Your Unit Size Is on Trial

Unit size is a contract, not a feeling. #

Most staking disasters do not begin at the casino. They begin in the second paragraph of an internal monologue, somewhere between "I am very confident in this one" and "this is a five-unit play." The bettor has already made the bet by the time the size escalates. The escalation is the seal on a decision that the bettor knew, at the moment of writing the stake, would not survive a calm second pass. Unit size is the single most important decision a bettor makes, and it is the decision most often outsourced to mood. Mood is a poor staking model.

The historical record on retail-bettor staking is brutal. A 2023 review of 12,400 anonymized US retail accounts showed that bettors who varied stake sizes by more than 2x across a season (excluding documented Kelly adjustments) under-performed their flat-stake counterfactual by 7.8 units per 100 wagers, even when their pick selection was identical. The bets did not change. The sizes did. The sizes cost more than the bets did.

Kelly is the only adult in the room. #

The Kelly criterion is not a mystery. It is a one-line formula that tells a bettor what fraction of bankroll to wager given an edge and an odds line. Full Kelly is volatile; half-Kelly is the most commonly used variant in serious bankroll management, and quarter-Kelly is appropriate for any bettor whose edge estimates are noisy. A bettor with a 3% edge at standard juice should size roughly 1.4% of bankroll at half-Kelly. That is the math. The math does not care about the bettor’s mood, recent record, or whether the bettor "really likes" the play. Edge plus odds equals size.

The most common abuse of Kelly is selective application — using the formula on bets the bettor likes a lot and ignoring it on bets the bettor likes less. That practice converts Kelly from a staking system into a permission structure. The whole point of Kelly is that it removes the conviction-as-input variable. Once conviction is allowed back into the sizing decision, the system is no longer Kelly. It is freelancing with a calculator on standby. The freelancing always wins out, and the freelancing always loses money.

The escalation pattern is the same every time. #

Mood-driven stake escalation follows a predictable shape. A bettor loses two units across three wagers. The next bet, against his own staking rule, is sized at three units to recover. The bet wins; the escalation is reinforced. Three weeks later the same pattern produces a five-unit stake on a coin-flip side. That bet loses. The bettor has now lost the equivalent of nine flat-staked wagers in two transactions. The flat staker, with the same pick selection, would be down two units. The math difference is entirely the sizing decision, and the sizing decision was made in less than ninety seconds with a pulse rate north of one hundred.

The procedural counter is to lock the stake rule in writing before the slate begins. A flat one-unit baseline is fine. A confidence tier system is fine, provided the tiers are defined ahead of time and the criteria for each tier are observable. A half-Kelly framework is fine if the edge estimates are documented. Any rule that exists in writing before the screen lights up is acceptable. Any rule that emerges in real time, in the presence of a live ticket, is not a rule. It is a confession.

Bankroll is not the same number as net worth. #

A foundational error in retail staking is conflating bankroll with net worth or even with a checking account balance. Bankroll is the explicitly-allocated amount of money assigned to betting, with the understanding that it is at risk and is not needed for any other purpose. The unit size is a fraction of that bankroll. If a bettor has a $5,000 bankroll and a 1% unit size, a unit is $50. The unit is not $500 just because the bettor had a good week or just because $5,000 is a small fraction of net worth. Bankroll discipline only works if bankroll is a fixed number that is reviewed quarterly, not adjusted weekly.

Most retail bettors who blow up did not lose their bankroll. They lost the boundary between bankroll and savings. Once the boundary becomes permeable, the unit size becomes meaningless, because any losing run can be "funded" by a deposit that was not part of the original plan. The deposit pattern is the single most reliable predictor of long-run loss in retail betting data — accounts that deposit three or more times in a single month after losing weeks have a 12-month survival rate below 22%. The math punishes the boundary failure regardless of pick quality.

The stake should be boring even on the best bet of the week. #

A staking system that produces an interesting stake size has stopped being a staking system. The whole point of pre-defined sizing is that the stake is boring — predictable, repeatable, and unrelated to the bettor’s feeling about the specific wager. The best bet of the week is sized the same as the worst bet of the week within the same confidence tier. If the bettor disagrees with that outcome on a specific Saturday, the bettor is no longer running the staking system. The bettor is running freelance variance and pretending it is process.

The benefit of boring stakes is that they survive bad weeks without escalation, which is the single most important property of any retail staking framework. A bettor with a flat one-unit stake who goes 3-7 in a week loses 4.4 units (after vig). A bettor with mood-driven escalation who goes 3-7 in the same week, with the same picks, typically loses 9-14 units because the bigger losses were the ones the bettor felt most strongly about. Strong feelings and weak sides correlate. That correlation is the entire business model of the certainty merchant.

Closing argument from the bench. #

Unit size is the bettor’s primary asset, more important than pick selection by a wide margin. A bettor with a 50% pick rate and a disciplined flat stake will break roughly even after vig. A bettor with a 53% pick rate and mood-driven stakes will typically lose money over a season. The pick edge of three percentage points is worth less than the staking discipline that costs nothing to maintain. The certainty industry sells the picks because the staking discipline is not a product. The retail bettor who wants to actually beat the market should buy what is not being sold.

Stake rules must predate picks. If the size changes with your mood, you are not running a system. You are running theater.

One last consideration for the bettor running a documented confidence tier system rather than flat stakes. Tier discipline only works if the criteria for each tier are observable, written, and applied identically across the slate. A three-tier system (1u / 2u / 3u) with clear edge thresholds (1u for 1-2% edge, 2u for 2-4%, 3u for 4%+) can outperform flat staking by roughly 1.2 units per 100 wagers if the edge estimates are honest. The same tier system, applied loosely on feel, underperforms flat staking by roughly 2.8 units per 100 because the loose application correlates the largest stakes with the weakest edges. The structure works. The discretion does not.