@timmeck/brain-core 2.3.0 → 2.5.0

package/dist/meta-learning/engine.js

@@ -0,0 +1,275 @@
+import { getLogger } from '../utils/logger.js';
+// ── Migration ───────────────────────────────────────────
+export function runMetaLearningMigration(db) {
+    db.exec(`
+    CREATE TABLE IF NOT EXISTS meta_learning_snapshots (
+      id INTEGER PRIMARY KEY AUTOINCREMENT,
+      cycle INTEGER NOT NULL,
+      params TEXT NOT NULL,
+      metrics TEXT NOT NULL,
+      score REAL NOT NULL,
+      created_at TEXT NOT NULL DEFAULT (datetime('now'))
+    );
+
+    CREATE TABLE IF NOT EXISTS meta_learning_optimizations (
+      id INTEGER PRIMARY KEY AUTOINCREMENT,
+      param_name TEXT NOT NULL,
+      old_value REAL NOT NULL,
+      new_value REAL NOT NULL,
+      reason TEXT NOT NULL,
+      improvement REAL,
+      created_at TEXT NOT NULL DEFAULT (datetime('now'))
+    );
+
+    CREATE INDEX IF NOT EXISTS idx_meta_snapshots_score ON meta_learning_snapshots(score);
+    CREATE INDEX IF NOT EXISTS idx_meta_snapshots_cycle ON meta_learning_snapshots(cycle);
+  `);
+}
+// ── Engine ───────────────────────────────────────────────
+/**
+ * Meta-Learning Engine: observes learning cycle outcomes over time
+ * and auto-tunes hyperparameters using gradient-free optimization.
+ *
+ * Research approach: Bayesian-inspired parameter search.
+ * - After each learning cycle, record a snapshot (params + metrics + effectiveness score)
+ * - Periodically analyze: which parameter configurations produced the best scores?
+ * - Use perturbation: try small changes to parameters, measure improvement, keep what works
+ * - Implements "explore vs exploit": 80% exploit best known, 20% explore new configurations
+ */
+export class MetaLearningEngine {
+    db;
+    params;
+    logger = getLogger();
+    cycleCount = 0;
+    analyzeInterval; // analyze every N cycles
+    explorationRate; // % of cycles spent exploring (0-1)
+    constructor(db, params, config) {
+        this.db = db;
+        this.params = params;
+        runMetaLearningMigration(db);
+        this.analyzeInterval = config?.analyzeInterval ?? 5;
+        this.explorationRate = config?.explorationRate ?? 0.2;
+    }
+    /**
+     * Record a learning cycle's outcome. Called after each learning cycle.
+     * @param metrics - the raw metrics from the cycle (e.g. newPatterns, updatedRules, prunedRules)
+     * @param score - a composite effectiveness score (higher = better)
+     */
+    recordSnapshot(metrics, score) {
+        this.cycleCount++;
+        const currentParams = {};
+        for (const p of this.params) {
+            currentParams[p.name] = p.value;
+        }
+        this.db.prepare(`
+      INSERT INTO meta_learning_snapshots (cycle, params, metrics, score)
+      VALUES (?, ?, ?, ?)
+    `).run(this.cycleCount, JSON.stringify(currentParams), JSON.stringify(metrics), score);
+        const snapshot = {
+            cycle: this.cycleCount,
+            params: currentParams,
+            metrics,
+            score,
+        };
+        this.logger.debug(`Meta-learning snapshot #${this.cycleCount}: score=${score.toFixed(3)}`);
+        return snapshot;
+    }
+    /**
+     * Analyze snapshots and recommend parameter changes.
+     * Uses a simplified approach inspired by Bayesian optimization:
+     * - Group snapshots by parameter ranges
+     * - Find which ranges produced the best average scores
+     * - Recommend moving towards those ranges
+     */
+    analyze() {
+        const snapshots = this.getSnapshots(50); // last 50 cycles
+        if (snapshots.length < 5)
+            return []; // need minimum data
+        const recommendations = [];
+        for (const param of this.params) {
+            const rec = this.analyzeParameter(param, snapshots);
+            if (rec)
+                recommendations.push(rec);
+        }
+        return recommendations;
+    }
+    /**
+     * Apply recommendations: perturb parameters towards better configurations.
+     * Returns the parameters that were changed.
+     */
+    optimize() {
+        const recommendations = this.analyze();
+        const applied = [];
+        for (const rec of recommendations) {
+            if (rec.confidence < 0.3)
+                continue; // skip low-confidence recommendations
+            if (Math.abs(rec.expectedImprovement) < 0.01)
+                continue; // skip negligible improvements
+            const param = this.params.find(p => p.name === rec.name);
+            if (!param)
+                continue;
+            // Explore vs exploit: sometimes try random perturbations
+            const exploring = Math.random() < this.explorationRate;
+            let newValue;
+            if (exploring) {
+                // Random perturbation within bounds
+                const range = param.max - param.min;
+                newValue = param.value + (Math.random() - 0.5) * range * 0.2;
+            }
+            else {
+                // Move towards recommended value
+                newValue = rec.recommendedValue;
+            }
+            // Clamp to bounds
+            newValue = Math.max(param.min, Math.min(param.max, newValue));
+            // Record optimization
+            this.db.prepare(`
+        INSERT INTO meta_learning_optimizations (param_name, old_value, new_value, reason, improvement)
+        VALUES (?, ?, ?, ?, ?)
+      `).run(param.name, param.value, newValue, exploring ? 'exploration' : 'exploitation', rec.expectedImprovement);
+            this.logger.info(`Meta-learning: ${param.name} ${param.value.toFixed(4)} → ${newValue.toFixed(4)} (${exploring ? 'explore' : 'exploit'}, expected +${(rec.expectedImprovement * 100).toFixed(1)}%)`);
+            param.value = newValue;
+            rec.recommendedValue = newValue;
+            applied.push(rec);
+        }
+        return applied;
+    }
+    /**
+     * Run analysis + optimization if it's time. Call this after every learning cycle.
+     */
+    step(metrics, score) {
+        const snapshot = this.recordSnapshot(metrics, score);
+        let optimized = [];
+        if (this.cycleCount > 0 && this.cycleCount % this.analyzeInterval === 0) {
+            optimized = this.optimize();
+        }
+        return { snapshot, optimized };
+    }
+    /** Get current parameter values. */
+    getParams() {
+        const result = {};
+        for (const p of this.params) {
+            result[p.name] = p.value;
+        }
+        return result;
+    }
+    /** Update a parameter value externally (e.g. user override). */
+    setParam(name, value) {
+        const param = this.params.find(p => p.name === name);
+        if (!param)
+            return false;
+        param.value = Math.max(param.min, Math.min(param.max, value));
+        return true;
+    }
+    /** Get learning effectiveness status. */
+    getStatus() {
+        const snapshots = this.getSnapshots(100);
+        const optimizations = this.db.prepare('SELECT COUNT(*) as count FROM meta_learning_optimizations').get();
+        const scores = snapshots.map(s => s.score);
+        const currentScore = scores.length > 0 ? scores[0] : 0;
+        const bestScore = scores.length > 0 ? Math.max(...scores) : 0;
+        const worstScore = scores.length > 0 ? Math.min(...scores) : 0;
+        // Trend detection: compare average of last 5 vs previous 5
+        let trend = 'stable';
+        if (scores.length >= 10) {
+            const recent = scores.slice(0, 5).reduce((a, b) => a + b, 0) / 5;
+            const previous = scores.slice(5, 10).reduce((a, b) => a + b, 0) / 5;
+            const delta = recent - previous;
+            if (delta > 0.05)
+                trend = 'improving';
+            else if (delta < -0.05)
+                trend = 'declining';
+        }
+        return {
+            totalSnapshots: snapshots.length,
+            totalOptimizations: optimizations.count,
+            bestScore,
+            worstScore,
+            currentScore,
+            trend,
+            recommendations: this.analyze(),
+        };
+    }
+    /** Get optimization history. */
+    getHistory(limit = 20) {
+        return this.db.prepare('SELECT * FROM meta_learning_optimizations ORDER BY created_at DESC LIMIT ?').all(limit);
+    }
+    // ── Private ─────────────────────────────────────────
+    getSnapshots(limit) {
+        const rows = this.db.prepare('SELECT * FROM meta_learning_snapshots ORDER BY cycle DESC LIMIT ?').all(limit);
+        return rows.map(r => ({
+            id: r.id,
+            cycle: r.cycle,
+            params: JSON.parse(r.params),
+            metrics: JSON.parse(r.metrics),
+            score: r.score,
+            created_at: r.created_at,
+        }));
+    }
+    /**
+     * Analyze a single parameter's effect on learning effectiveness.
+     *
+     * Algorithm:
+     * 1. Divide the parameter's range into bins
+     * 2. For each bin, calculate the average score of snapshots in that bin
+     * 3. Find the bin with the highest average score
+     * 4. Recommend moving towards the center of that bin
+     */
+    analyzeParameter(param, snapshots) {
+        const numBins = 5;
+        const range = param.max - param.min;
+        if (range <= 0)
+            return null;
+        const binSize = range / numBins;
+        // Accumulate scores per bin
+        const bins = Array.from({ length: numBins }, () => ({ sum: 0, count: 0 }));
+        for (const snap of snapshots) {
+            const val = snap.params[param.name];
+            if (val === undefined)
+                continue;
+            const binIdx = Math.min(numBins - 1, Math.floor((val - param.min) / binSize));
+            bins[binIdx].sum += snap.score;
+            bins[binIdx].count++;
+        }
+        // Find best bin (minimum 2 samples)
+        let bestBin = -1;
+        let bestAvg = -Infinity;
+        let totalSamples = 0;
+        for (let i = 0; i < numBins; i++) {
+            if (bins[i].count < 2)
+                continue;
+            const avg = bins[i].sum / bins[i].count;
+            totalSamples += bins[i].count;
+            if (avg > bestAvg) {
+                bestAvg = avg;
+                bestBin = i;
+            }
+        }
+        if (bestBin === -1)
+            return null; // not enough data
+        // Calculate recommended value (center of best bin)
+        const recommended = param.min + (bestBin + 0.5) * binSize;
+        // Skip if current value is already in or near the best bin
+        const currentBin = Math.min(numBins - 1, Math.floor((param.value - param.min) / binSize));
+        if (currentBin === bestBin)
+            return null;
+        // Estimate improvement
+        const currentBinAvg = bins[currentBin].count > 0
+            ? bins[currentBin].sum / bins[currentBin].count
+            : 0;
+        const expectedImprovement = currentBinAvg > 0
+            ? (bestAvg - currentBinAvg) / currentBinAvg
+            : 0;
+        // Confidence: based on sample count in best bin
+        const confidence = Math.min(1, bins[bestBin].count / 10);
+        return {
+            name: param.name,
+            currentValue: param.value,
+            recommendedValue: recommended,
+            expectedImprovement,
+            confidence,
+            evidence: totalSamples,
+        };
+    }
+}
+//# sourceMappingURL=engine.js.map

package/dist/meta-learning/engine.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"engine.js","sourceRoot":"","sources":["../../src/meta-learning/engine.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,SAAS,EAAE,MAAM,oBAAoB,CAAC;AAwC/C,2DAA2D;AAE3D,MAAM,UAAU,wBAAwB,CAAC,EAAqB;IAC5D,EAAE,CAAC,IAAI,CAAC;;;;;;;;;;;;;;;;;;;;;;GAsBP,CAAC,CAAC;AACL,CAAC;AAED,4DAA4D;AAE5D;;;;;;;;;GASG;AACH,MAAM,OAAO,kBAAkB;IAOnB;IACA;IAPF,MAAM,GAAG,SAAS,EAAE,CAAC;IACrB,UAAU,GAAG,CAAC,CAAC;IACf,eAAe,CAAS,CAAI,yBAAyB;IACrD,eAAe,CAAS,CAAI,oCAAoC;IAExE,YACU,EAAqB,EACrB,MAAwB,EAChC,MAA+D;QAFvD,OAAE,GAAF,EAAE,CAAmB;QACrB,WAAM,GAAN,MAAM,CAAkB;QAGhC,wBAAwB,CAAC,EAAE,CAAC,CAAC;QAC7B,IAAI,CAAC,eAAe,GAAG,MAAM,EAAE,eAAe,IAAI,CAAC,CAAC;QACpD,IAAI,CAAC,eAAe,GAAG,MAAM,EAAE,eAAe,IAAI,GAAG,CAAC;IACxD,CAAC;IAED;;;;OAIG;IACH,cAAc,CAAC,OAA+B,EAAE,KAAa;QAC3D,IAAI,CAAC,UAAU,EAAE,CAAC;QAElB,MAAM,aAAa,GAA2B,EAAE,CAAC;QACjD,KAAK,MAAM,CAAC,IAAI,IAAI,CAAC,MAAM,EAAE,CAAC;YAC5B,aAAa,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,KAAK,CAAC;QAClC,CAAC;QAED,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC;;;KAGf,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,UAAU,EAAE,IAAI,CAAC,SAAS,CAAC,aAAa,CAAC,EAAE,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,EAAE,KAAK,CAAC,CAAC;QAEvF,MAAM,QAAQ,GAAqB;YACjC,KAAK,EAAE,IAAI,CAAC,UAAU;YACtB,MAAM,EAAE,aAAa;YACrB,OAAO;YACP,KAAK;SACN,CAAC;QAEF,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,2BAA2B,IAAI,CAAC,UAAU,WAAW,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;QAE3F,OAAO,QAAQ,CAAC;IAClB,CAAC;IAED;;;;;;OAMG;IACH,OAAO;QACL,MAAM,SAAS,GAAG,IAAI,CAAC,YAAY,CAAC,EAAE,CAAC,CAAC,CAAC,iBAAiB;QAC1D,IAAI,SAAS,CAAC,MAAM,GAAG,CAAC;YAAE,OAAO,EAAE,CAAC,CAAC,oBAAoB;QAEzD,MAAM,eAAe,GAA8B,EAAE,CAAC;QAEtD,KAAK,MAAM,KAAK,IAAI,IAAI,CAAC,MAAM,EAAE,CAAC;YAChC,MAAM,GAAG,GAAG,IAAI,CAAC,gBAAgB,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC;YACpD,IAAI,GAAG;gBAAE,eAAe,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QACrC,CAAC;QAED,OAAO,eAAe,CAAC;IACzB,CAAC;IAED;;;OAGG;IACH,QAAQ;QACN,MAAM,eAAe,GAAG,IAAI,CAAC,OAAO,EAAE,CAAC;QACvC,MAAM,OAAO,GAA8B,EAAE,CAAC;QAE9C,KAAK,MAAM,GAAG,IAAI,eAAe,EAAE,CAAC;YAClC,IAAI,GAAG,CAAC,UAAU,GAAG,GAAG;gBAAE,SAAS,CAAC,sCAAsC;YAC1E,IAAI,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,mBAAmB,CAAC,GAAG,IAAI;gBAAE,SAAS,CAAC,+BAA+B;YAEvF,MAAM,KAAK,GAAG,IAAI,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,GAAG,CAAC,IAAI,CAAC,CAAC;YACzD,IAAI,CAAC,KAAK;gBAAE,SAAS;YAErB,yDAAyD;YACzD,MAAM,SAAS,GAAG,IAAI,CAAC,MAAM,EAAE,GAAG,IAAI,CAAC,eAAe,CAAC;YACvD,IAAI,QAAgB,CAAC;YAErB,IAAI,SAAS,EAAE,CAAC;gBACd,oCAAoC;gBACpC,MAAM,KAAK,GAAG,KAAK,CAAC,GAAG,GAAG,KAAK,CAAC,GAAG,CAAC;gBACpC,QAAQ,GAAG,KAAK,CAAC,KAAK,GAAG,CAAC,IAAI,CAAC,MAAM,EAAE,GAAG,GAAG,CAAC,GAAG,KAAK,GAAG,GAAG,CAAC;YAC/D,CAAC;iBAAM,CAAC;gBACN,iCAAiC;gBACjC,QAAQ,GAAG,GAAG,CAAC,gBAAgB,CAAC;YAClC,CAAC;YAED,kBAAkB;YAClB,QAAQ,GAAG,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,GAAG,EAAE,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,GAAG,EAAE,QAAQ,CAAC,CAAC,CAAC;YAE9D,sBAAsB;YACtB,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC;;;OAGf,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,EAAE,KAAK,CAAC,KAAK,EAAE,QAAQ,EAAE,SAAS,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,cAAc,EAAE,GAAG,CAAC,mBAAmB,CAAC,CAAC;YAE/G,IAAI,CAAC,MAAM,CAAC,IAAI,CAAC,kBAAkB,KAAK,CAAC,IAAI,IAAI,KAAK,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,SAAS,eAAe,CAAC,GAAG,CAAC,mBAAmB,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC;YAErM,KAAK,CAAC,KAAK,GAAG,QAAQ,CAAC;YACvB,GAAG,CAAC,gBAAgB,GAAG,QAAQ,CAAC;YAChC,OAAO,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QACpB,CAAC;QAED,OAAO,OAAO,CAAC;IACjB,CAAC;IAED;;OAEG;IACH,IAAI,CAAC,OAA+B,EAAE,KAAa;QACjD,MAAM,QAAQ,GAAG,IAAI,CAAC,cAAc,CAAC,OAAO,EAAE,KAAK,CAAC,CAAC;QAErD,IAAI,SAAS,GAA8B,EAAE,CAAC;QAC9C,IAAI,IAAI,CAAC,UAAU,GAAG,CAAC,IAAI,IAAI,CAAC,UAAU,GAAG,IAAI,CAAC,eAAe,KAAK,CAAC,EAAE,CAAC;YACxE,SAAS,GAAG,IAAI,CAAC,QAAQ,EAAE,CAAC;QAC9B,CAAC;QAED,OAAO,EAAE,QAAQ,EAAE,SAAS,EAAE,CAAC;IACjC,CAAC;IAED,oCAAoC;IACpC,SAAS;QACP,MAAM,MAAM,GAA2B,EAAE,CAAC;QAC1C,KAAK,MAAM,CAAC,IAAI,IAAI,CAAC,MAAM,EAAE,CAAC;YAC5B,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,KAAK,CAAC;QAC3B,CAAC;QACD,OAAO,MAAM,CAAC;IAChB,CAAC;IAED,gEAAgE;IAChE,QAAQ,CAAC,IAAY,EAAE,KAAa;QAClC,MAAM,KAAK,GAAG,IAAI,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,IAAI,CAAC,CAAC;QACrD,IAAI,CAAC,KAAK;YAAE,OAAO,KAAK,CAAC;QACzB,KAAK,CAAC,KAAK,GAAG,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,GAAG,EAAE,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,GAAG,EAAE,KAAK,CAAC,CAAC,CAAC;QAC9D,OAAO,IAAI,CAAC;IACd,CAAC;IAED,yCAAyC;IACzC,SAAS;QACP,MAAM,SAAS,GAAG,IAAI,CAAC,YAAY,CAAC,GAAG,CAAC,CAAC;QACzC,MAAM,aAAa,GAAG,IAAI,CAAC,EAAE,CAAC,OAAO,CACnC,2DAA2D,CAC5D,CAAC,GAAG,EAAuB,CAAC;QAE7B,MAAM,MAAM,GAAG,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC;QAC3C,MAAM,YAAY,GAAG,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAE,CAAC,CAAC,CAAC,CAAC,CAAC;QACxD,MAAM,SAAS,GAAG,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,GAAG,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;QAC9D,MAAM,UAAU,GAAG,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,GAAG,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;QAE/D,2DAA2D;QAC3D,IAAI,KAAK,GAAyC,QAAQ,CAAC;QAC3D,IAAI,MAAM,CAAC,MAAM,IAAI,EAAE,EAAE,CAAC;YACxB,MAAM,MAAM,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC;YACjE,MAAM,QAAQ,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC;YACpE,MAAM,KAAK,GAAG,MAAM,GAAG,QAAQ,CAAC;YAChC,IAAI,KAAK,GAAG,IAAI;gBAAE,KAAK,GAAG,WAAW,CAAC;iBACjC,IAAI,KAAK,GAAG,CAAC,IAAI;gBAAE,KAAK,GAAG,WAAW,CAAC;QAC9C,CAAC;QAED,OAAO;YACL,cAAc,EAAE,SAAS,CAAC,MAAM;YAChC,kBAAkB,EAAE,aAAa,CAAC,KAAK;YACvC,SAAS;YACT,UAAU;YACV,YAAY;YACZ,KAAK;YACL,eAAe,EAAE,IAAI,CAAC,OAAO,EAAE;SAChC,CAAC;IACJ,CAAC;IAED,gCAAgC;IAChC,UAAU,CAAC,KAAK,GAAG,EAAE;QAQnB,OAAO,IAAI,CAAC,EAAE,CAAC,OAAO,CACpB,4EAA4E,CAC7E,CAAC,GAAG,CAAC,KAAK,CAAU,CAAC;IACxB,CAAC;IAED,uDAAuD;IAE/C,YAAY,CAAC,KAAa;QAChC,MAAM,IAAI,GAAG,IAAI,CAAC,EAAE,CAAC,OAAO,CAC1B,mEAAmE,CACpE,CAAC,GAAG,CAAC,KAAK,CAET,CAAC;QAEH,OAAO,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;YACpB,EAAE,EAAE,CAAC,CAAC,EAAE;YACR,KAAK,EAAE,CAAC,CAAC,KAAK;YACd,MAAM,EAAE,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC;YAC5B,OAAO,EAAE,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,OAAO,CAAC;YAC9B,KAAK,EAAE,CAAC,CAAC,KAAK;YACd,UAAU,EAAE,CAAC,CAAC,UAAU;SACzB,CAAC,CAAC,CAAC;IACN,CAAC;IAED;;;;;;;;OAQG;IACK,gBAAgB,CAAC,KAAqB,EAAE,SAA6B;QAC3E,MAAM,OAAO,GAAG,CAAC,CAAC;QAClB,MAAM,KAAK,GAAG,KAAK,CAAC,GAAG,GAAG,KAAK,CAAC,GAAG,CAAC;QACpC,IAAI,KAAK,IAAI,CAAC;YAAE,OAAO,IAAI,CAAC;QAE5B,MAAM,OAAO,GAAG,KAAK,GAAG,OAAO,CAAC;QAEhC,4BAA4B;QAC5B,MAAM,IAAI,GAAqC,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,OAAO,EAAE,EAAE,GAAG,EAAE,CAAC,CAAC,EAAE,GAAG,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,CAAC,CAAC,CAAC;QAE7G,KAAK,MAAM,IAAI,IAAI,SAAS,EAAE,CAAC;YAC7B,MAAM,GAAG,GAAG,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;YACpC,IAAI,GAAG,KAAK,SAAS;gBAAE,SAAS;YAEhC,MAAM,MAAM,GAAG,IAAI,CAAC,GAAG,CAAC,OAAO,GAAG,CAAC,EAAE,IAAI,CAAC,KAAK,CAAC,CAAC,GAAG,GAAG,KAAK,CAAC,GAAG,CAAC,GAAG,OAAO,CAAC,CAAC,CAAC;YAC9E,IAAI,CAAC,MAAM,CAAE,CAAC,GAAG,IAAI,IAAI,CAAC,KAAK,CAAC;YAChC,IAAI,CAAC,MAAM,CAAE,CAAC,KAAK,EAAE,CAAC;QACxB,CAAC;QAED,oCAAoC;QACpC,IAAI,OAAO,GAAG,CAAC,CAAC,CAAC;QACjB,IAAI,OAAO,GAAG,CAAC,QAAQ,CAAC;QACxB,IAAI,YAAY,GAAG,CAAC,CAAC;QAErB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,OAAO,EAAE,CAAC,EAAE,EAAE,CAAC;YACjC,IAAI,IAAI,CAAC,CAAC,CAAE,CAAC,KAAK,GAAG,CAAC;gBAAE,SAAS;YACjC,MAAM,GAAG,GAAG,IAAI,CAAC,CAAC,CAAE,CAAC,GAAG,GAAG,IAAI,CAAC,CAAC,CAAE,CAAC,KAAK,CAAC;YAC1C,YAAY,IAAI,IAAI,CAAC,CAAC,CAAE,CAAC,KAAK,CAAC;YAC/B,IAAI,GAAG,GAAG,OAAO,EAAE,CAAC;gBAClB,OAAO,GAAG,GAAG,CAAC;gBACd,OAAO,GAAG,CAAC,CAAC;YACd,CAAC;QACH,CAAC;QAED,IAAI,OAAO,KAAK,CAAC,CAAC;YAAE,OAAO,IAAI,CAAC,CAAC,kBAAkB;QAEnD,mDAAmD;QACnD,MAAM,WAAW,GAAG,KAAK,CAAC,GAAG,GAAG,CAAC,OAAO,GAAG,GAAG,CAAC,GAAG,OAAO,CAAC;QAE1D,2DAA2D;QAC3D,MAAM,UAAU,GAAG,IAAI,CAAC,GAAG,CAAC,OAAO,GAAG,CAAC,EAAE,IAAI,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,KAAK,GAAG,KAAK,CAAC,GAAG,CAAC,GAAG,OAAO,CAAC,CAAC,CAAC;QAC1F,IAAI,UAAU,KAAK,OAAO;YAAE,OAAO,IAAI,CAAC;QAExC,uBAAuB;QACvB,MAAM,aAAa,GAAG,IAAI,CAAC,UAAU,CAAE,CAAC,KAAK,GAAG,CAAC;YAC/C,CAAC,CAAC,IAAI,CAAC,UAAU,CAAE,CAAC,GAAG,GAAG,IAAI,CAAC,UAAU,CAAE,CAAC,KAAK;YACjD,CAAC,CAAC,CAAC,CAAC;QACN,MAAM,mBAAmB,GAAG,aAAa,GAAG,CAAC;YAC3C,CAAC,CAAC,CAAC,OAAO,GAAG,aAAa,CAAC,GAAG,aAAa;YAC3C,CAAC,CAAC,CAAC,CAAC;QAEN,gDAAgD;QAChD,MAAM,UAAU,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,OAAO,CAAE,CAAC,KAAK,GAAG,EAAE,CAAC,CAAC;QAE1D,OAAO;YACL,IAAI,EAAE,KAAK,CAAC,IAAI;YAChB,YAAY,EAAE,KAAK,CAAC,KAAK;YACzB,gBAAgB,EAAE,WAAW;YAC7B,mBAAmB;YACnB,UAAU;YACV,QAAQ,EAAE,YAAY;SACvB,CAAC;IACJ,CAAC;CACF"}

package/dist/research/autonomous-scheduler.d.ts

@@ -0,0 +1,124 @@
+import type Database from 'better-sqlite3';
+import { MetaLearningEngine } from '../meta-learning/engine.js';
+import type { HyperParameter } from '../meta-learning/engine.js';
+import { CausalGraph } from '../causal/engine.js';
+import { HypothesisEngine } from '../hypothesis/engine.js';
+export interface ResearchDiscovery {
+    id?: number;
+    type: 'causal_chain' | 'confirmed_hypothesis' | 'parameter_optimization' | 'anomaly' | 'root_cause';
+    title: string;
+    description: string;
+    confidence: number;
+    impact: number;
+    source: string;
+    data: Record<string, unknown>;
+    created_at?: string;
+}
+export interface ResearchCycleReport {
+    cycle: number;
+    timestamp: number;
+    causalEdgesFound: number;
+    causalChainsFound: number;
+    hypothesesGenerated: number;
+    hypothesesTested: number;
+    hypothesesConfirmed: number;
+    hypothesesRejected: number;
+    parametersOptimized: number;
+    discoveriesProduced: number;
+    duration: number;
+}
+export interface AutonomousResearchConfig {
+    /** How often to run the full research cycle (ms). Default: 600_000 (10 min) */
+    intervalMs?: number;
+    /** Initial delay before first cycle (ms). Default: 30_000 (30s) */
+    initialDelayMs?: number;
+    /** Brain name for event attribution. */
+    brainName: string;
+    /** Hyperparameters for meta-learning. */
+    hyperParams?: HyperParameter[];
+    /** Minimum causal edge strength to produce a discovery. Default: 0.3 */
+    minCausalStrength?: number;
+    /** Maximum discoveries per cycle. Default: 10 */
+    maxDiscoveriesPerCycle?: number;
+}
+export declare function runResearchDiscoveryMigration(db: Database.Database): void;
+/**
+ * Autonomous Research Scheduler: orchestrates the three research engines
+ * into a self-running discovery pipeline.
+ *
+ * Flow:
+ * 1. External code calls onLearningCycleComplete() after each learning cycle
+ *    → feeds metrics into MetaLearningEngine
+ *    → records causal events
+ *    → records observations for hypothesis engine
+ *
+ * 2. On a timer (every intervalMs), runs a full research cycle:
+ *    a. Run causal analysis → detect cause-effect relationships
+ *    b. Auto-generate hypotheses from observation patterns
+ *    c. Test all pending hypotheses
+ *    d. Run meta-learning optimization (if enough data)
+ *    e. Produce discoveries from confirmed hypotheses + causal chains
+ *    f. Store everything in SQLite
+ */
+export declare class AutonomousResearchScheduler {
+    private db;
+    private logger;
+    private timer;
+    private delayTimer;
+    private cycleCount;
+    private running;
+    readonly metaLearning: MetaLearningEngine;
+    readonly causalGraph: CausalGraph;
+    readonly hypothesisEngine: HypothesisEngine;
+    private brainName;
+    private intervalMs;
+    private initialDelayMs;
+    private minCausalStrength;
+    private maxDiscoveriesPerCycle;
+    constructor(db: Database.Database, config: AutonomousResearchConfig);
+    /** Start the autonomous research timer. */
+    start(): void;
+    /** Stop the autonomous research timer. */
+    stop(): void;
+    /**
+     * Call this after each learning cycle completes.
+     * Feeds the cycle results into all three research engines.
+     */
+    onLearningCycleComplete(metrics: Record<string, number>, score: number, eventType?: string): void;
+    /**
+     * Record a domain event for causal + hypothesis tracking.
+     * Call this from event listeners in each brain.
+     */
+    recordEvent(eventType: string, data?: Record<string, unknown>): void;
+    /**
+     * Run one full autonomous research cycle.
+     * This is the core algorithm — the system thinking about itself.
+     */
+    runCycle(): ResearchCycleReport;
+    /** Get all discoveries, optionally filtered by type. */
+    getDiscoveries(type?: string, limit?: number): ResearchDiscovery[];
+    /** Get cycle reports. */
+    getCycleReports(limit?: number): ResearchCycleReport[];
+    /** Get a comprehensive research status. */
+    getStatus(): {
+        cyclesCompleted: number;
+        totalDiscoveries: number;
+        discoveryBreakdown: Record<string, number>;
+        metaLearningStatus: ReturnType<MetaLearningEngine['getStatus']>;
+        causalAnalysis: ReturnType<CausalGraph['getAnalysis']>;
+        hypothesisSummary: ReturnType<HypothesisEngine['getSummary']>;
+        lastCycleReport: ResearchCycleReport | null;
+        isRunning: boolean;
+    };
+    private safeRunCycle;
+    /**
+     * Produce discoveries from research findings.
+     * This is where data becomes insight.
+     */
+    private produceDiscoveries;
+    private findSignificantRoots;
+    private isDuplicateDiscovery;
+    private storeDiscovery;
+    private storeCycleReport;
+    private emptyReport;
+}