Performance

This page describes what we measure on @cosyte/synth, how, and how to time it yourself. The shared method every Cosyte package follows (throughput, latency percentiles, memory, import time and bundle size, and how a pull request is compared with main) is written down in the performance standard in cosyte/config's decision records. The numbers for the current release are in Current numbers below.

What we measure

The benchmark suite in the repository's bench/ folder measures the built package on what callers ask of it: one artifact from a seed, and corpora of a hundred to tens of thousands of artifacts in each format, the synthetic patient identities under them, and the cost of a seed. It reads throughput, latency at the 50th, 95th, and 99th percentiles, allocations, peak memory, a long run for leaks, import time, and bundle size, on a GitHub-hosted runner.

Two CI jobs run on every pull request. A scaling gate fails the pull request when a corpus generator's cost grows faster than the number of artifacts it builds. A comparison times the pull request's build against main on one runner and reports the change; it is advisory and fails nothing.

What each call does

The work behind a call depends on which kind of call it is:

• A generator (generateAdt, generatePatient, generate837P, and the others) draws its values and calls the parser's builder. It does not read its output back through the parser.
• A round-trip helper (roundTrip, scriptRoundTrip, astmRoundTrip, and the others) serializes the artifact, parses it with the sibling parser, and serializes the parsed result again. The FHIR helper also validates the resource.
• A corpus (hl7Corpus, fhirCorpus, and the others) runs a generator and a round trip for every artifact, so it costs more per artifact than calling a generator alone.
• A quirk generator builds a clean artifact, applies the quirk, and parses the result once to check the intended warning.
• A pairing loop in @cosyte/synth/deid generates an artifact and runs @cosyte/deid over it.

If a test needs many fixtures and does not need their warnings, call a generator in a loop rather than build a corpus.

Import cost

The package root (@cosyte/synth) imports no parser. Each format subpath imports only its own parser, except @cosyte/synth/deid, which imports @cosyte/deid and five parsers. Import only the subpaths you use.

What these numbers are not

• Not a guarantee for your machine. Absolute times depend on the hardware and its load.
• Not a comparison with other libraries. That is a separate suite with its own written method.
• Not a load-testing tool. A seeded corpus is a fine input for a throughput test of your own code, but synth ships no benchmark runner.

Time it yourself

This program times one corpus of each HL7 v2 family on your machine. It prints the time and checks only that the corpus was built:

import { hl7Corpus } from "@cosyte/synth/hl7";

const start = performance.now();
const corpus = hl7Corpus({ seed: 1, count: 70 });
const elapsed = performance.now() - start;

console.log(`${String(corpus.artifacts.length)} messages in ${elapsed.toFixed(1)} ms`);
corpus.artifacts.length; // => 70

Current numbers

The numbers for this section are measured on main at each release and filled in when the release is built. They are never typed by hand.