{"id":"11460028-caf8-4c55-a07e-cf95b7837f56","arxiv_id":"2506.13779","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"Ball path curvature measured near release, not early peak curvature, is the stronger negative correlate of NBA free throw percentage.","lead":"Using NBA optical tracking data, this paper fits Bézier curves to 515 in-game free throws and finds that smoother ball paths near the release point are associated with higher free throw shooting percentages. The result gives coaches a measurable target, terminal ball-path smoothness, instead of the conventional emphasis on the early part of the shot.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The ℓ=5 result may be an artifact of the unvalidated release-point rule (max speed + 2 frames); terminal curvature is measured at that endpoint and no sensitivity analysis is reported.","rationale":"To support the central claim, it must be true that the terminal curvature metric measures the actual smoothness of the shooting motion near release rather than the location of an arbitrary cutoff. The paper identifies no ground truth for release and provides no sensitivity analysis, despite its own limitation noting start/end sensitivity. This is more load-bearing than the small sample or multiple testing because those issues weaken precision, whereas an endpoint artifact could reverse the sign or produce the R² gap spuriously. The reader's weakest assumption points to the same mechanism, and I agree. The paper has real strengths: the Bézier fitting is explicit, the WLS model is stated, and the results are consistent with the controlled study by Slegers & Love (2024). But the lack of any perturbation of the release rule leaves the main comparison unsecured. A simple re-analysis with varying offsets and an independent release anchor would settle it. I also note Eq. (3) is typeset in a way that appears to omit the time-weight τ^ℓ described in the prose; this is likely a typo, but it should be corrected because the metric definition is central. The appropriate verdict remains CONDITIONAL pending that sensitivity analysis; if the effect is robust, the paper becomes acceptable.","tokens_in":8323,"tokens_out":6493,"duration_ms":81037,"concrete_test":"Run the full pipeline (Bézier fit, σ(ℓ) for ℓ=3,5, WLS regression) while varying the release-point offset from max-speed frame −2 to +5 frames and using at least one independent release estimator (e.g., last frame with ball near the hand from manual annotation of 50 shots, or peak of vertical acceleration). Report β, p, and R² for each offset/estimator. If the ℓ=5 coefficient remains negative and its R² advantage over max curvature persists across all offsets and estimators, the release heuristic is not the driver; if the effect disappears or changes sign, the central claim is not supported. A secondary check should re-fit with Bézier orders n=6,8,10 to ensure endpoint sensitivity is not an artifact of curve order.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing assumption is the release-point definition in §4.1: release is taken as two frames (0.08 s) after the frame of maximum ball speed, justified only by the assertion that speed is maximized at release. All terminal curvature metrics, especially σ(5), which is dominated by curvature near the endpoint κ(1), depend directly on this choice. If the max-speed frame is noisy, or if its relation to true release differs systematically across players (e.g., high-FT% players have different acceleration profiles), then the comparison is not between like-for-like release mechanics. In addition, if true release occurs at max speed, the +2-frame rule explicitly extends the path into post-release flight, where curvature is very low; any systematic correlation between FT% and the timing of max speed relative to release would load the terminal metric in favor of high-FT% players. The paper's Limitations section concedes that metrics are sensitive to start/end selection and Bézier parameters, but no sensitivity analysis or external validation (e.g., marker-based release times on a subset) is provided. Because the central claim is that terminal curvature explains more between-player variance than max curvature, an artifact of the endpoint rule would invalidate that claim.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper analyzes 515 in-game NBA free throws from 35 players using Second Spectrum optical tracking. The authors fit eighth-order Bézier curves to the sagittal-plane ball path, compute max curvature and time-weighted curvature integrals σ(ℓ) for ℓ=0, 3, and 5, and regress player-level season FT% on each metric via WLS weighted by season free-throw attempts. They find negative associations for all metrics and report that the ℓ=5 terminal integral has the largest R² (0.330 vs 0.222 for max curvature), concluding that coaches should prioritize smoothing the end of the shooting motion. The manuscript is an extension of Slegers & Love (2024) to in-game data.","tokens_in":8506,"tokens_out":8288,"duration_ms":101483,"significance":"The curvature mathematics and the regression setup are standard and clearly described, and the use of an independent outcome (season FT%) rather than a same-sample kinematic proxy is a real strength. If the endpoint-definition concerns are resolved, the paper would provide a useful in-game validation that terminal path smoothness carries more information than peak curvature, with a concrete coaching implication. The authors are also explicit about limitations, including sensitivity of Bézier fits to parameter choices. At present, however, the central claim rests on endpoint and parameter choices that are not shown to be robust, so the strength of the evidence is not yet commensurate with the title's broad conclusion.","major_comments":[{"comment":"The release point is defined as two frames (0.08 s) after the frame of maximum ball speed, and this endpoint determines the terminal portion of the Bézier path used to compute σ(5). The assumption that speed is maximized at release is not validated, and no sensitivity analysis is reported for the two-frame lag. If the true release occurs at or before the max-speed frame, the terminal metric partly describes post-release flight; if the max-speed frame is noisy or its offset from true release correlates with FT%, the reported advantage of σ(5) over κ_max (R²=0.330 vs 0.222) could be an artifact of the endpoint rule. Please report results for lags of 0, 1, 2, and 3 frames, and validate the rule against known release times on a subset of shots if possible.","section":"Section 4.1; Table 2"},{"comment":"The start of the analyzed path is 'the frame in which the ball is furthest from the player's body,' but the data description states that only ball positions are available and explicitly says the authors have no information about the shooter's elbow or hand. The operational definition of 'furthest from the player's body' is therefore not reproducible from the described data. Because the Bézier fit and all curvature metrics depend on the start point, please specify how body position is determined or replace this rule with a purely data-based criterion.","section":"Section 4.1"},{"comment":"As printed, Eq. (3) defines σ(ℓ)=∫₀¹ (ℓ+1)κ(τ)^ℓ dτ, which is a power transform of curvature, not a time-weighted integral. This is inconsistent with the surrounding text ('more weight as time increases') and with Figure 4, which shows the product of a time weight and curvature. If the intended definition is σ(ℓ)=∫₀¹ (ℓ+1)τ^ℓ κ(τ) dτ (or equivalent), correct the formula and confirm that the reported σ(3) and σ(5) values were computed with that definition; as written, the manuscript does not reproduce the central metric.","section":"Equation (3), Section 4.2.3"},{"comment":"The conclusion that terminal curvature 'explains much more of the between-player variance' is based on comparing in-sample R² values across four metrics, with ℓ=5 appearing to be selected after inspecting results. No adjusted R², cross-validation, bootstrap, or interval estimate for R² is provided. With n=35 players, the difference between R²=0.330 and R²=0.222 may be within sampling variability, and the ranking should be shown to be stable (e.g., leave-one-player-out or bootstrap) before drawing a coaching recommendation.","section":"Section 5, Table 2"},{"comment":"Each player-level curvature predictor is an average of only about 15 tracked shots, and the WLS model accounts for uncertainty in the outcome (FT%) but not for measurement error in the predictor. Such errors-in-variables attenuation can affect different curvature metrics differently and can bias the R² ranking in Table 2. Please add a reliability analysis (e.g., split-half or within-player between-shot consistency) or an errors-in-variables sensitivity check.","section":"Section 3, Section 4.3"}],"minor_comments":[{"comment":"Section 4.3 states that the p-value tests the two-sided alternative H_A: β≠0, while the Table 2 caption says the test is one-sided (H_A: β<0). Please make these consistent.","section":"Section 4.3 vs Table 2 caption"},{"comment":"The phrase 'weighted averaged' should be 'weighted average'.","section":"Section 4.2.1"},{"comment":"The choice of Bézier order n=8 is justified only by visual inspection; a brief note on why n=6 (as in Slegers & Love, 2024) was not used, or a sensitivity check, would help the reader assess the robustness of the reported metrics.","section":"Section 4.2.1"}],"recommendation":"major_revision","confidential_remarks":"The paper is within scope for a sports-analytics venue. The self-citation to Slegers & Love (2024) is appropriate given the method builds directly on that work; I see no circularity concern. The main risk is the endpoint-definition artifact, which can be addressed with the requested sensitivity analyses; if those analyses show instability, the authors should soften the central claim accordingly."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Honest take: this paper does something genuinely new. It takes the Slegers-Love curvature framework from controlled jump shots and applies it to in-game NBA free throws, with a clear finding: terminal curvature is negatively associated with season FT%, explaining about 33% of between-player variance at ell=5, while max curvature explains about 22%. That is a real result, and the directionally opposite max-curvature sign relative to the lab study makes it worth reporting. The math is standard, the WLS setup is transparent, and the prose is refreshingly clear.\n\nThe soft spots are real but not fatal. The release-point definition—two frames after max ball speed—is the load-bearing joint. Terminal curvature is measured near that point, and the ell=5 integral is dominated by curvature at the end. If the max-speed frame is noisy or systematically related to FT%, the headline comparison could be an artifact. The Limitations section concedes sensitivity to start/end selection and Bezier order, but no sensitivity analysis is provided. That is the main thing I'd want fixed. The sample is 35 players with about 15 tracked shots each; that's small but fine for an exploratory study. They test four metrics and highlight the best without multiple-comparison adjustment, and they report p-values but not standard errors. Data and code are not available, which makes independent checking harder.\n\nThe circularity concern in the notes does not really land: the curvature metrics are measured from ball tracking and regressed against an independent outcome, season FT%. Self-citation is not a problem here.\n\nIf you decide to review it, send it out. The question is meaningful, the method is mostly sound, and the limitations are addressable. I'd ask for a sensitivity analysis on the release lag and Bezier order, standard errors, and ideally an out-of-sample or cross-validated check before treating the coaching advice as solid. It is not a breakthrough, but it is an honest, useful contribution.","headline":"New in-game curvature result, but the release-point rule is load-bearing and untested.","tokens_in":9068,"tokens_out":2598,"would_cite":false,"duration_ms":30773,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper claims that low curvature near release predicts NBA free-throw percentage, and that coaches should prioritize smoothing the end of the shot rather than its peak-curvature transition.","keywords":["free throw shooting","ball path curvature","Bézier curve fitting","basketball analytics","shooting smoothness","terminal curvature","NBA tracking data","weighted least squares"],"falsifier":"Re-run the weighted least squares regressions with the release point moved one frame earlier and one frame later relative to the maximum-speed frame. If the ℓ = 5 terminal-curvature coefficient stays near –48.7 with p ≈ 0.0003 and R² ≈ 0.33, the endpoint rule is not driving the result; if the effect shrinks, changes sign, or falls below significance, the central claim depends on the chosen release heuristic.","tokens_in":8073,"feed_emoji":"🏀","tokens_out":7299,"duration_ms":80078,"temperature":0.7,"pith_summary":"Using optical ball-tracking data from 515 in-game NBA free throws in the 2023-2024 regular season, this paper asks which part of the shooting motion must be smooth for smoothness to predict accuracy. The authors fit Bézier curves to each ball path, compute curvature along the fitted curve, and regress season free-throw percentage on curvature summaries. Their central finding is that curvature near release—terminal curvature—is negatively associated with free-throw percentage and explains substantially more between-player variance than maximum curvature earlier in the shot. The time-weighted curvature integral with the strongest end-weighting explains 33.0% of the variance in free-throw percentage, versus 22.2% for maximum curvature. If correct, the practical consequence is that coaches should focus on smoothing the end of the shot rather than the initial forward motion.","feed_headline":"Smooth shot endings predict NBA free throw accuracy","feed_subtitle":"Late-path curvature explains 33% of between-player variance in FT%, far more than peak curvature.","key_machinery":"The central object is the time-weighted curvature integral σ(ℓ) = ∫₀¹ (ℓ+1)κ(τ)^ℓ dτ, applied to an eighth-order Bézier curve fitted by least squares to the ball's sagittal-plane trajectory. κ(τ) is the standard planar curvature formula |x' z'' − z' x''| / ((x')² + (z')²)^{3/2}. The weighting factor ℓ shifts emphasis toward the end of the path: ℓ = 0 gives the average curvature, while larger ℓ makes the integral approximate terminal curvature κ(1). Per-player averages of these integrals enter a weighted least squares regression of season FT%, weighted by season free-throw attempts, so players with more attempts influence the fit more.","core_discovery":"The paper's claim, on its own terms, is that the smoothness of the ball's path as it approaches release carries real information about free-throw proficiency in NBA game conditions. In a weighted least squares regression of season FT% on player-average curvature, the time-weighted curvature integral at ℓ = 5 has a coefficient of –48.73 (p = 0.0003) and an R² of 0.330, while max curvature has a coefficient of –0.98 and an R² of 0.222. Both are negative and statistically significant, so higher curvature anywhere is associated with lower accuracy, but the terminal measure is the stronger predictor. This matches the prior controlled study's conclusion that terminal curvature matters most, while differing from it on the direction of max curvature. The authors read these results as evidence that coaches should treat end-of-path smoothness, not the peak-curvature transition, as the primary mechanical target, and that curvature metrics can be derived automatically from in-game optical tracking alone.","pith_inferences":["My inference: the absence of a sensitivity analysis around the 0.08-second release rule is the most direct threat; re-running the regressions with release pulled one or three frames after peak speed would show whether the ℓ = 5 result is stable.","My inference: the monotone increase in R² from ℓ = 0 (0.213) to ℓ = 3 (0.316) to ℓ = 5 (0.330) suggests a dose-response; an independent dataset of jump shots or next-season free throws could test whether weighting the end more heavily always increases explanatory power.","My inference: since release-speed variability was the accuracy proxy in the laboratory study, measuring the spread of estimated release speeds from the same tracking data could reveal whether terminal curvature works through speed control.","My inference: the framework could be turned into a coach-facing diagnostic that tracks a player's ℓ = 5 curvature across games and checks whether a targeted late-shot smoothing drill lowers it before FT% rises."],"forward_implications":["Coaches should target the final phase of the shot: a curvature metric that weights late-path bending explains 33% of between-player variance in FT%, more than any other tested metric.","Peak curvature alone is a weaker lever; its R² is 22.2%, so smoothing only the initial forward transition is less likely to move accuracy.","Shot smoothness can be quantified automatically from in-game optical tracking, without biomechanical sensors or hand-contact video.","The method carries over from controlled jump-shot experiments to elite game conditions, with the open discrepancy in max-curvature direction flagged for future work."],"supporting_citations":[{"why":"Supplies the Bézier-curvature framework and the terminal-curvature hypothesis that this paper adapts to in-game NBA data.","marker":"Slegers & Love (2024)"},{"why":"Defines optimal release conditions that frame why the release phase and its control matter for free-throw accuracy.","marker":"Tran & Silverberg (2008)"},{"why":"Shows elbow and wrist coordination variability increases late in the shot, motivating the emphasis on the end of the ball path.","marker":"Button et al. (2003)"},{"why":"Links last-moment coordination variability and release-speed deviations to missed free throws, supporting terminal curvature as an accuracy signal.","marker":"Mullineaux & Uhl (2010)"},{"why":"Provides evidence that muscle activation timing is individualized and that free-throw speed is not decisive, justifying path-based smoothness metrics.","marker":"Pakosz et al. (2021)"},{"why":"Gives the Bézier-curve definition and polynomial properties used to fit differentiable paths and compute curvature.","marker":"Baydas & Karakas (2019)"},{"why":"Provides the season free-throw percentages used as the outcome variable in the regressions.","marker":"Basketball Reference (2024)"},{"why":"Documents the Second Spectrum optical tracking system that generated the ball trajectory data.","marker":"National Basketball Association (2016)"}],"fun_headline_variants":["Terminal curvature, not peak, predicts NBA free throws","Smooth release beats peak curvature for free throw accuracy","Late shot curvature, not peak, drives free throw success","Terminal curvature explains a third of FT% variance","Coaches: watch end-of-shot curve, not the peak, for FT%"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that release occurs about 0.08 seconds after the ball reaches maximum speed, and therefore that 'terminal curvature' measured near that point is a true property of the shooter's late motion; if that timing rule is wrong or noisy, the stronger association between terminal curvature and free-throw percentage could be an artifact of where the path is cut.","fun_headline_variants_meta":{"raw":{"variants":["Terminal curvature, not peak, predicts NBA free throws","Smooth release beats peak curvature for free throw accuracy","Late shot curvature, not peak, drives free throw success","Terminal curvature explains a third of FT% variance","Coaches: watch end-of-shot curve, not the peak, for FT%"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001257,"raw_usage":{"total_tokens":5117,"prompt_tokens":878,"completion_tokens":4239,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":494,"completion_tokens_details":{"reasoning_tokens":4155}},"tokens_in":494,"tokens_out":4239,"duration_ms":33894,"temperature":1.0,"reasoning_tokens":4155,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T05:16:49.342742+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-run the weighted least squares regressions with the release point moved one frame earlier and one frame later relative to the maximum-speed frame. If the ℓ = 5 terminal-curvature coefficient stays near –48.7 with p ≈ 0.0003 and R² ≈ 0.33, the endpoint rule is not driving the result; if the effect shrinks, changes sign, or falls below significance, the central claim depends on the chosen release heuristic.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the Bézier-curvature framework and the terminal-curvature hypothesis that this paper adapts to in-game NBA data."},{"cited_title":"M., & Silverberg, L","cited_arxiv_id":null,"evidence_quote":"Defines optimal release conditions that frame why the release phase and its control matter for free-throw accuracy."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Shows elbow and wrist coordination variability increases late in the shot, motivating the emphasis on the end of the ball path."},{"cited_title":"R., & Uhl, T","cited_arxiv_id":null,"evidence_quote":"Links last-moment coordination variability and release-speed deviations to missed free throws, supporting terminal curvature as an accuracy signal."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides evidence that muscle activation timing is individualized and that free-throw speed is not decisive, justifying path-based smoothness metrics."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the Bézier-curve definition and polynomial properties used to fit differentiable paths and compute curvature."},{"cited_title":"2023-24 NBA Player Stats : Totals","cited_arxiv_id":null,"evidence_quote":"Provides the season free-throw percentages used as the outcome variable in the regressions."},{"cited_title":"NBA announces multiyear partnership with Sportradar and Second Spectrum","cited_arxiv_id":null,"evidence_quote":"Documents the Second Spectrum optical tracking system that generated the ball trajectory data."}],"review_version":1}