Hybrid distillation mode. Uses any provided feature semantics (e.g., Min/Max/Anchor) and infers missing values automatically during exploration. Balances accuracy and flexibility. Recommended default for most scenarios.
Strict
Strict distillation mode. Requires all feature semantics (Min, Max, etc.) to be explicitly defined. No inference or backfilling is performed. Ensures full control and predictability, but may reduce coverage if inputs are incomplete.