The measurement framework was designed to force plugins to set specific failure flag fields indicating each known problem, by translating any uncaught exceptions other than lsst.pipe.base.MeasurementError into warnings, and expecting the obnoxiousness of such warnings to force changes to the plugin when they arise.

But at some point (ages ago, in a huge surprise to me), those warnings got turned into debug logging here:

https://github.com/lsst/meas_base/blob/f28afc6cbd5c4ca3fd47296423c511aa68004590/python/lsst/meas/base/baseMeasurement.py#L465

That's a huge problem, because it means all of the plugins added in the past five years have been allowed to have unrecognized failure modes that provide no information to the user about what went wrong.

And we probably can't afford to just change that one line back to what it should be - the deluge of warnings needs to be addressed first, ideally as part of the same (big) ticket (to keep the problem from returning before we fix the log level).

When transforming particular failure modes for meeasurement plugins into new failure flags, note that we should probably avoid using MeasurementError, even if that's a bit easier, since we're now trying to get rid of that, too (DM-35237).