Use cases
Start with a decision someone needs to make, then connect the trend, the breakdown and the underlying rows.
Try a complete workflow
Start unfiltered, make one change, inspect the metric and exact records, then clear it. A view that publishes selection keeps alternatives visible; selecting an account can focus sibling charts while the selecting table retains other matching accounts.
| Example | Try this | Check the boundary |
|---|---|---|
| Storefront | Change period, select a region, inspect products and export. | Modeled aggregate sales and fixed-week timing; not a live store or individual-order feed. |
| SaaS | Select a plan, inspect account totals, then view charts as tables. | Chart tables are for exact inspection; return to charts or use the Plan filter to select. |
| Operations | Select a service, brush latency, inspect requests and clear. | Heatmap counts failed requests; the error-rate denominator is the matching population. |
| Dashboard JSON | Validate a title change, then an unknown component, then reset. | Invalid edits retain the last valid dashboard; no model is called. |
| Formatting | Change locale/time zone, inspect exact chart values and compare CSV modes. | Display precision does not change measurements or provide arbitrary-precision chart arithmetic. |
| Dataset explorer | Load a snapshot, choose measure/group, search and export CSV. | This application explicitly links its local table search/filters to profiles and charts. |
| Cohorts | Choose cohort/channel and an interval; inspect membership. | Only completed calendar months enter retention denominators. |
| Model evaluation | Change threshold and segment, then inspect false positives. | Outcome/detail filters do not change the evaluation denominator. |
| Assistant | Request a local proposal, inspect, Apply, Undo or Reset. | Local mode is deterministic; changes wait for Apply. |
| Worker explorer | Initialize a small source, filter and download measurements. | Source/matching/returned/plotted counts differ; bounded points are not a representative sample. |
| Studio | Load a source, map fields, inspect React and download a starter. | Exports save configuration; the resulting app fetches the source again. |
Explore a real public API
Explore an hourly forecast, inspect earthquake events or compare country-year life expectancy. API recipes preserve the source, units and missing-value semantics of each provider. Use Studio to arrange the views visually and download a React starter, or copy the helper into an existing app.
Commerce and revenue
Help an operator answer “What changed, and where?” Put revenue and order count above a daily trend. Clicking a region or channel narrows the trend, KPIs and product table together.
| Question | View | Data contract |
|---|---|---|
| What changed? | KPIGroup + Histogram | One row per order; the daily histogram sums amount per date bucket. |
| Where did it change? | BarList + FilterBar | Region and channel on the same rows. |
| Which products explain it? | DataTable | Group by product; sum amount; trend over date. |
Open the storefront example or follow Building a dashboard.
Product and SaaS analytics
The runnable SaaS example contains 96 fictional account snapshots. Compare monthly recurring revenue by plan and region, inspect licensed seats against MRR, and follow the selected accounts in the table. It is a snapshot comparison, not an activity time series.
Define metrics before drawing them: event count is not unique active users, and summed subscription snapshots are not monthly recurring revenue. The example sums one MRR value per account and uses an unweighted mean for adoption. Calculate distinct counts and cohort retention in your data layer; pass the resulting measures with explicit labels.
Open the SaaS example. Use Funnel when rows represent ordered stages, and Heatmap for a precomputed cohort matrix.
Operations and service health
The runnable operations example contains 360 fictional request records. Inspect latency with a histogram and service box plots, compare failed-request counts by service/region in a heatmap, and inspect the matching requests. The mean of its binary failure flag gives the error rate.
Use Histogram for a latency distribution and BoxPlot to compare groups. Compute percentiles from raw observations or a mergeable sketch; averaging p95 values does not yield a combined p95.
Open the operations example. Label data freshness, the time window, the latency unit, and whether the data is sampled.
Generated analytical interfaces
Give a model a bounded dataset catalog and the Quartile JSON Schema. Validate its returned spec, check the requested fields against your catalog, then render the approved views. The application owns data access and authorization.
Open the spec-driven example and read Generating dashboards with AI. The example edits and renders a spec locally; connecting a model is an application concern.
For reviewed changes to an existing workspace, use AI-assisted analysis: bounded context, validated actions and explicit Apply. A runnable server example connects your own model credentials; no hosted service is provisioned.
Scientific analysis and model inspection
Inspect missingness and distributions, reconcile cohort retention with unique users, or change a binary model threshold and inspect false positives. The science workspaces use deterministic original fixtures and explicit denominators.
Bring an intentional pandas or Polars snapshot through Dataframe and notebook handoff. Use DataExplorer for local filtering, grouping, controlled edits and exact CSV export.
Embedded customer reporting
Place a QuartileProvider around a reporting surface, map its CSS tokens to your product brand, and keep the rest of your application shell. Each report can own a separate Selection to avoid filters leaking into neighboring reports.
- Fetch only the rows the signed-in customer may read. Browser filtering is presentation state, not an access boundary.
- Use the same units, currency and date granularity across views that share a selection.
- Provide explicit loading, error and empty states, and expose a table for precise values.
Continue with Framework integration, Theming and Linked selections.