elid 0.1.0 → 0.3.0

package/elid_bg.js

@@ -180,6 +180,273 @@ function handleError(f, args) {
         wasm.__wbindgen_exn_store(idx);
     }
 }
+/**
+ * Compute the OSA (Optimal String Alignment) distance between two strings.
+ *
+ * Similar to Levenshtein but also considers transpositions as a single operation.
+ *
+ * # JavaScript Example
+ *
+ * ```javascript
+ * import { osaDistance } from 'elid';
+ *
+ * const distance = osaDistance("ca", "ac");
+ * console.log(distance); // 1 (transposition)
+ * ```
+ * @param {string} a
+ * @param {string} b
+ * @returns {number}
+ */
+export function osaDistance(a, b) {
+    const ptr0 = passStringToWasm0(a, wasm.__wbindgen_malloc, wasm.__wbindgen_realloc);
+    const len0 = WASM_VECTOR_LEN;
+    const ptr1 = passStringToWasm0(b, wasm.__wbindgen_malloc, wasm.__wbindgen_realloc);
+    const len1 = WASM_VECTOR_LEN;
+    const ret = wasm.osaDistance(ptr0, len0, ptr1, len1);
+    return ret >>> 0;
+}
+
+/**
+ * Compute the Levenshtein distance between two strings.
+ *
+ * Returns the minimum number of single-character edits needed to transform one string into another.
+ *
+ * # JavaScript Example
+ *
+ * ```javascript
+ * import { levenshtein } from 'elid';
+ *
+ * const distance = levenshtein("kitten", "sitting");
+ * console.log(distance); // 3
+ * ```
+ * @param {string} a
+ * @param {string} b
+ * @returns {number}
+ */
+export function levenshtein(a, b) {
+    const ptr0 = passStringToWasm0(a, wasm.__wbindgen_malloc, wasm.__wbindgen_realloc);
+    const len0 = WASM_VECTOR_LEN;
+    const ptr1 = passStringToWasm0(b, wasm.__wbindgen_malloc, wasm.__wbindgen_realloc);
+    const len1 = WASM_VECTOR_LEN;
+    const ret = wasm.levenshtein(ptr0, len0, ptr1, len1);
+    return ret >>> 0;
+}
+
+function passArrayJsValueToWasm0(array, malloc) {
+    const ptr = malloc(array.length * 4, 4) >>> 0;
+    for (let i = 0; i < array.length; i++) {
+        const add = addToExternrefTable0(array[i]);
+        getDataViewMemory0().setUint32(ptr + 4 * i, add, true);
+    }
+    WASM_VECTOR_LEN = array.length;
+    return ptr;
+}
+/**
+ * Find all matches above a threshold score.
+ *
+ * Returns an array of objects with index and score for all candidates above the threshold.
+ *
+ * # JavaScript Example
+ *
+ * ```javascript
+ * import { findMatchesAboveThreshold } from 'elid';
+ *
+ * const candidates = ["apple", "application", "apply", "banana"];
+ * const matches = findMatchesAboveThreshold("app", candidates, 0.5);
+ * console.log(matches); // [{ index: 0, score: 0.907 }, { index: 1, score: 0.830 }, ...]
+ * ```
+ * @param {string} query
+ * @param {string[]} candidates
+ * @param {number} threshold
+ * @returns {any}
+ */
+export function findMatchesAboveThreshold(query, candidates, threshold) {
+    const ptr0 = passStringToWasm0(query, wasm.__wbindgen_malloc, wasm.__wbindgen_realloc);
+    const len0 = WASM_VECTOR_LEN;
+    const ptr1 = passArrayJsValueToWasm0(candidates, wasm.__wbindgen_malloc);
+    const len1 = WASM_VECTOR_LEN;
+    const ret = wasm.findMatchesAboveThreshold(ptr0, len0, ptr1, len1, threshold);
+    return ret;
+}
+
+/**
+ * Compute the Hamming distance between two SimHash values.
+ *
+ * Returns the number of differing bits. Lower values = higher similarity.
+ *
+ * # JavaScript Example
+ *
+ * ```javascript
+ * import { simhash, simhashDistance } from 'elid';
+ *
+ * const hash1 = simhash("iPhone 14");
+ * const hash2 = simhash("iPhone 15");
+ * const distance = simhashDistance(hash1, hash2);
+ *
+ * console.log(distance); // Low number = similar
+ * ```
+ * @param {number} hash1
+ * @param {number} hash2
+ * @returns {number}
+ */
+export function simhashDistance(hash1, hash2) {
+    const ret = wasm.simhashDistance(hash1, hash2);
+    return ret >>> 0;
+}
+
+let cachedFloat64ArrayMemory0 = null;
+
+function getFloat64ArrayMemory0() {
+    if (cachedFloat64ArrayMemory0 === null || cachedFloat64ArrayMemory0.byteLength === 0) {
+        cachedFloat64ArrayMemory0 = new Float64Array(wasm.memory.buffer);
+    }
+    return cachedFloat64ArrayMemory0;
+}
+
+function passArrayF64ToWasm0(arg, malloc) {
+    const ptr = malloc(arg.length * 8, 8) >>> 0;
+    getFloat64ArrayMemory0().set(arg, ptr / 8);
+    WASM_VECTOR_LEN = arg.length;
+    return ptr;
+}
+
+function takeFromExternrefTable0(idx) {
+    const value = wasm.__wbindgen_externrefs.get(idx);
+    wasm.__externref_table_dealloc(idx);
+    return value;
+}
+/**
+ * Encode an embedding with a maximum output string length constraint.
+ *
+ * Calculates the optimal precision and dimension settings to fit within
+ * the specified character limit while maximizing fidelity.
+ *
+ * # Parameters
+ *
+ * - `embedding`: Float64 array of embedding values (64-2048 dimensions)
+ * - `max_chars`: Maximum output string length in characters
+ *
+ * # Returns
+ *
+ * A base32hex-encoded ELID string guaranteed to be <= max_chars in length.
+ *
+ * # JavaScript Example
+ *
+ * ```javascript
+ * import { encodeElidMaxLength } from 'elid';
+ *
+ * const embedding = new Float64Array(768).fill(0.1);
+ *
+ * // Fit in 100 characters (e.g., for database column constraints)
+ * const elid = encodeElidMaxLength(embedding, 100);
+ * console.log(elid.length <= 100); // true
+ *
+ * // Fit in 50 characters (more compression)
+ * const shortElid = encodeElidMaxLength(embedding, 50);
+ * ```
+ * @param {Float64Array} embedding
+ * @param {number} max_chars
+ * @returns {string}
+ */
+export function encodeElidMaxLength(embedding, max_chars) {
+    let deferred3_0;
+    let deferred3_1;
+    try {
+        const ptr0 = passArrayF64ToWasm0(embedding, wasm.__wbindgen_malloc);
+        const len0 = WASM_VECTOR_LEN;
+        const ret = wasm.encodeElidMaxLength(ptr0, len0, max_chars);
+        var ptr2 = ret[0];
+        var len2 = ret[1];
+        if (ret[3]) {
+            ptr2 = 0; len2 = 0;
+            throw takeFromExternrefTable0(ret[2]);
+        }
+        deferred3_0 = ptr2;
+        deferred3_1 = len2;
+        return getStringFromWasm0(ptr2, len2);
+    } finally {
+        wasm.__wbindgen_free(deferred3_0, deferred3_1, 1);
+    }
+}
+
+function _assertClass(instance, klass) {
+    if (!(instance instanceof klass)) {
+        throw new Error(`expected instance of ${klass.name}`);
+    }
+}
+/**
+ * Compute Levenshtein distance with custom options.
+ *
+ * # JavaScript Example
+ *
+ * ```javascript
+ * import { levenshteinWithOpts, SimilarityOptions } from 'elid';
+ *
+ * const opts = new SimilarityOptions();
+ * opts.setCaseSensitive(false);
+ * opts.setTrimWhitespace(true);
+ *
+ * const distance = levenshteinWithOpts("  HELLO  ", "hello", opts);
+ * console.log(distance); // 0
+ * ```
+ * @param {string} a
+ * @param {string} b
+ * @param {SimilarityOptions} opts
+ * @returns {number}
+ */
+export function levenshteinWithOpts(a, b, opts) {
+    const ptr0 = passStringToWasm0(a, wasm.__wbindgen_malloc, wasm.__wbindgen_realloc);
+    const len0 = WASM_VECTOR_LEN;
+    const ptr1 = passStringToWasm0(b, wasm.__wbindgen_malloc, wasm.__wbindgen_realloc);
+    const len1 = WASM_VECTOR_LEN;
+    _assertClass(opts, SimilarityOptions);
+    var ptr2 = opts.__destroy_into_raw();
+    const ret = wasm.levenshteinWithOpts(ptr0, len0, ptr1, len1, ptr2);
+    return ret >>> 0;
+}
+
+/**
+ * Check if an ELID can be decoded back to an embedding.
+ *
+ * Returns true if the ELID was encoded with a FullVector profile
+ * (lossless, compressed, or max_length), false otherwise.
+ *
+ * # Parameters
+ *
+ * - `elid_str`: A valid ELID string (base32hex encoded)
+ *
+ * # Returns
+ *
+ * `true` if decodeElidToEmbedding will return an embedding, `false` otherwise.
+ *
+ * # JavaScript Example
+ *
+ * ```javascript
+ * import { encodeElid, encodeElidLossless, isElidReversible, ElidProfile } from 'elid';
+ *
+ * const embedding = new Float64Array(768).fill(0.1);
+ *
+ * // Mini128 is NOT reversible
+ * const mini128Elid = encodeElid(embedding, ElidProfile.Mini128);
+ * console.log(isElidReversible(mini128Elid)); // false
+ *
+ * // Lossless IS reversible
+ * const losslessElid = encodeElidLossless(embedding);
+ * console.log(isElidReversible(losslessElid)); // true
+ * ```
+ * @param {string} elid_str
+ * @returns {boolean}
+ */
+export function isElidReversible(elid_str) {
+    const ptr0 = passStringToWasm0(elid_str, wasm.__wbindgen_malloc, wasm.__wbindgen_realloc);
+    const len0 = WASM_VECTOR_LEN;
+    const ret = wasm.isElidReversible(ptr0, len0);
+    if (ret[2]) {
+        throw takeFromExternrefTable0(ret[1]);
+    }
+    return ret[0] !== 0;
+}
+
 /**
  * Compute the SimHash fingerprint of a string.
  *
@@ -211,173 +478,481 @@ export function simhash(text) {
     return ret;
 }
 
+function getArrayU8FromWasm0(ptr, len) {
+    ptr = ptr >>> 0;
+    return getUint8ArrayMemory0().subarray(ptr / 1, ptr / 1 + len);
+}
 /**
- * Compute the normalized Levenshtein similarity between two strings.
+ * Decode an ELID string to raw bytes.
  *
- * Returns a value between 0.0 (completely different) and 1.0 (identical).
+ * Returns the raw byte representation of an ELID, including the header
+ * and payload bytes. Useful for custom processing or debugging.
+ *
+ * # Parameters
+ *
+ * - `elid_str`: A valid ELID string (base32hex encoded)
+ *
+ * # Returns
+ *
+ * A Uint8Array containing the raw bytes (header + payload).
  *
  * # JavaScript Example
  *
  * ```javascript
- * import { normalizedLevenshtein } from 'elid';
+ * import { decodeElid } from 'elid';
  *
- * const similarity = normalizedLevenshtein("hello", "hallo");
- * console.log(similarity); // ~0.8
+ * const bytes = decodeElid("012345abcdef...");
+ * console.log(bytes); // Uint8Array [...]
  * ```
- * @param {string} a
- * @param {string} b
+ * @param {string} elid_str
+ * @returns {Uint8Array}
+ */
+export function decodeElid(elid_str) {
+    const ptr0 = passStringToWasm0(elid_str, wasm.__wbindgen_malloc, wasm.__wbindgen_realloc);
+    const len0 = WASM_VECTOR_LEN;
+    const ret = wasm.decodeElid(ptr0, len0);
+    if (ret[3]) {
+        throw takeFromExternrefTable0(ret[2]);
+    }
+    var v2 = getArrayU8FromWasm0(ret[0], ret[1]).slice();
+    wasm.__wbindgen_free(ret[0], ret[1] * 1, 1);
+    return v2;
+}
+
+/**
+ * Get metadata about a FullVector ELID.
+ *
+ * Returns an object containing information about how the ELID was encoded,
+ * including original dimensions, precision, and dimension mode.
+ *
+ * # Parameters
+ *
+ * - `elid_str`: A valid ELID string (base32hex encoded)
+ *
+ * # Returns
+ *
+ * An object with metadata fields, or null if not a FullVector ELID.
+ *
+ * # JavaScript Example
+ *
+ * ```javascript
+ * import { encodeElidCompressed, getElidMetadata } from 'elid';
+ *
+ * const embedding = new Float64Array(768).fill(0.1);
+ * const elid = encodeElidCompressed(embedding, 0.5);
+ *
+ * const meta = getElidMetadata(elid);
+ * if (meta) {
+ *     console.log(meta.originalDims);  // 768
+ *     console.log(meta.encodedDims);   // depends on compression
+ *     console.log(meta.isLossless);    // false
+ * }
+ * ```
+ * @param {string} elid_str
+ * @returns {any}
+ */
+export function getElidMetadata(elid_str) {
+    const ptr0 = passStringToWasm0(elid_str, wasm.__wbindgen_malloc, wasm.__wbindgen_realloc);
+    const len0 = WASM_VECTOR_LEN;
+    const ret = wasm.getElidMetadata(ptr0, len0);
+    if (ret[2]) {
+        throw takeFromExternrefTable0(ret[1]);
+    }
+    return takeFromExternrefTable0(ret[0]);
+}
+
+/**
+ * Encode an embedding for cross-dimensional comparison.
+ *
+ * Projects the embedding to a common dimension space, allowing comparison
+ * between embeddings of different original dimensions (e.g., 256d vs 768d).
+ *
+ * # Parameters
+ *
+ * - `embedding`: Float64 array of embedding values (64-2048 dimensions)
+ * - `common_dims`: Target dimension space (all vectors projected here)
+ *
+ * # Returns
+ *
+ * A base32hex-encoded ELID string.
+ *
+ * # JavaScript Example
+ *
+ * ```javascript
+ * import { encodeElidCrossDimensional, decodeElidToEmbedding } from 'elid';
+ *
+ * // Different sized embeddings from different models
+ * const embedding256 = new Float64Array(256).fill(0.1);
+ * const embedding768 = new Float64Array(768).fill(0.1);
+ *
+ * // Project both to 128-dim common space
+ * const elid1 = encodeElidCrossDimensional(embedding256, 128);
+ * const elid2 = encodeElidCrossDimensional(embedding768, 128);
+ *
+ * // Now they can be compared directly (both decode to 128 dims)
+ * const dec1 = decodeElidToEmbedding(elid1);
+ * const dec2 = decodeElidToEmbedding(elid2);
+ * // Both have length 128
+ * ```
+ * @param {Float64Array} embedding
+ * @param {number} common_dims
+ * @returns {string}
+ */
+export function encodeElidCrossDimensional(embedding, common_dims) {
+    let deferred3_0;
+    let deferred3_1;
+    try {
+        const ptr0 = passArrayF64ToWasm0(embedding, wasm.__wbindgen_malloc);
+        const len0 = WASM_VECTOR_LEN;
+        const ret = wasm.encodeElidCrossDimensional(ptr0, len0, common_dims);
+        var ptr2 = ret[0];
+        var len2 = ret[1];
+        if (ret[3]) {
+            ptr2 = 0; len2 = 0;
+            throw takeFromExternrefTable0(ret[2]);
+        }
+        deferred3_0 = ptr2;
+        deferred3_1 = len2;
+        return getStringFromWasm0(ptr2, len2);
+    } finally {
+        wasm.__wbindgen_free(deferred3_0, deferred3_1, 1);
+    }
+}
+
+/**
+ * Compute the Hamming distance between two ELID strings.
+ *
+ * Returns the number of differing bits between two Mini128 ELIDs.
+ * This distance is proportional to the angular distance between the
+ * original embeddings (lower = more similar).
+ *
+ * # Requirements
+ *
+ * Both ELIDs must use the Mini128 profile.
+ *
+ * # Parameters
+ *
+ * - `elid1`: First ELID string
+ * - `elid2`: Second ELID string
+ *
+ * # Returns
+ *
+ * Hamming distance (0-128). 0 means identical, 128 means completely different.
+ *
+ * # JavaScript Example
+ *
+ * ```javascript
+ * import { encodeElid, elidHammingDistance, ElidProfile } from 'elid';
+ *
+ * const elid1 = encodeElid(embedding1, ElidProfile.Mini128);
+ * const elid2 = encodeElid(embedding2, ElidProfile.Mini128);
+ *
+ * const distance = elidHammingDistance(elid1, elid2);
+ * if (distance < 20) {
+ *     console.log("Very similar embeddings!");
+ * }
+ * ```
+ * @param {string} elid1
+ * @param {string} elid2
  * @returns {number}
  */
-export function normalizedLevenshtein(a, b) {
-    const ptr0 = passStringToWasm0(a, wasm.__wbindgen_malloc, wasm.__wbindgen_realloc);
+export function elidHammingDistance(elid1, elid2) {
+    const ptr0 = passStringToWasm0(elid1, wasm.__wbindgen_malloc, wasm.__wbindgen_realloc);
     const len0 = WASM_VECTOR_LEN;
-    const ptr1 = passStringToWasm0(b, wasm.__wbindgen_malloc, wasm.__wbindgen_realloc);
+    const ptr1 = passStringToWasm0(elid2, wasm.__wbindgen_malloc, wasm.__wbindgen_realloc);
     const len1 = WASM_VECTOR_LEN;
-    const ret = wasm.normalizedLevenshtein(ptr0, len0, ptr1, len1);
-    return ret;
+    const ret = wasm.elidHammingDistance(ptr0, len0, ptr1, len1);
+    if (ret[2]) {
+        throw takeFromExternrefTable0(ret[1]);
+    }
+    return ret[0] >>> 0;
 }
 
 /**
- * Compute the best matching similarity between two strings.
+ * Find the best match for a query string in an array of candidates.
  *
- * Runs multiple algorithms and returns the highest score.
+ * Returns an object with the index and similarity score of the best match.
  *
  * # JavaScript Example
  *
  * ```javascript
- * import { bestMatch } from 'elid';
+ * import { findBestMatch } from 'elid';
  *
- * const score = bestMatch("hello", "hallo");
- * console.log(score); // ~0.8
+ * const candidates = ["apple", "application", "apply"];
+ * const result = findBestMatch("app", candidates);
+ * console.log(result); // { index: 0, score: 0.907 }
  * ```
- * @param {string} a
- * @param {string} b
- * @returns {number}
+ * @param {string} query
+ * @param {string[]} candidates
+ * @returns {object}
  */
-export function bestMatch(a, b) {
-    const ptr0 = passStringToWasm0(a, wasm.__wbindgen_malloc, wasm.__wbindgen_realloc);
+export function findBestMatch(query, candidates) {
+    const ptr0 = passStringToWasm0(query, wasm.__wbindgen_malloc, wasm.__wbindgen_realloc);
     const len0 = WASM_VECTOR_LEN;
-    const ptr1 = passStringToWasm0(b, wasm.__wbindgen_malloc, wasm.__wbindgen_realloc);
+    const ptr1 = passArrayJsValueToWasm0(candidates, wasm.__wbindgen_malloc);
     const len1 = WASM_VECTOR_LEN;
-    const ret = wasm.bestMatch(ptr0, len0, ptr1, len1);
+    const ret = wasm.findBestMatch(ptr0, len0, ptr1, len1);
     return ret;
 }
 
 /**
- * Compute the Jaro-Winkler similarity between two strings.
+ * Encode an embedding with percentage-based compression.
+ *
+ * The retention percentage (0.0-1.0) controls how much information is preserved:
+ * - 1.0 = lossless (Full32 precision, all dimensions)
+ * - 0.5 = half precision and/or half dimensions
+ * - 0.25 = quarter precision and/or quarter dimensions
+ *
+ * The algorithm optimizes for dimension reduction first (which preserves
+ * more geometric relationships) before reducing precision.
+ *
+ * # Parameters
+ *
+ * - `embedding`: Float64 array of embedding values (64-2048 dimensions)
+ * - `retention_pct`: Information retention percentage (0.0-1.0)
+ *
+ * # Returns
+ *
+ * A base32hex-encoded ELID string.
+ *
+ * # JavaScript Example
+ *
+ * ```javascript
+ * import { encodeElidCompressed } from 'elid';
+ *
+ * const embedding = new Float64Array(768).fill(0.1);
+ *
+ * // 50% retention - good balance of size and fidelity
+ * const elid = encodeElidCompressed(embedding, 0.5);
+ *
+ * // 25% retention - smaller but less accurate
+ * const smallElid = encodeElidCompressed(embedding, 0.25);
+ * ```
+ * @param {Float64Array} embedding
+ * @param {number} retention_pct
+ * @returns {string}
+ */
+export function encodeElidCompressed(embedding, retention_pct) {
+    let deferred3_0;
+    let deferred3_1;
+    try {
+        const ptr0 = passArrayF64ToWasm0(embedding, wasm.__wbindgen_malloc);
+        const len0 = WASM_VECTOR_LEN;
+        const ret = wasm.encodeElidCompressed(ptr0, len0, retention_pct);
+        var ptr2 = ret[0];
+        var len2 = ret[1];
+        if (ret[3]) {
+            ptr2 = 0; len2 = 0;
+            throw takeFromExternrefTable0(ret[2]);
+        }
+        deferred3_0 = ptr2;
+        deferred3_1 = len2;
+        return getStringFromWasm0(ptr2, len2);
+    } finally {
+        wasm.__wbindgen_free(deferred3_0, deferred3_1, 1);
+    }
+}
+
+/**
+ * Compute the Jaro similarity between two strings.
  *
  * Returns a value between 0.0 (completely different) and 1.0 (identical).
- * Gives more favorable ratings to strings with common prefixes.
+ * Particularly effective for short strings like names.
  *
  * # JavaScript Example
  *
  * ```javascript
- * import { jaroWinkler } from 'elid';
+ * import { jaro } from 'elid';
  *
- * const similarity = jaroWinkler("martha", "marhta");
- * console.log(similarity); // ~0.961
+ * const similarity = jaro("martha", "marhta");
+ * console.log(similarity); // ~0.944
  * ```
  * @param {string} a
  * @param {string} b
  * @returns {number}
  */
-export function jaroWinkler(a, b) {
+export function jaro(a, b) {
     const ptr0 = passStringToWasm0(a, wasm.__wbindgen_malloc, wasm.__wbindgen_realloc);
     const len0 = WASM_VECTOR_LEN;
     const ptr1 = passStringToWasm0(b, wasm.__wbindgen_malloc, wasm.__wbindgen_realloc);
     const len1 = WASM_VECTOR_LEN;
-    const ret = wasm.jaroWinkler(ptr0, len0, ptr1, len1);
+    const ret = wasm.jaro(ptr0, len0, ptr1, len1);
     return ret;
 }
 
 /**
- * Compute the Hamming distance between two SimHash values.
+ * Encode an embedding using lossless full vector encoding.
  *
- * Returns the number of differing bits. Lower values = higher similarity.
+ * Preserves the exact embedding values (32-bit float precision) and all dimensions.
+ * This produces the largest output but allows exact reconstruction.
+ *
+ * # Parameters
+ *
+ * - `embedding`: Float64 array of embedding values (64-2048 dimensions)
+ *
+ * # Returns
+ *
+ * A base32hex-encoded ELID string that can be decoded back to the original embedding.
  *
  * # JavaScript Example
  *
  * ```javascript
- * import { simhash, simhashDistance } from 'elid';
+ * import { encodeElidLossless, decodeElidToEmbedding } from 'elid';
  *
- * const hash1 = simhash("iPhone 14");
- * const hash2 = simhash("iPhone 15");
- * const distance = simhashDistance(hash1, hash2);
+ * const embedding = new Float64Array(768).fill(0.1);
+ * const elid = encodeElidLossless(embedding);
  *
- * console.log(distance); // Low number = similar
+ * // Later, recover the exact embedding
+ * const recovered = decodeElidToEmbedding(elid);
+ * // recovered is identical to embedding
  * ```
- * @param {number} hash1
- * @param {number} hash2
- * @returns {number}
+ * @param {Float64Array} embedding
+ * @returns {string}
  */
-export function simhashDistance(hash1, hash2) {
-    const ret = wasm.simhashDistance(hash1, hash2);
-    return ret >>> 0;
+export function encodeElidLossless(embedding) {
+    let deferred3_0;
+    let deferred3_1;
+    try {
+        const ptr0 = passArrayF64ToWasm0(embedding, wasm.__wbindgen_malloc);
+        const len0 = WASM_VECTOR_LEN;
+        const ret = wasm.encodeElidLossless(ptr0, len0);
+        var ptr2 = ret[0];
+        var len2 = ret[1];
+        if (ret[3]) {
+            ptr2 = 0; len2 = 0;
+            throw takeFromExternrefTable0(ret[2]);
+        }
+        deferred3_0 = ptr2;
+        deferred3_1 = len2;
+        return getStringFromWasm0(ptr2, len2);
+    } finally {
+        wasm.__wbindgen_free(deferred3_0, deferred3_1, 1);
+    }
 }
 
 /**
- * Compute the normalized SimHash similarity between two strings.
+ * Decode an ELID string back to an embedding vector.
  *
- * Returns a value between 0.0 (completely different) and 1.0 (identical).
+ * Only works for ELIDs encoded with a FullVector profile (lossless,
+ * compressed, or max_length). Returns null for non-reversible profiles
+ * like Mini128, Morton, or Hilbert.
+ *
+ * # Parameters
+ *
+ * - `elid_str`: A valid ELID string (base32hex encoded)
+ *
+ * # Returns
+ *
+ * A Float64Array containing the decoded embedding, or null if the ELID
+ * is not reversible.
+ *
+ * Note: If dimension reduction was used during encoding, the decoded
+ * embedding will be in the reduced dimension space, not the original.
  *
  * # JavaScript Example
  *
  * ```javascript
- * import { simhashSimilarity } from 'elid';
+ * import { encodeElidLossless, decodeElidToEmbedding, isElidReversible } from 'elid';
  *
- * const similarity = simhashSimilarity("iPhone 14", "iPhone 15");
- * console.log(similarity); // ~0.9 (very similar)
+ * const embedding = new Float64Array(768).fill(0.1);
+ * const elid = encodeElidLossless(embedding);
  *
- * const similarity2 = simhashSimilarity("iPhone", "Galaxy");
- * console.log(similarity2); // ~0.4 (different)
+ * if (isElidReversible(elid)) {
+ *     const recovered = decodeElidToEmbedding(elid);
+ *     console.log(recovered.length); // 768
+ * }
  * ```
- * @param {string} a
- * @param {string} b
- * @returns {number}
+ * @param {string} elid_str
+ * @returns {any}
  */
-export function simhashSimilarity(a, b) {
-    const ptr0 = passStringToWasm0(a, wasm.__wbindgen_malloc, wasm.__wbindgen_realloc);
+export function decodeElidToEmbedding(elid_str) {
+    const ptr0 = passStringToWasm0(elid_str, wasm.__wbindgen_malloc, wasm.__wbindgen_realloc);
     const len0 = WASM_VECTOR_LEN;
-    const ptr1 = passStringToWasm0(b, wasm.__wbindgen_malloc, wasm.__wbindgen_realloc);
-    const len1 = WASM_VECTOR_LEN;
-    const ret = wasm.simhashSimilarity(ptr0, len0, ptr1, len1);
-    return ret;
+    const ret = wasm.decodeElidToEmbedding(ptr0, len0);
+    if (ret[2]) {
+        throw takeFromExternrefTable0(ret[1]);
+    }
+    return takeFromExternrefTable0(ret[0]);
 }
 
-function _assertClass(instance, klass) {
-    if (!(instance instanceof klass)) {
-        throw new Error(`expected instance of ${klass.name}`);
+/**
+ * Encode an embedding vector to an ELID string.
+ *
+ * Converts a high-dimensional embedding (64-2048 dimensions) into a compact,
+ * sortable identifier. The ELID preserves locality properties for efficient
+ * similarity search.
+ *
+ * # Parameters
+ *
+ * - `embedding`: Float64 array of embedding values (64-2048 dimensions)
+ * - `profile`: Encoding profile (Mini128, Morton10x10, or Hilbert10x10)
+ *
+ * # Returns
+ *
+ * A base32hex-encoded ELID string suitable for storage and comparison.
+ *
+ * # JavaScript Example
+ *
+ * ```javascript
+ * import { encodeElid, ElidProfile } from 'elid';
+ *
+ * // OpenAI embeddings are 1536 dimensions
+ * const embedding = await getEmbedding("Hello world");
+ * const elid = encodeElid(embedding, ElidProfile.Mini128);
+ * console.log(elid); // "012345abcdef..."
+ * ```
+ * @param {Float64Array} embedding
+ * @param {ElidProfile} profile
+ * @returns {string}
+ */
+export function encodeElid(embedding, profile) {
+    let deferred3_0;
+    let deferred3_1;
+    try {
+        const ptr0 = passArrayF64ToWasm0(embedding, wasm.__wbindgen_malloc);
+        const len0 = WASM_VECTOR_LEN;
+        const ret = wasm.encodeElid(ptr0, len0, profile);
+        var ptr2 = ret[0];
+        var len2 = ret[1];
+        if (ret[3]) {
+            ptr2 = 0; len2 = 0;
+            throw takeFromExternrefTable0(ret[2]);
+        }
+        deferred3_0 = ptr2;
+        deferred3_1 = len2;
+        return getStringFromWasm0(ptr2, len2);
+    } finally {
+        wasm.__wbindgen_free(deferred3_0, deferred3_1, 1);
     }
 }
+
 /**
- * Compute Levenshtein distance with custom options.
+ * Compute the normalized SimHash similarity between two strings.
+ *
+ * Returns a value between 0.0 (completely different) and 1.0 (identical).
  *
  * # JavaScript Example
  *
  * ```javascript
- * import { levenshteinWithOpts, SimilarityOptions } from 'elid';
+ * import { simhashSimilarity } from 'elid';
  *
- * const opts = new SimilarityOptions();
- * opts.setCaseSensitive(false);
- * opts.setTrimWhitespace(true);
+ * const similarity = simhashSimilarity("iPhone 14", "iPhone 15");
+ * console.log(similarity); // ~0.9 (very similar)
  *
- * const distance = levenshteinWithOpts("  HELLO  ", "hello", opts);
- * console.log(distance); // 0
+ * const similarity2 = simhashSimilarity("iPhone", "Galaxy");
+ * console.log(similarity2); // ~0.4 (different)
  * ```
  * @param {string} a
  * @param {string} b
- * @param {SimilarityOptions} opts
  * @returns {number}
  */
-export function levenshteinWithOpts(a, b, opts) {
+export function simhashSimilarity(a, b) {
     const ptr0 = passStringToWasm0(a, wasm.__wbindgen_malloc, wasm.__wbindgen_realloc);
     const len0 = WASM_VECTOR_LEN;
     const ptr1 = passStringToWasm0(b, wasm.__wbindgen_malloc, wasm.__wbindgen_realloc);
     const len1 = WASM_VECTOR_LEN;
-    _assertClass(opts, SimilarityOptions);
-    var ptr2 = opts.__destroy_into_raw();
-    const ret = wasm.levenshteinWithOpts(ptr0, len0, ptr1, len1, ptr2);
-    return ret >>> 0;
+    const ret = wasm.simhashSimilarity(ptr0, len0, ptr1, len1);
+    return ret;
 }
 
 /**
@@ -411,137 +986,84 @@ export function hamming(a, b) {
 }
 
 /**
- * Compute the Jaro similarity between two strings.
+ * Compute the best matching similarity between two strings.
  *
- * Returns a value between 0.0 (completely different) and 1.0 (identical).
- * Particularly effective for short strings like names.
+ * Runs multiple algorithms and returns the highest score.
  *
  * # JavaScript Example
  *
  * ```javascript
- * import { jaro } from 'elid';
+ * import { bestMatch } from 'elid';
  *
- * const similarity = jaro("martha", "marhta");
- * console.log(similarity); // ~0.944
+ * const score = bestMatch("hello", "hallo");
+ * console.log(score); // ~0.8
  * ```
  * @param {string} a
  * @param {string} b
  * @returns {number}
  */
-export function jaro(a, b) {
+export function bestMatch(a, b) {
     const ptr0 = passStringToWasm0(a, wasm.__wbindgen_malloc, wasm.__wbindgen_realloc);
     const len0 = WASM_VECTOR_LEN;
     const ptr1 = passStringToWasm0(b, wasm.__wbindgen_malloc, wasm.__wbindgen_realloc);
     const len1 = WASM_VECTOR_LEN;
-    const ret = wasm.jaro(ptr0, len0, ptr1, len1);
+    const ret = wasm.bestMatch(ptr0, len0, ptr1, len1);
     return ret;
 }
 
 /**
- * Compute the Levenshtein distance between two strings.
+ * Compute the Jaro-Winkler similarity between two strings.
  *
- * Returns the minimum number of single-character edits needed to transform one string into another.
+ * Returns a value between 0.0 (completely different) and 1.0 (identical).
+ * Gives more favorable ratings to strings with common prefixes.
  *
  * # JavaScript Example
  *
  * ```javascript
- * import { levenshtein } from 'elid';
+ * import { jaroWinkler } from 'elid';
  *
- * const distance = levenshtein("kitten", "sitting");
- * console.log(distance); // 3
+ * const similarity = jaroWinkler("martha", "marhta");
+ * console.log(similarity); // ~0.961
  * ```
  * @param {string} a
  * @param {string} b
  * @returns {number}
  */
-export function levenshtein(a, b) {
+export function jaroWinkler(a, b) {
     const ptr0 = passStringToWasm0(a, wasm.__wbindgen_malloc, wasm.__wbindgen_realloc);
     const len0 = WASM_VECTOR_LEN;
     const ptr1 = passStringToWasm0(b, wasm.__wbindgen_malloc, wasm.__wbindgen_realloc);
     const len1 = WASM_VECTOR_LEN;
-    const ret = wasm.levenshtein(ptr0, len0, ptr1, len1);
-    return ret >>> 0;
+    const ret = wasm.jaroWinkler(ptr0, len0, ptr1, len1);
+    return ret;
 }
 
 /**
- * Compute the OSA (Optimal String Alignment) distance between two strings.
+ * Compute the normalized Levenshtein similarity between two strings.
  *
- * Similar to Levenshtein but also considers transpositions as a single operation.
+ * Returns a value between 0.0 (completely different) and 1.0 (identical).
  *
  * # JavaScript Example
  *
  * ```javascript
- * import { osaDistance } from 'elid';
+ * import { normalizedLevenshtein } from 'elid';
  *
- * const distance = osaDistance("ca", "ac");
- * console.log(distance); // 1 (transposition)
+ * const similarity = normalizedLevenshtein("hello", "hallo");
+ * console.log(similarity); // ~0.8
  * ```
  * @param {string} a
  * @param {string} b
  * @returns {number}
  */
-export function osaDistance(a, b) {
+export function normalizedLevenshtein(a, b) {
     const ptr0 = passStringToWasm0(a, wasm.__wbindgen_malloc, wasm.__wbindgen_realloc);
     const len0 = WASM_VECTOR_LEN;
     const ptr1 = passStringToWasm0(b, wasm.__wbindgen_malloc, wasm.__wbindgen_realloc);
     const len1 = WASM_VECTOR_LEN;
-    const ret = wasm.osaDistance(ptr0, len0, ptr1, len1);
-    return ret >>> 0;
-}
-
-function passArrayJsValueToWasm0(array, malloc) {
-    const ptr = malloc(array.length * 4, 4) >>> 0;
-    for (let i = 0; i < array.length; i++) {
-        const add = addToExternrefTable0(array[i]);
-        getDataViewMemory0().setUint32(ptr + 4 * i, add, true);
-    }
-    WASM_VECTOR_LEN = array.length;
-    return ptr;
-}
-/**
- * Find all matches above a threshold score.
- *
- * Returns an array of objects with index and score for all candidates above the threshold.
- *
- * # JavaScript Example
- *
- * ```javascript
- * import { findMatchesAboveThreshold } from 'elid';
- *
- * const candidates = ["apple", "application", "apply", "banana"];
- * const matches = findMatchesAboveThreshold("app", candidates, 0.5);
- * console.log(matches); // [{ index: 0, score: 0.907 }, { index: 1, score: 0.830 }, ...]
- * ```
- * @param {string} query
- * @param {string[]} candidates
- * @param {number} threshold
- * @returns {any}
- */
-export function findMatchesAboveThreshold(query, candidates, threshold) {
-    const ptr0 = passStringToWasm0(query, wasm.__wbindgen_malloc, wasm.__wbindgen_realloc);
-    const len0 = WASM_VECTOR_LEN;
-    const ptr1 = passArrayJsValueToWasm0(candidates, wasm.__wbindgen_malloc);
-    const len1 = WASM_VECTOR_LEN;
-    const ret = wasm.findMatchesAboveThreshold(ptr0, len0, ptr1, len1, threshold);
+    const ret = wasm.normalizedLevenshtein(ptr0, len0, ptr1, len1);
     return ret;
 }
 
-let cachedFloat64ArrayMemory0 = null;
-
-function getFloat64ArrayMemory0() {
-    if (cachedFloat64ArrayMemory0 === null || cachedFloat64ArrayMemory0.byteLength === 0) {
-        cachedFloat64ArrayMemory0 = new Float64Array(wasm.memory.buffer);
-    }
-    return cachedFloat64ArrayMemory0;
-}
-
-function passArrayF64ToWasm0(arg, malloc) {
-    const ptr = malloc(arg.length * 8, 8) >>> 0;
-    getFloat64ArrayMemory0().set(arg, ptr / 8);
-    WASM_VECTOR_LEN = arg.length;
-    return ptr;
-}
-
 let cachedUint32ArrayMemory0 = null;
 
 function getUint32ArrayMemory0() {
@@ -588,31 +1110,98 @@ export function findSimilarHashes(query_hash, candidate_hashes, max_distance) {
 }
 
 /**
- * Find the best match for a query string in an array of candidates.
+ * Dimension handling mode for full vector encoding.
  *
- * Returns an object with the index and similarity score of the best match.
+ * Controls whether to preserve original dimensions, reduce them,
+ * or project to a common space for cross-dimensional comparison.
  *
  * # JavaScript Example
  *
  * ```javascript
- * import { findBestMatch } from 'elid';
+ * import { ElidDimensionMode, encodeElidFullVector } from 'elid';
  *
- * const candidates = ["apple", "application", "apply"];
- * const result = findBestMatch("app", candidates);
- * console.log(result); // { index: 0, score: 0.907 }
+ * // Preserve all dimensions
+ * // Reduce to fewer dimensions for smaller output
+ * // Common space for comparing different-sized embeddings
  * ```
- * @param {string} query
- * @param {string[]} candidates
- * @returns {object}
+ * @enum {0 | 1 | 2}
  */
-export function findBestMatch(query, candidates) {
-    const ptr0 = passStringToWasm0(query, wasm.__wbindgen_malloc, wasm.__wbindgen_realloc);
-    const len0 = WASM_VECTOR_LEN;
-    const ptr1 = passArrayJsValueToWasm0(candidates, wasm.__wbindgen_malloc);
-    const len1 = WASM_VECTOR_LEN;
-    const ret = wasm.findBestMatch(ptr0, len0, ptr1, len1);
-    return ret;
-}
+export const ElidDimensionMode = Object.freeze({
+    /**
+     * Preserve all original dimensions (no projection)
+     */
+    Preserve: 0, "0": "Preserve",
+    /**
+     * Reduce dimensions using random projection
+     */
+    Reduce: 1, "1": "Reduce",
+    /**
+     * Project to common space for cross-dimensional comparison
+     */
+    Common: 2, "2": "Common",
+});
+/**
+ * ELID encoding profile for vector embeddings.
+ *
+ * Choose a profile based on your use case:
+ * - `Mini128`: Fast 128-bit SimHash, good for similarity via Hamming distance
+ * - `Morton10x10`: Z-order curve encoding, good for range queries
+ * - `Hilbert10x10`: Hilbert curve encoding, best locality preservation
+ *
+ * # JavaScript Example
+ *
+ * ```javascript
+ * import { ElidProfile, encodeElid } from 'elid';
+ *
+ * const embedding = new Float64Array(768).fill(0.1);
+ * const elid = encodeElid(embedding, ElidProfile.Mini128);
+ * ```
+ * @enum {0 | 1 | 2}
+ */
+export const ElidProfile = Object.freeze({
+    /**
+     * 128-bit SimHash (cosine similarity via Hamming distance)
+     */
+    Mini128: 0, "0": "Mini128",
+    /**
+     * Morton/Z-order curve encoding (10 dims, 10 bits each)
+     */
+    Morton10x10: 1, "1": "Morton10x10",
+    /**
+     * Hilbert curve encoding (10 dims, 10 bits each)
+     */
+    Hilbert10x10: 2, "2": "Hilbert10x10",
+});
+/**
+ * Precision options for full vector encoding.
+ *
+ * Controls how many bits are used to represent each dimension value.
+ * Higher precision means more accurate reconstruction but larger output.
+ *
+ * # JavaScript Example
+ *
+ * ```javascript
+ * import { ElidVectorPrecision, encodeElidWithPrecision } from 'elid';
+ *
+ * const embedding = new Float64Array(768).fill(0.1);
+ * // Full32 = lossless, Half16 = smaller with minimal error
+ * ```
+ * @enum {0 | 1 | 2}
+ */
+export const ElidVectorPrecision = Object.freeze({
+    /**
+     * Full 32-bit float (lossless, 4 bytes per dimension)
+     */
+    Full32: 0, "0": "Full32",
+    /**
+     * 16-bit half-precision float (2 bytes per dimension)
+     */
+    Half16: 1, "1": "Half16",
+    /**
+     * 8-bit quantized (1 byte per dimension, ~1% error)
+     */
+    Quant8: 2, "2": "Quant8",
+});
 
 const SimilarityOptionsFinalization = (typeof FinalizationRegistry === 'undefined')
     ? { register: () => {}, unregister: () => {} }
@@ -757,6 +1346,11 @@ export function __wbg_new_e17d9f43105b08be() {
     return ret;
 };
 
+export function __wbg_new_with_length_cd045ed0a87d4dd6(arg0) {
+    const ret = new Float64Array(arg0 >>> 0);
+    return ret;
+};
+
 export function __wbg_set_3f1d0b984ed272ed(arg0, arg1, arg2) {
     arg0[arg1] = arg2;
 };
@@ -775,6 +1369,10 @@ export function __wbg_set_c2abbebe8b9ebee1() { return handleError(function (arg0
     return ret;
 }, arguments) };
 
+export function __wbg_set_index_a0c01b257dd824f8(arg0, arg1, arg2) {
+    arg0[arg1 >>> 0] = arg2;
+};
+
 export function __wbindgen_cast_2241b6af4c4b2941(arg0, arg1) {
     // Cast intrinsic for `Ref(String) -> Externref`.
     const ret = getStringFromWasm0(arg0, arg1);