Expand description
/api/learning: whether learning works in this collective, week by week, from the logs Borg keeps
anyway (no state of its own, so the history is there from the first view).
Each section says where its numbers come from: team when they fold synced logs (stats.jsonl
rollups, lessons.jsonl) in a team collective, this machine
when they fold the local event log (stats/events.jsonl, which keeps the lesson sources) or the
collective is local.
- lessons written per source (session, review, regret, veto, queen, import, …), and the team’s lessons written (from lessons.jsonl, each lesson once) and merged;
- asked lessons’ verdicts: changed the work, ignored, corrected anyway (the asked→changed rate);
- vetoes and repeated vetoes (the same “no” said twice: a lesson that did not hold).
The weeks run from the first counted one to the current one (partial_week). The history before
live counting began (stats.rs backfill) derives no merges, repeated vetoes or lesson sources:
those series are null before live_from, unknown rather than zero.