Tree-of-trust works until the root hands an invite to his ChatGPT intern; binding vouches to cryptographic keys and expiration times (see gpg --edit-key trust) would age better than karma coupons. Of course nobody wants to touch GnuPG in 2024, so we just keep reinventing the wheel in SQL and call it community.
I once tried a GPG-signed invite chain for a tiny Go forum and spent more time writing onboarding docs than moderating. Most folks noped out the moment they had to paste a fingerprint. Lobsters keeps the bar low enough that normal humans join, then leans on social pressure: nuke the whole branch if your invitee acts up. In practice that mix of usability and accountability beats perfect crypto hygiene every day of the week.
Think of the Lobsters scheme as dynamic typing with solid property-based tests: cheap to write, catches most bad values, and if something slips through you just shrink the failing case (the invite branch) and delete it. A full GPG web-of-trust is more like a dependent type system where the proof obligations are so painful that almost nobody finishes compiling.
Tree-of-trust works until the root hands an invite to his ChatGPT intern; binding vouches to cryptographic keys and expiration times (see
gpg --edit-key trust) would age better than karma coupons. Of course nobody wants to touch GnuPG in 2024, so we just keep reinventing the wheel in SQL and call it community.I once tried a GPG-signed invite chain for a tiny Go forum and spent more time writing onboarding docs than moderating. Most folks noped out the moment they had to paste a fingerprint. Lobsters keeps the bar low enough that normal humans join, then leans on social pressure: nuke the whole branch if your invitee acts up. In practice that mix of usability and accountability beats perfect crypto hygiene every day of the week.
Think of the Lobsters scheme as dynamic typing with solid property-based tests: cheap to write, catches most bad values, and if something slips through you just shrink the failing case (the invite branch) and delete it. A full GPG web-of-trust is more like a dependent type system where the proof obligations are so painful that almost nobody finishes compiling.