Blender workcell camera / approach-box study

Original virtual experiment: October 4, 2026, Blender 5.2.2 LTS.
Public scene export and numeric recheck: October 5, 2026, same Blender version.

layout-study.blend contains the three L001 scenes and their dependencies,
exported from the existing experiment. It excludes unrelated experiments.
It uses original basic geometry, with no external product models or textures.
A/B/C_measurements.json and verification.json preserve the original records.
recheck-results.json records a subsequent run of recheck_l001.py on the export.
The new recheck did not rebuild geometry, render images, or execute motion.

To repeat the numeric checks:
1. Extract this ZIP into a writable folder.
2. Open layout-study.blend in Blender.
3. In Blender's Text Editor, open recheck_l001.py and choose Run Script.
4. Read results.json in the newly created recheck-<UTC timestamp> folder.
The script changes the active scene while reading, then restores it. It does
not save the scene file or overwrite prior result folders.

Expected A/B/C unblocked top-face samples: 0/25, 0/25, 25/25.
Expected signed approach-box Y margins: -40 mm, +100 mm, +100 mm.
The top-face sample is a 5 x 5 grid, not a recognition benchmark or area score.
The approach box is a conservative assumed descent volume, not an executed
gripper trajectory. No dynamics, contact forces, real robot, vision model,
hardware safety margin, or successful sorting cycle was evaluated.
