Single recovery code instead of backup lists — feedback welcome

Most products issue 5–10 backup codes for account recovery.

I implemented a different model in Toqen:

• a single 15-character recovery code
• 31-symbol alphabet (ambiguous characters excluded)
• ~74 bits of entropy
• CSPRNG generation
• no modulo bias
• strict per-identifier rate limiting
• automatic rotation after use

Search space:

31^15 ≈ 1.92 × 10^22
~74 bits entropy.

Under an online threat model, brute-force becomes mathematically impractical.

The idea is to reduce complexity without reducing security.

Would genuinely appreciate constructive critique:

• threat model gaps
• distributed attack risks
• UX trade-offs
• anything that looks fragile

You can try the implementation here:
https://toqen.app

Interested in honest feedback.