Unit tests

The pure logic behind the hooks, tested in Node without a device: 53 tests across six files, run on every push.

A hook needs React and a phone, but the decisions inside it often do not. Which Pixel generation has UWB, how barometric pressure becomes altitude, how two embeddings are compared — these are pure functions, and they are where a wrong answer passes every static check. The type checker cannot tell that generation >= 12 should have been >= 11. A test can.

npm test          # node --test "test/**/*.test.ts"

Node 24 runs the TypeScript directly, so the suite adds no dependency. It runs in npm run verify and in CI on every push.

What is covered

File	Tests	What it proves
capabilities.test.ts	16	resolveCapabilities gives the right hardware per Pixel generation, reports nothing on a non-Pixel rather than guessing, opens Android API gates at the right level, and says whether its answer came from the model table or the device. verifyCapabilities lets a real device probe override the table.
observability.test.ts	23	traced returns and records on success, rethrows and counts on failure; tracedSafe returns the fallback but never swallows the failure; noteExpected is counted and deliberately not logged; normalizeError turns anything thrown into a message; the diagnostics getters sort and filter correctly; resetObservability clears everything.
altimeter.test.ts	5	The barometric formula behind useAltimeter.
embeddings.test.ts	4	The cosine-similarity maths behind useEmbeddings.
radios.test.ts	3	Logic in useWifi7MLO and useWifiRTT.
charging.test.ts	2	Logic shared by useChargingIntelligence and useADPF.

The capabilities suite has been mutation-checked: changing generation >= 11 to >= 12 in capabilities.ts fails two tests, which is the evidence that it tests the table rather than merely running it.

What is not covered

Anything that needs a renderer or real silicon. useObservability is the one React-dependent export in the observability module and is left out for that reason. Whether a hook actually reads hardware on a Pixel is the job of on-device verification with ARTEMIS.

One gap is worth naming because the tests look as though they cover it. The observability suite calls traced sequentially, and sequential calls are correlated correctly. Overlapping calls are not, and nothing here exercises them — see Traces for what goes wrong and how it was measured.