What the PBZ variants did not preserve

Repeated filtering reduced bad baskets but often removed the behavior that made the original approach interesting. That failure narrows the next question.

PBZ research produced an important recurring failure: filters that removed damaging baskets often removed much of the original behaviour as well.

That result is useful. It suggests the difficult part is not simply recognizing a bad entry with another indicator. The unresolved problem may sit in the lifecycle of a basket and its exposure to persistent trends.

The work remains on the watchlist. Future tests should begin from that narrower claim instead of restarting a broad parameter search.

PBZNegative evidence