Finite tokens are promises, not exact estimators

Let $J=J^{†}$ be a finite-dimensional source observable. Write

A token promise might specify a class $D_{\mathrm{prom}}$, a code assignment $t:D_{\mathrm{prom}}\to \mathcal{T}$, and a table satisfying a task condition on that class. Supplying $t(\rho )$ and constructing an instrument that obtains it from an unknown $\rho$ are different operations.