If your computer is playing perfect GTO where the rules are the same for both players, it is essentially assuming that the other player is also playing GTO. To do otherwise would mean that it would be playing exploitively against the other players assumed non GTO strategy which would make the computer itself exploitable. But you can stipulate to the computer that the rules of the game have changed such that the opponent MUST play certain hands badly. For instance he must always raise preflop with 72, 83, and 92 offsuit. Given that, the computer will alter its strategy from the original GTO strategy to a revised strategy based on those new assumptions.

My question is what are those new assumptions? Does it assume that it is playing a handicap match with Doug Polk who is required to make those raises? If so, its counterstrategy has to be against the Polk strategy that will include far more changes than just his preflop raises. Polk will adjust strategy on every street because of the requiredplay of those three hands. And if he is doing it perfectly, his adjustments will include his knowledge that you adjust.

Is that how these nodelocking machines work? If the answer is "yes" that is a feather in the cap of the programmers. But there is a problem. When you notice that a player is raising a lot preflop with hands that he shouldn't, running to your computer to ask how you should adjust to this will not result in finding a great counter strategy. Unless, of course you are sitting across the table from Doug.

Solvers are just EV-maximizing algorithms. Each "agent" in the solver is trying to make the most money possible. The solver isn't trying to play GTO, it just so happens that when you pit two exploitative algorithms against each other, they will eventually reach an equilibrium where neither can improve their payoff by changing their strategy.

When you nodelock, you force a certain strategy within one node (decision point) in the game tree. Every node after that is unconstrained. Hell, the agent may even change prior strategies leading up to the nodelock to minimize the damage of the terrible strategy you've forced it to play. You can lock multiple nodes, but the problem is there are so many runouts, and each runout has a separate strategy. Trying to lock them all is impractical.

You've touched upon one of the greatest weaknesses in modern solvers. Multi-street nodelocking and strategy trend modeling, is still in its infancy.

The latest techniques involve the use of "incentives" - an artificial bounty to incentivize agents to prefer certain actions. For example, if you artificially boost the EV of calling, that player will prefer to call more often, but must still contend with the opponent's counters to exploit this tendency.

Can somebody make this into one thread? Sorry I screwed up.