Frequently asked questions

• What is Stockkarp?

  Stockkarp is a VGC learning tool that grades the decisions in your Pokémon Showdown! replays.

• How does Stockkarp work?

  Stockkarp uses a top-human-level VGC playing model to review Pokémon Showdown! replays and provide a turn-by-turn analysis. It combines board-state valuations and opponent prediction decisions to grade moves, suggest alternatives, and showcase what could have happened in other plausible scenarios.

  It works best on open team sheet games, where the model can more easily and efficiently understand the possible sets and game states. Closed team sheet games work too: paste your team and Stockkarp reviews your side, guessing your opponent's sets. Expect more variance.

• How accurate are Stockkarp reviews?

  While Stockkarp is powered by a neural network with top-level playing capabilities, the goal of the product is to illustrate the impact of every in-game decision so that reviews can be understood and trusted without having to put your faith in the bot.

• Why did my review miss something?

  Unlike chess, Pokémon is a game where information is hidden from the opponent. Stockkarp's model predicts and samples to fill in missing information in the most likely way, but this means reviews will always be limited by the model's understanding of opponent behavior and the stats and sets Pokémon typically use. Rare scenarios can catch the model off guard.

  Think you've found an issue with a review? Send feedback to [email protected].

• Does Stockkarp work with my anti-meta team?

  Stockkarp learned the format mostly by playing teams that have done well historically. While the team pool was diverse and the model can generalize what it learned, the further a game strays from those teams, the less experience the model has to draw on. Rare Pokémon or unusual sets and spreads can be misjudged, both when grading your choices and when predicting your opponent's.

  Reviews of these games can still catch clear mistakes, but read them with more caution.

• Can Stockkarp be used to cheat?

  Our goal is for the competitive integrity of online competitive Pokémon to be maintained. Stockkarp is a post-game learning tool. It does not and will never expose any custom-position analysis that can be maliciously used in live games. Only signed, Showdown-hosted replays get reviewed by our system.

• Are my reviews private?

  Of course! Your reviews, teams, and battle choices are private and secure. The Stockkarp model will never be trained on users' private battle reviews.

• How was Stockkarp trained and tested?

  Stockkarp's base model was trained using self-play, not human games. After base model training, some public human data was used to fine-tune the model's team preview choices and predictions to better fit human tendencies. No private games were used in the training process.

  To measure the model's strength against real players, we ran a bot on the Showdown ladder. The bot ran one game at a time, supervised. Opponents' data from those games was not used for training. The bot is registered on Pokémon Showdown!'s bot registry.

Something not covered here? Write to [email protected]. Privacy · Terms