The Bicycle Guide to Modern Poker Theory Building an Unbeatable Strategy Based on GTO Principles

1. 1Fix the spot before you judge anything: positions, stack depth (the passages' MTT buckets use 20, 30 and 40bb), pot size, and the bet sizes allowed. A range is only balanced relative to a defined tree, so write the tree first.
2. 2Build the range hand by hand and assign every hand to an equity bucket by hand-vs-range equity: Strong ≥75%, Good 50–74%, Weak 33–49%, Trash <33%. Recompute the buckets for each board type. Hand value is dynamic: top pair is 82% on AQT but 60% on single-suited T98, and two pair on AQT has more equity than a set on T98.
3. 3Compare your bucket distribution with the solver benchmark (averages across all 1,755 flops, GTO MTT ranges, 20–40bb). IP (BN–UTG) is about 6% trash, 22% strong, 29% weak, 43% good. BB is nearly half trash, about 7% strong, 23% weak, 25% good. A large gap from the benchmark means you are playing the wrong range, so fix the range before you tune the bet size.
4. 4Check coverage in each action line. Your betting range needs strong hands plus enough weaker hands to bluff, and your checking range must still hold hands that can call or continue. If one bucket is empty in a line, an observant opponent can attack that line.
5. 5Set the defense side by the toy-game logic. In the Clairvoyance game, a $100 bet into a $100 pot is called 50% of the time by KK, and Player 1's EV is $75 against Player 2's $25. Your continue frequency has to hold up against the sizing you face, so do not fold more than the bet size allows.
6. 6Decide whether to deviate from equilibrium only after you have the baseline. GTO cannot lose or win EV against a GTO opponent, and it gains against leaks without any adjustment. Put the cost of an exploit in numbers first, using the Table 11 trade-off (extra EV gained versus EV lost if the opponent counter-exploits).
7. 7If you do deviate, prefer the small shift (MinES). In the toy game, a Villain who calls 55% instead of 50% earns Hero no extra EV when Hero stays at GTO. Move only as far as your read justifies, and write down the read that justifies the move.
8. 8Review by spot, not by result. After each session, log which spots had a bucket gap, a frequency you cannot justify, or a deviation made without a read. Re-run those spots in a solver or EV tree tool.

Take a BB defending against an IP opener on an AQT flop, then on T98. Bucket your hands first. Top pair is Strong at 82% equity on AQT but only Good at 60% on single-suited T98, so the same hand belongs in a different bucket on each board. Two pair on AQT out-equities a set on T98, so you cannot rank hands by category name. Next, compare the whole BB range with the benchmark: about 7% strong, 25% good, 23% weak, and nearly half trash (<33%). If your BB range on these boards shows far more than 7% strong hands, you are probably defending too tightly, since the benchmark BB range is mostly weak and trash. Then test the exploit temptation. Against a Villain who over-calls, Table 11 shows MES (maximally exploitative strategy) gaining 6.3bb/100 over GTO but giving back −2.5bb/100 to a counter-exploit. Against a Villain who under-calls by 50%, MES gains 24.6bb/100 but is exposed to −62.4bb/100, and the shove frequency jumps from 28% to 100%. Stay near equilibrium unless the read is strong and the shift is small.

• If your range has far more or far fewer strong (≥75%) hands than the benchmark for your position (IP about 22%, BB about 7%), then rebuild the range for that board before touching bet sizes.
• If a hand's category changes between boards (top pair, set, two pair), then re-bucket it by hand-vs-range equity on that specific board rather than carrying over a label.
• If the opponent's leak is small (the toy game's 55% call instead of 50%), then stay at or near GTO or make a minimal adjustment. Full MES is not worth the exposure.
• If a full exploit would sharply change your strategy (shoving 100% instead of 28%), then expect an observant opponent to spot it and counter-exploit. Require a high-certainty read first.
• If you are unsure of your read, then default to the equilibrium strategy. A wrong read drops you to a weaker strategy that loses value.

• Judging hands by category (top pair is strong, a set is the nuts) rather than equity vs range. The tell: you play top pair the same on AQT and T98 despite 82% vs 60% equity.
• Dropping the abstraction mismatch. Tools built for a different tree (e.g. Nash calculators that forbid calls or assume checks down) give wrong ranges for deeper stacks. The tell: your opening ranges are wider than solver output and you rely on a shove/fold chart at stacks that allow non-all-in sizes.
• Chasing maximum exploitation on weak evidence. The tell: large, abrupt frequency swings after a small sample, which is the 100% vs 28% shoving pattern the passages warn can be counter-exploited.
• Assuming GTO should earn from a leak by itself. The tell: you expect extra EV against a loose caller while playing the unchanged equilibrium, and then feel the strategy is failing. The toy game shows GTO earns no extra at 55% calling.
• Ignoring the shape of the defender's range. The tell: you defend BB as if it had the same strong and good hand share as IP, when the benchmark has the BB nearly half trash.

• ▸Spot: positions, stack depth (20/30/40bb for MTT benchmark), pot size, bet sizes allowed. Define the tree before analysis.
• ▸Board and type: write the flop and mark it dynamic or static; hand values are board-dependent.
• ▸My range: % in each bucket (Strong ≥75 / Good 50–74 / Weak 33–49 / Trash <33).
• ▸Benchmark range: the same four percentages (IP about 22/43/29/6; BB about 7/25/23/near half trash).
• ▸Biggest gap: which bucket deviates most from the benchmark and in which direction.
• ▸Coverage check by line: does each action line (bet, check, call, raise) contain strong and weaker hands? List any empty bucket.
• ▸Facing-bet defense: bet size faced, continue frequency chosen, and justification (compare with the toy game's 50% vs a pot-sized bet).
• ▸Opponent leak observed: sample size, the specific frequency, and the baseline frequency it deviates from.
• ▸Deviation plan: none / minimal (MinES) / full (MES). Record the Table 11-style trade-off: expected gain vs the cost if counter-exploited.

1. 1Record position first: who is IP, who is OOP, and who is the pre-flop aggressor. The source defines a c-bet as the last aggressor's bet and a donk bet as an OOP lead into that aggressor, so you cannot label a bet without this.
2. 2Write blinds and antes, then stack depth in big blinds. Convert chips to bb (1,000,000 chips with a 100,000 BB is 10bb). Stack depth is always expressed in bb.
3. 3Compute effective stack: the smallest stack among the active players. In a heads-up pot it is the most either player can lose. In a multiway pot, note that blockers matter more than raw hand strength and default to passive play.
4. 4Compute SPR from the effective stack and the pot at the start of the street. SPR is a variable in the betting decision alongside information advantage and range composition. Recompute it on every street.
5. 5Log the board precisely, with ranks and suits. Note which draws are possible, which hands are the nuts or effective nuts, and whether the board is dynamic or static. Check the suit pattern: a flush draw is a bluffing hand you may or may not block.
6. 6Write the action history as lines (x/c, x/r, b/b, b/b/b) with bet sizes in bb and as a fraction of pot. Update each opponent's range after each action instead of assigning a single hand.
7. 7Classify your own hand against that range: nuts, effective nuts, bluff catcher, semi-bluff, air. Note which villain holdings you block, then compare the price you are offered (call risk versus pot) with your equity.
8. 8Write one sentence of plan for the remaining streets, even a simple one. Write it from the objective variables rather than from how strong the hand feels.

The source's river spot: pot with 18.375bb effective, board 8♥ 7♠ 5♠ J♥ 2♣, two players. Villain moves all-in for 1.7 × pot (31bb). Hero holds A♦A♣. Position, stack and bet size are fixed, and the action history shows Villain betting a river where every draw missed. Hero is holding a pure bluff catcher: the hand beats only bluffs. Hero's hand holds no hearts or spades, so Hero does not block the obvious bluffing hands, the flush draws. That makes a bluff a little more likely, so the call is good. Villain shows T♥3♥, a busted draw, and Hero wins. The conclusion comes from combining board, bet size, history and blockers. AA's strength in a vacuum is not what decides it.

• If you are unsure about the flop strategy and stacks are short or middle, default to a 1/3-pot c-bet; if the opponent folds or raises a lot, bet small at high frequency, and if they call a lot, size up.
• If you are unsure whether to bet or check, ask what happens when you are raised; if the hand hates a check-raise, check.
• If the pot is multiway, play passively, bet draws rarely unless they are to the nuts, and value blockers over raw hand strength.
• If the BB has no donk-betting range on good donking flops, c-bet less than equilibrium suggests, because you will face more check-raises.
• If you hold a bluff catcher on the river and your hand blocks none of the bluffing combos (no hearts or spades against missed flush draws), lean toward calling more.

• Judging a hand in a vacuum: the tell is naming only your hand's strength (AA 'is strong') without stating board, bet size and opponent line.
• Using chips instead of bb, or nominal stack instead of effective stack: the tell is that you can't state the maximum you can lose or the SPR.
• Treating a board class as a rule, as in 'the BB never donk bets': the tell is surprise when the solver donks 654r at 67% on average yet A76r at 0.3%.
• Pinning the villain to one hand instead of a range that narrows with each action: the tell is that your read doesn't change from flop to river.
• Ignoring your own blockers on the river: the tell is that you never ask which bluffs you remove, such as flush draws when you hold no hearts or spades.

• ▸Position and roles: Hero seat, Villain seat, who is IP/OOP, who is the pre-flop aggressor.
• ▸Blinds/antes: Write the structure and convert all amounts to bb.
• ▸Effective stack (bb): The smallest stack among active players.
• ▸Pot and SPR: Pot size in bb, effective stack divided by pot, recomputed each street.
• ▸Board: Ranks and suits; note draws that exist and any nuts or effective nuts.
• ▸Action line: Each street's actions (x/c, x/r, b/b) with sizes in bb and as a pot fraction.
• ▸Villain range (current): Which hands remain after each action, in plain categories.
• ▸Hero hand class and blockers: Nuts, bluff catcher, semi-bluff or air, and which villain combos you block.
• ▸Plan and EV check: One-sentence plan for remaining streets; the call/bet risk against %W × $W − %L × $R.

1. 1Define the decision point first. EV starts from the moment you choose your next action, so money already in the pot is not yours. List every option available (fold, call, raise to each size, all-in) before you evaluate any of them.
2. 2Anchor on folding. The EV of folding is always 0, so every other action is judged by whether it beats 0 and beats the other candidates. Always compare an action against all other possibilities, not just against folding.
3. 3Assign Villain a range, not a hand. Use a standard range for the player type: the passages use a 20% rejamming range for a SB regular and a standard 16% open for a UTG regular. This is your exploitative input, and it should be a stated assumption you can revise.
4. 4Get your equity vs that range from an equity calculator. This gives %W. Then %L = 100% − %W.
5. 5Count the money honestly. $W is what you can win, including dead money (blinds plus your own raise). $R is the extra amount you must put in. Compute EV = (%W × $W) − (%L × $R).
6. 6Discount for equity realization if the hand will see more streets. Raw equity only turns into EV fully when no one folds before showdown. Hands that capture less of the pot than their equity share under-realize. Judge hands like 9♣5♦ by how playable they are, not by raw equity alone.
7. 7For shoves and bluffs, compute the break-even fold frequency. Compare the fold equity you need against the fold frequency you estimate from Villain's range. Take the action only if your estimate clears the threshold.
8. 8Check your own range's exposure. Ask what information the action gives away. Ranges with hands that are never raised, or mixed limp/raise ranges, make you easier to read. Prefer the simpler line when the mix adds no value.
9. 9Review by win rate, not by single results. Multiply EV in bb by 100 to get bb/100. Log hands where you chose a lower-EV option, and study those spots with an equity calculator or EV tree.

Passage example: $100 3-max Jackpot Sit & Go, stacks BN 35bb, SB 15bb, BB 25bb, no ante. Hero on the BN holds A♠5♠ and raises 2x. The SB, a regular, shoves 15bb and the BB folds. Assign the SB a standard 20% rejamming range. An equity calculator gives A♠5♠ 43.31% against it, so %W = 43.31% and %L = 56.69%. The win amount is the SB stack plus dead money: $W = 15 + 1 + 2 = 18bb. The amount to call is $R = 13bb. EV(call) = 0.4331 × 18 − 0.5669 × 13 ≈ +0.4bb, against 0bb for folding, so calling is the higher-EV decision. Multiplied by 100, that is about 40bb/100 over a player who always folds this spot. The decision depends on the range assumption: if the SB's real range is tighter than 20%, your equity drops and the margin shrinks.

• If EV(call) > 0, which is also the same as your equity beating the pot-odds break-even, then call, because the highest-EV option beats fold at 0. Verify first that the Villain range you assigned is realistic.
• If you are shoving as a bluff, then compare the fold equity you need to the fold frequency you estimate. In the T♦5♣ example, Hero needs about 51% folds and estimates 52%, so the shove is only barely profitable and fails if the estimate slips.
• If the hand will play post-flop and tends to under-realize equity (like 9♣5♦ out of position), then do not treat raw equity as EV. Compare call and fold with realization in mind, and rule out a 3-bet bluff when the opener's range defends well against 3-bets.
• If an action pattern reveals your range (for example, never raising T9s or medium pairs), then simplify it, such as with a raise-only entry strategy from early position, so opponents cannot narrow your hand.
• If the spot is a tournament decision, then remember chips are not directly proportional to dollars. Use cash-style EV in bb as a guide and treat tournament value separately.

• Counting your own raise or the money already in as at risk. The tell: your $R is larger than the amount you actually have to put in, so everything looks -EV.
• Using a made-up opponent range without checking it. The tell: you describe Villain as 'bluffing a lot' but cannot name a range or percentage.
• Treating equity as EV. The tell: you call with hands that have decent equity but are out of position and fold often or play poorly post-flop, which means they under-realize.
• Choosing between only the action you like and folding. The tell: you never list other raise sizes or the all-in as alternatives, so the choice is never compared against the full set of options.
• Judging decisions by outcomes or by short samples. The tell: you change your line after a lost flip or a downswing, even though results deviate from expectation through variance and EV only shows up over the long run.

• ▸Situation: stacks, blinds, position, and action so far (one line).
• ▸Options considered: fold, call, each raise size, and all-in. Fold = 0 EV by definition.
• ▸Villain range assumption: the percentage or chart you assign and why (e.g. 20% rejam range).
• ▸My equity vs that range (%W): from an equity calculator. %L = 100% − %W.
• ▸$W (what I can win): villain bet or stack plus all dead money.
• ▸$R (what I must risk): extra chips I must put in.
• ▸EV = (%W × $W) − (%L × $R): result in bb and as bb/100 (EV × 100).
• ▸For shoves/bluffs, fold equity needed vs fold frequency estimated: needed ___%, estimated ___%, margin ___.
• ▸Equity realization note and range exposure: does the hand under- or over-realize, and does the action give away information about my range?

1. 1Define unexploitability as a property of the whole strategy. A strategy is a full specification of the action you take at every decision point, so audit the tree, not the single hand. For each node ask: if a clairvoyant opponent knew my frequencies here, what would they do?
2. 2Fix the spot before you balance anything. Write down the positions, the stack depth, the preflop actions and both ranges. Example: CO opens 2.25x at 40bb with 31.2%, and BB defends with a 56.8% call range.
3. 3Run range-vs-range equity on the flop before choosing a line. Preflop equity does not carry over. In the source, the CO's 58.5% preflop equity drops to 49.67% on 8♥7♠5♠, while the BB rises from 41.5% to 50.33%. Your overall range equity sets how aggressive your whole range can afford to be.
4. 4For each candidate line, name the opponent's best counter. High-frequency c-bet: the BB can check-raise very effectively. Check back: the BB can bet aggressively on many turns and rivers. If a line has an obvious counter, you cannot play it at 100% frequency.
5. 5Use mixed strategies where a pure line is punishable. Because the same action is never taken 100% of the time at the node, the opponent cannot single out one counter. On bad boards, lean toward passive play and keep the check-back range from collapsing to weak hands only.
6. 6Fill the check-back range with hands that hold up across many turns. In the source, A♦A♣ checks behind because it can call most turn bets and bluff-catch effectively. That lets the check-back range defend itself against the BB's aggression.
7. 7Check your bluff-catchers against blockers. A♦A♣ is a better call when it holds no hearts or spades, because it does not block the BB's flush-draw bluffs.
8. 8Verify rather than assert. A claim that a strategy is GTO needs a scientific, mathematical check. Use a solver and read its Nash distance (ε-Nash) as the measure of exploitability. The source lists that AI tools cannot guarantee unexploitability, so treat AI output as advice, not proof.

CO opens 2.25x at 40bb with a 31.2% range, and BB defends 56.8% by calling, 9.7% by 3-betting and folding 33.5%. The flop is 8♥7♠5♠. Range equity moves from CO 58.5% to 49.67%, and from BB 41.5% to 50.33%, so the BB is now a slight favorite with straights, sets, two pairs and combo draws. A high-frequency c-bet exposes the CO to a strong check-raise. A pure check-back range invites aggressive turn and river betting. Hero holds A♦A♣, which has 70.83% equity, but still checks behind with a high frequency of hands that improve on many turns. Hero plays it as a bluff-catcher. The turn is J♥ and Villain bets the full pot of 6.125bb. Hero calls. The river is 2♣ and Villain shoves 31bb into 18.375bb (1.7x pot). Hero calls because the draws missed and Hero holds no hearts or spades. Villain shows T♥3♥, a busted draw. The balanced check-back range defended itself, and AA was one of its best calling hands.

• If your range equity falls below the opponent's on the flop (e.g. 49.67% vs 50.33%), then cut your c-bet frequency and play more cautiously, because high-frequency c-betting invites check-raises.
• If you check back, then keep hands in the range that can call turn bets and bluff-catch, so the opponent cannot bet turns and rivers aggressively for free.
• If the opponent has an obvious counter-strategy to a line, then mix that line with another action instead of taking it 100% of the time.
• If a bluff-catcher blocks the opponent's bluffing hands (e.g. flush draws), then expect fewer bluffs and call less eagerly. If it blocks none, then it is a more attractive call.
• If someone claims a strategy is GTO or near-GTO, then require a Nash-distance or solver check. Do not accept AI-tool output as proof of unexploitability.

• Balancing a single hand instead of the whole tree. The tell: you justify a play only by that hand's equity (70.83% for AA) and ignore the range's frequencies.
• Using preflop equity to set postflop aggression. The tell: you c-bet at the same frequency on every board, even when range equity swings by 8-9 points.
• Checking back only weak hands. The tell: opponents bet turns and rivers aggressively against your check-back range and you have few hands that can call.
• Treating AI-tool output as unexploitable. The tell: your 'balanced' strategy comes from a tool that cannot explain itself, changes over time and offers no verification.
• Memorizing results instead of principles. The source notes that memorizing every equity matchup or running exact calculations at the table is impossible. The tell: you freeze on boards you have not studied instead of reasoning from range equity and which lines have counters.

• ▸Spot: positions, stack depth, preflop actions (e.g. CO 2.25x vs BB call, 40bb).
• ▸Hero range % / Villain range %: the starting range sizes for the spot.
• ▸Flop texture: the board and whether it is static or dynamic.
• ▸Range equity shift: Hero and Villain equity preflop vs on this flop.
• ▸Candidate line A and the opponent's best counter: e.g. high-frequency c-bet and check-raise.
• ▸Candidate line B and the opponent's best counter: e.g. check back and aggressive turn/river betting.
• ▸Mix plan: the frequency you assign to each line and which hands go in each.
• ▸Check-back protection: the hands that can call turn bets and bluff-catch, and which blockers they hold.
• ▸Verification: the solver or Nash-distance figure that confirms the plan, or a note that it is still unverified.

1. 1Define the unit first: win rate in bb/100 for cash games. The pot you compete for each orbit is 150bb/100 (1bb + 0.5bb of blinds per hand), and an equal split at a 6-max table would give each seat 25bb/100. Treat 25bb/100 as the reference point and not as a target.
2. 2Log every hand with tracking software. Online, it reports your win rate and the exact rake paid. Live, estimate: hands per year = hands per session × sessions per week × weeks, and win rate = winnings ÷ big blind ÷ hands × 100. Record the sample size beside every win-rate figure.
3. 3Grade the number by sample size. At 100,000 hands you only get an idea of your real win rate. A sample of at least 1,000,000 hands is needed before it counts as statistically significant. Below that, treat the figure as provisional and do not use it to move stakes.
4. 4Split your win rate by position. Your overall rate is the average of your per-position rates. The blinds start at -100bb/100 (BB) and -50bb/100 (SB), so they stay negative. The button should be highest, and rates should fall as you move away from it. A position that breaks this pattern points to a leak.
5. 5Net out the rake. You must recover the -150bb/100 lost to the blinds each round plus the rake, which is usually 5-10bb/100. Read each site's rake structure, rakeback deals and VIP programs before choosing where to play. At some online micro stakes the rake is high enough that even strong players cannot win.
6. 6Audit decisions, not results. Against a GTO opponent, any action that is not the highest EV for a given hand lowers your win rate in that spot. Compare each action against all alternatives, because a +EV play is not automatically the best one. Review spots in a solver such as PioSOLVER and note which deviations cost you.
7. 7Count the most common leaks: playing too few hands profitably, or playing hands with a low average profitability. Win rate rises with the number of hands you can play profitably and with the average profitability of each.
8. 8In tournaments, use win rate as a diagnostic only. Chips are not worth a proportional amount of dollars, so cash win rate does not translate directly to profit. A rate lower than expected still flags a skill leak.

A live player is dealt about 300 hands in a 10-hour session, plays five days a week, four weeks a month. That is about 6,000 hands a month and 72,000 a year. She wins $100,000 at $5/$10 blinds. Dividing by the $10 big blind gives 10,000bb, and 72,000 hands is 720 blocks of 100, so her win rate is about 13.9bb/100. The figure is real but unproven, because 72,000 hands is below even the 100,000-hand mark where a rate starts to look meaningful, and far below the 1,000,000 needed for significance. The source also notes that live games are softer than online games, so this rate is not comparable with an online one. An online player at 3,000 hands a day plays 720,000 hands a year, which is as many hands as a live player sees in 10 years. That online player reaches a significant sample in roughly 1.4 years, while the live player would need about 14 years. Before judging her skill, she should confirm that the rate holds as the sample grows, and then check it against the position pattern: button highest, blinds negative.

• If your sample is under 100,000 hands, then do not draw conclusions about your win rate. If it is under 1,000,000, treat it as an estimate and avoid big stake or bankroll moves based on it.
• If a play is +EV, then still compare it with every other action before choosing. Shoving AA is +EV but costs 136bb/100 more than raising to 2.5bb in the source's solver example.
• If a site's rake is high relative to your win rate (5-10bb/100 is typical), then check rakeback and VIP deals or change sites before changing your strategy.
• If a position's win rate departs from the expected pattern (button best, blinds negative), then review that position's ranges and decisions for leaks.
• If you play tournaments and your results lag what your cash win rate suggests, then use win rate as a leak finder and not as a profit forecast, since chip value is not proportional to dollar value.

• Judging skill on a small sample. The tell: you call yourself a winner or loser after tens of thousands of hands. The author blamed bad luck, and later credited good luck, for results that were mostly variance.
• Moving up in stakes without enough bankroll after a hot run. The tell: you jump to $1/$2 or $2/$5 after a single tournament win or a good month. The author went broke twice this way.
• Ignoring rake and the blinds' drag. The tell: you track gross winnings and not rake paid, and you cannot say why a break-even player is actually losing. The blinds cost 150bb/100 per round and rake adds 5-10bb/100.
• Treating any +EV play as correct. The tell: you defend an action because it makes money, without checking the highest-EV alternative. Shoving AA is +EV yet worse than raising by 136bb/100.
• Applying cash win-rate logic directly to tournaments, or abandoning study after early success. The tell: you equate chips with dollars, or you stop studying because you have already read the best books. The author did the latter and then blasted his bankroll.

• ▸Game / stakes / site: record where the hands were played so results are comparable.
• ▸Total hands in sample: use tracker data online, or sessions × hands/session × weeks live.
• ▸Net winnings (currency): total results over the sample, after rake.
• ▸Big blind size: used to convert winnings into bb.
• ▸Win rate (bb/100): winnings ÷ big blind ÷ hands × 100.
• ▸Sample status: below 100,000 / 100,000-999,999 / 1,000,000+ hands, and treat it as provisional until the last.
• ▸Rake paid (bb/100): from tracking software, compared against the typical 5-10bb/100.
• ▸Win rate by position (BN, CO, HJ, LJ, SB, BB): confirm button highest and blinds negative, and flag deviations.
• ▸Top three leaks found: spots where a lower-EV action was chosen than a solver's best action, with the solver cost in bb/100.