Estimation ledger

The named parameter and known calibration inputs are

The achieved and comparison quantities are

together with the finite-confidence radius $R_n(\alpha )$, or its implementation-adjusted counterpart.

The denominator $b$ is indispensable. If the calibration were changed so that $b=0$, the selected Bernoulli statistic would be independent of $\theta$, and its inference contract would fail. This is a statement about that statistic; some other record factor could still contain information in a different interface.

In particular, $\gamma =1$ would make the antiqubit readout constant, while the normal-qubit marginal could remain source-dependent. The main theorem avoids this degeneracy and does not conflate loss of one statistic with loss of every accessible parameter.