Forget & the purge

Most memory systems cannot actually forget — they mark, hide, or down-rank, and the content lingers in an index somewhere. memcell's forget is a real deletion, because a memory you cannot truly remove is a liability, not a feature.

what forgetting does the one operation history does not survive

Forgetting a statement deletes it and its evidence from the record, permanently. The action says so before you take it — there is no undo, no trash, no soft-delete waiting to resurface. This is the deliberate opposite of everything else in the product, where history is append-only: rewording keeps the old text, superseding keeps the chain, settling a disagreement keeps the dispute. Forgetting keeps nothing, on purpose.

the purge contract every projection acknowledges, or the delete did not happen

A statement lives in more than one place while it serves: the record, and the retrieval structures built over it — vector, full-text, graph. Forget is bound by the purge contract: the deletion lands in the record and every projection acknowledges the purge. A projection that cannot confirm it has dropped the content is a bug, not an inconvenience.

when to forget, and when not wrong is not the same as sensitive

Forget is for content that should not exist in the record: something filed by mistake, something you are bound to keep out. A statement that is merely wrong should be reported against or superseded instead — a failed claim that falls out of service still teaches the record; a forgotten one teaches nothing.