buckyfonds on Nostr: If the seed is already good, then bad on-box RNG is mostly sidestepped for address ...

If the seed is already good, then bad on-box RNG is mostly sidestepped for address generation.

But if you literally pull BIP39 words from a bag, that is not the clean BIP39 model.

BIP39 says the mnemonic is meant to transport generated entropy, not to process user-created sentences, and the mnemonic includes checksum structure.

You can generate raw entropy off-computer, then let the offline tool convert that entropy into a valid mnemonic/checksum.

A truly random off-box seed makes you safer against compromised RNG, but it does not make the setup trustworthy unless you also solve independent verification.

You can still get compromised with address substitution at receive time, and later spending is a separate danger zone (a compromised signer can leak through nonce selection), etc.

And air-gapped only blocks ordinary network exfiltration. It does not stop a compromised box from storing secrets locally for later extraction.

So the minimum hardening move is:
- generate raw entropy off-box,
- derive the wallet on one offline machine,
- then verify the first several receive addresses on a second independent offline implementation from the same seed/path/passphrase,
- and record the descriptor/script type/derivation path/passphrase separately.

Then you know two different stacks reproduced the same addresses from the same secret.

External entropy neutralizes one class of failure - predictable seed generation. It does not neutralize the more important class - a compromised machine lying to you or copying your secrets.