Sample-Size Traps and Regression to the Mean
Catalogs the everyday statistical errors players make off tiny samples: judging a strategy change after 5k hands, crediting a coach for a 10-buy-in heater the month after hiring, concluding a new stake is 'soft' or 'rigg
Assumptions: applies to all formats; builds on the confidence interval lesson. Quantitative claims use repo tool outputs recorded in content/mathlogs/mental-game-bankroll/variance-mathematics.md.
Sample-Size Traps and Regression to the Mean
Sample-Size Traps and Regression to the Mean is part of the Variance Mathematics module. Catalogs the everyday statistical errors players make off tiny samples: judging a strategy change after 5k hands, crediting a coach for a 10-buy-in heater the month after hiring, concluding a new stake is 'soft' or 'rigged' from a weekend, and benching a profitable bluff line because it failed three times in a row. Explains regression to the mean and why extreme stretches are followed by ordinary ones regardless of any change you made, with the worked example of a player whose winrate 'jumps' from 2 to 9bb/100 in the 20k hands after a coaching package — a swing entirely explicable by noise at 90bb/100 stddev. Second worked scenario: properly evaluating a preflop strategy change by isolating the affected spots (where 8k occurrences may suffice) instead of overall winrate (where 200k hands would be needed). Teaches the habit of asking 'what sample would I need?' before drawing any conclusion. The quiz tests spotting the sample-size fallacy in five player statements and choosing the correct evaluation method for a strategy change. The lesson has one job: turn a concept that players usually treat as soft into a rule that survives cards, emotion, and account balance. Poker does not give instant feedback. You can make a clean fold and watch the river card that would have saved you. You can make a bad call and get rewarded. A mental-game, bankroll, study, or performance rule is useful only if it helps you act correctly before the result tells a persuasive but unreliable story.
The working definition here is simple: expected value and sample noise are separate quantities. That sentence should not stay abstract. It should decide whether you keep playing, quit, move down, review a hand, change a drill, take a break, or reject a conclusion from a small sample. The player who says he understands the idea but keeps violating it in the exact pressure spot has not learned it yet. He has only recognized the words. Real learning shows up when the next hand is dealt and the old impulse is present but no longer in charge.
This topic also belongs inside poker strategy, not beside it. A player on tilt opens too many hands, calls rivers too quickly, attacks the wrong opponent, or refuses a move-down. A player with bad bankroll rules passes up good games when frightened and takes bad shots when desperate. A player with loose study systems watches content without changing frequencies. A tired player misreads ranges and calls it a cooler. The leak appears in the betting line even when the source is off-table. That is why every lesson in this track uses hand examples as well as written protocols.
The Core Model
Start with the baseline. Before the session, define the rule in neutral language. During the session, identify the trigger early enough that the rule can still work. After the session, write down whether the rule changed a decision. The loop is baseline, trigger, action, review. Baseline is your normal strategy or operating rule. Trigger is the condition that tells you the rule is relevant. Action is the exact behavior you will take. Review is the evidence that decides whether the rule stays, changes, or gets retired.
For Sample-Size Traps and Regression to the Mean, the baseline comes from the curriculum scope. The danger is not ignorance of the topic. Most players already know a slogan that sounds right. They know not to tilt, to use a bankroll, to study more, to sleep, or to avoid judging small samples. The danger is that the slogan has no teeth. A rule with teeth names the exact situation. It says what you do when the trigger appears. It also says what you refuse to do, because poker pressure is best at making exceptions sound reasonable.
A practical rule should fit on one line. For this lesson, write: when the trigger in this lesson appears, I will apply expected value and sample noise are separate quantities, then I will state the sample size, winrate assumption, standard deviation, and conclusion limit in writing. If that sentence feels too plain, that is a feature. Complicated rules collapse during real sessions. The goal is a decision aid that can be used while a timer runs, a dealer waits, or three tables beep at once.
What Changes In Real Hands
In the first hand, the important decision is not whether Hero eventually wins the pot. The important decision is whether Hero notices the correct trigger before acting. Hero stacks off correctly against AK and loses, adding one buy-in of noise to a long graph. That is exactly the kind of moment where players smuggle emotion into analysis. The mind offers a poker-sounding reason: the opponent is out of line, the graph owes me, this stake is easy, I studied this spot yesterday, I cannot fold this strong a hand, or I need the buy-in back. The rule interrupts that story.
Hero's line is correct only if the poker facts and the operating rule agree. Range, position, stack depth, pot odds, and opponent evidence still matter. Mental game does not excuse bad strategy. Bankroll discipline does not make a profitable call unprofitable inside a single hand. Study discipline does not mean forcing a newly learned heuristic into a node where it does not apply. The right use is narrower and stronger: once the strategic facts are named, the rule prevents a non-strategic force from changing the action.
A useful review note for this hand would not say, 'I ran bad' or 'I stayed disciplined.' It would say what changed. For example: trigger noticed on the river, breathing reset used, range rebuilt, fold made despite anger. Or: shot rule fired after losing the allowed buy-ins, stake closed, no exceptions. Or: study note converted into a drill rather than a vague promise. The note needs an observable behavior because future you cannot audit a feeling.
The second hand is the adversarial case. Hero wins three large pots in 600 hands and is tempted to claim a new winrate. This matters because many poker rules look sensible in clean examples and break when the evidence is partial. A player can weaponize any lesson against himself. He can call reckless play courage, call fear discipline, call laziness balance, call obsession work ethic, or call a heater proof of a new skill. The fix is to ask what would have made the conclusion false. If the answer is 'nothing,' the rule is not a rule. It is a rationalization.
In this hand, Hero applies the smallest adjustment that solves the actual problem. If the issue is tilt, the smallest adjustment might be sitting out one orbit rather than quitting the entire day. If the issue is bankroll, it might be ending the shot at the prewritten stop rather than swearing off the stake forever. If the issue is study, it might be reviewing one repeated node rather than rebuilding a whole strategy. If the issue is performance, it might be closing two tables rather than grinding until attention fully collapses. Small, timely adjustments beat dramatic reforms made after damage is already done.
Quantitative Guardrails
This module uses exact variance-tool outputs logged in the mathlog: a 5bb/100 winner with 90bb/100 standard deviation still loses a 1,000-hand sample 43.03% of the time; a 4bb/100 observed winner over 50,000 hands has a 95% confidence interval from -3.89 to 11.89 bb/100; over 250,000 hands the same observed rate narrows to 0.47 to 7.53 bb/100. Risk of ruin examples are also tool-checked: a 2bb/100 winner with 100bb/100 standard deviation has 44.93% ruin risk on 20 buy-ins, 13.53% on 50 buy-ins, and 1.83% on 100 buy-ins when the simplified no-move-down formula is used.
These numbers are not motivational quotes. They are guardrails against false conclusions. If a winning player can lose a normal sample often, then anger after a losing session is not evidence. If a confidence interval still includes losing after tens of thousands of hands, then a short heater is not proof that a strategy change worked. If simplified risk-of-ruin jumps when winrate falls or standard deviation rises, then shot-taking must be more conservative for thin edges and high-variance formats. The point is not to memorize every output. The point is to stop treating surprise as information when the math already predicted wide outcomes.
The assumptions matter. The variance tool uses normal approximations for sample results and a classic risk-of-ruin formula that assumes a positive constant winrate, fixed standard deviation, no withdrawals, and no moving down. Real poker is messier. Moving down reduces practical ruin risk. Withdrawals increase it. A worse game reduces winrate. More tables can increase mistakes and standard deviation. Tournaments add payout concentration that cash-game units do not capture cleanly. Use the outputs as disciplined baselines, then make practical rules more conservative when your life or format demands it.
Protocol
The protocol for Sample-Size Traps and Regression to the Mean has five steps.
- Name the trigger before the session starts. It may be a feeling, a graph event, a bankroll threshold, a study leak, a fatigue sign, or a relationship pressure.
- Name the first observable signal. Do not rely on 'I will know.' Use behavior: snap-calling, opening extra tables, checking the cashier, skipping reviews, ignoring breaks, or changing stakes outside the plan.
- Apply the prewritten action. state the sample size, winrate assumption, standard deviation, and conclusion limit in writing.
- Record the poker decision that changed. The record should include the hand, the trigger, the action taken, and whether the rule protected EV.
- Review on a schedule, never inside the emotional state that created the problem.
This protocol is deliberately mechanical because the failure mode is predictable. In calm study, everyone thinks they will be reasonable. Under pressure, they negotiate. They call the current spot special. They say the rule was meant for other days, other opponents, other stakes, or other emotional states. That negotiation is where bankrolls and winrates leak. The rule exists so you do not have to invent policy while tilted, tired, broke, euphoric, bored, or embarrassed.
Common Failure Modes
The first failure is result-first reasoning. Hero wins after violating the rule and decides the violation was insight. Hero loses after following the rule and decides the rule is too tight. Both conclusions are bad. A single result has almost no authority in poker, especially in the formats covered by this track. Judge the rule by whether it produced the correct decision under the information available at the time.
The second failure is overcorrection. A player identifies one leak and swings to the opposite leak. Revenge tilt becomes passivity. Loose shot-taking becomes permanent fear. Passive content consumption becomes endless solver browsing. Long sessions become avoidance dressed up as balance. The better correction is calibrated. Fix the actual leak, then measure whether the fix is working.
The third failure is vague journaling. 'Played bad because tilted' is almost useless. 'After losing QQ to AK, I opened 38% from early position for the next two orbits and called two river bets without ranging; correction is sit out one orbit after the second physical signal' is useful. Specific notes create future actions. Vague notes create guilt.
The fourth failure is using identity as evidence. Being smart, hard-working, experienced, or usually disciplined does not protect the current decision. Poker punishes identity stories because cards do not care who deserves what. The rule for Sample-Size Traps and Regression to the Mean should be applied because the trigger appeared, not because you feel like the kind of player who needs it.
Practice Assignment
Before your next session, write a one-line rule for this lesson using the form: 'When I notice X, I will do Y, and I will review it at Z.' Pick one hand from the session that touched the topic even if it was not a large pot. Reconstruct the decision without the result. Ask whether the rule changed anything. If it changed nothing, either the trigger did not appear or the rule is too vague. Tighten it.
After three sessions, look for drift. Drift is when the written rule remains in the notebook while your actual play quietly returns to the old pattern. The fix is not more inspiration. The fix is smaller triggers, earlier signals, and more concrete actions. Sample-Size Traps and Regression to the Mean is complete when the rule can be used in an ordinary hand under ordinary pressure, not when the idea sounds convincing on a calm afternoon.