How Novus calculates
Methodology
Normalization
Provider records retain source URL, provider item, economic level, observed date, original unit and currency, fetch time, confidence, and a deterministic raw-record hash. Invalid records remain rejected rather than receiving guessed values. The observed date is the provider's publication period; the fetch time only states when Novus checked that source and never turns an older quote into today's price.
Comparability
Novus compares observations only when provider, item, economic level, unit, and currency match. Producer, wholesale, retail, CPI, and index signals are not blended without an explicit conversion model.
Scores
Percentage change is current minus prior, divided by prior. Confidence is recalculated per comparable series on a 0–100 scale: 35% coverage, 25% cadence-aware freshness, 25% source-record quality, and 15% comparison-field completeness. Priority is a bounded 0–100 blend of change severity, purchasing weight, and confidence. Spend impact applies the percentage movement to matching imported spend.
Freshness and fluctuations
Daily sources are current only inside 48 hours. Weekly, fortnightly, dekadal, monthly, annual, and mixed sources use their documented publication windows. Novus reports movement between two like-for-like observed periods; it does not claim that the latest official observation is an executable price today.
Failure behavior
Each provider receives an isolated health and ingestion result. Operational requires at least one normalized observation inside the expected publication window. A reachable catalog with zero normalized rows is degraded or delayed, not operational. Disabled or unavailable integrations report their state; they do not emit fallback observations unless the documented country coverage uses a real global provider.