If I were running a Ponzi scheme, and I needed to generate fake results, I might ask developers to create a program that generated a month's worth of random trades (perhaps explaining that this was a way to test the expected commission + slippage of a strategy that covered a given volume and mix of securities). Then I'd run it twenty times and pick the top result. Being in the 95% percentile every month was close to what Madoff did; he never showed extraordinary profits (except to people who were apparently in on it), but did show incredibly consistent returns of ~1% per month.

I don't think that's entirely clear. I've written my fair share of test data generation code that could have been used to fudge as many statements as you might want. One in particular even let the user enter targets (x% growth), so we'd have 'appropriate' data to demonstrate the charts and projection pieces.

So I'd like to think it's possible that they didn't realize they were complicit in fraud until after they saw it applied.

Taking the hush money kinda seals it though.