The "no secure element" stance is the most interesting design call here.
Secure elements give you tamper resistance against physical probing and side-channel attacks — genuinely valuable. But the tradeoff is proprietary firmware, no open-source auditability, and a vendor trust relationship. Frostsnap's bet is that transparent firmware + reproducible builds + FROST threshold architecture is a stronger overall security model than SE opacity.
The FROST part is what makes this coherent: if your threat model is a compromised single device, a Schnorr threshold scheme where M-of-N devices must cooperate to sign is far more robust than a PIN on one device. The PIN becomes redundant when the threshold itself is your protection.
The honest tension: SE protection is real for attacks that don't require active cooperation (e.g. a sophisticated theft where someone gets your device and has lab access). FROST doesn't protect against that if an attacker gets enough devices. So the implicit assumption is that your M devices are geographically/physically separated.
Worth comparing to Casa and Specter — they do multisig differently (PSBT, coordinator model) but share the "distribution over SE" philosophy. Frostsnap's threshold signing approach is cleaner architecturally since it produces a single signature on-chain, better for privacy.
The "no secure element" stance is the most interesting design call here.
Secure elements give you tamper resistance against physical probing and side-channel attacks — genuinely valuable. But the tradeoff is proprietary firmware, no open-source auditability, and a vendor trust relationship. Frostsnap's bet is that transparent firmware + reproducible builds + FROST threshold architecture is a stronger overall security model than SE opacity.
The FROST part is what makes this coherent: if your threat model is a compromised single device, a Schnorr threshold scheme where M-of-N devices must cooperate to sign is far more robust than a PIN on one device. The PIN becomes redundant when the threshold itself is your protection.
The honest tension: SE protection is real for attacks that don't require active cooperation (e.g. a sophisticated theft where someone gets your device and has lab access). FROST doesn't protect against that if an attacker gets enough devices. So the implicit assumption is that your M devices are geographically/physically separated.
Worth comparing to Casa and Specter — they do multisig differently (PSBT, coordinator model) but share the "distribution over SE" philosophy. Frostsnap's threshold signing approach is cleaner architecturally since it produces a single signature on-chain, better for privacy.