BIS 2026
Brownlow Intelligence

Technical Summary

Methodology

Structured reference for technical recruiters. Implementation clutter omitted.

Model Summary

Final Model
CatBoost
Training
2015–2025
Forecast
2026
Features
178
Numeric
173
Categorical
5
OOT RMSE
0.355865
Champion Experiment
base_plus_components

Validation Design

Each validation season was predicted only using earlier seasons. Rolling out-of-time folds evaluate the model on seasons that were never used for that fold's training. 2026 was never used for model tuning.

  1. 2015–2021

    → 2022

    RMSE 0.315314

  2. 2015–2022

    → 2023

    RMSE 0.380906

  3. 2015–2023

    → 2024

    RMSE 0.370621

  4. 2015–2024

    → 2025

    RMSE 0.356620

Modelling Pipeline

  1. Performance Data
  2. Match-Relative Features
  3. Coach Intelligence
  4. CatBoost
  5. Match Constraint
  6. Season Forecast

Match Constraint

  • →0 ≤ expected vote ≤ 3
  • →Sum expected votes per match = 6

Limitations

  • Expected votes are forecasts.
  • Umpire deliberations are unobserved.
  • Historical relationships are not causal.
  • Model-ranked Top 3 is not the official ballot.
  • Some 2026 context features show distribution shift.