Memorable keeps the path that worked and hands it back before the next run.
The request is matched to a stored routine before the model starts reasoning.
The routine runs whole: lights, volume, the film. Nothing is re-explained.
The outcome is stored. A routine that changed is kept as a revision beside the old one.
Ability traces come in at the end of each run. On OpenHome, 100% task completion with procedural memory, up from 70%.
A routine taught in a past session replays as one procedure, not a re-derived plan.