Every statistic must be explainable.

Day and week boundaries use the goal timezone. Missing data remains unknown. Percent changes should not be shown when a zero or tiny baseline would make them misleading. Previously issued receipts need versioned inputs before long-term comparisons are presented as immutable records.

The founding cohort will be used to learn activation language, logging friction, privacy concerns and the emotional effect of accountability. Aggregate research will be published only when it is methodologically useful and cannot expose sensitive behavioural data.