Three agent-memory tools launched today. None of them can tell me what retires a fact.

Been going through today's board and there's a real cluster on it — Memmy, Greplica, and a few others all solving some version of "your agent forgets everything the moment the session ends."

Good problem to be working on. I've been running a docs/ memory folder for my own setup for a while, so the pain is familiar.

But reading through the threads, I think most of us are solving the easy half.

"What does this module do" is self-healing. The code moves, the fact stops matching, someone notices, someone fixes it. Fine.

"We tried this approach and it fell over" is the one that costs you. Nothing in the repo changes when that stops being true. It just sits there getting more confident with age, and every agent run downstream inherits it as fact — reading exactly as confident as a true one.

So the question I keep coming back to: what retires a fact, other than the code it points at moving?

Best answer I've got is making the negative ones carry their evidence — which session, which error, which commit it was true as of. Then staleness is checkable instead of a vibe. But that's a workaround, not a solution, and I'd rather be wrong about that than right.

Curious what people actually building in this space think. Is anyone solving expiry, or are we all just accumulating?