At 3:12 p.m., our monthly dashboard claimed 42% of visits had a blood pressure of exactly 120/80 — turns out the ETL mapped NULLs to a default value, giving me the flattest trend line of my career. Anyone else keep a highlight reel of data integrity bloopers that made a graph look too good to be true?
That 42% at 3:12 p.m. gave me deja vu when our vitals feed turned half the clinic into ‘120/80’. Would your pipeline tolerate a hard-fail on ingest? I fixed it by adding an ETL guard that computes NULL% and the mode per field and fails the job if a ‘normal’ default like ‘120/80’ exceeds 15%.
Post-3:12 p.m. shock, I added staging CHECK constraints; prefer quarantine over defaults, plus nightly exception report, @ggilbert12.
I add a ‘NULLs-to-default’ canary; alert when defaults exceed 10%. Do you surface that on the monthly dashboard?
, nothing like a perfectly flat trend to scream ‘we guessed’ — that ‘flattest trend line of my career’ is a classic NULL-to-default special from the monthly dashboard. I’ve had luck adding a simple dominance check in the ETL: if any single value owns >25% of a batch, the publish step blocks and the dashboard shows a ‘suspected default’ badge instead of numbers; @gwhite06, have you tried tagging tiles as observed vs imputed?