diff --git a/fjs/basen/proof.f.mjs b/fjs/basen/proof.f.mjs
index b65da0af8..3e0a94582 100644
--- a/fjs/basen/proof.f.mjs
+++ b/fjs/basen/proof.f.mjs
@@ -4,15 +4,10 @@ import { baseN } from './module.f.mjs'
 
 const hex = baseN(4n, '0123456789abcdef')
 
-const cb32 = baseN(5n, '0123456789abcdefghjkmnpqrstvwxyz', c => {
-    const lower = c.toLowerCase()
-    switch (lower) {
-        case 'i': { return '1' }
-        case 'l': { return '1' }
-        case 'o': { return '0' }
-        default: { return lower }
-    }
-})
+// A synthetic normalizer keeps this proof focused on `baseN`'s mechanism
+// rather than duplicating the rules owned by a concrete codec.
+const normalizedHex = baseN(4n, '0123456789abcdef', c =>
+    c === 'x' ? 'a' : c === 'y' ? 'z' : c.toLowerCase())
 
 // Sample input for the `big` proof below: 262 144 `f` characters decode into a
 // 1 Mibit (`maxLength`) vector.
@@ -48,16 +43,13 @@ export const proof = {
     },
     normalizeHit: () => {
         // 'A' lowercases to 'a' — same vector as the lowercase input.
-        const a = cb32.stringToVec('A')
-        const b = cb32.stringToVec('a')
+        const a = normalizedHex.stringToVec('A')
+        const b = normalizedHex.stringToVec('a')
         assertEq(a, b, [a, b])
-        // Crockford folds: i,l → 1 and o → 0.
-        assertEq(cb32.stringToVec('I'), cb32.stringToVec('1'), 'I→1')
-        assertEq(cb32.stringToVec('l'), cb32.stringToVec('1'), 'l→1')
-        assertEq(cb32.stringToVec('o'), cb32.stringToVec('0'), 'o→0')
+        assertEq(normalizedHex.stringToVec('x'), a, 'x→a')
     },
     normalizeMiss: () => {
-        assertEq(cb32.stringToVec('u'), null, 'unknown char should return null')
+        assertEq(normalizedHex.stringToVec('y'), null, 'normalizing to an unknown char should return null')
     },
     // Decodes a 1 Mibit hex string. With the O(n log n) `listToVec` builder this
     // runs in well under a second (was ~13 s node / ~43 s bun under the old
diff --git a/fjs/basen/todo/proof-crockford-copy.md b/fjs/basen/todo/proof-crockford-copy.md
deleted file mode 100644
index 27656126f..000000000
--- a/fjs/basen/todo/proof-crockford-copy.md
+++ /dev/null
@@ -1,57 +0,0 @@
-## proof-crockford-copy. `basen/proof.f.mjs` copies cbase32's alphabet and normalizer
-
-**Priority:** P4
-**Status:** open
-
-### Problem
-
-`fjs/basen/proof.f.mjs:7-15` re-declares the Crockford Base32 codec —
-character-for-character the alphabet and fold table that
-`fjs/basen/cbase32/module.f.mjs` owns:
-
-```js
-// fjs/basen/proof.f.mjs:7-15
-const cb32 = baseN(5n, '0123456789abcdefghjkmnpqrstvwxyz', c => {
-    const lower = c.toLowerCase()
-    switch (lower) {
-        case 'i': { return '1' }
-        case 'l': { return '1' }
-        case 'o': { return '0' }
-        default: { return lower }
-    }
-})
-
-// fjs/basen/cbase32/module.f.mjs:14, :21-29 — the owner
-const m = '0123456789abcdefghjkmnpqrstvwxyz'
-const normalizeChar = c => {
-    const lower = c.toLowerCase()
-    switch (lower) {
-        case 'i': { return '1' }
-        ...
-```
-
-The Crockford i/l→1, o→0 rule now has two definitions. If `cbase32` changes,
-`basen`'s `normalizeHit`/`normalizeMiss` proofs keep silently testing the old
-one and stay green.
-
-### Proposal
-
-Either direction works; pick one:
-
-1. Export the codec parameters (`alphabet`, `normalizeChar`) from `cbase32`
-   and build `basen/proof`'s `cb32` from the imports. Note the layering: a
-   lower module's proof importing from a child package — acceptable for a
-   proof, since `cbase32` already depends on `basen` only at runtime, not the
-   reverse.
-2. If that layering is unwanted: replace the copy with a deliberately
-   synthetic normalizing alphabet (e.g. 4-bit with one fold rule) so the
-   `basen` normalization proofs test the *mechanism* without mirroring a real
-   codec's data.
-
-Option 2 is simpler and keeps `basen`'s proof self-contained; the proof only
-needs *a* normalizer, not Crockford's.
-
-### Tasks
-
-- [ ] Replace `basen/proof.f.mjs`'s `cb32` per one of the options above.
-- [ ] `fjs t` — basen and cbase32 proofs pass.