Interrupted 2026-10-03 ~18:03 by usage limit; resumed ~21:18. Earlier outputs kept untouched.
State found at resume:
- source/: TEST-PLAN, HYPOTHESIS-AND-CRITERIA, prompts, text300 (done)
- outputs/facts/: raw primary-source pages saved, NOT yet analysed
- outputs/raw/control-openrouter-gemma4-26b-a4b.json (+control.log): 15/15 trials done (case 6 raw DONE; not yet scored)
- .local/models/gemma4-e4b/gemma-4-E4B_q4_0-it.gguf: 5154941280 B, equals HF listed size (download complete)
- llama.cpp build present, not run yet
| Case | Status |
|---|---|
| 1 app fact check | in progress |
| 2 model name | not started (analysis) |
| 3 claim check | not started |
| 4 run locally | not started |
| 5 matrix | not started |
| 6 control | raw done, scoring pending |
| 7 phone guide | not started |
| RESULTS.md / demo | not started |
Update 21:37
- Case 1-3: raw facts analysed (see outputs/facts-analysis.md when written).
- Case 4: gemma-4-E4B Q4_0 GGUF loads and generates on CPU, but only ~0.8-2.4 tok/s while machine load was 10-16 (see raw/bench-threads.txt, raw/bench-taskset.txt, raw/local.attempt1.log). Attempt 1 (6 threads) stopped at 6/15 trials.
- Case 5: attempt 2 running via outputs/run_local.sh (-t 4); done marker raw/local.done; results raw/local-gemma4-e4b-q4_0.json