Verified
Telixon measures three properties on every push: conformance with Google libphonenumber, performance, and payload size. Each section names the method and the command that reproduces it.
Conformance
Section titled “Conformance”Every query method is compared with Google libphonenumber. The oracle is version-matched: it runs Google’s own JavaScript source at the exact commit the engine’s metadata was built from, so a mismatch is a real difference and never version drift.
Three gates cover it: the full corpus, one number from every distinct case the library can
produce, and a one-million-input sample. The current run stands at 315,988 comparisons with
zero divergences: twelve methods, each matching on 11,232 of 11,232 corpus inputs, across all 245
regions. isValidForRegion is set-compared per case, 692,672 region checks. The allowlist of
accepted divergences is empty.
Weekly, the airtight proof enumerates the complete geographic input domain: 1,838,775,900 inputs, split across ten shards.
pnpm conformanceThe parity dashboard publishes every run.
Performance
Section titled “Performance”The benchmark measures parsing and query throughput, cold and warm, plus per-keystroke controller latency. The comparison baseline is the google-libphonenumber npm package, a third-party wrapper of Google’s port. Wall-time numbers depend on hardware, so this page quotes none. CodSpeed tracks instrumented regressions.
pnpm benchThe benchmark dashboard publishes each run.
size-limit gates the entry sizes. The engine tables ship outside the entries. The
async builds load them on demand; the sync entry embeds them.
| Artifact | Measured | Budget |
|---|---|---|
@telixon/core Node entry |
18.79 kB brotli | 22 kB |
@telixon/core browser entry |
21.34 kB brotli | 22 kB |
| Engine tables, loaded on demand | 123 kB gzip | none |
pnpm size