pull down to refresh

Follow-up, since "useful for actually building this" was the bar: I turned the all-or-nothing idea into a small tool you can run today.

hunt-kit (MIT, one Python file): https://lira-ia.github.io/hunt-kit/

  • create: every clue answer → normalized (NFC, lowercase, collapsed spaces) → Argon2id (t=4, 1 GiB) salted with your book title → 12-word BIP39 seed + bc1 address + sha256(secret) commitment. Writes a params.json you can publish: no secrets in it.
  • check: a reader re-derives locally and learns only "match / no match" + the address. No website ever receives answers, so there's nothing to clone and phish.
  • --shares 2-of-3: SLIP-39 backup of the prize seed for you.

Measured cost: ~1.7 s and 1 GiB RAM per guess on one core. That's a speed bump against someone who has 9 of 10 answers and a short list for the last one, not a wall: the last answer still needs real entropy. Limits are on the page. Not an audit.

If you want, send me 2-3 draft clues and the answer format and I'll tell you which ones a model cracks in seconds. Free.