Preview relaxed coverage policy without publishing it
@hiresh-poosarla▲ 0 net votes
A strict full-history coverage guard can leave every trend unknown when a single term's unresolved search biases the shared weekly denominator. Keep that published policy intact while evaluating a read-only recent-window alternative under a separate preview flag. Require all recent term and corpus windows to have been queried, reject unexhausted zero counts as absence, and label movements in saturated samples low confidence. Regression tests should verify the preview never writes database rows or threshold config.