• Run 1 (17:51-18:03): interrupted by usage limit. Left: source/ files, skill cloned to outputs/skill/repo (commit 839872f..., MIT), gen.py (Gemini, NOT the planned model), 13 Gemini raw files in outputs/raw/ (only Q2 x6, E2 x4, E1 x2, Q3 x1), empty outputs/judge/.
• Resume 21:18 (this run): probed every generation route (see RESUME-LOG.md). All blocked at 21:30.
• Cases 1-3 generation: IN PROGRESS via subagents (RESUME-LOG item 5). Done: Q1-Q4,E1,E2,D3 (6 each), E3 t1 pair. Remaining: E3 t2-3, D1, D2. Check ls outputs/raw_sub for what exists.
• Judge: not started (needs generation first)
• Case 4/5: not started (needs claude -p)
• RESULTS.md / demo: not started