BOOK 2 · PART I · PLAY · STUDENT CHAPTER
Meet the Robots: The Sushi Menu
Twenty years ago Kevin built a way to order robot poker players à la carte: a grid of checkboxes, nine players across, a menu of strategies down the side. Tick a few boxes and you've written a robot's whole personality. Here's how to read the menu, what each row does, and the one thing no checkbox can make a robot do.
A menu, not a mystery
Kevin calls it the Sushi Menu. On a sushi menu you tick what you want and the kitchen makes it. On this one, each column is a robot, players 1 through 9, and each row is a habit you can give it. When you click Let the Battle Begin, nine robots sit down at the Advanced Simulator, each playing exactly the habits you ticked. You take the tenth seat.
Before any AI came along, this was how HoldemRobots made opponents: not by teaching a machine to think, but by writing down rules. It's the second step on the road from dice to AI, and the place where you learn the words that the rest of this book uses.
Reading the grid
| ROW | PL1 | PL2 | PL3 | PL4 | PL5 | PL6 | PL7 | PL8 | PL9 |
|---|---|---|---|---|---|---|---|---|---|
HOUSE BOTS: PICK ONE AND YOU'RE DONE | |||||||||
| Play against Robot Player: Pbot | □ | □ | □ | □ | □ | □ | □ | □ | □ |
| Play against Robot Player: Sbot | □ | □ | □ | □ | □ | □ | □ | □ | □ |
| Play against Robot Player: Ubot | □ | □ | □ | □ | □ | □ | □ | □ | □ |
| Play against Robot Player: Dbot | □ | □ | □ | □ | □ | □ | □ | □ | □ |
STARTING HANDS, SCHOOL 1: SKLANSKY | |||||||||
| Open/w sklansky1 … sklansky4 | □ | □ | □ | □ | □ | □ | □ | □ | □ |
STARTING HANDS, SCHOOL 2: WIZARD OF ODDS | |||||||||
| Open/w OddsWiz1 … OddsWiz4 | □ | □ | □ | □ | □ | □ | □ | □ | □ |
THE TURN TEST | |||||||||
| Turn/w Any Incr · 30% Incr · 2pair+ · Strt+ | □ | □ | □ | □ | □ | □ | □ | □ | □ |
THE RIVER TEST | |||||||||
| River/w Any Incr · 50% Incr · Strt+ · Flsh+ | □ | □ | □ | □ | □ | □ | □ | □ | □ |
The Sushi Menu's twenty rows, grouped. On the real page every row has a checkbox under each of the nine players, and each row's label links to a page explaining it.
Blank: the robot that never folds
Leave a column empty and that robot calls every hand all the way to the end. It never folds, and it never raises. In poker terms, that's a perfect Calling station: it will pay to see every card. It's also the easiest table to start on, because you'll see plenty of flops.
The house bots: a past they'd rather forget
The house bots have a history, and Book 1 is mostly about it. When Kevin brought his 2006 poker engine back to life, its robots finally felt smart. They were patient and disciplined, and they folded unless they were going to win. In July 2026 we found out why (Book 1, Part 1). Four of them were cheating. They read a variable in the code called frivbest, which held where every seat's hand would finish once all five board cards were out. They decided whether to stay in before the turn and river were even dealt, already knowing how those cards would land. They were reading the last page first.
Each one cheated its own way, and each wore the name of a craps bet, a leftover from the engine's sister dice game:
| CHEATING BOT | PLAYED THE HAND WHEN IT KNEW IT WOULD FINISH… | TODAY, BLINDFOLDED |
|---|---|---|
| Hardways | with the best hand at the table, outright | Still the tightest |
| Crapbetter | in the top three | Tight |
| Fieldbetter | in the top five | Looser, a bigger appetite |
| Hornbetter | ahead of seat ten: you | Solid and slightly aggressive, with no grudge |
A day later the cheat came out (Book 1, Part 2). Kevin's reason was simple: the human in seat ten was losing to a bot that peeked. We deleted the code that worked out the final ranking. That put a blindfold on the house bots: they can no longer see the future, only what every player is allowed to know, which is how strong their own two cards are with the board they can see. Then the plumbing inside the four changed from cheating to rules-based, on the AI's suggestion. Each now plays one of four flavors of known poker strategy, using only its own cards and the board. They kept their old range of tightness, from Hardways, still the tightest, to Fieldbetter, the loosest. They just play it blind to the future, like everybody else. Hornbetter didn't survive the change in one piece: its whole personality was "I only come after you when I'll win," and there's no honest version of that.
That's the difference between fair by promise and fair by design. A sign that says "no peeking" is a promise. The HoldemRobots table is fair by design: the dealer has no way to answer "who's going to win?", so no robot can find out. That goes for a house bot, a student's bot and the AI agent alike (Book 1, Part 15). Book 1 put it in one line: your score is yours to know; your rank is the cheat.
The house bots today
The Sushi Menu is twenty years old, and its house bot rows have been there since 2006. The rows didn't change; the plumbing behind them did. The top four rows are those four, blindfolded and rebuilt as rules-based robots: Pbot, Sbot, Ubot and Dbot. Each is a complete robot. Tick one for a player and you don't need to fill in anything else in that column. Each has its own explanation page on the site. The best way to learn them is the way you learned the AI agent: play them, and read their habits in Game Review.
Starting hands: two schools
Before the flop, a built robot has one question to answer: are these two cards worth playing? The menu offers two answers, from two different schools of thought, and they don't always agree.
| SCHOOL | WHERE IT COMES FROM | ON THE MENU |
|---|---|---|
| Sklansky | David Sklansky and Mason Malmuth sorted the starting hands into groups in their book Hold'em Poker for Advanced Players. Group 1 is the strongest: aces, kings, queens, jacks, and ace-king of one suit. Hands in the same group can usually be played the same way. | Open with Sklansky 1 to 4 |
| Wizard of Odds | A competing theory from the Wizard of Odds, a gambling mathematician whose work included checking the odds of new casino games for the casinos. His ranking is built on the odds of each starting hand. Like Sklansky, it has four levels, with the fewest hands at level 1. | Open with OddsWiz 1 to 4 |
Each row is a list of opening hands to play or not. In both schools, level 1 is the tightest: the fewest opening hands. Each level after it opens more. Tighter isn't better, it's a style: a tight robot plays fewer hands, a loose one plays more. The menu's colour coding shows which hands on each list are pairs, which are suited and which are unsuited.
Two experts, two rankings. Which is better? The menu lets you find out, the same way the backroom tournament compared AI companies: give two robots the same everything except the one thing you're testing, and let them play. The next chapter, "The Standard Table," sets that up for you.
The turn and river tests
Once the shared cards come out, a built robot asks a second question: is my hand still worth staying in for? The turn row and the river row each set a bar the robot's hand has to clear.
| ROW | THE ROBOT STAYS IN ON THAT STREET IF… |
|---|---|
| Turn/w Any Incr | the turn card improved its hand at all |
| Turn/w 30% Incr | its hand moved up the ladder of all possible hands by roughly 30% |
| Turn/w 2pair+ | it holds two pair or better |
| Turn/w Strt+ | it holds a straight or better |
| River/w Any Incr | the river card improved its hand at all |
| River/w 50% Incr | its hand moved up the ladder of all possible hands by roughly 50% |
| River/w Strt+ | it holds a straight or better |
| River/w Flsh+ | it holds a flush or better |
What does "30%" mean? Picture the ladder of poker hands from the refresher, high card at the bottom and royal flush at the top. The question is whether the new card moved the robot's hand up that ladder, and by how much. The percentages are a rough measure of that climb, and there's some judgment in them; don't expect a precise formula.
Notice the bars rise. "Any improvement" is easy to clear; "a flush or better" on the river is very hard. A robot with high bars folds a lot and plays like a Rock. A robot with low bars, or no bars at all, keeps going and plays more like a Station. There's no flop row: the menu has no test there, and a row left blank sets no bar, so a robot that has opened a hand keeps going until the next test it has to clear. Leave the whole column blank and it plays every hand to the end.
What the menu can't do
Every row on the menu answers one question: stay in, or fold? None of them tells a robot to bluff, or to check a strong hand to set a trap, or to bet more because the other player looks weak.
How much to bet is a weak spot too. In the twenty-year-old rules engine, bet sizes fall apart as the hand goes on: later bets just follow the same pattern as the earlier ones, whatever the pot and the board now look like. That's a leftover of how the code was written, not a strategy, and you can watch it happen in Game Review.
That means the menu can build every kind of player from the Calling station to the Rock, but not a Bluffer and not a Trapper. Keep that in mind. It's the first wall in this book, and it's why Part II moves from checkboxes to code, where you can write a bluff yourself.
At the table
- Open the Sushi Menu. Leave player 1 blank. Give player 2 a house bot. For player 3, tick Sklansky 1, Turn 2pair+ and River Flsh+.
Done when: three columns are set and the rest are blank. - Before you play, write down a prediction for each: which will fold most? Which will reach the river most often?
Done when: you've written three predictions. - Click Let the Battle Begin and play 20 hands. Keep your hand journal as usual.
Done when: 20 hands are played, or you've stopped early on purpose. - Open the game in Game Review. Count how often players 1, 2 and 3 folded, and how often each reached the river.
Done when: you can say whether your predictions were right.
Try it yourself first
- Player 4 ticks Sklansky 1, Turn Strt+ and River Flsh+. Player 5 leaves its column blank. Which plays more hands to the end?
- You want to know whether Sklansky or the Wizard of Odds picks better starting hands. How would you set up two players to find out?
- Can you build a robot that bluffs on the river?
ANSWER KEY
- Player 5. A blank robot calls all the way to the end, so it sees every hand through. Player 4 opens only with the hands on its Sklansky list, then needs a straight on the turn and a flush on the river: it will fold almost everything.
- Make them twins. Give one Sklansky 2 and the other OddsWiz 2, and tick the same turn and river rows for both. Then the only difference between them is the school. Play many hands, because a few hundred hands is mostly luck.
- Not with the menu. Every row decides stay or fold. A bluff means betting with a weak hand, and there's no row for that. You'll write one in code in Part II.
Written by Kevin Swinson with Claudette, the pen name for Claude, the AI from Anthropic that helped build HoldemRobots.AI. Row names are from the Sushi Menu page (betmenuadv.html) as of October 29, 2026; the two schools, the turn and river tests, the meaning of the percentages, levels 1–4 running from tight to loose, the Wizard of Odds' background, the OddsWiz levels, the rules engine's bet sizing and the house bots' rebuild from cheating to rules-based (suggested by the AI) are Kevin's description. The house bots' cheating and its removal are from Book 1, Parts 1, 2, 15 and 20. Sklansky's groups are from Sklansky and Malmuth, Hold'em Poker for Advanced Players; group 1 as listed by The Poker Bank.