Most "confidence scores" in fantasy sports products are a heuristic dressed up as a number — high/medium/low mapped to 85/65/45 with nothing behind it. PlayCaller's NFL confidence score is a real model, and the API exposes the same number the validation report does.
What's behind the number
The NFL confidence model is a logistic regression trained on 20,585 observations across 5 NFL seasons (2021-2025). Walk-forward validated AUC-ROC is 0.697, with a 4.61x decile lift and a 54.9% hit rate in the top decile. Every score returned by the API traces back to that same measured model — there's no separate "display" number that diverges from what was actually validated.
Getting it from the API
The fastest path is the pre-composed intelligence dossier:
GET /v1/feed?player=patrick_mahomes
X-PlayCaller-Key: pc_live_your_key_here
One call returns the confidence score, a full signal breakdown showing exactly what drove it, a heat signal (a separate, directionally-validated model — 6.0 percentage-point quartile lift — that flags players whose opportunity metrics are converging before the market prices it in), and a synthesis sentence that's checked against the underlying signal data before it's returned. /v1/feed has no rate cap on the Pro Insights tier, so it's built to be called per-player rather than batched.
Why the score can come back null
If a player is confirmed out, the API returns an availability flag with no confidence score attached, rather than scoring a player who isn't going to play. That's a deliberate choice — a number with no real predictive content is worse than no number.
Ready to build? Start a free 14-day Developer Sandbox — no credit card, API key in under a minute.