I’ve tracked my Pokémon TCG matches for the past month. Not only did I record the outcome of each game I play on the official Pokémon TCG Online Live platform, but I also logged key details. These include the decks used by both myself and my opponent, the number of prizes left at the end of the game, and whether I had the first move. This data-driven approach goes beyond simple record-keeping. By analyzing my performance trends and understanding my success against varying deck archetypes, I'm obtaining insightful data that assists me in refining my future game strategies. In this article, I'll present the results I discovered after analyzing my data using Python and ChatGPT.
The Lost Zone Experiment
The decks I’ve been focusing on, and the protagonists of this analysis, are of the archetype Lost Box. This type of deck centers on a single strategy: get as many cards as possible in the Lost Zone. The reasons behind this are far beyond the scope of this article, but for now, let’s say that good things happen when you have banished 10 cards to this ominous Lost Zone.
I’ve played 97 matches using six different flavors of Lost Box decks. Each of these is slightly different from the others, as I usually change one or two cards with each tweak. My overall win ratio with this archetype is a satisfactory 67%, which has been steady throughout these 97 matches. You can see in Figure 1 how this ratio has evolved over time and the specific deck version I used at that point.
My first online match using a Lost Box deck ended with a victory. I named this deck "P," after a friend who provided this decklist to me. I soon switched to another variant, termed “Lost Box,” which I used for six matches. Then I changed to what has become my favorite variation, the “Dragonite” variation I used more than 60 times, achieving a win rate of 65% (Figure 2). The release of a new card set, Paldea Evolved (PE), introduced some beneficial cards to the Lost Box archetype, which I’ve been exploring with two different decks. However, they were met with resistance, and my win ratio dropped a bit.
My least successful matchup is against what’s known as the Lugia deck (Figure 3), where I've managed only four victories out of ten encounters. This doesn't surprise me, considering the community acknowledges Lugia as a challenging matchup for Lost Box decks, particularly for the variant I frequently play. On the other hand, my most successful matchups are against the popular new Chien-Pao deck and another one that centers around Miraidon, an electric Pokémon.
The goal of a Pokemon TCG game is to take the six prizes you lay down when the game starts. You get these prizes by knocking out your opponent’s Pokemon. Depending on the Pokémon defeated, players can claim 1, 2, or 3 prizes. On average, I’m left with 4.5 prizes when losing matches, with a standard deviation of 1.9, suggesting that there are numerous instances where I lose without claiming any prizes (Figure 4). Many of these matches involve early concessions due to a poor starting hand or early-game mistakes from which recovery seems unlikely. It's often more productive to concede and progress to the next game, a tactic frequently employed in tournaments where best-of-three rounds within a 50-minute limit are standard.
Conversely, in victorious matches, my opponents typically have an average of 4.1 remaining prizes, with a similar standard deviation of 1.9 (Figure 5).
Notes to myself
Beyond the numbers, I've also been recording qualitative data in the form of notes for each match. These notes have served as a rich source of insights, capturing key moments, mistakes, and learnings from each game. Given the textual nature of this data, analyzing these notes is far from straightforward. A simple count of keywords and phrases like “bad game” would overlook the context embedded in these notes.
So, I resorted to ChatGPT and let it analyze my notes. I provided the chatbot with my comments and asked it to summarize them. Here are its highlights, which I have paraphrased below:
- Concessions: A significant number of your matches ended with either you or your opponent conceding. In some cases, you conceded due to mistakes or a poor start, while in others, your opponent conceded, possibly due to your strong play (thanks for the encouragement, ChatGPT).
- Mistakes and Learning: You often noted when you made mistakes, such as attacking with the wrong Pokémon, setting the wrong energies, or forgetting to use certain abilities. These notes suggest that you're actively learning from your mistakes and looking for ways to improve your gameplay.
- Strategy and Tactics: You frequently commented on your strategies and tactics, such as going for certain Pokémon, managing energies, and handling damage spread. You also made notes to remind yourself of certain strategies, like hitting the [opponent’s] Comfey (this is a Pokémon) and growing the Lost Zone ASAP, or going for the small [Pokémon] ones first.
- Gameplay Reflections: Some notes reflect on the overall gameplay, such as noting a good game, an intense match, or a close game. You also mentioned when you had a great start or when the opponent had a bad deck.
- Specific Cards and Abilities: You often mentioned specific cards and abilities, like Manaphy, Greninja, Drapion, Colress, and Mirage Gate. These notes might indicate key cards or abilities in your deck or important aspects of your gameplay.
- Opponent Behavior: You noted when your opponent was inactive or when they conceded after certain events.
- Self-Encouragement: Some notes serve as reminders or encouragement to yourself, such as "Take it easy and slow," "Don't despair," and "Remember always to know if you’ll be able to retreat."
ChatGPT's analysis aligns with my experience. My notes typically recount the significant events and results of each match, particularly if they were extraordinary in some way. If I misplayed or did a good play using a particular card, I took note of it. If my opponent did something I wasn’t expecting, I took note of it. If I did something silly and needed to vent, I took note of it—a reminder of my misjudgments to help ensure I don't repeat them in the future.
As I prepare for the 2023 Pokémon North America International Championships, my first international tournament, I'm leveraging the power of data to refine my game. Tracking and analyzing my matches has given me valuable insights into my performance, particularly with the Lost Box deck archetype. This data-driven approach, supported by tools like Python and ChatGPT, has been a fun way to learn my strengths and weaknesses. I've confirmed my favorable matchups, identified Lugia as a challenging opponent, and found that, theoretically, I should win 67% of my matches—which won’t happen as I face the world’s best, but I’ll definitely do my best (and read my notes before the 50 minutes starts ticking).
You can find the Python notebook I wrote to analyze the data at: [https://github.com/juandes/data-experiments/blob/main/pokemon-tcg-lost-box/nb.ipynb].