Stage-4a archive cells: human review

Two reviews over the scored 300-scene re-batch. Review 1 checks each question type's cell definitions (the three axes, their bin edges and labels). Review 2 opens every occupied cell and shows every question that landed in it, with its video, so you can judge whether the cell's members share the condition its name claims and whether each question is fair.

Corpus: 299 scenes, 5320 gate-passed questions of live types, 361 occupied of 704 cells. Bins cut at 240×135@16, the settings the re-batch was rendered and measured at. Membership derived with Builder's axis functions (learning/cell_review/derive_members.py); Builder's canonical export should reproduce it exactly. Verdicts are stored in this browser only — export them from Progress / export when you stop.

Suggested order: for each type, definitions first, then its cells. Model answers are hidden on cell pages by default; judge fairness before revealing them.