XLSX Benchmarks & Evidence

This page exists for one reason: CmdCal XLSX should be judged with evidence, not category claims.

The numbers below come from the in-repo benchmark and teardown work already captured during the XLSX buildout. They are useful for directionally comparing package shape, local generation speed, and current limits.

They are not a claim that every environment, dataset, or spreadsheet app will behave identically.

Benchmark Environment

Date: 2026-03-27
Node.js: v22.22.1
Platform: darwin arm64
Python: 3.13.5

Small Package Comparison

This run used one Revenue sheet with 5 rows and mixed strings, numbers, and dates.

Metric	ExcelJS Buffer	ExcelJS Stream	SheetJS	openpyxl	CmdCal
File size (bytes)	6640	6529	16791	5083	5264
ZIP entry count	10	10	10	9	10
`styles.xml` size	1624	1624	1199	2711	609
String cell encoding	shared strings	shared strings	raw string	inline string	shared strings

What this shows:

CmdCal was already competitive on package size in this run
the generated stylesheet was materially smaller than the other JS libraries in this comparison
shared strings stayed enabled without inflating the package shape

Large Workbook Smoke Benchmarks

These runs come from the in-package benchmark harness and are best read as local generation evidence, not a final CI reference baseline.

Dataset	p50	p95	max	rows/sec (p95)	cells/sec (p95)	file size
`large-10k`	330 ms	340 ms	340 ms	29,435	588,703	845,255 bytes
`large-50k`	1,789 ms	1,820 ms	1,820 ms	27,466	549,325	4,233,213 bytes
`large-100k`	1,965 ms	1,979 ms	1,979 ms	50,527	505,272	4,515,979 bytes

Styled Workload Snapshot

The styled benchmark matrix is intentionally stricter, and some targets are still above goal.

Current local highlights:

styled 10K×20 generation: about 937.2ms
styled 50K×20 generation: about 5055.9ms
column-width computation on 50K rows: about 173.1ms

This is useful because it shows where the real remaining local performance work lives:

styled serialization overhead
styles.xml size efficiency
width computation cost

What This Evidence Does Prove

CmdCal XLSX is not a toy serializer
package shape is already competitive on representative local runs
the engine has real benchmarking, preflight, planning, and render-metrics surfaces
large deterministic workbook generation is already practical

What This Evidence Does Not Yet Prove

final desktop-app compatibility across every target viewer
public renderStream(...) performance, because that surface is not public yet
full styled benchmark target attainment across the entire Phase 2 matrix

Current Validation Gaps

These are still blocked on real target apps or dedicated environments:

Excel for Windows verification
Excel for Mac verification
Google Sheets import verification
Apple Numbers verification
LibreOffice round-trip verification on a machine with a working app bundle

How To Reproduce The Shape Of This Evidence

Want runnable sample files: XLSX Examples & Downloads
Want the comparison angle: CmdCal XLSX vs ExcelJS
Want the planning APIs behind these numbers: Templates & Large Workbooks