RelayConditioned

CV-MDI: the two-party state the untrusted Bell relay leaves behind

10/10 passed in 4ms

Test	What it does	Result
test_cov_apart	Two attacks sharing one chi share cvmdi_rate to the last bit while their covariances and per-attack rates differ by more than 1e-3.	✅ pass
test_cov_asymptote	cvmdi_point approaches cvmdi_rate monotonically in va, to 3e-5 at va = 1e5.	✅ pass
test_cov_conditions	cvmdi_cov matches the same optics driven through the public Gaussian layer to 1e-12.	✅ pass
test_cov_correlated	A harming correlation costs key monotonically and beats two independent cloners, the opposite sign aiding detection.	✅ pass
test_cov_lossless	Lossless arms leave a pure two-mode state: chi_e under 1e-5, key the whole of I_AB.	✅ pass
test_cov_matches	cvmdi_cov reproduces the published closed form to 1e-12, correlated ancillas included.	✅ pass
test_cov_physical	cvmdi_cov satisfies V + i*Omega/2 >= 0 over 27 combinations of arm loss, environment variance and Eve correlation.	✅ pass
test_cov_rebuilds	The (i_ab, chi_e, key) triple rebuilt from the 4x4 and beta alone agrees with cvmdi_point to 1e-12.	✅ pass
test_cov_structure	The projection onto q_A = q_B, p_A = -p_B leaves a positive x-x and a negative p-p correlation, x-p even unless the attack is correlated.	✅ pass
test_cov_vacuum	The 4x4 is in SHOT-NOISE UNITS, vacuum 1 against the Gaussian layer's 1/2, so an unmodulated sender leaves the identity whatever the arms did.	✅ pass