In a roulette game, each outcome is drawn from the same, well understood distribution, so we can easily calculate the probability that a sequence of outcomes is observed when the null hypothesis distribution applies.
In trading this is much harder, given the null hypothesis distribution is unknown and hard to estimate, and trades aren't i.i.d. since regimes come and go and therefore the luck of a specific style can be autocorrelated over a number of years. Trades are also correlated in the cross section through betas.
I'm going to go ahead and say that it's an impossible task if we're dealing with 10-20 year timeframes and the fund had a low frequency style. It's just way too easy to be lucky over a course of multiple years. All you really need is 1 lucky call combined with index tracking on the remaining capital to outperform ("overweight AMZN").
So you can't prove much about very passive investors.
But if someone is making a 2:1 payout decision every month, and manages to be right 75% of the time, you can be quite sure it's not luck.
If there aren't enough exploitable inefficiencies in an entire career to make that proof, then that's an interesting data point all on its own.
What you're requesting is that this happens in a specific way. That specific way might not be the way that is optimal.