A missing reading is not a zero reading
An environmental sensor stops reporting for an hour. If the API substitutes zero for missing temperatures, downstream analysis may show an impossible cooling event. Missing-data treatment preserves the difference between absence of evidence and an observed value.
Classify missingness
Device failure, delayed transmission, maintenance and quality rejection can create different gaps.
Preserve original observations
Do not silently overwrite missing periods with fabricated measurements.
Choose treatment for the purpose
Some analyses may use interpolation or model-based estimates, but assumptions and limitations should be explicit.
Expose quality flags
Clients should know which values are observed, estimated, stale or unavailable.
Consider temporal alignment
Different reporting intervals and delayed batches can complicate comparisons.
Review downstream effects
Missing data may affect averages, alerts and public displays differently.
Break-test: a long gap is filled with a straight line
Can users distinguish estimated values from actual measurements and judge whether the method is appropriate?
Connected knowledge
See Time Series Alignment, Stale Data Alerting and NEA