Filters and interpretation¶
The explorer groups choices as pathogen → logical source → signal, with a checkbox
for each available release/provider/geography variant under each signal. COVID-19,
influenza, and RSV come first, then combined and ARI measures, other respiratory
viruses, and all-cause capacity; sources and signals are alphabetical within each
pathogen. The shared selection policy assigns the pathogen (pathogen_of). NHSN
and NSSP each combine their source products into one group, whose summary lists
the providers present (CDC, Delphi, Forecast Hub); each hub retains its own group
and target definitions.
See Shared selection for the measure and file allowlists.
Searching covers providers, dataset titles, source paths, column names, and
complete series labels. Search and filters can restrict the available variants.
Delphi variants are highlighted in the selector and can be selected alongside
CDC variants from the same signal.
Signal filters¶
| Filter | Meaning |
|---|---|
| Cadence | Catalog temporal_resolution: daily, weekly, monthly, or sample-based |
| Revision history | Whether the catalog marks the source as historically versioned |
| Spatial support | Native state data or an explicitly labeled national context |
| Data currency | Whether the last event date falls within its cadence-specific window |
Currency thresholds are 14 days for daily, 35 for weekly, 75 for monthly, and 45 for sample-based data. “Current” describes the indexed event dates; it is not a guarantee that a publisher will continue maintaining a product.
Filters are evaluated by the local API before pagination. The API reports variant counts; the UI additionally reports the distinct signals among returned variants. The UI requests all matching variants. Native-state support is the default and Reset restores it. Choose Any support or National broadcast for national-only demographic and RSV-NET products.
Series interpretation¶
Low-cardinality fields such as pathogen, subtype, model, target, unit, and
quantile define separate dimension variants. Publisher season, respseason,
and season_week fields are metadata and do not artificially split a covariate
across seasons.
Repeated raw rows for one state, series, event date, and latest vintage are shown as an unweighted mean. Tooltips expose the contributing row count. County, HSA, HHS, catchment, and wastewater-site rows are excluded by the shared post-intake filter before indexing.
Scaling¶
Divide by mean divides each selected series by its mean over the common window, the dates covered by every selected series, so all series average 1 there. The y-axis fits the common window; a longer series with larger values outside it leaves the top of the canvas. The axis title names the window. Revisions use the latest version's divisor. It does not alter the index or raw data, and it is not the normalization contract for model training.