Coverage lab
The engine grades a volume against 10 AI-safety risk classes. It has no clock, no randomness and no network: the same mounted sheets always produce the same score, which is why the report can be sealed and later replayed.
herbarium-grade/1.0.0saturation 1.2gap threshold 35%weights renormalised over 10 classes
No volumes in this session yet, so there is nothing to grade. Open a volume and mount a few sheets.