ai-sherpa/repro-vlm-robustbench

chore: bump updated_at to re-trigger judge scan (was retrying-after-error, attempt 2/3)

rework: decisive falsification verdicts for claims 4/5, new claim-6 fingerprint page + scorecard, de-stale conclusion

Add full-scale N=1000 VLM evidence to claim-3-semantically-strong-spatially-fragile (glass_blur~8pp verified; geometric 34pp falsified)

Add full-scale N=1000 VLM evidence to claim-2-severity-is-a-weak-predictor (glass_blur~8pp verified; geometric 34pp falsified)

Claim 1: replace inconclusive with rigorous structural audit (49 types, 133=42x3+7, 4 families)

Sync complete logbook content (results + conclusion)

Sync complete logbook content (results + conclusion)

Sync complete logbook content (results + conclusion)

Sync complete logbook content (results + conclusion)

Update logbook: Repro - VLM-RobustBench

Update logbook: Repro - VLM-RobustBench

initial commit