ExpCaoVacuum

Tier B: Cao et al. 2026, the vacuum-decoy COW' measured from 25 to 100 km

8/8 passed in 1971ms

Test	What it does	Result
test_cow_certified	On all eight records the certified bound sits 19.7%-26.2% under the closed form, 1.36-3.85 times the asymptotic key length, 102.3 against 39.5 bit/s on the refined 100 km record; neither compares with their finite-size 29.0 bit/s.	✅ pass
test_cow_closure	Z-basis totals follow from their hardware to 1.3% at all four distances with data size in two-pulse rounds, and 58 kb over 29.0 bit/s closes at 2000 s.	✅ pass
test_cow_leak	The two records differ only in the |0>|0> counts, so the difference of their key lengths cancels every reconciliation and epsilon term; the model reproduces it to 3.3%-7.3%.	✅ pass
test_cow_ports	The |a>|a> gain is 0.98-0.99 of mu and the four Z gains within 11% of mu/4, fixing src/sdp.rs's ports convention against a doubled flux's 0.5, with 99% visibility.	✅ pass
test_cow_prefactor	Eqs. (5)-(7) with Eq. (7)'s prefactor equal _core.sdp_analytic to 1e-12, while Eq. (6) as printed reads 0.529-0.558 and aborts at all four published distances.	✅ pass
test_cow_record	Gao's closed form on their eight gains gives 0.929-0.978 of the phase error their key rate implies through Eq. (1), always below, and 4.8% and 5.1% low at 100 km with nothing fitted.	✅ pass
test_cow_reflection	Declared miss: the measured |0>|0> gain is 1142 times the 25 km dark floor, attributed to Michelson reflections, so the forward model reads 21%-35% under the recorded phase error.	✅ pass
test_cow_vacuum	The two records share six identical counts, and the two |0>|0> gains move the phase error by 8.3%-21.3%, most at 25 km.	✅ pass