Trail Making Tests in Extended Reality: A Critical Review with Implications for Cognitive Ageing
Panagiotis Kourtesis
Abstract
Extended reality (XR) versions of the Trail Making Test and Color Trails Test are promising for assessing and training cognition in older adults. However, each XR adaptation is a new instrument, whose display, interaction technique, timing rules and layouts may change what it measures. This critical narrative review maps these implementations by software lineage and appraises their evidence for assessment and training against common criteria. The Trail Making Test in Virtual Reality (TMT-VR), whose source code was inspected, serves as a measurement case. XR trail-making tests showed convergence with paper and computerised versions, varying by test part and sample, and clinical group differences (e.g., Parkinson's disease, mild cognitive impairment). However, gaps in reliability (e.g., randomised forms), validation (e.g., norms, prospective samples) and reporting (e.g., error handling, build identity) emerged, and no training benefit attributable to the trail-making component was identified. The TMT-VR case shows how interaction technique, timing rule, error handling and unlogged layouts shape a score's meaning. Future tests should log layouts and events, estimate alternate-form reliability and undergo prospective validation, and training trials should isolate the trail-making component. Several XR trail-making tests have moved beyond feasibility, and reported tolerability during supervised sessions supports further evaluation in older adults.
Review reports
xPeerd (AI review)
AI Review (xPeerd — DAReview)
Summary
This manuscript is a critical narrative review with a structured evidence map on extended-reality trail-making tests, aiming to determine what the current XR literature supports for cognitive assessment and training, to use TMT-VR as a measurement case, and to specify what XR trail tests must demonstrate before clinical or training claims are justified. The review is conceptually timely, technically informed, and stronger than many descriptive XR reviews because it distinguishes lineages, implementations, configurations, and studies, and because it explicitly frames design choices such as layout, depth, confirmation rule, and error handling as measurement-relevant variables rather than mere interface details. The manuscript also shows unusual attention to source inspection and technical provenance.
Its main strengths are conceptual clarity, a useful comparative inventory, and a disciplined insistence that immersive adaptations alter the construct being measured. Its main weaknesses are methodological transparency, reproducibility of the literature search, conflict-of-interest exposure in the TMT-VR case focus, incomplete release of some verification outputs, and occasional overreach from heterogeneous and partly inaccessible evidence. The manuscript is broadly coherent and scientifically grounded, but several claims would benefit from tighter delimitation, stronger separation between verified evidence and interpretive inference, and fuller procedural transparency before publication.
Potential Major Revisions
- Clarify and narrow the evidentiary status of the review design
The manuscript states, “This is a critical narrative review with a structured evidence map… It is not a systematic or scoping review” in Section 2. That is acceptable in principle, but the manuscript also makes strong field-level conclusions and prescriptive recommendations that resemble outputs commonly expected from more reproducible review designs. This creates tension between the declared method and the inferential weight of the conclusions.
A major revision should more sharply distinguish:
- what is established by formal, reproducible evidence extraction,
- what is authorial synthesis,
- what is technical inspection of one focal implementation,
- and what remains provisional because of incomplete source access.
This matters because the manuscript also admits that “The query strings and per-database counts were not logged, so the search cannot be reproduced as a database protocol” and that “one author screened and charted the records, without duplicate screening or logged queries.” These are not minor limitations; they materially constrain confidence in claims about completeness, balance, and absence of evidence. The discussion and conclusions should therefore be rewritten to avoid implying exhaustive coverage where the method does not support that implication.
- Strengthen reproducibility of the review process
The review’s own standards for XR measurement validity are demanding, but the manuscript’s evidence-gathering process is less reproducible than the standards it applies to others. The sentence “The query strings and per-database counts were not logged” is especially important. Since the manuscript aims to influence how the field evaluates XR neuropsychological tools, the review process itself should be auditable at a higher standard.
A major revision should add, either in the manuscript or supplement:
- retrospective reconstruction of database queries where possible,
- explicit database list and search dates in one consolidated table,
- screening rules translated into operational inclusion/exclusion criteria,
- counts by source type and access level,
- and a clearer flow from candidate records to included evidence families.
Without this, the paper risks undercutting its own argument about auditability and reporting completeness.
- Address conflict-of-interest and self-referential risk more directly
The manuscript openly states, “The author developed the TMT-VR and is an author of the studies of it that are synthesised in Section 4.” This disclosure is appropriate, but the methodological consequences are not sufficiently handled in the main critical framing. The manuscript’s second stated purpose is “to use the Trail Making Test in Virtual Reality (TMT-VR), whose source code and study records were available for inspection, as a measurement case.” That creates a real asymmetry: one lineage receives privileged inspection unavailable for many comparators.
This does not invalidate the manuscript, but it does require stricter safeguards against favorable imbalance. The paper should explicitly separate:
- claims based on published cross-lineage evidence,
- claims based on inspected TMT-VR source/build artifacts,
- and claims based on unpublished or nonreleased internal verification outputs.
That last issue is especially salient because the manuscript concedes: “The generated layout samples, the candidate form bank and the verification outputs described in Section 4.3 have not been released.” If these materials support substantive comparative or design conclusions, readers need access or the claims need to be softened.
- Recalibrate claims that depend on inaccessible or partial sources
The review includes several implementations known only from abstracts, indexed records, titles, or inaccessible supplements. For example, Table S3.3 identifies records “not read in full,” including technical reports and abstract-level records; one cited study is “known from its abstract and indexed records only, so the display, layout, statistics and any paper comparator could not be checked.”
This is handled responsibly in parts of the supplement, but the main text should more visibly distinguish:
- full-text verified evidence,
- abstract-only evidence,
- source-code-inspected evidence,
- and unverified claims retained only for mapping.
At present, some narrative synthesis may appear more uniformly grounded than it is. A major revision should standardize evidentiary labels in the main text or tables so readers can see which conclusions rest on checkable full reports and which do not.
- Tighten causal or quasi-causal language about design features and score interpretation
A strong conceptual contribution of the paper is the claim that XR changes trail-making at several measurement-relevant points. The manuscript writes that geometry, confirmation rules, and layout variation alter scores, and that “a rule that stops a wrong selection from advancing the trail does not stop the selection itself, so completion times and error counts are not comparable across implementations unless each states whether wrong inputs are registered, counted, corrected or ignored.”
This is plausible and often persuasive, but the wording sometimes shifts from “candidate difficulty factor” or “bearing on interpretation” toward stronger functional conclusions than the comparative evidence can always sustain. The manuscript should be more explicit when a point is:
- theoretically compelled,
- empirically demonstrated,
- or inferred from design analysis alone.
That distinction is central to methodological rigor. Otherwise, readers may mistake well-reasoned construct arguments for empirically established effect sizes or validated score consequences.
- Better justify the training-related conclusions
The manuscript includes both assessment and training uses in eligibility criteria. However, the retrieved evidence on training appears sparse, heterogeneous, and often package-level rather than task-specific. The supplement notes, for example, “near transfer within the trail-making family; broader outcomes and follow-up not combined with a comparator and a defined dose,” and for GAME2AWE, “cognitive outcomes exploratory.”
The training-related conclusions should therefore be restrained and clearly secondary unless the manuscript can demonstrate a stronger task-level causal basis. As currently framed, the review is much stronger on measurement and validation than on intervention efficacy.
- Discuss possible paper-mill or fabrication risk signals, while noting none are decisive
I do not see strong evidence of fabrication in the manuscript itself. On the contrary, the paper often foregrounds inconsistencies in the source literature rather than hiding them. For example, it notes discrepancies such as “One discussion sentence states 57, whereas the abstract, Methods and figure support 56,” and “The correlation statistic is labelled Spearman in the Results and Pearson in the abstract and figure legends.”
These observations are a strength of the review, not a fraud signal in this manuscript. Still, because the paper adopts a highly critical tone toward evidence quality, it should briefly explain how inconsistencies were weighted and when they were treated as reporting error versus reasons to discount substantive findings. That would improve coherence and help avoid impressionistic appraisal.
Anti-fraud warning: No clear paper-mill pattern is evident in the manuscript. The work fits a recognizable scientific paradigm in neuropsychology/HCI review scholarship. The main integrity risk is not fabrication but asymmetric evidentiary access, because one focal lineage is author-developed and technically inspected beyond what was possible for many comparators.
Potential Minor Revisions
- AI-content analysis
The acknowledgments state: “During the preparation of this work the author used Anthropic Claude … and OpenAI Codex … These tools assisted with organising the literature records and extracting and verifying their data, coding the layout emulation and the verification scripts, language refinement, and manuscript drafting and editing.”
Estimated percentage of AI-generated or AI-assisted prose/content: 20–35%.
Highlighted AI-detected sections:
- The acknowledgments explicitly disclose AI assistance in “language refinement” and “manuscript drafting and editing.”
- Stylistically, dense, highly regular sentence patterns in methods/limitations prose may reflect strong editorial AI assistance, though that is not itself problematic.
- No obvious hallucinated bibliography pattern is evident from the retrieved excerpts, and the manuscript appears to maintain domain coherence.
Assessed epistemic impact:
- Low to moderate.
- The disclosed AI use does not, by itself, undermine the manuscript’s contribution.
- The greater epistemic concern is whether AI-assisted extraction/verification introduced unnoticed classification or summarization errors in the evidence map.
- The manuscript would benefit from one sentence clarifying how AI-assisted record organization and extraction were quality-checked against source documents.
- Wording and style refinements
- “This review asks what the evidence on XR trail-making tests supports…” is clear, but the three aims that follow could be converted into numbered aims for readability.
- “Statements that a property was not found refer to the sources identified by this review” is careful, but could be shortened for readability and moved closer to the limitations caveat.
- Some supplementary notes are stronger and clearer than corresponding main-text synthesis; consider relocating one or two key cautionary examples into the main Results or Discussion.
- Typographic, consistency, and reporting issues
- Numerical/sample consistency should be checked carefully where the manuscript reports sample sizes or summarized lineage totals, especially because the review criticizes inconsistencies in the underlying literature. For example, the supplement flags source-level discrepancies such as 56 versus 57 participants in one study.
- Ensure consistent naming of statistical coefficients where source literature is being summarized. The supplement notes a source where a correlation is labelled Spearman in one place and Pearson elsewhere; the manuscript should ensure its own descriptors do not unintentionally propagate source inconsistency.
- Check consistency between “AR/MR,” “augmented or mixed reality,” and “AR trail-making” terminology across tables and prose.
- The manuscript should verify punctuation and spacing around statistical notation and Greek symbols in all tables, especially where excerpts show formatting density around rho values and alpha thresholds.
- If pages are not formally numbered in some supplements, page counting should be standardized in citations to avoid ambiguity for readers and reviewers.
- Formatting issues
- The manuscript relies heavily on supplementary materials for core transparency. Consider moving one consolidated summary table of evidence status into the main manuscript. Table 4 appears useful for this function.
- Several long tables are information-rich but visually dense. Improving row banding, indentation, or grouping by evidence type may improve readability.
- A visual legend distinguishing full-text, abstract-level, title-level, and source-inspected evidence would materially improve interpretation.
Recommendations
-
Reframe the manuscript as a high-value critical mapping review rather than a quasi-systematic definitive field synthesis unless the search record can be reconstructed more fully. This would align the claims with the actual method.
-
Add a transparent evidence-status key across all main findings: full text verified, abstract-level only, technical artifact inspected, or unavailable/inaccessible. This would substantially improve epistemic clarity.
-
Separate more rigorously the manuscript’s three layers of argument:
- field-level synthesis,
- TMT-VR technical case analysis,
- and normative recommendations for future XR test validation.
That separation would reduce perceived self-advocacy and strengthen neutrality.
-
Release, if possible, the “generated layout samples, the candidate form bank and the verification outputs described in Section 4.3,” or else reduce dependence on them in the argument.
-
Add one concise subsection on bias control, explicitly addressing how the author managed interpretation risks arising from being both reviewer and developer of one focal system.
-
Moderate all training-effect conclusions unless stronger intervention-specific evidence is added. The paper is currently most convincing on construct validity and comparability problems, not on rehabilitation efficacy.
-
Preserve the manuscript’s strongest contribution: its argument that XR implementation details are not cosmetic but psychometrically constitutive. That idea is important, persuasive, and worth emphasizing even more clearly in the abstract and conclusion.
Reviewer3 (AI review)
AI Review (reviewer3.com)
Overview
Appraisal of XR Trail-Making Implementations
Contribution
This review provides a structured synthesis and comparative appraisal of extended reality adaptations of trail-making tests across software lineages, highlighting critical gaps in normative data, alternate-form reliability, and prospective clinical validation.
Identification of Technical Determinants of Cognitive Scores
Contribution
Through a source-code inspection of the TMT-VR framework, the study demonstrates how interaction modalities, confirmation rules, error-handling logic, and unlogged layout variations introduce substantial unmeasured variance into extended reality cognitive scores.
Comments (8)
1. Inconsistent count of reports printing a build identifier for TMT-VR — Minor
two reports print Unity 2022.3 LTS with the XR Interaction Toolkit and OpenXR without a build identifier; the lifespan study prints FOCUS (HEAD), 13 March 2026, Unity 6000.2.10f1; the MCI study states one task version for all participants without a build identifier; head-gaze with 0.5-s dwell; seated; Greek audio
Table 4 (p. 10) and Supplementary Table S7.1 (p. 33, 34) state for TMT-VR (9 reports) that a build identifier is reported "in three reports". This matches Table S4.1 (p. 30), where the two VIVE Pro Eye PC reports print a build identifier ("build identifier printed in both reports") and one VIVE Focus Vision standalone report prints one ("the lifespan study names the examined build; the other three reports do not print one"), totaling 2 + 1 = 3 reports. However, in Table S4.1 under "Paper-reported configuration" for the VIVE Focus Vision standalone (4 reports), it says: "two reports print Unity 2022.3 LTS with the XR Interaction Toolkit and OpenXR without a build identifier; the lifespan study prints FOCUS (HEAD), 13 March 2026, Unity 6000.2.10f1; the MCI study states one task version for all participants without a build identifier". While the counts (2 + 1 = 3) are consistent, Table 4 and Table S7.1 say "build identifier in three reports" without noting that in two of the other reports, a conflicting Unity engine version (Unity 2022.3 LTS) was printed compared to the examined build (Unity 6000.2.10f1), which is a direct discrepancy between the paper-reported configuration and the examined build in Table S4.1 that is not flagged as a shortcoming or discrepancy in Section 4.1.
2. Contradiction in AR-TMT depth variation across tables — Editorial
2026). The AR-TMT places targets on physical panels in a 180 ° arc around a seated participant, under spacing constraints and distraction conditions (Baek et al., 2025a, 2025b).
In Section 3.1 (p. 4), Table 1 (p. 5), and Section 3.4 (p. 9), the AR-TMT (Baek et al., 2025a, 2025b) is described only as placing targets on physical panels in a 180° arc, while Table S7.3 (p. 37) states that "The demonstration paper describes floating targets at varying depths, which differs from the panel layout of the full study." However, Table S3.1 (p. 27) under "Layout and depth" for AR-TMT lists "targets on physical panels in a 180° arc; constrained random positions; distraction stages; mid-air stage at 0.8–3 m" without distinguishing the demo from the full study, whereas Table 1 (p. 5) omits the mid-air depth stage entirely ("TMT A and B on physical panels; constrained random positions; distraction"). More importantly, Section 3.2 (p. 6) and Section 3.4 (p. 9) treat AR-TMT's validation study as having a single panel format, leaving it unclear whether the mid-air stage at 0.8–3 m in Table S3.1 was part of the evaluated 34-participant study or only the demo paper.
3. Search completion date in text mismatches Supplementary Table S1.1 — Editorial
The library was assembled by the author between September and 24 October 2026 from database searches and from the reference lists of held papers, using trail- making, immersion and training terms. The query strings and per-database counts were not logged, so the search cannot be reproduced as a database protocol. Supple- mentary Section S1 reports it instead by its intakes, dates and record counts, with a reconciliation of the candidate records that a gap analysis identified. The evidence map holds 152 records with 37 fields (File S1). Of these, 144
Section 2 states that the 152-record library was assembled between September and 24 October 2026. According to Supplementary Table S1.1, however, the working map (140 records) and gap analysis were completed on 25 October 2026, and 12 of the 152 records—including an eligible XR implementation (Ding et al., 2026) and five full-text VR-CTT reports—were identified, appended, and read on 26 October 2026.
4. Contradictory integration status of the seeded placement repair — Editorial
For the TMT-VR, a seeded repair of the placement routine, these logging components and a bank of 4,000 candidate forms had been written and checked outside the Unity Editor. These components had not been integrated into the application or tested in the Unity Editor or on a headset as of 26 October 2026 (Supplementary Section S5). The repair is designed to reproduce every
Section 4.3 states that the seeded placement repair had not been integrated into the application or tested in the Unity Editor, and Table S5.1 similarly records "no" under the "Unity Editor" column for the seeded repair. However, Table S5.1 also states that the seeded repair was included in release 1.1.0 ("in 1.1.0, with the seeded repair"), and the Data Availability Statement notes that release 1.1.0 includes compiled standalone and PC builds attached to the release, which could not have been built without integrating the repair into the application and compiling it in Unity.
5. Inconsistent fixed versus unstated layout coding across tables — Editorial
Technical: layout logging and reproducibility N/A (fixed layout) OS (layouts not logged; placement defect) NR N/A (fixed tasks) N/A (fixed layout)
In Supplementary Table S7.1, the Motion-capture CTT (Ding et al., 2026) and the VR-TMT for epilepsy screening (Steininger et al., 2026) are both exempted from the layout logging and reproducibility criterion as "N/A (fixed tasks)" and "N/A (fixed layout)", respectively. However, Supplementary Table S3.1 codes the "Randomised layout" column for both implementations as "n/s" (not stated); if whether their layouts were randomised was not stated, layout logging cannot be ruled "N/A" for being fixed, whereas if their layouts are known to be fixed, Table S3.1 should record "no" rather than "n/s".
6. Mismatch between reported tolerances, SDs, and coefficients of variation — Minor
The bank holds 4,000 distinct layouts in four pools of 1,000 (practice A, test A, practice B, test B), selected from fresh layouts of the seeded repair by frozen criteria. Every form lies within ± 0.10 SD of the pilot medians on path length (87.21 ± 0.90 m), summed angular separation (503.9 ± 6.8 ° ) and the Fitts-type index (74.99 ± 0.36 bits), has 16 ± 1 depth-plane changes and horizontal and vertical extents within ± 0.5 SD, and shows no substantial overlap under an inscribed-sphere approximation at eye heights of 1.10, 1.20, 1.30 and 1.36 m. An independent recalculation confirmed that all 4,000 forms meet these criteria. Within the bank, the coefficients of variation are 0.60% for path length, 0.77% for angular separation and 0.27% for the index, against 10%, 13% and 5% among fresh layouts. Selection shifts other properties:
In Supplementary Section S6, the candidate form bank is described as lying within ±0.10 SD of pilot medians on path length (reported as 87.21 ± 0.90 m), angular separation (503.9 ± 6.8°), and the Fitts-type index (74.99 ± 0.36 bits), alongside CVs of 0.60%, 0.77%, and 0.27%. Here, ±0.90 m, ±6.8°, and ±0.36 bits are the half-widths of the selection bounds (0.10 of the fresh-layout SDs of 8.9 m, 68°, and 3.6 bits), whereas the CVs of 0.60%, 0.77%, and 0.27% correspond to the actual standard deviations of the truncated distributions (roughly 0.52 m, 3.9°, and 0.20 bits); labeling the ± values explicitly as selection bounds rather than SDs would prevent reader confusion.
7. Unreleased form bank and verification outputs limit auditability — Minor
English or PC versions is reported. The generated layout samples, the candidate form bank and the verification outputs described in Section 4.3 have not been released.
While the review emphasizes technical verification and open auditability, the Data Availability Statement notes that the 10,000 generated layout samples, the 4,000-form candidate bank, and the standalone emulation verification scripts/outputs described in Section 4.3 and Supplementary Section S6 have not been publicly released. Depositing these artifacts alongside the Zenodo records would allow independent verification of the reported Kolmogorov–Smirnov tests, exception rates, and form-bank CVs.
8. Internal inconsistency in reported path-length variance shares — Minor
Under the assumption that completion time is proportional to path length, simulated layout dispersion would account for about 11% of the completion-time variance in early adulthood, 4% in middle and late adulthood and 2% across the lifespan sample, with per-part shares of 14–16% and 6–7% (Kourtesis et al., in press). When random path noise of that size was added to the lifespan data
In Supplementary Section S6, the simulated layout dispersion is stated to account for 11% of completion-time variance in early adulthood, 4% in middle and late adulthood, and 2% across the lifespan sample, yet the same sentence states "with per-part shares of 14–16% and 6–7%". Because a single score in a given age group can only have one per-part variance share (unless 11% and 4% refer to a two-part average or transformed scale while 14–16% and 6–7% refer to raw per-part times), the distinction between these two sets of percentages should be explicitly clarified.
XSci (AI review)
Summary
This critical narrative review maps extended-reality (XR) adaptations of the Trail Making Test and the Color Trails Test, groups them by software lineage, and appraises each against one set of criteria covering reporting, method, measurement, technical verification and human factors. Across some fourteen implementations it finds convergence with paper and computerised trail tests varying by part and sample, clinical group differences in two lineages, retest reliability in one lineage on a fixed layout, and no normative data, prospective validation or attributable training benefit. The Trail Making Test in Virtual Reality (TMT-VR), whose source code and project records the author inspected, serves as a measurement case tracing how the confirmation rule, the timing boundary, the error definition and unlogged random layouts enter a score.
Contribution and Significance
The core contribution is the source-level measurement case of Section 4 and Section S6. Reviews of assessment software are written from published descriptions; this one is written from the code. It shows that the 0.5-s dwell adds 12 s to each part and cancels from B−A but not from B/A; that the error-counting window opens before the clock, so a pause on a wrong target during search is scored as an error; that the placement routine shuffles with a list sort using a random comparison, which violates the sort's contract and can leave a part without targets; and that the candidate field is used unevenly by depth plane (39%, 1%, 12%, 1%, 7%, 40%), with a path-length coefficient of variation of about 10% entering every score. Figure 1 generalises these into a schema other developers can apply.
Relative to the closest prior work the novelty is clear: Pieri et al. (2023) and Mancuso et al. (2024) establish the norms and openness gaps at field level but do not treat trail making or software identity, and Vlake et al. (2024) and R. M. Bauer et al. (2012) supply criteria the manuscript applies rather than extends. The inventory itself is useful but incremental, and the prescriptive conclusions follow from established measurement standards, so their force comes from the demonstration rather than the recommendations.
Strengths
- The TMT-VR case is quantitatively self-consistent and unusually candid: the placement-exception arithmetic in Section S6 checks out, and Section 4.1 discloses that three of the author's own published reports give a dwell of 2 s or 1.5 s where the examined project files set 0.5 s.
- Section 2's separation of lineage, configuration, publication and study is carried through consistently, and Table S4.1 keeps paper-reported, examined and study-linked configurations apart rather than overwriting history with a later folder.
- Table 3's four-level framing of the training question — content, programme outcomes, attribution, untrained transfer — locates the attribution gap precisely instead of asserting it.
- Section 3.1 establishes that completion times and error counts are not comparable across implementations unless each states whether wrong inputs are registered, counted, corrected or ignored.
Recommendation
Major Revision for Virtual Worlds. The manuscript makes a contribution that matters to this journal's readership and supports it; no central claim collapses, and every remedy lies within the review's own approach. What stands in the way is that the headline findings are absence claims resting on a search that cannot be reproduced and was shown to be incomplete, that the appraisal applies common criteria to uncommon evidence, and that the most distinctive analyses cannot be checked because their outputs are withheld.
Major Concerns
-
The absence claims rest on a search that cannot be reproduced and was shown to be incomplete (Section 2; Tables S1.1–S1.2; Section 6). The abstract and Section 7 state that normative data, prospective validation, reliability of randomised forms and attributable training benefit "were not identified". Yet Section 2 states that "The query strings and per-database counts were not logged, so the search cannot be reproduced as a database protocol", and that one author screened without a second screener; Table S1.1 records that an independent check of an earlier draft identified "one eligible implementation missed by the search (Ding et al., 2026)". Several of the 41 candidate records in Table S1.2 were never obtained, and 8 of 152 were read only as abstracts or titles. The bounding phrase "the sources identified by this review" is the right instinct, but readers will take the abstract at face value. Either log one reproducible query per database with its yield, obtain the outstanding records and have a second screener verify the inventory, or restate the absence claims so that they describe this library rather than the field.
-
The common appraisal is built on uncommon evidence, and the deepest-audited lineage is the author's own (Section 2; Tables 1, 4, S7.1–S7.2; Conflicts of Interest). Section 2 states that "Every implementation was appraised against the same criteria", but the evidence per implementation ranges from an inspected source tree, builds and project records (TMT-VR) to a publisher abstract. The consequence is visible in Table S7.1: the TMT-VR carries "OS" on timing verification, scoring verification and layout logging where every other lineage carries "NR". These are not on one scale — one records a shortcoming found, the other that nobody looked — yet Table 4's "Technical verification" column reads as a property of the instruments. Authorship compounds this: the author wrote all nine TMT-VR reports, four of which are a submitted manuscript, two preprints and an in-press article, yet Tables 1 and 4 place that lineage beside the VR-CTT's peer-reviewed record without distinction.
The remedy need not be large: record per implementation the access the appraisal had and the peer-review status of each report, state in the legends of Tables 4 and S7.1 that OS and NR are not comparable, and apply the technical criteria at source level to one implementation the author did not build — the AR-TMT links a public repository.
-
The most distinctive analyses are new primary work whose outputs are withheld (Section 4.3; Table S5.1; Section S6; Data Availability Statement). Section S6 reports ten thousand emulated layouts per part, per-plane placement shares, exception rates per 10^6 placements, modelled layout contributions to completion-time variance, and a 4,000-form bank. This is original simulation and audit work, not review synthesis, yet the Data Availability Statement records that "The generated layout samples, the candidate form bank and the verification outputs described in Section 4.3 have not been released", and Table S5.1 adds that the seeded repair, the logging components and the form-bank selector are written as source but not integrated, run in the Unity Editor or tested on a headset. A review that asks every other group to log, verify and release should not leave its own central quantitative claims unverifiable. Identify Section 4.3 and S6 as original analysis, give the emulation procedure in enough detail to repeat it, and deposit the scripts and outputs alongside the released source.
Minor Comments
- Table 3 reports impossible intervals from Qin et al. (2025) — .71 (.18 to .124) and .74 (.44 to .104) — and Yen and Chiu's g −.05 (−.37 to .46), whose interval is not centred on its estimate.
- In Table 4 the VR-CTT human-factors cell ends with the TMT-VR text ("SUS, user experience, acceptability and CSQ-VR measured; seated and standing; adults to 90 years and MCI"), misattributing measures the VR-CTT did not collect.
- Section 4.1 reports a published dwell of 2 s or 1.5 s against 0.5 s in the examined project files, which Table S4.1 leaves "unresolved"; say what follows for the three affected reports.
- Section 4.2's note that "one adult in ten stood up once during a part to locate a target, with the clock running" should be read against Section S6's 0.8 targets per layout partly hidden from the seat, which together impose a postural demand on less mobile older adults.
- The exception rates in Section S6 come from an emulation with Unity's Mono class library, whereas the deployed build is Android IL2CPP; state whether the same rate is expected there.
- Correct the front matter: the "Funding" heading is empty and its text sits under "Informed Consent Statement", "Not applicable." is duplicated at the ethics statement, and affiliation c appears twice with conflicting postcodes.
- Given the title's promise for cognitive ageing, the abstract should state what Section 3.4 establishes: only two lineages have evaluated adults over 65, and none studied dementia.
- The recommendation to log every layout, seed, event, interrupted dwell and head pose needs an estimate of its storage and runtime cost on a standalone headset.
- Section 5's proposed seated game in which "players release laser or water pistols at balloons" is a design proposal, not a review finding, and should be marked as such.
- Roughly fourteen reference-list entries are the author's own work; consider a non-self source where the citation is generic.
- A coverage diagram of the evidence map would make Section 2 and Table S1.1 easier to follow.
- Copy-editing: "AAUC" in Table 2 should read ΔAUC; the Dahmen et al. (2017) DOI differs between the two reference lists; "Goumpoulos" appears in Table S7.3; and Section 2's "Criteria were scaled to the use of an implementation claims" needs repair.