Five Hands That Defined Phil Ivey — Graded

Hook

Phil Ivey made his name on hands where the cameras were rolling and the pots were enormous.

We put five of them through a solver 20 years later. The math says something more interesting than "he was right all along."

Three of the five, the solver would play exactly the way Ivey did. Two, it wouldn't — and both of those exceptions hide their own logic. One is a read that beat another player who was also off-GTO. The other is a sizing detail in a hand where the cards, not the sizing, decided the pot.

Across all five, Ivey never loses on a bad read. He wins two on reads. He loses three: one to a correct fold, one to a cooler, one to a bad river as an 89% favorite. The Ivey legend isn't what most poker fans think it is.

Hand 1 — Chris Moneymaker, 2003 WSOP Main Event

Ten players left. Ivey seven seats from the final table. In some alternate poker history, he makes it and the boom happens differently — or doesn't happen at all.

Solver verdict: standard value line. This is one of the three hands where the solver would play the way Ivey did.

Ivey went into the turn with roughly 23 big blinds behind in a 17.5 BB pot — a stack-to-pot ratio of 1.3. At that depth with a boat, the math stops being about sizing choices and becomes commit-or-fold. A small raise leaves an irrelevant fraction of chips behind; folding is unthinkable. The check-raise all-in is the line the solver plays itself. Ivey's 9♥9♦ even blocks nine combinations of Moneymaker's strongest possible holdings (Q9 boats, pocket nines) while leaving the A♣Q♦ trips wide open — near-perfect value-targeting.

The loss was the river, not the decision. Moneymaker needed a specific kind of card. Find the ones that would have done it:

The A♥ hit. Ivey's stack went to Moneymaker, and the 2003 Main Event got the final table that sparked the boom.

Hand 2 — Paul Jackson, 2005 Monte Carlo Millions

Heads-up for $1 million. Cardplayer called it "the craziest bluff in poker history." Both players, it turns out, had Jack-high by board pair and nothing of their own — and neither knew it.

Solver verdict: both players off-GTO. Ivey's five-bet shove with Queen-high on this board isn't something the solver would ever do at equilibrium — the spot is so rarely reached under balanced play that the combo-specific verdict is thin (see methodology caveat). The range-level finding is robust: on J-J-7 flop, a balanced 3-bet range is weighted 50%+ trips and boats, which makes Ivey's 5-bet a hero's-bet-only play. Jackson's 3-bet with Six-Five offsuit is a 100%-frequency fold in solver terms — plain range deviation.

Ivey read Jackson's air correctly. Jackson read Ivey's air as strength. Ivey wins the pot. Not via math. Via the meta-game of two players both deviating from GTO, where the one with the better read takes the pot.

Hand 3 — Tom Dwan, High Stakes Poker Season 6 — $676,900

The most famous fold on televised poker. A four-bet pot. Three barrels. A three-minute tank. A pair of sixes that happened to be the best hand at the table.

Solver verdict: clean fold at the range level. The specific combo (A♦6♦) arrives at this river after a 4-bet pot and a triple-barrel — a sequence with very low equilibrium reach, so the combo-level figure ("100% fold") is directional rather than precision-calibrated. The range-level verdict is the strong claim, and it's unambiguous: at a 66% pot bet, Ivey needs to defend 60% of his range (minimum defense frequency), and his range has enough sets + two-pair + top-pair-strong-kicker to hit that 60% quota without ever stooping to bottom pair. Go through the hand categories yourself:

Ivey folded the winner. The solver says the fold was automatic. The three-minute tank wasn't a calculation Ivey was doing; it was a calculation he was resisting. The mystique is the time spent, not the decision.

Hand 4 — Tom Dwan, Million Dollar Cash Game London 2009 — $1.1M

The largest pot ever won in a televised poker game. Guinness-recognized. A wheel on the turn, dominated by a 7-high straight. Chips were going in one way or another.

Solver verdict: call is preferred over shove at the range level. The exact A2-wheel combo on this runout in a 3-bet pot has vanishingly small equilibrium reach, so the combo-level EV figures below are directional. The range-level finding drives the verdict and holds: Dwan's turn raising range here is ~55% semi-bluffs that Ivey's wheel crushes and ~25% straights that beat him. Shoving folds out the profitable part; calling captures it. Before reading on, pick Ivey's play:

Did the sizing cost Ivey the hand? No. Once the turn brought both straights, Ivey loses this pot to Dwan's seven-high regardless. The solver's "call vs shove" preference is about size of loss, not fact of loss. This is a cooler. The solver's critique is about chip-extraction theory on a hand where chips are going in no matter what.

Hand 5 — Mike Matusow, Poker After Dark — January 22, 2007

Poker After Dark, $120,000 winner-take-all. Four clubs on the river. Ivey has no club. The most-cited Ivey-as-aggressor hand on TV — and, as it turns out, one the solver would make itself.

Solver verdict: range structure supports the check-raise. The exact A♠4♠-two-pair-no-club combo at this river node is solver-rare, so the combo-specific "is this check-raise in the solver's range here?" answer has wide confidence. The robust finding is Matusow's range-level shape: his betting range on this river is extremely top-heavy and, critically, contains zero bluffs. Against that range, a credible check-raise forces non-nut flushes to fold. Before reading on, guess Matusow's bluff frequency:

Against a capped value range with no bluffs, check-raising forces non-nut flushes to fold. Ivey's A♠4♠ also has blocker merit: the 4♠ removes nine of Matusow's boat combos (J-4, A-4).

Matusow's fold was also correct. The "worst fold in poker history" narrative is wrong. Against a player who can attack capped ranges — and Ivey demonstrably can — the king-high flush is a mandatory release. Both players are playing the spot as the solver would.

The pattern

Five hands. Two wins. Three losses. None of the losses is a skill error.

Across all five, the solver approves Ivey's action on three of them — the Moneymaker check-raise with 9s-full, the Dwan tank-fold, the Matusow check-raise — and disagrees on two: the Monte Carlo bluff war and the $1.1M turn shove. On the two disagreements, the deviations hide their own logic. Jackson's three-bet with Six-Five offsuit for half his chips on a paired board was itself a 100% solver fold; Ivey's shove only works because Jackson was playing the same off-GTO game. And the $1.1M turn shove loses the pot whether Ivey shoves or calls — the cards, not the sizing, decided.

This is the Ivey legend the solver confirms, 20 years later, with his opponents' cards face up:

Phil Ivey's iconic hands are mostly the hands the solver would have made itself. The exceptions were reads — and they were correct. When he lost, it was the river or the cooler, not his reads.

That's a better legend than "he was always right." It's the one that survives contact with the math. More interesting, too: Ivey wasn't ahead of the solver. He was making the solver's decisions before solvers existed — and when the decision was something else, it was because he saw something the solver couldn't.

Methodology & caveats

Hands graded: Each hand was constructed as a game-state JSON from public televised footage and cross-referenced poker-media commentary (sources per hand listed in the article's companion provenance files). Game states were run through QuintAce's decision_analysis_tool at each hand's key decision node. Outputs cited include solver action frequencies, opponent range compositions, minimum defense frequencies, and blocker-effect data.

Source triangulation (every fact cross-checked against at least two independent sources; only aligned facts used): The Moneymaker hand required a full correction on first pass — Moneymaker, not Ivey, was the preflop raiser; Jason Lester was a third player (with pocket Tens, folded on flop); and Ivey's turn action was a check-raise all-in facing Moneymaker's $200K bet, not an open shove. Paul Jackson's hole cards at Monte Carlo 2005 (where one source cited Ace-Jack) were confirmed as 6♠5♦ via ESPN, PokerTube, and Cardplayer convergence; the preflop action was also corrected — Jackson limped, Ivey raised with Q-8 suited, Jackson called (not a limped-check pot). The High Stakes Poker Dwan hand episode was confirmed as Season 6 Episode 5, with Ivey's holdings confirmed as A♦6♦. The $1.1M MDCG preflop sizing was corrected to Dwan open $6K / Ivey three-bet $23K (from earlier estimates that were too large). Where sources disagreed on a specific detail, the detail was either dropped or replaced with the version that appeared in multiple sources.

Solver model: QuintAce's cash NLHE solver. For tournament hands (Moneymaker WSOP ME, Paul Jackson Monte Carlo final), the solver approximates — it doesn't model ICM (bubble pressure, pay jumps). For heads-up play already locked into top-two payouts (Hand 2), the approximation is tight. For the Moneymaker final-table bubble (Hand 1), ICM would add defensive pressure on Ivey as the short stack; the solver's "chip-EV" grade is the lower-bound reference, not the full tournament-EV answer.

Reconstructive sizing: Preflop action was reconstructed to match each hand's documented flop pot, since public commentary doesn't always preserve exact preflop sizings. Reconstruction is pressure-tested against the visible pot size at the decision point; minor preflop sizing differences do not change the decision-node solver outputs.

What "mistake" means in this article: the solver grades actions by their expected value over distributions of cards. When we report an action as a "0% GTO frequency" play, we mean the balanced solver strategy would never choose that action. That is not the same as "Ivey lost this hand because of this action." In Hand 4, the solver's disagreement with Ivey's sizing does not change the outcome of the specific hand — the result is determined by the cards. The solver is a theory grader, not a results grader.

Reach verification and low-reach caveats (important for this article). Every hero-specific solver output carries a hero_reach value — the frequency at which this exact hand reaches this exact node under equilibrium play. Reach ≥ 40% is the target for a confident combo-level verdict; ≥ 5% is the minimum trust threshold; below 5% the output is extrapolated from a sparsely-explored part of the solver's solution tree and should be read as directional rather than precisely calibrated. For this article's five hands, reach was measured via solver_tool:

Why Hands 2-5 are low-reach: these are iconic televised hands where at least one player was playing non-GTO by the solver's definition. In equilibrium simulations the solver rarely explores spots where (a) both players take strongly polarized / off-GTO actions that wouldn't occur at equilibrium frequency, or (b) the exact combo + stack + board conjunction is rare. The solver's retrospective verdict on such hands is informative — but the robust claim is at the range level, not the specific-combo level.

Concretely for this article: the claim "Ivey's calling range against Dwan's triple-barrel is sets + two-pair, and pair-of-sixes isn't in it" is a range-level finding and holds independent of reach. The claim "the solver folds A♦6♦ exactly 100% here" is a combo-level statement from a ~0%-reach node — directionally correct (bottom pair is far below the calling threshold) but not a precisely calibrated frequency. The same shift — emphasize range, hedge combo — applies to Hands 2, 4, and 5.

The range-level findings are what the article primarily leans on. The widgets (5-category classifier, 3-way EV picker, frequency dial) all present range-composition arguments rather than combo-specific verdicts — deliberately so, in light of the reach findings.

On the 9% river figure (Hand 1): with Moneymaker holding A♣Q♦ on a Q♥6♠Q♠9♣ turn, his outs to improve past Ivey's 9s-full are the three remaining Aces and the one remaining Queen — 4 outs of 44 unseen cards, approximately 9%. Some popular commentary on the hand cites 15–17%; that reflects probability counted from an earlier street or a broader reading of outs. We use the direct-outs figure throughout.

Solver grading by QuintAce. Hand provenance and per-hand solver runs documented in the hands/ folder of this article's workspace.