EstimateBounds

Phase 4: cv_bounds planes, Monte Carlo coverage, and what the old estimator overstated

5/5 passed in 430ms

Test	What it does	Result
test_detector_costs_key	Against the detector-free variance, Bob's measured variance widens the interval and lowers the probe rate ($I_{AB}$ true, $\chi$ bounded) from 0.0662 to 0.0536 bit/symbol at $n={10}^7$ (24%), 6.3% at ${10}^8$, 6.9x at ${10}^6$.	✅ pass
test_finite_uses_bounds	cv_finite's (t_min, xi_max) is cv_bounds fed Bob's measured variance with ${\chi }_{det}$ referred back out, rebuilt here to match.	✅ pass
test_interval_covers	Over 800 Monte Carlo rounds the worst-case pair covers the truth at 0.979 for $ϵ=0.05$ and 0.883 for $ϵ=0.2$, each above $1-ϵ$.	✅ pass
test_no_key_at_small_block	At $n={10}^4$ to ${10}^6$ the reported rate is zero and the oracle rate positive.	✅ pass
test_plane_is_bobs	sigma2 at Bob's ADC times $2/\eta$ is $1+{\chi }_{det}+\hat{T}\hat{\xi }$, the channel-output plane cv_bounds takes.	✅ pass