{"id":"942113a2-40c4-4e88-bf2e-76f481bd6b3b","arxiv_id":"2508.00200","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":6.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":3,"one_line_summary":"Applying Regularized Adjusted Plus Minus to F1 data, the paper attributes 64% of race outcome variance to constructors across the 2014-2024 Hybrid Era.","lead":"This paper describes a linear-model decomposition of Formula 1 race results into driver and constructor contributions, reporting that constructors explain 64% of outcome variance in the 2014-2024 Hybrid Era. A sports-analytics audience would read it for a new way to compare drivers across teams.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Because the attached full text is an unrelated paper, the RAPM pipeline and the additivity assumption behind the 64.0% constructor-variance claim cannot be checked; UNVERDICTED is appropriate.","rationale":"Good-faith reading: the abstract promises an RAPM extension to Formula 1 with a headline variance split. For that claim to hold, the estimated coefficients must correspond to attributable contributions; this requires additive separability, a regularization scheme that does not distort the decomposition, and a stable variance computation. None of those conditions can be checked because the attached full text is an unrelated classification paper, so the reader's UNVERDICTED verdict is correct. I partially agree with the reader's weakest-assumption identification: additive separability is indeed the most scientifically load-bearing premise, but the immediate blocker is that the submitted text contains no methods or results for the F1 study. A further concern is that the 64.0% figure is not model-free: ridge shrinkage and LOESS smoothing can redistribute explained variance between driver and constructor terms, so even a correctly submitted paper would need to show sensitivity to regularization. I would not reject the work based on the abstract alone; instead I would condition any future acceptance on correction of the submission and on reporting of additivity and regularization checks. Given the current evidence, no substantive verdict is possible, so UNVERDICTED is the appropriate outcome.","tokens_in":5236,"tokens_out":2923,"duration_ms":32381,"concrete_test":"After obtaining the actual F1 manuscript, reconstruct the RAPM specification and recompute the variance decomposition on the same 2014–2024 dataset under two conditions: (1) the reported ridge penalty and a penalty 10 times smaller/larger, and (2) an additive model versus an additive-plus-interaction model that includes driver-by-constructor terms for the top teams. If the constructor share moves by more than a few percentage points under either condition, the headline 64.0% is not robust and the claim must be tempered.","verdict_should_be":"UNVERDICTED","load_bearing_attack":"The only manuscript text attached to arXiv:2508.00200 is a different paper on robust classification; no methods, equations, or experiments for the F1 RAPM analysis are available. The abstract's headline figure—'constructors explain 64.0% of the variance in race outcomes'—therefore rests on an unobservable pipeline. The most substantive scientific risk, if the correct text were retrieved, is the additive-separability premise of RAPM: the claim is a variance decomposition of race outcome into constructor plus driver terms, but ridge regularization and LOESS smoothing alter coefficient magnitudes and can change the split. Without evidence that the model is additive, that interactions are negligible, and that the 64.0% figure is stable under regularization, the number is not yet a quantity about Formula 1; it is a quantity about a particular estimator. The abstract provides no tests of additivity, no cross-validation of the smoothing parameters, and no comparison against a model with driver-constructor interactions.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","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.","tokens_in":5410,"tokens_out":2717,"duration_ms":27504,"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":[{"comment":"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.","section":"Full text (entire document)"},{"comment":"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.","section":"Abstract, second sentence and results"},{"comment":"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.","section":"Abstract, RAPM methodology mention"}],"minor_comments":[{"comment":"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.","section":"Abstract, references"},{"comment":"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.","section":"General publication integrity"}],"recommendation":"reject","confidential_remarks":"This appears to be a submission-level error: the correct F1 paper is missing entirely. In principle, a resubmission with the actual RAPM text and full methods could be considered, but the current manuscript is not a coherent or reviewable submission. I recommend rejection of this version, with the editor free to invite a fresh submission if the correct manuscript is available."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The main thing you should know: arXiv:2508.00200 currently points to a full text that is not the paper described in the metadata or abstract. The attached manuscript is about robust classification under noisy labels using NNK neighborhoods, by a different set of authors. The F1 RAPM paper is only an abstract, so no methods, equations, validation, or data are available for inspection.\n\nWhat is genuinely new, based on the abstract: applying Regularized Adjusted Plus-Minus, previously used in basketball and hockey, to Formula 1 with time-decay and LOESS smoothing is a real domain extension. The headline number — constructors explain 64.0% of race-outcome variance in the Hybrid Era — is a concrete, falsifiable claim that would be worth examining if the actual analysis were in front of us. The abstract is honest about its lineage (Sill 2010, Jacoby 2000), and the framing as a decomposition rather than a prediction is not misleading on its face.\n\nThe soft spot is load-bearing and not the authors' methodology: the manuscript attached to the abstract is a different paper. That makes the current submission effectively unverdictable. Even if the correct text were uploaded, the key scientific risks would be the additive-separability assumption of RAPM and the sensitivity of the 64.0% split to ridge regularization and LOESS smoothing. Those are real concerns, but they are secondary here because we cannot see whether the paper addresses them. The reader's scores, particularly the low soundness and the \"unverdictable\" status, are appropriate given what is available. I would not infer that the F1 paper is flawed; I would only say that it cannot be assessed in this form.\n\nWho this is for: sports-analytics readers interested in driver valuation or team strategy. The abstract alone could be a seed for discussion, but no one can evaluate the actual contribution until the correct manuscript is submitted.\n\nRecommendation: do not send this to peer review yet. The desk editor should return the paper to the authors to correct the file — the metadata says 2508.00200, but the body is clearly the wrong document. Once the correct text is available, it deserves a serious referee look, because the question is interesting and the methodology is appropriate for the task. As it stands, the mismatch itself is a reason to withhold review.","headline":"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.","tokens_in":5869,"tokens_out":1374,"would_cite":false,"duration_ms":15283,"reading_group":"no","serious_thinker":"unclear","would_accept_peer_review":false},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["62J07"],"pacs":[],"model":"deepseek-v4-flash","headline":"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.","keywords":["Formula 1","RAPM","ridge regression","variance decomposition","driver and constructor effects","Hybrid Engine Era","LOESS smoothing","sports analytics"],"falsifier":"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.","tokens_in":5029,"feed_emoji":"🏎️","tokens_out":5671,"duration_ms":51022,"temperature":0.7,"pith_summary":"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.","feed_headline":"Constructors explain 64% of F1 race variance","feed_subtitle":"A 2014-2024 statistical model separates car and driver impact, sharpening driver comparisons.","key_machinery":"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.","core_discovery":"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.","pith_inferences":["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."],"forward_implications":["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."],"supporting_citations":[{"why":"Supplies the Regularized Adjusted Plus Minus methodology that the paper extends to Formula 1 race outcomes.","marker":"(Sill 2010)"},{"why":"Provides the LOESS smoothing technique used to model how driver and constructor coefficients change over time.","marker":"(Jacoby 2000)"}],"fun_headline_variants":["Car beats driver: 64% of F1 variance from constructors","F1 car matters most: constructors own 64% of outcome variance","64% of F1 race variance tied to constructor, not driver","In F1, car > driver: 64% variance from constructor","Hybrid-era F1: constructor dominates with 64% race variance"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"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.","fun_headline_variants_meta":{"raw":{"variants":["Car beats driver: 64% of F1 variance from constructors","F1 car matters most: constructors own 64% of outcome variance","64% of F1 race variance tied to constructor, not driver","In F1, car > driver: 64% variance from constructor","Hybrid-era F1: constructor dominates with 64% race variance"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000277,"raw_usage":{"total_tokens":1644,"prompt_tokens":930,"completion_tokens":714,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":546,"completion_tokens_details":{"reasoning_tokens":618}},"tokens_in":546,"tokens_out":714,"duration_ms":6152,"temperature":1.0,"reasoning_tokens":618,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T10:17:37.012243+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"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.","supporting_citations":[],"review_version":1}