@stevenvo780/st-lang 4.14.0 → 4.15.0

package/dist/reasoning/lemma-rag/embedding.js

@@ -0,0 +1,243 @@
+"use strict";
+// ============================================================
+// HashEmbedding — embedding determinístico R^256 sobre features AST
+//
+// Sin dependencias de red ni modelos externos. Features extraídas
+// directamente de la representación string de la fórmula:
+//   1. Frecuencias de conectivos top-level (∧ ∨ ¬ → ↔ □ ◇ ∀ ∃ = ⊥ ⊤)
+//   2. Átomos/identificadores hasheados a posiciones del vector
+//   3. Profundidad de cuantificadores (∀ ∃)
+//   4. Tipos de operadores (modal, aritmético, set-theory, etc.)
+//   5. Firma de esqueleto: secuencia de categorías de tokens
+//
+// Propiedades garantizadas:
+//   - Determinístico: misma entrada → mismo vector
+//   - α-invariante por canonicalización previa
+//   - Dimensión fija: EMBEDDING_DIM (256)
+//   - Sin NaN/Infinity
+// ============================================================
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.defaultProvider = exports.HashEmbeddingProvider = void 0;
+exports.normalizeEmbedding = normalizeEmbedding;
+exports.hashEmbed = hashEmbed;
+const types_1 = require("./types");
+// --------------- Hashing FNV-1a 32-bit (determinístico) ----------------
+function fnv1a(str) {
+    let hash = 2166136261;
+    for (let i = 0; i < str.length; i++) {
+        hash ^= str.charCodeAt(i);
+        hash = (hash * 16777619) >>> 0;
+    }
+    return hash;
+}
+/** Proyecta un string a un índice [0, DIM). */
+function strToIndex(str, dim) {
+    return fnv1a(str) % dim;
+}
+/** Proyecta un string a un índice con desplazamiento de salt. */
+function strToIndexSalted(str, salt, dim) {
+    return fnv1a(salt + str) % dim;
+}
+// --------------- Extracción de features --------------------------------
+const CONNECTIVE_SYMBOLS = ['∧', '∨', '¬', '→', '↔', '□', '◇', '∀', '∃', '=', '⊥', '⊤', '⊢', '⇒'];
+const MODAL_OPS = new Set(['□', '◇']);
+const QUANTIFIER_OPS = new Set(['∀', '∃']);
+const ARITH_OPS = new Set(['+', '·', '<', '≤', '>', '≠', 'S']);
+const SET_OPS = new Set(['∈', '∉', '⊆', '⊂', '∪', '∩', '∅']);
+const IDENT_RE = /[A-Za-z_][A-Za-z0-9_]*/g;
+const LOGIC_RESERVED = new Set([
+    'not', 'and', 'or', 'implies', 'iff', 'forall', 'exists',
+    'true', 'false', 'True', 'False', 'TRUE', 'FALSE',
+]);
+function extractFeatures(formula) {
+    const connectiveFreq = {};
+    const atomFreq = {};
+    let quantifierDepth = 0;
+    let currentQDepth = 0;
+    let modalCount = 0;
+    let arithCount = 0;
+    let setCount = 0;
+    let parenDepth = 0;
+    let maxParenDepth = 0;
+    // Skeleton: primeros 16 tokens categorizados
+    const skeleton = [];
+    for (const ch of formula) {
+        if (ch === '(') {
+            parenDepth++;
+            if (parenDepth > maxParenDepth)
+                maxParenDepth = parenDepth;
+        }
+        else if (ch === ')') {
+            parenDepth = Math.max(0, parenDepth - 1);
+        }
+        if (CONNECTIVE_SYMBOLS.includes(ch)) {
+            connectiveFreq[ch] = (connectiveFreq[ch] ?? 0) + 1;
+            if (QUANTIFIER_OPS.has(ch)) {
+                currentQDepth++;
+                if (currentQDepth > quantifierDepth)
+                    quantifierDepth = currentQDepth;
+                if (skeleton.length < 16)
+                    skeleton.push('q');
+            }
+            else if (MODAL_OPS.has(ch)) {
+                modalCount++;
+                if (skeleton.length < 16)
+                    skeleton.push('m');
+            }
+            else if (ARITH_OPS.has(ch)) {
+                arithCount++;
+                if (skeleton.length < 16)
+                    skeleton.push('n');
+            }
+            else if (SET_OPS.has(ch)) {
+                setCount++;
+                if (skeleton.length < 16)
+                    skeleton.push('s');
+            }
+            else {
+                if (skeleton.length < 16)
+                    skeleton.push('c');
+            }
+        }
+    }
+    // Cuantificadores: restamos al salir del scope aproximado
+    // (simplificado: depth = max alcanzada, no tracking exacto)
+    // Extraer átomos con regex
+    let m;
+    IDENT_RE.lastIndex = 0;
+    while ((m = IDENT_RE.exec(formula)) !== null) {
+        const tok = m[0];
+        if (!LOGIC_RESERVED.has(tok)) {
+            atomFreq[tok] = (atomFreq[tok] ?? 0) + 1;
+            if (skeleton.length < 16)
+                skeleton.push('a');
+        }
+    }
+    // Aritmética también detecta dígitos
+    for (const ch of formula) {
+        if (ARITH_OPS.has(ch)) {
+            arithCount++;
+            if (skeleton.length < 16)
+                skeleton.push('n');
+        }
+    }
+    const skeletonHash = fnv1a(skeleton.join(''));
+    return {
+        connectiveFreq,
+        atomFreq,
+        quantifierDepth,
+        modalCount,
+        arithCount,
+        setCount,
+        totalLen: formula.length,
+        parenDepth: maxParenDepth,
+        skeletonHash,
+    };
+}
+// --------------- Construcción del vector embedding ---------------------
+/**
+ * Construye un embedding R^256 determinístico a partir de la fórmula.
+ * El vector NO está normalizado aquí — la normalización se hace en
+ * `normalizeEmbedding`.
+ */
+function buildRawEmbedding(features, formula) {
+    const vec = new Float32Array(types_1.EMBEDDING_DIM);
+    // ---- Bloque 1: conectivos (dim 0..15) ----
+    // Cada conectivo tiene una posición fija
+    const connectiveSlots = {
+        '∧': 0, '∨': 1, '¬': 2, '→': 3, '↔': 4,
+        '□': 5, '◇': 6, '∀': 7, '∃': 8, '=': 9,
+        '⊥': 10, '⊤': 11, '⊢': 12, '⇒': 13,
+    };
+    for (const [sym, slot] of Object.entries(connectiveSlots)) {
+        const freq = features.connectiveFreq[sym] ?? 0;
+        if (freq > 0)
+            vec[slot] = (vec[slot] ?? 0) + Math.log1p(freq);
+    }
+    // ---- Bloque 2: átomos hasheados (dim 16..127) ----
+    for (const [atom, freq] of Object.entries(features.atomFreq)) {
+        // Hash doble para reducir colisiones
+        const idx1 = 16 + (strToIndex(atom, 56));
+        const idx2 = 72 + (strToIndex(atom, 56));
+        const idx3 = 16 + (strToIndexSalted(atom, 'b', 56));
+        vec[idx1] = (vec[idx1] ?? 0) + Math.log1p(freq) * 0.5;
+        vec[idx2] = (vec[idx2] ?? 0) + Math.log1p(freq) * 0.3;
+        vec[idx3] = (vec[idx3] ?? 0) + Math.log1p(freq) * 0.2;
+    }
+    // ---- Bloque 3: profundidad y métricas estructurales (dim 128..143) ----
+    vec[128] = features.quantifierDepth * 0.5;
+    vec[129] = Math.log1p(features.modalCount);
+    vec[130] = Math.log1p(features.arithCount);
+    vec[131] = Math.log1p(features.setCount);
+    vec[132] = Math.log1p(features.totalLen) * 0.1;
+    vec[133] = features.parenDepth * 0.3;
+    // ---- Bloque 4: firma de esqueleto (dim 144..159) ----
+    // Esparcimos el hash del esqueleto en 16 posiciones con salts distintos
+    for (let i = 0; i < 16; i++) {
+        const pos = 144 + i;
+        const contrib = Math.sin(features.skeletonHash * (i + 1)) * 0.5 + 0.5;
+        vec[pos] = (vec[pos] ?? 0) + contrib;
+    }
+    // ---- Bloque 5: bigrams de conectivos hasheados (dim 160..191) ----
+    // Captura patrones co-ocurrencia entre pares de conectivos
+    const connKeys = Object.keys(features.connectiveFreq);
+    for (let i = 0; i < connKeys.length; i++) {
+        for (let j = i + 1; j < connKeys.length; j++) {
+            const pair = (connKeys[i] ?? '') + (connKeys[j] ?? '');
+            const pairIdx = 160 + (fnv1a(pair) % 32);
+            const freq = Math.min(features.connectiveFreq[connKeys[i] ?? ''] ?? 0, features.connectiveFreq[connKeys[j] ?? ''] ?? 0);
+            vec[pairIdx] = (vec[pairIdx] ?? 0) + Math.log1p(freq) * 0.4;
+        }
+    }
+    // ---- Bloque 6: n-grams de caracteres de la fórmula (dim 192..255) ----
+    // Captura la "textura" local de la fórmula
+    for (let i = 0; i < formula.length - 1; i++) {
+        const bigram = formula[i] + formula[i + 1];
+        const pos = 192 + (fnv1a(bigram) % 64);
+        vec[pos] = (vec[pos] ?? 0) + 0.1;
+    }
+    return vec;
+}
+/** Normaliza un vector a norma L2 = 1. Si la norma es 0 devuelve el vector de ceros. */
+function normalizeEmbedding(vec) {
+    let norm = 0;
+    for (let i = 0; i < vec.length; i++)
+        norm += (vec[i] ?? 0) * (vec[i] ?? 0);
+    if (norm === 0)
+        return vec;
+    const invNorm = 1 / Math.sqrt(norm);
+    const out = new Float32Array(vec.length);
+    for (let i = 0; i < vec.length; i++)
+        out[i] = (vec[i] ?? 0) * invNorm;
+    return out;
+}
+/**
+ * Genera el embedding de una fórmula/texto.
+ *
+ * El embedding es:
+ *   1. Determinístico (misma entrada → mismo vector)
+ *   2. Normalizado L2
+ *   3. Sin NaN/Infinity
+ *
+ * Para queries de texto libre (no fórmulas puras), el texto se trata
+ * igual — los átomos y operadores se extraen con las mismas reglas.
+ */
+function hashEmbed(text) {
+    const features = extractFeatures(text);
+    const raw = buildRawEmbedding(features, text);
+    return normalizeEmbedding(raw);
+}
+/**
+ * EmbeddingProvider basado en HashEmbedding.
+ * Implementa la interfaz EmbeddingProvider para Stage 2.
+ */
+class HashEmbeddingProvider {
+    dim = types_1.EMBEDDING_DIM;
+    embed(text) {
+        return hashEmbed(text);
+    }
+}
+exports.HashEmbeddingProvider = HashEmbeddingProvider;
+/** Instancia singleton reutilizable. */
+exports.defaultProvider = new HashEmbeddingProvider();
+//# sourceMappingURL=embedding.js.map

package/dist/reasoning/lemma-rag/embedding.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"embedding.js","sourceRoot":"","sources":["../../../src/reasoning/lemma-rag/embedding.ts"],"names":[],"mappings":";AAAA,+DAA+D;AAC/D,oEAAoE;AACpE,EAAE;AACF,kEAAkE;AAClE,0DAA0D;AAC1D,qEAAqE;AACrE,gEAAgE;AAChE,4CAA4C;AAC5C,iEAAiE;AACjE,6DAA6D;AAC7D,EAAE;AACF,4BAA4B;AAC5B,mDAAmD;AACnD,+CAA+C;AAC/C,0CAA0C;AAC1C,uBAAuB;AACvB,+DAA+D;;;AAwN/D,gDAQC;AAaD,8BAIC;AA/OD,mCAAwC;AAGxC,0EAA0E;AAE1E,SAAS,KAAK,CAAC,GAAW;IACxB,IAAI,IAAI,GAAG,UAAU,CAAC;IACtB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,GAAG,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QACpC,IAAI,IAAI,GAAG,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC;QAC1B,IAAI,GAAG,CAAC,IAAI,GAAG,QAAQ,CAAC,KAAK,CAAC,CAAC;IACjC,CAAC;IACD,OAAO,IAAI,CAAC;AACd,CAAC;AAED,+CAA+C;AAC/C,SAAS,UAAU,CAAC,GAAW,EAAE,GAAW;IAC1C,OAAO,KAAK,CAAC,GAAG,CAAC,GAAG,GAAG,CAAC;AAC1B,CAAC;AAED,iEAAiE;AACjE,SAAS,gBAAgB,CAAC,GAAW,EAAE,IAAY,EAAE,GAAW;IAC9D,OAAO,KAAK,CAAC,IAAI,GAAG,GAAG,CAAC,GAAG,GAAG,CAAC;AACjC,CAAC;AAED,0EAA0E;AAE1E,MAAM,kBAAkB,GAAG,CAAC,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,CAAC,CAAC;AAClG,MAAM,SAAS,GAAG,IAAI,GAAG,CAAC,CAAC,GAAG,EAAE,GAAG,CAAC,CAAC,CAAC;AACtC,MAAM,cAAc,GAAG,IAAI,GAAG,CAAC,CAAC,GAAG,EAAE,GAAG,CAAC,CAAC,CAAC;AAC3C,MAAM,SAAS,GAAG,IAAI,GAAG,CAAC,CAAC,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,CAAC,CAAC,CAAC;AAC/D,MAAM,OAAO,GAAG,IAAI,GAAG,CAAC,CAAC,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,CAAC,CAAC,CAAC;AAC7D,MAAM,QAAQ,GAAG,yBAAyB,CAAC;AAC3C,MAAM,cAAc,GAAG,IAAI,GAAG,CAAC;IAC7B,KAAK,EAAE,KAAK,EAAE,IAAI,EAAE,SAAS,EAAE,KAAK,EAAE,QAAQ,EAAE,QAAQ;IACxD,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,OAAO;CAClD,CAAC,CAAC;AAuBH,SAAS,eAAe,CAAC,OAAe;IACtC,MAAM,cAAc,GAA2B,EAAE,CAAC;IAClD,MAAM,QAAQ,GAA2B,EAAE,CAAC;IAC5C,IAAI,eAAe,GAAG,CAAC,CAAC;IACxB,IAAI,aAAa,GAAG,CAAC,CAAC;IACtB,IAAI,UAAU,GAAG,CAAC,CAAC;IACnB,IAAI,UAAU,GAAG,CAAC,CAAC;IACnB,IAAI,QAAQ,GAAG,CAAC,CAAC;IACjB,IAAI,UAAU,GAAG,CAAC,CAAC;IACnB,IAAI,aAAa,GAAG,CAAC,CAAC;IAEtB,6CAA6C;IAC7C,MAAM,QAAQ,GAAa,EAAE,CAAC;IAE9B,KAAK,MAAM,EAAE,IAAI,OAAO,EAAE,CAAC;QACzB,IAAI,EAAE,KAAK,GAAG,EAAE,CAAC;YACf,UAAU,EAAE,CAAC;YACb,IAAI,UAAU,GAAG,aAAa;gBAAE,aAAa,GAAG,UAAU,CAAC;QAC7D,CAAC;aAAM,IAAI,EAAE,KAAK,GAAG,EAAE,CAAC;YACtB,UAAU,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,UAAU,GAAG,CAAC,CAAC,CAAC;QAC3C,CAAC;QAED,IAAI,kBAAkB,CAAC,QAAQ,CAAC,EAAE,CAAC,EAAE,CAAC;YACpC,cAAc,CAAC,EAAE,CAAC,GAAG,CAAC,cAAc,CAAC,EAAE,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC;YACnD,IAAI,cAAc,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,CAAC;gBAC3B,aAAa,EAAE,CAAC;gBAChB,IAAI,aAAa,GAAG,eAAe;oBAAE,eAAe,GAAG,aAAa,CAAC;gBACrE,IAAI,QAAQ,CAAC,MAAM,GAAG,EAAE;oBAAE,QAAQ,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;YAC/C,CAAC;iBAAM,IAAI,SAAS,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,CAAC;gBAC7B,UAAU,EAAE,CAAC;gBACb,IAAI,QAAQ,CAAC,MAAM,GAAG,EAAE;oBAAE,QAAQ,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;YAC/C,CAAC;iBAAM,IAAI,SAAS,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,CAAC;gBAC7B,UAAU,EAAE,CAAC;gBACb,IAAI,QAAQ,CAAC,MAAM,GAAG,EAAE;oBAAE,QAAQ,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;YAC/C,CAAC;iBAAM,IAAI,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,CAAC;gBAC3B,QAAQ,EAAE,CAAC;gBACX,IAAI,QAAQ,CAAC,MAAM,GAAG,EAAE;oBAAE,QAAQ,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;YAC/C,CAAC;iBAAM,CAAC;gBACN,IAAI,QAAQ,CAAC,MAAM,GAAG,EAAE;oBAAE,QAAQ,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;YAC/C,CAAC;QACH,CAAC;IACH,CAAC;IAED,0DAA0D;IAC1D,4DAA4D;IAE5D,2BAA2B;IAC3B,IAAI,CAAyB,CAAC;IAC9B,QAAQ,CAAC,SAAS,GAAG,CAAC,CAAC;IACvB,OAAO,CAAC,CAAC,GAAG,QAAQ,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,KAAK,IAAI,EAAE,CAAC;QAC7C,MAAM,GAAG,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC;QACjB,IAAI,CAAC,cAAc,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC;YAC7B,QAAQ,CAAC,GAAG,CAAC,GAAG,CAAC,QAAQ,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC;YACzC,IAAI,QAAQ,CAAC,MAAM,GAAG,EAAE;gBAAE,QAAQ,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QAC/C,CAAC;IACH,CAAC;IAED,qCAAqC;IACrC,KAAK,MAAM,EAAE,IAAI,OAAO,EAAE,CAAC;QACzB,IAAI,SAAS,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,CAAC;YACtB,UAAU,EAAE,CAAC;YACb,IAAI,QAAQ,CAAC,MAAM,GAAG,EAAE;gBAAE,QAAQ,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QAC/C,CAAC;IACH,CAAC;IAED,MAAM,YAAY,GAAG,KAAK,CAAC,QAAQ,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,CAAC;IAE9C,OAAO;QACL,cAAc;QACd,QAAQ;QACR,eAAe;QACf,UAAU;QACV,UAAU;QACV,QAAQ;QACR,QAAQ,EAAE,OAAO,CAAC,MAAM;QACxB,UAAU,EAAE,aAAa;QACzB,YAAY;KACb,CAAC;AACJ,CAAC;AAED,0EAA0E;AAE1E;;;;GAIG;AACH,SAAS,iBAAiB,CAAC,QAAqB,EAAE,OAAe;IAC/D,MAAM,GAAG,GAAG,IAAI,YAAY,CAAC,qBAAa,CAAC,CAAC;IAE5C,6CAA6C;IAC7C,yCAAyC;IACzC,MAAM,eAAe,GAA2B;QAC9C,GAAG,EAAE,CAAC,EAAE,GAAG,EAAE,CAAC,EAAE,GAAG,EAAE,CAAC,EAAE,GAAG,EAAE,CAAC,EAAE,GAAG,EAAE,CAAC;QACtC,GAAG,EAAE,CAAC,EAAE,GAAG,EAAE,CAAC,EAAE,GAAG,EAAE,CAAC,EAAE,GAAG,EAAE,CAAC,EAAE,GAAG,EAAE,CAAC;QACtC,GAAG,EAAE,EAAE,EAAE,GAAG,EAAE,EAAE,EAAE,GAAG,EAAE,EAAE,EAAE,GAAG,EAAE,EAAE;KACnC,CAAC;IACF,KAAK,MAAM,CAAC,GAAG,EAAE,IAAI,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,eAAe,CAAC,EAAE,CAAC;QAC1D,MAAM,IAAI,GAAG,QAAQ,CAAC,cAAc,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;QAC/C,IAAI,IAAI,GAAG,CAAC;YAAE,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;IAChE,CAAC;IAED,qDAAqD;IACrD,KAAK,MAAM,CAAC,IAAI,EAAE,IAAI,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAC,EAAE,CAAC;QAC7D,qCAAqC;QACrC,MAAM,IAAI,GAAG,EAAE,GAAG,CAAC,UAAU,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC,CAAC;QACzC,MAAM,IAAI,GAAG,EAAE,GAAG,CAAC,UAAU,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC,CAAC;QACzC,MAAM,IAAI,GAAG,EAAE,GAAG,CAAC,gBAAgB,CAAC,IAAI,EAAE,GAAG,EAAE,EAAE,CAAC,CAAC,CAAC;QACpD,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,GAAG,GAAG,CAAC;QACtD,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,GAAG,GAAG,CAAC;QACtD,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,GAAG,GAAG,CAAC;IACxD,CAAC;IAED,0EAA0E;IAC1E,GAAG,CAAC,GAAG,CAAC,GAAG,QAAQ,CAAC,eAAe,GAAG,GAAG,CAAC;IAC1C,GAAG,CAAC,GAAG,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,UAAU,CAAC,CAAC;IAC3C,GAAG,CAAC,GAAG,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,UAAU,CAAC,CAAC;IAC3C,GAAG,CAAC,GAAG,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC;IACzC,GAAG,CAAC,GAAG,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,QAAQ,CAAC,GAAG,GAAG,CAAC;IAC/C,GAAG,CAAC,GAAG,CAAC,GAAG,QAAQ,CAAC,UAAU,GAAG,GAAG,CAAC;IAErC,wDAAwD;IACxD,wEAAwE;IACxE,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,EAAE,EAAE,CAAC,EAAE,EAAE,CAAC;QAC5B,MAAM,GAAG,GAAG,GAAG,GAAG,CAAC,CAAC;QACpB,MAAM,OAAO,GAAG,IAAI,CAAC,GAAG,CAAC,QAAQ,CAAC,YAAY,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,GAAG,GAAG,GAAG,CAAC;QACtE,GAAG,CAAC,GAAG,CAAC,GAAG,CAAC,GAAG,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,GAAG,OAAO,CAAC;IACvC,CAAC;IAED,qEAAqE;IACrE,2DAA2D;IAC3D,MAAM,QAAQ,GAAG,MAAM,CAAC,IAAI,CAAC,QAAQ,CAAC,cAAc,CAAC,CAAC;IACtD,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,QAAQ,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QACzC,KAAK,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,QAAQ,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;YAC7C,MAAM,IAAI,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC;YACvD,MAAM,OAAO,GAAG,GAAG,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,GAAG,EAAE,CAAC,CAAC;YACzC,MAAM,IAAI,GAAG,IAAI,CAAC,GAAG,CACnB,QAAQ,CAAC,cAAc,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,IAAI,CAAC,EAC/C,QAAQ,CAAC,cAAc,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,IAAI,CAAC,CAChD,CAAC;YACF,GAAG,CAAC,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,GAAG,GAAG,CAAC;QAC9D,CAAC;IACH,CAAC;IAED,yEAAyE;IACzE,2CAA2C;IAC3C,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,OAAO,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;QAC5C,MAAM,MAAM,GAAG,OAAO,CAAC,CAAC,CAAE,GAAG,OAAO,CAAC,CAAC,GAAG,CAAC,CAAE,CAAC;QAC7C,MAAM,GAAG,GAAG,GAAG,GAAG,CAAC,KAAK,CAAC,MAAM,CAAC,GAAG,EAAE,CAAC,CAAC;QACvC,GAAG,CAAC,GAAG,CAAC,GAAG,CAAC,GAAG,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,GAAG,GAAG,CAAC;IACnC,CAAC;IAED,OAAO,GAAG,CAAC;AACb,CAAC;AAED,wFAAwF;AACxF,SAAgB,kBAAkB,CAAC,GAAiB;IAClD,IAAI,IAAI,GAAG,CAAC,CAAC;IACb,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,GAAG,CAAC,MAAM,EAAE,CAAC,EAAE;QAAE,IAAI,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC;IAC3E,IAAI,IAAI,KAAK,CAAC;QAAE,OAAO,GAAG,CAAC;IAC3B,MAAM,OAAO,GAAG,CAAC,GAAG,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IACpC,MAAM,GAAG,GAAG,IAAI,YAAY,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC;IACzC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,GAAG,CAAC,MAAM,EAAE,CAAC,EAAE;QAAE,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,GAAG,OAAO,CAAC;IACtE,OAAO,GAAG,CAAC;AACb,CAAC;AAED;;;;;;;;;;GAUG;AACH,SAAgB,SAAS,CAAC,IAAY;IACpC,MAAM,QAAQ,GAAG,eAAe,CAAC,IAAI,CAAC,CAAC;IACvC,MAAM,GAAG,GAAG,iBAAiB,CAAC,QAAQ,EAAE,IAAI,CAAC,CAAC;IAC9C,OAAO,kBAAkB,CAAC,GAAG,CAAC,CAAC;AACjC,CAAC;AAED;;;GAGG;AACH,MAAa,qBAAqB;IACvB,GAAG,GAAG,qBAAa,CAAC;IAE7B,KAAK,CAAC,IAAY;QAChB,OAAO,SAAS,CAAC,IAAI,CAAC,CAAC;IACzB,CAAC;CACF;AAND,sDAMC;AAED,wCAAwC;AAC3B,QAAA,eAAe,GAAG,IAAI,qBAAqB,EAAE,CAAC"}

package/dist/reasoning/lemma-rag/index-store.d.ts

@@ -0,0 +1,35 @@
+import type { CuratedLemma } from '../../tooling/lemma-library/types';
+import type { EmbeddedLemma, Embedding, EmbeddingProvider } from './types';
+export declare class IndexStore {
+    private readonly provider;
+    private entries;
+    private byId;
+    /** inverted index: token → set of lemma IDs (para BM25). */
+    private invertedIndex;
+    /** document frequency: token → número de documentos que lo contienen. */
+    private docFreq;
+    constructor(provider: EmbeddingProvider);
+    /** Añade un lema al índice. Idempotente si ya existe (mismo id). */
+    add(lemma: CuratedLemma): EmbeddedLemma;
+    /** Añade múltiples lemas. */
+    addAll(lemmas: CuratedLemma[]): void;
+    /** Devuelve todos los EmbeddedLemma. */
+    all(): EmbeddedLemma[];
+    /** Busca por ID. */
+    getById(id: string): EmbeddedLemma | undefined;
+    /** Número de lemas indexados. */
+    size(): number;
+    /** Devuelve el document frequency de un token (para BM25 IDF). */
+    getDocFreq(token: string): number;
+    /** Total de documentos. */
+    totalDocs(): number;
+    /** Average document length (para BM25). */
+    avgDocLength(): number;
+    /** Filtra los entries por dominio. */
+    byDomain(domain: string): EmbeddedLemma[];
+    /** Devuelve los IDs de lemas que contienen el token (inverted index). */
+    docsForToken(token: string): Set<string>;
+    /** Devuelve el embedding de un vector de consulta generado por el provider. */
+    embedQuery(text: string): Embedding;
+}
+//# sourceMappingURL=index-store.d.ts.map

package/dist/reasoning/lemma-rag/index-store.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"index-store.d.ts","sourceRoot":"","sources":["../../../src/reasoning/lemma-rag/index-store.ts"],"names":[],"mappings":"AAOA,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,mCAAmC,CAAC;AACtE,OAAO,KAAK,EAAE,aAAa,EAAE,SAAS,EAAE,iBAAiB,EAAE,MAAM,SAAS,CAAC;AAiB3E,qBAAa,UAAU;IAQT,OAAO,CAAC,QAAQ,CAAC,QAAQ;IAPrC,OAAO,CAAC,OAAO,CAAuB;IACtC,OAAO,CAAC,IAAI,CAAyC;IACrD,4DAA4D;IAC5D,OAAO,CAAC,aAAa,CAAuC;IAC5D,yEAAyE;IACzE,OAAO,CAAC,OAAO,CAAkC;gBAEpB,QAAQ,EAAE,iBAAiB;IAExD,oEAAoE;IACpE,GAAG,CAAC,KAAK,EAAE,YAAY,GAAG,aAAa;IAyBvC,6BAA6B;IAC7B,MAAM,CAAC,MAAM,EAAE,YAAY,EAAE,GAAG,IAAI;IAIpC,wCAAwC;IACxC,GAAG,IAAI,aAAa,EAAE;IAItB,oBAAoB;IACpB,OAAO,CAAC,EAAE,EAAE,MAAM,GAAG,aAAa,GAAG,SAAS;IAI9C,iCAAiC;IACjC,IAAI,IAAI,MAAM;IAId,kEAAkE;IAClE,UAAU,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM;IAIjC,2BAA2B;IAC3B,SAAS,IAAI,MAAM;IAInB,2CAA2C;IAC3C,YAAY,IAAI,MAAM;IAOtB,sCAAsC;IACtC,QAAQ,CAAC,MAAM,EAAE,MAAM,GAAG,aAAa,EAAE;IAIzC,yEAAyE;IACzE,YAAY,CAAC,KAAK,EAAE,MAAM,GAAG,GAAG,CAAC,MAAM,CAAC;IAIxC,+EAA+E;IAC/E,UAAU,CAAC,IAAI,EAAE,MAAM,GAAG,SAAS;CAGpC"}

package/dist/reasoning/lemma-rag/index-store.js

@@ -0,0 +1,105 @@
+"use strict";
+// ============================================================
+// IndexStore — índice in-memory de lemas embebidos
+//
+// Almacena EmbeddedLemma[] + inverted index de tokens para BM25.
+// Soporta inserción, borrado y búsqueda por embedding o por token.
+// ============================================================
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.IndexStore = void 0;
+const tokenize_1 = require("../../tooling/lemma-library/tokenize");
+/** Construye el term-frequency bag para un lema (name + tags + statement). */
+function buildTermFreq(lemma) {
+    const tokens = [
+        ...(0, tokenize_1.tokenize)(lemma.name),
+        ...lemma.tags.flatMap((t) => (0, tokenize_1.tokenize)(t)),
+        ...(0, tokenize_1.tokenize)(lemma.statement),
+    ];
+    const termFreq = new Map();
+    for (const tok of tokens) {
+        termFreq.set(tok, (termFreq.get(tok) ?? 0) + 1);
+    }
+    return { termFreq, docLength: tokens.length };
+}
+class IndexStore {
+    provider;
+    entries = [];
+    byId = new Map();
+    /** inverted index: token → set of lemma IDs (para BM25). */
+    invertedIndex = new Map();
+    /** document frequency: token → número de documentos que lo contienen. */
+    docFreq = new Map();
+    constructor(provider) {
+        this.provider = provider;
+    }
+    /** Añade un lema al índice. Idempotente si ya existe (mismo id). */
+    add(lemma) {
+        if (this.byId.has(lemma.id)) {
+            return this.byId.get(lemma.id);
+        }
+        const embedding = this.provider.embed(lemma.statement);
+        const { termFreq, docLength } = buildTermFreq(lemma);
+        const entry = { lemma, embedding, termFreq, docLength };
+        this.entries.push(entry);
+        this.byId.set(lemma.id, entry);
+        // Actualizar inverted index + docFreq
+        for (const tok of termFreq.keys()) {
+            let bucket = this.invertedIndex.get(tok);
+            if (!bucket) {
+                bucket = new Set();
+                this.invertedIndex.set(tok, bucket);
+            }
+            bucket.add(lemma.id);
+            this.docFreq.set(tok, (this.docFreq.get(tok) ?? 0) + 1);
+        }
+        return entry;
+    }
+    /** Añade múltiples lemas. */
+    addAll(lemmas) {
+        for (const l of lemmas)
+            this.add(l);
+    }
+    /** Devuelve todos los EmbeddedLemma. */
+    all() {
+        return this.entries;
+    }
+    /** Busca por ID. */
+    getById(id) {
+        return this.byId.get(id);
+    }
+    /** Número de lemas indexados. */
+    size() {
+        return this.entries.length;
+    }
+    /** Devuelve el document frequency de un token (para BM25 IDF). */
+    getDocFreq(token) {
+        return this.docFreq.get(token) ?? 0;
+    }
+    /** Total de documentos. */
+    totalDocs() {
+        return this.entries.length;
+    }
+    /** Average document length (para BM25). */
+    avgDocLength() {
+        if (this.entries.length === 0)
+            return 1;
+        let total = 0;
+        for (const e of this.entries)
+            total += e.docLength;
+        return total / this.entries.length;
+    }
+    /** Filtra los entries por dominio. */
+    byDomain(domain) {
+        return this.entries.filter((e) => e.lemma.domain === domain);
+    }
+    /** Devuelve los IDs de lemas que contienen el token (inverted index). */
+    docsForToken(token) {
+        return this.invertedIndex.get(token) ?? new Set();
+    }
+    /** Devuelve el embedding de un vector de consulta generado por el provider. */
+    embedQuery(text) {
+        return this.provider.embed(text);
+    }
+}
+exports.IndexStore = IndexStore;
+//# sourceMappingURL=index-store.js.map

package/dist/reasoning/lemma-rag/index-store.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index-store.js","sourceRoot":"","sources":["../../../src/reasoning/lemma-rag/index-store.ts"],"names":[],"mappings":";AAAA,+DAA+D;AAC/D,mDAAmD;AACnD,EAAE;AACF,iEAAiE;AACjE,mEAAmE;AACnE,+DAA+D;;;AAI/D,mEAAgE;AAEhE,8EAA8E;AAC9E,SAAS,aAAa,CAAC,KAAmB;IACxC,MAAM,MAAM,GAAG;QACb,GAAG,IAAA,mBAAQ,EAAC,KAAK,CAAC,IAAI,CAAC;QACvB,GAAG,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,IAAA,mBAAQ,EAAC,CAAC,CAAC,CAAC;QACzC,GAAG,IAAA,mBAAQ,EAAC,KAAK,CAAC,SAAS,CAAC;KAC7B,CAAC;IACF,MAAM,QAAQ,GAAG,IAAI,GAAG,EAAkB,CAAC;IAC3C,KAAK,MAAM,GAAG,IAAI,MAAM,EAAE,CAAC;QACzB,QAAQ,CAAC,GAAG,CAAC,GAAG,EAAE,CAAC,QAAQ,CAAC,GAAG,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IAClD,CAAC;IACD,OAAO,EAAE,QAAQ,EAAE,SAAS,EAAE,MAAM,CAAC,MAAM,EAAE,CAAC;AAChD,CAAC;AAED,MAAa,UAAU;IAQQ;IAPrB,OAAO,GAAoB,EAAE,CAAC;IAC9B,IAAI,GAA+B,IAAI,GAAG,EAAE,CAAC;IACrD,4DAA4D;IACpD,aAAa,GAA6B,IAAI,GAAG,EAAE,CAAC;IAC5D,yEAAyE;IACjE,OAAO,GAAwB,IAAI,GAAG,EAAE,CAAC;IAEjD,YAA6B,QAA2B;QAA3B,aAAQ,GAAR,QAAQ,CAAmB;IAAG,CAAC;IAE5D,oEAAoE;IACpE,GAAG,CAAC,KAAmB;QACrB,IAAI,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC,EAAE,CAAC;YAC5B,OAAO,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAE,CAAC;QAClC,CAAC;QACD,MAAM,SAAS,GAAG,IAAI,CAAC,QAAQ,CAAC,KAAK,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC;QACvD,MAAM,EAAE,QAAQ,EAAE,SAAS,EAAE,GAAG,aAAa,CAAC,KAAK,CAAC,CAAC;QACrD,MAAM,KAAK,GAAkB,EAAE,KAAK,EAAE,SAAS,EAAE,QAAQ,EAAE,SAAS,EAAE,CAAC;QAEvE,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;QACzB,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,EAAE,KAAK,CAAC,CAAC;QAE/B,sCAAsC;QACtC,KAAK,MAAM,GAAG,IAAI,QAAQ,CAAC,IAAI,EAAE,EAAE,CAAC;YAClC,IAAI,MAAM,GAAG,IAAI,CAAC,aAAa,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;YACzC,IAAI,CAAC,MAAM,EAAE,CAAC;gBACZ,MAAM,GAAG,IAAI,GAAG,EAAE,CAAC;gBACnB,IAAI,CAAC,aAAa,CAAC,GAAG,CAAC,GAAG,EAAE,MAAM,CAAC,CAAC;YACtC,CAAC;YACD,MAAM,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;YACrB,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,GAAG,EAAE,CAAC,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;QAC1D,CAAC;QAED,OAAO,KAAK,CAAC;IACf,CAAC;IAED,6BAA6B;IAC7B,MAAM,CAAC,MAAsB;QAC3B,KAAK,MAAM,CAAC,IAAI,MAAM;YAAE,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;IACtC,CAAC;IAED,wCAAwC;IACxC,GAAG;QACD,OAAO,IAAI,CAAC,OAAO,CAAC;IACtB,CAAC;IAED,oBAAoB;IACpB,OAAO,CAAC,EAAU;QAChB,OAAO,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;IAC3B,CAAC;IAED,iCAAiC;IACjC,IAAI;QACF,OAAO,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC;IAC7B,CAAC;IAED,kEAAkE;IAClE,UAAU,CAAC,KAAa;QACtB,OAAO,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;IACtC,CAAC;IAED,2BAA2B;IAC3B,SAAS;QACP,OAAO,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC;IAC7B,CAAC;IAED,2CAA2C;IAC3C,YAAY;QACV,IAAI,IAAI,CAAC,OAAO,CAAC,MAAM,KAAK,CAAC;YAAE,OAAO,CAAC,CAAC;QACxC,IAAI,KAAK,GAAG,CAAC,CAAC;QACd,KAAK,MAAM,CAAC,IAAI,IAAI,CAAC,OAAO;YAAE,KAAK,IAAI,CAAC,CAAC,SAAS,CAAC;QACnD,OAAO,KAAK,GAAG,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC;IACrC,CAAC;IAED,sCAAsC;IACtC,QAAQ,CAAC,MAAc;QACrB,OAAO,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC,MAAM,KAAK,MAAM,CAAC,CAAC;IAC/D,CAAC;IAED,yEAAyE;IACzE,YAAY,CAAC,KAAa;QACxB,OAAO,IAAI,CAAC,aAAa,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,IAAI,GAAG,EAAE,CAAC;IACpD,CAAC;IAED,+EAA+E;IAC/E,UAAU,CAAC,IAAY;QACrB,OAAO,IAAI,CAAC,QAAQ,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;IACnC,CAAC;CACF;AAxFD,gCAwFC"}

package/dist/reasoning/lemma-rag/index.d.ts

@@ -0,0 +1,9 @@
+export type { Embedding, EmbeddedLemma, QueryResult, RAGOptions, EmbeddingProvider, RecallResult } from './types';
+export { EMBEDDING_DIM } from './types';
+export { hashEmbed, normalizeEmbedding, HashEmbeddingProvider, defaultProvider } from './embedding';
+export { IndexStore } from './index-store';
+export { cosineSimilarity, bm25Score, hybridScore, retrieveTopK } from './retrieval';
+export { LemmaRAG, queryLemmas, globalRAG, resetGlobalRAG } from './query';
+export { runBenchmark, formatBenchmarkResults, BENCHMARK_PAIRS, } from './benchmarks';
+export type { BenchmarkPair } from './benchmarks';
+//# sourceMappingURL=index.d.ts.map

package/dist/reasoning/lemma-rag/index.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/reasoning/lemma-rag/index.ts"],"names":[],"mappings":"AAQA,YAAY,EAAE,SAAS,EAAE,aAAa,EAAE,WAAW,EAAE,UAAU,EAAE,iBAAiB,EAAE,YAAY,EAAE,MAAM,SAAS,CAAC;AAClH,OAAO,EAAE,aAAa,EAAE,MAAM,SAAS,CAAC;AAExC,OAAO,EAAE,SAAS,EAAE,kBAAkB,EAAE,qBAAqB,EAAE,eAAe,EAAE,MAAM,aAAa,CAAC;AAEpG,OAAO,EAAE,UAAU,EAAE,MAAM,eAAe,CAAC;AAE3C,OAAO,EAAE,gBAAgB,EAAE,SAAS,EAAE,WAAW,EAAE,YAAY,EAAE,MAAM,aAAa,CAAC;AAErF,OAAO,EAAE,QAAQ,EAAE,WAAW,EAAE,SAAS,EAAE,cAAc,EAAE,MAAM,SAAS,CAAC;AAE3E,OAAO,EACL,YAAY,EACZ,sBAAsB,EACtB,eAAe,GAChB,MAAM,cAAc,CAAC;AACtB,YAAY,EAAE,aAAa,EAAE,MAAM,cAAc,CAAC"}

package/dist/reasoning/lemma-rag/index.js

@@ -0,0 +1,34 @@
+"use strict";
+// ============================================================
+// lemma-rag — entrypoint público
+//
+// Sistema RAG semántico sobre la biblioteca de lemas ST.
+// Stage 1: HashEmbedding determinístico R^256 (sin deps de red).
+// Stage 2 (pendiente): ONNX transformers (MiniLM-L6).
+// ============================================================
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.BENCHMARK_PAIRS = exports.formatBenchmarkResults = exports.runBenchmark = exports.resetGlobalRAG = exports.globalRAG = exports.queryLemmas = exports.LemmaRAG = exports.retrieveTopK = exports.hybridScore = exports.bm25Score = exports.cosineSimilarity = exports.IndexStore = exports.defaultProvider = exports.HashEmbeddingProvider = exports.normalizeEmbedding = exports.hashEmbed = exports.EMBEDDING_DIM = void 0;
+var types_1 = require("./types");
+Object.defineProperty(exports, "EMBEDDING_DIM", { enumerable: true, get: function () { return types_1.EMBEDDING_DIM; } });
+var embedding_1 = require("./embedding");
+Object.defineProperty(exports, "hashEmbed", { enumerable: true, get: function () { return embedding_1.hashEmbed; } });
+Object.defineProperty(exports, "normalizeEmbedding", { enumerable: true, get: function () { return embedding_1.normalizeEmbedding; } });
+Object.defineProperty(exports, "HashEmbeddingProvider", { enumerable: true, get: function () { return embedding_1.HashEmbeddingProvider; } });
+Object.defineProperty(exports, "defaultProvider", { enumerable: true, get: function () { return embedding_1.defaultProvider; } });
+var index_store_1 = require("./index-store");
+Object.defineProperty(exports, "IndexStore", { enumerable: true, get: function () { return index_store_1.IndexStore; } });
+var retrieval_1 = require("./retrieval");
+Object.defineProperty(exports, "cosineSimilarity", { enumerable: true, get: function () { return retrieval_1.cosineSimilarity; } });
+Object.defineProperty(exports, "bm25Score", { enumerable: true, get: function () { return retrieval_1.bm25Score; } });
+Object.defineProperty(exports, "hybridScore", { enumerable: true, get: function () { return retrieval_1.hybridScore; } });
+Object.defineProperty(exports, "retrieveTopK", { enumerable: true, get: function () { return retrieval_1.retrieveTopK; } });
+var query_1 = require("./query");
+Object.defineProperty(exports, "LemmaRAG", { enumerable: true, get: function () { return query_1.LemmaRAG; } });
+Object.defineProperty(exports, "queryLemmas", { enumerable: true, get: function () { return query_1.queryLemmas; } });
+Object.defineProperty(exports, "globalRAG", { enumerable: true, get: function () { return query_1.globalRAG; } });
+Object.defineProperty(exports, "resetGlobalRAG", { enumerable: true, get: function () { return query_1.resetGlobalRAG; } });
+var benchmarks_1 = require("./benchmarks");
+Object.defineProperty(exports, "runBenchmark", { enumerable: true, get: function () { return benchmarks_1.runBenchmark; } });
+Object.defineProperty(exports, "formatBenchmarkResults", { enumerable: true, get: function () { return benchmarks_1.formatBenchmarkResults; } });
+Object.defineProperty(exports, "BENCHMARK_PAIRS", { enumerable: true, get: function () { return benchmarks_1.BENCHMARK_PAIRS; } });
+//# sourceMappingURL=index.js.map

package/dist/reasoning/lemma-rag/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../../src/reasoning/lemma-rag/index.ts"],"names":[],"mappings":";AAAA,+DAA+D;AAC/D,iCAAiC;AACjC,EAAE;AACF,yDAAyD;AACzD,iEAAiE;AACjE,sDAAsD;AACtD,+DAA+D;;;AAG/D,iCAAwC;AAA/B,sGAAA,aAAa,OAAA;AAEtB,yCAAoG;AAA3F,sGAAA,SAAS,OAAA;AAAE,+GAAA,kBAAkB,OAAA;AAAE,kHAAA,qBAAqB,OAAA;AAAE,4GAAA,eAAe,OAAA;AAE9E,6CAA2C;AAAlC,yGAAA,UAAU,OAAA;AAEnB,yCAAqF;AAA5E,6GAAA,gBAAgB,OAAA;AAAE,sGAAA,SAAS,OAAA;AAAE,wGAAA,WAAW,OAAA;AAAE,yGAAA,YAAY,OAAA;AAE/D,iCAA2E;AAAlE,iGAAA,QAAQ,OAAA;AAAE,oGAAA,WAAW,OAAA;AAAE,kGAAA,SAAS,OAAA;AAAE,uGAAA,cAAc,OAAA;AAEzD,2CAIsB;AAHpB,0GAAA,YAAY,OAAA;AACZ,oHAAA,sBAAsB,OAAA;AACtB,6GAAA,eAAe,OAAA"}

package/dist/reasoning/lemma-rag/query.d.ts

@@ -0,0 +1,48 @@
+import type { CuratedLemma } from '../../tooling/lemma-library/types';
+import type { EmbeddingProvider, QueryResult, RAGOptions } from './types';
+import { IndexStore } from './index-store';
+/**
+ * LemmaRAG — instancia principal del sistema RAG.
+ *
+ * Flujo de uso:
+ *   const rag = new LemmaRAG();
+ *   rag.index(lemmas);
+ *   const results = rag.query('negation implies double negation', { k: 3 });
+ */
+export declare class LemmaRAG {
+    private store;
+    constructor(provider?: EmbeddingProvider);
+    /** Indexa una lista de lemas. Idempotente por id. */
+    index(lemmas: CuratedLemma[]): void;
+    /** Número de lemas indexados. */
+    size(): number;
+    /**
+     * Realiza una búsqueda semántica híbrida.
+     *
+     * Flujo:
+     *   1. Embed(query) → query_vec
+     *   2. Cosine similarity contra todos los vectores en el store
+     *   3. BM25 sobre los tokens de la query
+     *   4. Score híbrido: 0.7 * cosine + 0.3 * bm25_norm
+     *   5. Top-k con score ≥ minScore
+     */
+    query(queryText: string, opts?: RAGOptions): QueryResult[];
+    /** Acceso directo al IndexStore (para tests y benchmarks). */
+    getStore(): IndexStore;
+}
+/**
+ * Retorna (creando si necesario) la instancia global del RAG.
+ * Útil para uso quick-start sin gestionar el ciclo de vida.
+ */
+export declare function globalRAG(): LemmaRAG;
+/** Resetea la instancia global (útil en tests). */
+export declare function resetGlobalRAG(): void;
+/**
+ * Función de conveniencia: indexa lemas en la instancia global y
+ * ejecuta la query. Si `rag` no se pasa, usa la global.
+ *
+ * @example
+ *   const results = queryLemmas(standardLibrary().all(), 'double negation');
+ */
+export declare function queryLemmas(lemmas: CuratedLemma[], queryText: string, opts?: RAGOptions, rag?: LemmaRAG): QueryResult[];
+//# sourceMappingURL=query.d.ts.map

package/dist/reasoning/lemma-rag/query.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"query.d.ts","sourceRoot":"","sources":["../../../src/reasoning/lemma-rag/query.ts"],"names":[],"mappings":"AAQA,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,mCAAmC,CAAC;AACtE,OAAO,KAAK,EAAE,iBAAiB,EAAE,WAAW,EAAE,UAAU,EAAE,MAAM,SAAS,CAAC;AAC1E,OAAO,EAAE,UAAU,EAAE,MAAM,eAAe,CAAC;AAO3C;;;;;;;GAOG;AACH,qBAAa,QAAQ;IACnB,OAAO,CAAC,KAAK,CAAa;gBAEd,QAAQ,GAAE,iBAAmC;IAIzD,qDAAqD;IACrD,KAAK,CAAC,MAAM,EAAE,YAAY,EAAE,GAAG,IAAI;IAInC,iCAAiC;IACjC,IAAI,IAAI,MAAM;IAId;;;;;;;;;OASG;IACH,KAAK,CAAC,SAAS,EAAE,MAAM,EAAE,IAAI,GAAE,UAAe,GAAG,WAAW,EAAE;IAY9D,8DAA8D;IAC9D,QAAQ,IAAI,UAAU;CAGvB;AAMD;;;GAGG;AACH,wBAAgB,SAAS,IAAI,QAAQ,CAGpC;AAED,mDAAmD;AACnD,wBAAgB,cAAc,IAAI,IAAI,CAErC;AAED;;;;;;GAMG;AACH,wBAAgB,WAAW,CACzB,MAAM,EAAE,YAAY,EAAE,EACtB,SAAS,EAAE,MAAM,EACjB,IAAI,GAAE,UAAe,EACrB,GAAG,CAAC,EAAE,QAAQ,GACb,WAAW,EAAE,CAIf"}

package/dist/reasoning/lemma-rag/query.js

@@ -0,0 +1,92 @@
+"use strict";
+// ============================================================
+// query.ts — API pública del sistema RAG
+//
+// Expone:
+//   - LemmaRAG: clase principal que encapsula store + retrieval
+//   - queryLemmas: función de conveniencia stateless (instancia lazy)
+// ============================================================
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.LemmaRAG = void 0;
+exports.globalRAG = globalRAG;
+exports.resetGlobalRAG = resetGlobalRAG;
+exports.queryLemmas = queryLemmas;
+const index_store_1 = require("./index-store");
+const retrieval_1 = require("./retrieval");
+const embedding_1 = require("./embedding");
+const DEFAULT_K = 5;
+const DEFAULT_COSINE_WEIGHT = 0.7;
+/**
+ * LemmaRAG — instancia principal del sistema RAG.
+ *
+ * Flujo de uso:
+ *   const rag = new LemmaRAG();
+ *   rag.index(lemmas);
+ *   const results = rag.query('negation implies double negation', { k: 3 });
+ */
+class LemmaRAG {
+    store;
+    constructor(provider = embedding_1.defaultProvider) {
+        this.store = new index_store_1.IndexStore(provider);
+    }
+    /** Indexa una lista de lemas. Idempotente por id. */
+    index(lemmas) {
+        this.store.addAll(lemmas);
+    }
+    /** Número de lemas indexados. */
+    size() {
+        return this.store.size();
+    }
+    /**
+     * Realiza una búsqueda semántica híbrida.
+     *
+     * Flujo:
+     *   1. Embed(query) → query_vec
+     *   2. Cosine similarity contra todos los vectores en el store
+     *   3. BM25 sobre los tokens de la query
+     *   4. Score híbrido: 0.7 * cosine + 0.3 * bm25_norm
+     *   5. Top-k con score ≥ minScore
+     */
+    query(queryText, opts = {}) {
+        const k = opts.k ?? DEFAULT_K;
+        const cosineWeight = opts.cosineWeight ?? DEFAULT_COSINE_WEIGHT;
+        const minScore = opts.minScore ?? 0;
+        if (this.store.size() === 0)
+            return [];
+        const queryVec = this.store.embedQuery(queryText);
+        return (0, retrieval_1.retrieveTopK)(queryText, queryVec, this.store, k, cosineWeight, opts.domain, minScore);
+    }
+    /** Acceso directo al IndexStore (para tests y benchmarks). */
+    getStore() {
+        return this.store;
+    }
+}
+exports.LemmaRAG = LemmaRAG;
+// --------------- Instancia lazy global (stateful) ----------------------
+let _globalRAG;
+/**
+ * Retorna (creando si necesario) la instancia global del RAG.
+ * Útil para uso quick-start sin gestionar el ciclo de vida.
+ */
+function globalRAG() {
+    if (!_globalRAG)
+        _globalRAG = new LemmaRAG();
+    return _globalRAG;
+}
+/** Resetea la instancia global (útil en tests). */
+function resetGlobalRAG() {
+    _globalRAG = undefined;
+}
+/**
+ * Función de conveniencia: indexa lemas en la instancia global y
+ * ejecuta la query. Si `rag` no se pasa, usa la global.
+ *
+ * @example
+ *   const results = queryLemmas(standardLibrary().all(), 'double negation');
+ */
+function queryLemmas(lemmas, queryText, opts = {}, rag) {
+    const instance = rag ?? new LemmaRAG();
+    instance.index(lemmas);
+    return instance.query(queryText, opts);
+}
+//# sourceMappingURL=query.js.map

package/dist/reasoning/lemma-rag/query.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"query.js","sourceRoot":"","sources":["../../../src/reasoning/lemma-rag/query.ts"],"names":[],"mappings":";AAAA,+DAA+D;AAC/D,yCAAyC;AACzC,EAAE;AACF,UAAU;AACV,gEAAgE;AAChE,sEAAsE;AACtE,+DAA+D;;;AAwE/D,8BAGC;AAGD,wCAEC;AASD,kCASC;AA9FD,+CAA2C;AAC3C,2CAA2C;AAC3C,2CAA8C;AAE9C,MAAM,SAAS,GAAG,CAAC,CAAC;AACpB,MAAM,qBAAqB,GAAG,GAAG,CAAC;AAElC;;;;;;;GAOG;AACH,MAAa,QAAQ;IACX,KAAK,CAAa;IAE1B,YAAY,WAA8B,2BAAe;QACvD,IAAI,CAAC,KAAK,GAAG,IAAI,wBAAU,CAAC,QAAQ,CAAC,CAAC;IACxC,CAAC;IAED,qDAAqD;IACrD,KAAK,CAAC,MAAsB;QAC1B,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;IAC5B,CAAC;IAED,iCAAiC;IACjC,IAAI;QACF,OAAO,IAAI,CAAC,KAAK,CAAC,IAAI,EAAE,CAAC;IAC3B,CAAC;IAED;;;;;;;;;OASG;IACH,KAAK,CAAC,SAAiB,EAAE,OAAmB,EAAE;QAC5C,MAAM,CAAC,GAAG,IAAI,CAAC,CAAC,IAAI,SAAS,CAAC;QAC9B,MAAM,YAAY,GAAG,IAAI,CAAC,YAAY,IAAI,qBAAqB,CAAC;QAChE,MAAM,QAAQ,GAAG,IAAI,CAAC,QAAQ,IAAI,CAAC,CAAC;QAEpC,IAAI,IAAI,CAAC,KAAK,CAAC,IAAI,EAAE,KAAK,CAAC;YAAE,OAAO,EAAE,CAAC;QAEvC,MAAM,QAAQ,GAAG,IAAI,CAAC,KAAK,CAAC,UAAU,CAAC,SAAS,CAAC,CAAC;QAElD,OAAO,IAAA,wBAAY,EAAC,SAAS,EAAE,QAAQ,EAAE,IAAI,CAAC,KAAK,EAAE,CAAC,EAAE,YAAY,EAAE,IAAI,CAAC,MAAM,EAAE,QAAQ,CAAC,CAAC;IAC/F,CAAC;IAED,8DAA8D;IAC9D,QAAQ;QACN,OAAO,IAAI,CAAC,KAAK,CAAC;IACpB,CAAC;CACF;AA3CD,4BA2CC;AAED,0EAA0E;AAE1E,IAAI,UAAgC,CAAC;AAErC;;;GAGG;AACH,SAAgB,SAAS;IACvB,IAAI,CAAC,UAAU;QAAE,UAAU,GAAG,IAAI,QAAQ,EAAE,CAAC;IAC7C,OAAO,UAAU,CAAC;AACpB,CAAC;AAED,mDAAmD;AACnD,SAAgB,cAAc;IAC5B,UAAU,GAAG,SAAS,CAAC;AACzB,CAAC;AAED;;;;;;GAMG;AACH,SAAgB,WAAW,CACzB,MAAsB,EACtB,SAAiB,EACjB,OAAmB,EAAE,EACrB,GAAc;IAEd,MAAM,QAAQ,GAAG,GAAG,IAAI,IAAI,QAAQ,EAAE,CAAC;IACvC,QAAQ,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;IACvB,OAAO,QAAQ,CAAC,KAAK,CAAC,SAAS,EAAE,IAAI,CAAC,CAAC;AACzC,CAAC"}

package/dist/reasoning/lemma-rag/retrieval.d.ts

@@ -0,0 +1,33 @@
+import type { Embedding, EmbeddedLemma, QueryResult } from './types';
+import type { IndexStore } from './index-store';
+/**
+ * Cosine similarity entre dos vectores NORMALIZADOS (norma L2 = 1).
+ * En ese caso, es simplemente el producto punto.
+ * Devuelve un valor en [-1, 1], saturado a [0, 1] para usarse como score.
+ */
+export declare function cosineSimilarity(a: Embedding, b: Embedding): number;
+/**
+ * BM25 Okapi score de una query contra un documento en el store.
+ * Devuelve un score ≥ 0.
+ */
+export declare function bm25Score(queryText: string, doc: EmbeddedLemma, store: IndexStore): number;
+/**
+ * Score híbrido: `cosineWeight * cosine + (1 - cosineWeight) * bm25_norm`.
+ *
+ * BM25 se normaliza dividiendo por `maxBm25` (máximo observado en el batch).
+ * Si maxBm25 === 0, el componente BM25 aporta 0.
+ */
+export declare function hybridScore(cosine: number, bm25: number, maxBm25: number, cosineWeight: number): number;
+/**
+ * Recupera los top-k lemas del store para una query dada.
+ *
+ * @param queryText  - Texto de la query (fórmula o descripción)
+ * @param queryVec   - Embedding pre-calculado de queryText
+ * @param store      - IndexStore con los lemas indexados
+ * @param k          - Número de resultados
+ * @param cosineWeight - Peso del componente cosine (default 0.7)
+ * @param domain     - Filtro opcional de dominio
+ * @param minScore   - Score mínimo para incluir
+ */
+export declare function retrieveTopK(queryText: string, queryVec: Embedding, store: IndexStore, k: number, cosineWeight: number, domain?: string, minScore?: number): QueryResult[];
+//# sourceMappingURL=retrieval.d.ts.map