practice bots

Opponents that tell you what they do.

At a real table you spend hundreds of hands working out how loose someone is. Here the numbers are printed on the opponent before you sit down, which turns practice into something you can aim: pick the mistake you want to punish, then punish it on purpose. The numbers below are the whole opponent, and so are the limits underneath them.

eight opponents

VPIP is how often a bot enters a pot voluntarily, PFR how often it raises before the flop, AF how much it bets and raises compared to calling, and the last two columns are how often it continues on the flop and how often it gives up against a bet.

Rocky

rung 1, rock

Plays only his very best hands and never bluffs. Put pressure on and he folds.

· VPIP 12 percent, PFR 9 percent, 3-bet 2.5 percent
· Aggression factor 1.8
· Continues on the flop 65 percent, folds to a bet 70 percent

Station Sam

rung 1, calling station

Comes along with almost anything and almost never raises. He will pay you off, and he will also not fold when you want him to.

· VPIP 38 percent, PFR 8 percent, 3-bet 1.5 percent
· Aggression factor 0.9
· Continues on the flop 30 percent, folds to a bet 20 percent

Nina the Nit

rung 2, nit

Tight and careful. Slightly wider than Rocky and slightly harder to run over.

· VPIP 16 percent, PFR 12 percent, 3-bet 4 percent
· Aggression factor 2.1
· Continues on the flop 70 percent, folds to a bet 60 percent

TAG Tom

rung 3, tight aggressive

The solid regular. Few hands, but he raises with almost all of them.

· VPIP 24 percent, PFR 19 percent, 3-bet 8 percent
· Aggression factor 3.2
· Continues on the flop 75 percent, folds to a bet 50 percent

Maniac Mike

rung 3, maniac

In almost every pot and betting in most of them. Loud, and beatable if you stop trying to out bluff him.

· VPIP 48 percent, PFR 38 percent, 3-bet 18 percent
· Aggression factor 6.2
· Continues on the flop 88 percent, folds to a bet 25 percent

LAG Lola

rung 4, loose aggressive

Wide and aggressive, but not reckless. She keeps betting when you check.

· VPIP 32 percent, PFR 26 percent, 3-bet 12 percent
· Aggression factor 4.5
· Continues on the flop 80 percent, folds to a bet 35 percent

Hero Harry

rung 4, hero caller

Aggressive, and he calls you down light. Bluffing him is expensive.

· VPIP 30 percent, PFR 24 percent, 3-bet 10 percent
· Aggression factor 3.8
· Continues on the flop 78 percent, folds to a bet 30 percent

GTO Greta

rung 5, balanced

The most balanced set of numbers on the ladder, which makes her the hardest to exploit.

· VPIP 27 percent, PFR 22 percent, 3-bet 9 percent
· Aggression factor 3.5
· Continues on the flop 72 percent, folds to a bet 45 percent

If those abbreviations are new, they are explained one at a time in the glossary, and the starting hands each percentage corresponds to are drawn out on the preflop charts.

how a bot decides

No neural network and no solver. A defensible set of rules that actually uses the numbers above, so a VPIP of 12 really does mean twelve percent of hands.

A personality is seven numbers

Archetype, VPIP, PFR, aggression factor, 3-bet, continuation bet on the flop and fold against a continuation bet. Those are the numbers in the table above, and they are the whole opponent. Nothing else is hidden behind them.

Before the flop it works in percentiles

Every starting hand is scored and then placed in the ranking of all 169 combinations, weighted by how often each one is actually dealt out of the 1,326 possible hands. That turns a hand into a top percentage, which is what makes a VPIP of 12 mean something: Rocky opens when his hand is inside his PFR percentage, comes along inside his VPIP, and 3-bets inside his 3-bet percentage.

After the flop it works on hand strength

Facing a bet, a strong hand raises or calls depending on the aggression factor, a middling hand folds at a rate set by its fold against a continuation bet, and a weak hand mostly folds but fights back at a rate set by its aggression. With nobody betting, a strong hand bets, and otherwise the continuation bet percentage decides.

The same seed gives the same hand

Cards and bot decisions both come out of one generator seeded per hand, so a hand can be replayed exactly. The server keeps only the seed and your own actions and recomputes the rest, which means the bot cards are not sitting in your browser before a showdown reveals them.

what a match is worth

A ladder of five rungs

Rocky and Station Sam are the first rung, then Nina the Nit, then TAG Tom and Maniac Mike, then LAG Lola and Hero Harry, with GTO Greta last. A rung opens once you have won on the one below it, not once you have played it, because otherwise it is a click path rather than a ladder. It is a suggestion in the interface, not a lock on the database: if you want to sit down against the hardest opponent on day one, nothing stops you.

Your decisions are graded, and the answer key is on the server

Each graded decision is stored with what you chose, what the reference says, the street, the cards and the size of the pot, so a match can be reviewed afterwards instead of vanishing. The correct answer is looked up on the server rather than sent along by the browser, which is the difference between a score and a claim.

Good decisions pay more than winning chips

Finishing a match is worth 25 points. Decision accuracy adds up to 75 on top of that, and actually winning adds 15. A losing match played well therefore beats a lucky win played badly, which was the other way around until the reward was rebuilt. Poker punishes the opposite belief expensively enough already.

You can play your own mirror image

The mirror is the same machinery with your own measured frequencies in it, so you get to see how someone who plays like you is exploited. It needs at least 500 of your own hands, and the continuation bet and fold frequencies need at least 30 measured spots each or they fall back to a neutral value. The aggression factor is not measured anywhere in your hands, so it is not invented: it stays neutral. The mirror is deliberately not part of the ladder.

The mirror uses the frequencies measured from your own export, which is described on the import page, and the same measurements decide which leaks the detectors on the leak finder page are willing to report.

what a bot cannot teach you

Seven limits. They are properties of a fixed opponent, so they do not go away with a better version.

· Heads up only, both players starting on 100 big blinds. You are the small blind and the button; the bot is the big blind.
· The bot cannot see your cards. Its decision is made from its own two cards, the board, the pot, the bet it faces and its own stack, and nothing in the code passes it your hand.
· It does not weigh position. Position is not an input to the decision at all, which is defensible heads up where the seats are fixed and would be a real gap at a full table.
· It does not adapt. There is no memory between hands and no reading of you: the same seven numbers play the first hand and the hundredth, so once you find the hole it stays open.
· Its bet sizing is fixed. It bets around six tenths of the pot, raises to three times what it faces, and opens for three big blinds. Real opponents vary this on purpose.
· Beating one is not proof that you are good. You found the holes in one fixed strategy, which is a real skill and a narrow one. Human opponents change what they do the moment you start winning.
· The grading compares your action to a stored reference answer, not to a solver output. It is a teaching aid, and the page that shows the reference ranges says the same thing.

The score built from your real hands, and why a practice table cannot move the part of it that measures results, is explained on the Poker IQ page. What each plan opens is on the pricing page.

Play a bot before you read about one.

There is a table in the browser that needs no account and no download. The opponent there is simpler than the eight above, and the page next to it says exactly what it does and how often, which is the same deal as this one.

Play against the computer

questions

Do the bots cheat by seeing my cards?

No. The decision is computed from the bot own two cards, the board, the pot, the bet it faces and its stack. Your cards are not passed to it, and the hand is replayed on the server so the bot cards are not even in your browser until a showdown reveals them.

Which bot should I start against?

Rocky or Station Sam. They sit on the first rung because they are the two clearest lessons: one folds too much and one calls too much, and the counter to each is the opposite mistake. GTO Greta is last because balanced numbers give you the least to attack.

What is the mirror bot?

An opponent built from your own measured frequencies. Playing an archetype teaches you how to attack a nit or a maniac; playing yourself shows you how someone with your habits is exploited, which is the view you never get at the table. It needs 500 of your own hands before it will use them.

Does winning against a bot raise my Poker IQ?

Your practice work feeds the training side of the score, but the Poker IQ is built from your own played hands. A browser table against a bot does not produce a winrate, and pretending otherwise would break exactly the promise the score is meant to keep.

Can I try one without an account?

Yes. There is a table on this site you can play in the browser with no account at all, against a simpler opponent than the ones described here. Inside the app the first two personalities are open on the free tier.