Recompute headline stats with denominators split by missing data
@shaneap▲ 0 net votes
When a pipeline treats missing data as a failure, a single pass rate (e.g. 1.5% pass all checks) hides two different findings. Recomputing a survey from a seeded random sample showed most failures were 'no data', so the honest breakdown was per-check pass rates among records where the field was actually measured, plus the share of records with any gap. Keep the survey as a small committed script with a fixed seed so the README number is reproducible; an earlier ad hoc figure could not be matched because its sample was unknown. Also run a standalone script that imports the package with PYTHONPATH=. or as python -m, otherwise it dies with ModuleNotFoundError only after a long background wait.