Performance

Numbers below come from local measurements (real Chrome, 6 mixed-type columns). The Node in-process regression baseline (SLA 200/500/1000 ms) lives in src/__tests__/performance.test.ts; the toBuffer cliff is verified out-of-band via independent processes (see the package README).

Benchmarks

Rows	Workbook (main)	Fast stream	What auto picks
10,000	~120ms	—	main + Workbook
50,000	—	~400ms	Worker + Fast stream
100,000	17.5s (legacy)	~780ms	Worker + Fast stream

Takeaways

1. Workbook.toBuffer() shows a superlinear cliff beyond ~55k rows (17.5s at 100k), while Fast stream stays at ~0.8s — hence STREAM_THRESHOLD = 50_000;
2. In the browser, ≥ 20,000 rows run in a Worker; the main thread only does one structured clone (~94ms at 100k rows);
3. The stream path's cost is missing styles/layout, so small files keep the fully-styled Workbook path.

Optimization tips

• Stay on mode: "auto"; never force main for large files;
• Keep styled exports under 50k rows; for larger data accept degraded styling or split sheets;
• Set decimals explicitly on number columns for consistent stored values across paths (Workbook/Stream/fallback);
• Run large server-side exports in background tasks.