Model specification: Statistical and calibration contract

Before data collection, declare:

1. a fixed number $N$ of attempted runs per nonidentity Pauli setting, with no deletion, postselection, or adaptive stopping;

2. the same channel instance and the same supplied oracle in every run, with independent fresh preparations and admitted environmental resets;

3. independent observations within each setting;

4. a bound $b$ on the absolute bias of each measured expectation from its ideal Pauli expectation on $\omega$;

5. a target Choi matrix ${\omega }_U=J_{\mathrm{n}}({\mathrm{Ad}}_U)$ known exactly or with an explicitly added target-description allowance.

The bias allowance includes preparation, basis-change, detection, and relevant calibration errors. It is not inferred from record agreement alone.