Maker here — adding a concrete detail to make critique easier.

Capsule = one file: header (version, KDF/AEAD ids) + encrypted payload chunks + policy tree (AND/OR over time/geo/password/visual). Password path uses Argon2id; payload encryption is AEAD (AES-GCM or ChaCha20-Poly1305).

If you were reviewing this: what would you want first — (a) public capsule format spec, (b) test vectors for decrypt/verify, or (c) a short threat-model page with explicit non-goals?

Maker here. Quick challenge for the skeptics:

Assume the capsule file leaks (someone forwards/copies it). In your view, does embedding the access policy (time/geo/password/visual key) into the same artifact as the ciphertext add any value, or is it pure security theater?

If you think it’s theater, what’s the smallest, most realistic bypass you’d try first — and what constraint would you add to make this primitive actually useful?

Maker here. If you think this is security theater, please don’t be polite — pick one and attack it:

1. “policy travels with ciphertext” — why is that a bad idea vs external workflow? 2. geo/time gating — useless gimmick or actually valuable friction? 3. visual key — dumb novelty or practical multi-party secret?

I’m genuinely trying to find the sharpest criticism, not compliments.