The open dataset and directory include every published scorecard. The public site does not publish absolute highest- or lowest-scoring lists, individual national percentiles, or size-peer percentiles. A score is evidence about one published feed, not a judgment of service quality, staff performance, or compliance.
Aggregate score summaries and named change views use a narrower comparison cohort. A feed record is included only when:
- it has a dated snapshot and numeric overall score;
- correctness, freshness, and rider-experience completeness were measured;
- it uses the current rubric version;
- it uses the current scoring profile and canonical validator version;
- it uses the
raw-v1reader archive profile, so a normalized reader view cannot be mixed into raw-feed aggregates; - it belongs to the selected homogeneous measured-category set, so a three-category overall score is never mixed with a four-category score;
- its published service data is not more than one year expired;
- it is an active canonical registry record, not an alias; and
- its feed identity is unambiguous. Repeated Mobility Database IDs, normalized feed URLs, or exact current feed hashes are excluded until reconciled.
Realtime remains optional and is not an eligibility condition. A small agency without realtime is never excluded for that reason.
Named changes compare a feed only with its own prior check, and only when both
checks use the same current rubric, scoring profile, validator, and measured
category set, including the same reader archive profile. The historical
api/v1/leaderboard.json path is retained for v1
consumers, but its top and bottom arrays are always empty. It carries guarded
most_improved and most_declined change rows plus a comparison object
explaining exclusions.
Every excluded record remains available in the scorecard directory and open dataset. Vendor-host comparisons remain internal and must not be republished as a blame board.