Blackjack simulator for practicing card counting and testing strategies

Hi everyone,

I've been studying blackjack advantage play and card counting for a while and noticed that many training tools are either very basic or mainly designed for casual players.

So I decided to build my own blackjack training simulator.

It includes:

• Card counting training with True Count
• Full blackjack gameplay trainer
• Simulation mode for testing strategies
• Detailed statistics tracking (ROI, wager, win rate, TC filtered stats)
• Replay system with mistake analysis

The app is called "TrueCount - Blackjack Trainer" and it is currently available on the Microsoft Store.

I'm mainly looking for feedback from people who practice card counting or advantage play.
If anyone here trains with simulators I'd love to hear what features you think are most useful.

The PRO version is currently free until March 20 if anyone wants to try all the features.

One more feature I forgot to mention:

It can simulate different betting strategies and compare results over many hands. You can also set stop conditions, such as stopping at a target profit, a maximum loss, or after a certain number of hands.

It also supports TC/signal-based analysis, including threshold splits and TC buckets.

I’m interested, but the terminologies, such as ROI and TC filtered stats, are unfamiliar to me. We often use expected value (EV) and variance per 100 hands to describe the simulation results. Can you make a website for me to try out?

Thanks for the feedback.

I agree that EV and variance per 100 hands are commonly used metrics when discussing blackjack simulations. I'm currently making some adjustments to the app, and I may include EV and variance per 100 hands in a future update if everything goes smoothly.

At the moment it's a desktop application available on the Microsoft Store, so there isn't a web version yet.

Besides the basic simulation results, the app currently allows comparing different players or strategies, testing betting ramps, and applying TC filters to the results. It also includes some session-based analysis such as stop-win / stop-loss simulations and cumulative result plots across trials.

If you have experience using other blackjack simulators, I'd be very interested to hear which statistics or analysis features you find most useful.

I’ve intensively tried another blackjack simulator called blackjack-hack that was developed by Phil. The only thing I’m not very sure about this simulator is how the variance is calculated in the software part. The game of blackjack includes two parts, the main game and the insurance bet. If we neglect the insurance and play the main game only, there is an EV and a variance of the main game; however, if we play both the main game and the insurance game, there is a combined EV and a combined variance of the whole game. I want to know how you would treat this combination.

Good question.

At the moment my simulator does not include the insurance bet. Each round currently models only the main game (including doubles and splits), and the net outcome of the round is recorded directly from that.

So in the current version the EV and variance come purely from the main game. If insurance were included, then as you mentioned the total EV and variance would be the combination of the two parts (main game + insurance), either through the variance decomposition or simply by recording the combined per-round outcome in the simulation.

The main reason I haven't implemented insurance yet is that I'm building the application alone, and adding it would require quite a bit of additional work in the backend: the game flow, strategy logic, and the interaction with counting systems would all need to be extended.

Also, my understanding is that insurance is negative EV most of the time unless the count is high, so I prioritized other parts of the simulator first.

If the simulator ends up being useful to enough players, I would definitely consider adding it later. My current idea would be to include insurance as part of the strategy logic so that different insurance strategies could be tested and compared in simulations. But realistically that probably won’t happen in the short term.

Thanks for the question — I actually just pushed an update that adds EV per 100 hands and variance per 100 hands to the statistics.

The way the statistics are generated in my app is through Monte Carlo simulation. The simulator runs the gameplay hand-by-hand according to the rules and records the outcome of each hand during the run.

After the simulation finishes, the results are saved and the Stats section analyzes those recorded results. From there it calculates metrics like EV/100 and variance/100, and it can also analyze things like strategy comparisons, betting ramps, TC-filtered results, and session-level outcomes such as stop-win / stop-loss simulations.

So instead of only returning a theoretical number, the simulator stores the simulation results and allows different types of analysis on them afterwards.

If you’ve used other simulators like blackjack-hack or CVCX, I’d be very interested to hear what statistics or analysis tools you find most useful in practice.

I’m interested in this saying “The simulator runs the gameplay hand-by-hand according to the rules and records the outcome of each hand during the run.” Let me use an example to ask.

The hand situation is 8-8 vs. A with the original wager of $10. First, player added $5 on the insurance but lost it. Second, player added $10 to split the 8-8 pair into two hands. Third, the first split hand got a card 3, so player doubled it. Fourth, the second split hand got a card 3, so player doubled it too. Unfortunately, player lost both split hands.

In the above example, how do you “record” the EV and variance of this particular hand? If we neglect the insurance, what are the corresponding numbers?

In my simulator, the record is currently kept per round per player, not per individual final hand after splits.
For each round I record values such as:
[ bet_multiplier_initial, total_units_wagered, net_units_won, outcome, num_hands, num_splits, num_doubles, the pre-bet count / signal ]

Using these round-level records:
1. EV calculation
EV is calculated from the total profit/loss and the total number of final hands:
EV per hand = total net units won ÷ total number of hands
EV per 100 hands = 100 × EV per hand
So effectively the simulator aggregates all rounds and divides the total result by the total number of resolved hands.

2. Variance calculation
The variance in my app is currently calculated using round-level totals, specifically net_units_won and num_hands.
In other words, each round contributes a single observation consisting of its total result and the number of final hands. Variance is then estimated from these round-level observations rather than from each individual hand outcome. Because the simulator does not store the result of every final hand separately, this is not the exact orthodox hand-level variance. It is an approximation.
One limitation is that variance can be slightly underestimated. For example, if one split hand wins and another loses within the same round, the round-level net result may partially cancel out, which hides the internal fluctuation. So the variance reported by the simulator should be interpreted as a practical approximation useful for statistical summaries and comparisons, rather than the exact theoretical variance per hand. To compute the orthodox variance, the simulator would need to record each final hand outcome individually after all splits and doubles are resolved. That would allow exact hand-level EV and variance calculations.
In other words, the simulator treats each round as a single statistical observation, while the number of resolved hands in that round is used to normalize EV per hand.

3. Insurance
Currently my backend does not implement insurance rules, so the simulator cannot calculate insurance outcomes yet.
If insurance is added later, the proper way to evaluate it would be to compare simulation results:
with an insurance strategy enabled, and with no insurance strategy and then measure the difference between the two strategies.

I’d like to use the above example hand 8-8 vs. A to try the calculation.

With insurance:
EV = -0.5-2-2=-4.5 units;
Variance = (-4.5)^2=20.25.

Without insurance:
EV = -2-2=-4 units;
Variance = (-4)^2=16.

The overall expected value is approximately 0 after billions of hands in any case; however, the overall variance will increase a lot with insurance. Is this right?

I think there is an important distinction here between realized outcome, expected value, and variance.

In your example, −4.5 (with insurance) or −4 (without insurance) is the realized result of that round, not the EV, and not the outcome of an individual hand.

Also, variance is not simply X². In general it is
Var(X) = E(X²) − [E(X)]²
So for a single observed round, you can say the realized result was -4 or -4.5, but that alone does not give the variance.

Another important point is that my simulator records results at the round level, not at the individual hand level.
When splits occur, a single round may contain multiple final hands. Instead of recording each final hand separately, the simulator records the total result of the round and the number of final hands that occurred in that round.
For example, ignoring insurance, this round would be recorded approximately as:
net_units_won = -4, num_hands = 2
So this round is treated as one statistical observation, while num_hands is used to normalize the EV calculation.
Therefore its contribution to the EV summary would be: -4 ÷ 2 = -2 units per hand

Also, I would not say the overall expected value becomes approximately 0 after many hands. Rather, after many hands the sample mean converges to the true EV of the strategy, and that EV can be negative, positive, or near zero depending on the rules and strategy used.

Here is the argument I’ve argued many times. Some people say taking insurance will lower the overall variance but at the cost of losing money. I doubt it. It seems to me like it will increase both the variance and the EV.

Also, in the simulators I tried, the number of hands within a round was often not used to calculate the EV. You say “Therefore its contribution to the EV summary would be: -4 ÷ 2 = -2 units per hand,” but I would not divide by 2.

A couple of points.

First, −4 or −4.5 in that example is just the realized outcome of that round, not the EV.

Second, variance is not X². In general Var(X)=E(X²)−[E(X)]², so a single outcome does not define the variance.

Regarding the division by 2: it depends on how the simulator defines the observation unit. In my simulator the observation unit is a round, which may contain multiple final hands after splits, so I normalize the result by the number of final hands.

Also, EV does not approach zero with more hands. The sample mean converges to the true EV of the strategy.