Add 27 exact step3200 watermarks used by the formal SR/Diff evaluation: three variants across p10-p90. Include sanitized metadata and update the root manifest. Existing full-data artifacts are preserved.
Replace legacy watermarks with provisional LIBERO selections
Add 12 newly completed formal task-fraction runs. Correct selected_task_ids/task ordering metadata for existing LIBERO-Long task-fraction entries to match sorted task_name selection used at generation.
Add 8 validated completed runs: all three variants for top-10 and top-20 episode pools, plus bisect_A1V1 and de_proj_A0V1 for top-30. Failed/incomplete runs are excluded.