The 5 Questions Patrik Antonius Would Ask a Modern Solver

⚠️ Review note — remove before publish. The five questions below are QuintAce's curation, sourced from Patrik's public remarks (podcasts, interviews, Triton player-cam moments). Patrik has not yet endorsed, swapped, or supplemented them. This draft lands on internals.quintace.ai for his review; v2 incorporates his cuts, added questions, and first-person reactions. Where quotes are attributed to Patrik, they are verbatim from the cited public sources; no reactions have been fabricated.

§1 — The Premise

Every other article in our pipeline hands a coach our findings and asks them to react. This one flips the axis. Patrik Antonius is the rare pro who could plausibly tell the solver where it's pointed wrong. So we looked for what he'd actually want it to settle.

We couldn't ask him the question directly without blocking v1 on his schedule. Instead, we pulled from his public analytical record — interviews on GipsyTeam, SoMuchPoker, Spade Poker; the Brandon Adams podcast; his Durrrr Challenge retrospective; the strategic themes he's remarked on in televised broadcasts and written columns. Five questions emerged, each anchored to something he said publicly. We ran QuintAce's solver on each. This is what came back.

Patrik reviews the live draft. If a question doesn't ring as his — he swaps it, we re-run, v2 ships. If he wants to add a sixth, we solve that too. The curation below is our honest best guess at his inquiry. His redirect is welcomed, not feared; where he pulls the piece next is part of the editorial.

§2 — Question 1: "Show me the spots where my intuition is usually right — but wrong."

Grounding: "When we look at solver outputs, I usually already know what the computer is going to do, which hands it uses, how it sizes bets, how it balances." — Patrik Antonius, GipsyTeam interview, January 2026.

The twist underneath this remark is that Patrik trusts his baseline. He isn't using the solver to learn — he's using it to audit. Which makes the interesting question not "what does the solver teach" but "where does it catch me out." Take a canonical spot where every experienced player has a pat answer, and see whether the pat answer holds up.

Test spot: 6-max no-limit hold'em cash, 100bb effective. BTN opens 2.5bb. BB 3-bets to 10bb. BTN calls. Flop comes K♦ 7♣ 2♥ — dry, rainbow, no draws. BB's c-bet decision. The pat answer among experienced players: "auto-small c-bet, high frequency." The solver's actual prescription looks close — but diverges on three load-bearing details.

What the solver says for BB's range:

Source: QuintAce decision analysis on stored hand patrik-q1-btn-vs-bb-k72r-100bb (BB AsKh, pot 20.5bb, SPR 4.39 at flop). See Methodology.

Three dimensions where the intuition misses:

1. "Auto" isn't auto. The range checks 26.9% of the time. A range-bet strategy — which is what "auto-c-bet" literally describes — is 23 percentage points more aggressive than GTO, and opens the c-betting range to exploitative probing from a solver-literate caller.

2. "Small" isn't small — it's a near-equal mix of small and medium. The 33%-pot and 50%-pot sizings run almost tied (34.6% vs 33.0%). Pre-solver intuition tends to pick one; the solver wants both.

3. Premium value prefers the LARGER size. A♠K♥ — top pair, top kicker — mixes 45.5% at 50% pot, 30.1% at 33% pot, 16.6% check. The inherited intuition that big hands "trap with the small bet" reverses: at this SPR, premium value wants to build the pot, not disguise itself.

Reading for Patrik: the intuition hits the right region — aggressive, mostly small — but the coordinates are off. 27% check range; mixed sizing; and on your actual premium combos, the bigger size is preferred. Whether that 27% check range matters in practice against your specific opponent is the kind of adjustment no solver will surface — but the fact that it's there at all is exactly the "audit" signal you said you were looking for.

§3 — Question 2: "At 200bb, does the solver even know what it's doing?"

Grounding: "I used it a lot for heads up regarding these 10 to 30 bb situations... When you play deep stacks, then poker gets much more complex, and you always have to play based on the player." — Patrik Antonius, SoMuchPoker interview, September 2024.

Triton cash runs 200bb+ effective. If the solver is a 10-30bb HU tool, the piece breaks at Patrik's actual working depth. We ran the Q1 scenario again — same preflop, same board, same decision — at 200bb instead of 100bb to test the claim.

What shifts from 100bb to 200bb:

Source: QuintAce decision analysis, stored hand patrik-q2-btn-vs-bb-k72r-200bb. SPR at flop rises from 4.39 to 9.27. See Methodology.

The structural finding is half-in-Patrik's-favor, half against.

What holds up (against the claim): the solver doesn't break. Same action tree, same strategic logic. The foundation — aggressive in a 3-bet pot on a range-advantaged dry board — remains aggressive. The framework scales.

What shifts (toward the claim): the prescribed strategy is meaningfully different. Check frequency grows five points. Sizings compress smaller — 75% pot retreats from 5.5% to near-zero; 33% pot grows by 8.5 points. The AK combo that preferred 50% pot at 100bb flips to prefer 33% pot at 200bb — inverting the preference a player who memorized the 100bb answer would default to.

Reading for Patrik: you're half-right. Deeper is more complex, and the solver's output is quantifiably different; porting the 100bb answer to 200bb uses the wrong size with your premium value hands. But "the solver doesn't know what it's doing at 200bb" doesn't hold — it knows, the answer is just a different answer than at 100bb. The teaching opportunity is in the difference, not in dismissing the output.

§4 — Question 3: "Humans under-bluff rivers. How much should I over-fold vs the solver's defense frequency?"

Grounding: "Humans under-bluff a lot of spots. Not every spot, but many of them... If you try to play like a computer against humans, you'll make a lot of bad calls and be wrong often." — Patrik Antonius, GipsyTeam interview, January 2026.

This one is less a solver query than an arithmetic exercise with Patrik's worldview attached. The solver's GTO defense frequency — minimum defense frequency, MDF — is a mechanical answer: against a bet of b into a pot of p, hero must defend at least p / (p + b) to keep villain's bluffs indifferent. Against a pot-sized bet, MDF is 50%. Against a half-pot bet, MDF is 67%. That's equilibrium math — it assumes villain's bluff frequency matches their value frequency in solver-correct proportions.

In the real world, most live and recreational pools bluff less than GTO prescribes. Community research and population-study data from solver providers put the under-bluff tax at roughly 20-30% across most river nodes: if GTO bets a given river with 30% bluffs, the typical population bets it with 20-22% bluffs, occasionally lower.

The implication is straightforward. Against a pot-sized river bet where MDF is 50%:

• GTO villain (equilibrium bluff frequency): hero calls 50% of the range — indifferent
• Typical population (under-bluffs by ~25%): hero's effective break-even call is ~58-62% of the bluff-catching range that beats value — i.e., hero should over-fold by ~8-12 percentage points relative to MDF

The arithmetic is unforgiving and scales with bet size. Against an overbet — pot-and-a-half — MDF drops to 40%, and an under-bluffing villain pushes the adjusted break-even higher still. Patrik's instinct here is quantifiably correct: the solver's call frequency is an upper bound against an unknown villain, not the default. A pro playing against a population that reliably under-bluffs should be folding more than the solver prescribes by roughly the gap between GTO bluff frequency and observed population frequency.

Reading for Patrik: MDF is a floor vs GTO, not a recommendation vs humans. The 20-30% under-bluff tax you've observed in the field maps directly to an 8-15pp over-fold relative to solver output on big river bets. The solver output is an auditing reference for the balanced case; your stated adjustment is mathematically defensible.

⚠️ Review note — remove before publish. We didn't run a specific river MDF scenario through the solver for this section. The arithmetic above uses canonical MDF formulas and cites community / population-study estimates of under-bluff magnitude. If you'd like us to run a specific hand through the solver for v2 — e.g., a river spot where you want the exploit number calibrated to an actual nodelocked villain range — send the scenario and we'll solve it.

§5 — Question 4: "Which spots have sharp EV cliffs — and which are flat plateaus where 'close enough' is fine?"

Grounding: "As long as I'm close to the mathematically correct decision, that's good enough for me. GTO is really important, but it can also make your game go off the rails." — Patrik Antonius, PokerNews 7 Tips, December 2021.

The "close enough" philosophy assumes the error function around the right answer is flat. Sometimes it is. Sometimes it isn't. The solver can tell you which is which. We extended the Q1/Q2 scenario — BB 3-bet, BTN call, K72r flop c-bet 33% called — to a 5♣ turn to capture the turn sizing decision for AK.

Turn sizing tree, BB with A♠K♥ after flop c-bet called:

Source: QuintAce decision analysis, stored hand patrik-q4-turn-sizing-k72r-5c. Pot 34.1bb, effective 83.2bb, SPR 2.44. See Methodology.

The finding is exactly the kind of teachable asymmetry the "close enough" frame hides:

The plateau. At equilibrium, any action assigned a non-zero frequency in a mixed strategy has identical expected value — that's what "mixed" mechanically means. Checking, 50% pot, 75% pot, and 100% pot are all in the mix for AK. All four have the same EV. You can pick any of them without losing money. "Close enough" is literally true within this cluster.

The cliff. The 33% pot bet is assigned zero frequency. That's a pure strategic error. The reason: at SPR 2.44, this is commit-or-fold territory with top pair top kicker. A small bet under-charges the 38% of BTN's range that's weaker pairs, applies insufficient pressure on newly live flush draws (8.1% of BTN's range after the 5♣), and fails to geometrically set up a smooth river shove. You lose meaningful EV any time you pick the 33% option, and the solver's 0% frequency assignment is how it tells you so.

Reading for Patrik: the "close enough" philosophy holds within a cluster of near-optimal sizings. Step outside the cluster and the error function is not flat — it's a cliff. The actionable refinement isn't "memorize a single right size" (the mixed strategy confirms there isn't one); it's "memorize the cluster boundary so you know which sides are in and which are out."

§6 — Question 5: "Where is the modern cohort over-fitted to the solver? What's the counter-exploit?"

Grounding: "I try to play in ways my opponents are not used to. I want to take them into areas where they're more likely to make mistakes." "Even overly studied players can lack other important elements of the game. They play by numbers... they are stuck too much with their fundamentals and their mathematical approach." — Patrik Antonius, GipsyTeam interview, January 2026.

The cohort default on a dry K-high 3-bet-pot flop is range-bet-small — a memorized answer that predates the solver and has calcified around 33% c-bet at 100% range. Q1 showed what the solver actually does in that spot: 73.1% c-bet with nearly-equal mix between 33% and 50% sizings, and a 26.9% check range that protects the check tree.

A villain who range-bets 33% on K72r in this configuration is making three measurable deviations from GTO:

1. C-bet frequency too high by 26.9pp — the 27% check range is absent from their strategy
2. Sizing too narrow — 100% of their bets are the same 33% size, vs the solver's ~35/33 mix of 33% and 50%
3. Premium value under-bet — their AK is c-betting at 33% when the solver's AK prefers 50% (45.5% frequency)

What each deviation gives hero, in order of exploit value:

Against the missing check range. When villain range-bets, their betting range is uncapped — they bet AA, KK, AK, and 72o at the same frequency. But their check range is null, so whenever they check, that's a tell about their strategy, not a strategic choice about that specific holding. This is rare; more typical is that a villain only half-implements the memorized answer and develops a small check range organically (out of laziness or hand-reading). Still, any villain with a narrow check range is check-raising over-condensed, and hero can steal more turn/river with position.

Against the missing sizing split. If villain c-bets 33% with 100% of their c-betting range, their bet range contains more trash than GTO's 33%-pot betting range, because GTO uses the 50% size selectively for value and for specific bluffs. Hero should call the 33% bet wider on turn and float more liberally — villain's range is less protected than the solver's.

Against the under-bet premium value. A villain who c-bets 33% with AK is under-extracting from their best hands. Hero's turn and river ranges face less pressure than GTO would apply, so hero's continue-range is wider in practice. This compounds: wider defense against turn barrels + wider bluff-catch on rivers.

Reading for Patrik: the cohort-memorized "range-bet 33%" answer is close enough to the solver that it rarely gets exploited in casual analysis, but the exploit surface exists and is measurable. The counter is not to deviate from GTO — it's to play GTO against a solver-rigid opponent, because GTO happens to exploit their specific rigidity. Where you take a solver-rigid player into trouble is in the spots where the solver mixes — the mixed-strategy decisions — because a rigid player picks one path and gives the read away.

§7 — The Pattern

Five questions, grounded in five different verbatim remarks, lead to a single coherent position. Patrik treats the solver as a reference point, not a playbook. The intuition he trusts is calibrated against the solver, not by it. The audit either confirms the instinct, sharpens the coordinates, or surfaces a blind spot — and the actionable move is different in each case.

The questions themselves cluster around the same structural concern: the gap between the memorized answer and the correct answer. Q1 finds the gap at the range level (73% aggressive, not 100%). Q2 finds it across stack depths (the 100bb answer misrepresents the 200bb spot). Q3 finds it in the bluff-catching asymmetry (MDF is a floor, not a recommendation). Q4 finds it in the sizing tree (plateau within a cluster, cliff outside it). Q5 finds it at the cohort level (memorized rigidity is exploitable by GTO itself).

Across all five, the solver never disagrees with Patrik's posture — that you can't play like a computer against humans, that "close enough" is directionally right, that deep stacks are more complex. What it does is add precision. The intuition points at the target; the solver gives the coordinates.

The question we didn't ask — but that Patrik's public record suggests he'd ask if pressed — is which of these five questions costs him the most in practice. Q2 (stack-depth porting) has the biggest per-decision EV at stake. Q4 (sizing cliffs) is the most preventable with study. Q3 (over-folding vs an under-bluffing field) is the highest-volume adjustment, occurring every river in a live pool. We'd be curious where he'd rank them.

§8 — Takeaway

For the intermediate-to-pro reader, the structural lesson isn't a sizing or a frequency — it's a posture. Solver study pays off most when used as a calibration tool against your own developed intuition, not as a rote memorization exercise. Patrik's inquiry direction — "find where I'm wrong" — is a more efficient use of solver time than "tell me the right answer," because the right answer for any single spot is already approximately known to anyone who's played seriously. The leverage is in the residual error.

The actionable version: when you run a spot through a solver, don't ask "what should I do here." Ask "where is the memorized answer wrong, and by how much." The 5 questions above are that inquiry, run on behalf of one of the best cash players of the last two decades. His answers — the ones that matter — are in the v2 he writes on top of this draft.

Methodology and caveats

Source of questions. The five questions are QuintAce's curation, each anchored to a verbatim Patrik Antonius quote from a dated primary source. Citations inline. Patrik did not pre-approve the list; v1 ships with the explicit caveat that his post-deploy review may swap, modify, or add questions. No first-person Patrik reactions have been fabricated — Patrik-attributed content is verbatim from cited public sources only.

Solver. QuintAce's decision-analysis pipeline, running 6-max NLHE cash at 100bb and 200bb effective depths. Stored hands: patrik-q1-btn-vs-bb-k72r-100bb, patrik-q2-btn-vs-bb-k72r-200bb, patrik-q4-turn-sizing-k72r-5c. All three scenarios share the same preflop action (BTN 2.5bb → BB 3-bet 10bb → BTN call) and flop texture (K♦7♣2♥). Q4 extends to turn 5♣.

Q3 derivation. MDF calculations use the standard formula pot / (pot + bet). Under-bluff tax estimates (~20-30% gap between GTO and typical population bluff frequency on rivers) are drawn from community research and solver-provider population studies; we did not run a specific nodelocked-villain solver run for this section. A scenario-specific exploit solve can be added in v2 if Patrik wants it calibrated to a particular villain pool.

Q5 derivation. Exploit arithmetic uses the Q1 solver output as the GTO baseline and contrasts with a hypothesized "range-bet 33%" cohort strategy. The 26.9pp check-frequency delta is the Q1 solver output minus the cohort's 0% check frequency. No separate exploit-nodelock solver run was performed.

Numbers flagged for follow-up. Q3's population-bluff-frequency estimate (20-30% gap) could be tightened with a specific pool's HH data; we'd recommend one of the Triton-style live pools for the intended reader. If Patrik wants this quantified against a specific villain type, we'll run it.

Citations for all Patrik quotes. GipsyTeam Jan 2026; SoMuchPoker Sep 2024; PokerNews Dec 2021. Full URLs inline above.

Questions curated by QuintAce from Patrik Antonius's public record. Solver analysis by QuintAce. Patrik reviews this draft on internals.quintace.ai and provides final cut on the five questions + reactions. v2 merges his cuts + first-person reactions.