agentic-flow 3.0.0-alpha.1 → 3.0.0

package/dist/packages/agentdb/src/controllers/CausalMemoryGraph.js

@@ -1,527 +0,0 @@
-/**
- * CausalMemoryGraph - Causal Reasoning over Agent Memories
- *
- * Implements intervention-based reasoning rather than correlation.
- * Stores p(y|do(x)) estimates and tracks causal uplift across episodes.
- *
- * Based on:
- * - Pearl's do-calculus and causal inference
- * - Uplift modeling from A/B testing
- * - Instrumental variable methods
- *
- * v2.0.0-alpha.3 Features:
- * - HyperbolicAttention for tree-structured causal chain retrieval
- * - Poincaré embeddings for hierarchical relationships
- * - Feature flag: ENABLE_HYPERBOLIC_ATTENTION (default: false)
- * - 100% backward compatible with fallback to standard retrieval
- */
-import { NodeIdMapper } from '../utils/NodeIdMapper.js';
-import { AttentionService } from '../utils/LegacyAttentionAdapter.js';
-export class CausalMemoryGraph {
-    db;
-    graphBackend; // GraphBackend or GraphDatabaseAdapter
-    attentionService;
-    embedder;
-    config;
-    /**
-     * Constructor supports both v1 (legacy) and v2 (with attention) modes
-     *
-     * v1 mode: new CausalMemoryGraph(db)
-     * v2 mode: new CausalMemoryGraph(db, graphBackend, embedder, config)
-     */
-    constructor(db, graphBackend, embedder, config) {
-        this.db = db;
-        this.graphBackend = graphBackend;
-        this.embedder = embedder;
-        this.config = {
-            ENABLE_HYPERBOLIC_ATTENTION: false,
-            ...config,
-        };
-        // Initialize AttentionService if embedder provided
-        if (embedder && this.config.ENABLE_HYPERBOLIC_ATTENTION) {
-            this.attentionService = new AttentionService(db, {
-                hyperbolic: {
-                    enabled: true,
-                    ...this.config.hyperbolicConfig,
-                },
-            });
-        }
-    }
-    /**
-     * Add a causal edge between memories
-     */
-    async addCausalEdge(edge) {
-        // Use GraphDatabaseAdapter if available (AgentDB v2)
-        if (this.graphBackend && 'createCausalEdge' in this.graphBackend) {
-            const graphAdapter = this.graphBackend;
-            // Create embedding for causal mechanism
-            const mechanismText = edge.mechanism || `${edge.fromMemoryType}-${edge.toMemoryType} causal link`;
-            const embedding = new Float32Array(384).fill(0); // Placeholder - would use embedder in production
-            // Convert episode IDs to string format expected by graph database
-            // Use NodeIdMapper to get full node IDs from numeric IDs
-            const mapper = NodeIdMapper.getInstance();
-            const fromNodeId = typeof edge.fromMemoryId === 'string'
-                ? edge.fromMemoryId
-                : (mapper.getNodeId(edge.fromMemoryId) || `${edge.fromMemoryType}-${edge.fromMemoryId}`);
-            const toNodeId = typeof edge.toMemoryId === 'string'
-                ? edge.toMemoryId
-                : (mapper.getNodeId(edge.toMemoryId) || `${edge.toMemoryType}-${edge.toMemoryId}`);
-            const graphEdge = {
-                from: fromNodeId,
-                to: toNodeId,
-                mechanism: mechanismText,
-                uplift: edge.uplift || 0,
-                confidence: edge.confidence,
-                sampleSize: edge.sampleSize || 0
-            };
-            const edgeId = await graphAdapter.createCausalEdge(graphEdge, embedding);
-            // Convert string ID to numeric ID for compatibility
-            // Extract numeric ID from string format "type-number" or return hash
-            if (typeof edgeId === 'number') {
-                return edgeId;
-            }
-            // Parse numeric ID from string format like "edge-123"
-            const numMatch = String(edgeId).match(/(\d+)$/);
-            return numMatch ? parseInt(numMatch[1], 10) : Math.abs(this.hashString(String(edgeId)));
-        }
-        // Fallback to SQLite
-        const stmt = this.db.prepare(`
-      INSERT INTO causal_edges (
-        from_memory_id, from_memory_type, to_memory_id, to_memory_type,
-        similarity, uplift, confidence, sample_size,
-        evidence_ids, confounder_score,
-        mechanism, metadata
-      ) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
-    `);
-        const result = stmt.run(edge.fromMemoryId, edge.fromMemoryType, edge.toMemoryId, edge.toMemoryType, edge.similarity, edge.uplift || null, edge.confidence, edge.sampleSize || null, edge.evidenceIds ? JSON.stringify(edge.evidenceIds) : null, edge.confounderScore || null, edge.mechanism || null, edge.metadata ? JSON.stringify(edge.metadata) : null);
-        return result.lastInsertRowid;
-    }
-    /**
-     * Create a causal experiment (A/B test)
-     */
-    createExperiment(experiment) {
-        const stmt = this.db.prepare(`
-      INSERT INTO causal_experiments (
-        name, hypothesis, treatment_id, treatment_type, control_id,
-        start_time, sample_size, status, metadata
-      ) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?)
-    `);
-        const result = stmt.run(experiment.name, experiment.hypothesis, experiment.treatmentId, experiment.treatmentType, experiment.controlId || null, experiment.startTime, experiment.sampleSize, experiment.status, experiment.metadata ? JSON.stringify(experiment.metadata) : null);
-        return result.lastInsertRowid;
-    }
-    /**
-     * Record an observation in an experiment
-     */
-    recordObservation(observation) {
-        const stmt = this.db.prepare(`
-      INSERT INTO causal_observations (
-        experiment_id, episode_id, is_treatment, outcome_value, outcome_type, context
-      ) VALUES (?, ?, ?, ?, ?, ?)
-    `);
-        stmt.run(observation.experimentId, observation.episodeId, observation.isTreatment ? 1 : 0, observation.outcomeValue, observation.outcomeType, observation.context ? JSON.stringify(observation.context) : null);
-        // Update sample size
-        this.db.prepare(`
-      UPDATE causal_experiments
-      SET sample_size = sample_size + 1
-      WHERE id = ?
-    `).run(observation.experimentId);
-    }
-    /**
-     * Calculate uplift for an experiment
-     */
-    calculateUplift(experimentId) {
-        // Get treatment and control observations
-        const observations = this.db.prepare(`
-      SELECT is_treatment, outcome_value
-      FROM causal_observations
-      WHERE experiment_id = ?
-    `).all(experimentId);
-        const treatmentValues = observations
-            .filter(o => o.is_treatment === 1)
-            .map(o => o.outcome_value);
-        const controlValues = observations
-            .filter(o => o.is_treatment === 0)
-            .map(o => o.outcome_value);
-        if (treatmentValues.length === 0 || controlValues.length === 0) {
-            return { uplift: 0, pValue: 1.0, confidenceInterval: [0, 0] };
-        }
-        // Calculate means
-        const treatmentMean = this.mean(treatmentValues);
-        const controlMean = this.mean(controlValues);
-        const uplift = treatmentMean - controlMean;
-        // Calculate standard errors
-        const treatmentSE = this.standardError(treatmentValues);
-        const controlSE = this.standardError(controlValues);
-        const pooledSE = Math.sqrt(treatmentSE ** 2 + controlSE ** 2);
-        // t-statistic and p-value (two-tailed)
-        const tStat = uplift / pooledSE;
-        const df = treatmentValues.length + controlValues.length - 2;
-        const pValue = 2 * (1 - this.tCDF(Math.abs(tStat), df));
-        // 95% confidence interval
-        const tCritical = this.tInverse(0.025, df);
-        const marginOfError = tCritical * pooledSE;
-        const confidenceInterval = [
-            uplift - marginOfError,
-            uplift + marginOfError
-        ];
-        // Update experiment with results
-        this.db.prepare(`
-      UPDATE causal_experiments
-      SET treatment_mean = ?,
-          control_mean = ?,
-          uplift = ?,
-          p_value = ?,
-          confidence_interval_low = ?,
-          confidence_interval_high = ?,
-          status = 'completed'
-      WHERE id = ?
-    `).run(treatmentMean, controlMean, uplift, pValue, confidenceInterval[0], confidenceInterval[1], experimentId);
-        return { uplift, pValue, confidenceInterval };
-    }
-    /**
-     * Query causal effects
-     */
-    queryCausalEffects(query) {
-        const { interventionMemoryId, interventionMemoryType, outcomeMemoryId, minConfidence = 0.5, minUplift = 0.0 } = query;
-        let sql = `
-      SELECT * FROM causal_edges
-      WHERE from_memory_id = ?
-        AND from_memory_type = ?
-        AND confidence >= ?
-        AND ABS(uplift) >= ?
-    `;
-        const params = [
-            interventionMemoryId,
-            interventionMemoryType,
-            minConfidence,
-            minUplift
-        ];
-        if (outcomeMemoryId) {
-            sql += ' AND to_memory_id = ?';
-            params.push(outcomeMemoryId);
-        }
-        sql += ' ORDER BY ABS(uplift) * confidence DESC';
-        const rows = this.db.prepare(sql).all(...params);
-        return rows.map(row => this.rowToCausalEdge(row));
-    }
-    /**
-     * Get causal chain (multi-hop reasoning)
-     *
-     * v2: Uses HyperbolicAttention if enabled for tree-structured retrieval
-     * v1: Falls back to recursive CTE with standard scoring
-     *
-     * @param fromMemoryId - Starting memory node
-     * @param toMemoryId - Target memory node
-     * @param maxDepth - Maximum chain depth (default: 5)
-     * @returns Ranked causal chains with paths, uplift, and confidence
-     */
-    async getCausalChain(fromMemoryId, toMemoryId, maxDepth = 5) {
-        // v2: Use HyperbolicAttention if enabled
-        if (this.attentionService && this.embedder) {
-            return this.getCausalChainWithAttention(fromMemoryId, toMemoryId, maxDepth);
-        }
-        // v1: Legacy recursive CTE
-        const chains = this.db.prepare(`
-      WITH RECURSIVE chain(from_id, to_id, depth, path, total_uplift, min_confidence) AS (
-        SELECT
-          from_memory_id,
-          to_memory_id,
-          1,
-          from_memory_id || '->' || to_memory_id,
-          uplift,
-          confidence
-        FROM causal_edges
-        WHERE from_memory_id = ? AND confidence >= 0.5
-
-        UNION ALL
-
-        SELECT
-          chain.from_id,
-          ce.to_memory_id,
-          chain.depth + 1,
-          chain.path || '->' || ce.to_memory_id,
-          chain.total_uplift + ce.uplift,
-          MIN(chain.min_confidence, ce.confidence)
-        FROM chain
-        JOIN causal_edges ce ON chain.to_id = ce.from_memory_id
-        WHERE chain.depth < ?
-          AND ce.confidence >= 0.5
-          AND chain.path NOT LIKE '%' || ce.to_memory_id || '%'
-      )
-      SELECT path, total_uplift, min_confidence
-      FROM chain
-      WHERE to_id = ?
-      ORDER BY total_uplift DESC
-      LIMIT 10
-    `).all(fromMemoryId, maxDepth, toMemoryId);
-        return chains.map(row => ({
-            path: row.path.split('->').map(Number),
-            totalUplift: row.total_uplift,
-            confidence: row.min_confidence
-        }));
-    }
-    /**
-     * Hash a string to a positive integer
-     * Used for converting string IDs to numeric IDs for backward compatibility
-     */
-    hashString(str) {
-        let hash = 0;
-        for (let i = 0; i < str.length; i++) {
-            const char = str.charCodeAt(i);
-            hash = ((hash << 5) - hash) + char;
-            hash = hash & hash; // Convert to 32bit integer
-        }
-        return Math.abs(hash);
-    }
-    /**
-     * Get causal chain with HyperbolicAttention (v2 feature)
-     *
-     * Uses Poincaré embeddings to model hierarchical causal relationships.
-     * Retrieves chains based on hyperbolic distance in embedding space.
-     *
-     * @private
-     */
-    async getCausalChainWithAttention(fromMemoryId, toMemoryId, maxDepth) {
-        // Get all candidate chains using CTE
-        const candidateChains = this.db.prepare(`
-      WITH RECURSIVE chain(from_id, to_id, depth, path, total_uplift, min_confidence) AS (
-        SELECT
-          from_memory_id,
-          to_memory_id,
-          1,
-          from_memory_id || '->' || to_memory_id,
-          uplift,
-          confidence
-        FROM causal_edges
-        WHERE from_memory_id = ? AND confidence >= 0.5
-
-        UNION ALL
-
-        SELECT
-          chain.from_id,
-          ce.to_memory_id,
-          chain.depth + 1,
-          chain.path || '->' || ce.to_memory_id,
-          chain.total_uplift + ce.uplift,
-          MIN(chain.min_confidence, ce.confidence)
-        FROM chain
-        JOIN causal_edges ce ON chain.to_id = ce.from_memory_id
-        WHERE chain.depth < ?
-          AND ce.confidence >= 0.5
-          AND chain.path NOT LIKE '%' || ce.to_memory_id || '%'
-      )
-      SELECT path, total_uplift, min_confidence
-      FROM chain
-      WHERE to_id = ?
-      LIMIT 50
-    `).all(fromMemoryId, maxDepth, toMemoryId);
-        if (candidateChains.length === 0) {
-            return [];
-        }
-        // Get embeddings for query (from node)
-        const fromEpisode = this.db.prepare('SELECT task, output FROM episodes WHERE id = ?').get(fromMemoryId);
-        const queryText = fromEpisode ? `${fromEpisode.task}: ${fromEpisode.output}` : '';
-        const queryEmbedding = await this.embedder.embed(queryText);
-        // Get embeddings and hierarchy levels for all chain nodes
-        const allNodeIds = new Set();
-        candidateChains.forEach((chain) => {
-            const path = chain.path.split('->').map(Number);
-            path.forEach(id => allNodeIds.add(id));
-        });
-        const nodeEmbeddings = new Map();
-        const hierarchyLevels = new Map();
-        for (const nodeId of allNodeIds) {
-            const episode = this.db.prepare('SELECT task, output FROM episodes WHERE id = ?').get(nodeId);
-            if (episode) {
-                const text = `${episode.task}: ${episode.output}`;
-                const embedding = await this.embedder.embed(text);
-                nodeEmbeddings.set(nodeId, embedding);
-                // Calculate hierarchy level (depth from root)
-                const level = candidateChains
-                    .filter((chain) => chain.path.includes(String(nodeId)))
-                    .reduce((minDepth, chain) => {
-                    const path = chain.path.split('->').map(Number);
-                    const idx = path.indexOf(nodeId);
-                    return Math.min(minDepth, idx);
-                }, maxDepth);
-                hierarchyLevels.set(nodeId, level);
-            }
-        }
-        // Prepare keys, values, and hierarchy for attention
-        const nodeList = Array.from(allNodeIds);
-        const keys = new Float32Array(nodeList.length * 384);
-        const values = new Float32Array(nodeList.length * 384);
-        const hierarchyArray = [];
-        nodeList.forEach((nodeId, idx) => {
-            const embedding = nodeEmbeddings.get(nodeId);
-            keys.set(embedding, idx * 384);
-            values.set(embedding, idx * 384);
-            hierarchyArray.push(hierarchyLevels.get(nodeId) || 0);
-        });
-        // Apply HyperbolicAttention
-        const queries = new Float32Array(384);
-        queries.set(queryEmbedding);
-        const attentionResult = await this.attentionService.hyperbolicAttention(queries, keys, values, hierarchyArray);
-        // Re-rank chains by attention weights
-        const rankedChains = candidateChains
-            .map((chain) => {
-            const path = chain.path.split('->').map(Number);
-            // Calculate average attention weight for nodes in path
-            const avgWeight = path.reduce((sum, nodeId) => {
-                const idx = nodeList.indexOf(nodeId);
-                return sum + (idx >= 0 ? attentionResult.weights[idx] : 0);
-            }, 0) / path.length;
-            return {
-                path,
-                totalUplift: chain.total_uplift,
-                confidence: chain.min_confidence * avgWeight, // Boost confidence by attention
-                attentionMetrics: {
-                    hyperbolicDistance: attentionResult.distances,
-                    computeTimeMs: attentionResult.metrics.computeTimeMs,
-                },
-            };
-        })
-            .sort((a, b) => b.confidence - a.confidence)
-            .slice(0, 10);
-        return rankedChains;
-    }
-    /**
-     * Calculate causal gain: E[outcome|do(treatment)] - E[outcome]
-     */
-    calculateCausalGain(treatmentId, outcomeType) {
-        // Get episodes where treatment was applied
-        const withTreatment = this.db.prepare(`
-      SELECT AVG(CASE WHEN ? = 'reward' THEN reward
-                     WHEN ? = 'success' THEN success
-                     WHEN ? = 'latency' THEN latency_ms
-                END) as avg_outcome
-      FROM episodes
-      WHERE id IN (
-        SELECT to_memory_id FROM causal_edges
-        WHERE from_memory_id = ? AND confidence >= 0.6
-      )
-    `).get(outcomeType, outcomeType, outcomeType, treatmentId);
-        // Get baseline (no treatment)
-        const baseline = this.db.prepare(`
-      SELECT AVG(CASE WHEN ? = 'reward' THEN reward
-                     WHEN ? = 'success' THEN success
-                     WHEN ? = 'latency' THEN latency_ms
-                END) as avg_outcome
-      FROM episodes
-      WHERE id NOT IN (
-        SELECT to_memory_id FROM causal_edges
-        WHERE from_memory_id = ?
-      )
-    `).get(outcomeType, outcomeType, outcomeType, treatmentId);
-        const causalGain = (withTreatment?.avg_outcome || 0) - (baseline?.avg_outcome || 0);
-        // Get most confident edge for mechanism
-        const edge = this.db.prepare(`
-      SELECT mechanism, confidence
-      FROM causal_edges
-      WHERE from_memory_id = ?
-      ORDER BY confidence DESC
-      LIMIT 1
-    `).get(treatmentId);
-        return {
-            causalGain,
-            confidence: edge?.confidence || 0,
-            mechanism: edge?.mechanism || 'unknown'
-        };
-    }
-    /**
-     * Detect confounders using correlation analysis
-     */
-    detectConfounders(edgeId) {
-        const edge = this.db.prepare('SELECT * FROM causal_edges WHERE id = ?').get(edgeId);
-        if (!edge) {
-            return { confounders: [] };
-        }
-        // Find memories correlated with both treatment and outcome
-        // This is a simplified version - production would use proper statistical tests
-        const potentialConfounders = this.db.prepare(`
-      SELECT DISTINCT e.id, e.task
-      FROM episodes e
-      WHERE e.id != ? AND e.id != ?
-        AND e.session_id IN (
-          SELECT session_id FROM episodes WHERE id = ?
-          UNION
-          SELECT session_id FROM episodes WHERE id = ?
-        )
-    `).all(edge.from_memory_id, edge.to_memory_id, edge.from_memory_id, edge.to_memory_id);
-        const confounders = potentialConfounders.map((conf) => {
-            // Calculate correlation scores (simplified)
-            const treatmentCorr = this.calculateCorrelation(conf.id, edge.from_memory_id);
-            const outcomeCorr = this.calculateCorrelation(conf.id, edge.to_memory_id);
-            const confounderScore = Math.sqrt(treatmentCorr ** 2 * outcomeCorr ** 2);
-            return {
-                memoryId: conf.id,
-                correlationWithTreatment: treatmentCorr,
-                correlationWithOutcome: outcomeCorr,
-                confounderScore
-            };
-        }).filter(c => c.confounderScore > 0.3);
-        // Update edge with confounder score
-        if (confounders.length > 0) {
-            const maxConfounderScore = Math.max(...confounders.map(c => c.confounderScore));
-            this.db.prepare(`
-        UPDATE causal_edges
-        SET confounder_score = ?
-        WHERE id = ?
-      `).run(maxConfounderScore, edgeId);
-        }
-        return { confounders };
-    }
-    // ========================================================================
-    // Private Helper Methods
-    // ========================================================================
-    rowToCausalEdge(row) {
-        return {
-            id: row.id,
-            fromMemoryId: row.from_memory_id,
-            fromMemoryType: row.from_memory_type,
-            toMemoryId: row.to_memory_id,
-            toMemoryType: row.to_memory_type,
-            similarity: row.similarity,
-            uplift: row.uplift,
-            confidence: row.confidence,
-            sampleSize: row.sample_size,
-            evidenceIds: row.evidence_ids ? JSON.parse(row.evidence_ids) : undefined,
-            experimentIds: row.experiment_ids ? JSON.parse(row.experiment_ids) : undefined,
-            confounderScore: row.confounder_score,
-            mechanism: row.mechanism,
-            metadata: row.metadata ? JSON.parse(row.metadata) : undefined
-        };
-    }
-    mean(values) {
-        return values.reduce((a, b) => a + b, 0) / values.length;
-    }
-    variance(values) {
-        const avg = this.mean(values);
-        return values.reduce((sum, val) => sum + (val - avg) ** 2, 0) / values.length;
-    }
-    standardError(values) {
-        return Math.sqrt(this.variance(values) / values.length);
-    }
-    tCDF(t, df) {
-        // Simplified t-distribution CDF (use proper stats library in production)
-        // This is an approximation
-        return 0.5 + 0.5 * Math.sign(t) * (1 - Math.pow(1 + t * t / df, -df / 2));
-    }
-    tInverse(p, df) {
-        // Simplified inverse t-distribution (use proper stats library)
-        // Approximation for 95% CI
-        return 1.96; // Standard normal approximation
-    }
-    calculateCorrelation(id1, id2) {
-        // Simplified correlation calculation
-        // In production, use proper correlation metrics
-        const sharedSessions = this.db.prepare(`
-      SELECT COUNT(DISTINCT e1.session_id) as shared
-      FROM episodes e1
-      JOIN episodes e2 ON e1.session_id = e2.session_id
-      WHERE e1.id = ? AND e2.id = ?
-    `).get(id1, id2);
-        return Math.min(sharedSessions?.shared || 0, 1.0);
-    }
-}
-//# sourceMappingURL=CausalMemoryGraph.js.map

package/dist/packages/agentdb/src/controllers/CausalMemoryGraph.js.map

@@ -1 +0,0 @@
-{"version":3,"file":"CausalMemoryGraph.js","sourceRoot":"","sources":["../../../../../../packages/agentdb/src/controllers/CausalMemoryGraph.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;GAgBG;AAGH,OAAO,EAAE,YAAY,EAAE,MAAM,0BAA0B,CAAC;AACxD,OAAO,EAAE,gBAAgB,EAAkC,MAAM,oCAAoC,CAAC;AAiFtG,MAAM,OAAO,iBAAiB;IACpB,EAAE,CAAW;IACb,YAAY,CAAO,CAAC,uCAAuC;IAC3D,gBAAgB,CAAoB;IACpC,QAAQ,CAAoB;IAC5B,MAAM,CAA0B;IAExC;;;;;OAKG;IACH,YACE,EAAY,EACZ,YAAkB,EAClB,QAA2B,EAC3B,MAAgC;QAEhC,IAAI,CAAC,EAAE,GAAG,EAAE,CAAC;QACb,IAAI,CAAC,YAAY,GAAG,YAAY,CAAC;QACjC,IAAI,CAAC,QAAQ,GAAG,QAAQ,CAAC;QACzB,IAAI,CAAC,MAAM,GAAG;YACZ,2BAA2B,EAAE,KAAK;YAClC,GAAG,MAAM;SACV,CAAC;QAEF,mDAAmD;QACnD,IAAI,QAAQ,IAAI,IAAI,CAAC,MAAM,CAAC,2BAA2B,EAAE,CAAC;YACxD,IAAI,CAAC,gBAAgB,GAAG,IAAI,gBAAgB,CAAC,EAAE,EAAE;gBAC/C,UAAU,EAAE;oBACV,OAAO,EAAE,IAAI;oBACb,GAAG,IAAI,CAAC,MAAM,CAAC,gBAAgB;iBAChC;aACF,CAAC,CAAC;QACL,CAAC;IACH,CAAC;IAED;;OAEG;IACH,KAAK,CAAC,aAAa,CAAC,IAAgB;QAClC,qDAAqD;QACrD,IAAI,IAAI,CAAC,YAAY,IAAI,kBAAkB,IAAI,IAAI,CAAC,YAAY,EAAE,CAAC;YACjE,MAAM,YAAY,GAAG,IAAI,CAAC,YAA2C,CAAC;YAEtE,wCAAwC;YACxC,MAAM,aAAa,GAAG,IAAI,CAAC,SAAS,IAAI,GAAG,IAAI,CAAC,cAAc,IAAI,IAAI,CAAC,YAAY,cAAc,CAAC;YAClG,MAAM,SAAS,GAAG,IAAI,YAAY,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,iDAAiD;YAElG,kEAAkE;YAClE,yDAAyD;YACzD,MAAM,MAAM,GAAG,YAAY,CAAC,WAAW,EAAE,CAAC;YAE1C,MAAM,UAAU,GAAG,OAAO,IAAI,CAAC,YAAY,KAAK,QAAQ;gBACtD,CAAC,CAAC,IAAI,CAAC,YAAY;gBACnB,CAAC,CAAC,CAAC,MAAM,CAAC,SAAS,CAAC,IAAI,CAAC,YAAY,CAAC,IAAI,GAAG,IAAI,CAAC,cAAc,IAAI,IAAI,CAAC,YAAY,EAAE,CAAC,CAAC;YAE3F,MAAM,QAAQ,GAAG,OAAO,IAAI,CAAC,UAAU,KAAK,QAAQ;gBAClD,CAAC,CAAC,IAAI,CAAC,UAAU;gBACjB,CAAC,CAAC,CAAC,MAAM,CAAC,SAAS,CAAC,IAAI,CAAC,UAAU,CAAC,IAAI,GAAG,IAAI,CAAC,YAAY,IAAI,IAAI,CAAC,UAAU,EAAE,CAAC,CAAC;YAErF,MAAM,SAAS,GAAoB;gBACjC,IAAI,EAAE,UAAU;gBAChB,EAAE,EAAE,QAAQ;gBACZ,SAAS,EAAE,aAAa;gBACxB,MAAM,EAAE,IAAI,CAAC,MAAM,IAAI,CAAC;gBACxB,UAAU,EAAE,IAAI,CAAC,UAAU;gBAC3B,UAAU,EAAE,IAAI,CAAC,UAAU,IAAI,CAAC;aACjC,CAAC;YAEF,MAAM,MAAM,GAAG,MAAM,YAAY,CAAC,gBAAgB,CAAC,SAAS,EAAE,SAAS,CAAC,CAAC;YACzE,oDAAoD;YACpD,qEAAqE;YACrE,IAAI,OAAO,MAAM,KAAK,QAAQ,EAAE,CAAC;gBAC/B,OAAO,MAAM,CAAC;YAChB,CAAC;YACD,sDAAsD;YACtD,MAAM,QAAQ,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC;YAChD,OAAO,QAAQ,CAAC,CAAC,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,UAAU,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC;QAC1F,CAAC;QAED,qBAAqB;QACrB,MAAM,IAAI,GAAG,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC;;;;;;;KAO5B,CAAC,CAAC;QAEH,MAAM,MAAM,GAAG,IAAI,CAAC,GAAG,CACrB,IAAI,CAAC,YAAY,EACjB,IAAI,CAAC,cAAc,EACnB,IAAI,CAAC,UAAU,EACf,IAAI,CAAC,YAAY,EACjB,IAAI,CAAC,UAAU,EACf,IAAI,CAAC,MAAM,IAAI,IAAI,EACnB,IAAI,CAAC,UAAU,EACf,IAAI,CAAC,UAAU,IAAI,IAAI,EACvB,IAAI,CAAC,WAAW,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC,CAAC,CAAC,IAAI,EAC1D,IAAI,CAAC,eAAe,IAAI,IAAI,EAC5B,IAAI,CAAC,SAAS,IAAI,IAAI,EACtB,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,IAAI,CACrD,CAAC;QAEF,OAAO,MAAM,CAAC,eAAyB,CAAC;IAC1C,CAAC;IAED;;OAEG;IACH,gBAAgB,CAAC,UAA4B;QAC3C,MAAM,IAAI,GAAG,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC;;;;;KAK5B,CAAC,CAAC;QAEH,MAAM,MAAM,GAAG,IAAI,CAAC,GAAG,CACrB,UAAU,CAAC,IAAI,EACf,UAAU,CAAC,UAAU,EACrB,UAAU,CAAC,WAAW,EACtB,UAAU,CAAC,aAAa,EACxB,UAAU,CAAC,SAAS,IAAI,IAAI,EAC5B,UAAU,CAAC,SAAS,EACpB,UAAU,CAAC,UAAU,EACrB,UAAU,CAAC,MAAM,EACjB,UAAU,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,UAAU,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,IAAI,CACjE,CAAC;QAEF,OAAO,MAAM,CAAC,eAAyB,CAAC;IAC1C,CAAC;IAED;;OAEG;IACH,iBAAiB,CAAC,WAA8B;QAC9C,MAAM,IAAI,GAAG,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC;;;;KAI5B,CAAC,CAAC;QAEH,IAAI,CAAC,GAAG,CACN,WAAW,CAAC,YAAY,EACxB,WAAW,CAAC,SAAS,EACrB,WAAW,CAAC,WAAW,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAC/B,WAAW,CAAC,YAAY,EACxB,WAAW,CAAC,WAAW,EACvB,WAAW,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,WAAW,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,IAAI,CACjE,CAAC;QAEF,qBAAqB;QACrB,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC;;;;KAIf,CAAC,CAAC,GAAG,CAAC,WAAW,CAAC,YAAY,CAAC,CAAC;IACnC,CAAC;IAED;;OAEG;IACH,eAAe,CAAC,YAAoB;QAKlC,yCAAyC;QACzC,MAAM,YAAY,GAAG,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC;;;;KAIpC,CAAC,CAAC,GAAG,CAAC,YAAY,CAAU,CAAC;QAE9B,MAAM,eAAe,GAAG,YAAY;aACjC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,YAAY,KAAK,CAAC,CAAC;aACjC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC;QAE7B,MAAM,aAAa,GAAG,YAAY;aAC/B,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,YAAY,KAAK,CAAC,CAAC;aACjC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC;QAE7B,IAAI,eAAe,CAAC,MAAM,KAAK,CAAC,IAAI,aAAa,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YAC/D,OAAO,EAAE,MAAM,EAAE,CAAC,EAAE,MAAM,EAAE,GAAG,EAAE,kBAAkB,EAAE,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,CAAC;QAChE,CAAC;QAED,kBAAkB;QAClB,MAAM,aAAa,GAAG,IAAI,CAAC,IAAI,CAAC,eAAe,CAAC,CAAC;QACjD,MAAM,WAAW,GAAG,IAAI,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC;QAC7C,MAAM,MAAM,GAAG,aAAa,GAAG,WAAW,CAAC;QAE3C,4BAA4B;QAC5B,MAAM,WAAW,GAAG,IAAI,CAAC,aAAa,CAAC,eAAe,CAAC,CAAC;QACxD,MAAM,SAAS,GAAG,IAAI,CAAC,aAAa,CAAC,aAAa,CAAC,CAAC;QACpD,MAAM,QAAQ,GAAG,IAAI,CAAC,IAAI,CAAC,WAAW,IAAI,CAAC,GAAG,SAAS,IAAI,CAAC,CAAC,CAAC;QAE9D,uCAAuC;QACvC,MAAM,KAAK,GAAG,MAAM,GAAG,QAAQ,CAAC;QAChC,MAAM,EAAE,GAAG,eAAe,CAAC,MAAM,GAAG,aAAa,CAAC,MAAM,GAAG,CAAC,CAAC;QAC7D,MAAM,MAAM,GAAG,CAAC,GAAG,CAAC,CAAC,GAAG,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC;QAExD,0BAA0B;QAC1B,MAAM,SAAS,GAAG,IAAI,CAAC,QAAQ,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC;QAC3C,MAAM,aAAa,GAAG,SAAS,GAAG,QAAQ,CAAC;QAC3C,MAAM,kBAAkB,GAAqB;YAC3C,MAAM,GAAG,aAAa;YACtB,MAAM,GAAG,aAAa;SACvB,CAAC;QAEF,iCAAiC;QACjC,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC;;;;;;;;;;KAUf,CAAC,CAAC,GAAG,CACJ,aAAa,EACb,WAAW,EACX,MAAM,EACN,MAAM,EACN,kBAAkB,CAAC,CAAC,CAAC,EACrB,kBAAkB,CAAC,CAAC,CAAC,EACrB,YAAY,CACb,CAAC;QAEF,OAAO,EAAE,MAAM,EAAE,MAAM,EAAE,kBAAkB,EAAE,CAAC;IAChD,CAAC;IAED;;OAEG;IACH,kBAAkB,CAAC,KAAkB;QACnC,MAAM,EACJ,oBAAoB,EACpB,sBAAsB,EACtB,eAAe,EACf,aAAa,GAAG,GAAG,EACnB,SAAS,GAAG,GAAG,EAChB,GAAG,KAAK,CAAC;QAEV,IAAI,GAAG,GAAG;;;;;;KAMT,CAAC;QAEF,MAAM,MAAM,GAAU;YACpB,oBAAoB;YACpB,sBAAsB;YACtB,aAAa;YACb,SAAS;SACV,CAAC;QAEF,IAAI,eAAe,EAAE,CAAC;YACpB,GAAG,IAAI,uBAAuB,CAAC;YAC/B,MAAM,CAAC,IAAI,CAAC,eAAe,CAAC,CAAC;QAC/B,CAAC;QAED,GAAG,IAAI,yCAAyC,CAAC;QAEjD,MAAM,IAAI,GAAG,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,GAAG,MAAM,CAAU,CAAC;QAE1D,OAAO,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC,IAAI,CAAC,eAAe,CAAC,GAAG,CAAC,CAAC,CAAC;IACpD,CAAC;IAED;;;;;;;;;;OAUG;IACH,KAAK,CAAC,cAAc,CAAC,YAAoB,EAAE,UAAkB,EAAE,WAAmB,CAAC;QASjF,yCAAyC;QACzC,IAAI,IAAI,CAAC,gBAAgB,IAAI,IAAI,CAAC,QAAQ,EAAE,CAAC;YAC3C,OAAO,IAAI,CAAC,2BAA2B,CAAC,YAAY,EAAE,UAAU,EAAE,QAAQ,CAAC,CAAC;QAC9E,CAAC;QAED,2BAA2B;QAC3B,MAAM,MAAM,GAAG,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;KAgC9B,CAAC,CAAC,GAAG,CAAC,YAAY,EAAE,QAAQ,EAAE,UAAU,CAAU,CAAC;QAEpD,OAAO,MAAM,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;YACxB,IAAI,EAAE,GAAG,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,MAAM,CAAC;YACtC,WAAW,EAAE,GAAG,CAAC,YAAY;YAC7B,UAAU,EAAE,GAAG,CAAC,cAAc;SAC/B,CAAC,CAAC,CAAC;IACN,CAAC;IAED;;;OAGG;IACK,UAAU,CAAC,GAAW;QAC5B,IAAI,IAAI,GAAG,CAAC,CAAC;QACb,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,GAAG,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;YACpC,MAAM,IAAI,GAAG,GAAG,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC;YAC/B,IAAI,GAAG,CAAC,CAAC,IAAI,IAAI,CAAC,CAAC,GAAG,IAAI,CAAC,GAAG,IAAI,CAAC;YACnC,IAAI,GAAG,IAAI,GAAG,IAAI,CAAC,CAAC,2BAA2B;QACjD,CAAC;QACD,OAAO,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;IACxB,CAAC;IAED;;;;;;;OAOG;IACK,KAAK,CAAC,2BAA2B,CACvC,YAAoB,EACpB,UAAkB,EAClB,QAAgB;QAUhB,qCAAqC;QACrC,MAAM,eAAe,GAAG,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;KA+BvC,CAAC,CAAC,GAAG,CAAC,YAAY,EAAE,QAAQ,EAAE,UAAU,CAAU,CAAC;QAEpD,IAAI,eAAe,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YACjC,OAAO,EAAE,CAAC;QACZ,CAAC;QAED,uCAAuC;QACvC,MAAM,WAAW,GAAG,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC,gDAAgD,CAAC,CAAC,GAAG,CAAC,YAAY,CAAQ,CAAC;QAC/G,MAAM,SAAS,GAAG,WAAW,CAAC,CAAC,CAAC,GAAG,WAAW,CAAC,IAAI,KAAK,WAAW,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QAClF,MAAM,cAAc,GAAG,MAAM,IAAI,CAAC,QAAS,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC;QAE7D,0DAA0D;QAC1D,MAAM,UAAU,GAAG,IAAI,GAAG,EAAU,CAAC;QACrC,eAAe,CAAC,OAAO,CAAC,CAAC,KAAU,EAAE,EAAE;YACrC,MAAM,IAAI,GAAG,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC;YAChD,IAAI,CAAC,OAAO,CAAC,EAAE,CAAC,EAAE,CAAC,UAAU,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,CAAC;QACzC,CAAC,CAAC,CAAC;QAEH,MAAM,cAAc,GAAG,IAAI,GAAG,EAAwB,CAAC;QACvD,MAAM,eAAe,GAAG,IAAI,GAAG,EAAkB,CAAC;QAElD,KAAK,MAAM,MAAM,IAAI,UAAU,EAAE,CAAC;YAChC,MAAM,OAAO,GAAG,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC,gDAAgD,CAAC,CAAC,GAAG,CAAC,MAAM,CAAQ,CAAC;YACrG,IAAI,OAAO,EAAE,CAAC;gBACZ,MAAM,IAAI,GAAG,GAAG,OAAO,CAAC,IAAI,KAAK,OAAO,CAAC,MAAM,EAAE,CAAC;gBAClD,MAAM,SAAS,GAAG,MAAM,IAAI,CAAC,QAAS,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;gBACnD,cAAc,CAAC,GAAG,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC;gBAEtC,8CAA8C;gBAC9C,MAAM,KAAK,GAAG,eAAe;qBAC1B,MAAM,CAAC,CAAC,KAAU,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC,QAAQ,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC;qBAC3D,MAAM,CAAC,CAAC,QAAgB,EAAE,KAAU,EAAE,EAAE;oBACvC,MAAM,IAAI,GAAG,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC;oBAChD,MAAM,GAAG,GAAG,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;oBACjC,OAAO,IAAI,CAAC,GAAG,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC;gBACjC,CAAC,EAAE,QAAQ,CAAC,CAAC;gBAEf,eAAe,CAAC,GAAG,CAAC,MAAM,EAAE,KAAK,CAAC,CAAC;YACrC,CAAC;QACH,CAAC;QAED,oDAAoD;QACpD,MAAM,QAAQ,GAAG,KAAK,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC;QACxC,MAAM,IAAI,GAAG,IAAI,YAAY,CAAC,QAAQ,CAAC,MAAM,GAAG,GAAG,CAAC,CAAC;QACrD,MAAM,MAAM,GAAG,IAAI,YAAY,CAAC,QAAQ,CAAC,MAAM,GAAG,GAAG,CAAC,CAAC;QACvD,MAAM,cAAc,GAAa,EAAE,CAAC;QAEpC,QAAQ,CAAC,OAAO,CAAC,CAAC,MAAM,EAAE,GAAG,EAAE,EAAE;YAC/B,MAAM,SAAS,GAAG,cAAc,CAAC,GAAG,CAAC,MAAM,CAAE,CAAC;YAC9C,IAAI,CAAC,GAAG,CAAC,SAAS,EAAE,GAAG,GAAG,GAAG,CAAC,CAAC;YAC/B,MAAM,CAAC,GAAG,CAAC,SAAS,EAAE,GAAG,GAAG,GAAG,CAAC,CAAC;YACjC,cAAc,CAAC,IAAI,CAAC,eAAe,CAAC,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC;QACxD,CAAC,CAAC,CAAC;QAEH,4BAA4B;QAC5B,MAAM,OAAO,GAAG,IAAI,YAAY,CAAC,GAAG,CAAC,CAAC;QACtC,OAAO,CAAC,GAAG,CAAC,cAAc,CAAC,CAAC;QAE5B,MAAM,eAAe,GAAG,MAAM,IAAI,CAAC,gBAAiB,CAAC,mBAAmB,CACtE,OAAO,EACP,IAAI,EACJ,MAAM,EACN,cAAc,CACf,CAAC;QAEF,sCAAsC;QACtC,MAAM,YAAY,GAAG,eAAe;aACjC,GAAG,CAAC,CAAC,KAAU,EAAE,EAAE;YAClB,MAAM,IAAI,GAAG,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC;YAEhD,uDAAuD;YACvD,MAAM,SAAS,GAAG,IAAI,CAAC,MAAM,CAAC,CAAC,GAAW,EAAE,MAAc,EAAE,EAAE;gBAC5D,MAAM,GAAG,GAAG,QAAQ,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;gBACrC,OAAO,GAAG,GAAG,CAAC,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC,eAAe,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;YAC7D,CAAC,EAAE,CAAC,CAAC,GAAG,IAAI,CAAC,MAAM,CAAC;YAEpB,OAAO;gBACL,IAAI;gBACJ,WAAW,EAAE,KAAK,CAAC,YAAY;gBAC/B,UAAU,EAAE,KAAK,CAAC,cAAc,GAAG,SAAS,EAAE,gCAAgC;gBAC9E,gBAAgB,EAAE;oBAChB,kBAAkB,EAAE,eAAe,CAAC,SAAS;oBAC7C,aAAa,EAAE,eAAe,CAAC,OAAO,CAAC,aAAa;iBACrD;aACF,CAAC;QACJ,CAAC,CAAC;aACD,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,UAAU,GAAG,CAAC,CAAC,UAAU,CAAC;aAC3C,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;QAEhB,OAAO,YAAY,CAAC;IACtB,CAAC;IAED;;OAEG;IACH,mBAAmB,CAAC,WAAmB,EAAE,WAA6C;QAKpF,2CAA2C;QAC3C,MAAM,aAAa,GAAG,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC;;;;;;;;;;KAUrC,CAAC,CAAC,GAAG,CAAC,WAAW,EAAE,WAAW,EAAE,WAAW,EAAE,WAAW,CAAQ,CAAC;QAElE,8BAA8B;QAC9B,MAAM,QAAQ,GAAG,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC;;;;;;;;;;KAUhC,CAAC,CAAC,GAAG,CAAC,WAAW,EAAE,WAAW,EAAE,WAAW,EAAE,WAAW,CAAQ,CAAC;QAElE,MAAM,UAAU,GAAG,CAAC,aAAa,EAAE,WAAW,IAAI,CAAC,CAAC,GAAG,CAAC,QAAQ,EAAE,WAAW,IAAI,CAAC,CAAC,CAAC;QAEpF,wCAAwC;QACxC,MAAM,IAAI,GAAG,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC;;;;;;KAM5B,CAAC,CAAC,GAAG,CAAC,WAAW,CAAQ,CAAC;QAE3B,OAAO;YACL,UAAU;YACV,UAAU,EAAE,IAAI,EAAE,UAAU,IAAI,CAAC;YACjC,SAAS,EAAE,IAAI,EAAE,SAAS,IAAI,SAAS;SACxC,CAAC;IACJ,CAAC;IAED;;OAEG;IACH,iBAAiB,CAAC,MAAc;QAQ9B,MAAM,IAAI,GAAG,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC,yCAAyC,CAAC,CAAC,GAAG,CAAC,MAAM,CAAQ,CAAC;QAE3F,IAAI,CAAC,IAAI,EAAE,CAAC;YACV,OAAO,EAAE,WAAW,EAAE,EAAE,EAAE,CAAC;QAC7B,CAAC;QAED,2DAA2D;QAC3D,+EAA+E;QAC/E,MAAM,oBAAoB,GAAG,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC;;;;;;;;;KAS5C,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,cAAc,EAAE,IAAI,CAAC,YAAY,EAAE,IAAI,CAAC,cAAc,EAAE,IAAI,CAAC,YAAY,CAAU,CAAC;QAEhG,MAAM,WAAW,GAAG,oBAAoB,CAAC,GAAG,CAAC,CAAC,IAAS,EAAE,EAAE;YACzD,4CAA4C;YAC5C,MAAM,aAAa,GAAG,IAAI,CAAC,oBAAoB,CAAC,IAAI,CAAC,EAAE,EAAE,IAAI,CAAC,cAAc,CAAC,CAAC;YAC9E,MAAM,WAAW,GAAG,IAAI,CAAC,oBAAoB,CAAC,IAAI,CAAC,EAAE,EAAE,IAAI,CAAC,YAAY,CAAC,CAAC;YAC1E,MAAM,eAAe,GAAG,IAAI,CAAC,IAAI,CAAC,aAAa,IAAI,CAAC,GAAG,WAAW,IAAI,CAAC,CAAC,CAAC;YAEzE,OAAO;gBACL,QAAQ,EAAE,IAAI,CAAC,EAAE;gBACjB,wBAAwB,EAAE,aAAa;gBACvC,sBAAsB,EAAE,WAAW;gBACnC,eAAe;aAChB,CAAC;QACJ,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,eAAe,GAAG,GAAG,CAAC,CAAC;QAExC,oCAAoC;QACpC,IAAI,WAAW,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAC3B,MAAM,kBAAkB,GAAG,IAAI,CAAC,GAAG,CAAC,GAAG,WAAW,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,eAAe,CAAC,CAAC,CAAC;YAChF,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC;;;;OAIf,CAAC,CAAC,GAAG,CAAC,kBAAkB,EAAE,MAAM,CAAC,CAAC;QACrC,CAAC;QAED,OAAO,EAAE,WAAW,EAAE,CAAC;IACzB,CAAC;IAED,2EAA2E;IAC3E,yBAAyB;IACzB,2EAA2E;IAEnE,eAAe,CAAC,GAAQ;QAC9B,OAAO;YACL,EAAE,EAAE,GAAG,CAAC,EAAE;YACV,YAAY,EAAE,GAAG,CAAC,cAAc;YAChC,cAAc,EAAE,GAAG,CAAC,gBAAgB;YACpC,UAAU,EAAE,GAAG,CAAC,YAAY;YAC5B,YAAY,EAAE,GAAG,CAAC,cAAc;YAChC,UAAU,EAAE,GAAG,CAAC,UAAU;YAC1B,MAAM,EAAE,GAAG,CAAC,MAAM;YAClB,UAAU,EAAE,GAAG,CAAC,UAAU;YAC1B,UAAU,EAAE,GAAG,CAAC,WAAW;YAC3B,WAAW,EAAE,GAAG,CAAC,YAAY,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC,SAAS;YACxE,aAAa,EAAE,GAAG,CAAC,cAAc,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,cAAc,CAAC,CAAC,CAAC,CAAC,SAAS;YAC9E,eAAe,EAAE,GAAG,CAAC,gBAAgB;YACrC,SAAS,EAAE,GAAG,CAAC,SAAS;YACxB,QAAQ,EAAE,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,SAAS;SAC9D,CAAC;IACJ,CAAC;IAEO,IAAI,CAAC,MAAgB;QAC3B,OAAO,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC;IAC3D,CAAC;IAEO,QAAQ,CAAC,MAAgB;QAC/B,MAAM,GAAG,GAAG,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QAC9B,OAAO,MAAM,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,GAAG,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,GAAG,GAAG,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC;IAChF,CAAC;IAEO,aAAa,CAAC,MAAgB;QACpC,OAAO,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,MAAM,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC;IAC1D,CAAC;IAEO,IAAI,CAAC,CAAS,EAAE,EAAU;QAChC,yEAAyE;QACzE,2BAA2B;QAC3B,OAAO,GAAG,GAAG,GAAG,GAAG,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC,GAAG,EAAE,EAAE,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC,CAAC;IAC5E,CAAC;IAEO,QAAQ,CAAC,CAAS,EAAE,EAAU;QACpC,+DAA+D;QAC/D,2BAA2B;QAC3B,OAAO,IAAI,CAAC,CAAC,gCAAgC;IAC/C,CAAC;IAEO,oBAAoB,CAAC,GAAW,EAAE,GAAW;QACnD,qCAAqC;QACrC,gDAAgD;QAChD,MAAM,cAAc,GAAG,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC;;;;;KAKtC,CAAC,CAAC,GAAG,CAAC,GAAG,EAAE,GAAG,CAAQ,CAAC;QAExB,OAAO,IAAI,CAAC,GAAG,CAAC,cAAc,EAAE,MAAM,IAAI,CAAC,EAAE,GAAG,CAAC,CAAC;IACpD,CAAC;CACF","sourcesContent":["/**\n * CausalMemoryGraph - Causal Reasoning over Agent Memories\n *\n * Implements intervention-based reasoning rather than correlation.\n * Stores p(y|do(x)) estimates and tracks causal uplift across episodes.\n *\n * Based on:\n * - Pearl's do-calculus and causal inference\n * - Uplift modeling from A/B testing\n * - Instrumental variable methods\n *\n * v2.0.0-alpha.3 Features:\n * - HyperbolicAttention for tree-structured causal chain retrieval\n * - Poincaré embeddings for hierarchical relationships\n * - Feature flag: ENABLE_HYPERBOLIC_ATTENTION (default: false)\n * - 100% backward compatible with fallback to standard retrieval\n */\n\nimport type { GraphDatabaseAdapter, CausalEdge as GraphCausalEdge } from '../backends/graph/GraphDatabaseAdapter.js';\nimport { NodeIdMapper } from '../utils/NodeIdMapper.js';\nimport { AttentionService, type HyperbolicAttentionConfig } from '../utils/LegacyAttentionAdapter.js';\nimport { EmbeddingService } from './EmbeddingService.js';\n\n// Database type from db-fallback\ntype Database = any;\n\n/**\n * Configuration for CausalMemoryGraph\n */\nexport interface CausalMemoryGraphConfig {\n  /** Enable hyperbolic attention for causal chains (default: false) */\n  ENABLE_HYPERBOLIC_ATTENTION?: boolean;\n  /** Hyperbolic attention configuration */\n  hyperbolicConfig?: Partial<HyperbolicAttentionConfig>;\n}\n\nexport interface CausalEdge {\n  id?: number;\n  fromMemoryId: number;\n  fromMemoryType: 'episode' | 'skill' | 'note' | 'fact';\n  toMemoryId: number;\n  toMemoryType: 'episode' | 'skill' | 'note' | 'fact';\n\n  // Metrics\n  similarity: number;\n  uplift?: number; // E[y|do(x)] - E[y]\n  confidence: number;\n  sampleSize?: number;\n\n  // Evidence\n  evidenceIds?: string[];\n  experimentIds?: string[];\n  confounderScore?: number;\n\n  // Explanation\n  mechanism?: string;\n  metadata?: Record<string, any>;\n}\n\nexport interface CausalExperiment {\n  id?: number;\n  name: string;\n  hypothesis: string;\n  treatmentId: number;\n  treatmentType: string;\n  controlId?: number;\n\n  // Design\n  startTime: number;\n  endTime?: number;\n  sampleSize: number;\n\n  // Results\n  treatmentMean?: number;\n  controlMean?: number;\n  uplift?: number;\n  pValue?: number;\n  confidenceIntervalLow?: number;\n  confidenceIntervalHigh?: number;\n\n  status: 'running' | 'completed' | 'failed';\n  metadata?: Record<string, any>;\n}\n\nexport interface CausalObservation {\n  experimentId: number;\n  episodeId: number;\n  isTreatment: boolean;\n  outcomeValue: number;\n  outcomeType: 'reward' | 'success' | 'latency';\n  context?: Record<string, any>;\n}\n\nexport interface CausalQuery {\n  interventionMemoryId: number;\n  interventionMemoryType: string;\n  outcomeMemoryId?: number;\n  minConfidence?: number;\n  minUplift?: number;\n}\n\nexport class CausalMemoryGraph {\n  private db: Database;\n  private graphBackend?: any; // GraphBackend or GraphDatabaseAdapter\n  private attentionService?: AttentionService;\n  private embedder?: EmbeddingService;\n  private config: CausalMemoryGraphConfig;\n\n  /**\n   * Constructor supports both v1 (legacy) and v2 (with attention) modes\n   *\n   * v1 mode: new CausalMemoryGraph(db)\n   * v2 mode: new CausalMemoryGraph(db, graphBackend, embedder, config)\n   */\n  constructor(\n    db: Database,\n    graphBackend?: any,\n    embedder?: EmbeddingService,\n    config?: CausalMemoryGraphConfig\n  ) {\n    this.db = db;\n    this.graphBackend = graphBackend;\n    this.embedder = embedder;\n    this.config = {\n      ENABLE_HYPERBOLIC_ATTENTION: false,\n      ...config,\n    };\n\n    // Initialize AttentionService if embedder provided\n    if (embedder && this.config.ENABLE_HYPERBOLIC_ATTENTION) {\n      this.attentionService = new AttentionService(db, {\n        hyperbolic: {\n          enabled: true,\n          ...this.config.hyperbolicConfig,\n        },\n      });\n    }\n  }\n\n  /**\n   * Add a causal edge between memories\n   */\n  async addCausalEdge(edge: CausalEdge): Promise<number> {\n    // Use GraphDatabaseAdapter if available (AgentDB v2)\n    if (this.graphBackend && 'createCausalEdge' in this.graphBackend) {\n      const graphAdapter = this.graphBackend as any as GraphDatabaseAdapter;\n\n      // Create embedding for causal mechanism\n      const mechanismText = edge.mechanism || `${edge.fromMemoryType}-${edge.toMemoryType} causal link`;\n      const embedding = new Float32Array(384).fill(0); // Placeholder - would use embedder in production\n\n      // Convert episode IDs to string format expected by graph database\n      // Use NodeIdMapper to get full node IDs from numeric IDs\n      const mapper = NodeIdMapper.getInstance();\n\n      const fromNodeId = typeof edge.fromMemoryId === 'string'\n        ? edge.fromMemoryId\n        : (mapper.getNodeId(edge.fromMemoryId) || `${edge.fromMemoryType}-${edge.fromMemoryId}`);\n\n      const toNodeId = typeof edge.toMemoryId === 'string'\n        ? edge.toMemoryId\n        : (mapper.getNodeId(edge.toMemoryId) || `${edge.toMemoryType}-${edge.toMemoryId}`);\n\n      const graphEdge: GraphCausalEdge = {\n        from: fromNodeId,\n        to: toNodeId,\n        mechanism: mechanismText,\n        uplift: edge.uplift || 0,\n        confidence: edge.confidence,\n        sampleSize: edge.sampleSize || 0\n      };\n\n      const edgeId = await graphAdapter.createCausalEdge(graphEdge, embedding);\n      // Convert string ID to numeric ID for compatibility\n      // Extract numeric ID from string format \"type-number\" or return hash\n      if (typeof edgeId === 'number') {\n        return edgeId;\n      }\n      // Parse numeric ID from string format like \"edge-123\"\n      const numMatch = String(edgeId).match(/(\\d+)$/);\n      return numMatch ? parseInt(numMatch[1], 10) : Math.abs(this.hashString(String(edgeId)));\n    }\n\n    // Fallback to SQLite\n    const stmt = this.db.prepare(`\n      INSERT INTO causal_edges (\n        from_memory_id, from_memory_type, to_memory_id, to_memory_type,\n        similarity, uplift, confidence, sample_size,\n        evidence_ids, confounder_score,\n        mechanism, metadata\n      ) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)\n    `);\n\n    const result = stmt.run(\n      edge.fromMemoryId,\n      edge.fromMemoryType,\n      edge.toMemoryId,\n      edge.toMemoryType,\n      edge.similarity,\n      edge.uplift || null,\n      edge.confidence,\n      edge.sampleSize || null,\n      edge.evidenceIds ? JSON.stringify(edge.evidenceIds) : null,\n      edge.confounderScore || null,\n      edge.mechanism || null,\n      edge.metadata ? JSON.stringify(edge.metadata) : null\n    );\n\n    return result.lastInsertRowid as number;\n  }\n\n  /**\n   * Create a causal experiment (A/B test)\n   */\n  createExperiment(experiment: CausalExperiment): number {\n    const stmt = this.db.prepare(`\n      INSERT INTO causal_experiments (\n        name, hypothesis, treatment_id, treatment_type, control_id,\n        start_time, sample_size, status, metadata\n      ) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?)\n    `);\n\n    const result = stmt.run(\n      experiment.name,\n      experiment.hypothesis,\n      experiment.treatmentId,\n      experiment.treatmentType,\n      experiment.controlId || null,\n      experiment.startTime,\n      experiment.sampleSize,\n      experiment.status,\n      experiment.metadata ? JSON.stringify(experiment.metadata) : null\n    );\n\n    return result.lastInsertRowid as number;\n  }\n\n  /**\n   * Record an observation in an experiment\n   */\n  recordObservation(observation: CausalObservation): void {\n    const stmt = this.db.prepare(`\n      INSERT INTO causal_observations (\n        experiment_id, episode_id, is_treatment, outcome_value, outcome_type, context\n      ) VALUES (?, ?, ?, ?, ?, ?)\n    `);\n\n    stmt.run(\n      observation.experimentId,\n      observation.episodeId,\n      observation.isTreatment ? 1 : 0,\n      observation.outcomeValue,\n      observation.outcomeType,\n      observation.context ? JSON.stringify(observation.context) : null\n    );\n\n    // Update sample size\n    this.db.prepare(`\n      UPDATE causal_experiments\n      SET sample_size = sample_size + 1\n      WHERE id = ?\n    `).run(observation.experimentId);\n  }\n\n  /**\n   * Calculate uplift for an experiment\n   */\n  calculateUplift(experimentId: number): {\n    uplift: number;\n    pValue: number;\n    confidenceInterval: [number, number];\n  } {\n    // Get treatment and control observations\n    const observations = this.db.prepare(`\n      SELECT is_treatment, outcome_value\n      FROM causal_observations\n      WHERE experiment_id = ?\n    `).all(experimentId) as any[];\n\n    const treatmentValues = observations\n      .filter(o => o.is_treatment === 1)\n      .map(o => o.outcome_value);\n\n    const controlValues = observations\n      .filter(o => o.is_treatment === 0)\n      .map(o => o.outcome_value);\n\n    if (treatmentValues.length === 0 || controlValues.length === 0) {\n      return { uplift: 0, pValue: 1.0, confidenceInterval: [0, 0] };\n    }\n\n    // Calculate means\n    const treatmentMean = this.mean(treatmentValues);\n    const controlMean = this.mean(controlValues);\n    const uplift = treatmentMean - controlMean;\n\n    // Calculate standard errors\n    const treatmentSE = this.standardError(treatmentValues);\n    const controlSE = this.standardError(controlValues);\n    const pooledSE = Math.sqrt(treatmentSE ** 2 + controlSE ** 2);\n\n    // t-statistic and p-value (two-tailed)\n    const tStat = uplift / pooledSE;\n    const df = treatmentValues.length + controlValues.length - 2;\n    const pValue = 2 * (1 - this.tCDF(Math.abs(tStat), df));\n\n    // 95% confidence interval\n    const tCritical = this.tInverse(0.025, df);\n    const marginOfError = tCritical * pooledSE;\n    const confidenceInterval: [number, number] = [\n      uplift - marginOfError,\n      uplift + marginOfError\n    ];\n\n    // Update experiment with results\n    this.db.prepare(`\n      UPDATE causal_experiments\n      SET treatment_mean = ?,\n          control_mean = ?,\n          uplift = ?,\n          p_value = ?,\n          confidence_interval_low = ?,\n          confidence_interval_high = ?,\n          status = 'completed'\n      WHERE id = ?\n    `).run(\n      treatmentMean,\n      controlMean,\n      uplift,\n      pValue,\n      confidenceInterval[0],\n      confidenceInterval[1],\n      experimentId\n    );\n\n    return { uplift, pValue, confidenceInterval };\n  }\n\n  /**\n   * Query causal effects\n   */\n  queryCausalEffects(query: CausalQuery): CausalEdge[] {\n    const {\n      interventionMemoryId,\n      interventionMemoryType,\n      outcomeMemoryId,\n      minConfidence = 0.5,\n      minUplift = 0.0\n    } = query;\n\n    let sql = `\n      SELECT * FROM causal_edges\n      WHERE from_memory_id = ?\n        AND from_memory_type = ?\n        AND confidence >= ?\n        AND ABS(uplift) >= ?\n    `;\n\n    const params: any[] = [\n      interventionMemoryId,\n      interventionMemoryType,\n      minConfidence,\n      minUplift\n    ];\n\n    if (outcomeMemoryId) {\n      sql += ' AND to_memory_id = ?';\n      params.push(outcomeMemoryId);\n    }\n\n    sql += ' ORDER BY ABS(uplift) * confidence DESC';\n\n    const rows = this.db.prepare(sql).all(...params) as any[];\n\n    return rows.map(row => this.rowToCausalEdge(row));\n  }\n\n  /**\n   * Get causal chain (multi-hop reasoning)\n   *\n   * v2: Uses HyperbolicAttention if enabled for tree-structured retrieval\n   * v1: Falls back to recursive CTE with standard scoring\n   *\n   * @param fromMemoryId - Starting memory node\n   * @param toMemoryId - Target memory node\n   * @param maxDepth - Maximum chain depth (default: 5)\n   * @returns Ranked causal chains with paths, uplift, and confidence\n   */\n  async getCausalChain(fromMemoryId: number, toMemoryId: number, maxDepth: number = 5): Promise<{\n    path: number[];\n    totalUplift: number;\n    confidence: number;\n    attentionMetrics?: {\n      hyperbolicDistance: number[];\n      computeTimeMs: number;\n    };\n  }[]> {\n    // v2: Use HyperbolicAttention if enabled\n    if (this.attentionService && this.embedder) {\n      return this.getCausalChainWithAttention(fromMemoryId, toMemoryId, maxDepth);\n    }\n\n    // v1: Legacy recursive CTE\n    const chains = this.db.prepare(`\n      WITH RECURSIVE chain(from_id, to_id, depth, path, total_uplift, min_confidence) AS (\n        SELECT\n          from_memory_id,\n          to_memory_id,\n          1,\n          from_memory_id || '->' || to_memory_id,\n          uplift,\n          confidence\n        FROM causal_edges\n        WHERE from_memory_id = ? AND confidence >= 0.5\n\n        UNION ALL\n\n        SELECT\n          chain.from_id,\n          ce.to_memory_id,\n          chain.depth + 1,\n          chain.path || '->' || ce.to_memory_id,\n          chain.total_uplift + ce.uplift,\n          MIN(chain.min_confidence, ce.confidence)\n        FROM chain\n        JOIN causal_edges ce ON chain.to_id = ce.from_memory_id\n        WHERE chain.depth < ?\n          AND ce.confidence >= 0.5\n          AND chain.path NOT LIKE '%' || ce.to_memory_id || '%'\n      )\n      SELECT path, total_uplift, min_confidence\n      FROM chain\n      WHERE to_id = ?\n      ORDER BY total_uplift DESC\n      LIMIT 10\n    `).all(fromMemoryId, maxDepth, toMemoryId) as any[];\n\n    return chains.map(row => ({\n      path: row.path.split('->').map(Number),\n      totalUplift: row.total_uplift,\n      confidence: row.min_confidence\n    }));\n  }\n\n  /**\n   * Hash a string to a positive integer\n   * Used for converting string IDs to numeric IDs for backward compatibility\n   */\n  private hashString(str: string): number {\n    let hash = 0;\n    for (let i = 0; i < str.length; i++) {\n      const char = str.charCodeAt(i);\n      hash = ((hash << 5) - hash) + char;\n      hash = hash & hash; // Convert to 32bit integer\n    }\n    return Math.abs(hash);\n  }\n\n  /**\n   * Get causal chain with HyperbolicAttention (v2 feature)\n   *\n   * Uses Poincaré embeddings to model hierarchical causal relationships.\n   * Retrieves chains based on hyperbolic distance in embedding space.\n   *\n   * @private\n   */\n  private async getCausalChainWithAttention(\n    fromMemoryId: number,\n    toMemoryId: number,\n    maxDepth: number\n  ): Promise<{\n    path: number[];\n    totalUplift: number;\n    confidence: number;\n    attentionMetrics: {\n      hyperbolicDistance: number[];\n      computeTimeMs: number;\n    };\n  }[]> {\n    // Get all candidate chains using CTE\n    const candidateChains = this.db.prepare(`\n      WITH RECURSIVE chain(from_id, to_id, depth, path, total_uplift, min_confidence) AS (\n        SELECT\n          from_memory_id,\n          to_memory_id,\n          1,\n          from_memory_id || '->' || to_memory_id,\n          uplift,\n          confidence\n        FROM causal_edges\n        WHERE from_memory_id = ? AND confidence >= 0.5\n\n        UNION ALL\n\n        SELECT\n          chain.from_id,\n          ce.to_memory_id,\n          chain.depth + 1,\n          chain.path || '->' || ce.to_memory_id,\n          chain.total_uplift + ce.uplift,\n          MIN(chain.min_confidence, ce.confidence)\n        FROM chain\n        JOIN causal_edges ce ON chain.to_id = ce.from_memory_id\n        WHERE chain.depth < ?\n          AND ce.confidence >= 0.5\n          AND chain.path NOT LIKE '%' || ce.to_memory_id || '%'\n      )\n      SELECT path, total_uplift, min_confidence\n      FROM chain\n      WHERE to_id = ?\n      LIMIT 50\n    `).all(fromMemoryId, maxDepth, toMemoryId) as any[];\n\n    if (candidateChains.length === 0) {\n      return [];\n    }\n\n    // Get embeddings for query (from node)\n    const fromEpisode = this.db.prepare('SELECT task, output FROM episodes WHERE id = ?').get(fromMemoryId) as any;\n    const queryText = fromEpisode ? `${fromEpisode.task}: ${fromEpisode.output}` : '';\n    const queryEmbedding = await this.embedder!.embed(queryText);\n\n    // Get embeddings and hierarchy levels for all chain nodes\n    const allNodeIds = new Set<number>();\n    candidateChains.forEach((chain: any) => {\n      const path = chain.path.split('->').map(Number);\n      path.forEach(id => allNodeIds.add(id));\n    });\n\n    const nodeEmbeddings = new Map<number, Float32Array>();\n    const hierarchyLevels = new Map<number, number>();\n\n    for (const nodeId of allNodeIds) {\n      const episode = this.db.prepare('SELECT task, output FROM episodes WHERE id = ?').get(nodeId) as any;\n      if (episode) {\n        const text = `${episode.task}: ${episode.output}`;\n        const embedding = await this.embedder!.embed(text);\n        nodeEmbeddings.set(nodeId, embedding);\n\n        // Calculate hierarchy level (depth from root)\n        const level = candidateChains\n          .filter((chain: any) => chain.path.includes(String(nodeId)))\n          .reduce((minDepth: number, chain: any) => {\n            const path = chain.path.split('->').map(Number);\n            const idx = path.indexOf(nodeId);\n            return Math.min(minDepth, idx);\n          }, maxDepth);\n\n        hierarchyLevels.set(nodeId, level);\n      }\n    }\n\n    // Prepare keys, values, and hierarchy for attention\n    const nodeList = Array.from(allNodeIds);\n    const keys = new Float32Array(nodeList.length * 384);\n    const values = new Float32Array(nodeList.length * 384);\n    const hierarchyArray: number[] = [];\n\n    nodeList.forEach((nodeId, idx) => {\n      const embedding = nodeEmbeddings.get(nodeId)!;\n      keys.set(embedding, idx * 384);\n      values.set(embedding, idx * 384);\n      hierarchyArray.push(hierarchyLevels.get(nodeId) || 0);\n    });\n\n    // Apply HyperbolicAttention\n    const queries = new Float32Array(384);\n    queries.set(queryEmbedding);\n\n    const attentionResult = await this.attentionService!.hyperbolicAttention(\n      queries,\n      keys,\n      values,\n      hierarchyArray\n    );\n\n    // Re-rank chains by attention weights\n    const rankedChains = candidateChains\n      .map((chain: any) => {\n        const path = chain.path.split('->').map(Number);\n\n        // Calculate average attention weight for nodes in path\n        const avgWeight = path.reduce((sum: number, nodeId: number) => {\n          const idx = nodeList.indexOf(nodeId);\n          return sum + (idx >= 0 ? attentionResult.weights[idx] : 0);\n        }, 0) / path.length;\n\n        return {\n          path,\n          totalUplift: chain.total_uplift,\n          confidence: chain.min_confidence * avgWeight, // Boost confidence by attention\n          attentionMetrics: {\n            hyperbolicDistance: attentionResult.distances,\n            computeTimeMs: attentionResult.metrics.computeTimeMs,\n          },\n        };\n      })\n      .sort((a, b) => b.confidence - a.confidence)\n      .slice(0, 10);\n\n    return rankedChains;\n  }\n\n  /**\n   * Calculate causal gain: E[outcome|do(treatment)] - E[outcome]\n   */\n  calculateCausalGain(treatmentId: number, outcomeType: 'reward' | 'success' | 'latency'): {\n    causalGain: number;\n    confidence: number;\n    mechanism: string;\n  } {\n    // Get episodes where treatment was applied\n    const withTreatment = this.db.prepare(`\n      SELECT AVG(CASE WHEN ? = 'reward' THEN reward\n                     WHEN ? = 'success' THEN success\n                     WHEN ? = 'latency' THEN latency_ms\n                END) as avg_outcome\n      FROM episodes\n      WHERE id IN (\n        SELECT to_memory_id FROM causal_edges\n        WHERE from_memory_id = ? AND confidence >= 0.6\n      )\n    `).get(outcomeType, outcomeType, outcomeType, treatmentId) as any;\n\n    // Get baseline (no treatment)\n    const baseline = this.db.prepare(`\n      SELECT AVG(CASE WHEN ? = 'reward' THEN reward\n                     WHEN ? = 'success' THEN success\n                     WHEN ? = 'latency' THEN latency_ms\n                END) as avg_outcome\n      FROM episodes\n      WHERE id NOT IN (\n        SELECT to_memory_id FROM causal_edges\n        WHERE from_memory_id = ?\n      )\n    `).get(outcomeType, outcomeType, outcomeType, treatmentId) as any;\n\n    const causalGain = (withTreatment?.avg_outcome || 0) - (baseline?.avg_outcome || 0);\n\n    // Get most confident edge for mechanism\n    const edge = this.db.prepare(`\n      SELECT mechanism, confidence\n      FROM causal_edges\n      WHERE from_memory_id = ?\n      ORDER BY confidence DESC\n      LIMIT 1\n    `).get(treatmentId) as any;\n\n    return {\n      causalGain,\n      confidence: edge?.confidence || 0,\n      mechanism: edge?.mechanism || 'unknown'\n    };\n  }\n\n  /**\n   * Detect confounders using correlation analysis\n   */\n  detectConfounders(edgeId: number): {\n    confounders: Array<{\n      memoryId: number;\n      correlationWithTreatment: number;\n      correlationWithOutcome: number;\n      confounderScore: number;\n    }>;\n  } {\n    const edge = this.db.prepare('SELECT * FROM causal_edges WHERE id = ?').get(edgeId) as any;\n\n    if (!edge) {\n      return { confounders: [] };\n    }\n\n    // Find memories correlated with both treatment and outcome\n    // This is a simplified version - production would use proper statistical tests\n    const potentialConfounders = this.db.prepare(`\n      SELECT DISTINCT e.id, e.task\n      FROM episodes e\n      WHERE e.id != ? AND e.id != ?\n        AND e.session_id IN (\n          SELECT session_id FROM episodes WHERE id = ?\n          UNION\n          SELECT session_id FROM episodes WHERE id = ?\n        )\n    `).all(edge.from_memory_id, edge.to_memory_id, edge.from_memory_id, edge.to_memory_id) as any[];\n\n    const confounders = potentialConfounders.map((conf: any) => {\n      // Calculate correlation scores (simplified)\n      const treatmentCorr = this.calculateCorrelation(conf.id, edge.from_memory_id);\n      const outcomeCorr = this.calculateCorrelation(conf.id, edge.to_memory_id);\n      const confounderScore = Math.sqrt(treatmentCorr ** 2 * outcomeCorr ** 2);\n\n      return {\n        memoryId: conf.id,\n        correlationWithTreatment: treatmentCorr,\n        correlationWithOutcome: outcomeCorr,\n        confounderScore\n      };\n    }).filter(c => c.confounderScore > 0.3);\n\n    // Update edge with confounder score\n    if (confounders.length > 0) {\n      const maxConfounderScore = Math.max(...confounders.map(c => c.confounderScore));\n      this.db.prepare(`\n        UPDATE causal_edges\n        SET confounder_score = ?\n        WHERE id = ?\n      `).run(maxConfounderScore, edgeId);\n    }\n\n    return { confounders };\n  }\n\n  // ========================================================================\n  // Private Helper Methods\n  // ========================================================================\n\n  private rowToCausalEdge(row: any): CausalEdge {\n    return {\n      id: row.id,\n      fromMemoryId: row.from_memory_id,\n      fromMemoryType: row.from_memory_type,\n      toMemoryId: row.to_memory_id,\n      toMemoryType: row.to_memory_type,\n      similarity: row.similarity,\n      uplift: row.uplift,\n      confidence: row.confidence,\n      sampleSize: row.sample_size,\n      evidenceIds: row.evidence_ids ? JSON.parse(row.evidence_ids) : undefined,\n      experimentIds: row.experiment_ids ? JSON.parse(row.experiment_ids) : undefined,\n      confounderScore: row.confounder_score,\n      mechanism: row.mechanism,\n      metadata: row.metadata ? JSON.parse(row.metadata) : undefined\n    };\n  }\n\n  private mean(values: number[]): number {\n    return values.reduce((a, b) => a + b, 0) / values.length;\n  }\n\n  private variance(values: number[]): number {\n    const avg = this.mean(values);\n    return values.reduce((sum, val) => sum + (val - avg) ** 2, 0) / values.length;\n  }\n\n  private standardError(values: number[]): number {\n    return Math.sqrt(this.variance(values) / values.length);\n  }\n\n  private tCDF(t: number, df: number): number {\n    // Simplified t-distribution CDF (use proper stats library in production)\n    // This is an approximation\n    return 0.5 + 0.5 * Math.sign(t) * (1 - Math.pow(1 + t * t / df, -df / 2));\n  }\n\n  private tInverse(p: number, df: number): number {\n    // Simplified inverse t-distribution (use proper stats library)\n    // Approximation for 95% CI\n    return 1.96; // Standard normal approximation\n  }\n\n  private calculateCorrelation(id1: number, id2: number): number {\n    // Simplified correlation calculation\n    // In production, use proper correlation metrics\n    const sharedSessions = this.db.prepare(`\n      SELECT COUNT(DISTINCT e1.session_id) as shared\n      FROM episodes e1\n      JOIN episodes e2 ON e1.session_id = e2.session_id\n      WHERE e1.id = ? AND e2.id = ?\n    `).get(id1, id2) as any;\n\n    return Math.min(sharedSessions?.shared || 0, 1.0);\n  }\n}\n"]}