k2dex learns teambuilding patterns from thousands of real tournament teams by fitting a statistical physics Maximum Entropy model that captures which Pokémon attract and repel each other in the teambuilding process. Use it to complete rosters, diagnose team synergy, and explore format-wide patterns.
Give a partial roster and let the model suggest optimal completions. Supports fast greedy fill and full parallel-tempered sampling.
→§02Team Rater & AnalysisRate a full team of six: score and coherence rating, synergies and anti-synergies, and suggested swaps.
→§03Metagame ModelGeneral information about how the k2dex model sees the current metagame, with the most used teams and strongest synergies and anti-synergies.
→§04Pokémon IndexPer-Pokémon breakdowns: usage, best teammates, item builds, and the strongest synergies and anti-synergies for every Pokémon in the format.
→Write-ups on the ideas, findings, and implementation details behind k2dex.
Release notes for the rebuilt k2dex: one model per regulation, a six-slot roster editor with Pokémon-only pins, the Anchor Strength knob, and the Boltzmann-learned model that made the Bias Adjustment slider obsolete.
An interactive comparison showing why the k2dex model gives better team recommendations than the raw co-occurrence counting every other teambuilder uses.
An interactive explainer on the statistical physics behind k2dex. Walk through Ising models, Metropolis sampling, parallel tempering, and mean field with live simulations.