LoggerJS Benchmark Matrix

Last updated: 2026-06-18 (two machine rows hand-merged; per-machine JSON artifacts live in the gitignored benchmarks/matrix/, so the aggregate command cannot regenerate a row whose artifact is not present locally).

This table is the checked-in evidence matrix for loggerjs-vs-pino Node hot-path claims. Ratios are paired per-round latency medians from the interleaved A/B harness, not one-off sequential-run ratios. A ratio below 1.00x means the LoggerJS path had lower latency than pino on that machine. The two rows below deliberately disagree — that is the point: the ranking is CPU/V8-dependent (LoggerJS faster on M1 Max, pino faster on M4 Pro), so no universal "faster than pino" claim is supportable.

Label	Platform	CPU	Node	Git	Runs	Pino ns	Lean ns	Prepared ns	Lean / pino	Prepared / pino	Result
macbookpro-node22	darwin/arm64	Apple M1 Max	v22.21.1	5d7e4e3bf423	5	286	244	223	0.843x (118.6%)	0.773x (129.3%)	LoggerJS lean + prepared faster
m4pro-node22	darwin/arm64	Apple M4 Pro	v22.22.2	1d3d51c21f86	6	197	223	211	1.137x (87.9%)	1.054x (94.9%)	pino faster (lean + prepared)

Row Details

Label	Memory	Dependencies	Sampling	Baseline spread	Prepared / lean
macbookpro-node22	64 GB	pino 10.3.1, winston 3.19.0, LogTape 2.1.3	5 runs, 120 rounds x 5000 ops, 100000 warmup	21.2%	0.919x (108.8%)
m4pro-node22	24 GB	pino 10.3.1, winston 3.19.0, LogTape 2.1.5	6 runs, 120 rounds x 5000 ops, 100000 warmup	41.9%	0.942x (106.1%)

Evidence Coverage

Requirement	Status	Rows
At least one non-Apple-Silicon runtime	Missing	darwin/arm64
At least two Node major versions	Missing	22

Both gates remain Missing: the M4 Pro row is a second Apple-Silicon CPU on Node 22, so it does not satisfy the non-Apple or second-Node-major requirements. It does, however, already make the universal-claim guard concrete — within the same OS/arch/Node-major, swapping only the CPU (M1 Max → M4 Pro) flips the result from "LoggerJS faster" to "pino faster". Until a non-Apple and a second Node-major row are added, keep performance wording CPU-scoped and never frame it as "LoggerJS is always faster than pino".

Reproduce

pnpm build
pnpm bench:matrix -- --runs=5 --rounds=120 --label="$(hostname)-node22"

# after copying artifacts from other machines into benchmarks/matrix/
pnpm bench:matrix:aggregate -- benchmarks/matrix --out docs/BENCHMARK-MATRIX.md

For non-Apple-Silicon and multi-Node evidence, run the manual GitHub Actions workflow:

gh workflow run benchmark-matrix.yml -f runs=5 -f rounds=120 -f batch=5000 -f warmup=100000

Download the benchmark-matrix-aggregate artifact, review the generated benchmark-matrix-ci.md, and commit it to this file only if the rows are from the intended machine/runtime combinations. Do not hand-write benchmark rows without the matching JSON artifacts.

Notes:

• The matrix proves only the listed machine/runtime/dependency combinations. Do not turn it into a universal "always faster than pino" claim.
• Add new rows when testing new CPUs, operating systems, Node/V8 versions, or pino releases.
• If a row is captured from a dirty worktree, mark the Git column with *.