Integration view
What goes where
Each stage scored 0–3 for where it runs and which technique families it relies on. Values come from the where and tech fields of each stage data file, so this matrix updates when a stage is edited. The Edge column takes the higher of Edge and Robot.
Dots: 0 not used · 1 occasional · 2 common · 3 centralFilter chips dim rows whose stage is not central to the selected technique or location.
How to read it
Read a column top to bottom to see where a technique family lives in the pipeline. RL is concentrated in simulation, sim↔real and the control layer; imitation dominates data, curation and model training; world models now touch simulation, training, evaluation and the flywheel; in-context learning shows up wherever prompts, retrieval or coaching enter (models, deployment, flywheel). Read a row left to right to see what infrastructure and skills a stage needs.