How the compatibility numbers are measured#

live/corpus-refusals.json measures which refusals fire and how often across the corpus. Compatibility reference copies no count from it, because a copied count goes stale the moment the corpus re-runs.

That measured ranking is why the static-evaluability rule (see Identity) leads the reference page. Several of the most frequent refusals are that one rule under different diagnostics.

Do not read the fixture or module-example populations as a compatibility rate. Module examples/ directories demonstrate a module’s full surface, so they lean far harder on variables, conditionals and dynamic blocks than a configuration describing one deployment, and refuse almost across the board. Those populations are marked as a ranking, settled by #118. One population can honestly carry a rate since #147, whole deployment root modules published by their operators, pinned by commit, marked reads_as: rate.

Run choudoufu live-check on your own configuration rather than inferring anything from the corpus.