Check it without taking us on trust.

A copy of the record is kept each time it is published. Each copy is named by the date and the first characters of its own fingerprint. Any two versions can be compared line by line.

Only probability calls locked no later than publication enter the headline score. Earlier published calls stay visible and score separately. Calls entered after their windows closed stay visible for audit and never enter the headline score.

Candidate statements that failed the standard are counted by reason. Selection cannot disappear from the denominator. Their text stays private because it includes working material.

The record holds 77 call families. A family defaults to its source report until a narrower family is set. Several calls driven by one event are not presented as several independent tests. The checks below are generated from the same record. A non-zero count stays visible until it is corrected.