDUP_CACHE_SIZE = 10000;
+/** Maximum number of metadata keys allowed per evaluation (CR-M2) */
+export const MAX_METADATA_KEYS = 50;
+/** Maximum length of a metadata key (CR-M2) */
+export const MAX_METADATA_KEY_LENGTH = 128;
+/** File name for persisted hook stats */
+export const STATS_FILE_NAME = 'hook-stats.json';
+/** File name for persisted webhook configs */
+export const WEBHOOK_CONFIGS_FILE_NAME = 'webhook-configs.json';
+//# sourceMappingURL=llm-judge-constants.js.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"llm-judge-constants.js","sourceRoot":"","sources":["../../../src/lib/judge/llm-judge-constants.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH,OAAO,EACL,KAAK,EACL,OAAO,EACP,kBAAkB,EAClB,OAAO,GACR,MAAM,kBAAkB,CAAC;AAE1B,+EAA+E;AAC/E,8BAA8B;AAC9B,+EAA+E;AAE/E,2CAA2C;AAC3C,MAAM,CAAC,MAAM,eAAe,GAAG,KAAK,CAAC;AAErC,mCAAmC;AACnC,MAAM,CAAC,MAAM,iBAAiB,GAAG,EAAE,CAAC;AAEpC,mDAAmD;AACnD,MAAM,CAAC,MAAM,cAAc,GAAG,EAAE,CAAC;AAEjC,4FAA4F;AAC5F,MAAM,CAAC,MAAM,YAAY,GAAG,IAAI,CAAC;AAEjC;8EAC8E;AAC9E,MAAM,CAAC,MAAM,6BAA6B,GAAG,GAAG,CAAC;AAEjD,iDAAiD;AACjD,MAAM,CAAC,MAAM,sBAAsB,GAAG,KAAK,CAAC;AAE5C,iCAAiC;AACjC,MAAM,CAAC,MAAM,cAAc,GAAG,CAAC,CAAC;AAEhC,+EAA+E;AAC/E,SAAS;AACT,+EAA+E;AAE/E,uCAAuC;AACvC,MAAM,CAAC,MAAM,gBAAgB,GAAG,CAAC,CAAC;AAElC,uCAAuC;AACvC,MAAM,CAAC,MAAM,gBAAgB,GAAG,CAAC,CAAC;AAElC,yDAAyD;AACzD,MAAM,CAAC,MAAM,oBAAoB,GAAG,CAAC,CAAC;AAEtC,MAAM,mBAAmB,GAAG,CAAC,CAAC;AAC9B,MAAM,mBAAmB,GAAG,CAAC,CAAC;AAE9B,sCAAsC;AACtC,MAAM,CAAC,MAAM,mBAAmB,GAAG;IACjC,gBAAgB;IAChB,mBAAmB;IACnB,oBAAoB;IACpB,mBAAmB;IACnB,gBAAgB;CACR,CAAC;AAEX,uDAAuD;AACvD,MAAM,CAAC,MAAM,kBAAkB,GAAG,gBAAgB,GAAG,gBAAgB,CAAC;AAEtE,8EAA8E;AAC9E,MAAM,CAAC,MAAM,0BAA0B,GAAG,GAAG,CAAC;AAE9C,qDAAqD;AACrD,MAAM,CAAC,MAAM,gBAAgB,GAAG,CAAC,CAAC;AAElC,qDAAqD;AACrD,MAAM,CAAC,MAAM,gBAAgB,GAAG,CAAC,CAAC;AAElC,2FAA2F;AAC3F,MAAM,CAAC,MAAM,qBAAqB,GAAG,GAAG,CAAC;AAEzC,+EAA+E;AAC/E,MAAM;AACN,+EAA+E;AAE/E,oFAAoF;AACpF,MAAM,CAAC,MAAM,6BAA6B,GAAG,CAAC,CAAC;AAE/C,yEAAyE;AACzE,MAAM,CAAC,MAAM,oBAAoB,GAAG,EAAE,CAAC;AAEvC,4EAA4E;AAC5E,MAAM,CAAC,MAAM,2BAA2B,GAAG,GAAG,CAAC;AAE/C,+EAA+E;AAC/E,uCAAuC;AACvC,+EAA+E;AAE/E,qCAAqC;AACrC,MAAM,CAAC,MAAM,oBAAoB,GAAG,CAAC,CAAC;AAEtC,qCAAqC;AACrC,MAAM,CAAC,MAAM,oBAAoB,GAAG,CAAC,CAAC;AAEtC,gDAAgD;AAChD,MAAM,CAAC,MAAM,iCAAiC,GAAG,CAAC,CAAC;AAEnD,4DAA4D;AAC5D,MAAM,CAAC,MAAM,gCAAgC,GAAG,KAAK,CAAC;AAEtD,+CAA+C;AAC/C,MAAM,CAAC,MAAM,oBAAoB,GAAG,CAAC,CAAC;AAEtC,2EAA2E;AAC3E,6EAA6E;AAC7E,wEAAwE;AACxE,MAAM,CAAC,MAAM,cAAc,GAAG,OAAO,CAAC,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC;AAE9D,+EAA+E;AAC/E,oBAAoB;AACpB,+EAA+E;AAE/E,kEAAkE;AAClE,MAAM,CAAC,MAAM,oBAAoB,GAAG,GAAG,CAAC;AAExC,kEAAkE;AAClE,MAAM,CAAC,MAAM,yBAAyB,GAAG,GAAG,CAAC;AAE7C,+EAA+E;AAC/E,kBAAkB;AAClB,+EAA+E;AAE/E,wFAAwF;AACxF,MAAM,CAAC,MAAM,uBAAuB,GAAG,GAAG,CAAC;AAE3C,0FAA0F;AAC1F,MAAM,CAAC,MAAM,yBAAyB,GAAG,GAAG,CAAC;AAE7C,sGAAsG;AACtG,MAAM,CAAC,MAAM,8BAA8B,GAAG,GAAG,CAAC;AAElD,gGAAgG;AAChG,MAAM,CAAC,MAAM,0BAA0B,GAAG,IAAI,CAAC;AAE/C,sDAAsD;AACtD,MAAM,CAAC,MAAM,2BAA2B,GAAG,GAAG,CAAC;AAE/C,qDAAqD;AACrD,MAAM,CAAC,MAAM,2BAA2B,GAAG,GAAG,CAAC;AAE/C,uDAAuD;AACvD,MAAM,CAAC,MAAM,0BAA0B,GAAG,YAAY,CAAC;AAEvD,sDAAsD;AACtD,MAAM,CAAC,MAAM,0BAA0B,GAAG,CAAC,aAAa,EAAE,cAAc,EAAE,gBAAgB,CAAC,CAAC;AAE5F,6DAA6D;AAC7D,MAAM,CAAC,MAAM,sCAAsC,GAAG,GAAG,CAAC;AAE1D,gEAAgE;AAChE,MAAM,CAAC,MAAM,mCAAmC,GAAG,GAAG,CAAC;AAEvD,sGAAsG;AACtG,MAAM,CAAC,MAAM,sCAAsC,GAAG,CAAC,CAAC;AAExD,+EAA+E;AAC/E,OAAO;AACP,+EAA+E;AAE/E,iFAAiF;AACjF,MAAM,CAAC,MAAM,iBAAiB,GAAG,GAAG,CAAC;AAErC,+EAA+E;AAC/E,SAAS;AACT,+EAA+E;AAE/E,kFAAkF;AAClF,MAAM,CAAC,MAAM,+BAA+B,GAAG,IAAI,CAAC;AAEpD,2DAA2D;AAC3D,MAAM,CAAC,MAAM,qBAAqB,GAAG,GAAG,CAAC;AAEzC,8BAA8B;AAC9B,MAAM,CAAC,MAAM,cAAc,GAAG,OAAO,CAAC;AAEtC,kDAAkD;AAClD,MAAM,CAAC,MAAM,WAAW,GAAG,CAAC,CAAC;AAE7B,2DAA2D;AAC3D,MAAM,CAAC,MAAM,qBAAqB,GAAG,OAAO,CAAC,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC;AAErE,gDAAgD;AAChD,MAAM,CAAC,MAAM,sBAAsB,GAAG,CAAC,CAAC;AAExC,+EAA+E;AAC/E,wCAAwC;AACxC,+EAA+E;AAE/E,iEAAiE;AACjE,MAAM,CAAC,MAAM,eAAe,GAAG,KAAK,CAAC,EAAE,GAAG,KAAK,CAAC,EAAE,CAAC;AAEnD,kDAAkD;AAClD,MAAM,CAAC,MAAM,2BAA2B,GAAG,kBAAkB,GAAG,KAAK,CAAC,EAAE,CAAC;AAEzE,sDAAsD;AACtD,MAAM,CAAC,MAAM,oBAAoB,GAAG,KAAK,CAAC;AAE1C,qEAAqE;AACrE,MAAM,CAAC,MAAM,iBAAiB,GAAG,EAAE,CAAC;AAEpC,+CAA+C;AAC/C,MAAM,CAAC,MAAM,uBAAuB,GAAG,GAAG,CAAC;AAE3C,yCAAyC;AACzC,MAAM,CAAC,MAAM,eAAe,GAAG,iBAAiB,CAAC;AAEjD,8CAA8C;AAC9C,MAAM,CAAC,MAAM,yBAAyB,GAAG,sBAAsB,CAAC"}

package/dist/lib/judge/llm-judge-dag.d.ts

@@ -0,0 +1,57 @@
+/**
+ * DAG-based LLM evaluation framework.
+ *
+ * Supports structured multi-hop evaluation via a directed acyclic graph of
+ * judgement, task, and verdict nodes.
+ */
+import type { LLMProvider, TestCase, EvalResult } from './llm-as-judge.js';
+export type VerdictNode = {
+    type: 'verdict';
+    score: number;
+    label: string;
+};
+export type TaskNode = {
+    type: 'task';
+    instruction: string;
+    next: string;
+};
+export type BinaryJudgementNode = {
+    type: 'binary_judgement';
+    criteria: string;
+    trueChild: string;
+    falseChild: string;
+};
+export type NonBinaryJudgementNode = {
+    type: 'non_binary_judgement';
+    criteria: string;
+    verdicts: Map<string, string>;
+};
+/**
+ * Routes traversal based on a numeric score from the LLM (0–10 scale).
+ * Branches to `aboveChild` if score >= `scoreThreshold`, otherwise `belowChild`.
+ */
+export type ConditionalJudgementNode = {
+    type: 'conditional_judgement';
+    criteria: string;
+    scoreThreshold: number;
+    aboveChild: string;
+    belowChild: string;
+};
+export type DAGNode = VerdictNode | TaskNode | BinaryJudgementNode | NonBinaryJudgementNode | ConditionalJudgementNode;
+export interface DAGEvalConfig {
+    name: string;
+    nodes: Map<string, DAGNode>;
+    rootId: string;
+    temperature?: number;
+    /** Throw if traversal visits more than this many nodes. Prevents runaway chains in deep trees. */
+    maxDepth?: number;
+    /**
+     * Edge weight multipliers keyed as `"fromNodeId:toNodeId"`.
+     * Applied to the final verdict score (product of all traversed edge weights).
+     * Defaults to 1.0 for unspecified edges. Values should be in (0, 1] for discount semantics.
+     */
+    edgeWeights?: Map<string, number>;
+}
+export declare function validateDAGConfig(config: DAGEvalConfig): void;
+export declare function dagEval(llm: LLMProvider, config: DAGEvalConfig, testCase: TestCase, timeoutMs?: number): Promise<EvalResult>;
+//# sourceMappingURL=llm-judge-dag.d.ts.map

package/dist/lib/judge/llm-judge-dag.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"llm-judge-dag.d.ts","sourceRoot":"","sources":["../../../src/lib/judge/llm-judge-dag.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH,OAAO,KAAK,EAAE,WAAW,EAAE,QAAQ,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAc3E,MAAM,MAAM,WAAW,GAAG;IACxB,IAAI,EAAE,SAAS,CAAC;IAChB,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;CACf,CAAC;AAEF,MAAM,MAAM,QAAQ,GAAG;IACrB,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,MAAM,CAAC;IACpB,IAAI,EAAE,MAAM,CAAC;CACd,CAAC;AAEF,MAAM,MAAM,mBAAmB,GAAG;IAChC,IAAI,EAAE,kBAAkB,CAAC;IACzB,QAAQ,EAAE,MAAM,CAAC;IACjB,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,EAAE,MAAM,CAAC;CACpB,CAAC;AAEF,MAAM,MAAM,sBAAsB,GAAG;IACnC,IAAI,EAAE,sBAAsB,CAAC;IAC7B,QAAQ,EAAE,MAAM,CAAC;IACjB,QAAQ,EAAE,GAAG,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;CAC/B,CAAC;AAEF;;;GAGG;AACH,MAAM,MAAM,wBAAwB,GAAG;IACrC,IAAI,EAAE,uBAAuB,CAAC;IAC9B,QAAQ,EAAE,MAAM,CAAC;IACjB,cAAc,EAAE,MAAM,CAAC;IACvB,UAAU,EAAE,MAAM,CAAC;IACnB,UAAU,EAAE,MAAM,CAAC;CACpB,CAAC;AAEF,MAAM,MAAM,OAAO,GAAG,WAAW,GAAG,QAAQ,GAAG,mBAAmB,GAAG,sBAAsB,GAAG,wBAAwB,CAAC;AAEvH,MAAM,WAAW,aAAa;IAC5B,IAAI,EAAE,MAAM,CAAC;IACb,KAAK,EAAE,GAAG,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IAC5B,MAAM,EAAE,MAAM,CAAC;IACf,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,kGAAkG;IAClG,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB;;;;OAIG;IACH,WAAW,CAAC,EAAE,GAAG,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;CACnC;AA6BD,wBAAgB,iBAAiB,CAAC,MAAM,EAAE,aAAa,GAAG,IAAI,CA4C7D;AAMD,wBAAsB,OAAO,CAC3B,GAAG,EAAE,WAAW,EAChB,MAAM,EAAE,aAAa,EACrB,QAAQ,EAAE,QAAQ,EAClB,SAAS,CAAC,EAAE,MAAM,GACjB,OAAO,CAAC,UAAU,CAAC,CAsHrB"}

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

@@ -0,0 +1,217 @@
+/**
+ * DAG-based LLM evaluation framework.
+ *
+ * Supports structured multi-hop evaluation via a directed acyclic graph of
+ * judgement, task, and verdict nodes.
+ */
+import { TEXT_PREVIEW_LIMIT } from '../core/constants.js';
+// ============================================================================
+// Constants
+// ============================================================================
+const SCORE_MAX = 10;
+const MAX_TASK_INPUT_CHARS = TEXT_PREVIEW_LIMIT * 10;
+// ============================================================================
+// Validation
+// ============================================================================
+function getChildIds(node) {
+    switch (node.type) {
+        case 'verdict':
+            return [];
+        case 'task':
+            return [node.next];
+        case 'binary_judgement':
+            return [node.trueChild, node.falseChild];
+        case 'non_binary_judgement':
+            return Array.from(node.verdicts.values());
+        case 'conditional_judgement':
+            return [node.aboveChild, node.belowChild];
+    }
+}
+function edgeWeight(weights, fromId, toId) {
+    const w = weights?.get(`${fromId}:${toId}`) ?? 1;
+    if (w <= 0 || w > 1) {
+        throw new Error(`Edge weight for '${fromId}:${toId}' must be in (0, 1], got ${w}`);
+    }
+    return w;
+}
+export function validateDAGConfig(config) {
+    const { nodes, rootId } = config;
+    if (!nodes.has(rootId)) {
+        throw new Error(`Invalid DAG config: root node '${rootId}' does not exist in nodes map`);
+    }
+    // Check all child references exist and validate node-level fields
+    for (const [id, node] of nodes) {
+        for (const childId of getChildIds(node)) {
+            if (!nodes.has(childId)) {
+                throw new Error(`Invalid DAG config: node '${id}' references non-existent child node '${childId}'`);
+            }
+        }
+        if (node.type === 'conditional_judgement') {
+            if (node.scoreThreshold < 0 || node.scoreThreshold > SCORE_MAX) {
+                throw new Error(`Invalid DAG config: node '${id}' scoreThreshold ${node.scoreThreshold} must be in [0, ${SCORE_MAX}]`);
+            }
+        }
+    }
+    // Cycle detection via DFS
+    const visited = new Set();
+    const inStack = new Set();
+    function dfs(nodeId) {
+        if (inStack.has(nodeId)) {
+            throw new Error(`Invalid DAG config: cycle detected at node '${nodeId}'`);
+        }
+        if (visited.has(nodeId))
+            return;
+        visited.add(nodeId);
+        inStack.add(nodeId);
+        const node = nodes.get(nodeId);
+        if (!node)
+            throw new Error(`Invalid DAG config: unknown node '${nodeId}'`);
+        for (const childId of getChildIds(node)) {
+            dfs(childId);
+        }
+        inStack.delete(nodeId);
+    }
+    dfs(rootId);
+}
+// ============================================================================
+// Traversal
+// ============================================================================
+export async function dagEval(llm, config, testCase, timeoutMs) {
+    validateDAGConfig(config);
+    const controller = timeoutMs !== undefined ? new AbortController() : undefined;
+    let timeoutHandle;
+    if (controller) {
+        timeoutHandle = setTimeout(() => {
+            controller.abort();
+        }, timeoutMs);
+    }
+    try {
+        const trail = [];
+        let currentId = config.rootId;
+        let currentInput = testCase.output;
+        let accumulatedWeight = 1;
+        while (true) {
+            const node = config.nodes.get(currentId);
+            if (!node)
+                throw new Error(`DAG traversal: unknown node '${currentId}'`);
+            trail.push(currentId);
+            if (config.maxDepth !== undefined && trail.length > config.maxDepth) {
+                throw new Error(`DAG traversal exceeded maxDepth of ${config.maxDepth}`);
+            }
+            if (node.type === 'verdict') {
+                const score = Math.max(0, Math.min(1, (node.score / SCORE_MAX) * accumulatedWeight));
+                const reason = `[${config.name}] traversal: ${trail.join(' -> ')}; verdict: ${node.label}`;
+                return { score, reason };
+            }
+            if (node.type === 'task') {
+                const safeInput = currentInput.slice(0, MAX_TASK_INPUT_CHARS);
+                const prompt = `${node.instruction}\n\nInput: ${safeInput}`;
+                const response = await llm.generate(prompt, {
+                    temperature: config.temperature,
+                    signal: controller?.signal,
+                });
+                currentInput = response.text.trim();
+                const nextId = node.next;
+                accumulatedWeight *= edgeWeight(config.edgeWeights, currentId, nextId);
+                currentId = nextId;
+                continue;
+            }
+            if (node.type === 'binary_judgement') {
+                const prompt = buildBinaryPrompt(node.criteria, testCase.input, currentInput);
+                const response = await llm.generate(prompt, {
+                    temperature: config.temperature,
+                    signal: controller?.signal,
+                });
+                const answer = response.text.trim().toLowerCase();
+                const isTrue = answer === 'true' || answer.startsWith('true');
+                const nextId = isTrue ? node.trueChild : node.falseChild;
+                accumulatedWeight *= edgeWeight(config.edgeWeights, currentId, nextId);
+                currentId = nextId;
+                continue;
+            }
+            if (node.type === 'non_binary_judgement') {
+                const keys = Array.from(node.verdicts.keys());
+                const prompt = buildNonBinaryPrompt(node.criteria, testCase.input, currentInput, keys);
+                const response = await llm.generate(prompt, {
+                    temperature: config.temperature,
+                    signal: controller?.signal,
+                });
+                const answer = response.text.trim().toLowerCase();
+                const matched = keys.find(k => {
+                    const pattern = new RegExp(`\\b${k.toLowerCase()}\\b`);
+                    return pattern.test(answer);
+                });
+                if (!matched) {
+                    throw new Error(`No matching verdict found in LLM response '${answer}'. Expected one of: ${keys.join(', ')}`);
+                }
+                const nextId = node.verdicts.get(matched);
+                if (!nextId)
+                    throw new Error(`DAG traversal: no verdict mapping for '${matched}'`);
+                accumulatedWeight *= edgeWeight(config.edgeWeights, currentId, nextId);
+                currentId = nextId;
+                continue;
+            }
+            if (node.type === 'conditional_judgement') {
+                const prompt = buildConditionalPrompt(node.criteria, testCase.input, currentInput);
+                const response = await llm.generate(prompt, {
+                    temperature: config.temperature,
+                    signal: controller?.signal,
+                });
+                // Expected format: a line containing only a number from 0 to SCORE_MAX (e.g. "7.5" or "8").
+                // If the LLM includes surrounding text, scan all lines for the first fully-numeric line.
+                const lines = response.text.trim().split('\n');
+                const parsed = lines.reduce((acc, line) => {
+                    if (!isNaN(acc))
+                        return acc;
+                    const trimmed = line.trim();
+                    if (!/^-?\d+(\.\d+)?$/.test(trimmed))
+                        return acc;
+                    return parseFloat(trimmed);
+                }, NaN);
+                if (isNaN(parsed)) {
+                    throw new Error(`ConditionalJudgement node could not parse numeric score from LLM response. ` +
+                        `Expected a line containing only a number from 0 to ${SCORE_MAX}. Got: '${response.text.trim().slice(0, 100)}'`);
+                }
+                const clampedScore = Math.max(0, Math.min(SCORE_MAX, parsed));
+                const nextId = clampedScore >= node.scoreThreshold ? node.aboveChild : node.belowChild;
+                accumulatedWeight *= edgeWeight(config.edgeWeights, currentId, nextId);
+                currentId = nextId;
+                continue;
+            }
+            throw new Error(`Unknown node type encountered`);
+        }
+    }
+    finally {
+        if (timeoutHandle !== undefined) {
+            clearTimeout(timeoutHandle);
+        }
+    }
+}
+// ============================================================================
+// Prompt Helpers
+// ============================================================================
+function buildBinaryPrompt(criteria, input, output) {
+    return `Evaluate the following based on this criteria: ${criteria}
+
+Input: ${input}
+Output: ${output}
+
+Answer with only "true" or "false".`;
+}
+function buildNonBinaryPrompt(criteria, input, output, options) {
+    return `Evaluate the following based on this criteria: ${criteria}
+
+Input: ${input}
+Output: ${output}
+
+Answer with exactly one of: ${options.join(', ')}.`;
+}
+function buildConditionalPrompt(criteria, input, output) {
+    return `Evaluate the following based on this criteria: ${criteria}
+
+Input: ${input}
+Output: ${output}
+
+Respond with a single number from 0 to 10.`;
+}
+//# sourceMappingURL=llm-judge-dag.js.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"llm-judge-dag.js","sourceRoot":"","sources":["../../../src/lib/judge/llm-judge-dag.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAGH,OAAO,EAAE,kBAAkB,EAAE,MAAM,sBAAsB,CAAC;AAE1D,+EAA+E;AAC/E,YAAY;AACZ,+EAA+E;AAE/E,MAAM,SAAS,GAAG,EAAE,CAAC;AACrB,MAAM,oBAAoB,GAAG,kBAAkB,GAAG,EAAE,CAAC;AA4DrD,+EAA+E;AAC/E,aAAa;AACb,+EAA+E;AAE/E,SAAS,WAAW,CAAC,IAAa;IAChC,QAAQ,IAAI,CAAC,IAAI,EAAE,CAAC;QAClB,KAAK,SAAS;YACZ,OAAO,EAAE,CAAC;QACZ,KAAK,MAAM;YACT,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACrB,KAAK,kBAAkB;YACrB,OAAO,CAAC,IAAI,CAAC,SAAS,EAAE,IAAI,CAAC,UAAU,CAAC,CAAC;QAC3C,KAAK,sBAAsB;YACzB,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,MAAM,EAAE,CAAC,CAAC;QAC5C,KAAK,uBAAuB;YAC1B,OAAO,CAAC,IAAI,CAAC,UAAU,EAAE,IAAI,CAAC,UAAU,CAAC,CAAC;IAC9C,CAAC;AACH,CAAC;AAED,SAAS,UAAU,CAAC,OAAwC,EAAE,MAAc,EAAE,IAAY;IACxF,MAAM,CAAC,GAAG,OAAO,EAAE,GAAG,CAAC,GAAG,MAAM,IAAI,IAAI,EAAE,CAAC,IAAI,CAAC,CAAC;IACjD,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC;QACpB,MAAM,IAAI,KAAK,CAAC,oBAAoB,MAAM,IAAI,IAAI,4BAA4B,CAAC,EAAE,CAAC,CAAC;IACrF,CAAC;IACD,OAAO,CAAC,CAAC;AACX,CAAC;AAED,MAAM,UAAU,iBAAiB,CAAC,MAAqB;IACrD,MAAM,EAAE,KAAK,EAAE,MAAM,EAAE,GAAG,MAAM,CAAC;IAEjC,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,MAAM,CAAC,EAAE,CAAC;QACvB,MAAM,IAAI,KAAK,CAAC,kCAAkC,MAAM,+BAA+B,CAAC,CAAC;IAC3F,CAAC;IAED,kEAAkE;IAClE,KAAK,MAAM,CAAC,EAAE,EAAE,IAAI,CAAC,IAAI,KAAK,EAAE,CAAC;QAC/B,KAAK,MAAM,OAAO,IAAI,WAAW,CAAC,IAAI,CAAC,EAAE,CAAC;YACxC,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,OAAO,CAAC,EAAE,CAAC;gBACxB,MAAM,IAAI,KAAK,CAAC,6BAA6B,EAAE,yCAAyC,OAAO,GAAG,CAAC,CAAC;YACtG,CAAC;QACH,CAAC;QACD,IAAI,IAAI,CAAC,IAAI,KAAK,uBAAuB,EAAE,CAAC;YAC1C,IAAI,IAAI,CAAC,cAAc,GAAG,CAAC,IAAI,IAAI,CAAC,cAAc,GAAG,SAAS,EAAE,CAAC;gBAC/D,MAAM,IAAI,KAAK,CAAC,6BAA6B,EAAE,oBAAoB,IAAI,CAAC,cAAc,mBAAmB,SAAS,GAAG,CAAC,CAAC;YACzH,CAAC;QACH,CAAC;IACH,CAAC;IAED,0BAA0B;IAC1B,MAAM,OAAO,GAAG,IAAI,GAAG,EAAU,CAAC;IAClC,MAAM,OAAO,GAAG,IAAI,GAAG,EAAU,CAAC;IAElC,SAAS,GAAG,CAAC,MAAc;QACzB,IAAI,OAAO,CAAC,GAAG,CAAC,MAAM,CAAC,EAAE,CAAC;YACxB,MAAM,IAAI,KAAK,CAAC,+CAA+C,MAAM,GAAG,CAAC,CAAC;QAC5E,CAAC;QACD,IAAI,OAAO,CAAC,GAAG,CAAC,MAAM,CAAC;YAAE,OAAO;QAEhC,OAAO,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC;QACpB,OAAO,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC;QAEpB,MAAM,IAAI,GAAG,KAAK,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC;QAC/B,IAAI,CAAC,IAAI;YAAE,MAAM,IAAI,KAAK,CAAC,qCAAqC,MAAM,GAAG,CAAC,CAAC;QAC3E,KAAK,MAAM,OAAO,IAAI,WAAW,CAAC,IAAI,CAAC,EAAE,CAAC;YACxC,GAAG,CAAC,OAAO,CAAC,CAAC;QACf,CAAC;QAED,OAAO,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;IACzB,CAAC;IAED,GAAG,CAAC,MAAM,CAAC,CAAC;AACd,CAAC;AAED,+EAA+E;AAC/E,YAAY;AACZ,+EAA+E;AAE/E,MAAM,CAAC,KAAK,UAAU,OAAO,CAC3B,GAAgB,EAChB,MAAqB,EACrB,QAAkB,EAClB,SAAkB;IAElB,iBAAiB,CAAC,MAAM,CAAC,CAAC;IAE1B,MAAM,UAAU,GAAG,SAAS,KAAK,SAAS,CAAC,CAAC,CAAC,IAAI,eAAe,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC;IAC/E,IAAI,aAAwD,CAAC;IAE7D,IAAI,UAAU,EAAE,CAAC;QACf,aAAa,GAAG,UAAU,CAAC,GAAG,EAAE;YAC9B,UAAU,CAAC,KAAK,EAAE,CAAC;QACrB,CAAC,EAAE,SAAS,CAAC,CAAC;IAChB,CAAC;IAED,IAAI,CAAC;QACH,MAAM,KAAK,GAAa,EAAE,CAAC;QAC3B,IAAI,SAAS,GAAG,MAAM,CAAC,MAAM,CAAC;QAC9B,IAAI,YAAY,GAAG,QAAQ,CAAC,MAAM,CAAC;QACnC,IAAI,iBAAiB,GAAG,CAAC,CAAC;QAE1B,OAAO,IAAI,EAAE,CAAC;YACZ,MAAM,IAAI,GAAG,MAAM,CAAC,KAAK,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC;YACzC,IAAI,CAAC,IAAI;gBAAE,MAAM,IAAI,KAAK,CAAC,gCAAgC,SAAS,GAAG,CAAC,CAAC;YACzE,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;YAEtB,IAAI,MAAM,CAAC,QAAQ,KAAK,SAAS,IAAI,KAAK,CAAC,MAAM,GAAG,MAAM,CAAC,QAAQ,EAAE,CAAC;gBACpE,MAAM,IAAI,KAAK,CAAC,sCAAsC,MAAM,CAAC,QAAQ,EAAE,CAAC,CAAC;YAC3E,CAAC;YAED,IAAI,IAAI,CAAC,IAAI,KAAK,SAAS,EAAE,CAAC;gBAC5B,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,CAAC,IAAI,CAAC,KAAK,GAAG,SAAS,CAAC,GAAG,iBAAiB,CAAC,CAAC,CAAC;gBACrF,MAAM,MAAM,GAAG,IAAI,MAAM,CAAC,IAAI,gBAAgB,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,cAAc,IAAI,CAAC,KAAK,EAAE,CAAC;gBAC3F,OAAO,EAAE,KAAK,EAAE,MAAM,EAAE,CAAC;YAC3B,CAAC;YAED,IAAI,IAAI,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;gBACzB,MAAM,SAAS,GAAG,YAAY,CAAC,KAAK,CAAC,CAAC,EAAE,oBAAoB,CAAC,CAAC;gBAC9D,MAAM,MAAM,GAAG,GAAG,IAAI,CAAC,WAAW,cAAc,SAAS,EAAE,CAAC;gBAC5D,MAAM,QAAQ,GAAG,MAAM,GAAG,CAAC,QAAQ,CAAC,MAAM,EAAE;oBAC1C,WAAW,EAAE,MAAM,CAAC,WAAW;oBAC/B,MAAM,EAAE,UAAU,EAAE,MAAM;iBAC3B,CAAC,CAAC;gBACH,YAAY,GAAG,QAAQ,CAAC,IAAI,CAAC,IAAI,EAAE,CAAC;gBACpC,MAAM,MAAM,GAAG,IAAI,CAAC,IAAI,CAAC;gBACzB,iBAAiB,IAAI,UAAU,CAAC,MAAM,CAAC,WAAW,EAAE,SAAS,EAAE,MAAM,CAAC,CAAC;gBACvE,SAAS,GAAG,MAAM,CAAC;gBACnB,SAAS;YACX,CAAC;YAED,IAAI,IAAI,CAAC,IAAI,KAAK,kBAAkB,EAAE,CAAC;gBACrC,MAAM,MAAM,GAAG,iBAAiB,CAAC,IAAI,CAAC,QAAQ,EAAE,QAAQ,CAAC,KAAK,EAAE,YAAY,CAAC,CAAC;gBAC9E,MAAM,QAAQ,GAAG,MAAM,GAAG,CAAC,QAAQ,CAAC,MAAM,EAAE;oBAC1C,WAAW,EAAE,MAAM,CAAC,WAAW;oBAC/B,MAAM,EAAE,UAAU,EAAE,MAAM;iBAC3B,CAAC,CAAC;gBACH,MAAM,MAAM,GAAG,QAAQ,CAAC,IAAI,CAAC,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;gBAClD,MAAM,MAAM,GAAG,MAAM,KAAK,MAAM,IAAI,MAAM,CAAC,UAAU,CAAC,MAAM,CAAC,CAAC;gBAC9D,MAAM,MAAM,GAAG,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,IAAI,CAAC,UAAU,CAAC;gBACzD,iBAAiB,IAAI,UAAU,CAAC,MAAM,CAAC,WAAW,EAAE,SAAS,EAAE,MAAM,CAAC,CAAC;gBACvE,SAAS,GAAG,MAAM,CAAC;gBACnB,SAAS;YACX,CAAC;YAED,IAAI,IAAI,CAAC,IAAI,KAAK,sBAAsB,EAAE,CAAC;gBACzC,MAAM,IAAI,GAAG,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,EAAE,CAAC,CAAC;gBAC9C,MAAM,MAAM,GAAG,oBAAoB,CAAC,IAAI,CAAC,QAAQ,EAAE,QAAQ,CAAC,KAAK,EAAE,YAAY,EAAE,IAAI,CAAC,CAAC;gBACvF,MAAM,QAAQ,GAAG,MAAM,GAAG,CAAC,QAAQ,CAAC,MAAM,EAAE;oBAC1C,WAAW,EAAE,MAAM,CAAC,WAAW;oBAC/B,MAAM,EAAE,UAAU,EAAE,MAAM;iBAC3B,CAAC,CAAC;gBACH,MAAM,MAAM,GAAG,QAAQ,CAAC,IAAI,CAAC,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;gBAClD,MAAM,OAAO,GAAG,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE;oBAC5B,MAAM,OAAO,GAAG,IAAI,MAAM,CAAC,MAAM,CAAC,CAAC,WAAW,EAAE,KAAK,CAAC,CAAC;oBACvD,OAAO,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;gBAC9B,CAAC,CAAC,CAAC;gBACH,IAAI,CAAC,OAAO,EAAE,CAAC;oBACb,MAAM,IAAI,KAAK,CAAC,8CAA8C,MAAM,uBAAuB,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;gBAChH,CAAC;gBACD,MAAM,MAAM,GAAG,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;gBAC1C,IAAI,CAAC,MAAM;oBAAE,MAAM,IAAI,KAAK,CAAC,0CAA0C,OAAO,GAAG,CAAC,CAAC;gBACnF,iBAAiB,IAAI,UAAU,CAAC,MAAM,CAAC,WAAW,EAAE,SAAS,EAAE,MAAM,CAAC,CAAC;gBACvE,SAAS,GAAG,MAAM,CAAC;gBACnB,SAAS;YACX,CAAC;YAED,IAAI,IAAI,CAAC,IAAI,KAAK,uBAAuB,EAAE,CAAC;gBAC1C,MAAM,MAAM,GAAG,sBAAsB,CAAC,IAAI,CAAC,QAAQ,EAAE,QAAQ,CAAC,KAAK,EAAE,YAAY,CAAC,CAAC;gBACnF,MAAM,QAAQ,GAAG,MAAM,GAAG,CAAC,QAAQ,CAAC,MAAM,EAAE;oBAC1C,WAAW,EAAE,MAAM,CAAC,WAAW;oBAC/B,MAAM,EAAE,UAAU,EAAE,MAAM;iBAC3B,CAAC,CAAC;gBACH,4FAA4F;gBAC5F,yFAAyF;gBACzF,MAAM,KAAK,GAAG,QAAQ,CAAC,IAAI,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;gBAC/C,MAAM,MAAM,GAAG,KAAK,CAAC,MAAM,CAAS,CAAC,GAAG,EAAE,IAAI,EAAE,EAAE;oBAChD,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC;wBAAE,OAAO,GAAG,CAAC;oBAC5B,MAAM,OAAO,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC;oBAC5B,IAAI,CAAC,iBAAiB,CAAC,IAAI,CAAC,OAAO,CAAC;wBAAE,OAAO,GAAG,CAAC;oBACjD,OAAO,UAAU,CAAC,OAAO,CAAC,CAAC;gBAC7B,CAAC,EAAE,GAAG,CAAC,CAAC;gBACR,IAAI,KAAK,CAAC,MAAM,CAAC,EAAE,CAAC;oBAClB,MAAM,IAAI,KAAK,CACb,6EAA6E;wBAC7E,sDAAsD,SAAS,WAAW,QAAQ,CAAC,IAAI,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,GAAG,CAChH,CAAC;gBACJ,CAAC;gBACD,MAAM,YAAY,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,SAAS,EAAE,MAAM,CAAC,CAAC,CAAC;gBAC9D,MAAM,MAAM,GAAG,YAAY,IAAI,IAAI,CAAC,cAAc,CAAC,CAAC,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC,CAAC,IAAI,CAAC,UAAU,CAAC;gBACvF,iBAAiB,IAAI,UAAU,CAAC,MAAM,CAAC,WAAW,EAAE,SAAS,EAAE,MAAM,CAAC,CAAC;gBACvE,SAAS,GAAG,MAAM,CAAC;gBACnB,SAAS;YACX,CAAC;YAED,MAAM,IAAI,KAAK,CAAC,+BAA+B,CAAC,CAAC;QACnD,CAAC;IACH,CAAC;YAAS,CAAC;QACT,IAAI,aAAa,KAAK,SAAS,EAAE,CAAC;YAChC,YAAY,CAAC,aAAa,CAAC,CAAC;QAC9B,CAAC;IACH,CAAC;AACH,CAAC;AAED,+EAA+E;AAC/E,iBAAiB;AACjB,+EAA+E;AAE/E,SAAS,iBAAiB,CAAC,QAAgB,EAAE,KAAa,EAAE,MAAc;IACxE,OAAO,kDAAkD,QAAQ;;SAE1D,KAAK;UACJ,MAAM;;oCAEoB,CAAC;AACrC,CAAC;AAED,SAAS,oBAAoB,CAAC,QAAgB,EAAE,KAAa,EAAE,MAAc,EAAE,OAAiB;IAC9F,OAAO,kDAAkD,QAAQ;;SAE1D,KAAK;UACJ,MAAM;;8BAEc,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC;AACpD,CAAC;AAED,SAAS,sBAAsB,CAAC,QAAgB,EAAE,KAAa,EAAE,MAAc;IAC7E,OAAO,kDAAkD,QAAQ;;SAE1D,KAAK;UACJ,MAAM;;2CAE2B,CAAC;AAC5C,CAAC"}

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

@@ -0,0 +1,8 @@
+/**
+ * DAGEval Tests — RED phase
+ *
+ * Tests for the not-yet-implemented dagEval function and DAGEvalConfig types.
+ * All tests are expected to fail until llm-judge-dag.ts is implemented.
+ */
+export {};
+//# sourceMappingURL=llm-judge-dag.test.d.ts.map

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

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