bench

Benchmarks

Micro-benchmarks for N34.apply and N34.stream throughput.

Files

File	What it measures
apply.bench.ts	Full corpus, per-language, and synthetic line-count scaling for apply
stream.bench.ts	Edge-case corpus, per-spec, and chunk-width scaling for stream
output.txt	Reference run captured on Apple M3 Pro, Deno 2.9.5

Run

deno task bench

Or bench one file directly:

deno bench --allow-read bench/apply.bench.ts
deno bench --allow-read bench/stream.bench.ts

Corpora

• apply reads specs from tests/code/ - 100 language samples across the file corpus.
• stream reads chunk specs from tests/edge/ - edge cases covering wrappers, whitespace, and streaming shapes.

Synthetic ladders (64 / 256 / 1024 / 4096 lines) verify scaling. Stream bench also sweeps chunk widths (1 / 16 / 256) to mimic LLM token cadence.

Notes

• No network calls, all inputs load from local test JSON.
• apply corpus and stream corpus groups run as baselines for relative iter/s.
• Timings vary with CPU load and thermal state; treat output.txt as a snapshot, not a contract.