Engineering, production, governance, cost, communication · insight-0302
Classify data-quality checks into blocking and warning tiers before they fire
- Evidence type
- practitioner heuristic
- Difficulty
- easy
- Roles
- ML engineer, Data analyst
- Industries
- aviation, cross-industry
Summary: situation, action, reason, boundary
Situation
A new validation suite fails the nightly pipeline on a minor null-rate rise, and the business misses its morning report.
Action
Tag each check as blocking (schema break, duplicated keys, impossible values in critical fields) or warning (small distribution shifts), route warnings to an owner, and review tiers after each incident.
Reason
Severity tiers keep critical protections active while avoiding unnecessary outages.
Boundary
Data pipelines feeding models, reports and decision systems.
Symptom or risk
If every check blocks, teams disable checks; if none block, corrupt data reaches models and dashboards.
Exceptions
Regulatory reports may require blocking on any check failure.
Trade-offs
Warnings can be ignored; track time to resolution.
Pitfalls
Thresholds set once on clean historical data that normal seasonality breaches.
Example
Synthetic airline raw data contain load factors above 1 and negative bookings: impossible values that should block.
Lessons and concepts
Claims and their sources
Great Expectations runs declarative data-quality Expectations as validation steps in pipelines.