How Should a Dashboard Handle Conflicting Data Sources?
From the earlier reference collection. Read the current software blog.
When sources disagree, explain their definitions, timing, coverage, and selection rules instead of blending incompatible values silently.
Establish why values differ
Two sources may disagree because they measure different things, cover different areas, update on different schedules, or apply different correction rules. Before choosing a value, compare the definitions and timestamps. A station observation from 10 a.m. and a model estimate for the whole day should not be treated as competing readings of the same kind.
For example, one outage feed may count customer reports while another counts utility-confirmed incidents. Their totals can both be valid while describing different stages of a process. Label those measures separately rather than averaging them into a number that belongs to neither source.
Document the selection rule
If the dashboard needs one displayed value, state how it selects the source. It might prefer a designated primary feed, use the newest successful update, or fall back to another source after a defined delay. The rule should be understandable and predictable, with the selected source named near the metric.
A “latest timestamp wins” rule is not always enough. A newer source may have less geographic coverage or a different measurement method. If selection depends on both recency and quality, explain the criteria and make relevant exceptions visible.
Preserve provenance and disagreement
When readers benefit from seeing both values, provide a comparison with source names, timestamps, units, and coverage. Explain the known reason for disagreement when it is established. If the reason is not known, say so; do not invent a reconciliation story.
Keep provenance in downloads as well as on screen. A CSV column identifying the source and observation time allows later users to trace a value. If the dashboard normalizes units or filters duplicate records, document that processing so the transformation does not appear to be a raw source value.
Avoid false consensus
Combining sources can make a result look more robust, but an average is meaningful only when the inputs are comparable and the method makes sense. A sensor and a forecast, or a report count and a verified count, may require separate display. A disagreement can itself be useful information about coverage or method.
A trustworthy dashboard makes the choice visible and lets readers inspect the evidence behind it. When the sources cannot be reconciled responsibly, present the difference and its limits rather than hiding it in a polished single number.