OMNISIGHT FIELD GUIDE
How to Verify a Data-Driven Claim in Seven Steps
A repeatable checklist for checking timestamps, source independence, exact evidence, revisions, arithmetic, and counterarguments.
1. Identify the exact claim
Rewrite the headline as a statement that could be proven wrong. Specify the entity, metric, geography, and time window. Vague phrases such as demand is surging or risk is rising cannot be audited until those terms are defined.
2. Check when the evidence became observable
Compare source event time, first-seen time, and ingestion time. A record downloaded today may describe an event from years ago. Training and backtests should use the earliest reliable availability time, not a later corrected timestamp.
3. Open the exact source item
A search page, API endpoint, or publisher home page is not item-level evidence. Open the specific filing, docket, paper, award, advisory, observation, or job posting and confirm that its contents match the claim.
4. Count independent original sources
Syndicated copies do not provide independent confirmation. Resolve each citation to its original upstream host. For a feature claim, OmniSight requires at least two independent hosts that refer to the same coherent subject.
5. Reproduce the number
Record filters, units, exclusions, denominators, time zones, and revision state. Withhold comparisons for partial windows. A chart should carry the same as-of time and source bundle as the text beside it.
6. Write the strongest counterpoint
Look for collection changes, seasonality, backfills, duplicate publishers, policy timing, and plausible non-causal explanations. A good counterpoint does not weaken analysis; it tells the reader what would change the conclusion.
7. Separate observation from implication
Publish the observed fact first. Label interpretation as interpretation, and do not convert metadata completeness into truth or market impact. When evidence is insufficient, publish the monitoring state and the missing proof instead of forcing a story.