Findings are the primary outputs of a Compass session — specific query results that have been given a title, description, and optional chart visualization.
How Findings Are Created
Every successful database query in a session is stored as a potential finding. The AI can use the PROMOTE_FINDING tool to give a query result a title, description, and optional chart configuration, which surfaces it in the Findings pane.
Editing Findings
Session owners can edit findings using the Edit button on each result card. Editable fields include:
- Title and Description — Describe what the result shows.
- Chart configuration — Add or modify a chart visualization.
- Column Formatting — Set per-column display labels, unit types (number, dollar, percent), and decimal precision. Labels replace raw column names in chart series, axis titles, tooltips, and table headers. Formatting applies to chart values and table cells.
- Default Display — Choose whether to show the chart, table, or both by default.
Findings can also be removed, which clears the title and description but retains the underlying query data.
CSV Export
Users with Allow PHI Download rights can export a finding's data table as a CSV file using the Export button on the result card toolbar. The exported file contains the raw query results exactly as stored — the same data shown in the data table.
Presentation Membership
Findings can be added to presentations in two ways:
- Add button — Each finding's toolbar includes an Add button that opens a dropdown of presentations you own or can edit. Select one to add the finding, or click New Presentation to create a presentation inline and add the finding to it in one step.
- Edit panel — The edit panel includes a Presentations section listing all presentations you own or can edit. Check a presentation to add the finding; uncheck to remove it.
See Presentations for details.
Data Integrity
Results presented in the main conversation panel are generated by the LLM and carry the inherent risk of hallucination — the AI may summarize, round, or fabricate numbers in its narrative responses.
Findings, by contrast, provide a layer of data integrity:
- Data tables contain the exact response returned by the database for the given SQL query. Compass stores and displays these results directly, with no LLM intermediation. That said, the SQL itself is AI-generated and may contain logical errors, so users should always review the query to confirm it answers the intended question correctly.
- Charts are hallucination-free visualizations rendered directly from the data table. The LLM's role is limited to choosing which fields to chart and which chart type to use when it promotes a result — the plotted values are always drawn straight from the underlying data.
This design means that while the conversational narrative should be treated as interpretive guidance, the findings panel is the authoritative source for verified query output.
Charts and Visualizations
Five chart types are supported:
- Bar — Category-based horizontal bar charts with up to 4 measure columns.
- Column — Category-based vertical bar charts. Same options as Bar (categoryFields, measureFields, stacked, altMeasureFields, dataLabels) but renders bars vertically.
- Timeseries — Line charts over time, with optional dimension grouping for multiple lines.
- Scatter — Two-variable scatter plots with optional point labels.
- Single Value — A single prominent number or value, suitable for KPIs.
All chart types support optional per-field formatting via the fieldFormat property in chart instructions. This lets the AI (or a user via the edit screen) specify unit type (dollar, percent, or number), decimal precision (0–6), and a display label for any column. Formatted values appear in tooltips, axis labels, data labels, series names, and data table cells. Display labels replace raw column names wherever they appear.