1: \begin{definition}[$\statalg_{\epsilon}$]

2: Given piecewise constant algorithm $\alg$ and $\epsilon > 0$, the {\em statistically ironed algorithm} $\statalg_{\epsilon}$ for $\alg$ with error $\epsilon$ is:

3: \begin{enumerate}

4: %\item Take $\epsilon' = \epsilon/4n$ (this term is motivated in the proof of Theorem \ref{thm.main.eps-bne}, below).

5: %\item Construct $\discalg_{\epsilon'}$, the discretized version of $\alg$.

6: \item Find the estimated allocation rule $\ests(\cdot)$ for $\alg$.

7: \item Find monotonizing intervals $\intsets$ for $\ests(\cdot)$ (as described in Section \ref{sec.iron}).

8:  %$\discalg_{\epsilon'}$.

9: \item Run $\ialg_{\intsets}(\vals)$.

10: %\item For each agent $i$, for the allocation rule $\esti$, compute the intervals $\inti$ from the definition of the ideal ironed algorithm $\ialg$.

11: %\item For each agent $i$, if $\vali \in I \in \inti$,

12: %  draw $\estvali~\sim {\disti}[I]$.

13: %\item run $\discalg_{\epsilon'}(\estvals)$.

14: \end{enumerate}

15: \end{definition}

16: