Prediction Accuracy
After a round finishes, we score the predictions against actual fantasy points (FIFA feed) — error, ranking, scouting-bonus calibration, and how the recommended XI did. The hints feed back into tuning the model.
attacking-output multiplier the tuner learned per position (1.0 = no change)
predicted chance to clear threshold vs what actually happened
| Bucket | Predicted | Observed | n |
|---|---|---|---|
| 0-20% | 10% | 14% | 445 |
| 20-40% | 28% | 29% | 201 |
| 40-60% | 49% | 30% | 47 |
| 60-80% | 68% | 29% | 7 |
- XI actual points (capt ×2)
- 35.0
- Captain
- Ryan Gravenberch (2 pts)
- Captain rank in XI
- #6 of 11
- Best captain (hindsight)
- Ramin Rezaeian
- Avg player (field)
- 3.04 pts
where the model flatters players — chasing differentials is only right if the low-owned buckets aren't inflated
Every squad's real score vs the hindsight-best XIs, plus every big miss classified to the model factor that caused it — so each round makes the next projection better.
- DEFMateo Chávez0.6 →16
Model gave him only 4% to start, but he played and returned 16. The appearance model missed his role.
- FWDYoane Wissa3.4 →17
Model gave him only 51% to start, but he played and returned 17. The appearance model missed his role.
- MIDPape Gueye4.1 →17
Scored 17 largely through tackle/chance volume the per-90 rates didn't credit (model expected 0.5 pts from those routes).
- FWDNicolas Pépé2.6 →14
Low goal prior (12%) but hauled 14 — likely a finishing spike. Check role support before chasing.
- MIDRiyad Mahrez4.6 →16
Low goal prior (9%) but hauled 16 — likely a finishing spike. Check role support before chasing.
- MIDSebastian Berhalter1.9 →13
Model gave him only 27% to start, but he played and returned 13. The appearance model missed his role.
- MIDCrysencio Summerville13.5 →1
Ceiling pick projected 13.5 but returned 1. High-variance blank; the miss cost is amplified if captained.
- MIDMichael Olise11.8 →2
Ceiling pick projected 11.8 but returned 2. High-variance blank; the miss cost is amplified if captained.
- MIDDaichi Kamada11.5 →2
Ceiling pick projected 11.5 but returned 2. High-variance blank; the miss cost is amplified if captained.
- MIDRyan Gravenberch11.4 →2
Ceiling pick projected 11.4 but returned 2. High-variance blank; the miss cost is amplified if captained.
- MIDMohamed Salah9.8 →1
Ceiling pick projected 9.8 but returned 1. High-variance blank; the miss cost is amplified if captained.
- MIDBrahim Díaz10.8 →2
Ceiling pick projected 10.8 but returned 2. High-variance blank; the miss cost is amplified if captained.
- 9×Ceiling pick blanked — captaincy/exposure sizing, not projection
- 7×Start-probability model (appearances / rotation signal)
- 2×One-off finishing spike — do NOT over-correct, check role support
- 1×Per-90 tackle / chance-created rates (volume routes)
- 1×Variance blank — played, no returns; only act if it repeats
backtested 4 Aug, 08:15 UTC