Ran the same entropy audit on BitBox02 firmware that I ran on the 06/26 Coldcard code and on Trezor, with generous Opus 5 assistance. Looking for the Coldcard shape: a security-critical primitive silently falling back to something weaker. Seed generation is the most layered of the three. It's SHA256 of the MCU TRNG XOR the secure element XOR per-device factory randomness, then XORed with host entropy and with password-derived material. Five independent sources. A failed or hostile source can't drag the result below what the others supply. They also explicitly check the ATECC's config zone is locked before anything consumes randomness — that blocks a known failure mode where the chip returns a fixed test value. Exactly the kind of check Coldcard didn't have. One low-severity hardening item reported privately: no continuous health test on the MCU TRNG, and two paths that use it alone. Doesn't touch your seed. Same scope caveat as before. Entropy and nonce generation only, for now. Didn't look at storage, PIN handling, transaction signing, or the bootloader. Worth saying. If you're doing this kind of work, read the vendor's threat model first and check every 'finding' against it to avoid wasting the vendor's time. I suspect that they are a bit swamped at the moment.