smackastan.github.io/set-trainer/

Have you ever sat around a table playing Set with your family, only to watch in frustration as someone else casually snatched up three cards before you even registered a single pattern?

I built the Set Trainer: a custom web application to apply proven learning principles to master the game, with the goal of giving you everything you need to beat your family at Set.

The Mathematics of Set

To build an effective training tool, I first needed to understand the statistical landscape of the game. A standard Set deck consists of 81 cards, with each card possessing four attributes (number, color, shape, shading), and three possible values per attribute. A valid “set” consists of three cards where each attribute is either entirely the same or completely different across the cards.

Based on these combinations, I classified sets by difficulty into four categories:

1. Easy (1 difference): 1 attribute is different, 3 are the same.
2. Medium (2 differences): 2 attributes are different, 2 are the same.
3. Hard (3 differences): 3 attributes are different, 1 is the same.
4. “Impossible” (4 differences): All 4 attributes are different.

Using combinatorial math and verifying with a million-trial Monte Carlo simulation, I found some fascinating insights about a standard 12-card hand. According to the data, you will find at least one “Easy” set a little more than 25% of the time.

Why is this important? Because when an easy set is on the board, you don’t want to waste time scanning for complex patterns. Conversely, this means that 75% of the time, you will be forced to find a medium, hard, or “impossible” set. An effective training tool therefore needs to balance identifying the easy layups with aggressively practicing the complex structures that make up the vast majority of the game.

Deliberate Practice: Tailoring the Learning Experience

Cognitive science tells us that people learn best through deliberate practice—focusing deeply on specific weaknesses rather than mindlessly repeating entire games.

To implement this, I programmed distinct difficulty modes into the Set Trainer. If you struggle with recognizing simple patterns, you can enable a mode that guarantees an easy set is always on the board. Once you have mastered the basics, you can disable easy sets entirely. By forcing the algorithm to only generate medium and hard sets, you artificially increase the density of complex problems, removing the “noise” of simple sets. This allows you to tailor your learning experience entirely to the areas you struggle with the most.

Mastery through The “Fundamental Theorem of Set”

The centerpiece of the Set Trainer relies on a mathematical property of the game that I like to call the “Fundamental Theorem of Set”: Given any two cards in the deck, there always exists exactly one unique third card that completes the set.

In a high-speed game, the players who win aren’t just scanning randomly; they are matching two cards in their periphery and mentally projecting the exact third card they need, then searching the board for that specific card.

To train this specific cognitive skill, I built Trainer Mode. In this mode, the application displays just two random cards, and you must visualize and construct the missing third card. Once you determine what it must look like, you type it in.

Practicing this fundamental theorem explicitly trains your brain to calculate missing pieces instantaneously. This transition from broad visual scanning to targeted visual searching is the single most important skill in accelerating your reaction time during a real game.

Conclusion

Building the Set Trainer was an exercise in bridging the gap between statistical analysis and human psychology. By identifying the underlying probabilities of the game and creating isolated environments for deliberate practice, this project serves as a comprehensive system for hacking the learning curve of Set.

Give the trainer a try, focus on your weaknesses, and prepare to bring a highly-optimized, data-driven approach to your next family game night.