D3

Last updated: Mar 25, 2026

---

Rationale

D3.js and Billboard.js are used to render pre-computed analytics in the platform dashboard.

The main reasons why we chose them over other alternatives are:

• They are open source.
• D3 is the de facto standard for data-driven documents and gives full control over rendering and DOM; generators use it for selection and layout (e.g., d3.select, d3.selectAll).
• Billboard.js is built on D3 and provides ready-made chart types (line, bar, stacked bar, etc.) that match our needs and reduce custom code.
• The stack fits our analytics pipeline: pre-computed data is stored, then served to the front end; no server-side chart rendering is required.
• Charts run in the browser from static JSON produced by generators.

Alternatives

The following alternatives are being considered:

Evidence.dev

• Evidence.dev is an open-source BI framework (Markdown + SQL) that could leverage Snowflake, which we already use.
• Evidence Studio requires an Enterprise license to embed charts as iframes, blocking the primary integration surface (Graphic component).
• Evidence Studio does not support GitLab versioning; report definitions live only on Evidence's platform, conflicting with the team's standard workflow.
• Evidence OSS has no built-in access control; securing chart pages would require a custom reverse proxy or auth layer, adding significant infrastructure work.

Plotly

• Plotly is an open source charting library.
• It does not require a new service; charts can be built by using a SQL query to snowflake without pre-computing the data.
• Charts are interactive out of the box (hover, zoom, pan, click events).
• With the current D3/Billboard approach each graph takes normally under 1 second; Plotly charts take 2-5 seconds per graph without alru_cache, degrading the dashboard experience.
• The Plotly.js CDN bundle is ~3 MB, adding significant weight to each chart response inside sandboxed iframes.

Apache ECharts

• Apache ECharts offers many chart types and is another option worth considering.
• We stayed with D3 and Billboard for their smaller API surface and existing generator integration.

Usage

We use D3.js and Billboard.js to render analytics in the platform:

• Pre-computed chart data is loaded from storage and sent to the client.
• The front end loads D3 and Billboard and runs chart the generator logic for each visualization.
• Generators consume the data and use D3 and Billboard to draw the chart on the page.