← All posts
4/15/2026

The Kelly Criterion for Blackjack: How Much to Bet at Each True Count

A practical guide to applying Kelly and half-Kelly bet sizing in blackjack, including exact bet bands by true count and why full Kelly destroys most bankrolls.

If you've learned basic strategy and a counting system and then started playing flat bets, you're a counter in name only. The edge from counting comes entirely from betting more when you have the edge. Every hour you spend flat-betting through positive counts is money you're leaving on the table — and yet most new counters over-correct and ramp too aggressively, which lights their bankroll on fire from variance.

The practical answer is the Kelly criterion, specifically half-Kelly. This post is the math, the exact bet bands by true count, and why full Kelly is a trap.

The Kelly formula in one line

For a game with edge e and variance v, the Kelly bet fraction is:

f = e / v

For blackjack, per-hand variance is approximately 1.3 (a bit over 1 because of doubles and splits). That means full-Kelly bet fraction is roughly:

f ≈ edge × 0.77

At +1% edge, full Kelly says bet ~0.77% of bankroll. At +2% edge, ~1.5%. Scale by bankroll to get dollars.

Why full Kelly destroys bankrolls

Full Kelly maximizes expected log-wealth over an infinite sequence of bets. In theory, optimal. In practice, the path to that optimum runs through 50% drawdowns that almost nobody can tolerate emotionally or financially.

Simulation after simulation of Kelly-bet blackjack shows:

Half-Kelly is the practical answer: cut every Kelly bet fraction in half. You give up roughly 25% of long-run growth rate but dramatically reduce variance. You still make money in the long run, you just don't need nerves of steel to stay in your chair.

Quarter-Kelly exists for very conservative players. Extra safety, more EV left on the table. Use it if half-Kelly still swings you too hard.

The half-Kelly ramp by true count — 6D S17 DAS LS

This is the practical bet table for a typical Vegas Strip game. Each TC has a bet band in units (multiples of your base unit):

| TC | Bet range (units) | |---|---| | ≤ 1 | 1 | | 2 | 1–2 | | 3 | 2–4 | | 4 | 4–8 | | ≥ 5 | 8–12 (capped) |

Read as: at TC ≤ 1, flat-bet the minimum. At TC +3, push to 2–4 units. At TC ≥ 5, cap out at 12 units (or your table's max, or the upper bound of "this won't draw heat," whichever is lowest).

This is APTrainer's canonical ramp, used by the landing test scoring and the grind simulator's coached mode. It's approximately half-Kelly for a player with 1000-unit bankroll. Adjust for:

Units and bankroll sizing

Start from bankroll and work down:

Why 1/1000? Roughly, a 1/1000 sizing at a 1–12 spread on 6D S17 DAS LS yields ~1% risk of ruin over hundreds of hours. Smaller bankrolls = larger RoR; larger bankrolls = less risk but lower percentage growth.

Spread choice

Wider spreads capture more EV. They also draw pit attention and increase variance. Common choices:

The EV gain from 1–12 → 1–20 is real but single-digit percentage. The heat increase is enormous. Most sustainable counters stay at 1–12.

Risk of ruin (RoR) basics

RoR is the probability you lose your entire bankroll before doubling it. Depends on:

Rules of thumb for 6D S17 DAS LS at a 1–12 spread:

Halving the spread from 1–12 to 1–6 roughly halves the variance — and thus halves the RoR at the same bankroll — at the cost of maybe 30% of the EV.

The takeaway: 1000 units at 1–12 half-Kelly is the canonical "reasonably safe" baseline. If your bankroll is smaller, either reduce spread, reduce Kelly fraction, or accept higher RoR.

Flat-betting and Wonging

Two alternative bet-sizing strategies show up often:

Flat-betting the count: betting the same amount every hand regardless of TC. This is not a counter strategy — it returns the raw house edge. There's no reason to count if you flat-bet. Some players flat-bet as "learning practice" without real money; fine. Don't play real money flat while counting.

Wonging (back-counting): standing behind the table, tracking the count silently, and sitting down only when TC crosses a threshold (commonly +2 or +3). Benefits: EV per hand played is much higher. Drawbacks: hands per hour drops dramatically (you skip most of the shoe), most casinos prohibit mid-shoe entry specifically to stop this, and it's obvious — other players and pit staff notice the back-counter.

Wonging pairs well with counting but is situational. Most modern counters use a hybrid: play through normally at positive counts, leave the table if the TC goes deeply negative.

The APTrainer bet-sizing check

Our scoring uses the half-Kelly ramp above. If you bet 2 units at TC +1, you get 0 points (too aggressive — the band is exactly 1). If you bet 1 unit at TC +5, you get 0 points (too conservative — the band is 8–12). Bet 6 units at TC +4? Correct — 4–8 range.

The landing test scores your bet sizing across 15 hands, with 30 points dedicated to betting + true-count estimation. The bet-sizing training module drills the ramp with 25 scenarios and progression questions.

Common bet-sizing mistakes

Flat-betting the count. The single most common mistake. Counting without varying bets returns the base house edge; you're doing the mental work of a counter for a non-counter's result.

Over-spreading on negative counts. Some counters press bets after a string of losses — Martingale thinking sneaking in. Never press on a negative count. The house edge is already working against you; larger bets just lose faster. Bet the minimum and wait for the count to turn.

Under-betting on strong positives. The opposite failure: a +5 TC shoe appears, and the counter bets 4 units out of fear of variance. The counter edge lives in the upper half of the ramp. Not pressing at +5 leaves most of the EV on the table. If you can't stomach 8–12 units at TC +5, reduce your unit size so the dollar amount is palatable — don't collapse the ramp.

Splitting your bankroll into too many trips. Risk of ruin assumes you play the whole bankroll in one long session. If you only commit $500 to a given trip at a $25 unit, your trip RoR is much higher than your bankroll-long RoR. Either travel with more bankroll, shorten sessions, or reduce unit size.

Following a bet that lost. "I was betting 4 units, I lost, so I'll bet 4 units again to get it back." No — you bet according to the count at the start of the next hand. If the count is now +1 and you should bet 1 unit, bet 1 unit. Previous-hand outcome is irrelevant.

Ignoring table variance. A $25 unit ±1 unit per hand means about ±$30 variance per hand. A $100 unit at a 1–12 spread means variance of several hundred dollars per hand at peak bets. If the table's dollar swings freak you out mid-session, your unit size is too big for your risk tolerance — shrink it.

Get these right and the counter edge actually materializes. Get them wrong and you're counting for no profit — the single most common mistake in new counters who master the arithmetic but never get the bet ramp right.