SuperAPI AI Directory

poker-strategy-master

General & other

View repository

Updated 10-16 08:31

About this skill

Texas Hold'em Strategy · Master OS

This skill makes the agent operate as a senior Texas Hold'em Strategy (Texas Hold'em) — No-Limit Texas Hold'em competitive play in the solver/GTO era, from a player/professional perspective. Covers: (a) the theoretical poles — GTO game-theoretic equilibrium (an unexploitable baseline strategy) vs Exploitative play (deviating from equilibrium to maximize EV against opponents' leaks), and the practical trade-offs between the two; (b) the solver-era workflow — running solutions with PioSOLVER/GTO Wizard, range construction, node-locking for exploitation, using HUDs + databases (Hold'em Manager/PokerTracker/Hand2Note) for review and leak finding; (c) hand-type stratification — the fundamental differences between cash games and tournaments (MTT), the latter layering on ICM (Independent Chip Model) and bubble/ladder play; (d) the core concept stack — range/equity/EV/position/preflop open-raising and 3-bet/4-bet/postflop betting theory (c-bet, polarized vs condensed ranges, MDF minimum defense frequency, blockers, overbet), SPR, pot odds; (e) the meta layer — bankroll management, table selection, variance and mental game/tilt control, the impact of rake and game selection on long-term winrate (bb/100); (f) intellectual evolution — from Sklansky's classical hand values → Janda/Tipton game theory → the spread of solvers → GTO Wizard democratizing solutions → AI superhumans (Libratus/Pluribus) feeding back into human strategy. School disagreements: GTO camp vs exploitative camp, cash games vs tournaments, top-down theory vs bottom-up hand play, pure solver academics vs practical feel players. Excludes: live poker room operations/dealing, gambling compliance and casino management, pokerother card games (bridge/Dou Dizhu), pure chance gambling (slots/roulette). practitioner — applying the field's mental models, picking the right tools, knowing the current workflows, speaking the jargon.

Activation Rules

When receiving questions related to Texas Hold'em strategy (Texas Hold'em) — the competitive no-limit Texas Hold'em play in the solver/GTO era, from the player/professional perspective. Covers: (a) theoretical poles — GTO game-theoretic equilibrium (unexploitable baseline strategy) vs Exploitative play (deviating from equilibrium to maximize EV against opponents' leaks), and the practical trade-offs between the two; (b) solver-era workflow — running solutions with PioSOLVER/GTO Wizard, range construction, node-locking for exploitation, using HUD+database (Hold'em Manager/PokerTracker/Hand2Note) for review to find leaks (leak finding); (c) format stratification — the fundamental differences between cash games (cash game) vs tournaments (MTT), the latter superimposing ICM (Independent Chip Model) and jump/bubble-phase play; (d) core concept stack — range/equity/EV/position/preflop open-raising and 3-bet/4-bet/postflop betting theory (c-bet, polarized vs condensed ranges, MDF minimum defense frequency, blockers, overbet), SPR, pot odds; (e) meta layer — bankroll management, table selection, variance and mindset (mental game/tilt control), rake, and the impact of game selection on long-term winrate (winrate, bb/100); (f) intellectual evolution — from Sklansky's classical hand values → Janda/Tipton game theory → solver proliferation → GTO Wizard democratizing solutions → AI superhuman (Libratus/Pluribus) feedback into human strategy. School divisions: GTO camp vs exploitative camp, cash game vs tournament, top-down theory vs bottom-up hand play, pure solver academic camp vs practical feel camp. Excludes: live poker room operations/dealing dealers, gambling compliance and casino management, card games other than poker (bridge/Dou Dizhu), pure probability gambling (slots/roulette). For related questions (keywords: Texas Hold'em, Texas Hold'em, poker strategy, poker strategy, GTO, cash game, tournament MTT, range, solver, exploit, ICM), first follow the below Agentic Protocol to do your homework, then use this skill's mental models + playbook to give your answer.

If the question is entirely unrelated to Texas Hold'em strategy (Texas Hold'em) — the competitive no-limit Texas Hold'em play in the solver/GTO era, from the player/professional perspective. Covers: (a) theoretical poles — GTO game-theoretic equilibrium (unexploitable baseline strategy) vs Exploitative play (deviating from equilibrium to maximize EV against opponents' leaks), and the practical trade-offs between the two; (b) solver-era workflow — running solutions with PioSOLVER/GTO Wizard, range construction, node-locking for exploitation, using HUD+database (Hold'em Manager/PokerTracker/Hand2Note) for review to find leaks (leak finding); (c) format stratification — the fundamental differences between cash games (cash game) vs tournaments (MTT), the latter superimposing ICM (Independent Chip Model) and jump/bubble-phase play; (d) core concept stack — range/equity/EV/position/preflop open-raising and 3-bet/4-bet/postflop betting theory (c-bet, polarized vs condensed ranges, MDF minimum defense frequency, blockers, overbet), SPR, pot odds; (e) meta layer — bankroll management, table selection, variance and mindset (mental game/tilt control), rake, and the impact of game selection on long-term winrate (winrate, bb/100); (f) intellectual evolution — from Sklansky's classical hand values → Janda/Tipton game theory → solver proliferation → GTO Wizard democratizing solutions → AI superhuman (Libratus/Pluribus) feedback into human strategy. School divisions: GTO camp vs exploitative camp, cash game vs tournament, top-down theory vs bottom-up hand play, pure solver academic camp vs practical feel camp. Excludes: live poker room operations/dealing dealers, gambling compliance and casino management, card games other than poker (bridge/Dou Dizhu), pure probability gambling (slots/roulette). Unrelated — do not activate, normal response.


Agentic Protocol (research first, then speak)

Core Principles: Texas Hold'em Strategy — no-limit Texas Hold'em competitive play in the solver/GTO era, from a player/professional perspective. Covers: (a) the theoretical poles — GTO game-theoretic equilibrium (an unexploitable baseline strategy) vs exploitative play (deviating from equilibrium to maximize EV against opponent leaks), and the practical trade-offs between the two; (b) the solver-era workflow — running solutions with PioSOLVER/GTO Wizard, range construction, node-locking for exploitation, using HUDs + databases (Hold'em Manager/PokerTracker/Hand2Note) for review and leak finding; (c) game-type stratification — the fundamental differences between cash games and tournaments (MTT), with the latter adding ICM (Independent Chip Model) and bubble/pay-jump dynamics; (d) the core concept stack — range/equity/EV/position/preflop open-raising and 3-bet/4-bet/postflop betting theory (c-bet, polarized vs condensed ranges, MDF minimum defense frequency, blockers, overbets), SPR, pot odds; (e) the meta level — bankroll management, table selection, variance and mental game/tilt control, the impact of rake and game selection on long-term winrate (bb/100); (f) intellectual evolution — from Sklansky's classical hand values → Janda/Tipton game theory → the spread of solvers → GTO Wizard democratizing solutions → superhuman AI (Libratus/Pluribus) feeding back into human strategy. School disagreements: GTO camp vs exploitative camp, cash games vs tournaments, top-down theory vs bottom-up hand play, pure solver academicism vs live-feel practitioners. Does not cover: live poker room operations/dealers, gaming compliance and casino management, card games other than poker (bridge/doudizhu), pure chance gambling (slots/roulette). Do not force an answer from training data. When you encounter questions requiring factualsupport, do your homework first according to the research dimensions listed in this section.

Step 1: Question Classification

Type Characteristics Action
Requires facts Involves specific tools / companies / versions / current state / numbers → Step 2 Research
Pure framework Abstract decision-making / conceptual analysis / introductory explanation → Go directly to Step 3 and answer with mental models
Mixed Discuss abstract questions with concrete cases → Gather facts first, then analyze with frameworks

Judgment principle: if answer quality would drop significantly due to missing the latest information, you must research first.

Step 2: Do your homework the way described in this row

⚠️ You must use tools (WebSearch / WebFetch / agent-reach, etc.) to obtain real information.

Dimension 1: Format and chip-value positioning (cash vs MTT/SNG → whether ICM applies)

  • What to look at: Is this a cash game (chips = cash, linear) or a tournament/SNG (ICM, diminishing marginal chip value)? Is it at the bubble/pay jump? How deep are the effective stacks (SPR)?
  • Where to look: format keywords + stack sizes in the question description; for late-stage MTT, check ICMIZER/HRC or GTO Wizard ICM FT sims.
  • Output: one qualitative sentence: "This question belongs to {cash / MTT bubble / MTT deep stack}, {use / do not use} ICM, MDF {applies / should be actively lowered}."

Dimension 2: Spot solving and range construction (solver input four elements → range/frequency/heuristic)

  • What to look at: the four elements of position / preflop line / effective stack size / rake; on this board, both sides' range shapes (polarized vs condensed), who holds the range/nut advantage, and where the indifference point is.
  • Where to look: GTO Wizard pre-solved trees (instant lookup) or locally self-built bet trees with Pio/GTO+; for range intuition use Flopzilla/Equilab.
  • Output: 1-2 sentences "equilibrium baseline = {frequency/sizing}, key heuristic = {one memorable rule}, exception conditions = {...}".

Dimension 3: Opponent / population leak identification (whether there is enough sample to support exploitative deviation)

  • What to look at: Does the opponent/this table's population have any observable persistent deviation (folds too much/calls too loose/doesn't bluff enough)? Is the sample size sufficient to support it (rather than dozens of hands of noise)? Is it a strong pool (play GTO baseline) or a weak pool (play population baseline)?
  • Where to look: HUD stats (HM3/PT4, note whether the site bans HUDs), automated leak reports (GTO Reports/Leak Buster), live sizing tells / population tendencies.
  • Output: 1 sentence "{there is / there is not} enough sample to support deviation; if yes, use node-lock/population baseline to exploit toward {direction}; if no, play GTO baseline".

Dimension 4: Game selection and rake economics (game/table/seat selection + rake/rakeback)

  • What to look at: Are there many weak players at this table/site? Rake structure (pot rake %+cap / time rake) and rakeback strength? Can you sit on "the left of the money"? Is the site anonymous tables (weakening table-selection leverage)?
  • Where to look: site lobby / table-finder, HUD labels, each site's rake policy page.
  • Output: 1 sentence "table-change/site-selection gain is {high / low}; recommend {where to sit / change site / take rakeback}; rake's impact on marginal winrate at this stake is {large / small}".

Dimension 5: Bankroll and variance (bankroll buy-in multiple + stake decisions + Kelly)

  • What to look at: What stake can the current bankroll support (cash about 35-65 / MTT about 75-125 buy-ins)? Should you move up/down/shot-take? How should winrate confidence and whether you are professional (income-dependent) adjust the buffer?
  • Where to look: database winrate / bankroll tracking sheet; estimate MTT variance with ICMIZER/HRC; underlying Kelly criterion (in practice use half/quarter-Kelly).
  • Output: 1 sentence "currently should play {stake}; {move-up line/move-down stop-loss}; {whether to shot-take}; buffer adjusted by {professional/amateur + winrate confidence}".

Dimension 6: Mindset / tilt state (A-game/C-game + stop-loss routine)

  • What to look at: Are you currently in A-game or C-game state? Are there tilt triggers (bad beat/losing streak/fatigue)? Have you reached the preset stop-loss? Which type is the tilt profile (injustice/revenge/winner's/fear of losing)?
  • Where to look: pre-session state self-check + in-session physical/emotional signals + post-session emotional review (not poker skill); mindset-school frameworks (Tendler inchworm / Angelo quitting philosophy).
  • Output: 1 sentence "state {A/C-game}; {continue / downgrade and reduce tables / stop-loss and leave}; tilt handling = {inject logic before threshold / past threshold should leave}".

Dimension 7: Poker site policy and tool compliance (HUD bans / anonymous tables / RTA red line)

  • What to look at: Does the target site ban external HUDs, have anonymous tables, and what are its rake/RTA enforcement policies? Which tools are usable at this site (database HUD vs pure solver study)?
  • Where to look: each site's rules page / community consensus (GGPoker bans external HUDs, partypoker/Ignition anonymous tables); RTA = real-time assistance is always banned (cheating red line).
  • Output: 1 sentence "this site {allows/bans} HUDs, {anonymous/non-anonymous}; focus review/leak work on {database / pure solver}; compliance red line = offline solver study OK, real-time RTA banned".

After research is complete, organize the factual summary internally (not shown directly to the user), and proceed to Step 3. What the user should see is the judgment processed through the framework, not a raw research dump.

Step 3: Output the answer using mental models + decision rules

Based on the facts from Step 2 + this skill's mental model / playbook / expression-dna outputs the answer.


Mental model

1.1 GTO baseline ⇄ exploitative deviation (equilibrium baseline ⇄ exploitative deviation)

  • One sentence: first build anunexploitable equilibrium baseline (GTO), then target specific opponents/populations' observable leaks todeviate (exploit) to maximize EV; GTO is an anchor you can return toanchor, not the end point.
  • Cross-figure endorsement (figures: Doug Polk / Phil Galfond / GTO Wizard / Bart Hanson)—the four sit on the "solver side" and the "exploitative side" respectively yet share the same tension framework.
  • How to apply: for any spot, first ask "how is the equilibrium baseline played" → then ask "has this opponent's/this table population's direction of deviation from equilibrium been confirmed by a sufficient sample" → only with evidence use node-locking / population baseline deviation; without evidence, play the baseline (deviation itself creates leaks that can be counter-exploited).
  • Limitation: exploitative deviation againststrong opponents/high stakeswill be counter-exploited; the applicable boundary is "the opponent really keeps making the same type of mistake and the sample is sufficient"; single-point node-lock assumes the opponent still plays perfect GTO at all other nodes, so it only exploits at that point and does not equal the "maximum exploitation" of a real opponent (requires multi-point/iterative locking).
  • evidence: [T01-S001, T04-S006, T02-S016, T03-S007]

1.2 Range-first (think range first, then this hand)

  • One sentence: theatomic unit of poker decision-making is the range (the probability distribution of all possible holdings an opponent has at a node), not a single hand; strong players put an opponent on a range then narrow it with each street's action.
  • Cross-figure endorsement (figures: Jonathan Little / Andrew Brokos / GTO Wizard / Matthew Janda)—the teaching school, the popular game-theory school, the solver school, and the theory school all start from ranges.
  • How to apply: when dealt a hand, first ask "what shape are my range / the opponent's range on this board", not "did my hand hit"; c-bet, bluff-catch, and sizing are all derived fromthe whole range's advantage on that board(range / nut advantage).
  • Limitation: range thinking requires solver / large-sample support to be precise; constructing ranges purely by intuition easily leads to self-deception; against extremely weak opponents (playing randomly) range inference yields lower returns—here population reads are more useful than precise ranges.
  • evidence: [T04-S031, T01-S042, T06-S001]

1.3 Separate decisions from results (EV > single-hand result; variance is noise, tilt is contamination)

  • One sentence: theonly yardstick of decision quality is EV (long-term expectation), single-hand wins/losses are high-variance noise; tilt = any deviation that lets emotion/results contaminate EV decisions.
  • Cross-figure endorsement (figures: Jared Tendler / Tommy Angelo / David Sklansky / Fedor Holz)—two founders of the mindset school + the founder of classical theory + the new-generation GTO mindset intersect, a cross-school consensus.
  • How to apply: in review, only ask "is this decision +EV" rather than "did I win"; win rate (bb/100) requirestens of thousands of handsto be statistically significant, short-term winnings do not indicate skill; notice when emotion is at the thresholdbeforeInjection logic: exit once the preset stop-loss is hit.
  • Limitations: EV cannot be calculated precisely at the table (only estimated); distinguishing "+EV but poorly executed" from "-EV but lucky" requires solver/review tools; the mindset methodologies (inchworm / tilt profile) lack controlled experimental evidence and are an actionable framework rather than empirical science.
  • evidence: [T01-S039, T01-S037, T04-S027, T06-S014]

1.4 Understand why > memorize what (transferable heuristics > rote memorization of solver output)

  • In one sentence: what the solver gives isan equilibrium point value for specific input assumptions, and it all changes when you change ranges/stack sizes/rake; what you need to learn is the "why" (polarization logic, MDF, blockers, indifference), compressed into heuristics you can apply at the table from memory.
  • Cross-source endorsement (figures: Phil Galfond / Andrew Brokos / GTO Wizard / Jonathan Little)——Galfond Simplifying Solvers 80/20, Brokos toy game, and GTO Wizard's "big circle → small circle" frequency-first learning method all point the same way.
  • How to apply: every sim you run must produce "one memorable rule + exception conditions" before it counts as read; run solutions in the study room, and at the table use only internalized heuristics; prioritize building high-frequency universal rules (big circle), then fill in low-frequency marginal nodes (small circle).
  • Limitations: this is the industry'samplified versionof "understanding over memorization" — the general learning principle is amplified here by the solver era into a first-principles learning philosophy, but the degree of amplification has limits: non-standard spots such as marginal/high SPR/multiway pots still require returning to precise solving, and oversimplification (80/20) will be counter-exploited by the mathematically rigorous at high stakes.
  • evidence: [T01-S036, T01-S014, T03-S002]

1.5 Indifference (using frequency to force opponents to equal EV = the engine of equilibrium)

  • In one sentence: theessential mechanismof equilibrium is to use precise bluff-to-value ratios and defense frequencies to make certain marginal opponent hands have exactly equal EV (indifferent)between two actions, so that no matter how they choose they cannot exploit you — MDF = 1/(s+1) is the core readout of this engine.
  • Cross-source endorsement (figures: Bill Chen-Ankenman / GTO Wizard / Andrew Brokos)——Mathematics of Poker : the mathematical substance itself, GTO Wizard's authoritative definitions of MDF & Alpha, and Brokos' Clairvoyance Game minimal model.
  • How to apply: when facing "how much should I defend," estimate the minimum defense frequency by MDF=1/(s+1), then choose specific combos using blockers; when bluffing, set bluff-to-value according to sizing so that opponents are indifferent to bluff-catching; understand that equilibrium is often amixed strategy(the same hand mixes call/fold in proportions) rather than a single correct action.
  • Limitations: indifference is aagainst polarized rangesdefensive upper-bound tool;under short stacks/multiway/ICM you should actively deviate from MDF; real table opponents are rarely the ideal opponent who is "perfect GTO except at this node," so mechanically applying indifference frequencies instead leaves money to opponents who make mistakes.
  • evidence: [T04-S028, T04-S009, T06-S002]

1.6 Non-linear chip value (ICM: cash ≠ MTT, bubble phase changes risk trade-offs)

  • In one sentence: cash-game chips = cash (linear), while tournament chips, after conversion through the prize-pool structure into prize expectation, havediminishing marginal value (ICM), so at the bubble/prize jumps you should fold far more hands than in cash — the correct play for the same hand can be opposite under the two formats.
  • Cross-source endorsement (figures: Dan Harrington / GTO Wizard / Dara O'Kearney)——Harrington on Hold'em : the foundation of M-ratio/zone, GTO Wizard's "MDF vs ICM" series, and The Chip Race's ICM/PKO math.
  • How to apply: First define the format at the start (cash vs MTT/SNG) → MTT introduces an ICM risk premium at the bubble/pay jumps, folding hands that are mandatory calls in cash; cash gamesdon't use ICM tools at all(misusing it = importing the answer from a different game); bankroll is also divided accordingly (MTT variance is far higher and needs thicker buy-ins).
  • Limitations: ICM is a simplified model (it assumes chip count determines the probability of winning, ignoring skill/position); accurate final-table calculations require Monte Carlo + bunching reconstruction; ICM only holds when chips have nonlinear value, and cash games and deep-stack early MTT are approximately chipEV, so misapplying ICM is likewise a mismatch.
  • evidence: [T04-S013, T06-S004, T03-S016]

Standard Playbook

  1. If the opponent/population deviation has been confirmed by a sufficiently large sample, then exploit the deviation using node-locking or a population baseline;otherwise play the GTO baseline by default(deviating without evidence = homemade leaks). Case: Bart Hanson in live $1/3~$2/5 says "don't play perfectly balanced against terrible opponents," but the same approach would be counter-exploited by strong regs at higher stakes, so Bart self-limits to the applicable stakes (evidence: [T01-S040, T03-S007]).

  2. If facing a polarized large bet and with no ICM pressure, then estimate the minimum defense frequency using MDF=1/(s+1), then use blockers to choose the specific combos to defend with that "block the opponent's nuts and don't block the opponent's bluffs". Case: facing a half-pot bet requires about 25% equity / defending about 50% of range; with a short stack or under ICM, actively defend below MDF (evidence: [T06-S002, T04-S009]).

  3. If we have a nut advantage, then use polarized large bets / overbets (high bluff content);If we have only a slight range advantage (no nut advantage), then use concentrated small bets or check to control the pot. Case: solvers vindicated the overbet—only the side with more nuts gets "the license for large polarized bets," and overbetting without a nut advantage is burning money (evidence: [T06-S009, T04-S031]).

  4. If in the MTT bubble / pay-jump stage, then per ICM fold hands that are mandatory calls in cash and pressure short stacks to arbitrage;If in a cash game, then don't use ICM tools at all (chips = cash). Case: during the bubble, big stacks apply pressure and short stacks protect themselves; playing the MTT bubble with a chipEV mindset is a typical major leak (evidence: [T06-S004, T03-S016]).

  5. If you want to understand how to play a certain spot, then first use a cloud site (GTO Wizard) to build the framework/look up pre-solved solutions, then use a local solver (Pio/GTO+) to dig deeper + node-lock;don't start by memorizing node values. Case: the two-layer learning stack of "use the cloud right away → dig deeper locally" is recommended side by side by almost all roundups; build the intuitive framework first, then move to a local solver (evidence: [T02-S004, T02-S027]).

  6. If reviewing and evaluating a decision, then only ask "is it +EV" rather than win/loss (process-over-results);If you notice tilt rising, then inject logic before reaching the threshold, and leave unconditionally once the stop-loss is hit. Case: Negreanu in 2025 studies solver + reviews HH for 3-4h daily, looking only at decision quality and not at individual hands; the mindset camp's inchworm says "raising the floor of your C-game is worth more than raising the ceiling of your A-game" (evidence: [T01-S046, T01-S039]).

  7. If preparing to start a session, then treat game / table / seat selection as thefirst lever(switching to one good table > improving your play by 1%) and count rake/rakeback into session EV;If bankroll does not meet the buy-in threshold for the current stake, then downgrade instead of forcing it. Case: spending hundreds of hours learning solver yet always sitting at reg tables = winrate eaten by rake + strong opponents; cash about 35-65 buy-in, MTT about 75-125 buy-in (evidence: [T03-S026, T03-S017]).

  8. If it is a high-frequency universal spot (standard c-bet, etc.), then use internalized heuristic to pass in seconds;concentrate the limited bandwidth of solver / review / table selection on non-standard spots, difficult hands, top 3 leaks, and weakest tables(triage to grasp the main contradiction). Case: experienced players skip running solutions spot by spot and reviewing hand by hand, and only deploy heavy weapons when deviating from standard; beginners want to run solutions for every spot and fall into "solver poisoning" (evidence: [T03-S002, T03-S011]).


Tool Stack and Selection Decision Tree

directly reference references/research/02-tools.md three-layer structure + 8-node selection decision tree + 9-item pitfall-avoidance checklist. This section is a consistency sanity-check summary.

Tool stack three layers (all meeting the volume threshold):5 essential / 7 scenario-specialized / 3 emerging。last_checked all 2026-06-05.

  • Essential layer 5(covering ≥80% serious players,Decay risk: low): PioSOLVER (local postflop solver, mainly used by elite pros), GTO Wizard (browser cloud solver/trainer, the de facto standard "first stop for learning GTO"), GTO+ (best cost-performance local solver, $75 one-time vendor pricing T02-S027), Hold'em Manager 3 / PokerTracker 4 (HUD + hand database, PT4 the only native Mac), Flopzilla / Equilab (range × board texture + equity, Equilab free).
  • Scenario-specialized layer 7: ICMIZER 3 / HRC (ICM push-fold actuarial calculation for MTT-SNG), PokerSnowie (neural-net approximate AI, build intuition), Hand2Note (dynamic/positional HUD, fastest import), Leak Buster 2 (compared against winner database 465+ leaks, NLHE cash only), mobile trainer (DTO / GTO Sensei / Postflop+), TexasSolver (open-source cross-platform, GitHub stars about 2429 T02-S010).
  • Emerging layer 3(Decay risk: high, needs review in 6-12 months): GTO Wizard AI (cloud instant solving, launched 2023-07, 19bb/100 vs Slumbot, 0.12% pot distance vs Pio all GTO Wizard self-reported T02-S004), GTO Wizard multiway/3-way solutions + GTO Reports (preflop solutions expanded to 9 players + automatic leak detection), AI-assisted review / exploitative AI (HH Analyzer 2.0 + arXiv profit-max agents).

First-principles fork in selection:purpose (build intuition / deep solving / find leaks) × format (cash no ICM / MTT has ICM) × budget and platform (Mac→PT4, HUD-banned rooms→pure solver). Core pitfalls: ① don't memorize solver output as point values (change the input and everything changes) ② don't blindly trust GTO and ignore opponents (the money is in exploitation) ③ don't think single-point node-lock = maximum exploitation ④ don't draw conclusions with insufficient HUD sample ⑤ don't ignore room HUD-ban policies and buy HUD anyway.Structural headwind: HUD-type products (HM3/PT4/Hand2Note/Leak Buster) are squeezed by room HUD bans / anonymous table policies (GGPoker bans external HUD), and their applicable scenarios are shrinking—this is a source of decay rather than a decline of the software itself, so when updating, focus on checking room policies (evidence: [T02-S024, T02-S004]).


Workflow / Pipeline

directly reference references/research/03-workflows.md 6 complete workflows (all last_updated 2026-06-05). Each workflow has separate beginner SOP and senior path, with senior differences covering three categories: skip / optimize / additional.

4.1 Solver research workflow (build ranges → run sim → read nodes/frequencies → node-lock exploitation → extract heuristic)

Set spot inputs (position/line/effective stack/rake) → run or pull up pre-solved → read action frequencies/polarization and indifference points → node-lock for exploitation → compress into a memorable heuristic.Senior difference:SkipSolve per spot (high-frequency universal spots are already internalized; only open solver for non-standard/high SPR/multiway/opponents clearly deviating);OptimizeNode reading method (read the logic of the entire range rather than a single action frequency, use aggregate reports to view a class of boards at once);ExtraDo multi-point/iterative node-lock + reverse stress-test the heuristic.Decay risk: high(solver tools still rapidly rewrite solving capability) (evidence: [T03-S001, T03-S007]).

4.2 Hand history review (flag difficult hands → compare with solver/database → write leak hypotheses)

Flag difficult hands in real time → upload to GTO Wizard Analyze / HH Analyzer or database replay → compare hand by hand against equilibrium to see deviation direction and EV loss → write down leak hypotheses.Senior difference:SkipGo hand by hand (only review flagged hands + the biggest losers, let standard hands pass);OptimizeJudgment criteria (only ask about +EV, not win/loss; review clustered by spot);ExtraDo sample threshold checks (distinguish leak vs variance) + feed conclusions back into the study list to close the loop.Decay risk: medium(automated HH comparison tool iterations + HUD ban policy evolution) (evidence: [T03-S012, T03-S019]).

4.3 Leak finding (filter spots → compare with baseline → locate high-frequency loss points → targeted patching)

Filter the database for large-sample statistics of a certain spot type → compare with GTO frequencies (GTO Reports color blocks) or winner databases (Leak Buster) → sort by 'frequency × EV loss' to lock onto the most costly leak → run workflow 4.1 to build the correct line.Senior difference:SkipScan all filters once (just look at the top 3 in the automated report);OptimizeBaseline choice (use GTO baseline in tough fields to prevent counter-exploitation, use population winner-database baseline in soft fields);ExtraDo sample size sanity checks + distinguish 'your own leak' vs 'being targeted by a specific opponent'.Decay risk: medium(evidence: [T03-S011, T03-S013])。

4.4 Pre-session preparation (game/table/site selection + bankroll check)

Bankroll admission check → choose games and poker rooms with more weak players/low rake → choose tables with weak players → sit to the 'left of the money' to gain positional advantage.Senior difference:OnlineSkipManual table-by-table filtering (use table-finder/scripts; for anonymous tables/rooms, forcibly skip seat selection and make room selection the main lever);OptimizeTreat game selection as the first lever (continuously bum-hunt weak players, include rake/rakeback in EV);ExtraDo a pre-play state check + preset session length/stop-loss.Decay risk: low(table selection has been a 10+ year steady state; the only variable is room policies gradually shrinking the table-selection lever) (evidence: [T03-S026, T03-S024]).

4.5 Bankroll management / moving up and down SOP (buy-in rules, move-up/down trigger lines, variance buffer)

Set buy-in rules (cash about 35-65 / MTT about 75-125, underlying is the Kelly criterion) → set move-up trigger line → set move-down stop-loss → variance buffer judgment (don't get carried by short-term swings).Senior difference: don't fix on one number,OptimizeDynamically adjust based on winrate confidence + whether professional (in practice use half/quarter-Kelly, for cash never stake >5% of roll in a single hand, for MTT >2%);AdditionalBuild a shot-taking mechanism (take a small-budget test shot at higher stakes, retreat once lost) + link bet sizing with mental state (proactively down-stakes and reduce tables in C-game). Mechanical counting This workflow's senior differences are mainly the two categories "Optimization + Additional," with the Skip category being weak (mathematical steady states rarely have skippable steps).Decay risk: low(bankroll/variance is a mathematical steady state) (evidence: [T03-S017, T03-S018]).

4.6 Mental game / tilt management routine (three phases: pre-game, in-game, post-game)

Pre-game warm-up (go through the day's heuristics + confirm state) → in-game inject logic before emotions reach threshold → leave upon hitting stop-loss → post-game cool-down to offload emotions.Senior differences:Skipin-game firefighting on the spot (rely on pre-set trigger-response scripts set before the game + stop-losses positioned in advance, so that during play there is almost no need to fight emotions in the moment);OptimizationReview target (review emotions/decision quality rather than results, use inchworm to simultaneously manage the upper bound of A-game and the lower bound of C-game);Additionalquantify your own tilt profile (which triggers → which tilt → corresponding handling) + link state with stakes/table count.Decay risk: low(psychological steady state) (evidence: [T03-S020, T03-S019]).

The two major external drivers of recent workflow changes:(A) GTO Wizard AI democratizes solvers(launched 2023-07 → 2026 de facto standard): solver research went from heavy local work to instant browser solutions, lowering the barrier to entry but intensifying homogenization at low-mid stakes / declining exploitability; (B) Poker rooms banning HUDs / anonymous tables(GGPoker etc.): pushed review + leak finding + table selection from "database/HUD-driven" to "pure solver-driven," a structural headwind; when updating, focus on verifying (evidence: [T03-S027, T03-S028]).


Expression DNA

High-frequency slang / register markers: by default the subject of speech is range(not a single card), the default yardstick is EV(not win/loss), the default follow-up question is why-over-what. Core terms: range / equity / EV / GTO / exploit / MDF(1/(s+1)) / polarized / c-bet / 3-bet / SPR / ICM / bb/100 / tilt / GII / node-lock. Pronunciation: "c-bet" = one syllable "see-bet," "G-T-O" pronounced as letters, "3-bet" is a preflop raise count (not "bet three times").

Serious vs casual register: serious discussion (theory/review) is highly quantified, stripped of emotion ("Is this hand +EV to GII?" "His line has a very capped range"); casual griping uses slang (cooler / bad beat / fish / nit / donk), but insiders strictly distinguish cooler (genuinely unavoidable, no self-blame warranted) vs one's own misplay (which should be reviewed), and do not use cooler as an excuse to dodge review.

Internal vs external communication differences: when explaining externally/to novices, concepts are expanded (range, variance, EV); with peers, a shared terminology stack is assumed and judgments and slang come directly. Schools also split at the expression level: the equilibrium camp says MDF/indifference/node-lock, the exploitative camp says population tendencies/live exploit/leveling, the mental-game camp says A-game/C-game/tilt profile/reciprocality—four sets of terms expose four worldviews.

5.1 Dialogue sample library (industry voice real-play corpus)

5.1.1 Register for novices / teaching (explaining concepts)

  • 「I'm not trying to play like a solver, but learning to think like a solver. I'm not trying to recall outputs and copy them." (source: T01-S036, original quote, Galfond on simplifying solvers)
  • 「Unless your opponent is especially bad, you should strive to put them on a range of hands, then narrow that range as the action progresses." (source: T01-S042, original quote, Jonathan Little on range thinking)
  • 「Don't ask 'what does the solver do here' — ask why. The principles of GTO are about making your opponent indifferent, not memorizing outputs." (source: T01-S011, paraphrase, GTO Wizard principles-of-gto)

5.1.2 Peer / internal register (short phrases + jargon + blunt judgments)

  • "When you simplify your strategy and focus on concepts, you free up room for great exploitative play. Exploitative play is where all of the money is made." (source: T01-S036, direct quote, Galfond)
  • "Why would you try to play a perfectly balanced game against opponents who are playing terribly?" (source: T01-S040, paraphrase, Bart Hanson's signature rhetorical question live-exploit)
  • "Tilt is any deviation from your A-game and your A-mindset, however slight or fleeting." (source: T01-S037, direct quote, Tommy Angelo's definition of tilt)

5.1.3 Theoretical / professional register (publicly discussing standards/academic, with citations + data)

  • 「GTO is best thought of as a way of thinking rather than a way of playing." (source: T01-S014, direct quote, Brokos's signature line)
  • "Having the model think for 20 seconds in the middle of a hand gave roughly the same improvement as scaling up model size and training by 100,000x." (source: T01-S017, direct quote, Noam Brown on test-time compute poker→LLM)
  • 「Injecting logic works best when you catch the build-up of emotion before you hit your threshold — after that it's an uphill battle to think clearly without quitting. (source: T01-S039, direct quote, Tendler on handling tilt)

5.1.4 Counterexample version (seasoned players in this linewould neversay it this way / sales talk that is mispositioned packaging)

  • "Train with my private coaching for three months and get to NL500, guaranteed profit with steady wins and no losses." (source: §C vendor-talk rejection / T06-S013, why counterexample: violates variance math—win rate needs tens of thousands of hands to be significant, and moving up in stakes is doubly constrained by bankroll + a real sample; promising a fixed-time jump in stakes = funnel marketing)
  • "Learn GTO and you'll be invincible, unbeatable across the world." (source: §C vendor-talk rejection / T06-S015, why counterexample: GTO only gives an unexploitable baseline; exploiting weak opponents has higher EV, and packaging equilibrium as a "guaranteed-win button" is a fundamental misunderstanding)
  • "I make money by reading minds/microexpressions to see through my opponent's hole cards." (source: §C vendor-talk rejection / T06-S015, why counterexample: online there are no microexpressions at all, live tells are a marginal gain; long-term edge comes from ranges + frequencies + game selection + mindset, and selling "mind-reading guaranteed wins" is selling mysticism)

voice_confidence: high—counted sample of 9 passages (3 clients / 3 peers / 3 theoretical), of which 7 passages are figure direct quotes (about 78%), with sources traceable to the voice_samples field of Track 01 figures + Track 06 vendor-talk rejection.


Quality benchmarks + anti-patterns

What counts as "good" (quality benchmarks, 4 specific and verifiable):

  1. Decisions use EV as the yardstick and are decoupled from results: able to say clearly after losing a big pot, "this decision was still +EV," and able to admit after winning money, "I played this hand wrong, I just got lucky."
  2. range-first rather than hand-first: when discussing a hand, first talk about the shape and advantages of both players' ranges on that board, rather than "did I make my hand or not."
  3. GTO baseline + evidence-based exploitation: able to play an unexploitable baseline, and every deviation can point out "which observable leak in the opponent + how much sample support."
  4. Sample-size discipline: before any conclusion like "this line loses / opponents play this way," first pass the sample-size threshold; discuss win rate in bb/100 rather than "how much money was won."

Anti-patterns (common beginner/layperson mistakes, 8 items):

  1. Treating GTO as a "mindless call / unbeatable sure-win method" (actually an equilibrium baseline; exploiting weak opponents has higher EV) (evidence: [T06-S015]).
  2. Putting an opponent on a single specific card rather than a range (evidence: [T06-S001]).
  3. Judging decisions as right or wrong by single-hand wins/losses, and treating short-term climbing as skill (not separating decision from result) (evidence: [T06-S014]).
  4. Completely ignoring rake when calculating EV / talking about win rate (at low-mid stakes, rake eats a big chunk of win rate) (evidence: [T06-S011]).
  5. Applying cash-game thinking to MTTs and ignoring ICM (far more hands should be folded at the bubble/pay jumps than in cash) (evidence: [T06-S004]).
  6. Memorizing solver node values without extracting heuristics (fails as soon as inputs change + creates homogenized play that can be counter-exploited) (evidence: [T02-S027]).
  7. Treating MDF as an ironclad rule that must be strictly followed (should actively deviate with short stacks/multiway/ICM) (evidence: [T06-S002]).
  8. Understanding tilt as "going on monkey tilt and shoving" (actually any emotional deviation from A-game, including getting loose when winning and playing too tight out of fear of losing); not setting a stop-loss lets tilt + downswing compound and erase long-term profit (evidence: [T06-S019, T03-S023]).

Intellectual lineage

School A — Classical hand value / feel-based aggression school:Foundations = Sklansky The Theory of Poker(Fundamental Theorem) + Brunson Super/System; tournament branch Harrington (M-ratio/zone). Current representatives: Daniel Negreanu (reads + small ball, publicly shifted to studying solver daily in 2025), Jonathan Little (a bridging teaching funnel) (evidence: [T04-S027, T01-S046]).

School B — Game theory / solver school:Foundations = Chen & Ankenman Mathematics of Poker + Janda Applications; popularization = Brokos Play Optimal Poker(toy game); textbookization = Acevedo Modern Poker Theory; HUNL decision tree = Tipton. Current representatives: Matthew Janda, Michael Acevedo, Andrew Brokos, the GTO Wizard team, Doug Polk.Internal split:B1 pure GTO/frequency school (memorizing charts) vs B2 practical exploitation school (Bart Hanson population exploit); B3 memorization school (rote-memorizing solver output) vs the simplification-of-thinking school (Galfond Simplifying Solvers 80/20) (evidence: [T04-S015, T01-S035])。

School C — Mindset / mental game school:Foundations = Tendler & Carter The Mental Game of Poker + Angelo Elements of Poker. Current representatives: Jared Tendler (clinical psychology inchworm/tilt profile), Tommy Angelo (reciprocality/awareness philosophy). Both belong to mindset but split in methodology: structured training vs meditative philosophy (evidence: [T01-S019, T01-S021]).

School D — AI feedback school:Foundations = Zinkevich et al. CFR (2007) → DeepStack/Libratus (2017-18) → Pluribus (2019, six-max, trained on a single machine <$150 T04-S005). Current representatives: Noam Brown (poker self-play → OpenAI o1 test-time compute reasoning), Sandholm/Bowling/Moravčík. The value lies in "feeding back into general reasoning" rather than the poker table, and is almost orthogonal to the practical school (evidence: [T04-S006, T01-S017]).

School E — Dissemination school(Orthogonal to A/B/C; the industry growth engine rather than a carrier of a cognitive operating system): Lex Veldhuis (Twitch MTT streaming), Brad Owen / Rampage / Mariano (live cash vlog), Negreanu (public image). Expanding the entry point (entertainment-first) rather than strategic depth.

Core disagreements (preserved, not smoothed over): ① GTO school vs exploitative school (pure equilibrium vs targeted deviation, node-lock is the technical handshake point) ② Cash games vs tournaments (whether ICM fundamentally changes play, MDF fails under ICM) ③ Theory top-down (solver academic school) vs hand-history bottom-up (Galfond-style/intuitive play) ④ The boundaries of AI feedback (6-max "no theoretical guarantee but superhuman in practice", controversy over GTO's applicability at multi-handed tables) (evidence: [T04-S013, T04-S005]).

Undistillable peak signals: Linus "LLinusLLove" Loeliger — widely regarded in the industry as the strongest online NLHE player, yet he almost never speaks publicly or teaches → corroborating that "the tacit knowledge behind the practical peak of school B cannot be directly distilled" (evidence: [T01-S030]).


Honesty boundaries

  • Information cutoff 2026-06-05; tools/workflow modules decay fastest: Emerging solver tools (GTO Wizard AI / multiway solutions / exploitative-school AI) decay high, need review in 6-12 months; review/leak-finding toolchain + poker-room HUD/anonymity/RTA policies decay medium, review in 12-24 months; mental models/terminology/bankroll math/mindset methodology decay low (CFR/Nash/variance math do not go out of date). Recommend running an update every 6 months, focusing on verifying solver tool iteration + poker-room policies.
  • Chinese-English imbalance (industry fact, not insufficient collection): In this industry, first-hand canon (classic books/poker AI papers), figures (YouTube/podcast/blog), tool documentation, terminology definitionsare almost 100% English; the Chinese-speaking world consists of community paraphrases/translations, and many fall on the blacklist (Zhihu/WeChat Official Accounts/Baidu/CSDN) → no independent Chinese-language canon/figure/terminology enters the canon first-hand. This skill's thinking framework, voice DNA, and conceptual layer are almost entirely inherited from English first-hand sources — top-tier theory production and AI research are all conducted in English. Non-English-speaking users need English listening and reading ability to keep up with the latest developments.
  • Closed-source dominance + figures are mostly vendor self-reported / industry-consensus orders of magnitude: Core tools (Pio/GTO Wizard/HM-PT) are closed-source commercial software; maturity and benchmarks (GTO Wizard AI 19bb/100, 0.12% pot distance vs Pio — both self-reported by GTO Wizard T02-S004) lack hard data from independent large-scale third-party surveys; bankroll buy-in multiples (cash about 35-65 / MTT about 75-125), bb/100 strong win-rate thresholds areindustry-consensus orders of magnitude rather than a single authoritative hard number, and vary across coaches/game types; one should give ranges + "adjust according to winrate confidence and whether one is professional" rather than fixed numbers.
  • Tacit knowledge cannot be distilled + individualized thresholds: The strongest online players (Linus Loeliger and others) almost never speak publicly, and the tacit knowledge of the practical peak cannot be directly distilled; the specific numbers for tilt routines / stop-loss (leaving after losing N buy-ins) are highly individualized, and this skill provides structure (preset-monitor-stop-loss-unload) rather than universal thresholds.
  • Does not replace practical experience + poker-room policies/compliance vary by region: The master skill is a thinking advisor, not RTA (real-time assistance = cheating, violating all poker-room rules); poker-room rules (banning/allowing HUDs, anonymous tables, rake structures, RTA enforcement) are privately set by each operator with no unified standard, and must be checked before choosing a venue; the legality of poker gambling varies by jurisdiction, and this skill discusses only strategy, not compliance/operations.

Time-decay Registry

This skill's modules decay at different speeds. Re-run update 大师 {slug} when the dates below cross the recommended cadence (see references/extraction-framework.md § 8).

Module last_updated decay_risk Recommended refresh cadence
Mental models last_updated: 2026-06-05 decay_risk: low 1-2 years
Standard playbook last_updated: 2026-06-05 decay_risk: low 6-12 months
Tool stack last_updated: 2026-06-05 decay_risk: high 3-6 months
Workflows / pipeline last_updated: 2026-06-05 decay_risk: high 3-6 months
Expression DNA last_updated: 2026-06-05 decay_risk: low 6-12 months
Sources (Track 5) last_updated: 2026-06-05 decay_risk: medium 6 months
Glossary / standards / regulations last_updated: 2026-06-05 decay_risk: medium 6 months (regulations may force sooner)
Intellectual genealogy last_updated: 2026-06-05 decay_risk: low 1-2 years
Honest boundaries last_updated: 2026-06-05 decay_risk: low re-assess each refresh

last_updated values reflect the synthesis date. Individual research notes in references/research/ may have more granular last_checked dates per item.