Predict the injury before the cheque.
Shield is a conservative, evidence-first injury-risk engine. Pre-transfer due-diligence on every target and in-season rotation guidance for every player you’ve already signed.
Why this exists
The most expensive mistake in football is buying an injury.
On average, top-flight clubs lose 12–15% of their wage bill to player days unavailable. That figure spikes after every window for the same reason: the buying club didn’t know what the selling club did.
Shield combines public match-load signals, public injury history, and (where clubs share it) biometric and GPS data, into a survival-analysis model that estimates a player’s 30/60/90-day risk profile.
It is intentionally conservative. The cost of flagging a player you skip is small. The cost of missing a player you sign hurt is enormous.
How Shield calibrates risk
- 0–15%Low. Risk roughly in line with peers at age and load.
- 16–30%Watch. One or two soft signals — recent load spike, single soft-tissue history.
- 31–55%Elevated. Multiple signals stacking — Shield will recommend a medical second opinion before transfer.
- 56%+High. Pause. Shield will refuse to clear pre-transfer due-diligence without an explicit override.
What goes in. What comes out.
Shield works with the data clubs already have — and gets sharper when clubs choose to share more.
Inputs
Public + opt-in private signals.
- Match minutes, competition density, days between matches
- Injury history (public reports + member-club submissions)
- Age curve relative to position-specific decline patterns
- Movement intensity (sprint counts, accelerations) from broadcast or club GPS
- Recovery quality — sleep, HRV, RPE — only when the club opts in
Private data is never used to train models without explicit, written consent.
Outputs
Decision-grade reports.
- 30 / 60 / 90-day injury probability with tiered severity
- Top-five risk drivers with feature attribution
- Rotation guidance — minutes ceiling, recovery window length
- Pre-transfer due-diligence PDF, signable by the medical lead
- Comparable injury-history players across leagues
Three layers, one verdict.
Layer 01
Survival analysis
Time-to-event modelling on injury timelines. Estimates the hazard function — the probability of injury in the next N days, conditional on not being injured today.
Layer 02
LSTM workload anomaly
A recurrent network reads the player’s last 12 weeks of load, flags acute:chronic spikes, and detects deviations from their personal baseline.
Layer 03
Random Forest ensemble
Final layer combines survival hazard, workload anomaly, and history features into a calibrated probability — with feature attributions for explainability.
Two moments where Shield earns its keep.
Due-diligence before signing.
Shield runs as the last gate before a Deal Room moves to financial offer. The sporting director, technical director, and head of medical see the same report. If the verdict is Elevated or High, the system requires an explicit override — logged, attributed, and timestamped.
Rotation guidance.
Shield refreshes weekly. Coaching staff see per-player risk movement, recommended minutes ceilings, and recovery-window flags. It does not replace the medical lead — it gives them defensible numbers to bring into the conversation with the head coach.
Shield is in private training. Founding clubs get first access.
The first 25 clubs to sign founding-member terms get Shield in pilot in 2026, with the option to feed in their own GPS and biometric data — under a strict data-sharing agreement, never used to train cross-club models.