Tabulator resource-sheet benchmark

This record defines the performance gate for Coleo's Tabulator migration. It measures the behavior Coleo owns instead of relying on a vendor maximum-row claim.

Pass thresholds

Workload	Automated threshold	Protected behavior
1,000 rows	50 ms	Resource-to-sheet projection
10,000 rows	250 ms	Resource-to-sheet projection
50,000 rows	1,250 ms	Resource-to-sheet projection
Interactive browser	30 s test budget	Initial render, virtual scrolling, dark editors, Golden Layout resizing
Live reconciliation	No history loss	React Query/server updates followed by undo and redo API calls
Repeated navigation	No runtime error	Tasks, Bugs, and Discovery mount/unmount through the browser suite

The 2026-08-03 reference run completed the projection workloads in 3.69 ms, 5.23 ms, and 28.00 ms respectively on the migration workstation. Thresholds remain deliberately wider so ordinary local and CI variance does not create a false regression.

The model thresholds run in src/web/__tests__/resource-sheet-model.test.ts. Playwright covers the rendering half of the gate with realistic Tasks, Bugs, plan-item, and Discovery data. The browser tests select and edit virtualized rows, resize a Golden Layout pane, reconcile mocked server responses, undo and redo edits and moves, and navigate repeatedly between resource sheets.

Implementation notes

• Tabulator renders only the visible DOM rows using its virtual renderer.
• React owns API resources and saved-view preferences.
• One stable Tabulator instance receives same-shape server and stream updates through updateData(), which does not add them to Coleo's history.
• Coleo owns edit and move history. Structural refreshes may replace rendered data without erasing user undo/redo actions.
• Column lookup and reconciliation use Map and Set indexes so a 50,000-row update does not introduce quadratic work.

Run the gate with:

bun test src/web/__tests__/resource-sheet-model.test.ts
bunx playwright test e2e/tasks.spec.ts