Loading…
SIGNAL AI / NASCAR CUP SERIES
Read the field.
Respect the uncertainty.
Track-aware race simulations and an open view of the evidence behind every estimate. Explore season stats and driver matchups on the separate research page.
Open the race lab ↗ Season stats & head-to-head →Free experimental release. This is a fixed, experimental statistical model. Ratings update with results; algorithm changes require validation. Probabilities are uncalibrated. Historical replay uses reconstructed fields and revised results.
Loading race archive…
01 / RACE SIMULATION
Explore the range of outcomes.
Results
| Driver | Start | Win | Top 5 | Top 10 | Top 20 | Expected finish | Place diff. | DNF |
|---|
Place differential = starting position minus simulated average finish. DNF follows recorded non-running status, including disqualifications; it is not a crash-only estimate. Variability controls are hypothetical scenarios.
SIMULATION / DRIVER & TRACK INPUTS
See what drives the forecast.
Run a scenario to inspect only the history available before that race. Lower percentile scores are better. Track starts below count completed, running finishes used for pace.
| Driver | Team | Prior starts | Track starts | Overall | Recent form | Track fit | Track type | Avg finish | Laps-led share |
|---|
03 / DATA HEALTH & MODEL EVIDENCE
A clear view of readiness.
Update status
Scheduled data updates run while this page is closed. An open dashboard checks status every minute. Failed updates preserve prior valid race records.
Historical coverage
Five completed seasons plus the current season. Coverage counts points races in the schedule; exhibitions and qualifying races are excluded. Raw feeds are retained for audit.
Walk-forward probability checks
Loading validation…
| Outcome | Model Brier | Equal-skill Brier | Relative improvement |
|---|
Lower Brier error is better. A negative improvement means the equal-skill baseline performed better. Results are retrospective, without confidence intervals; they do not establish calibrated probabilities or betting value.
Download validation evidence JSON ↓Model calculations & limitations
Finish percentile is normalized by field size. Running finishes estimate pace; non-running outcomes estimate a separate DNF risk. Recent results decay with a 365-day half-life; pre-2022 results receive 25% weight for later races to reduce cross-generation influence. Small samples shrink toward the driver's overall record or the field prior.
Pace blends overall history (45%), recent eight running finishes (20%), track category (20%), and exact track (15%). Available starting position contributes 8% to the final pace score. These are initial heuristic weights, not optimized coefficients. Monte Carlo ranks sampled pace among running cars and samples non-running finishes separately, with shared event risk and greater drafting-track variance.
Track classes use drafting flags, names, and distance per lap. They are coarse proxies: surface, layout, repaving, equipment, crew, pit strategy, cautions, and tire behavior are not fully represented. Team and historical laps-led share are research context, not independent model adjustments. Weather, practice speed, loop data, pit-stop metrics, DFS scoring, and stage/lap-by-lap simulation are not implemented in this release.
Historical tests use actual entrants and final grid positions, which were retrieved after the race. Within each test, ratings use earlier race dates only. Upcoming forecasts require a current field, inputs refreshed within 48 hours, and a future race date. They are archived with inputs for later evaluation. This model is not automatically promoted based on a single good result.