Add section 12: RoboCasa fine-tune (0/15 -> 7/15 on trained task, 0/45 held-out); qualify section 11's rotation-frame hypothesis

Complete section 11: all 5 RoboCasa tasks, zero-shot 0/60 on arm-only tasks

Add section 11 (RoboCasa zero-shot transfer); extend section 10 with pretrain_C fine-tune + B-vs-C head-to-head

ablation.md: section 7 - fixed vs adaptive gate comparison at matched call budget (17-30x fewer calls, no difference vs full-rate); documents the bootstrap-call counting bug found and fixed along the way

ablation.md: add section 10 - fine-tuning pretrain_B on LIBERO with the ft4 recipe (69.0% vs ft4's 85.2%, significant)

ablation.md: add pretrain_B (same-harness LIBERO+profile); pretrain_C flagged as unrunnable here (missing morph-conditioning code/descriptors), cite author's own A/B/C few-shot table instead

ablation.md: complete consistent-lag test (3 ages, fixes a filename-mapping bug that silently dropped the 1s row)

ablation.md: add pre-LIBERO checkpoint comparison (stage2/stage3/pretrain_A, all 0% on LIBERO closed-loop)

ablation.md: shuffle control, consistent-lag test (proprio staling makes it worse, not better)

ablation.md: add the causal staleness grid (semantic vs geometric) and the zeroed-latent control

Add ablation.md: corrected rate-separated evaluation, 500 paired episodes, system profile