G-Host: Unbreakable wall for a GTO bot?
G-Host: My poker system concept which is supposed to be an unbreakable barrier for any NLHE strategy including GTO. It mirrors the strategy of the player not by duplicating it, but instead by dealing certain cards to the table and making certain actions. Everything is based on cards dealt from the "Mother Deck" and cards and actions rating system. G-Host can simulate from 1 to 5 opponents at a table.
The Mother Deck is a 52 cards deck, where:
Spades and hearts are CARDS:
♠ Good card for you
♥ Bad card for you
Clubs and diamonds are ACTIONS:
♣ Good opponent action for you
♦ Bad opponent action for you
The higher the black card from the Mother Deck, the better the card / action will be for you. Ace = best possible, Deuce = slightly positive
The higher the red card from the Mother Deck, the worse the card / action will be for you. Ace = worst possible, Deuce = slightly negative
So the G-Host uses two decks - Mother Deck (52) and Table Deck (52). By dealing one card from the Mother Deck, G-Host decides what the next Table Card or action will be. If there is more than one Table Card (like hole cards or flop) required, they will still be dealt "one by one" card way. Everything is based on YOUR PERSPECTIVE, so for example the hole cards that G-Host holds will be dealt at the showdown instead of the beginning of the hand. Because G-Host mirrors your strategy, all the cards and actions G-Host will deal and make, will be based on GTO strategy, which is considered to be the best one out there.
For example:
Ace♣ = best action from your point of view against you (that G-Host will make), according to GTO strategy
Ace♦ = worst action from your point of view against you (that G-Host will make), according to GTO strategy
Ace♠ = perfect card for you (that G-Host will deal)
Ace♦ = worst possible card for you (that G-Host will deal)
In a 6 handed game, there will be 17 Table Cards required and the maximum total number of player actions will be 35.
As you can see, a precise cards / actions rating system will be required to make a perfect G-Host. But even if there are multiple ways of doing this, there will always be balance between GOOD and BAD, so it should always be fair. The information which card from the Mother Deck was dealt, should stay hidden for the GTO bot, so that it doesn't know what kind of opponent it is playing against.
In a situation where you make an All-in bluff for example against G-Host and you want him to fold, the worse hole cards you're holding, the higher black card will be required from the Mother Deck to make this fold. And if you want him to call, the better hole cards you're holding, also the higher black card will be required.
I hope I explained this all well, because it's 5 AM here and I'm not sleeping today 
4 Replies
In a 6 handed game, there will be 17 Table Cards required and the maximum total number of player actions will be 35.
CORRECTION: total number of player (G-Host) actions will be 26 since the Mother Deck is divided into two 26 decks:
Deck A ("cards") 26 cards (spades and hearts)
Deck B ("actions") 26 cards (clubs and diamonds)
I'm not sure if this will be enough for a 6 handed table, but it's definitely enough for heads up.
I'm currently doing preliminary testing phase here with real deck of cards and it's quite crazy to play HU against an invisible player who mirrors your understanding of the NLHE game. All you need is:
- one 52 cards deck
- some memory training (remembering Table Cards, bets, stacks and pot)
- being honest and not cheating on yourself.
The overall experience is then surprisingly good and gives hope that this could really work against the GTO bot in a perfect computerized version !!
Spooky stuff 
I've found one more mistake in post #1 (Ace♦), so I decided to make this graphic:
I'm sure it's possible to precisely define the rating system for computerized version, although the level of complexity would probably be very high. But I think the most important thing is to not search for the ultimate, golden way of doing this. There are many different ways you can define those factors and rating levels, so the main thing would be to just keep proper balance between red and black cards and set 13 different levels.
I'm not that deep into GTO math, but I think:
- in cards rating these things will be important: hand strength, odds, outs, draws
- in actions rating these things will be important: stack sizes, bet sizes, pot odds, pot size
GTO assigns a certain range to your opponent. G-Host doesn't hold any cards before the showdown, but the whole perspective on the interface gives an impression that it does. This is enough to trick a GTO bot. From the bot perspective, the game proceeds normally and all according to the rules. Bot only sees that cards are dealt and the opponent is making his actions and holds 2 cards, so it will try to win the usual way - applying GTO strategy.
Basically GTO bot doesn't know anything about it's opponent in reality. It only "thinks" it does. I believe it could only get a result close to zero profit. Just as a human can only luckbox a GTO bot with variance in a match, the same thing is with GTO vs G-Host.
