slopless 0.2.16
Density-based precision for two more rules, from the false-positive audit over a corpus of pre-AI human writing.
- simplicity and llm-vocabulary now fire only when complex/stock diction clusters, not on each single word. Both fire 0 on 100 Wikipedia files (were hundreds of per-instance hits); cluster detection retained. Single-word slop like "delve" is still caught by prohibited-words.