ClickJitter

Phase 5b: detector timing jitter, window loss and bin error apart

8/8 passed in 1684ms

Test	What it does	Result
test_bin_error_law	Contiguous bins lose nothing, so bin leak is an error: QBER is half the leak and contrast falls to 1 - leak.	✅ pass
test_gaussian_has_no_tail	A Gaussian core alone gives window loss and no bin error, the leak at a 79 ps core in a 100 ps window being exactly 0.0 by the 9-sigma cut, the true 29-sigma tail 7.2e-175.	✅ pass
test_grain_invariance	The convolution reads neighbours across chunk boundaries by absolute index, so the record is chunk-invariant.	✅ pass
test_jitter_off	Window and tail time set with zero jitter and tail weight reproduce a jitter-free run bit for bit, at zero loss and leak.	✅ pass
test_rejects_nonsense	Refused by name: a window wider than the symbol period, a tail weight with no time constant or above 1, a negative jitter, and a tail spanning too many slots.	✅ pass
test_split_closed_form	The engine's Gauss-Legendre panels agree to 1e-12 with weights from math.erf and the exponential's antiderivative, out to slots of order $e^{-100}$ where a rational erf's 1e-7 floor would read as leak.	✅ pass
test_tail_raises_qber	At a realistic core width only the diffusion tail reaches the next slot, and growing it raises QBER and window loss separately.	✅ pass
test_window_is_a_loss	A gated Gaussian core only loses counts: the rate falls by the reported window loss to an eta-scaled run's, QBER unmoved.	✅ pass