COVERAGE.md

Coverage

npm run test:coverage

Roughly 94% of statements and 98% of lines across src/core/.

Two files report far below that. Neither is an untested gap.

index.js reports 0%

A re-export barrel: no logic, only export { ... } from './x.js'. Jest instruments it, but nothing in it executes as a statement, so the counter stays at zero no matter how heavily the exports are used.

The barrel is still checked, by tests/smoke.mjs, which imports it and asserts every expected export exists and has the right type:

node tests/smoke.mjs

That check matters more than the percentage. A broken barrel fails every consumer while every unit test passes, because the unit tests import modules directly.

vendor.js reports about 41%

The dependency-injection boundary for the vendored browser libraries. Most uncovered lines are the failure paths that throw when a library was never registered. Their messages are long and specific on purpose: they are what a developer sees when a page loads bundles in the wrong order.

The registration path is covered. The failure paths are exercised where a test needs one. The rest exist to produce a good error, not to be measured.

What to watch

The numerical modules, not the total: statistics.js, structure.js, units.js, curve-fitting.js, outliers.js. A regression in any of those changes a published number.

One real gap: iso4.js at about 88% of statements, 80% of branches, 90% of functions. Unlike the two files above, that is genuinely untested code. It is also the newest module and the one smoke.mjs does not exercise.