Pith. sign in

REVIEW 3 major objections 2 minor 25 references

Predicting Formula 1 Race Outcomes: Decomposing the Roles of Drivers and Constructors through Linear Modeling

T0 review · 3 major / 2 minor · reviewed 2026-08-06 · deepseek-v4-flash

Pith's one-line read This paper claims that the constructor, not the driver, dominates Formula 1 race outcomes, explaining 64.0% of the variance in finishing results during the Hybrid Engine Era.

desk verdict The attached full text is a different paper about noisy-label classification, so the F1 RAPM claims are impossible to check; the abstract alone is not enough to warrant peer review. read the letter →

arxiv 2508.00200 v1 pith:HEGCZUMW submitted 2025-07-31 stat.AP

classification stat.AP MSC 62J07
keywords Formula1RAPMridgeregressionvariancedecompositiondriverandconstructoreffectsHybridEngineEraLOESSsmoothingsportsanalytics
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Formula 1 results are produced jointly by the car and the driver, and this paper tries to separate the two. It adapts a Regularized Adjusted Plus Minus (RAPM) model, a method used in basketball and hockey to isolate player impact, to race results from 2014 through 2024, using time-decayed ridge regression with LOESS smoothing to estimate driver and constructor coefficients over time. The central finding is that constructor team performance explains 64.0% of the variance in race outcomes in that era. This matters because the official points system only shows joint outcomes, making it difficult to compare drivers who never share the same car. A credible decomposition would let teams and analysts compare drivers across constructors and track how the relative weight of car versus driver changes by race phase and context.

What carries the argument

The central object is a time-decayed ridge regression trained on race results, of the kind known as Regularized Adjusted Plus Minus (RAPM): it assigns a coefficient to each driver and each constructor such that each observed finishing outcome is modeled as the sum of the two, then uses ridge penalties plus LOESS smoothing over time to keep the estimates stable and continuous. The mechanism carries the argument because the variance explained by constructor coefficients, relative to driver coefficients, is exactly what produces the 64.0% figure and its variation across race versus qualifying contexts.

What would settle it

Compute the residuals of the additive RAPM model for each driver-constructor pairing over a full season, and check whether residuals are systematically non-zero for specific pairings, for instance, a reigning champion in a backmarker car consistently finishing better or worse than the sum of their individual coefficients. A stable pattern of pairing-specific residuals across multiple seasons would falsify the additivity assumption on which the 64.0% attribution rests.

Watch

Extended reading notes

Core claim

On its own terms, the paper claims that in the Hybrid Engine Era (2014–2024) the constructor is the dominant factor in Formula 1 race outcomes, accounting for 64.0% of the variance explained by the model. It further claims that the constructor share rises in rank-agnostic cohorts such as top-ten points finishers and falls in qualifying sessions, where driver skill matters relatively more. This is presented as a demonstration that RAPM, originally built for team sports, transfers to a motorsport setting where the joint car-driver outcome can be linearly separated. By decomposing performance into individual driver and constructor metrics, the paper claims to create a framework for inter-constructor driver comparisons that the Formula 1 points system obscures.

Load-bearing premise

The load-bearing premise is that a driver's and a constructor's contributions to race outcomes combine additively, with no interaction between them; if great drivers can disproportionately rescue bad cars or if certain cars amplify certain driving styles, the 64.0% split would be biased.

Editorial extensions

If this is right

  • If the 64.0% figure holds, driver comparisons should weight car performance first, and driver quality can only be reliably assessed after stripping out the constructor component.
  • The method yields per-season and per-race coefficient trajectories, allowing teams to track when a car's advantage peaks relative to driver contributions.
  • Because qualifying is claimed to show decreased constructor importance, the model implies that driver skill shows up more in one-lap pace than in race finishing position.
  • The decomposition enables comparing drivers who switched teams, since the estimated constructor coefficients can be subtracted from observed results.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The 64.0% share is a property of the particular outcome variable, model, and era used here; switching from finishing position to points or to lap times could shift the car-versus-driver split.
  • The additive model's linear separability could be tested by looking for systematic interaction residuals, for example, whether a top driver in a weak car systematically over- or under-performs the summed coefficients; the paper does not report such a test.
  • The framework could be applied to other series with shared-car structure, such as MotoGP or endurance racing, or to other F1 eras, to check whether the constructor-dominant pattern is specific to the Hybrid Engine Era.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 2 minor

Summary. The manuscript, arXiv:2508.00200, is submitted under the title 'Predicting Formula 1 Race Outcomes: Decomposing the Roles of Drivers and Constructors through Linear Modeling' and its abstract describes a Regularized Adjusted Plus Minus (RAPM) analysis of Formula 1 races in the Hybrid Engine Era (2014-2024), reporting that constructors explain 64.0% of the variance in race outcomes. However, the full text provided is an entirely different paper: 'Robust Classification under Noisy Labels: A Geometry-Aware Reliability Framework for Foundation Models' by Ecem Bozkurt and Antonio Ortega, about label noise in foundation-model embeddings. There is no methodological description, no data, no equations, and no results related to Formula 1 anywhere in the submission. Consequently, the abstract's central claim, the RAPM pipeline, and all associated findings are unsupported by the submitted manuscript content.

Significance. If the reported results were substantiated, a rigorous variance decomposition of Formula 1 outcomes into driver and constructor contributions would be of genuine interest to sports analytics, team strategy, and driver evaluation. The proposed RAPM approach is standard in other sports and, if carefully applied with proper validation and robustness checks, could yield useful insights. However, because the submitted full text is a different paper entirely, the current submission contains no evidence that these results exist or are reproducible. The potential significance cannot be assessed without the actual F1 analysis. No strengths of implementation—reproducible code, machine-checked proofs, parameter-free derivations, or out-of-sample predictions—are present in the provided materials.

major comments (3)
  1. [Full text (entire document)] The complete full text is a paper by different authors on robust classification under noisy labels, with no connection to Formula 1, RAPM, drivers, or constructors. This is not a local formatting error; it means that every substantive claim in the abstract and title is unsupported by the submission. The 64.0% variance-explained figure cannot be checked, and the methodology is absent.
  2. [Abstract, second sentence and results] The abstract states that 'constructors explain 64.0% of the variance in race outcomes' but does not define the response variable (e.g., finishing position, points, or normalized rank), the model equation, or the estimation sample. Even taken on its own, the abstract gives no indication whether this is an in-sample fit statistic or an out-of-sample predictive metric. Since RAPM is a fitted regression with regularization, the reported R-squared is a property of the estimator and its tuning parameters, not a standalone empirical quantity.
  3. [Abstract, RAPM methodology mention] The abstract says the work 'extends a Regularized Adjusted Plus Minus (RAPM) methodology' but does not state or test the additive-separability assumption that underlies any plus-minus decomposition. If driver and constructor effects interact (e.g., specific drivers are better in specific cars, or car development changes driving style), the variance attribution in a purely additive model is not identified. Without any statement of this assumption or tests of it, the 64.0% figure is at best an artifact of the model specification rather than a measurement of Formula 1.
minor comments (2)
  1. [Abstract, references] The abstract cites 'Sill 2010' and 'Jacoby 2000' but the submission contains no reference list, so these citations cannot be verified or placed in context.
  2. [General publication integrity] The arXiv metadata indicates stat.AP, while the full text is a cs.LG paper. This mismatch should be resolved by the authors before any further submission.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity can be established from the provided text: the attached manuscript is a different paper, so the F1 RAPM derivation is absent.

full rationale

arXiv:2508.00200's abstract describes a Regularized Adjusted Plus-Minus analysis of Formula 1 outcomes and reports that 'constructors explain 64.0% of the variance in race outcomes in the Hybrid Engine Era.' The supplied FULL TEXT, however, is an unrelated paper on robust classification under noisy labels (Bozkurt and Ortega), containing no RAPM equations, no race dataset, no ridge or LOESS specifications, and no driver/constructor coefficient definitions. Without those elements there is no derivation chain to audit. The closest candidate—that the 64.0% figure is an in-sample variance explained rather than an out-of-sample prediction—cannot be verified or refuted from the text. Merely reporting a goodness-of-fit statistic from a fitted model is not circular unless a fitted parameter is renamed as a prediction, and none is visible here. The additivity assumption mentioned in the abstract is a substantive modeling assumption, not a circular step. No quote can be produced showing Equation X = Equation Y by construction or a self-citation carrying the load. Under hard rule 1, no circularity is alleged. Score 0 reflects absence of evidence of circularity, not an endorsement of the F1 analysis, whose methods are outside the supplied text.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

The abstract reveals no new entities, but the model depends on several free hyperparameters and strong domain assumptions about additivity and comparability of race outcomes. These are not justified in the abstract.

free parameters (3)
  • Time-decay parameter
    Controls how quickly older race results are down-weighted; likely chosen by validation.
  • Ridge regularization parameter
    Regularizes the regression; chosen via cross-validation or hand-tuning.
  • LOESS smoothing span
    Smooths coefficient trajectories over time; a smoothing hyperparameter.
assumptions (3)
  • domain assumption Driver and constructor contributions are additive and separable in determining race outcome.
    The RAPM decomposition requires a linear additive model; the abstract does not justify this assumption.
  • domain assumption Race results (points or finishing position) are a valid, comparable measure of performance across seasons and rule changes.
    The model treats race outcomes as a continuous response; the abstract does not address how different race formats or points systems are normalized.
  • domain assumption The RAPM methodology (Sill 2010) is applicable to F1 data.
    The paper extends a technique designed for basketball/hockey; transferability to F1 is assumed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Predicting Formula 1 Race Outcomes: Decomposing the Roles of Drivers and Constructors through Linear Modeling." pith.science (2026). https://pith.science/paper/HEGCZUMW

@misc{pith2026250800200,
  author       = {Pith},
  title        = {Pith review of: Predicting Formula 1 Race Outcomes: Decomposing the Roles of Drivers and Constructors through Linear Modeling},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HEGCZUMW}},
  note         = {Machine review of arXiv:2508.00200}
}
read the original abstract

Formula 1 performance is a combination of the car's ability and the driver's ability. While a given race or season can tell you how well a car and driver performed jointly, isolating the individual impact of the driver and constructor remains challenging. This paper extends a Regularized Adjusted Plus Minus (RAPM) methodology (Sill 2010), commonly used in basketball and hockey, to parse out individual driver and constructor impact. It employs a time-decayed ridge regression with LOESS (Jacoby 2000) smoothing to predict race results for the Hybrid Engine Era (2014 - 2024). By measuring the constructor and driver coefficients over time, we measure the relative individual impact of driver and constructor throughout the period. Results show that constructors explain 64.0% of the variance in race outcomes in the Hybrid Engine Era. Additionally, constructors have increased importance in benchmarked rank-agnostic cohorts (e.g., Top 10 points finishers) and decreased importance in qualifying. By decomposing performance into individual driver and constructor metrics, we create a robust framework for inter-constructor driver comparisons that the Formula 1 points system obfuscates. Our work enhances the understanding of driver and constructor contributions to race success, offering valuable insights for strategic decision-making in Formula 1.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

25 extracted references · 20 canonical work pages

  1. [1]

    On the opportunities and risks of foundation models,

    Rishi Bommasani, Drew A Hudson, Ehsan Adeli, Russ Altman, Simran Arora, Sydney von Arx, Michael S Bernstein, Jeannette Bohg, Antoine Bosselut, Emma Brunskill, et al., “On the opportunities and risks of foundation models,”arXiv preprint arXiv:2108.07258, 2021

  2. [2]

    Co-teaching: Robust training of deep neural networks with extremely noisy labels,

    Bo Han, Quanming Yao, Xingrui Yu, Gang Niu, Miao Xu, Weihua Hu, Ivor W Tsang, and Masashi Sugiyama, “Co-teaching: Robust training of deep neural networks with extremely noisy labels,” inNeurIPS, 2018

  3. [3]

    Mentornet: Learning data-driven curriculum for deep neural networks on noisy labels,

    Lu Jiang, Zhengyan Zhou, Thomas Leung, Li-Jia Li, and Li Fei-Fei, “Mentornet: Learning data-driven curriculum for deep neural networks on noisy labels,” inICML, 2018

  4. [4]

    Selfie: Refurbishing unclean samples for robust deep learning,

    Hwanjun Song, Minseok Kim, Dongmin Park, Yung Yi Shin, and Jae-Gil Lee, “Selfie: Refurbishing unclean samples for robust deep learning,” inICML, 2019

  5. [5]

    Dividemix: Learning with noisy labels as semi-supervised learning,

    Junnan Li, Richard Socher, and Steven C.H. Hoi, “Dividemix: Learning with noisy labels as semi-supervised learning,” inICLR, 2020

  6. [6]

    Combating noisy labels by agreement: A joint training method with co-regularization,

    Xiaobo Wei, Liyang Zhu, Xiaoshuang Li, Luc Van Gool, and Cees G.M. Snoek Seu, “Combating noisy labels by agreement: A joint training method with co-regularization,” inCVPR, 2020

  7. [7]

    Promix: Combating label noise via maximizing clean sample utility,

    Ruixuan Xiao, Yiwen Dong, Haobo Wang, Lei Feng, Runze Wu, Gang Chen, and Junbo Zhao, “Promix: Combating label noise via maximizing clean sample utility,”arXiv preprint arXiv:2207.10276, 2022

  8. [8]

    Badlabel: A robust perspective on evaluating and enhancing label-noise learning,

    Jingfeng Zhang, Bo Song, Haohan Wang, Bo Han, Tongliang Liu, Lei Liu, and Masashi Sugiyama, “Badlabel: A robust perspective on evaluating and enhancing label-noise learning,”IEEE transactions on pattern analysis and machine intelligence, vol. 46, no. 6, pp. 4398–4409, 2024

Show all 25 references
  1. [9]

    Manifold dividemix: A semi-supervised contrastive learning framework for severe label noise,

    Fahimeh Fooladgar, Minh Nguyen Nhat To, Parvin Mousavi, and Purang Abolmaesumi, “Manifold dividemix: A semi-supervised contrastive learning framework for severe label noise,” inCVPR, 2024, pp. 4012– 4021

  2. [10]

    Making deep neural networks robust to label noise: A loss correction approach,

    Giorgio Patrini, Alessandro Rozza, Aditya Krishna Menon, Richard Nock, and Lizhen Qu, “Making deep neural networks robust to label noise: A loss correction approach,” inCVPR, 2017

  3. [11]

    Joint optimization framework for learning with noisy labels,

    Daiki Tanaka, Daiki Ikami, Toshihiko Yamasaki, and Yutaka Matsui, “Joint optimization framework for learning with noisy labels,” inCVPR, 2018

  4. [12]

    Early-learning regularization: Quieting the confusion in early training,

    Shuo Liu, Zemin Zheng, Xingrui Yu, Bo Han, Gang Niu, Tongliang Liu, and Masashi Sugiyama, “Early-learning regularization: Quieting the confusion in early training,” inNeurIPS, 2020

  5. [13]

    Understanding deep learning (still) requires rethinking generalization,

    Chiyuan Zhang, Samy Bengio, Moritz Hardt, Benjamin Recht, and Oriol Vinyals, “Understanding deep learning (still) requires rethinking generalization,”Communications of the ACM, vol. 64, no. 3, pp. 107– 115, 2021

  6. [14]

    Robust loss functions under label noise for deep neural networks,

    Aritra Ghosh, Himanshu Kumar, and PS Sastry, “Robust loss functions under label noise for deep neural networks,” inAAAI, 2017

  7. [15]

    Generalized cross entropy loss for training deep neural networks with noisy labels,

    Zhilu Zhang and Mert R Sabuncu, “Generalized cross entropy loss for training deep neural networks with noisy labels,” inNeurIPS, 2018

  8. [16]

    Symmetric cross entropy for robust learning with noisy labels,

    Yisen Wang, Xingjun Ma, Zhiyu Chen, Yuan Luo, Jinfeng Yi, and James Bailey, “Symmetric cross entropy for robust learning with noisy labels,” inICCV, 2019

  9. [17]

    Normalized loss functions for learning with noisy labels,

    Xingjun Ma, Yisen Wang, Le Hou, Xuewei Liu, James Bailey, and Quanquan Gu, “Normalized loss functions for learning with noisy labels,” inICLR, 2020

  10. [18]

    Asymmetric loss functions for learning with noisy labels,

    Xiong Zhou, Xianming Liu, Junjun Jiang, Xin Gao, and Xiangyang Ji, “Asymmetric loss functions for learning with noisy labels,” in International conference on machine learning. PMLR, 2021, pp. 12846– 12856

  11. [19]

    Active negative loss: A robust framework for learning with noisy labels,

    Xichen Ye, Yifan Wu, Yiwen Xu, Xiaoqiang Li, Weizhong Zhang, and Yifan Chen, “Active negative loss: A robust framework for learning with noisy labels,”arXiv preprint arXiv:2412.02373, 2024

  12. [20]

    An embedding is worth a thousand noisy labels,

    Francesco Di Salvo, Sebastian Doerrich, Ines Rieger, and Christian Ledig, “An embedding is worth a thousand noisy labels,”Transactions on Machine Learning Research, 2025

  13. [21]

    Graph construction from data by non-negative kernel regression,

    Sarath Shekkizhar and Antonio Ortega, “Graph construction from data by non-negative kernel regression,” inICASSP, 2020, pp. 3892–3896

  14. [22]

    Revisiting local neighborhood methods in machine learning,

    Sarath Shekkizhar and Antonio Ortega, “Revisiting local neighborhood methods in machine learning,” in2021 IEEE Data Science and Learning Workshop (DSLW). IEEE, 2021, pp. 1–6

  15. [23]

    Learning multiple layers of features from tiny images,

    Alex Krizhevsky, Geoffrey Hinton, et al., “Learning multiple layers of features from tiny images,” 2009

  16. [24]

    Medmnist v2-a large-scale lightweight benchmark for 2d and 3d biomedical image classification,

    Jiancheng Yang, Rui Shi, Donglai Wei, Zequan Liu, Lin Zhao, Bilian Ke, Hanspeter Pfister, and Bingbing Ni, “Medmnist v2-a large-scale lightweight benchmark for 2d and 3d biomedical image classification,” Scientific Data, vol. 10, no. 1, pp. 41, 2023

  17. [25]

    Dinov2: Learning robust visual features without supervision,

    Maxime Oquab, Timoth ´ee Darcet, Th ´eo Moutakanni, Huy V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, et al., “Dinov2: Learning robust visual features without supervision,”arXiv preprint arXiv:2304.07193, 2023

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.