A number without its quality status can mislead
An environmental API returns a temperature value that was estimated after sensor failure. If the client treats it as directly observed, the resulting analysis may be wrong. Quality flags communicate how a value should be interpreted.
Define flag meanings
Observed, provisional, stale, estimated and invalid may require distinct definitions appropriate to the dataset.
Use consistent rules
Flag assignment should follow documented criteria rather than arbitrary manual choices.
Preserve provenance
Users should be able to understand relevant measurement, processing and correction history.
Test downstream interpretation
Applications must handle flags correctly instead of ignoring them or treating all values as equivalent.
Manage changes carefully
Adding or changing flag meanings may require versioning and consumer communication.
Audit exceptions
Review misclassified observations and recurring data-quality problems.
Break-test: an imputed value is marked as directly measured
Can governance detect the error and correct affected outputs?
Connected knowledge
See Imputation Disclosure, Missing Data Treatment and NEA.