portal-echarts-dashboards

Aggregate data server-side with Polars group_by_dynamic so the endpoint returns bucketed points, then enable sampling: 'lttb' and large: true on the ECharts series. Never send raw rows to the client; request a target bucket and max_points from the aggregation endpoint.

Wrap charts in LazyVChart so they stay out of the initial bundle, and import only the needed modules from echarts/core such as LineChart, GridComponent, and CanvasRenderer. Never use import * as echarts, which defeats tree-shaking.

Define routeRules in nuxt.config.ts with swr values per route, for example '/dashboard': { swr: 60 } for instant perceived loads with background revalidation. Use ssr: true instead for analyst views that need fresh data per filter change.

Listen to chart events like datazoom or bar clicks and call a store action such as workspace.applyDrillDown with the extracted dimension, for example a date range or product. This keeps dashboard and chat state synchronized through the shared Pinia store.

Stutter usually comes from deep reactivity on large point arrays or missing large-mode optimizations. Use Nuxt useFetch shallowRef defaults for series data, enable sampling and large mode, and measure frame drops with a requestAnimationFrame counter to verify the performance gate.

The agreed degradation is 500K pre-aggregated points if pan/zoom over the full million-point series shows jank on team hardware. This must be an explicit team decision documented in the user story, not a silent change.