These thresholds are intentionally conservative and intended as regression guards, not absolute hardware-independent benchmarks.
- Hit-testing smoke (
packages/core/src/__tests__/perf-smoke.test.ts):- 2000 dispatches on ~300-leaf tree: <= 200 ms
- Caret geometry smoke (
packages/core/src/__tests__/perf-smoke.test.ts):- 5000 lookups across many measured lines: <= 300 ms
- Geometry diff smoke (
packages/server/src/__tests__/protocol-perf-smoke.test.ts):- 60 burst diffs on a depth-4 breadth-4 tree: <= 500 ms
When intentionally changing algorithms:
- Measure locally and in CI.
- Update thresholds with rationale in commit/release notes.
- Ensure limits remain defensive enough to catch accidental regressions.