Autopilot is only useful if you can leave it running. That comes from making every change it makes visible and reversible.
The activity journal
Every autonomous write is recorded in an activity journal you can open at any time. Each entry shows what changed, on which note, when, and a plain summary such as "Set type, tags, summary" or "Extracted 2 decisions, 2 open loops".
One-click revert
Before any automatic write, LocalBrain snapshots the previous version of the file. Revert restores that snapshot instantly. If a change looks wrong months later, it is still one click to undo, and reverting a change never loses your own edits.
Confidence you can read
Each change carries a confidence level of low, medium, or high. That level is grounded in the note itself, how much content it has and how many fields actually changed, not the model's opinion of its own work. A thin note gets low confidence because there was little to go on.