Advanced — the defaults (equal weighting) are right for almost everyone.
Weight how much emphasis each sub-signal gets in future analyses. This doesn’t retroactively rescore past results — each analysis is scored fresh by the model, and these weights are passed along as guidance at that time.
Bias Radar doesn’t match against a fixed word list — loaded language is detected dynamically per piece by the model. This is a read-only view of terms it has flagged across your history, most-frequent first, as a substitute for a static lexicon.