{"id":"a96a0b2a-cbb2-47b9-927c-77e78db439ff","arxiv_id":"2507.21000","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":3.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":3,"one_line_summary":"A design proposal for a modular real-time eye tracking pipeline in the XR Space Framework to enable Dynamic Difficulty Adjustment, with no empirical validation.","lead":"This paper proposes an eye tracking module for the XR Space Framework to adjust task difficulty in virtual reality based on gaze and pupil data. It is a design proposal without experiments, so the promised human performance benefits are not yet demonstrated.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Fixed velocity thresholds in §3.2 are unlikely to separate saccades from smooth pursuit and VOR in dynamic HMD scenes, so the DDA metrics rest on an unvalidated assumption.","rationale":"The reader's weakest assumption already identified the fixed velocity thresholds as the key risk, and my analysis agrees with that assessment. The concern is load-bearing because Equations 1–3 and the threshold rules in Section 3.2 are the only mechanism connecting raw eye tracking to the saccade/fixation/pupil metrics that drive Dynamic Difficulty Adjustment. If those thresholds misclassify events under natural head movement or smooth pursuit, the inferred attention and cognitive-load states are unreliable, and the proposed biofeedback loop cannot deliver the claimed performance enhancement. The paper is explicitly a proposal, so the absence of an implementation or user study is not itself an inconsistency, but it does mean the central claim is unverified. I would keep the reader's CONDITIONAL verdict rather than moving to ACCEPT or REJECT, because the architecture is plausible and the problem is clearly scoped, but the empirical question of threshold validity must be resolved before the proposal can be considered effective. The proposed concrete test is a small, scripted validation that would directly settle whether the threshold assumption holds in the target environment.","tokens_in":5439,"tokens_out":3497,"duration_ms":42936,"concrete_test":"Record HMD eye-tracking data in a scripted VR scenario containing four segments: (1) fixation on a stationary target with the head still, (2) fixation while rotating the head, (3) smooth pursuit of a moving target, and (4) voluntary saccades, with ground-truth event labels from the script. Run the §3.2 velocity-threshold detector on each segment and compute precision/recall per event type. If head-fixation segments are labeled as saccades or pursuit segments are not labeled as fixations, the fixed thresholds are invalid for dynamic XR and the DDA module needs head/pursuit compensation before the central claim can be supported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The DDA pipeline's event classification (Equations 1–3, §3.2) labels samples as saccades above 250 deg/s and fixations below 3 deg/s. In an HMD, the gaze direction vector is typically expressed in head coordinates. During natural head movement while fixating a world-locked target, the vestibulo-ocular reflex rotates the eye opposite the head, so eye-in-head angular velocity can exceed 3 deg/s and be misclassified as a saccade. Conversely, smooth pursuit of moving objects in dynamic XR scenes produces sustained velocities well above the fixation threshold, and high-speed pursuit can overlap saccade velocities, so the same rule will misclassify pursuit. The paper itself flags 'complex moving visuals' as a challenge, but no head-movement compensation, pursuit handling, or adaptive threshold is described. Since saccade/fixation labels and the inferred attention/cognitive-load metrics are the sole inputs to Dynamic Difficulty Adjustment, the central claim depends on this unvalidated threshold assumption. Additionally, Section 3.4 repeatedly refers to DDA 'in the future adaptive environments', and no implementation, data, or user study is provided, so the enhancement claim is currently a design proposal rather than an established result.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes an eye tracking module for the XR Space Framework, intended to provide real-time gaze-based metrics (saccades, fixations, pupil diameter) that feed into Dynamic Difficulty Adjustment (DDA) for XR applications in training, screening, and teleoperation. It describes a workflow spanning calibration, data acquisition, preprocessing, processing, and DDA, and it uses velocity-threshold identification to classify eye movements. The paper includes equations for angular velocity and event classification, a high-level architecture based on abstract classes, and a discussion of threading efficiency. No implementation, empirical data, user study, or real-time performance measurements are presented; the text explicitly places DDA in \"the future adaptive environments.\"","tokens_in":5675,"tokens_out":4112,"duration_ms":49872,"significance":"If fully realized and validated, the proposed module could make a modest but useful contribution by standardizing real-time integration of eye tracking biofeedback into XR frameworks, potentially benefiting adaptive training and teleoperation systems. The paper's strengths are its clear workflow description, use of established velocity-threshold algorithms with citations, and explicit acknowledgment of current limitations. However, the central claim of enhancing human performance is not demonstrated: there is no evidence that the pipeline runs in real time, that the event classifications are accurate in dynamic HMD scenes, or that the inferred attention/cognitive-load metrics lead to correct DDA decisions. The significance is therefore prospective rather than established.","major_comments":[{"comment":"The fixed velocity thresholds (saccade above 250 deg/s, fixation below 3 deg/s) are taken from classical eye tracking literature and are applied without modification to HMD-based XR. In dynamic immersive scenes, the vestibulo-ocular reflex rotates the eye relative to the head during natural head movement, and smooth pursuit of moving objects produces sustained eye-in-head velocities that can exceed the fixation threshold and even overlap saccade velocities. The proposed module has no head-movement compensation, no pursuit handling, and no adaptive threshold mechanism, so the saccade/fixation labels and the derived attention/cognitive-load metrics may be unreliable. Because these metrics are the sole inputs to DDA, this unvalidated threshold assumption is load-bearing for the paper's central claim. The authors should describe coordinate-system handling, add support for smooth pursuit and VOR, and provide empirical validation in the target dynamic scenarios.","section":"3.2, Equations (1)-(3)"},{"comment":"The DDA stage is described only as a future component: the text states that the metrics \"can be used for Dynamic Difficulty Adjustment in the future adaptive environments.\" No implementation details, latency measurements, frame-rate impact data, or user study are reported. Consequently, there is no evidence that the pipeline runs in real time or that the proposed metrics lead to correct difficulty adjustments. To support the title's \"Effective Human Performance\" claim, the authors would need at least a proof-of-concept implementation with a pilot evaluation, even offline analysis of recorded sessions or a small within-subjects study comparing DDA against fixed-difficulty conditions.","section":"3.4, Figure 2"}],"minor_comments":[{"comment":"The heading \"Data Aquisition\" contains a typo and should read \"Data Acquisition.\"","section":"2.1"},{"comment":"The footnote \"Normalize vector has a length 1\" is grammatically incorrect; it should be \"A normalized vector has length 1.\"","section":"3.2"},{"comment":"The classification rule divides samples into saccades (above 250 deg/s) and fixations (below 3 deg/s), but leaves the range 3-250 deg/s unclassified; the paper should state how samples in this intermediate range are labeled in the processing pipeline.","section":"3.2"},{"comment":"The table lists \"Time to first fixation (TTFF)\" as a metric, but TTFF does not appear anywhere in the proposed workflow; either remove it or explain how it is integrated into the module.","section":"3.1, Table 1"},{"comment":"The claim that \"developers can easily extend these abstract classes\" is not substantiated; providing a minimal interface sketch or a brief example would make the universality claim more concrete.","section":"3.3"},{"comment":"Reference 1 (\"Beck, L.A.\") is incomplete and confusingly attributed; the work is Csikszentmihalyi's Flow, and the reference list entry should be corrected to the original source with full bibliographic details.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"This manuscript reads more as a position paper or an architecture sketch than as a completed research contribution. The central gap—no validation of the proposed DDA pipeline—is substantial but addressable if the authors add an implementation and at least a pilot study. Editors may also wish to consider whether the current level of empirical support matches the journal's typical expectations; if not, rejection might be considered. I recommend major_revision because the core idea is defensible and the missing evidence could be supplied within a revised manuscript."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Karpowicz et al. describes an architectural proposal for integrating real-time eye tracking into their XR Space Framework to drive Dynamic Difficulty Adjustment. It is not a validated system; there is no running implementation, no user study, and no benchmark. The paper's own language is careful, repeatedly saying DDA is for \"future adaptive environments,\" so it does not oversell. What it does well: it gives a clean, readable account of the pipeline (calibration, buffering, preprocessing, event detection, pupillometry, DDA), grounds the metrics in standard literature, and positions the work inside an existing framework with a plausible \"universal approach\" via abstract classes and interfaces. The equations are standard and correct. That is real value as a design document.\n\nThe soft spots are equally clear. The biggest is the one your stress-test flags: the velocity-threshold classification from Section 3.2 assumes static fixed thresholds for saccades (250 deg/s) and fixations (3 deg/s), but in an HMD the eye-in-head signal is contaminated by the vestibulo-ocular reflex during head movement, and smooth pursuit of moving objects easily produces sustained velocities above 3 deg/s without being saccades. Since the inferred attention and cognitive-load metrics depend entirely on this classification, and the paper offers no head-movement compensation or adaptive thresholding, the load-bearing input to the proposed DDA loop is unvalidated in exactly the conditions XR applications create. The paper acknowledges \"complex moving visuals\" as a challenge but does not address it. That matters, but it is a missing validation issue, not an internal contradiction, because the paper positions itself as a proposal.\n\nOther issues are minor: the \"universal\" claim is only supported by one SDK implementation and a mention of OpenXR; the multi-threading discussion is unsurprising; and the self-citations to their own prior systems are appropriate for context, not load-bearing. The paper would benefit from at least a proof-of-concept with real data showing the thresholds behave acceptably in head-mounted, moving scenes.\n\nThis deserves serious peer review as a design/position paper for a venue that accepts such contributions. An editor should not desk-reject it, but a referee should require either a working prototype with event-classification validation or an explicit narrowing of the claim. I would not cite it as a result, but I would mention it as a roadmap. Worth a reading group if your group works on adaptive XR or physiological sensing.","headline":"A clearly written design sketch for an XR eye-tracking DDA pipeline that is honest about its own preliminary status, but its central performance claim currently rests on an unvalidated event-classification threshold.","tokens_in":6214,"tokens_out":1453,"would_cite":false,"duration_ms":20832,"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 proposes a real-time eye-tracking module that would let XR systems adjust task difficulty to the user's gaze-based state.","keywords":["Immersive Systems","Virtual Reality","eXtended Reality","Human Performance","Psychophysiology","Eye tracking","Dynamic Difficulty Adjustment","Biofeedback"],"falsifier":"Run a controlled training scenario in a head-mounted display comparing task performance and self-reported engagement with dynamic difficulty adjustment driven by this eye-tracking module against a fixed-difficulty control; a second, more direct test is to record gaze in a moving three-dimensional scene with ground-truth saccade and fixation labels from a high-speed reference tracker and measure whether the velocity-threshold classifier keeps its accuracy when the scene moves.","tokens_in":5285,"feed_emoji":"👁️","tokens_out":7829,"duration_ms":79618,"temperature":0.7,"pith_summary":"The paper proposes an eye-tracking module for the XR Space Framework that would let immersive applications read attention, cognitive load, and engagement in real time from gaze data, and act on that state. The intended payoff is dynamic difficulty adjustment in XR-based training, screening, and teleoperation, keeping users in a flow state rather than bored or overwhelmed. The proposal covers a hardware-agnostic pipeline that detects saccades and fixations with velocity thresholds, monitors pupil diameter as an indicator of cognitive load, and runs the heavy data stages off the render thread. If the module performs as described, eye tracking would shift from a post-hoc measurement tool to a live control input for adaptive immersive systems.","feed_headline":"Eye tracking to auto-adjust XR task difficulty in real time","feed_subtitle":"Gaze metrics would let training, screening, and teleoperation adapt to user state in real time.","key_machinery":"The central mechanism is velocity-threshold identification, a method that computes the angular velocity of gaze from the angle between successive normalized gaze-direction vectors (Equations 1–3) and labels each sample a saccade above 250 deg/s or a fixation below 3 deg/s; pupil-diameter changes are tracked alongside as a cognitive-load signal. This event classification is what feeds the dynamic difficulty adjustment. Equally load-bearing is the module's abstraction layer, which defines abstract classes and interfaces so that the same pipeline can consume eye-tracking data from various HMDs and SDKs, with pupil-based metrics enabled only when the hardware provides them.","core_discovery":"The paper's central claim is that a modular eye-tracking pipeline, integrated with a real-time XR engine, can supply continuous, low-latency indicators of attention and cognitive load that are suitable for driving dynamic difficulty adjustment in immersive training, screening, and teleoperation. The proposed module classifies saccades and fixations from gaze-direction angular velocity using threshold rules, monitors pupil diameter as a cognitive-load proxy, and packages the processing in an abstraction layer that accepts data from different eye-tracking SDKs. The workflow keeps calibration, acquisition, preprocessing, processing, and adaptation as separate stages, with data-heavy stages running off the main render thread so that real-time performance is preserved.","pith_inferences":["The fixed velocity thresholds (250 deg/s and 3 deg/s) are taken from the eye-tracking literature and the paper does not validate them against dynamic head-mounted scenes with moving visuals; an in-situ calibration or trained classifier would be a natural extension before relying on the metrics for difficulty control.","The paper leaves the mapping from gaze metrics to difficulty change qualitative; a concrete scoring rule linking dwell time, saccade velocity, and pupil dilation to difficulty increments would make the biofeedback loop directly testable.","Pupil diameter is affected by scene luminance changes inside a headset, so combining it with the framework's ECG, GSR, and PPG channels could separate cognitive-load effects from light-induced constriction.","A benchmark comparing real-time event detection latency and accuracy against offline analysis of the same logged eye data would show whether the processing pipeline actually keeps pace with the sampling rate."],"forward_implications":["XR training systems could raise or lower task difficulty mid-session, guided by whether fixations land on task-relevant objects and whether gaze is scanning too quickly.","Screening applications could use dwell time and missed fixations to flag when a user needs more guidance before the difficulty level is increased.","Teleoperation setups could adapt display and control complexity in response to the operator's real-time cognitive load inferred from pupil and gaze metrics.","Because the module is built on abstract classes and interfaces, the same analysis pipeline could be reused with different head-mounted displays and eye-tracking SDKs without redesigning the logic.","Separating acquisition and processing into dedicated threads keeps the adaptation from degrading frame rate in the real-time graphics engine."],"supporting_citations":[{"why":"Provides the flow-theory rationale for keeping task difficulty matched to user competence, the basis for dynamic difficulty adjustment.","marker":"[1]"},{"why":"Demonstrates physiological signals being used to adapt game difficulty, the precedent the proposed module extends.","marker":"[2]"},{"why":"Supplies the velocity-threshold event-detection approach for gaze data in virtual reality, including calibration and vergence considerations.","marker":"[3]"},{"why":"Validates saccade detection thresholds with head-mounted display VR technology.","marker":"[4]"},{"why":"Describes the authors' prior XR system with 5-point eye-tracking calibration that the module builds on.","marker":"[5]"},{"why":"Shows the XR framework applied to human factors in space exploration, supporting the claimed application domains.","marker":"[7]"},{"why":"Shows physiological-based difficulty assessment in VR rehabilitation games, supporting the feasibility of real-time difficulty adaptation.","marker":"[12]"}],"fun_headline_variants":["Eye tracking biofeedback for adaptive XR training","Real-time gaze metrics adjust XR task difficulty","Gaze-driven adaptive XR for training and teleoperation","Pupil and gaze data tune XR workload in real time"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the fixed saccade and fixation velocity thresholds of 250 deg/s and 3 deg/s, taken from eye-tracking literature, remain valid inside dynamic head-mounted XR scenes where head movement and moving visuals change gaze patterns; if the thresholds misclassify events there, the inferred attention and cognitive-load metrics, and the difficulty adjustments built on them, become unreliable.","fun_headline_variants_meta":{"raw":{"variants":["Eye tracking biofeedback for adaptive XR training","Real-time gaze metrics adjust XR task difficulty","Gaze-driven adaptive XR for training and teleoperation","Pupil and gaze data tune XR workload in real time"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000173,"raw_usage":{"total_tokens":1260,"prompt_tokens":905,"completion_tokens":355,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":521,"completion_tokens_details":{"reasoning_tokens":290}},"tokens_in":521,"tokens_out":355,"duration_ms":4206,"temperature":1.0,"reasoning_tokens":290,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T13:02:59.591284+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run a controlled training scenario in a head-mounted display comparing task performance and self-reported engagement with dynamic difficulty adjustment driven by this eye-tracking module against a fixed-difficulty control; a second, more direct test is to record gaze in a moving three-dimensional scene with ground-truth saccade and fixation labels from a high-speed reference tracker and measure whether the velocity-threshold classifier keeps its accuracy when the scene moves.","supporting_citations":[{"cited_title":"flow: the psychology of optimal expe- rience (1992)","cited_arxiv_id":null,"evidence_quote":"Provides the flow-theory rationale for keeping task difficulty matched to user competence, the basis for dynamic difficulty adjustment."},{"cited_title":"Systems, Man and Cybernetics, Part A: Systems and Humans, IEEE Transactions on41, 1052 – 1063 (12 2011)","cited_arxiv_id":null,"evidence_quote":"Demonstrates physiological signals being used to adapt game difficulty, the precedent the proposed module extends."},{"cited_title":"Frontiers in Psychiatry11(2020)","cited_arxiv_id":null,"evidence_quote":"Validates saccade detection thresholds with head-mounted display VR technology."},{"cited_title":"In: Digital Interaction and Machine Intelligence","cited_arxiv_id":null,"evidence_quote":"Describes the authors' prior XR system with 5-point eye-tracking calibration that the module builds on."},{"cited_title":"In: Machine Intelligence and Digital Interaction Conference","cited_arxiv_id":null,"evidence_quote":"Shows the XR framework applied to human factors in space exploration, supporting the claimed application domains."},{"cited_title":"In: Proceedings of FDG 2023","cited_arxiv_id":null,"evidence_quote":"Shows physiological-based difficulty assessment in VR rehabilitation games, supporting the feasibility of real-time difficulty adaptation."}],"review_version":1}