rag-eval-node-ts 0.3.0

package/dist/index.js

@@ -0,0 +1,37 @@
+"use strict";
+// rag-eval-node-ts - Lightweight RAG evaluation metrics for CI/CD pipelines
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.tokenF1 = exports.tfidfSimilarity = exports.cosineSimilarity = exports.buildTfIdfVectors = exports.weightedNgramOverlap = exports.ngramOverlap = exports.getNgrams = exports.tokenize = exports.filterFactualSentences = exports.splitSentences = exports.computeMetric = exports.scoreHallucinationRate = exports.scoreAnswerCorrectness = exports.scoreContextRelevance = exports.scoreContextRecall = exports.scoreContextPrecision = exports.scoreAnswerRelevance = exports.scoreFaithfulness = exports.createEvaluator = exports.evaluateBatch = exports.evaluate = void 0;
+// Core evaluation functions
+var evaluate_1 = require("./evaluate");
+Object.defineProperty(exports, "evaluate", { enumerable: true, get: function () { return evaluate_1.evaluate; } });
+Object.defineProperty(exports, "evaluateBatch", { enumerable: true, get: function () { return evaluate_1.evaluateBatch; } });
+// Evaluator factory
+var evaluator_1 = require("./evaluator");
+Object.defineProperty(exports, "createEvaluator", { enumerable: true, get: function () { return evaluator_1.createEvaluator; } });
+// Metric functions
+var index_1 = require("./metrics/index");
+Object.defineProperty(exports, "scoreFaithfulness", { enumerable: true, get: function () { return index_1.scoreFaithfulness; } });
+Object.defineProperty(exports, "scoreAnswerRelevance", { enumerable: true, get: function () { return index_1.scoreAnswerRelevance; } });
+Object.defineProperty(exports, "scoreContextPrecision", { enumerable: true, get: function () { return index_1.scoreContextPrecision; } });
+Object.defineProperty(exports, "scoreContextRecall", { enumerable: true, get: function () { return index_1.scoreContextRecall; } });
+Object.defineProperty(exports, "scoreContextRelevance", { enumerable: true, get: function () { return index_1.scoreContextRelevance; } });
+Object.defineProperty(exports, "scoreAnswerCorrectness", { enumerable: true, get: function () { return index_1.scoreAnswerCorrectness; } });
+Object.defineProperty(exports, "scoreHallucinationRate", { enumerable: true, get: function () { return index_1.scoreHallucinationRate; } });
+Object.defineProperty(exports, "computeMetric", { enumerable: true, get: function () { return index_1.computeMetric; } });
+// Heuristic primitives
+var sentences_1 = require("./heuristic/sentences");
+Object.defineProperty(exports, "splitSentences", { enumerable: true, get: function () { return sentences_1.splitSentences; } });
+Object.defineProperty(exports, "filterFactualSentences", { enumerable: true, get: function () { return sentences_1.filterFactualSentences; } });
+var ngrams_1 = require("./heuristic/ngrams");
+Object.defineProperty(exports, "tokenize", { enumerable: true, get: function () { return ngrams_1.tokenize; } });
+Object.defineProperty(exports, "getNgrams", { enumerable: true, get: function () { return ngrams_1.getNgrams; } });
+Object.defineProperty(exports, "ngramOverlap", { enumerable: true, get: function () { return ngrams_1.ngramOverlap; } });
+Object.defineProperty(exports, "weightedNgramOverlap", { enumerable: true, get: function () { return ngrams_1.weightedNgramOverlap; } });
+var tfidf_1 = require("./heuristic/tfidf");
+Object.defineProperty(exports, "buildTfIdfVectors", { enumerable: true, get: function () { return tfidf_1.buildTfIdfVectors; } });
+Object.defineProperty(exports, "cosineSimilarity", { enumerable: true, get: function () { return tfidf_1.cosineSimilarity; } });
+Object.defineProperty(exports, "tfidfSimilarity", { enumerable: true, get: function () { return tfidf_1.tfidfSimilarity; } });
+var token_f1_1 = require("./heuristic/token-f1");
+Object.defineProperty(exports, "tokenF1", { enumerable: true, get: function () { return token_f1_1.tokenF1; } });
+//# sourceMappingURL=index.js.map

package/dist/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":";AAAA,4EAA4E;;;AAwB5E,4BAA4B;AAC5B,uCAAqD;AAA5C,oGAAA,QAAQ,OAAA;AAAE,yGAAA,aAAa,OAAA;AAEhC,oBAAoB;AACpB,yCAA8C;AAArC,4GAAA,eAAe,OAAA;AAExB,mBAAmB;AACnB,yCASyB;AARvB,0GAAA,iBAAiB,OAAA;AACjB,6GAAA,oBAAoB,OAAA;AACpB,8GAAA,qBAAqB,OAAA;AACrB,2GAAA,kBAAkB,OAAA;AAClB,8GAAA,qBAAqB,OAAA;AACrB,+GAAA,sBAAsB,OAAA;AACtB,+GAAA,sBAAsB,OAAA;AACtB,sGAAA,aAAa,OAAA;AAGf,uBAAuB;AACvB,mDAA+E;AAAtE,2GAAA,cAAc,OAAA;AAAE,mHAAA,sBAAsB,OAAA;AAC/C,6CAA6F;AAApF,kGAAA,QAAQ,OAAA;AAAE,mGAAA,SAAS,OAAA;AAAE,sGAAA,YAAY,OAAA;AAAE,8GAAA,oBAAoB,OAAA;AAChE,2CAAyF;AAAhF,0GAAA,iBAAiB,OAAA;AAAE,yGAAA,gBAAgB,OAAA;AAAE,wGAAA,eAAe,OAAA;AAC7D,iDAA+C;AAAtC,mGAAA,OAAO,OAAA"}

package/dist/metrics/answer-correctness.d.ts

@@ -0,0 +1,7 @@
+/**
+ * Answer correctness metric: is the answer factually correct vs ground truth?
+ * Requires groundTruth.
+ */
+import type { EvalSample, HeuristicOptions, MetricResult } from '../types';
+export declare function scoreAnswerCorrectness(sample: EvalSample, _options?: HeuristicOptions): Promise<MetricResult>;
+//# sourceMappingURL=answer-correctness.d.ts.map

package/dist/metrics/answer-correctness.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"answer-correctness.d.ts","sourceRoot":"","sources":["../../src/metrics/answer-correctness.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,KAAK,EAAE,UAAU,EAAE,gBAAgB,EAAE,YAAY,EAAc,MAAM,UAAU,CAAC;AAMvF,wBAAsB,sBAAsB,CAC1C,MAAM,EAAE,UAAU,EAClB,QAAQ,CAAC,EAAE,gBAAgB,GAC1B,OAAO,CAAC,YAAY,CAAC,CA2CvB"}

package/dist/metrics/answer-correctness.js

@@ -0,0 +1,51 @@
+"use strict";
+/**
+ * Answer correctness metric: is the answer factually correct vs ground truth?
+ * Requires groundTruth.
+ */
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.scoreAnswerCorrectness = scoreAnswerCorrectness;
+const token_f1_1 = require("../heuristic/token-f1");
+const ngrams_1 = require("../heuristic/ngrams");
+const DEFAULT_THRESHOLD = 0.6;
+async function scoreAnswerCorrectness(sample, _options) {
+    const start = Date.now();
+    const threshold = DEFAULT_THRESHOLD;
+    const signals = [];
+    if (!sample.groundTruth) {
+        return {
+            metricId: 'answerCorrectness',
+            score: null,
+            mode: 'heuristic',
+            passed: null,
+            threshold,
+            explanation: 'groundTruth is required for answerCorrectness but was not provided.',
+            signals,
+            llmCalls: 0,
+            durationMs: Date.now() - start,
+        };
+    }
+    const f1 = (0, token_f1_1.tokenF1)(sample.groundTruth, sample.answer);
+    const ngram = (0, ngrams_1.ngramOverlap)(sample.groundTruth, sample.answer, 1);
+    const score = 0.7 * f1 + 0.3 * ngram;
+    if (score < 0.5) {
+        signals.push({
+            id: 'answer-correctness-low',
+            metricId: 'answerCorrectness',
+            severity: 'warning',
+            message: `Answer correctness is low (score=${score.toFixed(3)}).`,
+        });
+    }
+    return {
+        metricId: 'answerCorrectness',
+        score,
+        mode: 'heuristic',
+        passed: score >= threshold,
+        threshold,
+        explanation: `Token F1: ${f1.toFixed(3)}, unigram overlap: ${ngram.toFixed(3)}, blended score (70%F1 + 30%ngram): ${score.toFixed(3)}.`,
+        signals,
+        llmCalls: 0,
+        durationMs: Date.now() - start,
+    };
+}
+//# sourceMappingURL=answer-correctness.js.map

package/dist/metrics/answer-correctness.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"answer-correctness.js","sourceRoot":"","sources":["../../src/metrics/answer-correctness.ts"],"names":[],"mappings":";AAAA;;;GAGG;;AAQH,wDA8CC;AAnDD,oDAAgD;AAChD,gDAAmD;AAEnD,MAAM,iBAAiB,GAAG,GAAG,CAAC;AAEvB,KAAK,UAAU,sBAAsB,CAC1C,MAAkB,EAClB,QAA2B;IAE3B,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;IACzB,MAAM,SAAS,GAAG,iBAAiB,CAAC;IACpC,MAAM,OAAO,GAAiB,EAAE,CAAC;IAEjC,IAAI,CAAC,MAAM,CAAC,WAAW,EAAE,CAAC;QACxB,OAAO;YACL,QAAQ,EAAE,mBAAmB;YAC7B,KAAK,EAAE,IAAI;YACX,IAAI,EAAE,WAAW;YACjB,MAAM,EAAE,IAAI;YACZ,SAAS;YACT,WAAW,EAAE,qEAAqE;YAClF,OAAO;YACP,QAAQ,EAAE,CAAC;YACX,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK;SAC/B,CAAC;IACJ,CAAC;IAED,MAAM,EAAE,GAAG,IAAA,kBAAO,EAAC,MAAM,CAAC,WAAW,EAAE,MAAM,CAAC,MAAM,CAAC,CAAC;IACtD,MAAM,KAAK,GAAG,IAAA,qBAAY,EAAC,MAAM,CAAC,WAAW,EAAE,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;IACjE,MAAM,KAAK,GAAG,GAAG,GAAG,EAAE,GAAG,GAAG,GAAG,KAAK,CAAC;IAErC,IAAI,KAAK,GAAG,GAAG,EAAE,CAAC;QAChB,OAAO,CAAC,IAAI,CAAC;YACX,EAAE,EAAE,wBAAwB;YAC5B,QAAQ,EAAE,mBAAmB;YAC7B,QAAQ,EAAE,SAAS;YACnB,OAAO,EAAE,oCAAoC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI;SAClE,CAAC,CAAC;IACL,CAAC;IAED,OAAO;QACL,QAAQ,EAAE,mBAAmB;QAC7B,KAAK;QACL,IAAI,EAAE,WAAW;QACjB,MAAM,EAAE,KAAK,IAAI,SAAS;QAC1B,SAAS;QACT,WAAW,EAAE,aAAa,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC,sBAAsB,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,uCAAuC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG;QACvI,OAAO;QACP,QAAQ,EAAE,CAAC;QACX,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK;KAC/B,CAAC;AACJ,CAAC"}

package/dist/metrics/answer-relevance.d.ts

@@ -0,0 +1,6 @@
+/**
+ * Answer relevance metric: is the answer relevant to the question?
+ */
+import type { EvalSample, HeuristicOptions, MetricResult } from '../types';
+export declare function scoreAnswerRelevance(sample: EvalSample, _options?: HeuristicOptions): Promise<MetricResult>;
+//# sourceMappingURL=answer-relevance.d.ts.map

package/dist/metrics/answer-relevance.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"answer-relevance.d.ts","sourceRoot":"","sources":["../../src/metrics/answer-relevance.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,EAAE,UAAU,EAAE,gBAAgB,EAAE,YAAY,EAAc,MAAM,UAAU,CAAC;AAMvF,wBAAsB,oBAAoB,CACxC,MAAM,EAAE,UAAU,EAClB,QAAQ,CAAC,EAAE,gBAAgB,GAC1B,OAAO,CAAC,YAAY,CAAC,CA6BvB"}

package/dist/metrics/answer-relevance.js

@@ -0,0 +1,37 @@
+"use strict";
+/**
+ * Answer relevance metric: is the answer relevant to the question?
+ */
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.scoreAnswerRelevance = scoreAnswerRelevance;
+const tfidf_1 = require("../heuristic/tfidf");
+const ngrams_1 = require("../heuristic/ngrams");
+const DEFAULT_THRESHOLD = 0.7;
+async function scoreAnswerRelevance(sample, _options) {
+    const start = Date.now();
+    const threshold = DEFAULT_THRESHOLD;
+    const signals = [];
+    const tfidf = (0, tfidf_1.tfidfSimilarity)(sample.question, sample.answer);
+    const ngram = (0, ngrams_1.ngramOverlap)(sample.question, sample.answer, 1);
+    const score = (tfidf + ngram) / 2;
+    if (score < 0.5) {
+        signals.push({
+            id: 'answer-relevance-low',
+            metricId: 'answerRelevance',
+            severity: 'warning',
+            message: `Answer may not be relevant to the question (score=${score.toFixed(3)}).`,
+        });
+    }
+    return {
+        metricId: 'answerRelevance',
+        score,
+        mode: 'heuristic',
+        passed: score >= threshold,
+        threshold,
+        explanation: `TF-IDF similarity: ${tfidf.toFixed(3)}, unigram overlap: ${ngram.toFixed(3)}, average: ${score.toFixed(3)}.`,
+        signals,
+        llmCalls: 0,
+        durationMs: Date.now() - start,
+    };
+}
+//# sourceMappingURL=answer-relevance.js.map

package/dist/metrics/answer-relevance.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"answer-relevance.js","sourceRoot":"","sources":["../../src/metrics/answer-relevance.ts"],"names":[],"mappings":";AAAA;;GAEG;;AAQH,oDAgCC;AArCD,8CAAqD;AACrD,gDAAmD;AAEnD,MAAM,iBAAiB,GAAG,GAAG,CAAC;AAEvB,KAAK,UAAU,oBAAoB,CACxC,MAAkB,EAClB,QAA2B;IAE3B,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;IACzB,MAAM,SAAS,GAAG,iBAAiB,CAAC;IACpC,MAAM,OAAO,GAAiB,EAAE,CAAC;IAEjC,MAAM,KAAK,GAAG,IAAA,uBAAe,EAAC,MAAM,CAAC,QAAQ,EAAE,MAAM,CAAC,MAAM,CAAC,CAAC;IAC9D,MAAM,KAAK,GAAG,IAAA,qBAAY,EAAC,MAAM,CAAC,QAAQ,EAAE,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;IAC9D,MAAM,KAAK,GAAG,CAAC,KAAK,GAAG,KAAK,CAAC,GAAG,CAAC,CAAC;IAElC,IAAI,KAAK,GAAG,GAAG,EAAE,CAAC;QAChB,OAAO,CAAC,IAAI,CAAC;YACX,EAAE,EAAE,sBAAsB;YAC1B,QAAQ,EAAE,iBAAiB;YAC3B,QAAQ,EAAE,SAAS;YACnB,OAAO,EAAE,qDAAqD,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI;SACnF,CAAC,CAAC;IACL,CAAC;IAED,OAAO;QACL,QAAQ,EAAE,iBAAiB;QAC3B,KAAK;QACL,IAAI,EAAE,WAAW;QACjB,MAAM,EAAE,KAAK,IAAI,SAAS;QAC1B,SAAS;QACT,WAAW,EAAE,sBAAsB,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,sBAAsB,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,cAAc,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG;QAC1H,OAAO;QACP,QAAQ,EAAE,CAAC;QACX,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK;KAC/B,CAAC;AACJ,CAAC"}

package/dist/metrics/context-precision.d.ts

@@ -0,0 +1,6 @@
+/**
+ * Context precision metric: are the retrieved contexts relevant to answering the question?
+ */
+import type { EvalSample, HeuristicOptions, MetricResult } from '../types';
+export declare function scoreContextPrecision(sample: EvalSample, _options?: HeuristicOptions): Promise<MetricResult>;
+//# sourceMappingURL=context-precision.d.ts.map

package/dist/metrics/context-precision.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"context-precision.d.ts","sourceRoot":"","sources":["../../src/metrics/context-precision.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,EAAE,UAAU,EAAE,gBAAgB,EAAE,YAAY,EAAc,MAAM,UAAU,CAAC;AAMvF,wBAAsB,qBAAqB,CACzC,MAAM,EAAE,UAAU,EAClB,QAAQ,CAAC,EAAE,gBAAgB,GAC1B,OAAO,CAAC,YAAY,CAAC,CAoDvB"}

package/dist/metrics/context-precision.js

@@ -0,0 +1,57 @@
+"use strict";
+/**
+ * Context precision metric: are the retrieved contexts relevant to answering the question?
+ */
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.scoreContextPrecision = scoreContextPrecision;
+const tfidf_1 = require("../heuristic/tfidf");
+const DEFAULT_THRESHOLD = 0.7;
+const LOW_RELEVANCE_THRESHOLD = 0.3;
+async function scoreContextPrecision(sample, _options) {
+    const start = Date.now();
+    const threshold = DEFAULT_THRESHOLD;
+    const signals = [];
+    if (sample.contexts.length === 0) {
+        return {
+            metricId: 'contextPrecision',
+            score: 0,
+            mode: 'heuristic',
+            passed: false,
+            threshold,
+            explanation: 'No context chunks provided.',
+            signals,
+            llmCalls: 0,
+            durationMs: Date.now() - start,
+        };
+    }
+    const relevances = [];
+    const lowChunks = [];
+    for (let i = 0; i < sample.contexts.length; i++) {
+        const relevance = (0, tfidf_1.tfidfSimilarity)(sample.question, sample.contexts[i]);
+        relevances.push(relevance);
+        if (relevance < LOW_RELEVANCE_THRESHOLD) {
+            lowChunks.push(i);
+        }
+    }
+    if (lowChunks.length > 0) {
+        signals.push({
+            id: 'context-precision-low-relevance-chunks',
+            metricId: 'contextPrecision',
+            severity: 'info',
+            message: `Context chunk(s) with low relevance (< ${LOW_RELEVANCE_THRESHOLD}): indices [${lowChunks.join(', ')}].`,
+        });
+    }
+    const score = relevances.reduce((a, b) => a + b, 0) / relevances.length;
+    return {
+        metricId: 'contextPrecision',
+        score,
+        mode: 'heuristic',
+        passed: score >= threshold,
+        threshold,
+        explanation: `Average TF-IDF relevance of ${sample.contexts.length} context chunk(s) to the question: ${score.toFixed(3)}.`,
+        signals,
+        llmCalls: 0,
+        durationMs: Date.now() - start,
+    };
+}
+//# sourceMappingURL=context-precision.js.map

package/dist/metrics/context-precision.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"context-precision.js","sourceRoot":"","sources":["../../src/metrics/context-precision.ts"],"names":[],"mappings":";AAAA;;GAEG;;AAQH,sDAuDC;AA5DD,8CAAqD;AAErD,MAAM,iBAAiB,GAAG,GAAG,CAAC;AAC9B,MAAM,uBAAuB,GAAG,GAAG,CAAC;AAE7B,KAAK,UAAU,qBAAqB,CACzC,MAAkB,EAClB,QAA2B;IAE3B,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;IACzB,MAAM,SAAS,GAAG,iBAAiB,CAAC;IACpC,MAAM,OAAO,GAAiB,EAAE,CAAC;IAEjC,IAAI,MAAM,CAAC,QAAQ,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACjC,OAAO;YACL,QAAQ,EAAE,kBAAkB;YAC5B,KAAK,EAAE,CAAC;YACR,IAAI,EAAE,WAAW;YACjB,MAAM,EAAE,KAAK;YACb,SAAS;YACT,WAAW,EAAE,6BAA6B;YAC1C,OAAO;YACP,QAAQ,EAAE,CAAC;YACX,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK;SAC/B,CAAC;IACJ,CAAC;IAED,MAAM,UAAU,GAAa,EAAE,CAAC;IAChC,MAAM,SAAS,GAAa,EAAE,CAAC;IAE/B,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,MAAM,CAAC,QAAQ,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QAChD,MAAM,SAAS,GAAG,IAAA,uBAAe,EAAC,MAAM,CAAC,QAAQ,EAAE,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC;QACvE,UAAU,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;QAC3B,IAAI,SAAS,GAAG,uBAAuB,EAAE,CAAC;YACxC,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACpB,CAAC;IACH,CAAC;IAED,IAAI,SAAS,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACzB,OAAO,CAAC,IAAI,CAAC;YACX,EAAE,EAAE,wCAAwC;YAC5C,QAAQ,EAAE,kBAAkB;YAC5B,QAAQ,EAAE,MAAM;YAChB,OAAO,EAAE,0CAA0C,uBAAuB,eAAe,SAAS,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI;SAClH,CAAC,CAAC;IACL,CAAC;IAED,MAAM,KAAK,GAAG,UAAU,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,GAAG,UAAU,CAAC,MAAM,CAAC;IAExE,OAAO;QACL,QAAQ,EAAE,kBAAkB;QAC5B,KAAK;QACL,IAAI,EAAE,WAAW;QACjB,MAAM,EAAE,KAAK,IAAI,SAAS;QAC1B,SAAS;QACT,WAAW,EAAE,+BAA+B,MAAM,CAAC,QAAQ,CAAC,MAAM,sCAAsC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG;QAC3H,OAAO;QACP,QAAQ,EAAE,CAAC;QACX,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK;KAC/B,CAAC;AACJ,CAAC"}

package/dist/metrics/context-recall.d.ts

@@ -0,0 +1,7 @@
+/**
+ * Context recall metric: do contexts cover the ground truth?
+ * Requires groundTruth.
+ */
+import type { EvalSample, HeuristicOptions, MetricResult } from '../types';
+export declare function scoreContextRecall(sample: EvalSample, _options?: HeuristicOptions): Promise<MetricResult>;
+//# sourceMappingURL=context-recall.d.ts.map

package/dist/metrics/context-recall.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"context-recall.d.ts","sourceRoot":"","sources":["../../src/metrics/context-recall.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,KAAK,EAAE,UAAU,EAAE,gBAAgB,EAAE,YAAY,EAAE,MAAM,UAAU,CAAC;AAO3E,wBAAsB,kBAAkB,CACtC,MAAM,EAAE,UAAU,EAClB,QAAQ,CAAC,EAAE,gBAAgB,GAC1B,OAAO,CAAC,YAAY,CAAC,CAyDvB"}

package/dist/metrics/context-recall.js

@@ -0,0 +1,66 @@
+"use strict";
+/**
+ * Context recall metric: do contexts cover the ground truth?
+ * Requires groundTruth.
+ */
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.scoreContextRecall = scoreContextRecall;
+const sentences_1 = require("../heuristic/sentences");
+const ngrams_1 = require("../heuristic/ngrams");
+const DEFAULT_THRESHOLD = 0.7;
+const COVERAGE_THRESHOLD = 0.3;
+async function scoreContextRecall(sample, _options) {
+    const start = Date.now();
+    const threshold = DEFAULT_THRESHOLD;
+    if (!sample.groundTruth) {
+        return {
+            metricId: 'contextRecall',
+            score: null,
+            mode: 'heuristic',
+            passed: null,
+            threshold,
+            explanation: 'groundTruth is required for contextRecall but was not provided.',
+            signals: [],
+            llmCalls: 0,
+            durationMs: Date.now() - start,
+        };
+    }
+    const gtSentences = (0, sentences_1.splitSentences)(sample.groundTruth);
+    if (gtSentences.length === 0 || sample.contexts.length === 0) {
+        return {
+            metricId: 'contextRecall',
+            score: 0,
+            mode: 'heuristic',
+            passed: false,
+            threshold,
+            explanation: 'No ground truth sentences or context chunks to evaluate.',
+            signals: [],
+            llmCalls: 0,
+            durationMs: Date.now() - start,
+        };
+    }
+    let coveredCount = 0;
+    for (const sentence of gtSentences) {
+        let maxOverlap = 0;
+        for (const ctx of sample.contexts) {
+            const overlap = (0, ngrams_1.ngramOverlap)(sentence, ctx, 1);
+            if (overlap > maxOverlap)
+                maxOverlap = overlap;
+        }
+        if (maxOverlap >= COVERAGE_THRESHOLD)
+            coveredCount++;
+    }
+    const score = coveredCount / gtSentences.length;
+    return {
+        metricId: 'contextRecall',
+        score,
+        mode: 'heuristic',
+        passed: score >= threshold,
+        threshold,
+        explanation: `${coveredCount}/${gtSentences.length} ground truth sentence(s) covered by contexts (overlap >= ${COVERAGE_THRESHOLD}): score=${score.toFixed(3)}.`,
+        signals: [],
+        llmCalls: 0,
+        durationMs: Date.now() - start,
+    };
+}
+//# sourceMappingURL=context-recall.js.map

package/dist/metrics/context-recall.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"context-recall.js","sourceRoot":"","sources":["../../src/metrics/context-recall.ts"],"names":[],"mappings":";AAAA;;;GAGG;;AASH,gDA4DC;AAlED,sDAAwD;AACxD,gDAAmD;AAEnD,MAAM,iBAAiB,GAAG,GAAG,CAAC;AAC9B,MAAM,kBAAkB,GAAG,GAAG,CAAC;AAExB,KAAK,UAAU,kBAAkB,CACtC,MAAkB,EAClB,QAA2B;IAE3B,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;IACzB,MAAM,SAAS,GAAG,iBAAiB,CAAC;IAEpC,IAAI,CAAC,MAAM,CAAC,WAAW,EAAE,CAAC;QACxB,OAAO;YACL,QAAQ,EAAE,eAAe;YACzB,KAAK,EAAE,IAAI;YACX,IAAI,EAAE,WAAW;YACjB,MAAM,EAAE,IAAI;YACZ,SAAS;YACT,WAAW,EAAE,iEAAiE;YAC9E,OAAO,EAAE,EAAE;YACX,QAAQ,EAAE,CAAC;YACX,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK;SAC/B,CAAC;IACJ,CAAC;IAED,MAAM,WAAW,GAAG,IAAA,0BAAc,EAAC,MAAM,CAAC,WAAW,CAAC,CAAC;IAEvD,IAAI,WAAW,CAAC,MAAM,KAAK,CAAC,IAAI,MAAM,CAAC,QAAQ,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC7D,OAAO;YACL,QAAQ,EAAE,eAAe;YACzB,KAAK,EAAE,CAAC;YACR,IAAI,EAAE,WAAW;YACjB,MAAM,EAAE,KAAK;YACb,SAAS;YACT,WAAW,EAAE,0DAA0D;YACvE,OAAO,EAAE,EAAE;YACX,QAAQ,EAAE,CAAC;YACX,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK;SAC/B,CAAC;IACJ,CAAC;IAED,IAAI,YAAY,GAAG,CAAC,CAAC;IACrB,KAAK,MAAM,QAAQ,IAAI,WAAW,EAAE,CAAC;QACnC,IAAI,UAAU,GAAG,CAAC,CAAC;QACnB,KAAK,MAAM,GAAG,IAAI,MAAM,CAAC,QAAQ,EAAE,CAAC;YAClC,MAAM,OAAO,GAAG,IAAA,qBAAY,EAAC,QAAQ,EAAE,GAAG,EAAE,CAAC,CAAC,CAAC;YAC/C,IAAI,OAAO,GAAG,UAAU;gBAAE,UAAU,GAAG,OAAO,CAAC;QACjD,CAAC;QACD,IAAI,UAAU,IAAI,kBAAkB;YAAE,YAAY,EAAE,CAAC;IACvD,CAAC;IAED,MAAM,KAAK,GAAG,YAAY,GAAG,WAAW,CAAC,MAAM,CAAC;IAEhD,OAAO;QACL,QAAQ,EAAE,eAAe;QACzB,KAAK;QACL,IAAI,EAAE,WAAW;QACjB,MAAM,EAAE,KAAK,IAAI,SAAS;QAC1B,SAAS;QACT,WAAW,EAAE,GAAG,YAAY,IAAI,WAAW,CAAC,MAAM,6DAA6D,kBAAkB,YAAY,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG;QAChK,OAAO,EAAE,EAAE;QACX,QAAQ,EAAE,CAAC;QACX,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK;KAC/B,CAAC;AACJ,CAAC"}

package/dist/metrics/context-relevance.d.ts

@@ -0,0 +1,6 @@
+/**
+ * Context relevance metric: are context chunks relevant to the question (stricter than precision)?
+ */
+import type { EvalSample, HeuristicOptions, MetricResult } from '../types';
+export declare function scoreContextRelevance(sample: EvalSample, options?: HeuristicOptions): Promise<MetricResult>;
+//# sourceMappingURL=context-relevance.d.ts.map

package/dist/metrics/context-relevance.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"context-relevance.d.ts","sourceRoot":"","sources":["../../src/metrics/context-relevance.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,EAAE,UAAU,EAAE,gBAAgB,EAAE,YAAY,EAAE,MAAM,UAAU,CAAC;AAM3E,wBAAsB,qBAAqB,CACzC,MAAM,EAAE,UAAU,EAClB,OAAO,CAAC,EAAE,gBAAgB,GACzB,OAAO,CAAC,YAAY,CAAC,CAwCvB"}

package/dist/metrics/context-relevance.js

@@ -0,0 +1,48 @@
+"use strict";
+/**
+ * Context relevance metric: are context chunks relevant to the question (stricter than precision)?
+ */
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.scoreContextRelevance = scoreContextRelevance;
+const ngrams_1 = require("../heuristic/ngrams");
+const DEFAULT_THRESHOLD = 0.6;
+const DEFAULT_CHUNK_RELEVANCE_THRESHOLD = 0.2;
+async function scoreContextRelevance(sample, options) {
+    const start = Date.now();
+    const threshold = DEFAULT_THRESHOLD;
+    const chunkRelevanceThreshold = options?.chunkRelevanceThreshold ?? DEFAULT_CHUNK_RELEVANCE_THRESHOLD;
+    const ngramSizes = options?.ngramSizes ?? [1, 2];
+    const ngramWeights = options?.ngramWeights ?? [0.7, 0.3];
+    if (sample.contexts.length === 0) {
+        return {
+            metricId: 'contextRelevance',
+            score: 0,
+            mode: 'heuristic',
+            passed: false,
+            threshold,
+            explanation: 'No context chunks provided.',
+            signals: [],
+            llmCalls: 0,
+            durationMs: Date.now() - start,
+        };
+    }
+    let relevantCount = 0;
+    for (const ctx of sample.contexts) {
+        const overlap = (0, ngrams_1.weightedNgramOverlap)(sample.question, ctx, ngramSizes, ngramWeights);
+        if (overlap >= chunkRelevanceThreshold)
+            relevantCount++;
+    }
+    const score = relevantCount / sample.contexts.length;
+    return {
+        metricId: 'contextRelevance',
+        score,
+        mode: 'heuristic',
+        passed: score >= threshold,
+        threshold,
+        explanation: `${relevantCount}/${sample.contexts.length} context chunk(s) meet relevance threshold (>= ${chunkRelevanceThreshold}): score=${score.toFixed(3)}.`,
+        signals: [],
+        llmCalls: 0,
+        durationMs: Date.now() - start,
+    };
+}
+//# sourceMappingURL=context-relevance.js.map

package/dist/metrics/context-relevance.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"context-relevance.js","sourceRoot":"","sources":["../../src/metrics/context-relevance.ts"],"names":[],"mappings":";AAAA;;GAEG;;AAQH,sDA2CC;AAhDD,gDAA2D;AAE3D,MAAM,iBAAiB,GAAG,GAAG,CAAC;AAC9B,MAAM,iCAAiC,GAAG,GAAG,CAAC;AAEvC,KAAK,UAAU,qBAAqB,CACzC,MAAkB,EAClB,OAA0B;IAE1B,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;IACzB,MAAM,SAAS,GAAG,iBAAiB,CAAC;IACpC,MAAM,uBAAuB,GAAG,OAAO,EAAE,uBAAuB,IAAI,iCAAiC,CAAC;IACtG,MAAM,UAAU,GAAG,OAAO,EAAE,UAAU,IAAI,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;IACjD,MAAM,YAAY,GAAG,OAAO,EAAE,YAAY,IAAI,CAAC,GAAG,EAAE,GAAG,CAAC,CAAC;IAEzD,IAAI,MAAM,CAAC,QAAQ,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACjC,OAAO;YACL,QAAQ,EAAE,kBAAkB;YAC5B,KAAK,EAAE,CAAC;YACR,IAAI,EAAE,WAAW;YACjB,MAAM,EAAE,KAAK;YACb,SAAS;YACT,WAAW,EAAE,6BAA6B;YAC1C,OAAO,EAAE,EAAE;YACX,QAAQ,EAAE,CAAC;YACX,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK;SAC/B,CAAC;IACJ,CAAC;IAED,IAAI,aAAa,GAAG,CAAC,CAAC;IACtB,KAAK,MAAM,GAAG,IAAI,MAAM,CAAC,QAAQ,EAAE,CAAC;QAClC,MAAM,OAAO,GAAG,IAAA,6BAAoB,EAAC,MAAM,CAAC,QAAQ,EAAE,GAAG,EAAE,UAAU,EAAE,YAAY,CAAC,CAAC;QACrF,IAAI,OAAO,IAAI,uBAAuB;YAAE,aAAa,EAAE,CAAC;IAC1D,CAAC;IAED,MAAM,KAAK,GAAG,aAAa,GAAG,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC;IAErD,OAAO;QACL,QAAQ,EAAE,kBAAkB;QAC5B,KAAK;QACL,IAAI,EAAE,WAAW;QACjB,MAAM,EAAE,KAAK,IAAI,SAAS;QAC1B,SAAS;QACT,WAAW,EAAE,GAAG,aAAa,IAAI,MAAM,CAAC,QAAQ,CAAC,MAAM,kDAAkD,uBAAuB,YAAY,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG;QAC/J,OAAO,EAAE,EAAE;QACX,QAAQ,EAAE,CAAC;QACX,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK;KAC/B,CAAC;AACJ,CAAC"}

package/dist/metrics/faithfulness.d.ts

@@ -0,0 +1,6 @@
+/**
+ * Faithfulness metric: is the answer supported by the retrieved contexts?
+ */
+import type { EvalSample, HeuristicOptions, MetricResult } from '../types';
+export declare function scoreFaithfulness(sample: EvalSample, options?: HeuristicOptions): Promise<MetricResult>;
+//# sourceMappingURL=faithfulness.d.ts.map

package/dist/metrics/faithfulness.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"faithfulness.d.ts","sourceRoot":"","sources":["../../src/metrics/faithfulness.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,EAAE,UAAU,EAAE,gBAAgB,EAAE,YAAY,EAAc,MAAM,UAAU,CAAC;AAOvF,wBAAsB,iBAAiB,CACrC,MAAM,EAAE,UAAU,EAClB,OAAO,CAAC,EAAE,gBAAgB,GACzB,OAAO,CAAC,YAAY,CAAC,CA0DvB"}

package/dist/metrics/faithfulness.js

@@ -0,0 +1,64 @@
+"use strict";
+/**
+ * Faithfulness metric: is the answer supported by the retrieved contexts?
+ */
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.scoreFaithfulness = scoreFaithfulness;
+const sentences_1 = require("../heuristic/sentences");
+const ngrams_1 = require("../heuristic/ngrams");
+const DEFAULT_THRESHOLD = 0.7;
+const LOW_SUPPORT_THRESHOLD = 0.3;
+async function scoreFaithfulness(sample, options) {
+    const start = Date.now();
+    const threshold = DEFAULT_THRESHOLD;
+    const ngramSizes = options?.ngramSizes ?? [1, 2];
+    const ngramWeights = options?.ngramWeights ?? [0.7, 0.3];
+    const sentences = (0, sentences_1.splitSentences)(sample.answer);
+    const signals = [];
+    if (sentences.length === 0 || sample.contexts.length === 0) {
+        return {
+            metricId: 'faithfulness',
+            score: 0,
+            mode: 'heuristic',
+            passed: false,
+            threshold,
+            explanation: 'No answer sentences or context chunks to evaluate.',
+            signals,
+            llmCalls: 0,
+            durationMs: Date.now() - start,
+        };
+    }
+    const sentenceScores = [];
+    for (let i = 0; i < sentences.length; i++) {
+        const sentence = sentences[i];
+        let maxOverlap = 0;
+        for (const ctx of sample.contexts) {
+            const overlap = (0, ngrams_1.weightedNgramOverlap)(sentence, ctx, ngramSizes, ngramWeights);
+            if (overlap > maxOverlap)
+                maxOverlap = overlap;
+        }
+        sentenceScores.push(maxOverlap);
+        if (maxOverlap < LOW_SUPPORT_THRESHOLD) {
+            signals.push({
+                id: `faithfulness-low-support-${i}`,
+                metricId: 'faithfulness',
+                severity: 'warning',
+                message: `Answer sentence has low context support (overlap=${maxOverlap.toFixed(3)}).`,
+                evidence: sentence,
+            });
+        }
+    }
+    const score = sentenceScores.reduce((a, b) => a + b, 0) / sentenceScores.length;
+    return {
+        metricId: 'faithfulness',
+        score,
+        mode: 'heuristic',
+        passed: score >= threshold,
+        threshold,
+        explanation: `Average max-context overlap across ${sentences.length} answer sentence(s): ${score.toFixed(3)}.`,
+        signals,
+        llmCalls: 0,
+        durationMs: Date.now() - start,
+    };
+}
+//# sourceMappingURL=faithfulness.js.map

package/dist/metrics/faithfulness.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"faithfulness.js","sourceRoot":"","sources":["../../src/metrics/faithfulness.ts"],"names":[],"mappings":";AAAA;;GAEG;;AASH,8CA6DC;AAnED,sDAAwD;AACxD,gDAA2D;AAE3D,MAAM,iBAAiB,GAAG,GAAG,CAAC;AAC9B,MAAM,qBAAqB,GAAG,GAAG,CAAC;AAE3B,KAAK,UAAU,iBAAiB,CACrC,MAAkB,EAClB,OAA0B;IAE1B,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;IACzB,MAAM,SAAS,GAAG,iBAAiB,CAAC;IACpC,MAAM,UAAU,GAAG,OAAO,EAAE,UAAU,IAAI,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;IACjD,MAAM,YAAY,GAAG,OAAO,EAAE,YAAY,IAAI,CAAC,GAAG,EAAE,GAAG,CAAC,CAAC;IAEzD,MAAM,SAAS,GAAG,IAAA,0BAAc,EAAC,MAAM,CAAC,MAAM,CAAC,CAAC;IAChD,MAAM,OAAO,GAAiB,EAAE,CAAC;IAEjC,IAAI,SAAS,CAAC,MAAM,KAAK,CAAC,IAAI,MAAM,CAAC,QAAQ,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC3D,OAAO;YACL,QAAQ,EAAE,cAAc;YACxB,KAAK,EAAE,CAAC;YACR,IAAI,EAAE,WAAW;YACjB,MAAM,EAAE,KAAK;YACb,SAAS;YACT,WAAW,EAAE,oDAAoD;YACjE,OAAO;YACP,QAAQ,EAAE,CAAC;YACX,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK;SAC/B,CAAC;IACJ,CAAC;IAED,MAAM,cAAc,GAAa,EAAE,CAAC;IAEpC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,SAAS,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QAC1C,MAAM,QAAQ,GAAG,SAAS,CAAC,CAAC,CAAC,CAAC;QAC9B,IAAI,UAAU,GAAG,CAAC,CAAC;QACnB,KAAK,MAAM,GAAG,IAAI,MAAM,CAAC,QAAQ,EAAE,CAAC;YAClC,MAAM,OAAO,GAAG,IAAA,6BAAoB,EAAC,QAAQ,EAAE,GAAG,EAAE,UAAU,EAAE,YAAY,CAAC,CAAC;YAC9E,IAAI,OAAO,GAAG,UAAU;gBAAE,UAAU,GAAG,OAAO,CAAC;QACjD,CAAC;QACD,cAAc,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC;QAEhC,IAAI,UAAU,GAAG,qBAAqB,EAAE,CAAC;YACvC,OAAO,CAAC,IAAI,CAAC;gBACX,EAAE,EAAE,4BAA4B,CAAC,EAAE;gBACnC,QAAQ,EAAE,cAAc;gBACxB,QAAQ,EAAE,SAAS;gBACnB,OAAO,EAAE,oDAAoD,UAAU,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI;gBACtF,QAAQ,EAAE,QAAQ;aACnB,CAAC,CAAC;QACL,CAAC;IACH,CAAC;IAED,MAAM,KAAK,GAAG,cAAc,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,GAAG,cAAc,CAAC,MAAM,CAAC;IAEhF,OAAO;QACL,QAAQ,EAAE,cAAc;QACxB,KAAK;QACL,IAAI,EAAE,WAAW;QACjB,MAAM,EAAE,KAAK,IAAI,SAAS;QAC1B,SAAS;QACT,WAAW,EAAE,sCAAsC,SAAS,CAAC,MAAM,wBAAwB,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG;QAC9G,OAAO;QACP,QAAQ,EAAE,CAAC;QACX,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK;KAC/B,CAAC;AACJ,CAAC"}

package/dist/metrics/hallucination-rate.d.ts

@@ -0,0 +1,7 @@
+/**
+ * Hallucination rate metric: does the answer contain claims not supported by any context?
+ * Score = 1 - hallucination_rate (higher is better).
+ */
+import type { EvalSample, HeuristicOptions, MetricResult } from '../types';
+export declare function scoreHallucinationRate(sample: EvalSample, options?: HeuristicOptions): Promise<MetricResult>;
+//# sourceMappingURL=hallucination-rate.d.ts.map

package/dist/metrics/hallucination-rate.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"hallucination-rate.d.ts","sourceRoot":"","sources":["../../src/metrics/hallucination-rate.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,KAAK,EAAE,UAAU,EAAE,gBAAgB,EAAE,YAAY,EAAc,MAAM,UAAU,CAAC;AAOvF,wBAAsB,sBAAsB,CAC1C,MAAM,EAAE,UAAU,EAClB,OAAO,CAAC,EAAE,gBAAgB,GACzB,OAAO,CAAC,YAAY,CAAC,CA0DvB"}

package/dist/metrics/hallucination-rate.js

@@ -0,0 +1,65 @@
+"use strict";
+/**
+ * Hallucination rate metric: does the answer contain claims not supported by any context?
+ * Score = 1 - hallucination_rate (higher is better).
+ */
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.scoreHallucinationRate = scoreHallucinationRate;
+const sentences_1 = require("../heuristic/sentences");
+const ngrams_1 = require("../heuristic/ngrams");
+const DEFAULT_THRESHOLD = 0.7;
+const DEFAULT_CLAIM_SUPPORT_THRESHOLD = 0.15;
+async function scoreHallucinationRate(sample, options) {
+    const start = Date.now();
+    const threshold = DEFAULT_THRESHOLD;
+    const claimSupportThreshold = options?.claimSupportThreshold ?? DEFAULT_CLAIM_SUPPORT_THRESHOLD;
+    const signals = [];
+    const sentences = (0, sentences_1.splitSentences)(sample.answer);
+    if (sentences.length === 0 || sample.contexts.length === 0) {
+        return {
+            metricId: 'hallucinationRate',
+            score: sample.contexts.length === 0 ? 0 : 1,
+            mode: 'heuristic',
+            passed: sample.contexts.length === 0 ? false : true,
+            threshold,
+            explanation: 'No answer sentences or context chunks to evaluate.',
+            signals,
+            llmCalls: 0,
+            durationMs: Date.now() - start,
+        };
+    }
+    let unsupportedCount = 0;
+    for (let i = 0; i < sentences.length; i++) {
+        const sentence = sentences[i];
+        let maxOverlap = 0;
+        for (const ctx of sample.contexts) {
+            const overlap = (0, ngrams_1.ngramOverlap)(sentence, ctx, 1);
+            if (overlap > maxOverlap)
+                maxOverlap = overlap;
+        }
+        if (maxOverlap < claimSupportThreshold) {
+            unsupportedCount++;
+            signals.push({
+                id: `hallucination-unsupported-${i}`,
+                metricId: 'hallucinationRate',
+                severity: 'critical',
+                message: `Answer sentence appears unsupported by any context (max overlap=${maxOverlap.toFixed(3)}).`,
+                evidence: sentence,
+            });
+        }
+    }
+    const hallucinationRate = unsupportedCount / sentences.length;
+    const score = 1 - hallucinationRate;
+    return {
+        metricId: 'hallucinationRate',
+        score,
+        mode: 'heuristic',
+        passed: score >= threshold,
+        threshold,
+        explanation: `${unsupportedCount}/${sentences.length} answer sentence(s) appear unsupported (hallucination rate=${hallucinationRate.toFixed(3)}, score=${score.toFixed(3)}).`,
+        signals,
+        llmCalls: 0,
+        durationMs: Date.now() - start,
+    };
+}
+//# sourceMappingURL=hallucination-rate.js.map

package/dist/metrics/hallucination-rate.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"hallucination-rate.js","sourceRoot":"","sources":["../../src/metrics/hallucination-rate.ts"],"names":[],"mappings":";AAAA;;;GAGG;;AASH,wDA6DC;AAnED,sDAAwD;AACxD,gDAAmD;AAEnD,MAAM,iBAAiB,GAAG,GAAG,CAAC;AAC9B,MAAM,+BAA+B,GAAG,IAAI,CAAC;AAEtC,KAAK,UAAU,sBAAsB,CAC1C,MAAkB,EAClB,OAA0B;IAE1B,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;IACzB,MAAM,SAAS,GAAG,iBAAiB,CAAC;IACpC,MAAM,qBAAqB,GAAG,OAAO,EAAE,qBAAqB,IAAI,+BAA+B,CAAC;IAChG,MAAM,OAAO,GAAiB,EAAE,CAAC;IAEjC,MAAM,SAAS,GAAG,IAAA,0BAAc,EAAC,MAAM,CAAC,MAAM,CAAC,CAAC;IAEhD,IAAI,SAAS,CAAC,MAAM,KAAK,CAAC,IAAI,MAAM,CAAC,QAAQ,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC3D,OAAO;YACL,QAAQ,EAAE,mBAAmB;YAC7B,KAAK,EAAE,MAAM,CAAC,QAAQ,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;YAC3C,IAAI,EAAE,WAAW;YACjB,MAAM,EAAE,MAAM,CAAC,QAAQ,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI;YACnD,SAAS;YACT,WAAW,EAAE,oDAAoD;YACjE,OAAO;YACP,QAAQ,EAAE,CAAC;YACX,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK;SAC/B,CAAC;IACJ,CAAC;IAED,IAAI,gBAAgB,GAAG,CAAC,CAAC;IAEzB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,SAAS,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QAC1C,MAAM,QAAQ,GAAG,SAAS,CAAC,CAAC,CAAC,CAAC;QAC9B,IAAI,UAAU,GAAG,CAAC,CAAC;QACnB,KAAK,MAAM,GAAG,IAAI,MAAM,CAAC,QAAQ,EAAE,CAAC;YAClC,MAAM,OAAO,GAAG,IAAA,qBAAY,EAAC,QAAQ,EAAE,GAAG,EAAE,CAAC,CAAC,CAAC;YAC/C,IAAI,OAAO,GAAG,UAAU;gBAAE,UAAU,GAAG,OAAO,CAAC;QACjD,CAAC;QAED,IAAI,UAAU,GAAG,qBAAqB,EAAE,CAAC;YACvC,gBAAgB,EAAE,CAAC;YACnB,OAAO,CAAC,IAAI,CAAC;gBACX,EAAE,EAAE,6BAA6B,CAAC,EAAE;gBACpC,QAAQ,EAAE,mBAAmB;gBAC7B,QAAQ,EAAE,UAAU;gBACpB,OAAO,EAAE,mEAAmE,UAAU,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI;gBACrG,QAAQ,EAAE,QAAQ;aACnB,CAAC,CAAC;QACL,CAAC;IACH,CAAC;IAED,MAAM,iBAAiB,GAAG,gBAAgB,GAAG,SAAS,CAAC,MAAM,CAAC;IAC9D,MAAM,KAAK,GAAG,CAAC,GAAG,iBAAiB,CAAC;IAEpC,OAAO;QACL,QAAQ,EAAE,mBAAmB;QAC7B,KAAK;QACL,IAAI,EAAE,WAAW;QACjB,MAAM,EAAE,KAAK,IAAI,SAAS;QAC1B,SAAS;QACT,WAAW,EAAE,GAAG,gBAAgB,IAAI,SAAS,CAAC,MAAM,8DAA8D,iBAAiB,CAAC,OAAO,CAAC,CAAC,CAAC,WAAW,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI;QAC7K,OAAO;QACP,QAAQ,EAAE,CAAC;QACX,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK;KAC/B,CAAC;AACJ,CAAC"}

package/dist/metrics/index.d.ts

@@ -0,0 +1,14 @@
+/**
+ * Metric dispatcher: route MetricId to the correct scoring function.
+ */
+import type { MetricId, EvalSample, HeuristicOptions, MetricResult } from '../types';
+import { scoreFaithfulness } from './faithfulness';
+import { scoreAnswerRelevance } from './answer-relevance';
+import { scoreContextPrecision } from './context-precision';
+import { scoreContextRecall } from './context-recall';
+import { scoreContextRelevance } from './context-relevance';
+import { scoreAnswerCorrectness } from './answer-correctness';
+import { scoreHallucinationRate } from './hallucination-rate';
+export { scoreFaithfulness, scoreAnswerRelevance, scoreContextPrecision, scoreContextRecall, scoreContextRelevance, scoreAnswerCorrectness, scoreHallucinationRate, };
+export declare function computeMetric(metricId: MetricId, sample: EvalSample, options?: HeuristicOptions): Promise<MetricResult>;
+//# sourceMappingURL=index.d.ts.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/metrics/index.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,EAAE,QAAQ,EAAE,UAAU,EAAE,gBAAgB,EAAE,YAAY,EAAE,MAAM,UAAU,CAAC;AACrF,OAAO,EAAE,iBAAiB,EAAE,MAAM,gBAAgB,CAAC;AACnD,OAAO,EAAE,oBAAoB,EAAE,MAAM,oBAAoB,CAAC;AAC1D,OAAO,EAAE,qBAAqB,EAAE,MAAM,qBAAqB,CAAC;AAC5D,OAAO,EAAE,kBAAkB,EAAE,MAAM,kBAAkB,CAAC;AACtD,OAAO,EAAE,qBAAqB,EAAE,MAAM,qBAAqB,CAAC;AAC5D,OAAO,EAAE,sBAAsB,EAAE,MAAM,sBAAsB,CAAC;AAC9D,OAAO,EAAE,sBAAsB,EAAE,MAAM,sBAAsB,CAAC;AAE9D,OAAO,EACL,iBAAiB,EACjB,oBAAoB,EACpB,qBAAqB,EACrB,kBAAkB,EAClB,qBAAqB,EACrB,sBAAsB,EACtB,sBAAsB,GACvB,CAAC;AAEF,wBAAsB,aAAa,CACjC,QAAQ,EAAE,QAAQ,EAClB,MAAM,EAAE,UAAU,EAClB,OAAO,CAAC,EAAE,gBAAgB,GACzB,OAAO,CAAC,YAAY,CAAC,CAiBvB"}

package/dist/metrics/index.js

@@ -0,0 +1,40 @@
+"use strict";
+/**
+ * Metric dispatcher: route MetricId to the correct scoring function.
+ */
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.scoreHallucinationRate = exports.scoreAnswerCorrectness = exports.scoreContextRelevance = exports.scoreContextRecall = exports.scoreContextPrecision = exports.scoreAnswerRelevance = exports.scoreFaithfulness = void 0;
+exports.computeMetric = computeMetric;
+const faithfulness_1 = require("./faithfulness");
+Object.defineProperty(exports, "scoreFaithfulness", { enumerable: true, get: function () { return faithfulness_1.scoreFaithfulness; } });
+const answer_relevance_1 = require("./answer-relevance");
+Object.defineProperty(exports, "scoreAnswerRelevance", { enumerable: true, get: function () { return answer_relevance_1.scoreAnswerRelevance; } });
+const context_precision_1 = require("./context-precision");
+Object.defineProperty(exports, "scoreContextPrecision", { enumerable: true, get: function () { return context_precision_1.scoreContextPrecision; } });
+const context_recall_1 = require("./context-recall");
+Object.defineProperty(exports, "scoreContextRecall", { enumerable: true, get: function () { return context_recall_1.scoreContextRecall; } });
+const context_relevance_1 = require("./context-relevance");
+Object.defineProperty(exports, "scoreContextRelevance", { enumerable: true, get: function () { return context_relevance_1.scoreContextRelevance; } });
+const answer_correctness_1 = require("./answer-correctness");
+Object.defineProperty(exports, "scoreAnswerCorrectness", { enumerable: true, get: function () { return answer_correctness_1.scoreAnswerCorrectness; } });
+const hallucination_rate_1 = require("./hallucination-rate");
+Object.defineProperty(exports, "scoreHallucinationRate", { enumerable: true, get: function () { return hallucination_rate_1.scoreHallucinationRate; } });
+async function computeMetric(metricId, sample, options) {
+    switch (metricId) {
+        case 'faithfulness':
+            return (0, faithfulness_1.scoreFaithfulness)(sample, options);
+        case 'answerRelevance':
+            return (0, answer_relevance_1.scoreAnswerRelevance)(sample, options);
+        case 'contextPrecision':
+            return (0, context_precision_1.scoreContextPrecision)(sample, options);
+        case 'contextRecall':
+            return (0, context_recall_1.scoreContextRecall)(sample, options);
+        case 'contextRelevance':
+            return (0, context_relevance_1.scoreContextRelevance)(sample, options);
+        case 'answerCorrectness':
+            return (0, answer_correctness_1.scoreAnswerCorrectness)(sample, options);
+        case 'hallucinationRate':
+            return (0, hallucination_rate_1.scoreHallucinationRate)(sample, options);
+    }
+}
+//# sourceMappingURL=index.js.map

package/dist/metrics/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../src/metrics/index.ts"],"names":[],"mappings":";AAAA;;GAEG;;;AAqBH,sCAqBC;AAvCD,iDAAmD;AASjD,kGATO,gCAAiB,OASP;AARnB,yDAA0D;AASxD,qGATO,uCAAoB,OASP;AARtB,2DAA4D;AAS1D,sGATO,yCAAqB,OASP;AARvB,qDAAsD;AASpD,mGATO,mCAAkB,OASP;AARpB,2DAA4D;AAS1D,sGATO,yCAAqB,OASP;AARvB,6DAA8D;AAS5D,uGATO,2CAAsB,OASP;AARxB,6DAA8D;AAS5D,uGATO,2CAAsB,OASP;AAGjB,KAAK,UAAU,aAAa,CACjC,QAAkB,EAClB,MAAkB,EAClB,OAA0B;IAE1B,QAAQ,QAAQ,EAAE,CAAC;QACjB,KAAK,cAAc;YACjB,OAAO,IAAA,gCAAiB,EAAC,MAAM,EAAE,OAAO,CAAC,CAAC;QAC5C,KAAK,iBAAiB;YACpB,OAAO,IAAA,uCAAoB,EAAC,MAAM,EAAE,OAAO,CAAC,CAAC;QAC/C,KAAK,kBAAkB;YACrB,OAAO,IAAA,yCAAqB,EAAC,MAAM,EAAE,OAAO,CAAC,CAAC;QAChD,KAAK,eAAe;YAClB,OAAO,IAAA,mCAAkB,EAAC,MAAM,EAAE,OAAO,CAAC,CAAC;QAC7C,KAAK,kBAAkB;YACrB,OAAO,IAAA,yCAAqB,EAAC,MAAM,EAAE,OAAO,CAAC,CAAC;QAChD,KAAK,mBAAmB;YACtB,OAAO,IAAA,2CAAsB,EAAC,MAAM,EAAE,OAAO,CAAC,CAAC;QACjD,KAAK,mBAAmB;YACtB,OAAO,IAAA,2CAAsB,EAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IACnD,CAAC;AACH,CAAC"}