[mvanhorn]

Summary

Add fleet-size/ to translate a publication page-load capture into cores needed at chosen request rates:

cores = (CPU ms/response) × (target responses/s) / 1000

Optional --headroom (a fraction; 0.3 is 30%) and --max-cores to flag cells over a budget. This is a translation of the retained one-core CPU/response column, not a multi-core run.

Why

The HTTP capacity page only prints cores for 10,000 responses/s. A different rate, or a reserve on top of the measured CPU cost, currently means recomputing that table by hand. Keeping the formula on the publication capture, and refusing smoke or generator-limited files, makes the 10,000 responses/s table a special case of the same command.

How to try it

node --test fleet-size/fleet-size.test.mjs
node fleet-size/run.mjs
node fleet-size/run.mjs --target-rps 1000,10000,50000 --headroom 0.3 --max-cores 16

The default invocation matches the published 10,000 responses/s table. --json writes the structured plan. Exit status 2 means at least one cell exceeded --max-cores.

Notes

• Quality CI runs node --test fleet-size/fleet-size.test.mjs.
• Root README, BENCHMARK_PROTOCOL.md, and page-load/README.md point at fleet-size/.
• Measured-source fingerprints are unchanged: no edits under package.json or the HTTP/renderer source lists.

Demo