`event.timeStamp` and `performance.now()` are on subtly different time bases: `GlobalObject::performance()` samples `Bun__readOriginTimer` first and `MonotonicTime::now()` second, so `m_timeOrigin` overshoots the true origin by δ ≥ 0 and every `timeStamp` reads δ ms lower than `performance.now()` would at the same instant. The zero-tolerance `toBeGreaterThanOrEqual(before)` / `first >= after` assertions here therefore rely on JS overhead exceeding δ — normally trivial, but a preemption between t

`expect(stderr).toBe("")` will flake on ASAN/debug CI lanes, which emit benign warnings on stderr — REVIEW.md's subprocess-test rule explicitly says to assert a combined `{ stdout, stderr, exitCode }` object instead. It also runs before `JSON.parse(stdout)`, so a spurious stderr failure hides the actual output; the `navigator.%s is a getter-only accessor` test just below already shows the preferred pattern (`stderr: expect.any(String)` inside a combined `toEqual`).