Predictions · honest backtest

2026 predicted XIs

The team our model would have picked before every match of the season, against what actually happened. Predicted before the match using only earlier matches (walk-forward).

Predicted before the match using only earlier matches (walk-forward). XIs are chosen without the 100-credit cap (Dream11 role limits and max 10 per team apply); the credits shown are what the picks would cost, using 2025 credits.

Model XI per match

831pts

+76 vs baseline

Baseline XI per match

755pts

last-5-games average

Best possible per match

1,244pts

model got 67% of it

Captain in top 2

27%

baseline 7%

Beat the baseline

52/73

71% of matches

Season total

60,675pts

baseline 55,129

Season so far

Cumulative fantasy points (thousands)

Model XI
Baseline XI
Best possible

Model minus baseline, per match

Best callMatch 17 · SRH v PBKS1,038 vs 503 baseline+536 Worst callMatch 54 · MI v RCB578 vs 954 baseline−376

How these were made

The model was retrained every 10 matches on earlier matches only, then predicted the next 10. It never saw a match before predicting it.
The optimiser picked the best 11 on the predicted mean: Dream11 role limits (1–8 each) and max 10 per team, captain ×2, vice-captain ×1.5, no credit cap.
The baseline XI used each player’s last-5-games average instead; best possible is picked after the match on actual points.
Every XI is scored with the points players actually made.

The final model behind live projections was trained on 2026 too, so it is never used on this page.

Open the Model Lab

2026 predicted XIs

Season so far

How these were made

Every match (73)