archetech / archon / 30922975031

94%

main: 87%

Pull #857

github

docs: record that there are two independent RNGs, not one

The "Randomness source" section described a single chain through
@noble/hashes randomBytes. That is only half of it: @noble/secp256k1 ships its
own randomBytes and is what backs randomPrivateKey, so vault keypairs never
touch @noble/hashes at all.

They are separate packages on separate version lines, which is the reason this
matters — a dependency bump can change one and leave the other untouched. A
single install can also resolve more than one copy of @noble/hashes, since
@noble/curves nests its own.

Both refuse to degrade, but by different mechanisms: @noble/hashes captures
globalThis.crypto at module load and throws, while @noble/secp256k1 reads it at
call time and dereferences it. Documents both, and points at the tests that now
assert the property of each.

Also records what those tests deliberately do not attempt: a degraded RNG still
produces output that looks random, so neither statistical testing nor any
assertion about the shape of a mnemonic would catch a substitution.

Verified while writing: randomPrivateKey is still randomBytes(L + 16) with
L = 32, so the existing "randomBytes(48), FIPS 186 B.4.1" claim holds — only
the attribution of which randomBytes was wrong.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

3609 of 4063 branches covered (88.83%)

Branch coverage included in aggregate %.

8230 of 8512 relevant lines covered (96.69%)

687.14 hits per line