The essay reads to me as 5000 words on why epistemic standards should be lowered for effective fraud investigation. Requiring retrospective evidence - the highest epistemic standard - is too slow and expensive, so the government needs to copy the private sector's epistemic standards where a collection of solid heuristics and smart observations is sufficient to preemptively block payment.

It's an insightful point, but it seems consistent with those epistemic standards that it's unimportant whether the 50% figure is valid in retrospect. The observation of "industrial-scale fraud" is sufficient to act, it doesn't need a retrospectively validated 50% figure and he would like you to get all the way off his back about it.