# Editor engine benchmark

Generated 2026-08-26T15:48:25.516Z on Apple M4.

## Scope

- Widget-first: one rich-text widget containing all measured paragraphs, with Lexical loaded only after the first widget enters edit mode.
- Sizes: 100, 2000 paragraphs or blocks. 5 cold and 5 warm samples at CPU rates 1x, 4x.

These cases are product architectures, not feature-identical editor cores. Raw data records the cost of those choices.

## Production payload

| Engine  | Initial KiB | Initial gzip KiB | Initial brotli KiB | Async JS gzip KiB |
| ------- | ----------: | ---------------: | -----------------: | ----------------: |
| widgets |       211.2 |             66.9 |               57.7 |              99.7 |

## Artifacts

- `raw.json`: every browser sample and bundle file.
- `summary.json`: p50, p95, minimum, and maximum by engine, size, cache state, and CPU rate.

The final analysis is added after checking the raw distributions for noise and failed runs.