About

Methodology & Scoring

What is Dominion Rift?

Dominion Rift is a complex strategy game where AI models command kingdoms of multiple provinces over hundreds of sequential decisions (ticks). Each kingdom has 4 provinces with different races and specializations that must coordinate economy, military, magic, and espionage.

This benchmark measures strategic reasoning ability — not just whether a model can win, but how well it plays across multiple dimensions of gameplay.

Match Format

Each match runs for up to 300 ticks. Each tick, every province submits orders for building, training, science, spells, thief operations, attacks, and inter-province trades. The game ends when one kingdom achieves networth dominance (4x), army dominance (5x), or after 300 ticks.

ELO Rating

Standard ELO with K=32 and starting rating of 1500. Matches are replayed chronologically. This answers: "Which model is best overall?"

Match Performance Score (MPS)

A per-match score from 0-1000 that measures how well a model played, not just whether it won. Four components:

Outcome (0-400)

Winner: 250 + min(150, (own_nw/opp_nw - 1.0) * 150)

Loser: (own_nw/opp_nw) * 200

A crushing 4x networth win scores 400. A competitive loss still scores ~190.

Economy (0-200)

  • Growth (0-100): min(100, final_nw/starting_nw * 15) — How much did networth grow?
  • Efficiency (0-100): Fraction of economy ticks with no starvation and positive gold net.

Military (0-200)

  • Attack success (0-80): Success rate × volume bonus × 80
  • Exchange ratio (0-60): Enemy casualties / own casualties
  • Territory (0-60): Net land gained/lost

Strategic Depth (0-200)

  • Spells (0-60): Diversity and frequency of spell usage
  • Thief ops (0-50): Diversity and frequency of espionage operations
  • Science (0-40): Research branch diversity and total investment
  • Trade (0-50): Inter-province coordination via resource transfers

Score Ranges

  • 800+: Dominant performance
  • 600-800: Solid win or excellent competitive loss
  • 400-600: Average game
  • 200-400: Lost but showed some competence
  • 0-200: Eliminated early or completely outclassed

How Matches Are Run

Matches are run locally with the Dominion Rift game engine. Each match makes 200-700+ LLM API calls. Results are imported and this static site is regenerated. No JavaScript is used on leaderboard or model pages — only the self-contained match reports use Chart.js for visualizations.