Card Counter Trainer
Keep a running count, Hi-Lo or Hi-Opt I, against a dealt shoe at four speeds, scored on accuracy.
This measures two things, whether you can hold an accurate running count (Hi-Lo or Hi-Opt I, whichever you picked) at dealing speed and, with the discard tray on, whether you can read the decks remaining that count has to be divided by. It is not a betting system, and a perfect score on either says nothing about whether counting would show a profit at a table you can actually get to. Separating the two is the honest part: it makes visible how much harder advantage play is at a real table than a trainer handing you a perfect divisor implies.
Tray
Shoe
Keep the Hi-Lo running count as cards deal.
Low (2–6) = +1 · Middle (7–9) = 0 · High (10–A) = −1
Your count
0
Combo
×1
none yet
Score
0
Best 0
Decks left
2.0
of 2
Use ↑ / ↓ (or the buttons) to track the count. At each checkpoint the shoe pauses, your current count is locked as your answer, then pick a bet size.
The bet ramp
Correct counts build a combo multiplier; the true count decides the bet (below 1 = min, 1–3 = mid, 3+ = max).
Where a Hi-Lo shoe crosses into player advantage: +0.72 where the dealer stands on soft 17, +1.07 where the dealer hits it.
What one point of true count is worth
Both systems, against both dealer rules, and where each shoe crosses from a house edge into player advantage. Your system is marked.
| System | Dealer rule | Per pointpct. pts | Crossovertrue count |
|---|---|---|---|
| Hi-LoYour system | Stands soft 17 | 0.56 | +0.72 |
| Hits soft 17 | 0.58 | +1.07 | |
| Hi-Opt I | Stands soft 17 | 0.58 | +0.70 |
| Hits soft 17 | 0.61 | +1.02 |
What is graded
You are graded only on cards you have seen. The twentieth card of a block gets its full time on screen and the checkpoint opens after it, so the answer is always checkable against the cards you were shown. If your count is off, the reveal says so and resets you to the shoe’s own count, which means one slip costs one checkpoint instead of every checkpoint left in the shoe.
The cut card
Penetration is how much of the shoe is dealt before the cut card comes out, and everything behind it is reshuffled unseen. It is the first thing a counter asks about a table: most of a shoe passes with the count near neutral, so the short deep stretches are where counting pays anything at all, and a tighter cut card is what takes them away.
Tray readings a shoe, typical cut card
Rookie, 2 decks1 of 3 checkpoints
Sharp, 4 decks3 of 7 checkpoints
Pro, 6 decks4 of 11 checkpoints
Legend, 8 decks5 of 15 checkpoints
78 cards are dealt before the cut card at your level. Your level is in gold.
The discard tray
With the discard tray on, decks left stops being printed and becomes a tray you read. Every third checkpoint asks what you make it, counting back from the cut card so the deepest one always asks, and a reading inside 0.5 of a deck is graded correct. The reading earns no points of its own, and the summary reports what it did to your true count rather than scoring it.
It is not free, though. The bet at that same checkpoint is graded against the shoe’s own true count, its running count over the exact decks left, so a reading that puts you the wrong side of a ramp boundary costs you the bet even when the reading was inside the band: at 1.5 decks left with a running count of +5 the shoe wants max, and a reading of 2.0 decks, which grades as correct, puts you on mid. Over 500 simulated shoes at six decks that happens at about one prompt in fifteen, and it is not spread evenly: below true count +1 the ramp has one step, so a misreading is free in the stretches where the count was worth nothing anyway and costs you only in the rich shoes counting is played for.
Deck count decides how much of a drill this is at all: Rookie asks once a shoe, at the cut card with half a deck left, where the slider’s floor leaves reading a full deck as the only half-deck miss available and that costs a rung on about one shoe in twelve. Sharp asks three times, Pro four, Legend five. None of those is a sample, which is why the summary does not report a rate off one shoe: it keeps your readings across shoes and reports the tray figures over all of them, with the shoe’s own underneath, and prints no percentage at all below two readings. The record lives in your browser and the summary has a control that clears it.
Counting perfectly against a tray you read a deck wrong is roughly the whole edge a realistic counter is credited with, which is the part a trainer that hands you the number cannot show you. One caveat on the score: a checkpoint that asks for a reading pays no speed bonus, since a clock that rewarded answering before you had looked would be scoring the opposite of the skill, so a tray shoe scores 25 a prompt lower and best scores are worth comparing within a mode rather than across the two.
Written and reviewed by Alexander Korsager · last reviewed September 26, 2026 · corrections
The idea
What this tool shows
The Hi-Lo system assigns +1 to low cards (2–6), 0 to middle cards (7–9) and −1 to high cards (10 through ace). Keep a running total as cards appear, then divide it by the decks still to be dealt. That second number is the true count, and it is the one a counter bets from, because a running count of +6 means something very different with five decks left than with one. This trainer also deals against Hi-Opt I, a second system that tags 3–6 as +1 and tens as −1 while leaving 2s and aces untagged: pick either from the System control before you deal, and the choice locks for the shoe.
The reason the true count is the number that matters is that it converts almost directly into edge, and the conversion rate is checkable.The Wizard of Odds publishes how much removing a single card of each rank moves player expectation in a six-deck game, and across the twenty cards per deck the Hi-Lo count tags as low and the twenty it tags as high, the average move is 0.094 of a percentage point on the game where the dealer stands on soft 17. That per-card effect grows as the shoe shrinks, and the true count divides by the same shrinking quantity, so the two cancel: one point of true count is worth about the same amount of edge at any depth in the shoe. What it is worth in absolute terms, and where that puts the crossover into player advantage, differs by counting system and by whether the dealer hits or stands on soft 17, four combinations rather than one, rendered in the table below the bet ramp rather than typed out here.
That is also why this trainer grades your bet the way it does.Its ramp asks for the minimum below +1, which is conservative for the liberal game and about right for the common one, and the same ramp grades both systems since their crossovers sit close enough together that a second ramp would not change what a bet call teaches. Holding an accurate count is the first of several skills, though, not the whole of it: the Wizard of Odds puts a realistic counter's advantage at 0.5% to 1.5%, and describes a counting bankroll as going up and down like a roller coaster in the short run, with the edge only showing over hundreds of hours of play.
FAQ
Questions players ask
How does the Hi-Lo count work?
Add +1 for each 2–6, subtract 1 for each 10, J, Q, K or A, and ignore 7–9. Divide the running count by the estimated decks remaining to get the true count, which is what you bet from.
What is Hi-Opt I, and how is it different from Hi-Lo?
Hi-Opt I adds +1 for each 3–6 and subtracts 1 for each 10, J, Q or K, leaving 2s and aces untagged along with 7–9. It is balanced, like Hi-Lo, and derived from the same removal-effect table this page uses for Hi-Lo: one point of true count is worth slightly more under Hi-Opt I than under Hi-Lo, on either dealer rule, and the table below the bet ramp has the exact figures for both. The tradeoff is that its main count carries no information about how many aces are left, which matters for betting since aces are worth more to a player than the low cards Hi-Opt I tags the same way. Players who want it at full strength commonly keep a separate ace count and adjust for it; this trainer grades the main count only and says so rather than quietly leaving that step out.
How much is one point of true count worth?
It depends on the counting system and on whether the dealer stands or hits on soft 17: four figures rather than one, each derived from the Wizard of Odds effect-of-removal table for a six-deck shoe and priced against this site's own house edge for that dealer rule. The table below the bet ramp has all four, plus where each one puts the shoe's crossover into player advantage.
Is a top score here enough to make money counting?
No. This page measures two skills of several. The count is one, and estimating decks remaining off a discard tray is the other, which the tray mode grades separately. Sizing a spread the table will tolerate, having a bankroll that survives the swings, and not being asked to leave are still separate problems it does not test. The Wizard of Odds puts a realistic counter's advantage at 0.5% to 1.5%, which is thin enough that it only shows up over hundreds of hours.
Why hide decks remaining behind a discard tray?
Because no table shows you that number. You read a tray, and your reading is a second error on top of the count, landing on the divisor rather than the running count. The two do not average, they multiply: a perfect count read against a tray you are a deck out on is wrong by the divisor alone. With the tray on, every third checkpoint asks what you make it, a reading within half a deck is graded correct, and the shoe summary reports count accuracy and tray accuracy apart so you can see which of the two is costing you.
Is card counting illegal?
No, it is not illegal to think. Casinos are private property, though, and can ask counters to leave or bar them.
Does counting work in online blackjack?
Generally no. Live-dealer and RNG games reshuffle frequently or every hand, which removes the deck composition edge counting relies on.
Methodology
How this is calculated
Nothing here is simulated or approximated.
The shoe is a real shuffled shoe of the deck count your level uses, dealt one card at a time, and the count you are checked against is the running count of the cards already shown under whichever system you picked before dealing: Hi-Lo (+1 for each 2 through 6, 0 for each 7, 8 and 9, −1 for each ten, face card and ace) or Hi-Opt I (+1 for each 3 through 6, 0 for 2, 7, 8, 9 and ace, −1 for each ten or face card).
A checkpoint grades that count over every card put on screen since the shoe started and over nothing else, which is why the card that completes a block is dealt and displayed for its normal interval before the checkpoint opens on it, and each reveal then resets your count to the shoe's own so a single miscount costs one checkpoint rather than every checkpoint after it.
The true count in the panel is that running count divided by the cards left, expressed in decks, for either system.
The bet you are graded on comes from a three-step ramp, minimum below true count +1, middle from +1 to +3, maximum above +3, read off the shoe's own true count rather than off whatever you were working from, and where that first boundary comes from is the part most trainers leave unsaid.
The Wizard of Odds publishes the change in player expectation from removing one card of each rank from a six-deck shoe.
Averaged over the forty cards per deck that Hi-Lo tags, weighted by how many of each rank a deck holds, that comes to 0.094 of a percentage point per card; the same average over the thirty-two cards Hi-Opt I tags comes to about 0.097.
The effect scales with how little shoe is left and the true count divides by the same quantity, so one point of true count is worth about the same amount of edge at any depth in the shoe, though not the same amount under every dealer rule.
The evidence table below the bet ramp has the derived value for both systems against both rules, and the crossover into player advantage each one implies against this site's own house edge for that rule.
The two systems sit close together at a given dealer rule, which is why one shared ramp grades either system honestly; it is the dealer rule, not the system, that moves the crossover enough to matter, which is what makes the ramp's first boundary conservative for the liberal game and about right for the common one.
Common misreadings4
- A top grade is not evidence that counting would show a profit somewhere real. The Wizard of Odds puts a realistic counter's advantage at 0.5% to 1.5% and describes the short run as a roller coaster, with the edge only reliable over hundreds of hours.
- A high true count is a shift in the odds, not a signal that a hand is due. Even several points above the crossover the player edge is a low single-digit percentage, which still loses plenty of individual hands.
- The count deciding your bet is not the count deciding your play. Betting more in a rich shoe while playing ordinary basic strategy is most of what counting is worth; the index plays that alter individual decisions are a separate skill this page neither teaches nor tests.
- None of this transfers to a game that reshuffles every hand. Online RNG blackjack and most live-dealer tables shuffle often enough that no count survives, so practice here is not a route to an edge there.
Assumptions8
- The shoe is shuffled uniformly at random and dealt to a cut card, not to its last card. The three depths are the three the Wizard of Odds runs its own Hi-Lo simulations at, 5, 4.5 and 4 decks of six, and the default is the 4.5 that same page calls the norm for a six-deck shoe. Everything behind the cut card is reshuffled unseen, as at a table, so every count you are graded on is one a real game would reach. Where the cut card falls is itself a checkpoint, so the deepest count of each shoe is always graded.
- Decks remaining is taken from the exact number of cards left. With the discard tray off it is printed, so the true count you work from is exact, which is not a number anyone has at a table. With it on the figure is never printed anywhere on the page: the tray is a magnitude you read, every third checkpoint asks what you make it, and a reading within half a deck is graded correct. The reading itself is awarded no points, and the shoe summary reports what it did to your divisor rather than turning it into a score. It is not free, though. The bet at that same checkpoint is graded against the shoe's own true count, its running count over the exact decks left, so a reading that puts you the wrong side of a ramp boundary costs you the bet even when the reading was inside the band: with 1.5 decks genuinely left and a running count of +5 the shoe wants the top bet at true count 3.3, while a reading of 2.0 decks, which grades as correct, puts you at 2.5 and on the middle bet. Played out over 500 shoes at six decks and the default cut card, a reader out by exactly the band at every prompt is graded correct at every one of them and is still put on the wrong rung at about one prompt in fifteen, which is at least one lost bet on roughly a fifth of shoes. That cost is not spread evenly, and where it falls is the part worth carrying to a table: below true count +1 the ramp has only one step, so a misreading costs nothing in the long stretches where the count was worth nothing either, and everything it does cost lands in the rich shoes counting is played for. A checkpoint that asks for a reading also pays no speed bonus, so a tray shoe scores 25 a prompt lower than the same play without one and the two modes' scores are not directly comparable.
- Those figures are a six-deck shoe's, and the level this page opens on is the one where the tray drill is smallest, so they are not what you meet first. Rookie deals two decks, and at the default cut card that is three checkpoints of which exactly one asks for a reading, at the cut card itself with half a deck left. The slider stops at half a deck, so at that single prompt the only half-deck misreading available to you is upward: read a full deck where half of one is there and you halve the true count you bet from, which costs a rung on about one shoe in twelve over the same 500 shoes. Deck count is therefore what decides how much of a drill a single shoe is: Sharp asks three times a shoe, Pro four, Legend five. The summary does not report a rate off a shoe for that reason. It carries the readings across shoes and reports the three tray figures over every reading you have taken, with the shoe's own figures kept underneath, and prints no percentage at all below two readings, where the honest output is the reading itself. That denominator is the fix for a figure that was unreadable at every level rather than only at this one: five readings is not a sample either, and a tray accuracy off one reading is 0% or 100% and nothing in between. The record is kept in your browser and nowhere else, and the summary carries a control that clears it.
- Half a deck as the tolerance is this site's own convention, not a transcribed one, and it is stated rather than dressed up as a citation because no published source was found giving a working tolerance as a figure. The reasoning is the cost of being wrong. A running count of +6 with three decks genuinely left is a true count of +2; read the tray as 3.5 and you bet from +1.71, read it as 2.5 and you bet from +2.40, read it as 2.0 and you bet from +3.00. At Hi-Lo's per-true-count-point value for the default stand-on-soft-17 game (the evidence table's first row), a half-deck misreading costs at most about half of that value in expectation, while a whole-deck one costs about the whole of it, which is more than the 0.5% floor the Wizard of Odds puts on a realistic counter's entire advantage. Half a deck is therefore the largest band whose cost stays inside the edge, and estimation.test.ts pins that arithmetic against the published removal table rather than against the sentence you are reading.
- The prompts fall at every third checkpoint, laid out backwards from the cut card so the deepest reading of every shoe is always asked for. Three is derived rather than chosen: checkpoints are twenty cards apart, which is 0.385 of a deck and less than the tolerance, so consecutive prompts could be answered with the previous answer and pass both times. Sixty cards is 1.154 decks, more than the band's full width, so an unchanged answer is guaranteed wrong at the second prompt and every reading is a new one.
- The removal figures behind the ramp are published for a six-deck game, in one column for the dealer standing on soft 17 and one for hitting it. This trainer deals two, four, six and eight decks, and the value of a true-count point shifts slightly across that range.
- Every checkpoint is graded on the running count and the bet alone, and both are graded against the shoe rather than against you. The bet the ramp wants is computed from the shoe's own running count over the exact decks left, so a bet that was right for the count you were holding is still marked wrong when that count was wrong. Varying how you play a hand with the count is a separate skill, and the basic-strategy chart it departs from lives on its own page.
- Hi-Opt I's main count leaves both 2s and aces untagged. That is what makes it a single-level, balanced system rather than a claim of equal strength to Hi-Lo: a real Hi-Opt I player who wants full betting accuracy commonly keeps a separate side count of aces and adjusts the true count for how many more or fewer than expected have appeared, because aces are worth more to a player than the low cards this system tags the same way as it tags nothing. This trainer grades the main count only. That is a real simplification, stated here rather than left for a player to discover the gap between a top grade here and a table where the ace count also matters.
Limits6
- Every value in the evidence table's per-true-count-point column is a linear approximation to a relationship that is not exactly linear, drawn from a table published to three decimal places. Treat each as the right size rather than an exact figure.
- The three-step ramp is a simplification chosen so that a bet call can be marked right or wrong. A real ramp depends on the rules, how deep the shoe is dealt and how large a spread the table tolerates, none of which this page models.
- The score, the combo multiplier and the letter grade are this page's own scoring. They rank you against your own previous attempts and against nothing published.
- Your count is checked every twenty cards. Carrying one through a full shoe with no checkpoint to reset against is harder than this makes it look.
- The discard tray is described to screen readers in eighths of the shoe, never in decks, which is about what a glance off an unmarked tray gives and is deliberately not converted for you. It is not too coarse to pass the tolerance band, and neither is the tray a sighted reader looks at: making the words vaguer would penalize one channel without making the drill measure more. Reading a tray to within half a deck at a trainer's pace is not the hard part of this skill. Doing it at a table, and living with what a misreading does to the number you bet from, is.
- Estimation accuracy and count accuracy are reported apart because they do not average, they multiply: the true count you read over the true count that was there equals your count over the shoe's count, times the shoe's decks over the decks you estimated. A perfect count against a tray misread by a deck is wrong by the divisor alone. The shoe summary states that relation and the numbers beside it are pinned against it by test rather than described near it.
Evidence
The removal figures the ramp is justified from are transcribed in count-value.ts from the Wizard of Odds effect-of-removal table for a six-deck shoe, retrieved 2026-09-09, with the source URL and the date in the file header. count-value.test.ts pins each of the twenty published figures as printed, checks that a whole deck's worth of them cancels to about zero as it must, and checks the property that makes the system work at all: every rank Hi-Lo tags moves expectation further, in the direction its tag says, than every rank it ignores.
The 0.094 per-card figure, and the evidence table's four per-true-count-point values and four crossovers, are all derived from that table by count-value.ts's own functions and rendered by the page rather than typed into it, and the thresholds the trainer actually grades against live in ramp.ts and are imported by the tool rather than restated in prose.
The Hi-Opt I tag table (0 for 2, +1 for 3 through 6, 0 for 7 through 9, −1 for ten, 0 for ace) was cross-checked against two independent sources, both retrieved 2026-09-14: Wikipedia's card-counting systems comparison table and CountingEdge's Hi-Opt I page, the latter also the source for the ace-side-count tradeoff stated in the assumptions above.
Its 0.097-per-card figure is derived from the same removal-effect table by the same function count-value.ts uses for Hi-Lo, generalized to take either tag table rather than special-cased, so the two systems' figures cannot drift apart from having been computed two different ways.
The two house edges behind the evidence table's crossovers come from this site's own blackjack engine, the same one behind the basic-strategy chart and the blackjack row of the house-edge table. A passing test means the arithmetic matches the published table, not that the linear approximation holds exactly at every count.
The tray drill's own figures are pinned the same way in estimation.test.ts: the half-deck tolerance is checked against that removal table rather than asserted, the prompt spacing is derived from the tolerance rather than picked, and the relation between count accuracy, tray accuracy and the resulting true-count error is asserted on every reading of a played shoe, so the sentence the summary prints is a statement about the numbers beside it.
What a tray reading costs on the score is pinned there too, because this page had it backwards until 2026-09-10: a shoe counted perfectly, with every reading graded correct, still loses bets wherever a reading crossed a ramp boundary, and the test that says so plays a reader who bets from the tray rather than from a divisor the mode exists to hide.
The one prompt in fifteen is that reader played over 500 shoes in the same test, at six decks, in both directions of misreading, so the figure this page quotes is measured by the file rather than typed into it.
What the default level does instead is measured there too, over the same 500 shoes, and it is a separate test because it was a separate mistake: until 2026-09-11 this page called the six-deck figure the default level's, and the four levels lived in a literal inside the component that no test could import.
They now live in levels.ts, which both the tool and that test read, so the level the page opens on and the level its figures were measured at cannot drift apart again.
Gambling can cause serious harm. The National Problem Gambling Helpline is free, confidential and open 24/7, call or text 1-800-MY-RESET (1-800-697-3738), or chat at ncpgambling.org/chat.
Related tools
Try next
- Blackjack Basic Strategy TrainerSet your table's rules, see the chart they produce and the cells that move, then drill it hand after hand.
- Blackjack Side BetsWhat 21+3, Perfect Pairs, Lucky Ladies and Royal Match cost, at your shoe size and pay table.
- House Edge ComparisonThe house edge on every common casino bet, side by side, and what it costs you per hour.
- Variance SimulatorWatch a downswing play out