Lugo Bots

How the ranking system works

Lugo Bots Rating System Glicko-2

This document explains why Lugo Bots moved to Glicko-2, the configuration we use, and which games affect rating.

Why we changed to Glicko-2

Lugo Bots previously used a simpler rating model that worked well at small scale but became less accurate as the competition grew.

We migrated to Glicko-2 because it:

  • Explicitly models rating uncertainty (Rating Deviation – RD)
  • Adapts faster for new or inactive bots
  • Stabilizes ratings for experienced bots
  • Handles upsets and uneven matchups more realistically
  • Is a proven system used in competitive environments

Glicko-2 configuration used by Lugo Bots

Storage format
All rating values are stored internally using scaled integers for precision and performance.

Initial values

Parameter Meaning Value
Initial rating Starting strength 1500.0
Initial RD Initial uncertainty 350.0
Initial volatility (σ) Rating instability 0.06

Constants

Constant What it controls Value
Tau (τ) How quickly volatility is allowed to change 0.5
Scale factor Glicko-2 scaling between rating units and μ/φ 173.7178
Minimum RD Prevents over-confidence (RD never goes below this) 30.0
Note
These values are based on standard Glicko-2 recommendations, with light tuning for competitive programming bots.

Which games affect the rating

Games that affect rating
  • Friendly games
  • Challenge games
  • Tournament games
Games that do NOT affect rating
  • Ghost games
  • Hidden-opponent games
  • Test / debug matches
  • Training or simulation matches
  • Any match with special non-competitive rules

This prevents ratings from being influenced by experimentation or internal testing.

Match weight and importance

Lugo Bots does not modify the Glicko-2 formulas directly to apply weights. Instead, higher-impact games apply multiple standard Glicko-2 updates, preserving the system’s mathematical integrity.

Match type Weight # of rating updates Explanation
Friendly 1.0× 1 Standard competitive match.
Challenge ~1.2× 2 Counts more than casual play, but less than tournaments.
Tournament ~1.5× 3 Higher stakes; ratings move faster and confidence updates more quickly.
How weighting works
The system applies the same match result versus the same opponent sequentially for the number of updates shown above (each update uses the previous update’s output).

Official bots do not affect rating

Matches against official Lugo Bots (Level 0, Level 1, Level 2, etc.) never affect ratings.

These bots exist for:

  • Training
  • Benchmarking
  • Testing
  • Onboarding new developers
Why exclude them?
This prevents rating inflation or deflation through farming AI opponents and keeps ratings focused on real competition between players.