Node 05 | The Self-Improvement LoopEvery decision a run makes is scored against what the world returned — merged or rejected. The greenlist is the memory: change it, and you change what the next run picks.
merge rate
0%
0 merged / 6 attempts
one repo · no verdict returned yet
The greenlist is the memory. Everything else a run does is logged for analysis; only the ranked set of repositories it mines feeds forward. Change the greenlist and you change what the next run does. Edits to it are human-gated — currently idling. The circle is fully wired; it just hasn’t completed a lap, because the world hasn’t returned a verdict yet.
01Collect the pillar scoreshow every step actually performed
02Calibrate against realitywhere the self-scores were off
03Log to the ledgerone queryable record
04Find the weak pillarspatterns across many runs
05Propose a greenlist editfix the picking pillars
06Push PR with improvementsfix the judgment pillars
07HITL · Human reviewapprove and the agent finishes the merge
