{"id":"942ad7fd-17b4-4382-9d8d-c7ed76dc863a","arxiv_id":"1908.06403","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A sensor platform and case study show that professional CS:GO players keep their gaze near the screen center and use movement keys more than amateurs, who tend to run and shoot simultaneously.","lead":"This paper presents a sensing platform that collects eye-tracking, heart-rate, keyboard, mouse, and game-telemetry data from eSports players, and demonstrates it on 15 CS:GO players. A generalist might read it to see whether simple gaze and input patterns can distinguish professional from amateur players in competitive gaming.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The gaze analysis rests on an untested zone-mapping assumption and lacks inferential statistics; the pro-vs-amateur difference may be a game-state or clustering artifact.","rationale":"The platform itself is a credible demonstration: it collects synchronized eye-tracking, heart-rate, mouse/keyboard, and demo data, and the data-collection infrastructure is a useful contribution even without strong comparative conclusions. The stress-test concern is specifically about the comparative gaze claim, which is the paper's headline result. That claim requires that a gaze sample's nearest UI-zone center reflects the zone the player actually attends to, and that any group difference in zone occupancy is due to skill rather than to role, map position, or round phase. Neither condition is checked. The paper's own Section V.A states the zone-attribution assumption explicitly and then builds all subsequent PCA and heatmap conclusions on it. With only 15 participants (5 vs 10) and no inferential statistics, the \"clear separation\" in Figure 6 is anecdotal; PCA separation can easily arise from a single outlier or from within-group correlations. This is not an accusation of fraud; it is a request for evidence that the stated assumption holds and that the difference is statistically stable. The reader's weakest_assumption flagged the same zone-mapping risk, so I agree. The verdict should remain CONDITIONAL: accept the platform contribution and the plausibility of the gaze difference, but require the additional validation or statistical support before treating the skill-difference rule as established. Also noting a minor internal inconsistency (30 Hz in Section III.A vs 60 Hz in Section V.A) as an additional reason for care, though not itself decisive.","tokens_in":8200,"tokens_out":5941,"duration_ms":67191,"concrete_test":"Re-run the comparison as a participant-level permutation test on per-participant mean probabilities for Zone 1 and Zone 2, stratifying by team role (terrorist vs counter-terrorist) and round phase, and using raw gaze coordinates rather than nearest-cluster assignment to define \"screen-center\" and \"radar\" regions. If the professional-amateur difference is not significant (p<0.05) after role/phase stratification, or disappears when raw-coordinate regions are used, the headline \"screen-center vs radar\" rule is not established.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central comparative claim depends on converting raw gaze coordinates into occupancy probabilities for nine manually chosen UI-zone clusters (Section V.A, Table 4). This conversion assumes that \"the gaze positions (x,y) are mainly caused by the specific zone of interest shown on the screen during the game\" — an assumption the paper does not validate. Because the nine centers define a fixed Voronoi partition of the 1920x1080 screen, any gaze at a moving enemy, weapon model, or empty world region is assigned to the nearest UI-zone center; e.g., a target near the lower-left corner is counted as \"radar\" gaze, and a target near screen center is counted as \"crosshair\" gaze. If professional and amateur players differ in team role, map position, or round phase, the observed separation in PCA (Figure 6) and the \"screen-center vs radar\" rule could reflect game-state differences rather than skill. No confidence intervals, permutation tests, or effect sizes are reported for the 5-pro vs 10-amateur comparison, and with n=15 eyeballing a PCA projection is not a statistical test. The subsection itself says future work needs specialized machine-learning methods for such data, undercutting the strength of the present claim.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents a sensing platform for collecting heterogeneous data from eSports athletes, integrating an eye tracker, a heart-rate monitor, a mouse/keyboard logger, and CS:GO telemetry, and demonstrates it in a case study with 5 professional and 10 amateur players in the CS:GO Retake mode. The main analytical claims are that professional athletes spend more time looking at the screen center and less time looking at the radar than amateurs (Section V.A, Figures 6 and 7), and that professionals use strafe keys (A/D) more while amateurs combine forward movement (W) with shooting (MOUSE1) (Section V.B, Figures 8 and 9). The platform description and data collection methodology are plausible, but the comparative claims rest on visual inspection of PCA projections and unvalidated gaze-to-UI-zone assignments, with no inferential statistics reported.","tokens_in":8448,"tokens_out":3532,"duration_ms":36035,"significance":"If the platform and the gaze/input differences were rigorously established, this would be a useful contribution to eSports analytics and pervasive sensing: the system enables synchronized multimodal data collection from real professional athletes in a controlled scenario, and the proposed gaze-zone attribution approach could support coach feedback and scouting. The paper also includes a reproducible experimental protocol and a concrete dataset description. However, the analytical results are preliminary: the central comparative claims currently lack statistical support, and the key assumption that gaze coordinates are caused by fixed UI zones is not validated. The platform itself appears functional, so the contribution is primarily the sensing infrastructure and the case-study setup, not yet a demonstrated behavioral rule.","major_comments":[{"comment":"The claim that professional athletes and amateur players are 'clearly' separated by the first two principal components is based entirely on visual inspection of a PCA projection of per-window gaze-zone distributions, with no statistical test, confidence intervals, or effect sizes. With only 5 professionals and 10 amateurs, and with multiple rolling windows per player (a repeated-measures structure), the apparent separation could be driven by individual differences or by non-independence of the windows. Please provide a formal test comparing the distributions of Zone 1 and Zone 2 occupancy between the two groups, for example a mixed-effects model with participant as a random effect, a permutation test that respects subject-level clustering, or a Mann-Whitney U test on per-player averages, and report effect sizes.","section":"Section V.A, Figure 6"},{"comment":"The gaze-to-zone mapping assumes that 'the gaze positions (x,y) are mainly caused by the specific zone of interest shown on the screen during the game,' but this assumption is not validated. The fixed nine cluster centers define a Voronoi-like partition of the 1920x1080 screen, so gaze at moving targets, empty world regions, or the weapon model will be assigned to the nearest UI-zone center (e.g., lower-left gaze is counted as radar, center gaze as crosshair). If professionals and amateurs differ in map position, round phase, or engagement distance, the reported center-versus-radar difference could reflect game-state differences rather than skill. Please validate the zone assignment (e.g., by comparing automated labels to manual annotation of gaze videos) or at minimum analyze the sensitivity of the conclusions to the choice of cluster centers and to the number of clusters k.","section":"Section V.A, Table 4 and assumption before Eq. (1)"},{"comment":"The mouse and keyboard analysis is qualitative: the claims that professionals 'often use' A/D keys and that amateurs 'often combine' W with MOUSE1 are supported only by visual inspection of two kernel-density plots, with no numerical summaries, no statistical tests, and no effect sizes. Please report quantitative features (e.g., fraction of time holding A/D, frequency or duration of W+MOUSE1 co-occurrence) for each group and test the group difference, ideally with the same repeated-measures care as recommended for the gaze data.","section":"Section V.B, Figures 8 and 9"},{"comment":"The conclusion states that 'Analysis of eSports data requires the development of specialized machine learning methods' and lists many open data-processing challenges, which is in tension with the strong comparative conclusions drawn earlier in Section V.A and V.B. Please either soften the claims (e.g., present the gaze and input results as exploratory) or provide the additional statistical evidence needed to support them.","section":"Section VI (Conclusion)"}],"minor_comments":[{"comment":"The sampling rate of the eye tracker is stated as 30 times per second in Section III.A but as 60 Hz in Section V.A; please reconcile these values and verify that the interpolation gap threshold of '6 missing consequent observations maximum' matches the actual sampling rate (0.1 s at 60 Hz, but 0.2 s at 30 Hz).","section":"Section III.A vs Section V.A"},{"comment":"The game scenario in Section IV.B describes 12 rounds, while Section V.A refers to '10 repeated rounds'; please clarify the actual number of rounds used in the analysis.","section":"Section IV.B vs Section V.A"},{"comment":"Zones 4, 6, and 8 are described as 'Right Area of Sight,' 'Left Area of Sight,' and 'Bottom Area of Sight' but are not visible in Figure 3; please explain the rationale for their placement and how they relate to the UI elements or to gameplay.","section":"Section V.A, Table 4"},{"comment":"The caption says 'principle components' but the correct term is 'principal components'; similarly, 'prepossessed' in Section V.A should be 'preprocessed.'","section":"Section V.A, Figure 6"},{"comment":"The axes of the KDE plots in Figures 8 and 9 are not clearly labeled in the figure captions; please specify what quantity is plotted on each axis (e.g., time in seconds, probability density).","section":"Section V.A, Figures 8 and 9"}],"recommendation":"major_revision","confidential_remarks":"The platform description is a reasonable engineering contribution, but the evidence for the behavioral claims is not yet at the level required for the paper's stated conclusions. The authors should either add rigorous statistical analysis and validate the zone-mapping assumption, or reframe the paper as a demonstration of the sensing system with exploratory case-study results. I would lean toward major revision rather than rejection because the infrastructure contribution appears sound and the requested analyses are within the scope of the manuscript."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: the platform is the real contribution; the gaze skill-difference result is plausible but rests on a hand-built zone mapping the paper never validates, and the 5-vs-10 comparison has no inferential statistics. That doesn't sink the paper, but you should treat the comparative claims as preliminary.\n\nWhat's new: a working integration of eye tracker, HRM, key/mouse logging, and CS:GO demo capture into one synchronized stream, demonstrated with a professional team (Monolith) in a fixed Retake scenario. That is genuinely useful for eSports sensing work. The authors document the hardware and preprocessing honestly, and the input-analysis rule (amateurs run-and-shoot, pros use strafe keys) is simple and tied to game mechanics.\n\nWhere it softens: the gaze analysis assigns every observed gaze point to one of nine manually defined UI-zone centers. The paper says \"by confirming this assumption one can then match the gaze position\" but never does the confirmation. Any gaze at a moving enemy or empty space gets snapped to the nearest zone, so the \"center vs radar\" story could partly reflect where players happen to be on the map or what phase the round is in, rather than a stable skill difference. The PCA separation is visually suggestive, but with n=15 and no confidence intervals, permutation tests, or effect sizes, it isn't a demonstrated result. The mouse/keyboard analysis is qualitative only. Also no artifacts are released, which makes the platform harder to reuse.\n\nThe stress-test note is fair on all these points; I don't think it overreaches. The paper doesn't claim more than a case study, and the conclusion explicitly defers to future ML methods, so the authors know the limits. The platform claim holds up; the behavioral differences are the weak spot. The repeated self-citations in the future-work section are a bit noisy but not a flaw for a systems paper.\n\nWho benefits: eSports analytics and HCI people who want an example of end-to-end sensing with a pro team. A serious referee shouldn't reject it; it deserves review, with requests for statistical grounding, validation or a caveated interpretation of the zone mapping, and ideally public data/code.","headline":"The platform is a real, reusable contribution; the pro-vs-amateur gaze finding is plausible but under-supported, and the unvalidated zone-mapping assumption is the main weakness.","tokens_in":8953,"tokens_out":2688,"would_cite":true,"duration_ms":30114,"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 professional and amateur CS:GO players can be separated by two measured behaviors: keeping gaze on the screen center rather than the radar, and avoiding forward movement while shooting.","keywords":["eSports","gaze analysis","eye tracking","Counter-Strike: Global Offensive","player skill assessment","sensing platform","mouse and keyboard input","wearable sensors"],"falsifier":"Run the same gaze pipeline on a round with a scripted enemy moving across the screen center: if eye-tracking samples that follow the moving enemy are assigned to the crosshair zone for both professionals and amateurs, then the claimed center-gaze difference may reflect target tracking rather than crosshair discipline, and the professional-amateur separation would need re-testing with the moving-target samples excluded.","tokens_in":8050,"feed_emoji":"🎮","tokens_out":6523,"duration_ms":61522,"temperature":0.7,"pith_summary":"This paper claims that a sensing platform that synchronizes eye tracking, heart-rate monitoring, mouse and keyboard logging, and in-game demo recording can gather the heterogeneous data needed to study eSports athletes in real matches. In a case study with fifteen Counter-Strike: Global Offensive players, the platform separates five professionals from ten amateurs using gaze alone: professionals keep their gaze at the screen center, near the aiming crosshair, while amateurs look more often at the game radar. The same data show that amateurs frequently press the forward key and the shoot button at the same time, an action the paper notes reduces shooting accuracy in the game engine, while professionals more often use left-right movement keys. If these behavioral differences hold, the platform offers a quantitative basis for scouting and assessing players rather than relying only on game statistics or a coach's experience.","feed_headline":"Pros stare at the crosshair; amateurs watch the radar","feed_subtitle":"Eye-tracking and input logs from a CS:GO sensing platform reveal measurable skill differences.","key_machinery":"The load-bearing object is the zone-occupancy distribution: a vector of nine probabilities, one per screen zone, computed from raw gaze coordinates by k-center clustering with nine hand-selected cluster centers (crosshair, radar, health/armor panel, weapon/ammo panel, kill/log, timer/player panel, and three zones around the crosshair). Gaze gaps shorter than 0.1 seconds are linearly interpolated, and the frequency of gaze in each zone is counted over a 15-second rolling window. Principal component analysis is applied to these distributions; its first two components are nearly sparse, with the dominant coordinate of the first component being the crosshair zone and that of the second being the radar zone, which is what makes the professional-amateur separation interpretable. The supporting input analysis uses two simple features: time spent pressing A or D, and time spent pressing W and MOUSE1 together.","core_discovery":"The central discovery is that professional and amateur CS:GO players can be distinguished by where they look and how they press keys, measured in a controlled Retake-mode scenario. The eye-tracker data are assigned to nine hand-chosen screen zones by k-center clustering, and the fraction of time spent in each zone is computed over 15-second windows; PCA on these zone-occupancy vectors shows a clear split between the two groups. The first principal component is dominated by the aiming-crosshair zone, and the second by the radar zone, leading the paper to conclude that professionals spend more time looking at the screen center and less time looking at the radar than amateurs. The input analysis adds that professionals use the left and right movement keys more often, whereas amateurs more often combine forward movement with shooting, which the game engine penalizes. The paper presents these as simple rules that can explain the skill difference observed in the data.","pith_inferences":["If the crosshair-versus-radar axis is the main separator, a cheaper deployment could use a coarse per-second zone estimator from a low-frequency eye tracker and still reproduce the skill split.","The fixed-zone model treats gaze as caused by static UI areas; mapping gaze onto 3D game-world objects (enemies, moving targets) would test whether center gaze is crosshair fixation or target tracking.","The two behavioral rules (center gaze, no run-and-shoot) could be combined into a single low-dimensional skill score and validated against external performance metrics such as round win rate or kill-death ratio."],"forward_implications":["A scout or coach could use gaze-zone occupancy, especially the crosshair-versus-radar balance, as a quantitative indicator of FPS player experience in similar scenarios.","The W+MOUSE1 simultaneous-press rate gives a second, independent behavioral marker that does not require an eye tracker.","The same sensor stack can be applied to other game disciplines and to larger groups to test whether the center-gaze rule generalizes beyond CS:GO Retake.","Amateur players could be given targeted feedback to reduce simultaneous run-and-shoot and to keep gaze near the crosshair, based directly on the observed professional pattern."],"supporting_citations":[{"why":"Supplies the precedent that raw mouse and keyboard input can predict player skill in a first-person shooter, which the input analysis builds on.","marker":"[8]"},{"why":"Reports that eye-movement patterns in FPS combat differ with playing experience, motivating gaze as a skill signal and the comparative study.","marker":"[11, 12]"},{"why":"Provides the principal component analysis method used to project gaze-zone distributions and identify the crosshair and radar as dominant components.","marker":"[18, 34]"},{"why":"Shows distribution of click intensity over screen zones as a mouse input feature, supporting the zone-based approach to input behavior.","marker":"[19]"}],"fun_headline_variants":["Pro gamers lock eyes on crosshair, amateurs on radar","Esports pros fixate on center, amateurs on radar","Eye-tracking reveals skill: pros aim eyes, amateurs radar","CS:GO pros vs amateurs: gaze and keys tell who's elite","Pros keep eyes on crosshair, amateurs on radar in CS:GO"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The analysis assumes that each raw gaze position is mainly caused by one of the fixed on-screen zones, so gaze at moving enemies, empty space, or off-screen areas gets assigned to the nearest hand-picked cluster center; if that mapping is wrong, the center-versus-radar conclusion could be an artifact.","fun_headline_variants_meta":{"raw":{"variants":["Pro gamers lock eyes on crosshair, amateurs on radar","Esports pros fixate on center, amateurs on radar","Eye-tracking reveals skill: pros aim eyes, amateurs radar","CS:GO pros vs amateurs: gaze and keys tell who's elite","Pros keep eyes on crosshair, amateurs on radar in CS:GO"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000324,"raw_usage":{"total_tokens":1767,"prompt_tokens":844,"completion_tokens":923,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":460,"completion_tokens_details":{"reasoning_tokens":844}},"tokens_in":460,"tokens_out":923,"duration_ms":8032,"temperature":1.0,"reasoning_tokens":844,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T12:45:45.847289+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the same gaze pipeline on a round with a scripted enemy moving across the screen center: if eye-tracking samples that follow the moving enemy are assigned to the crosshair zone for both professionals and amateurs, then the claimed center-gaze difference may reflect target tracking rather than crosshair discipline, and the professional-amateur separation would need re-testing with the moving-target samples excluded.","supporting_citations":[{"cited_title":"Buckley, K","cited_arxiv_id":null,"evidence_quote":"Supplies the precedent that raw mouse and keyboard input can predict player skill in a first-person shooter, which the input analysis builds on."},{"cited_title":"Iden- tifying game players with mouse biometrics","cited_arxiv_id":null,"evidence_quote":"Shows distribution of click intensity over screen zones as a mouse input feature, supporting the zone-based approach to input behavior."}],"review_version":1}