Well the first thing you need is some sort of randomiser and both players admitted to using them. I'm not an expert on Hold'em (only play Pot Limit Omaha) but there are only 169 distinct starting hands so many hands are mixed strategies (e.g. A7 suited could be a raise 64% of the time, a call 25% of the time and a fold 11% of the time in a particular spot). Mixed strategies can also happen in PLO but are
a) not really relevant as nobody is going to exploit you if you play one of the 16,432 possible hands a certain way in a certain spot and
b) often the result of solver output which hasn't fully converged.
Computing power is still an issue with a game as large as PLO. I have rented a server with 512 RAM in the past and even then you work with abstractions ("strength buckets") and simplified postflop lines (e.g. only allowing one or two bet sizings). Though once you have solid preflop ranges (preflop converges relatively quickly) you can set up postflop scenarios using these ranges and run them with mainstream hardware. But heads-up NL Hold'em which is what they were playing is solved for all practical purposes.
Strength buckets bring us to heuristics. A practical example from PLO: You hold AKKx with the ace of spades and get called by a player in position. Flop comes down Kxx with two spades giving you top set and a blocker to the nut flush draw. You check (the out of position player checks almost their entire range here as K high boards are much more favourable to the preflop caller), your opponent bets and you have to decide whether to raise or just call. From playing around with solvers I have found top set with the nut flush draw blocker virtually never raises here and always calls as raising sacrifices the value of your blocker which you want to use on future streets to potentially turn your hand into a bluff if another spade rolls off. This is my reading of the solver output of course. The solution doesn't come with explanations, it is merely the result of iterating through the game tree millions of times and see which strategy yields the highest EV. On the other hand loads of other KK combinations want to raise here and get as much money in as possible. Including top set with say a 9-high flush draw as the value of calling these mediocre draws out of position is marginal. These sort of heuristics can be studied (there is training software now as well which can process the raw data and make it visually appealing and comprehendable for humans) but it takes a lot of effort and anyone trying to get into the game at this point really has to be willing to grind a lot of hours. Don't think it's worth it and it's not getting better. Any type of live assistance is prohibited of course but there have been bots for as long as there's been online poker and the only thing that stands in the way of true GTO bots is the availability of sufficient computing power imho.
Don't know how well versed you are in poker. Hope that wasn't too confusing.