2026-05-18
Today we activated the first piece of v3 — the autonomous side of DARA. From this morning on, the system tends to itself: it reads what's new, finds connections between memories, cleans up dead references, and writes its own one-line summaries when a memory is missing one. No human in the loop. No queue of suggestions waiting for review.
The day wasn't clean.
Five hours after launch, the system had drawn five wrong connections between unrelated memories. It saw shared category words — "website", "marketing", "design" — and treated them as evidence of the same topic. Five false positives across as many memories.
We stopped it, made it stricter. Raised the threshold for what counts as overlap. Added a list of generic category words it should ignore. Rewrote the underlying judgment to demand a concrete shared subject — same project name, same person, same product — before suggesting any link. Smoke-tested. Relaunched.
By evening, the autonomous side was running cleanly on a scheduled task — survives shutdown, auto-starts on logon, restarts itself if it crashes. The fixes held: zero false candidates after the iteration.
This is what v3 is meant to feel like. Every step the system gets a bit more autonomous. You find where its judgment falls short. You iterate. You let it run again.
Today the housekeeping is delegated. Eyes and ears come next.
The plan doesn't require a perfect day one. It requires honest iteration.
— Javier