Drag the slider to ask: what would prediction quality look like for a tenant whose data only goes through this month?

Each slider position runs three live _evaluate queries against the current tenant's purchases table, with order_month: { $lte: cutoff } tacked onto the evaluate-step where. Aito's conditional probabilities then condition on only rows up to that cutoff — same shape as a younger tenant. No data manipulation; works against read-only keys.