diff --git a/src/lib/inspector/wordkey.test.ts b/src/lib/inspector/wordkey.test.ts
index 48f1bb2..ad80f00 100644
--- a/src/lib/inspector/wordkey.test.ts
+++ b/src/lib/inspector/wordkey.test.ts
@@ -8,10 +8,14 @@ import { normalizeWords, WORD_KEY_ROUNDS, wordKey } from './wordkey.ts';
 const CHAIN = '52db9ba70e0cc0f6eaf7803dd07447a1f5477735fd3f661792ba94600c84e971';
 
 describe('normalizeWords', () => {
-  it('ignores case, accents and extra spaces', () => {
-    const accented = `  ${String.fromCharCode(0xc1)}baco   ${String.fromCharCode(0xc1)}RBOL\tni${String.fromCharCode(0xf1)}o  `;
-    expect(normalizeWords(accented)).toEqual(['abaco', 'arbol', 'nino']);
+  it('ignores case, accents and extra spaces, as Go wordkey.Normalize does', () => {
+    const nbsp = String.fromCharCode(0xa0);
+    const tab = String.fromCharCode(9);
+    // The case of TestNormalize of datekeys-go: Σ in lower case on its own,
+    // not as a final sigma, and İ without its dot.
+    expect(normalizeWords(`  Ábaco${nbsp}ÁRBOL${tab}niño  ΣΑΣ  İ `)).toEqual(['abaco', 'arbol', 'nino', 'σασ', 'i']);
     expect(normalizeWords('   ')).toEqual([]);
+    expect(normalizeWords(`a${String.fromCharCode(0xfeff)}b`)).toEqual([`a${String.fromCharCode(0xfeff)}b`]);
   });
 });
 
@@ -21,6 +25,8 @@ describe('wordKey', () => {
     const key = await wordKey(words, CHAIN, 1000);
     const want = pbkdf2Sync('perro luna casa verde tren mar', `DateKeys llave de palabras v1|${CHAIN}|1000`, WORD_KEY_ROUNDS, 32, 'sha256');
     expect(Buffer.from(key).toString('hex')).toBe(want.toString('hex'));
+    // The vector of TestKeyMatchesTypeScript of datekeys-go.
+    expect(Buffer.from(key).toString('hex')).toBe('be74aecd9ea734bfece963597a2269188a4ea8a1247bc381dffbd6a38a2c6376');
     expect(Buffer.from(await wordKey(words, CHAIN, 1001)).equals(Buffer.from(key))).toBe(false);
   });
 });
diff --git a/src/lib/inspector/wordkey.ts b/src/lib/inspector/wordkey.ts
index e7a62b8..ea8ae4a 100644
--- a/src/lib/inspector/wordkey.ts
+++ b/src/lib/inspector/wordkey.ts
@@ -13,17 +13,29 @@ export const MIN_WORDS = 6;
 /** The rounds of PBKDF2-SHA256, OWASP's figure for 2023. */
 export const WORD_KEY_ROUNDS = 600_000;
 
-// The combining marks that NFD splits from accented letters.
-const MARKS = new RegExp(`[${String.fromCharCode(0x300)}-${String.fromCharCode(0x36f)}]`, 'g');
+// The white space at which the words are split: Go's unicode.IsSpace, so
+// that the CLI of the reference reads the same words.
+const SPACES = new Set([0x09, 0x0a, 0x0b, 0x0c, 0x0d, 0x20, 0x85, 0xa0, 0x1680, 0x2028, 0x2029, 0x202f, 0x205f, 0x3000]);
+for (let r = 0x2000; r <= 0x200a; r++) SPACES.add(r);
 
-/** The words of a text: in lower case, without accents, split by spaces. */
+/**
+ * The words of a text, as the CLI of the reference reads them (Go's
+ * wordkey.Normalize): its NFD without the combining marks U+0300 to U+036F,
+ * each code point in lower case on its own, split at white space.
+ */
 export function normalizeWords(text: string): string[] {
-  return text
-    .normalize('NFD')
-    .replace(MARKS, '')
-    .toLowerCase()
-    .split(/\s+/)
-    .filter((w) => w !== '');
+  const words: string[] = [];
+  let word = '';
+  for (const ch of text.normalize('NFD')) {
+    const r = ch.codePointAt(0)!;
+    if (r >= 0x300 && r <= 0x36f) continue;
+    if (SPACES.has(r)) {
+      if (word !== '') words.push(word);
+      word = '';
+    } else word += ch.toLowerCase();
+  }
+  if (word !== '') words.push(word);
+  return words;
 }
 
 /**