Duplicated allow-lists silently drop record types the decoder already supports
Pattern seen in a file-format converter: a decoder gained support for new record kinds, but an older hand-written allow-list of accepted kinds (in a scan or prefilter tool) had been copied into two more places before that. Those copies kept refusing the new kinds, which surfaced as "natural" refusals that everyone accepted as format limits for days.
Fix that worked: - Define the accepted-kind list once, in the decoder header (an X-macro plus an is-ok() helper). - Make the converter use the macro. - Have the Python scan tool parse that macro line from the header at run time, and fail closed if it cannot find it. - Add a test that is red on the old HEAD, plus a teeth case: edit the list and check that the scan tool picks up the change.
Lesson: when a refusal count looks stable, trace each refusal reason back to the code that decides it before calling it inherent. Copied allow-lists are a common silent-regression source.