mmClaw: Separating Spatial Readability from Action-Head Geometry

Imitation policies can expose readable spatial coordinates without using the corresponding probe directions to act. We study this distinction in mmClaw, an auditable, path-level radar testbed. Two teachers label identical requests using a public estimate or a return centroid; paired students differ only in supervision. Frozen policies and probes preserve a positive teacher-by-variable interaction on 48 new layouts: $D=0.156[0.125,0.236]$. We then decompose native evidence responses through the final action head. Coordinate-related editing can mix originally output-null changes into the selected response: covariance operators produce signed null-component projections of $0.207$ and $0.357$ for the two teachers. LEACE exhibits stronger fit-dependent cancellation, while coordinate increments become small after teacher-direction erasure. Nonlinear action-and-pose references and independently varied eraser/probe supports further distinguish task-label predictability from stable semantic use. These results support a supervision effect on relative readability, but not a unique coordinate-mediated action mechanism. We provide an output-consistency check, retained finite-action components, and explicit failure controls for interpreting representation edits.