The Proof

Accuracy
& Backtest

72% ROC-AUC out-of-sample on the unseen 2025 NFL injury set — validated, reproducible, and live. Every claim on this page can be reproduced from the public repo.

What the Industry Tracks.
What INJSUR.AI Adds.

Other platforms rank healthy players. Only INJSUR prices pre-season durability into your board and validates it against real injury data.

Capability Consensus Rankings & DFS Tools INJSUR.AI Why It Matters
Pre-Season Durability Scoring Not offered 20-feature frailty model scored per player Workload, contact rate, age, experience and injury history quantified before Week 1
Injury-Risk Ranking Accuracy No pre-season signal (effectively 50% — coin flip) 72% ROC-AUC out-of-sample on the unseen 2025 injury set Calibrated on 2024 labels, evaluated against 2025 with time-correct features — no leakage, fully reproducible from the repo
Position-Calibrated Risk Bands Generic "questionable / doubtful / out" labels Per-position sigmoid calibration (QB, RB, WR, TE, OL, DL, LB, DB) A high-risk RB and a high-risk QB mean the same thing on INJSUR’s scale
Historical Injury Integration Current-week status only 5 recency-weighted injury features (count, severity, games missed, structural, soft-tissue) A player with two ACL tears scores differently than a player with zero — before they even take the field

All INJSUR.AI metrics are computed from public NFL statistics. The model is calibrated on the 2024 notable-injury labels and evaluated out-of-sample against the unseen 2025 set. Run the backtest yourself in the app.

Out-of-sample · 2025 holdout
72% AUC
The Trained Frailty Model Against the Unseen 2025 Notable-Injury Set

Every player was scored by INJSUR’s frailty model using only pre-2025 history in the features. Across every (2025-injured · healthy) pair, the model ranked the injured player higher 72% of the time, and players flagged elevated-risk missed time at ~1.7× the league base rate. The fit never saw 2025 labels — this is the honest pre-season ranking accuracy, recomputed live on the Results tab.

Live · 2026 Season
In Production
The Same Model, Now Scoring Every 2026 Roster Pre-Season

The frailty model that posted those 2025 numbers is now scoring every player on every 2026 active roster — sourced from the official nflverse data feed and refreshed continuously. You see the exact same High-Risk / Medium / Low bands the backtest produced, applied to the season about to start.

Market-Wide Price Points vs. INJSUR.AI

Platform Tier Brands Monthly Cost Annual / Seasonal What You Get
Low-End Aggregators FantasyPros, Draft Sharks, RotoWire $3.99 – $11.99/mo
(annual lock-in required)
$72 – $130+/yr Generic rankings, consensus expert lists, basic draft advice
Betting-Focused Data Action Network PRO, BettingPros $14.99/wk or $24.99/mo $100 – $150/yr Line movements, public vs. sharp money %, sportsbook liabilities
Elite DFS / High-Stakes Establish The Run, RotoGrinders, The Solver $49.99/mo $300 – $500+/season Linear projections, optimizer tools, high-stakes content grids
INJSUR.AI INJSUR.AI $39/season $99 Lifetime
One-time. Forever.
Proprietary risk model scoring 20 workload, contact and injury-history features per player — what no one else tracks
vs. Low-End Aggregators
5× Cheaper

Consensus ranking services cost $72–$130/year. INJSUR gives you injury intelligence for $39/season — no annual commitment, no upsells.

vs. Elite DFS Tools
$99 vs. $500+

High-stakes players pay $300–$500 every season for projection optimizers. INJSUR.AI's $99 Lifetime Pass gives you the one variable they can't see — forever.

The Real Edge
$0 Elsewhere

No other platform publishes pre-season durability scores. You can pay $500/year for projections that assume every player stays healthy — or $39 for the tool that tells you who might not.

Out-of-Sample,
on a Season the Model Never Saw

Calibrated on 2024 injury labels, evaluated against the unseen 2025 set with time-correct features — no names, no IDs, no outcome leakage.

Ranking Accuracy · 2025 Holdout
72%
72% ROC-AUC. Across every (2025-injured · healthy) player pair, the model ranked the injured player higher 72% of the time — a coin flip is 50%, and this held out-of-sample on a season the model never saw.
Elevated-Risk Tier Lift
1.7×
Players flagged elevated-risk missed time at ~1.7× the league base rate in 2025 — about 1 in 3 flagged players vs ~1 in 5 league-wide.
Top-20% Capture
32%
The riskiest 20% of the pre-season ranking accounted for ~1/3 of all 2025 missed-game injuries — roughly 1.6× a random 20% slice.
Top-half recall
78%
78% of 2025 injured players ranked in the top half of the risk distribution before the season — the durable half produced just ~1 in 5 injuries.

Full honesty: 72% ranking accuracy still leaves 28% of injured-vs-healthy pairs misranked, and no model can predict a purely random contact injury. Every number is recomputed live on the Results tab and reproducible from the repo — not cherry-picked.

Backed by out-of-sample ranking accuracy The frailty model is calibrated on 2024 labels and then evaluated against the unseen 2025 injury list with time-correct features — no peeking at the season being scored.
72%
ROC-AUC ranking accuracy (2025 holdout)
32%
Top-20% capture rate (2025 holdout)
1.7×
Elevated-risk tier lift (2025 holdout)
78%
2025-injured ranked in top half pre-season