Pacing Variability, Not Half-Marathon Speed, Distinguishes Sub-3-Hour Success from Failure

A cluster analysis of 9,585 runners at the Berlin Marathon (1999–2025).

Authors

• Aldo Seffrin — Nova O2 Sports Science, São José dos Campos, Brazil
• Elias Villiger — University of Zurich, Switzerland
• Marília dos Santos Andrade — UNIFESP, Brazil
• Thomas Rosemann — University of Zurich, Switzerland
• Katja Weiss — University of Zurich, Switzerland
• Daniel Ferreira — Nova O2 Sports Science, São José dos Campos, Brazil
• Beat Knechtle — University of Zurich, Switzerland (corresponding author)

Status

Manuscript under technical check and editorial review at PLOS ONE.

Key Finding

The Pacing Execution Paradox: Among 9,585 runners who reached the half-marathon at identical sub-3h pace (~1:28:05), those who failed (finish >3:05) had 3.8x higher pacing variability (CV 11.5% vs 3.0%). K-Means clustering (k=2) identified two archetypes: "Positive Split / Fade" (98.3% success rate) and "Metabolic Crash" (71.3% failure rate). Pacing CV alone predicted failure with AUC = 0.965.

Reproducibility

Download the dataset from Zenodo (https://doi.org/10.5281/zenodo.19342683) and place in data/. See data/README.md for details.

pip install -r requirements.txt

Run notebooks in order:

CLEANING → COHORT_DEFINITION → DESCRIPTIVES → CLUSTERING → PREDICTION → FIGURES

Structure

data/                          # Dataset (download from Zenodo)
├── README.md                  # Reproduction instructions

notebooks/
├── CLEANING.ipynb             # Raw CSV → clean parquet
├── COHORT_DEFINITION.ipynb    # Golden Window matching → cohorts
├── DESCRIPTIVES.ipynb         # Table 1 & Table 2
├── CLUSTERING.ipynb           # K-Means pacing archetypes
├── PREDICTION.ipynb           # Logistic regression, Random Forest, ROC
└── FIGURES.ipynb              # Publication figures (TIFF 600 DPI)

figures/
├── Figure_1_Segment_Divergence.tiff
├── Figure_2_Pacing_Profiles.tiff
├── Figure_3_Cluster_Outcome.tiff
├── Figure_4_CV_by_Outcome.tiff
├── Figure_5_Start_vs_Finish.tiff
└── supplementary/
    ├── Supp_1_Clustering_Validation.tiff
    └── Supp_2_ROC_Pacing_CV.tiff

License

MIT — Copyright (c) 2026 Nova O2 Sports Science