causantic 0.2.0

package/dist/eval/code-nl-alignment.d.ts

@@ -0,0 +1,25 @@
+/**
+ * Code/NL pair extraction and alignment scoring.
+ *
+ * Measures how well a model captures the relationship between
+ * code chunks and their natural language explanations.
+ */
+import type { LabeledPair } from './annotation-schema.js';
+export interface AlignmentResult {
+    /** Mean angular distance of code/NL pairs. */
+    meanCodeNLDistance: number;
+    /** Mean angular distance of random baseline pairs. */
+    meanRandomDistance: number;
+    /** Ratio of code-NL distance to random distance. Lower = better alignment. */
+    alignmentRatio: number;
+    /** Number of code/NL pairs evaluated. */
+    pairCount: number;
+}
+/**
+ * Evaluate code-NL alignment quality.
+ *
+ * Compares the mean distance of code/NL pairs against a random baseline
+ * drawn from unrelated pairs.
+ */
+export declare function evaluateCodeNLAlignment(pairs: LabeledPair[], embeddings: Map<string, number[]>): AlignmentResult;
+//# sourceMappingURL=code-nl-alignment.d.ts.map

package/dist/eval/code-nl-alignment.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"code-nl-alignment.d.ts","sourceRoot":"","sources":["../../src/eval/code-nl-alignment.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAGH,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,wBAAwB,CAAC;AAE1D,MAAM,WAAW,eAAe;IAC9B,8CAA8C;IAC9C,kBAAkB,EAAE,MAAM,CAAC;IAC3B,sDAAsD;IACtD,kBAAkB,EAAE,MAAM,CAAC;IAC3B,8EAA8E;IAC9E,cAAc,EAAE,MAAM,CAAC;IACvB,yCAAyC;IACzC,SAAS,EAAE,MAAM,CAAC;CACnB;AAED;;;;;GAKG;AACH,wBAAgB,uBAAuB,CACrC,KAAK,EAAE,WAAW,EAAE,EACpB,UAAU,EAAE,GAAG,CAAC,MAAM,EAAE,MAAM,EAAE,CAAC,GAChC,eAAe,CAiCjB"}

package/dist/eval/code-nl-alignment.js

@@ -0,0 +1,49 @@
+/**
+ * Code/NL pair extraction and alignment scoring.
+ *
+ * Measures how well a model captures the relationship between
+ * code chunks and their natural language explanations.
+ */
+import { angularDistance } from '../utils/angular-distance.js';
+/**
+ * Evaluate code-NL alignment quality.
+ *
+ * Compares the mean distance of code/NL pairs against a random baseline
+ * drawn from unrelated pairs.
+ */
+export function evaluateCodeNLAlignment(pairs, embeddings) {
+    const codeNLPairs = pairs.filter((p) => p.label === 'code-nl-pair');
+    const unrelatedPairs = pairs.filter((p) => p.label === 'unrelated');
+    // Score code/NL pairs
+    const codeNLDistances = [];
+    for (const pair of codeNLPairs) {
+        const a = embeddings.get(pair.chunkIdA);
+        const b = embeddings.get(pair.chunkIdB);
+        if (a && b) {
+            codeNLDistances.push(angularDistance(a, b));
+        }
+    }
+    // Score random baseline from unrelated pairs
+    const randomDistances = [];
+    for (const pair of unrelatedPairs) {
+        const a = embeddings.get(pair.chunkIdA);
+        const b = embeddings.get(pair.chunkIdB);
+        if (a && b) {
+            randomDistances.push(angularDistance(a, b));
+        }
+    }
+    const meanCodeNL = mean(codeNLDistances);
+    const meanRandom = mean(randomDistances);
+    return {
+        meanCodeNLDistance: meanCodeNL,
+        meanRandomDistance: meanRandom,
+        alignmentRatio: meanRandom > 0 ? meanCodeNL / meanRandom : 1,
+        pairCount: codeNLDistances.length,
+    };
+}
+function mean(arr) {
+    if (arr.length === 0)
+        return 0;
+    return arr.reduce((a, b) => a + b, 0) / arr.length;
+}
+//# sourceMappingURL=code-nl-alignment.js.map

package/dist/eval/code-nl-alignment.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"code-nl-alignment.js","sourceRoot":"","sources":["../../src/eval/code-nl-alignment.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH,OAAO,EAAE,eAAe,EAAE,MAAM,8BAA8B,CAAC;AAc/D;;;;;GAKG;AACH,MAAM,UAAU,uBAAuB,CACrC,KAAoB,EACpB,UAAiC;IAEjC,MAAM,WAAW,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,KAAK,cAAc,CAAC,CAAC;IACpE,MAAM,cAAc,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,KAAK,WAAW,CAAC,CAAC;IAEpE,sBAAsB;IACtB,MAAM,eAAe,GAAa,EAAE,CAAC;IACrC,KAAK,MAAM,IAAI,IAAI,WAAW,EAAE,CAAC;QAC/B,MAAM,CAAC,GAAG,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;QACxC,MAAM,CAAC,GAAG,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;QACxC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;YACX,eAAe,CAAC,IAAI,CAAC,eAAe,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC;QAC9C,CAAC;IACH,CAAC;IAED,6CAA6C;IAC7C,MAAM,eAAe,GAAa,EAAE,CAAC;IACrC,KAAK,MAAM,IAAI,IAAI,cAAc,EAAE,CAAC;QAClC,MAAM,CAAC,GAAG,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;QACxC,MAAM,CAAC,GAAG,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;QACxC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;YACX,eAAe,CAAC,IAAI,CAAC,eAAe,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC;QAC9C,CAAC;IACH,CAAC;IAED,MAAM,UAAU,GAAG,IAAI,CAAC,eAAe,CAAC,CAAC;IACzC,MAAM,UAAU,GAAG,IAAI,CAAC,eAAe,CAAC,CAAC;IAEzC,OAAO;QACL,kBAAkB,EAAE,UAAU;QAC9B,kBAAkB,EAAE,UAAU;QAC9B,cAAc,EAAE,UAAU,GAAG,CAAC,CAAC,CAAC,CAAC,UAAU,GAAG,UAAU,CAAC,CAAC,CAAC,CAAC;QAC5D,SAAS,EAAE,eAAe,CAAC,MAAM;KAClC,CAAC;AACJ,CAAC;AAED,SAAS,IAAI,CAAC,GAAa;IACzB,IAAI,GAAG,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,CAAC,CAAC;IAC/B,OAAO,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,GAAG,GAAG,CAAC,MAAM,CAAC;AACrD,CAAC"}

package/dist/eval/collection-benchmark/graph-value.d.ts

@@ -0,0 +1,16 @@
+/**
+ * Graph value benchmarks.
+ *
+ * Quantifies what the knowledge graph adds beyond pure vector similarity.
+ * Measures source attribution, graph vs vector-only comparison, and
+ * edge type effectiveness.
+ */
+import type { GraphValueResult, BenchmarkSample, SkippedBenchmark } from './types.js';
+/**
+ * Run graph value benchmarks.
+ */
+export declare function runGraphValueBenchmarks(sample: BenchmarkSample, topK: number, onProgress?: (msg: string) => void): Promise<{
+    result: GraphValueResult;
+    skipped: SkippedBenchmark[];
+}>;
+//# sourceMappingURL=graph-value.d.ts.map

package/dist/eval/collection-benchmark/graph-value.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"graph-value.d.ts","sourceRoot":"","sources":["../../../src/eval/collection-benchmark/graph-value.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AASH,OAAO,KAAK,EACV,gBAAgB,EAGhB,eAAe,EACf,gBAAgB,EACjB,MAAM,YAAY,CAAC;AAGpB;;GAEG;AACH,wBAAsB,uBAAuB,CAC3C,MAAM,EAAE,eAAe,EACvB,IAAI,EAAE,MAAM,EACZ,UAAU,CAAC,EAAE,CAAC,GAAG,EAAE,MAAM,KAAK,IAAI,GACjC,OAAO,CAAC;IAAE,MAAM,EAAE,gBAAgB,CAAC;IAAC,OAAO,EAAE,gBAAgB,EAAE,CAAA;CAAE,CAAC,CAsJpE"}

package/dist/eval/collection-benchmark/graph-value.js

@@ -0,0 +1,162 @@
+/**
+ * Graph value benchmarks.
+ *
+ * Quantifies what the knowledge graph adds beyond pure vector similarity.
+ * Measures source attribution, graph vs vector-only comparison, and
+ * edge type effectiveness.
+ */
+import { getChunkById } from '../../storage/chunk-store.js';
+import { getOutgoingEdges, getIncomingEdges } from '../../storage/edge-store.js';
+import { assembleContext, } from '../../retrieval/context-assembler.js';
+/**
+ * Run graph value benchmarks.
+ */
+export async function runGraphValueBenchmarks(sample, topK, onProgress) {
+    const skipped = [];
+    // Source attribution: run retrieval and count by source
+    const sourceCounts = { vector: 0, keyword: 0, cluster: 0, graph: 0 };
+    let totalResults = 0;
+    let vectorOnlyResults = 0;
+    let totalGraphBoosted = 0;
+    const queryIds = sample.queryChunkIds.slice(0, Math.min(30, sample.queryChunkIds.length));
+    let processed = 0;
+    // Full pipeline results + vector-only results for comparison
+    const fullResults = [];
+    const vectorOnlyResponses = [];
+    for (const chunkId of queryIds) {
+        const chunk = getChunkById(chunkId);
+        if (!chunk)
+            continue;
+        onProgress?.(`[${++processed}/${queryIds.length}] Graph value analysis...`);
+        // Full pipeline
+        const fullResponse = await assembleContext({
+            query: chunk.content.slice(0, 500),
+            mode: 'recall',
+            projectFilter: chunk.sessionSlug,
+            maxTokens: 10000,
+            vectorSearchLimit: topK * 2,
+        });
+        fullResults.push(fullResponse);
+        totalGraphBoosted += fullResponse.graphBoosted ?? 0;
+        for (const result of fullResponse.chunks) {
+            totalResults++;
+            const source = result.source ?? 'vector';
+            sourceCounts[source]++;
+        }
+        // Vector-only (skip graph and clusters)
+        const vectorOnlyResponse = await assembleContext({
+            query: chunk.content.slice(0, 500),
+            mode: 'recall',
+            projectFilter: chunk.sessionSlug,
+            maxTokens: 10000,
+            vectorSearchLimit: topK * 2,
+            skipGraph: true,
+            skipClusters: true,
+        });
+        vectorOnlyResponses.push(vectorOnlyResponse);
+        vectorOnlyResults += vectorOnlyResponse.chunks.length;
+    }
+    // Source attribution
+    const sourceAttribution = {
+        vectorPercentage: totalResults > 0 ? sourceCounts.vector / totalResults : 0,
+        keywordPercentage: totalResults > 0 ? sourceCounts.keyword / totalResults : 0,
+        clusterPercentage: totalResults > 0 ? sourceCounts.cluster / totalResults : 0,
+        graphPercentage: totalResults > 0 ? sourceCounts.graph / totalResults : 0,
+        augmentationRatio: vectorOnlyResults > 0 ? totalResults / vectorOnlyResults : 1,
+    };
+    // Graph vs vector-only comparison using adjacent pairs as ground truth
+    let fullRecallAt10 = 0;
+    let vectorOnlyRecallAt10 = 0;
+    let uniqueGraphFinds = 0;
+    let comparisonCount = 0;
+    if (sample.adjacentPairs.length > 0) {
+        // Build a map: queryChunkId -> adjacent pairs
+        const adjacentMap = new Map();
+        for (const pair of sample.adjacentPairs) {
+            const list = adjacentMap.get(pair.queryChunkId) ?? [];
+            list.push(pair.adjacentChunkId);
+            adjacentMap.set(pair.queryChunkId, list);
+        }
+        for (let i = 0; i < queryIds.length; i++) {
+            const chunkId = queryIds[i];
+            const adjacents = adjacentMap.get(chunkId);
+            if (!adjacents || adjacents.length === 0)
+                continue;
+            const fullResultIds = new Set(fullResults[i]?.chunks.slice(0, topK).map(c => c.id) ?? []);
+            const vectorOnlyIds = new Set(vectorOnlyResponses[i]?.chunks.slice(0, topK).map(c => c.id) ?? []);
+            for (const adjId of adjacents) {
+                if (fullResultIds.has(adjId))
+                    fullRecallAt10++;
+                if (vectorOnlyIds.has(adjId))
+                    vectorOnlyRecallAt10++;
+                if (fullResultIds.has(adjId) && !vectorOnlyIds.has(adjId))
+                    uniqueGraphFinds++;
+                comparisonCount++;
+            }
+        }
+    }
+    const normalizedFullRecall = comparisonCount > 0 ? fullRecallAt10 / comparisonCount : 0;
+    const normalizedVectorOnlyRecall = comparisonCount > 0 ? vectorOnlyRecallAt10 / comparisonCount : 0;
+    const lift = normalizedVectorOnlyRecall > 0
+        ? (normalizedFullRecall - normalizedVectorOnlyRecall) / normalizedVectorOnlyRecall
+        : normalizedFullRecall > 0 ? 1 : 0;
+    // Edge type effectiveness
+    const edgeTypeEffectiveness = [];
+    const edgeTypeSurfaced = new Map();
+    const edgeTypeRecallHits = new Map();
+    // Count graph-sourced results by examining response metadata
+    for (const response of fullResults) {
+        for (const chunk of response.chunks) {
+            if (chunk.source === 'graph') {
+                // We can't directly track which edge type led to this result
+                // without additional instrumentation, so we approximate by
+                // looking at what edge types exist for this chunk
+                const edgeTypes = getEdgeTypesForChunk(chunk.id);
+                for (const type of edgeTypes) {
+                    edgeTypeSurfaced.set(type, (edgeTypeSurfaced.get(type) ?? 0) + 1);
+                }
+            }
+        }
+    }
+    const allEdgeTypes = [
+        'file-path', 'code-entity', 'explicit-backref', 'error-fragment',
+        'tool-output', 'adjacent', 'cross-session', 'brief', 'debrief',
+    ];
+    for (const type of allEdgeTypes) {
+        const surfaced = edgeTypeSurfaced.get(type) ?? 0;
+        if (surfaced > 0) {
+            edgeTypeEffectiveness.push({
+                type,
+                chunksSurfaced: surfaced,
+                recallContribution: totalResults > 0 ? surfaced / totalResults : 0,
+            });
+        }
+    }
+    edgeTypeEffectiveness.sort((a, b) => b.chunksSurfaced - a.chunksSurfaced);
+    return {
+        result: {
+            sourceAttribution,
+            fullRecallAt10: normalizedFullRecall,
+            vectorOnlyRecallAt10: normalizedVectorOnlyRecall,
+            uniqueGraphFinds,
+            graphBoostedCount: totalGraphBoosted,
+            lift,
+            edgeTypeEffectiveness,
+        },
+        skipped,
+    };
+}
+/**
+ * Get the reference types of edges connected to a chunk.
+ */
+function getEdgeTypesForChunk(chunkId) {
+    const outgoing = getOutgoingEdges(chunkId);
+    const incoming = getIncomingEdges(chunkId);
+    const types = new Set();
+    for (const edge of [...outgoing, ...incoming]) {
+        if (edge.referenceType)
+            types.add(edge.referenceType);
+    }
+    return [...types];
+}
+//# sourceMappingURL=graph-value.js.map

package/dist/eval/collection-benchmark/graph-value.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"graph-value.js","sourceRoot":"","sources":["../../../src/eval/collection-benchmark/graph-value.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAEH,OAAO,EAAE,YAAY,EAAE,MAAM,8BAA8B,CAAC;AAC5D,OAAO,EAAE,gBAAgB,EAAE,gBAAgB,EAAE,MAAM,6BAA6B,CAAC;AACjF,OAAO,EACL,eAAe,GAGhB,MAAM,sCAAsC,CAAC;AAU9C;;GAEG;AACH,MAAM,CAAC,KAAK,UAAU,uBAAuB,CAC3C,MAAuB,EACvB,IAAY,EACZ,UAAkC;IAElC,MAAM,OAAO,GAAuB,EAAE,CAAC;IAEvC,wDAAwD;IACxD,MAAM,YAAY,GAAG,EAAE,MAAM,EAAE,CAAC,EAAE,OAAO,EAAE,CAAC,EAAE,OAAO,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,CAAC;IACrE,IAAI,YAAY,GAAG,CAAC,CAAC;IACrB,IAAI,iBAAiB,GAAG,CAAC,CAAC;IAC1B,IAAI,iBAAiB,GAAG,CAAC,CAAC;IAE1B,MAAM,QAAQ,GAAG,MAAM,CAAC,aAAa,CAAC,KAAK,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,EAAE,EAAE,MAAM,CAAC,aAAa,CAAC,MAAM,CAAC,CAAC,CAAC;IAC1F,IAAI,SAAS,GAAG,CAAC,CAAC;IAElB,6DAA6D;IAC7D,MAAM,WAAW,GAAwB,EAAE,CAAC;IAC5C,MAAM,mBAAmB,GAAwB,EAAE,CAAC;IAEpD,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;QAC/B,MAAM,KAAK,GAAG,YAAY,CAAC,OAAO,CAAC,CAAC;QACpC,IAAI,CAAC,KAAK;YAAE,SAAS;QAErB,UAAU,EAAE,CAAC,IAAI,EAAE,SAAS,IAAI,QAAQ,CAAC,MAAM,2BAA2B,CAAC,CAAC;QAE5E,gBAAgB;QAChB,MAAM,YAAY,GAAG,MAAM,eAAe,CAAC;YACzC,KAAK,EAAE,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC;YAClC,IAAI,EAAE,QAAQ;YACd,aAAa,EAAE,KAAK,CAAC,WAAW;YAChC,SAAS,EAAE,KAAK;YAChB,iBAAiB,EAAE,IAAI,GAAG,CAAC;SAC5B,CAAC,CAAC;QACH,WAAW,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;QAC/B,iBAAiB,IAAI,YAAY,CAAC,YAAY,IAAI,CAAC,CAAC;QAEpD,KAAK,MAAM,MAAM,IAAI,YAAY,CAAC,MAAM,EAAE,CAAC;YACzC,YAAY,EAAE,CAAC;YACf,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,IAAI,QAAQ,CAAC;YACzC,YAAY,CAAC,MAAM,CAAC,EAAE,CAAC;QACzB,CAAC;QAED,wCAAwC;QACxC,MAAM,kBAAkB,GAAG,MAAM,eAAe,CAAC;YAC/C,KAAK,EAAE,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC;YAClC,IAAI,EAAE,QAAQ;YACd,aAAa,EAAE,KAAK,CAAC,WAAW;YAChC,SAAS,EAAE,KAAK;YAChB,iBAAiB,EAAE,IAAI,GAAG,CAAC;YAC3B,SAAS,EAAE,IAAI;YACf,YAAY,EAAE,IAAI;SACiD,CAAC,CAAC;QACvE,mBAAmB,CAAC,IAAI,CAAC,kBAAkB,CAAC,CAAC;QAC7C,iBAAiB,IAAI,kBAAkB,CAAC,MAAM,CAAC,MAAM,CAAC;IACxD,CAAC;IAED,qBAAqB;IACrB,MAAM,iBAAiB,GAAsB;QAC3C,gBAAgB,EAAE,YAAY,GAAG,CAAC,CAAC,CAAC,CAAC,YAAY,CAAC,MAAM,GAAG,YAAY,CAAC,CAAC,CAAC,CAAC;QAC3E,iBAAiB,EAAE,YAAY,GAAG,CAAC,CAAC,CAAC,CAAC,YAAY,CAAC,OAAO,GAAG,YAAY,CAAC,CAAC,CAAC,CAAC;QAC7E,iBAAiB,EAAE,YAAY,GAAG,CAAC,CAAC,CAAC,CAAC,YAAY,CAAC,OAAO,GAAG,YAAY,CAAC,CAAC,CAAC,CAAC;QAC7E,eAAe,EAAE,YAAY,GAAG,CAAC,CAAC,CAAC,CAAC,YAAY,CAAC,KAAK,GAAG,YAAY,CAAC,CAAC,CAAC,CAAC;QACzE,iBAAiB,EAAE,iBAAiB,GAAG,CAAC,CAAC,CAAC,CAAC,YAAY,GAAG,iBAAiB,CAAC,CAAC,CAAC,CAAC;KAChF,CAAC;IAEF,uEAAuE;IACvE,IAAI,cAAc,GAAG,CAAC,CAAC;IACvB,IAAI,oBAAoB,GAAG,CAAC,CAAC;IAC7B,IAAI,gBAAgB,GAAG,CAAC,CAAC;IACzB,IAAI,eAAe,GAAG,CAAC,CAAC;IAExB,IAAI,MAAM,CAAC,aAAa,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACpC,8CAA8C;QAC9C,MAAM,WAAW,GAAG,IAAI,GAAG,EAAoB,CAAC;QAChD,KAAK,MAAM,IAAI,IAAI,MAAM,CAAC,aAAa,EAAE,CAAC;YACxC,MAAM,IAAI,GAAG,WAAW,CAAC,GAAG,CAAC,IAAI,CAAC,YAAY,CAAC,IAAI,EAAE,CAAC;YACtD,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,eAAe,CAAC,CAAC;YAChC,WAAW,CAAC,GAAG,CAAC,IAAI,CAAC,YAAY,EAAE,IAAI,CAAC,CAAC;QAC3C,CAAC;QAED,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,QAAQ,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;YACzC,MAAM,OAAO,GAAG,QAAQ,CAAC,CAAC,CAAC,CAAC;YAC5B,MAAM,SAAS,GAAG,WAAW,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;YAC3C,IAAI,CAAC,SAAS,IAAI,SAAS,CAAC,MAAM,KAAK,CAAC;gBAAE,SAAS;YAEnD,MAAM,aAAa,GAAG,IAAI,GAAG,CAAC,WAAW,CAAC,CAAC,CAAC,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,IAAI,EAAE,CAAC,CAAC;YAC1F,MAAM,aAAa,GAAG,IAAI,GAAG,CAAC,mBAAmB,CAAC,CAAC,CAAC,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,IAAI,EAAE,CAAC,CAAC;YAElG,KAAK,MAAM,KAAK,IAAI,SAAS,EAAE,CAAC;gBAC9B,IAAI,aAAa,CAAC,GAAG,CAAC,KAAK,CAAC;oBAAE,cAAc,EAAE,CAAC;gBAC/C,IAAI,aAAa,CAAC,GAAG,CAAC,KAAK,CAAC;oBAAE,oBAAoB,EAAE,CAAC;gBACrD,IAAI,aAAa,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,aAAa,CAAC,GAAG,CAAC,KAAK,CAAC;oBAAE,gBAAgB,EAAE,CAAC;gBAC9E,eAAe,EAAE,CAAC;YACpB,CAAC;QACH,CAAC;IACH,CAAC;IAED,MAAM,oBAAoB,GAAG,eAAe,GAAG,CAAC,CAAC,CAAC,CAAC,cAAc,GAAG,eAAe,CAAC,CAAC,CAAC,CAAC,CAAC;IACxF,MAAM,0BAA0B,GAAG,eAAe,GAAG,CAAC,CAAC,CAAC,CAAC,oBAAoB,GAAG,eAAe,CAAC,CAAC,CAAC,CAAC,CAAC;IACpG,MAAM,IAAI,GAAG,0BAA0B,GAAG,CAAC;QACzC,CAAC,CAAC,CAAC,oBAAoB,GAAG,0BAA0B,CAAC,GAAG,0BAA0B;QAClF,CAAC,CAAC,oBAAoB,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IAErC,0BAA0B;IAC1B,MAAM,qBAAqB,GAA4B,EAAE,CAAC;IAC1D,MAAM,gBAAgB,GAAG,IAAI,GAAG,EAAkB,CAAC;IACnD,MAAM,kBAAkB,GAAG,IAAI,GAAG,EAAkB,CAAC;IAErD,6DAA6D;IAC7D,KAAK,MAAM,QAAQ,IAAI,WAAW,EAAE,CAAC;QACnC,KAAK,MAAM,KAAK,IAAI,QAAQ,CAAC,MAAM,EAAE,CAAC;YACpC,IAAI,KAAK,CAAC,MAAM,KAAK,OAAO,EAAE,CAAC;gBAC7B,6DAA6D;gBAC7D,2DAA2D;gBAC3D,kDAAkD;gBAClD,MAAM,SAAS,GAAG,oBAAoB,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;gBACjD,KAAK,MAAM,IAAI,IAAI,SAAS,EAAE,CAAC;oBAC7B,gBAAgB,CAAC,GAAG,CAAC,IAAI,EAAE,CAAC,gBAAgB,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;gBACpE,CAAC;YACH,CAAC;QACH,CAAC;IACH,CAAC;IAED,MAAM,YAAY,GAAoB;QACpC,WAAW,EAAE,aAAa,EAAE,kBAAkB,EAAE,gBAAgB;QAChE,aAAa,EAAE,UAAU,EAAE,eAAe,EAAE,OAAO,EAAE,SAAS;KAC/D,CAAC;IAEF,KAAK,MAAM,IAAI,IAAI,YAAY,EAAE,CAAC;QAChC,MAAM,QAAQ,GAAG,gBAAgB,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACjD,IAAI,QAAQ,GAAG,CAAC,EAAE,CAAC;YACjB,qBAAqB,CAAC,IAAI,CAAC;gBACzB,IAAI;gBACJ,cAAc,EAAE,QAAQ;gBACxB,kBAAkB,EAAE,YAAY,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,GAAG,YAAY,CAAC,CAAC,CAAC,CAAC;aACnE,CAAC,CAAC;QACL,CAAC;IACH,CAAC;IAED,qBAAqB,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,cAAc,GAAG,CAAC,CAAC,cAAc,CAAC,CAAC;IAE1E,OAAO;QACL,MAAM,EAAE;YACN,iBAAiB;YACjB,cAAc,EAAE,oBAAoB;YACpC,oBAAoB,EAAE,0BAA0B;YAChD,gBAAgB;YAChB,iBAAiB,EAAE,iBAAiB;YACpC,IAAI;YACJ,qBAAqB;SACtB;QACD,OAAO;KACR,CAAC;AACJ,CAAC;AAED;;GAEG;AACH,SAAS,oBAAoB,CAAC,OAAe;IAC3C,MAAM,QAAQ,GAAG,gBAAgB,CAAC,OAAO,CAAC,CAAC;IAC3C,MAAM,QAAQ,GAAG,gBAAgB,CAAC,OAAO,CAAC,CAAC;IAC3C,MAAM,KAAK,GAAG,IAAI,GAAG,EAAU,CAAC;IAChC,KAAK,MAAM,IAAI,IAAI,CAAC,GAAG,QAAQ,EAAE,GAAG,QAAQ,CAAC,EAAE,CAAC;QAC9C,IAAI,IAAI,CAAC,aAAa;YAAE,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC;IACxD,CAAC;IACD,OAAO,CAAC,GAAG,KAAK,CAAC,CAAC;AACpB,CAAC"}

package/dist/eval/collection-benchmark/health.d.ts

@@ -0,0 +1,12 @@
+/**
+ * Collection health benchmarks.
+ *
+ * Fast metrics that require no embedder — always runs even with --quick.
+ * Measures collection structure, graph density, cluster coverage, and orphan rate.
+ */
+import type { HealthResult } from './types.js';
+/**
+ * Run collection health benchmarks.
+ */
+export declare function runHealthBenchmarks(includeClusterQuality?: boolean): Promise<HealthResult>;
+//# sourceMappingURL=health.d.ts.map

package/dist/eval/collection-benchmark/health.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"health.d.ts","sourceRoot":"","sources":["../../../src/eval/collection-benchmark/health.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAYH,OAAO,KAAK,EACV,YAAY,EAKb,MAAM,YAAY,CAAC;AAYpB;;GAEG;AACH,wBAAsB,mBAAmB,CACvC,qBAAqB,GAAE,OAAe,GACrC,OAAO,CAAC,YAAY,CAAC,CA0KvB"}

package/dist/eval/collection-benchmark/health.js

@@ -0,0 +1,246 @@
+/**
+ * Collection health benchmarks.
+ *
+ * Fast metrics that require no embedder — always runs even with --quick.
+ * Measures collection structure, graph density, cluster coverage, and orphan rate.
+ */
+import { getAllChunks, getSessionIds, getDistinctProjects } from '../../storage/chunk-store.js';
+import { getEdgeCount, getAllEdges } from '../../storage/edge-store.js';
+import { getClusterCount, getAllClusters, getClusterChunkIds, } from '../../storage/cluster-store.js';
+import { vectorStore } from '../../storage/vector-store.js';
+import { angularDistance } from '../../utils/angular-distance.js';
+/**
+ * Compute the median of an array of numbers.
+ */
+function median(values) {
+    if (values.length === 0)
+        return 0;
+    const sorted = [...values].sort((a, b) => a - b);
+    const mid = Math.floor(sorted.length / 2);
+    return sorted.length % 2 !== 0 ? sorted[mid] : (sorted[mid - 1] + sorted[mid]) / 2;
+}
+/**
+ * Run collection health benchmarks.
+ */
+export async function runHealthBenchmarks(includeClusterQuality = false) {
+    const chunks = getAllChunks();
+    const chunkCount = chunks.length;
+    const projects = getDistinctProjects();
+    const projectCount = projects.length;
+    const sessionIds = getSessionIds();
+    const sessionCount = sessionIds.length;
+    const edgeCount = getEdgeCount();
+    const clusterCount = getClusterCount();
+    // Edge-to-chunk ratio
+    const edgeToChunkRatio = chunkCount > 0 ? edgeCount / chunkCount : 0;
+    // Cluster coverage: chunks that belong to at least one cluster
+    const allClusters = getAllClusters();
+    const chunksInClusters = new Set();
+    for (const cluster of allClusters) {
+        const memberIds = getClusterChunkIds(cluster.id);
+        for (const id of memberIds) {
+            chunksInClusters.add(id);
+        }
+    }
+    const clusterCoverage = chunkCount > 0 ? chunksInClusters.size / chunkCount : 0;
+    // Orphan chunks: no edges AND not in any cluster
+    const edges = getAllEdges();
+    const chunksWithEdges = new Set();
+    for (const edge of edges) {
+        chunksWithEdges.add(edge.sourceChunkId);
+        chunksWithEdges.add(edge.targetChunkId);
+    }
+    let orphanCount = 0;
+    for (const chunk of chunks) {
+        if (!chunksWithEdges.has(chunk.id) && !chunksInClusters.has(chunk.id)) {
+            orphanCount++;
+        }
+    }
+    const orphanChunkPercentage = chunkCount > 0 ? orphanCount / chunkCount : 0;
+    // Temporal span
+    let temporalSpan = null;
+    if (chunks.length > 0) {
+        const times = chunks.map(c => c.startTime).sort();
+        temporalSpan = { earliest: times[0], latest: times[times.length - 1] };
+    }
+    // Edge type distribution
+    const edgeTypeCounts = new Map();
+    for (const edge of edges) {
+        const type = edge.referenceType ?? 'unknown';
+        edgeTypeCounts.set(type, (edgeTypeCounts.get(type) ?? 0) + 1);
+    }
+    const edgeTypeDistribution = [...edgeTypeCounts.entries()]
+        .map(([type, count]) => ({
+        type: type,
+        count,
+        percentage: edgeCount > 0 ? count / edgeCount : 0,
+    }))
+        .sort((a, b) => b.count - a.count);
+    // Session size stats
+    let sessionSizeStats = null;
+    if (sessionCount > 0) {
+        const sessionChunkCounts = new Map();
+        for (const chunk of chunks) {
+            sessionChunkCounts.set(chunk.sessionId, (sessionChunkCounts.get(chunk.sessionId) ?? 0) + 1);
+        }
+        const counts = [...sessionChunkCounts.values()];
+        sessionSizeStats = {
+            min: Math.min(...counts),
+            max: Math.max(...counts),
+            mean: counts.reduce((a, b) => a + b, 0) / counts.length,
+            median: median(counts),
+        };
+    }
+    // Per-project breakdown
+    const perProject = [];
+    if (projectCount > 0) {
+        const projectChunkCounts = new Map();
+        const projectEdgeCounts = new Map();
+        const projectClusterCounts = new Map();
+        const projectChunksWithEdges = new Map();
+        const projectChunksInClusters = new Map();
+        for (const chunk of chunks) {
+            projectChunkCounts.set(chunk.sessionSlug, (projectChunkCounts.get(chunk.sessionSlug) ?? 0) + 1);
+        }
+        // Build chunk-to-project map
+        const chunkToProject = new Map();
+        for (const chunk of chunks) {
+            chunkToProject.set(chunk.id, chunk.sessionSlug);
+        }
+        for (const edge of edges) {
+            const proj = chunkToProject.get(edge.sourceChunkId);
+            if (proj) {
+                projectEdgeCounts.set(proj, (projectEdgeCounts.get(proj) ?? 0) + 1);
+                const edgeSet = projectChunksWithEdges.get(proj) ?? new Set();
+                edgeSet.add(edge.sourceChunkId);
+                edgeSet.add(edge.targetChunkId);
+                projectChunksWithEdges.set(proj, edgeSet);
+            }
+        }
+        for (const cluster of allClusters) {
+            const memberIds = getClusterChunkIds(cluster.id);
+            for (const id of memberIds) {
+                const proj = chunkToProject.get(id);
+                if (proj) {
+                    const clusterSet = projectClusterCounts.get(proj) ?? new Set();
+                    clusterSet.add(cluster.id);
+                    projectClusterCounts.set(proj, clusterSet);
+                    const inCluster = projectChunksInClusters.get(proj) ?? new Set();
+                    inCluster.add(id);
+                    projectChunksInClusters.set(proj, inCluster);
+                }
+            }
+        }
+        for (const project of projects) {
+            const pChunkCount = projectChunkCounts.get(project.slug) ?? 0;
+            const pEdgeCount = projectEdgeCounts.get(project.slug) ?? 0;
+            const pClusterCount = projectClusterCounts.get(project.slug)?.size ?? 0;
+            const pChunksWithEdges = projectChunksWithEdges.get(project.slug) ?? new Set();
+            const pChunksInClusters = projectChunksInClusters.get(project.slug) ?? new Set();
+            let orphans = 0;
+            for (const chunk of chunks) {
+                if (chunk.sessionSlug === project.slug &&
+                    !pChunksWithEdges.has(chunk.id) &&
+                    !pChunksInClusters.has(chunk.id)) {
+                    orphans++;
+                }
+            }
+            perProject.push({
+                slug: project.slug,
+                chunkCount: pChunkCount,
+                edgeCount: pEdgeCount,
+                clusterCount: pClusterCount,
+                orphanPercentage: pChunkCount > 0 ? orphans / pChunkCount : 0,
+            });
+        }
+    }
+    // Cluster quality (requires embedder/vectors loaded)
+    let clusterQuality = null;
+    if (includeClusterQuality && allClusters.length >= 2) {
+        clusterQuality = await computeClusterQuality(allClusters);
+    }
+    return {
+        chunkCount,
+        projectCount,
+        sessionCount,
+        edgeCount,
+        edgeToChunkRatio,
+        clusterCount,
+        clusterCoverage,
+        orphanChunkPercentage,
+        temporalSpan,
+        edgeTypeDistribution,
+        sessionSizeStats,
+        perProject,
+        clusterQuality,
+    };
+}
+/**
+ * Compute cluster quality metrics using stored embeddings.
+ */
+async function computeClusterQuality(clusters) {
+    // Find clusters with >= 2 members and centroids
+    const eligibleClusters = [];
+    for (const cluster of clusters) {
+        if (!cluster.centroid)
+            continue;
+        const memberIds = getClusterChunkIds(cluster.id);
+        if (memberIds.length >= 2) {
+            eligibleClusters.push({
+                id: cluster.id,
+                centroid: cluster.centroid,
+                memberIds,
+            });
+        }
+    }
+    if (eligibleClusters.length < 2)
+        return null;
+    // Intra-cluster similarity: mean pairwise cosine similarity within each cluster
+    let totalIntraSim = 0;
+    let intraCount = 0;
+    for (const cluster of eligibleClusters) {
+        const embeddings = [];
+        for (const id of cluster.memberIds) {
+            const vec = await vectorStore.get(id);
+            if (vec)
+                embeddings.push(vec);
+        }
+        if (embeddings.length < 2)
+            continue;
+        // Sample pairwise comparisons (cap at 100 pairs per cluster)
+        let pairCount = 0;
+        let simSum = 0;
+        for (let i = 0; i < embeddings.length && pairCount < 100; i++) {
+            for (let j = i + 1; j < embeddings.length && pairCount < 100; j++) {
+                const dist = angularDistance(embeddings[i], embeddings[j]);
+                simSum += 1 - dist; // Convert distance to similarity
+                pairCount++;
+            }
+        }
+        if (pairCount > 0) {
+            totalIntraSim += simSum / pairCount;
+            intraCount++;
+        }
+    }
+    const intraClusterSimilarity = intraCount > 0 ? totalIntraSim / intraCount : 0;
+    // Inter-cluster separation: mean cosine distance between centroids
+    let interSum = 0;
+    let interCount = 0;
+    for (let i = 0; i < eligibleClusters.length; i++) {
+        for (let j = i + 1; j < eligibleClusters.length; j++) {
+            interSum += angularDistance(eligibleClusters[i].centroid, eligibleClusters[j].centroid);
+            interCount++;
+        }
+    }
+    const interClusterSeparation = interCount > 0 ? interSum / interCount : 0;
+    // Coherence score: intra / (intra + inter)
+    const coherenceScore = (intraClusterSimilarity + interClusterSeparation) > 0
+        ? intraClusterSimilarity / (intraClusterSimilarity + interClusterSeparation)
+        : 0;
+    return {
+        intraClusterSimilarity,
+        interClusterSeparation,
+        coherenceScore,
+    };
+}
+//# sourceMappingURL=health.js.map

package/dist/eval/collection-benchmark/health.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"health.js","sourceRoot":"","sources":["../../../src/eval/collection-benchmark/health.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH,OAAO,EAAE,YAAY,EAAiB,aAAa,EAAE,mBAAmB,EAAE,MAAM,8BAA8B,CAAC;AAC/G,OAAO,EAAE,YAAY,EAAE,WAAW,EAAE,MAAM,6BAA6B,CAAC;AACxE,OAAO,EACL,eAAe,EACf,cAAc,EACd,kBAAkB,GACnB,MAAM,gCAAgC,CAAC;AACxC,OAAO,EAAE,WAAW,EAAE,MAAM,+BAA+B,CAAC;AAC5D,OAAO,EAAE,eAAe,EAAE,MAAM,iCAAiC,CAAC;AAUlE;;GAEG;AACH,SAAS,MAAM,CAAC,MAAgB;IAC9B,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,CAAC,CAAC;IAClC,MAAM,MAAM,GAAG,CAAC,GAAG,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IACjD,MAAM,GAAG,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;IAC1C,OAAO,MAAM,CAAC,MAAM,GAAG,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,GAAG,CAAC,CAAC,GAAG,MAAM,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC;AACrF,CAAC;AAED;;GAEG;AACH,MAAM,CAAC,KAAK,UAAU,mBAAmB,CACvC,wBAAiC,KAAK;IAEtC,MAAM,MAAM,GAAG,YAAY,EAAE,CAAC;IAC9B,MAAM,UAAU,GAAG,MAAM,CAAC,MAAM,CAAC;IACjC,MAAM,QAAQ,GAAG,mBAAmB,EAAE,CAAC;IACvC,MAAM,YAAY,GAAG,QAAQ,CAAC,MAAM,CAAC;IACrC,MAAM,UAAU,GAAG,aAAa,EAAE,CAAC;IACnC,MAAM,YAAY,GAAG,UAAU,CAAC,MAAM,CAAC;IACvC,MAAM,SAAS,GAAG,YAAY,EAAE,CAAC;IACjC,MAAM,YAAY,GAAG,eAAe,EAAE,CAAC;IAEvC,sBAAsB;IACtB,MAAM,gBAAgB,GAAG,UAAU,GAAG,CAAC,CAAC,CAAC,CAAC,SAAS,GAAG,UAAU,CAAC,CAAC,CAAC,CAAC,CAAC;IAErE,+DAA+D;IAC/D,MAAM,WAAW,GAAG,cAAc,EAAE,CAAC;IACrC,MAAM,gBAAgB,GAAG,IAAI,GAAG,EAAU,CAAC;IAC3C,KAAK,MAAM,OAAO,IAAI,WAAW,EAAE,CAAC;QAClC,MAAM,SAAS,GAAG,kBAAkB,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC;QACjD,KAAK,MAAM,EAAE,IAAI,SAAS,EAAE,CAAC;YAC3B,gBAAgB,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;QAC3B,CAAC;IACH,CAAC;IACD,MAAM,eAAe,GAAG,UAAU,GAAG,CAAC,CAAC,CAAC,CAAC,gBAAgB,CAAC,IAAI,GAAG,UAAU,CAAC,CAAC,CAAC,CAAC,CAAC;IAEhF,iDAAiD;IACjD,MAAM,KAAK,GAAG,WAAW,EAAE,CAAC;IAC5B,MAAM,eAAe,GAAG,IAAI,GAAG,EAAU,CAAC;IAC1C,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QACzB,eAAe,CAAC,GAAG,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC;QACxC,eAAe,CAAC,GAAG,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC;IAC1C,CAAC;IACD,IAAI,WAAW,GAAG,CAAC,CAAC;IACpB,KAAK,MAAM,KAAK,IAAI,MAAM,EAAE,CAAC;QAC3B,IAAI,CAAC,eAAe,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC,IAAI,CAAC,gBAAgB,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC,EAAE,CAAC;YACtE,WAAW,EAAE,CAAC;QAChB,CAAC;IACH,CAAC;IACD,MAAM,qBAAqB,GAAG,UAAU,GAAG,CAAC,CAAC,CAAC,CAAC,WAAW,GAAG,UAAU,CAAC,CAAC,CAAC,CAAC,CAAC;IAE5E,gBAAgB;IAChB,IAAI,YAAY,GAAgD,IAAI,CAAC;IACrE,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACtB,MAAM,KAAK,GAAG,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,IAAI,EAAE,CAAC;QAClD,YAAY,GAAG,EAAE,QAAQ,EAAE,KAAK,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,KAAK,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC,EAAE,CAAC;IACzE,CAAC;IAED,yBAAyB;IACzB,MAAM,cAAc,GAAG,IAAI,GAAG,EAAkB,CAAC;IACjD,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QACzB,MAAM,IAAI,GAAG,IAAI,CAAC,aAAa,IAAI,SAAS,CAAC;QAC7C,cAAc,CAAC,GAAG,CAAC,IAAI,EAAE,CAAC,cAAc,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IAChE,CAAC;IACD,MAAM,oBAAoB,GAA2B,CAAC,GAAG,cAAc,CAAC,OAAO,EAAE,CAAC;SAC/E,GAAG,CAAC,CAAC,CAAC,IAAI,EAAE,KAAK,CAAC,EAAE,EAAE,CAAC,CAAC;QACvB,IAAI,EAAE,IAAqB;QAC3B,KAAK;QACL,UAAU,EAAE,SAAS,GAAG,CAAC,CAAC,CAAC,CAAC,KAAK,GAAG,SAAS,CAAC,CAAC,CAAC,CAAC;KAClD,CAAC,CAAC;SACF,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,GAAG,CAAC,CAAC,KAAK,CAAC,CAAC;IAErC,qBAAqB;IACrB,IAAI,gBAAgB,GAA4B,IAAI,CAAC;IACrD,IAAI,YAAY,GAAG,CAAC,EAAE,CAAC;QACrB,MAAM,kBAAkB,GAAG,IAAI,GAAG,EAAkB,CAAC;QACrD,KAAK,MAAM,KAAK,IAAI,MAAM,EAAE,CAAC;YAC3B,kBAAkB,CAAC,GAAG,CAAC,KAAK,CAAC,SAAS,EAAE,CAAC,kBAAkB,CAAC,GAAG,CAAC,KAAK,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;QAC9F,CAAC;QACD,MAAM,MAAM,GAAG,CAAC,GAAG,kBAAkB,CAAC,MAAM,EAAE,CAAC,CAAC;QAChD,gBAAgB,GAAG;YACjB,GAAG,EAAE,IAAI,CAAC,GAAG,CAAC,GAAG,MAAM,CAAC;YACxB,GAAG,EAAE,IAAI,CAAC,GAAG,CAAC,GAAG,MAAM,CAAC;YACxB,IAAI,EAAE,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,GAAG,MAAM,CAAC,MAAM;YACvD,MAAM,EAAE,MAAM,CAAC,MAAM,CAAC;SACvB,CAAC;IACJ,CAAC;IAED,wBAAwB;IACxB,MAAM,UAAU,GAAuB,EAAE,CAAC;IAC1C,IAAI,YAAY,GAAG,CAAC,EAAE,CAAC;QACrB,MAAM,kBAAkB,GAAG,IAAI,GAAG,EAAkB,CAAC;QACrD,MAAM,iBAAiB,GAAG,IAAI,GAAG,EAAkB,CAAC;QACpD,MAAM,oBAAoB,GAAG,IAAI,GAAG,EAAuB,CAAC;QAC5D,MAAM,sBAAsB,GAAG,IAAI,GAAG,EAAuB,CAAC;QAC9D,MAAM,uBAAuB,GAAG,IAAI,GAAG,EAAuB,CAAC;QAE/D,KAAK,MAAM,KAAK,IAAI,MAAM,EAAE,CAAC;YAC3B,kBAAkB,CAAC,GAAG,CAAC,KAAK,CAAC,WAAW,EAAE,CAAC,kBAAkB,CAAC,GAAG,CAAC,KAAK,CAAC,WAAW,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;QAClG,CAAC;QAED,6BAA6B;QAC7B,MAAM,cAAc,GAAG,IAAI,GAAG,EAAkB,CAAC;QACjD,KAAK,MAAM,KAAK,IAAI,MAAM,EAAE,CAAC;YAC3B,cAAc,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,EAAE,KAAK,CAAC,WAAW,CAAC,CAAC;QAClD,CAAC;QAED,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;YACzB,MAAM,IAAI,GAAG,cAAc,CAAC,GAAG,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC;YACpD,IAAI,IAAI,EAAE,CAAC;gBACT,iBAAiB,CAAC,GAAG,CAAC,IAAI,EAAE,CAAC,iBAAiB,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;gBACpE,MAAM,OAAO,GAAG,sBAAsB,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,IAAI,GAAG,EAAE,CAAC;gBAC9D,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC;gBAChC,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC;gBAChC,sBAAsB,CAAC,GAAG,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;YAC5C,CAAC;QACH,CAAC;QAED,KAAK,MAAM,OAAO,IAAI,WAAW,EAAE,CAAC;YAClC,MAAM,SAAS,GAAG,kBAAkB,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC;YACjD,KAAK,MAAM,EAAE,IAAI,SAAS,EAAE,CAAC;gBAC3B,MAAM,IAAI,GAAG,cAAc,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;gBACpC,IAAI,IAAI,EAAE,CAAC;oBACT,MAAM,UAAU,GAAG,oBAAoB,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,IAAI,GAAG,EAAE,CAAC;oBAC/D,UAAU,CAAC,GAAG,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC;oBAC3B,oBAAoB,CAAC,GAAG,CAAC,IAAI,EAAE,UAAU,CAAC,CAAC;oBAE3C,MAAM,SAAS,GAAG,uBAAuB,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,IAAI,GAAG,EAAE,CAAC;oBACjE,SAAS,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;oBAClB,uBAAuB,CAAC,GAAG,CAAC,IAAI,EAAE,SAAS,CAAC,CAAC;gBAC/C,CAAC;YACH,CAAC;QACH,CAAC;QAED,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;YAC/B,MAAM,WAAW,GAAG,kBAAkB,CAAC,GAAG,CAAC,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YAC9D,MAAM,UAAU,GAAG,iBAAiB,CAAC,GAAG,CAAC,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YAC5D,MAAM,aAAa,GAAG,oBAAoB,CAAC,GAAG,CAAC,OAAO,CAAC,IAAI,CAAC,EAAE,IAAI,IAAI,CAAC,CAAC;YAExE,MAAM,gBAAgB,GAAG,sBAAsB,CAAC,GAAG,CAAC,OAAO,CAAC,IAAI,CAAC,IAAI,IAAI,GAAG,EAAE,CAAC;YAC/E,MAAM,iBAAiB,GAAG,uBAAuB,CAAC,GAAG,CAAC,OAAO,CAAC,IAAI,CAAC,IAAI,IAAI,GAAG,EAAE,CAAC;YAEjF,IAAI,OAAO,GAAG,CAAC,CAAC;YAChB,KAAK,MAAM,KAAK,IAAI,MAAM,EAAE,CAAC;gBAC3B,IAAI,KAAK,CAAC,WAAW,KAAK,OAAO,CAAC,IAAI;oBAClC,CAAC,gBAAgB,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC;oBAC/B,CAAC,iBAAiB,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC,EAAE,CAAC;oBACrC,OAAO,EAAE,CAAC;gBACZ,CAAC;YACH,CAAC;YAED,UAAU,CAAC,IAAI,CAAC;gBACd,IAAI,EAAE,OAAO,CAAC,IAAI;gBAClB,UAAU,EAAE,WAAW;gBACvB,SAAS,EAAE,UAAU;gBACrB,YAAY,EAAE,aAAa;gBAC3B,gBAAgB,EAAE,WAAW,GAAG,CAAC,CAAC,CAAC,CAAC,OAAO,GAAG,WAAW,CAAC,CAAC,CAAC,CAAC;aAC9D,CAAC,CAAC;QACL,CAAC;IACH,CAAC;IAED,qDAAqD;IACrD,IAAI,cAAc,GAA0B,IAAI,CAAC;IACjD,IAAI,qBAAqB,IAAI,WAAW,CAAC,MAAM,IAAI,CAAC,EAAE,CAAC;QACrD,cAAc,GAAG,MAAM,qBAAqB,CAAC,WAAW,CAAC,CAAC;IAC5D,CAAC;IAED,OAAO;QACL,UAAU;QACV,YAAY;QACZ,YAAY;QACZ,SAAS;QACT,gBAAgB;QAChB,YAAY;QACZ,eAAe;QACf,qBAAqB;QACrB,YAAY;QACZ,oBAAoB;QACpB,gBAAgB;QAChB,UAAU;QACV,cAAc;KACf,CAAC;AACJ,CAAC;AAED;;GAEG;AACH,KAAK,UAAU,qBAAqB,CAClC,QAA0D;IAE1D,gDAAgD;IAChD,MAAM,gBAAgB,GAAmE,EAAE,CAAC;IAE5F,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;QAC/B,IAAI,CAAC,OAAO,CAAC,QAAQ;YAAE,SAAS;QAChC,MAAM,SAAS,GAAG,kBAAkB,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC;QACjD,IAAI,SAAS,CAAC,MAAM,IAAI,CAAC,EAAE,CAAC;YAC1B,gBAAgB,CAAC,IAAI,CAAC;gBACpB,EAAE,EAAE,OAAO,CAAC,EAAE;gBACd,QAAQ,EAAE,OAAO,CAAC,QAAQ;gBAC1B,SAAS;aACV,CAAC,CAAC;QACL,CAAC;IACH,CAAC;IAED,IAAI,gBAAgB,CAAC,MAAM,GAAG,CAAC;QAAE,OAAO,IAAI,CAAC;IAE7C,gFAAgF;IAChF,IAAI,aAAa,GAAG,CAAC,CAAC;IACtB,IAAI,UAAU,GAAG,CAAC,CAAC;IAEnB,KAAK,MAAM,OAAO,IAAI,gBAAgB,EAAE,CAAC;QACvC,MAAM,UAAU,GAAe,EAAE,CAAC;QAClC,KAAK,MAAM,EAAE,IAAI,OAAO,CAAC,SAAS,EAAE,CAAC;YACnC,MAAM,GAAG,GAAG,MAAM,WAAW,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;YACtC,IAAI,GAAG;gBAAE,UAAU,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QAChC,CAAC;QAED,IAAI,UAAU,CAAC,MAAM,GAAG,CAAC;YAAE,SAAS;QAEpC,6DAA6D;QAC7D,IAAI,SAAS,GAAG,CAAC,CAAC;QAClB,IAAI,MAAM,GAAG,CAAC,CAAC;QACf,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,UAAU,CAAC,MAAM,IAAI,SAAS,GAAG,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC;YAC9D,KAAK,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,UAAU,CAAC,MAAM,IAAI,SAAS,GAAG,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC;gBAClE,MAAM,IAAI,GAAG,eAAe,CAAC,UAAU,CAAC,CAAC,CAAC,EAAE,UAAU,CAAC,CAAC,CAAC,CAAC,CAAC;gBAC3D,MAAM,IAAI,CAAC,GAAG,IAAI,CAAC,CAAC,iCAAiC;gBACrD,SAAS,EAAE,CAAC;YACd,CAAC;QACH,CAAC;QACD,IAAI,SAAS,GAAG,CAAC,EAAE,CAAC;YAClB,aAAa,IAAI,MAAM,GAAG,SAAS,CAAC;YACpC,UAAU,EAAE,CAAC;QACf,CAAC;IACH,CAAC;IAED,MAAM,sBAAsB,GAAG,UAAU,GAAG,CAAC,CAAC,CAAC,CAAC,aAAa,GAAG,UAAU,CAAC,CAAC,CAAC,CAAC,CAAC;IAE/E,mEAAmE;IACnE,IAAI,QAAQ,GAAG,CAAC,CAAC;IACjB,IAAI,UAAU,GAAG,CAAC,CAAC;IACnB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,gBAAgB,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QACjD,KAAK,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,gBAAgB,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;YACrD,QAAQ,IAAI,eAAe,CAAC,gBAAgB,CAAC,CAAC,CAAC,CAAC,QAAQ,EAAE,gBAAgB,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC;YACxF,UAAU,EAAE,CAAC;QACf,CAAC;IACH,CAAC;IACD,MAAM,sBAAsB,GAAG,UAAU,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,GAAG,UAAU,CAAC,CAAC,CAAC,CAAC,CAAC;IAE1E,2CAA2C;IAC3C,MAAM,cAAc,GAAG,CAAC,sBAAsB,GAAG,sBAAsB,CAAC,GAAG,CAAC;QAC1E,CAAC,CAAC,sBAAsB,GAAG,CAAC,sBAAsB,GAAG,sBAAsB,CAAC;QAC5E,CAAC,CAAC,CAAC,CAAC;IAEN,OAAO;QACL,sBAAsB;QACtB,sBAAsB;QACtB,cAAc;KACf,CAAC;AACJ,CAAC"}

package/dist/eval/collection-benchmark/history.d.ts

@@ -0,0 +1,27 @@
+/**
+ * Historical trending for benchmark runs.
+ *
+ * Stores benchmark results in SQLite and computes trends between runs.
+ */
+import type { CollectionBenchmarkResult, BenchmarkRunSummary, TrendReport } from './types.js';
+/**
+ * Store a completed benchmark run.
+ */
+export declare function storeBenchmarkRun(result: CollectionBenchmarkResult, configSnapshot?: Record<string, unknown>): void;
+/**
+ * Get past benchmark runs (most recent first).
+ */
+export declare function getBenchmarkHistory(limit?: number): BenchmarkRunSummary[];
+/**
+ * Get a full benchmark result by run ID.
+ */
+export declare function getBenchmarkRun(id: number): CollectionBenchmarkResult | null;
+/**
+ * Get the most recent benchmark run.
+ */
+export declare function getLatestBenchmarkRun(): CollectionBenchmarkResult | null;
+/**
+ * Compute trend between two benchmark runs.
+ */
+export declare function computeTrend(previous: CollectionBenchmarkResult, current: CollectionBenchmarkResult): TrendReport;
+//# sourceMappingURL=history.d.ts.map

package/dist/eval/collection-benchmark/history.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"history.d.ts","sourceRoot":"","sources":["../../../src/eval/collection-benchmark/history.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAGH,OAAO,KAAK,EACV,yBAAyB,EACzB,mBAAmB,EACnB,WAAW,EAEZ,MAAM,YAAY,CAAC;AAuBpB;;GAEG;AACH,wBAAgB,iBAAiB,CAC/B,MAAM,EAAE,yBAAyB,EACjC,cAAc,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GACvC,IAAI,CAcN;AAED;;GAEG;AACH,wBAAgB,mBAAmB,CAAC,KAAK,GAAE,MAAW,GAAG,mBAAmB,EAAE,CAsB7E;AAED;;GAEG;AACH,wBAAgB,eAAe,CAAC,EAAE,EAAE,MAAM,GAAG,yBAAyB,GAAG,IAAI,CAU5E;AAED;;GAEG;AACH,wBAAgB,qBAAqB,IAAI,yBAAyB,GAAG,IAAI,CAUxE;AAED;;GAEG;AACH,wBAAgB,YAAY,CAC1B,QAAQ,EAAE,yBAAyB,EACnC,OAAO,EAAE,yBAAyB,GACjC,WAAW,CAkEb"}