When does piecewise tire degradation modeling outperform linear models in Formula 1 race strategy? An empirical study across seven circuits using OpenF1 2023 telemetry with cross-validated model selection

Tire degradation modeling underpins pit-stop timing decisions in Formula 1 (F1) race strategy, yet the published literature has predominantly assumed a linear relationship between tire age and lap-time loss. This study tests that assumption against a physically constrained piecewise model with automated structural-break (“cliff”) detection, using 2023-season OpenF1 telemetry across seven circuits, yielding 14 compound–circuit combinations that meet a minimum sample threshold. A constraint requiring the post-break degradation rate to strictly exceed the pre-break rate excluded one physically implausible detection identified during the analysis. In-sample model comparison used the Bayesian Information Criterion (BIC), assigning the piecewise model five parameters to conservatively account for the degrees of freedom associated with grid-searching the break location. BIC did not favor the piecewise model for any of the 14 combinations; the closest case, Monaco Hard, showed ΔBIC = +3.0, favoring the linear model once the full search cost was accounted for. Five-fold cross-validation, applied independently across ten random seeds to assess stability, nonetheless revealed two combinations with a seed-stable out-of-sample improvement despite BIC favoring the linear model in both: Monaco Hard (10/10 seeds; 3.0% mean absolute error (MAE) improvement) and Singapore Hard (9/10 seeds; 0.9% MAE improvement). The remaining 12 combinations showed no such benefit. These results indicate that the linear model remains adequate for most compound–circuit combinations, that a conservative BIC can miss small, reproducible predictive improvements detectable only out-of-sample, and that piecewise modeling is associated with improved out-of-sample performance specifically for the Hard compound at two street circuits. We report this pattern descriptively, without a confirmed mechanistic explanation, and, to our knowledge, this is the first study to test linear versus piecewise tire degradation on open-access F1 telemetry using a conservatively parameterized BIC alongside seed-stability-checked cross-validation.