Model specification: Complete-record statistical contract

Before collecting records, specify:

1. the channel, its preparation, every branch and failure outcome, and the measurement settings;

2. $N$ attempted observations per setting, with independent fresh runs and no unpriced postselection or stopping rule;

3. a bias allowance $b_{\mathrm{cal}}$ for each oriented mean;

4. a half-diamond transfer allowance $b$ between the tested channel and any separately named target implementation.

Independence, calibration, and transfer are assumptions to be established for the experiment, not consequences of simulation.