That's why I suggested treating such orgs as an upper-bound, evaluating on realworld collected data on the intersection between "Amount of money in play" and "observed incidence rate of corruption"
Might even make it a 3D intersection, with a vertical axis of "Cost of administrative overhead", while you are at it.
It would then be possible to define a "rough" allowable size range, that by virtue of limits on size of operation, prevents that kind of wrongdoing, and which permits the maximal amount of proper action, and mitigates inefficiency from oppressive levels of oversight.
Rather than be predictive, it would be prescriptive in function. (Collect data first, use data to determine ideal size classes, enforce rules on size class to limit inappropriate behavior, restart cycle and amend as needed.)
If we combined that with some actually useful campaign finance reform (and killed super-pac bullshit with holy fire), we could limit the kind of international shennanigans that Clinton Foundation was up to, and by limiting the size of total operations to a monetary ceiling before requiring the charity to be split up, we can prevent "personal piggybank" shenanigans of Trump Foundation. (at the very least, we can detect it much more easily, because attempts to use it for large personal transactions/tax evasion, would stand out clearly, because the total fungibility of the foundation is seriously curtailed.)