@undefineds.co/xpod 0.3.57 → 0.3.61

package/dist/storage/rdf/RdfVectorIndex.js

@@ -1,9 +1,26 @@
 "use strict";
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.RdfVectorIndex = void 0;
+exports.vectorScoreComponents = vectorScoreComponents;
+exports.normalizeEmbedding = normalizeEmbedding;
+exports.parseEmbedding = parseEmbedding;
+exports.vectorMagnitude = vectorMagnitude;
+exports.vectorScore = vectorScore;
+exports.scoredVectorDistance = scoredVectorDistance;
+exports.insertVectorComponents = insertVectorComponents;
+exports.parsePath = parsePath;
+exports.parseSummaryMetadata = parseSummaryMetadata;
+exports.buildVectorOrderClause = buildVectorOrderClause;
+exports.vectorScoreSql = vectorScoreSql;
+exports.vectorDistanceSql = vectorDistanceSql;
+exports.vectorSquaredDistanceSql = vectorSquaredDistanceSql;
+exports.vectorThresholdSql = vectorThresholdSql;
+exports.applyResultWindow = applyResultWindow;
+exports.sqlNumber = sqlNumber;
 const node_fs_1 = require("node:fs");
 const node_path_1 = require("node:path");
 const SqliteRuntime_1 = require("../SqliteRuntime");
+const RdfSearchSourceFilter_1 = require("./RdfSearchSourceFilter");
 class RdfVectorIndex {
     constructor(options) {
         this.options = options;
@@ -45,12 +62,18 @@ class RdfVectorIndex {
         start_offset,
         end_offset,
         embedding_json,
+        summary_metadata,
         dimensions,
         magnitude,
+        provider,
         model,
+        model_version,
+        input_kind,
+        input_hash,
+        projection_policy_version,
         updated_at
       )
-      VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, strftime('%Y-%m-%dT%H:%M:%fZ', 'now'))
+      VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, strftime('%Y-%m-%dT%H:%M:%fZ', 'now'))
     `);
         const insertComponent = db.prepare(`
       INSERT INTO rdf_vector_components (
@@ -62,16 +85,48 @@ class RdfVectorIndex {
       VALUES (?, ?, ?, strftime('%Y-%m-%dT%H:%M:%fZ', 'now'))
     `);
         db.transaction(() => {
-            db.prepare(`
+            if (chunks.length === 0) {
+                db.prepare(`
+          DELETE FROM rdf_vector_components
+          WHERE chunk_id IN (
+            SELECT id FROM rdf_vector_chunks WHERE source_id = ?
+          )
+        `).run(sourceId);
+                db.prepare('DELETE FROM rdf_vector_chunks WHERE source_id = ?').run(sourceId);
+                return;
+            }
+            const deleteComponents = db.prepare(`
         DELETE FROM rdf_vector_components
         WHERE chunk_id IN (
-          SELECT id FROM rdf_vector_chunks WHERE source_id = ?
+          SELECT id FROM rdf_vector_chunks
+          WHERE source_id = ?
+            AND provider = ?
+            AND model = ?
+            AND model_version = ?
+            AND input_kind = ?
+            AND projection_policy_version = ?
         )
-      `).run(sourceId);
-            db.prepare('DELETE FROM rdf_vector_chunks WHERE source_id = ?').run(sourceId);
+      `);
+            const deleteChunks = db.prepare(`
+        DELETE FROM rdf_vector_chunks
+        WHERE source_id = ?
+          AND provider = ?
+          AND model = ?
+          AND model_version = ?
+          AND input_kind = ?
+          AND projection_policy_version = ?
+      `);
+            const deletedIdentities = new Set();
             for (const chunk of chunks) {
+                const identity = normalizeVectorIdentity(chunk);
+                const identityKey = vectorIdentityKey(identity);
+                if (!deletedIdentities.has(identityKey)) {
+                    deleteComponents.run(sourceId, identity.provider, identity.model, identity.modelVersion, identity.inputKind, identity.projectionPolicyVersion);
+                    deleteChunks.run(sourceId, identity.provider, identity.model, identity.modelVersion, identity.inputKind, identity.projectionPolicyVersion);
+                    deletedIdentities.add(identityKey);
+                }
                 const embedding = normalizeEmbedding(chunk.embedding);
-                const result = insertChunk.run(sourceId, chunk.chunkKey, chunk.ordinal, chunk.level, chunk.heading || null, JSON.stringify(chunk.path ?? []), chunk.content, chunk.startOffset, chunk.endOffset, JSON.stringify(embedding), embedding.length, vectorMagnitude(embedding), chunk.model ?? '');
+                const result = insertChunk.run(sourceId, chunk.chunkKey, chunk.ordinal, chunk.level, chunk.heading || null, JSON.stringify(chunk.path ?? []), chunk.content, chunk.startOffset, chunk.endOffset, JSON.stringify(embedding), chunk.summaryMetadata ? JSON.stringify(chunk.summaryMetadata) : null, embedding.length, vectorMagnitude(embedding), identity.provider, identity.model, identity.modelVersion, identity.inputKind, chunk.inputHash ?? '', identity.projectionPolicyVersion);
                 insertVectorComponents(insertComponent, Number(result.lastInsertRowid), embedding);
             }
         })();
@@ -110,7 +165,61 @@ class RdfVectorIndex {
             .map((row) => {
             const rowEmbedding = parseEmbedding(row.embedding_json);
             const distance = scoredVectorDistance(row, metric);
-            return this.toSearchResult(row, rowEmbedding, vectorScore(distance, metric), distance);
+            return this.toSearchResult(row, rowEmbedding, metric, vectorMagnitude(embedding), vectorScore(distance, metric), distance);
+        });
+    }
+    summaryLifecycle(options = {}) {
+        const params = [];
+        const conditions = ['chunk.summary_metadata IS NOT NULL'];
+        if (options.workspace) {
+            conditions.push('source.workspace = ?');
+            params.push(options.workspace);
+        }
+        (0, RdfSearchSourceFilter_1.appendRdfSearchSourceFilters)(options, conditions, params);
+        appendVectorIdentityFilters(options, conditions, params);
+        const limitSql = options.limit === undefined ? '' : 'LIMIT ?';
+        if (options.limit !== undefined) {
+            params.push(Math.max(0, options.limit));
+        }
+        const rows = this.requireDb().prepare(`
+      SELECT
+        chunk.id,
+        chunk.source_id,
+        source.source_key,
+        source.source,
+        source.workspace,
+        source.local_path,
+        source.content_type,
+        source.source_version,
+        source.source_hash,
+        chunk.chunk_key,
+        chunk.ordinal,
+        chunk.level,
+        chunk.heading,
+        chunk.path,
+        chunk.content,
+        chunk.start_offset,
+        chunk.end_offset,
+        chunk.embedding_json,
+        chunk.summary_metadata,
+        chunk.dimensions,
+        chunk.magnitude,
+        chunk.provider,
+        chunk.model,
+        chunk.model_version,
+        chunk.input_kind,
+        chunk.input_hash,
+        chunk.projection_policy_version,
+        chunk.updated_at
+      FROM rdf_vector_chunks chunk
+      JOIN rdf_vector_sources source ON source.id = chunk.source_id
+      WHERE ${conditions.join(' AND ')}
+      ORDER BY source.source ASC, chunk.ordinal ASC, chunk.chunk_key ASC
+      ${limitSql}
+    `).all(...params);
+        return rows.flatMap((row) => {
+            const entry = toSummaryLifecycleEntry(row);
+            return entry ? [entry] : [];
         });
     }
     estimateSearchCardinality(options) {
@@ -145,18 +254,8 @@ class RdfVectorIndex {
             conditions.push('source.workspace = ?');
             params.push(options.workspace);
         }
-        if (options.source) {
-            conditions.push('source.source = ?');
-            params.push(options.source);
-        }
-        if (options.sourcePrefix) {
-            conditions.push('source.source >= ? AND source.source < ?');
-            params.push(options.sourcePrefix, `${options.sourcePrefix}\uffff`);
-        }
-        if (options.model !== undefined) {
-            conditions.push('chunk.model = ?');
-            params.push(options.model);
-        }
+        (0, RdfSearchSourceFilter_1.appendRdfSearchSourceFilters)(options, conditions, params);
+        appendVectorIdentityFilters(options, conditions, params);
         const rows = this.requireDb().prepare(`
       SELECT COUNT(*) AS count
       FROM rdf_vector_chunks chunk
@@ -182,7 +281,11 @@ class RdfVectorIndex {
     modelDistribution() {
         const rows = this.requireDb().prepare(`
       SELECT
+        chunk.provider,
         chunk.model,
+        chunk.model_version,
+        chunk.input_kind,
+        chunk.projection_policy_version,
         chunk.dimensions,
         COUNT(DISTINCT chunk.source_id) AS source_count,
         COUNT(*) AS chunk_count,
@@ -190,11 +293,15 @@ class RdfVectorIndex {
         MAX(chunk.magnitude) AS max_magnitude,
         AVG(chunk.magnitude) AS average_magnitude
       FROM rdf_vector_chunks chunk
-      GROUP BY chunk.model, chunk.dimensions
-      ORDER BY chunk_count DESC, source_count DESC, chunk.model ASC, chunk.dimensions ASC
+      GROUP BY chunk.provider, chunk.model, chunk.model_version, chunk.input_kind, chunk.projection_policy_version, chunk.dimensions
+      ORDER BY chunk_count DESC, source_count DESC, chunk.provider ASC, chunk.model ASC, chunk.model_version ASC, chunk.input_kind ASC, chunk.projection_policy_version ASC, chunk.dimensions ASC
     `).all();
         return rows.map((row) => ({
+            provider: row.provider || undefined,
             model: row.model,
+            modelVersion: row.model_version || undefined,
+            inputKind: row.input_kind || undefined,
+            projectionPolicyVersion: row.projection_policy_version || undefined,
             dimensions: row.dimensions,
             sourceCount: row.source_count,
             chunkCount: row.chunk_count,
@@ -207,6 +314,7 @@ class RdfVectorIndex {
         this.requireDb().exec(`
       CREATE TABLE IF NOT EXISTS rdf_vector_sources (
         id INTEGER PRIMARY KEY AUTOINCREMENT,
+        source_key TEXT,
         source TEXT NOT NULL UNIQUE,
         workspace TEXT NOT NULL,
         local_path TEXT,
@@ -228,11 +336,26 @@ class RdfVectorIndex {
         start_offset INTEGER NOT NULL,
         end_offset INTEGER NOT NULL,
         embedding_json TEXT NOT NULL,
+        summary_metadata TEXT,
         dimensions INTEGER NOT NULL,
         magnitude REAL NOT NULL,
+        provider TEXT NOT NULL DEFAULT '',
         model TEXT NOT NULL,
+        model_version TEXT NOT NULL DEFAULT '',
+        input_kind TEXT NOT NULL DEFAULT '',
+        input_hash TEXT NOT NULL DEFAULT '',
+        projection_policy_version TEXT NOT NULL DEFAULT '',
         updated_at TEXT NOT NULL DEFAULT (strftime('%Y-%m-%dT%H:%M:%fZ', 'now')),
-        UNIQUE (source_id, chunk_key),
+        UNIQUE (
+          source_id,
+          chunk_key,
+          provider,
+          model,
+          model_version,
+          input_kind,
+          projection_policy_version,
+          input_hash
+        ),
         FOREIGN KEY (source_id) REFERENCES rdf_vector_sources(id)
       );
 
@@ -248,11 +371,178 @@ class RdfVectorIndex {
       CREATE INDEX IF NOT EXISTS rdf_vector_sources_workspace ON rdf_vector_sources(workspace);
       CREATE INDEX IF NOT EXISTS rdf_vector_sources_source ON rdf_vector_sources(source);
       CREATE INDEX IF NOT EXISTS rdf_vector_chunks_source ON rdf_vector_chunks(source_id, ordinal);
-      CREATE INDEX IF NOT EXISTS rdf_vector_chunks_model_dimensions ON rdf_vector_chunks(model, dimensions);
       CREATE INDEX IF NOT EXISTS rdf_vector_components_dimension ON rdf_vector_components(dimension, chunk_id);
     `);
+        this.ensureSourceKeyColumn();
+        this.ensureVectorIdentityColumns();
         this.backfillVectorComponents();
     }
+    ensureSourceKeyColumn() {
+        const db = this.requireDb();
+        const columns = db.prepare('PRAGMA table_info(rdf_vector_sources)').all();
+        if (!columns.some((column) => column.name === 'source_key')) {
+            db.exec('ALTER TABLE rdf_vector_sources ADD COLUMN source_key TEXT');
+        }
+        db.exec('UPDATE rdf_vector_sources SET source_key = source WHERE source_key IS NULL OR source_key = \'\';');
+    }
+    ensureVectorIdentityColumns() {
+        const db = this.requireDb();
+        const columns = db.prepare('PRAGMA table_info(rdf_vector_chunks)').all();
+        const names = new Set(columns.map((column) => column.name));
+        for (const [name, definition] of [
+            ['provider', 'TEXT NOT NULL DEFAULT \'\''],
+            ['model_version', 'TEXT NOT NULL DEFAULT \'\''],
+            ['input_kind', 'TEXT NOT NULL DEFAULT \'\''],
+            ['input_hash', 'TEXT NOT NULL DEFAULT \'\''],
+            ['projection_policy_version', 'TEXT NOT NULL DEFAULT \'\''],
+            ['summary_metadata', 'TEXT'],
+        ]) {
+            if (!names.has(name)) {
+                db.exec(`ALTER TABLE rdf_vector_chunks ADD COLUMN ${name} ${definition}`);
+            }
+        }
+        db.exec(`
+      UPDATE rdf_vector_chunks
+      SET
+        provider = COALESCE(provider, ''),
+        model_version = COALESCE(model_version, ''),
+        input_kind = COALESCE(input_kind, ''),
+        input_hash = COALESCE(input_hash, ''),
+        projection_policy_version = COALESCE(projection_policy_version, '')
+    `);
+        this.ensureVectorChunkIdentityUniqueConstraint();
+        db.exec(`
+      CREATE INDEX IF NOT EXISTS rdf_vector_chunks_model_dimensions
+      ON rdf_vector_chunks(provider, model, model_version, input_kind, projection_policy_version, dimensions)
+    `);
+    }
+    ensureVectorChunkIdentityUniqueConstraint() {
+        const db = this.requireDb();
+        const indexes = db.prepare('PRAGMA index_list(rdf_vector_chunks)').all();
+        const hasLegacyUnique = indexes.some((index) => {
+            if (!index.unique) {
+                return false;
+            }
+            const columns = db.prepare(`PRAGMA index_info(${sqliteStringLiteral(index.name)})`).all();
+            return columns.map((column) => column.name).join(',') === 'source_id,chunk_key';
+        });
+        if (!hasLegacyUnique) {
+            return;
+        }
+        db.exec(`
+      ALTER TABLE rdf_vector_components RENAME TO rdf_vector_components_legacy_unique;
+      ALTER TABLE rdf_vector_chunks RENAME TO rdf_vector_chunks_legacy_unique;
+
+      CREATE TABLE rdf_vector_chunks (
+        id INTEGER PRIMARY KEY AUTOINCREMENT,
+        source_id INTEGER NOT NULL,
+        chunk_key TEXT NOT NULL,
+        ordinal INTEGER NOT NULL,
+        level INTEGER NOT NULL,
+        heading TEXT,
+        path TEXT,
+        content TEXT NOT NULL,
+        start_offset INTEGER NOT NULL,
+        end_offset INTEGER NOT NULL,
+        embedding_json TEXT NOT NULL,
+        summary_metadata TEXT,
+        dimensions INTEGER NOT NULL,
+        magnitude REAL NOT NULL,
+        provider TEXT NOT NULL DEFAULT '',
+        model TEXT NOT NULL,
+        model_version TEXT NOT NULL DEFAULT '',
+        input_kind TEXT NOT NULL DEFAULT '',
+        input_hash TEXT NOT NULL DEFAULT '',
+        projection_policy_version TEXT NOT NULL DEFAULT '',
+        updated_at TEXT NOT NULL DEFAULT (strftime('%Y-%m-%dT%H:%M:%fZ', 'now')),
+        UNIQUE (
+          source_id,
+          chunk_key,
+          provider,
+          model,
+          model_version,
+          input_kind,
+          projection_policy_version,
+          input_hash
+        ),
+        FOREIGN KEY (source_id) REFERENCES rdf_vector_sources(id)
+      );
+
+      CREATE TABLE rdf_vector_components (
+        chunk_id INTEGER NOT NULL,
+        dimension INTEGER NOT NULL,
+        value REAL NOT NULL,
+        updated_at TEXT NOT NULL DEFAULT (strftime('%Y-%m-%dT%H:%M:%fZ', 'now')),
+        PRIMARY KEY (chunk_id, dimension),
+        FOREIGN KEY (chunk_id) REFERENCES rdf_vector_chunks(id)
+      );
+
+      INSERT INTO rdf_vector_chunks (
+        id,
+        source_id,
+        chunk_key,
+        ordinal,
+        level,
+        heading,
+        path,
+        content,
+        start_offset,
+        end_offset,
+        embedding_json,
+        summary_metadata,
+        dimensions,
+        magnitude,
+        provider,
+        model,
+        model_version,
+        input_kind,
+        input_hash,
+        projection_policy_version,
+        updated_at
+      )
+      SELECT
+        id,
+        source_id,
+        chunk_key,
+        ordinal,
+        level,
+        heading,
+        path,
+        content,
+        start_offset,
+        end_offset,
+        embedding_json,
+        NULL AS summary_metadata,
+        dimensions,
+        magnitude,
+        COALESCE(provider, ''),
+        model,
+        COALESCE(model_version, ''),
+        COALESCE(input_kind, ''),
+        COALESCE(input_hash, ''),
+        COALESCE(projection_policy_version, ''),
+        updated_at
+      FROM rdf_vector_chunks_legacy_unique;
+
+      INSERT INTO rdf_vector_components (
+        chunk_id,
+        dimension,
+        value,
+        updated_at
+      )
+      SELECT
+        chunk_id,
+        dimension,
+        value,
+        updated_at
+      FROM rdf_vector_components_legacy_unique;
+
+      DROP TABLE rdf_vector_components_legacy_unique;
+      DROP TABLE rdf_vector_chunks_legacy_unique;
+      CREATE INDEX IF NOT EXISTS rdf_vector_chunks_source ON rdf_vector_chunks(source_id, ordinal);
+      CREATE INDEX IF NOT EXISTS rdf_vector_components_dimension ON rdf_vector_components(dimension, chunk_id);
+    `);
+    }
     backfillVectorComponents() {
         const db = this.requireDb();
         const rows = db.prepare(`
@@ -291,6 +581,7 @@ class RdfVectorIndex {
         const db = this.requireDb();
         db.prepare(`
       INSERT INTO rdf_vector_sources (
+        source_key,
         source,
         workspace,
         local_path,
@@ -299,16 +590,17 @@ class RdfVectorIndex {
         source_hash,
         updated_at
       )
-      VALUES (?, ?, ?, ?, ?, ?, strftime('%Y-%m-%dT%H:%M:%fZ', 'now'))
+      VALUES (?, ?, ?, ?, ?, ?, ?, strftime('%Y-%m-%dT%H:%M:%fZ', 'now'))
       ON CONFLICT (source)
       DO UPDATE SET
+        source_key = COALESCE(excluded.source_key, rdf_vector_sources.source_key, excluded.source),
         workspace = excluded.workspace,
         local_path = excluded.local_path,
         content_type = excluded.content_type,
         source_version = excluded.source_version,
         source_hash = excluded.source_hash,
         updated_at = excluded.updated_at
-    `).run(source.source, source.workspace, source.localPath ?? null, source.contentType ?? null, source.sourceVersion ?? null, source.sourceHash ?? null);
+    `).run(source.sourceKey ?? source.source, source.source, source.workspace, source.localPath ?? null, source.contentType ?? null, source.sourceVersion ?? null, source.sourceHash ?? null);
         const row = db.prepare('SELECT * FROM rdf_vector_sources WHERE source = ?').get(source.source);
         if (!row) {
             throw new Error(`Failed to upsert RDF vector source: ${source.source}`);
@@ -326,7 +618,7 @@ class RdfVectorIndex {
             return 0;
         }
     }
-    toSearchResult(row, embedding, score, distance) {
+    toSearchResult(row, embedding, metric, queryMagnitude, score, distance) {
         return {
             source: row.source,
             workspace: row.workspace,
@@ -334,7 +626,9 @@ class RdfVectorIndex {
             contentType: row.content_type ?? undefined,
             sourceVersion: row.source_version ?? undefined,
             sourceHash: row.source_hash ?? undefined,
+            sourceKey: row.source_key || row.source,
             chunkKey: row.chunk_key,
+            retrievalPointKey: row.chunk_key,
             ordinal: row.ordinal,
             level: row.level,
             heading: row.heading ?? undefined,
@@ -343,7 +637,14 @@ class RdfVectorIndex {
             startOffset: row.start_offset,
             endOffset: row.end_offset,
             embedding,
+            provider: row.provider || undefined,
             model: row.model || undefined,
+            modelVersion: row.model_version || undefined,
+            inputKind: row.input_kind || undefined,
+            inputHash: row.input_hash || undefined,
+            projectionPolicyVersion: row.projection_policy_version || undefined,
+            summaryMetadata: parseSummaryMetadata(row.summary_metadata),
+            scoreComponents: vectorScoreComponents(row, metric, queryMagnitude, score, distance),
             score,
             distance,
         };
@@ -356,6 +657,21 @@ class RdfVectorIndex {
     }
 }
 exports.RdfVectorIndex = RdfVectorIndex;
+function vectorScoreComponents(row, metric, queryMagnitude, score, distance, backend) {
+    const distanceSquared = row.vector_distance_squared ?? undefined;
+    return {
+        sourceType: 'vector',
+        ...(backend ? { backend } : {}),
+        metric,
+        dimensions: row.dimensions,
+        score,
+        distance,
+        dotProduct: row.dot_product,
+        queryMagnitude,
+        candidateMagnitude: row.magnitude,
+        ...(distanceSquared !== undefined ? { distanceSquared } : {}),
+    };
+}
 function normalizeEmbedding(embedding) {
     return embedding.filter((value) => Number.isFinite(value));
 }
@@ -412,6 +728,116 @@ function parsePath(value) {
         return [];
     }
 }
+function parseSummaryMetadata(value) {
+    if (!value) {
+        return undefined;
+    }
+    try {
+        const parsed = JSON.parse(value);
+        if (!parsed ||
+            parsed.status !== 'summarized' ||
+            typeof parsed.provider !== 'string' ||
+            typeof parsed.model !== 'string' ||
+            typeof parsed.promptVersion !== 'string' ||
+            typeof parsed.originalChars !== 'number' ||
+            typeof parsed.summaryChars !== 'number' ||
+            typeof parsed.rounds !== 'number') {
+            return undefined;
+        }
+        return {
+            status: 'summarized',
+            provider: parsed.provider,
+            model: parsed.model,
+            promptVersion: parsed.promptVersion,
+            ...(typeof parsed.sourceHash === 'string' ? { sourceHash: parsed.sourceHash } : {}),
+            originalChars: parsed.originalChars,
+            summaryChars: parsed.summaryChars,
+            rounds: parsed.rounds,
+        };
+    }
+    catch {
+        return undefined;
+    }
+}
+function normalizeVectorIdentity(chunk) {
+    return {
+        provider: chunk.provider ?? '',
+        model: chunk.model ?? '',
+        modelVersion: chunk.modelVersion ?? '',
+        inputKind: chunk.inputKind ?? '',
+        projectionPolicyVersion: chunk.projectionPolicyVersion ?? '',
+    };
+}
+function vectorIdentityKey(identity) {
+    return [
+        identity.provider,
+        identity.model,
+        identity.modelVersion,
+        identity.inputKind,
+        identity.projectionPolicyVersion,
+    ].join('\u0000');
+}
+function sqliteStringLiteral(value) {
+    return `'${value.replace(/'/g, '\'\'')}'`;
+}
+function appendVectorIdentityFilters(options, conditions, params) {
+    if (options.provider !== undefined) {
+        conditions.push('chunk.provider = ?');
+        params.push(options.provider);
+    }
+    if (options.model !== undefined) {
+        conditions.push('chunk.model = ?');
+        params.push(options.model);
+    }
+    if (options.modelVersion !== undefined) {
+        conditions.push('chunk.model_version = ?');
+        params.push(options.modelVersion);
+    }
+    if (options.inputKind !== undefined) {
+        conditions.push('chunk.input_kind = ?');
+        params.push(options.inputKind);
+    }
+    if (options.inputHash !== undefined) {
+        conditions.push('chunk.input_hash = ?');
+        params.push(options.inputHash);
+    }
+    if (options.projectionPolicyVersion !== undefined) {
+        conditions.push('chunk.projection_policy_version = ?');
+        params.push(options.projectionPolicyVersion);
+    }
+}
+function toSummaryLifecycleEntry(row) {
+    const summaryMetadata = parseSummaryMetadata(row.summary_metadata);
+    if (!summaryMetadata) {
+        return undefined;
+    }
+    return {
+        source: row.source,
+        workspace: row.workspace,
+        localPath: row.local_path ?? undefined,
+        contentType: row.content_type ?? undefined,
+        sourceVersion: row.source_version ?? undefined,
+        sourceHash: row.source_hash ?? undefined,
+        sourceKey: row.source_key || row.source,
+        chunkKey: row.chunk_key,
+        retrievalPointKey: row.chunk_key,
+        ordinal: row.ordinal,
+        level: row.level,
+        heading: row.heading ?? undefined,
+        path: parsePath(row.path),
+        content: row.content,
+        startOffset: row.start_offset,
+        endOffset: row.end_offset,
+        provider: row.provider || undefined,
+        model: row.model || undefined,
+        modelVersion: row.model_version || undefined,
+        inputKind: row.input_kind || undefined,
+        inputHash: row.input_hash || undefined,
+        projectionPolicyVersion: row.projection_policy_version || undefined,
+        summaryMetadata,
+        updatedAt: row.updated_at,
+    };
+}
 function buildVectorScoredRowsQuery(embedding, metric, options) {
     const scored = buildVectorScoredBaseQuery(embedding, metric, options);
     const orderBy = buildVectorOrderClause(metric, options.orderBy);
@@ -458,18 +884,8 @@ function buildVectorScoredBaseQuery(embedding, metric, options) {
         conditions.push('source.workspace = ?');
         params.push(options.workspace);
     }
-    if (options.source) {
-        conditions.push('source.source = ?');
-        params.push(options.source);
-    }
-    if (options.sourcePrefix) {
-        conditions.push('source.source >= ? AND source.source < ?');
-        params.push(options.sourcePrefix, `${options.sourcePrefix}\uffff`);
-    }
-    if (options.model !== undefined) {
-        conditions.push('chunk.model = ?');
-        params.push(options.model);
-    }
+    (0, RdfSearchSourceFilter_1.appendRdfSearchSourceFilters)(options, conditions, params);
+    appendVectorIdentityFilters(options, conditions, params);
     const thresholdWhere = options.threshold === undefined
         ? ''
         : `WHERE ${vectorThresholdSql(metric, queryMagnitude, options.threshold)}`;
@@ -482,6 +898,7 @@ function buildVectorScoredBaseQuery(embedding, metric, options) {
         SELECT
           chunk.id,
           chunk.source_id,
+          source.source_key,
           source.source,
           source.workspace,
           source.local_path,
@@ -497,9 +914,15 @@ function buildVectorScoredBaseQuery(embedding, metric, options) {
           chunk.start_offset,
           chunk.end_offset,
           chunk.embedding_json,
+          chunk.summary_metadata,
           chunk.dimensions,
           chunk.magnitude,
+          chunk.provider,
           chunk.model,
+          chunk.model_version,
+          chunk.input_kind,
+          chunk.input_hash,
+          chunk.projection_policy_version,
           chunk.updated_at,
           SUM(component.value * query_vector.value) AS dot_product
         FROM rdf_vector_chunks chunk