Assumption: Conditional fixed-shot acquisition

A calibration record $\mathcal{Z}$ is collected before validation. Conditional on $\mathcal{Z}$:

1. $n\ge 1$, the calibration set $\mathcal{K}$, all scores, and all comparison allowances are fixed before the $n$ trial records are observed.

2. The trial records are independent. Their actual laws ${\tilde{p}}_i$ may differ, but on a calibration-measurable event $G$ they obey

   $$\eta \in \mathcal{K},d_{\mathrm{TV}}({\tilde{p}}_i,p_{\theta ,\eta })\le e(i=1,\dots ,n).$$

   Here $\theta ,\eta$ describe one fixed target family member.

3. The event has $\mathrm{Pr}(G^c)\le \beta$. Every attempted trial produces a complete record, including failure or rejection. There is no unaccounted postselection, validation-driven tuning, or adaptive stopping.