Subscribe to download
Slay-I

Slay-I uses a machine learning model trained on over 325, 000 fights to

1. Predict how much damage you will take in a fight

2. Evaluate adding, upgrading and removing cards from your deck according to average damage reduced in a fight

For more information on how the AI works, check out the article I wrote up on Medium [towardsdatascience.com].


Slay-I was trained on over 325, 000 fights from Spire Logs [spirelogs.com] and Jorbs . If you're interested in learning more about how the machine learning model works, checkout the Google Colab notebook [colab.research.google.com] and train it yourself!


Cards are evaluated based on the average damage reduced in a fight by taking the card. This average is calculated against your current act and next act independently. The calculation is an average between damage saved in boss(es), elites, and hard hallway fights, weighted more heavily towards fights that do more damage.

A percentile ranking is also calculated that compares a card against all the cards in a character's card pool. Percentiles are computed on a scale of 1 to 100. A card that is ranked in the 75th percentile is better than 75% of cards and a card in the 10th percentile is better than 10% of cards (aka is really bad).


Slay-I is generally within +/- 7 HP of how much damage you will take in a fight. For adding, removing, and upgrading cards, the score is the average damage saved in a fight. The model isn't perfect, but it does a pretty good job at evaluating cards!


Because the only data available to train the machine learning model with is from base game characters, the mod is only compatible with base game cards, relics and enemies (aka. Ironclad, Silent, Defect, and Watcher)