Accuracy and Validity of 3D Markerless Motion Capture Compared to Marker-Based Systems for Lower-Limb Biomechanical Assessment: A Systematic Review

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

< !-- obsidian -->

This systematic review evaluates markerless motion capture systems against marker-based references for lower-limb kinematic assessment during jump-landing tasks relevant to ACL injury screening. The topic is timely, the PRISMA framework is applied, and a broad range of validity metrics is reported. However, several methodological, statistical, and editorial issues require attention before the manuscript can be accepted.

1. PROSPERO Registration

The PROSPERO record (CRD420251011431) lists the review as incomplete despite the manuscript being submitted for publication. Authors must update the registration status to "Completed." and take into account works discovered throughout entire study. Additionally, the author list in the PROSPERO record differs from that in the manuscript; this discrepancy should be explained.

2. Absence of Methodological Quality Assessment

No formal quality appraisal of included studies is reported. For a systematic review evaluating measurement validity, a tool such as COSMIN or QAREL should be applied. Without this, all 14 studies are implicitly treated as methodologically equivalent, which is not warranted given the wide variation in sample sizes and designs. The authors should either apply a validated appraisal tool or provide a clear justification for its omission, and reflect any quality concerns in the conclusions.

3. Single-Participant Studies

Two included studies (Giblin et al. and McGroarty et al.) involved only a single participant. The authors acknowledge excluding these from the results synthesis, but their inclusion in the review itself is questionable. The authors should either exclude these studies entirely and update the PRISMA diagram, or explicitly justify their retention and discuss the implications for the review's conclusions.

4. Statistical Incoherence in Bland–Altman Data (Asaeda et al.)

By definition, the Bland–Altman bias must lie centrally between the limits of agreement, since LoA = bias ± 1.96 × SD. The values reported for Asaeda et al. (bias: −19.28°; LoA: −12.91° to +25.66°) violate this relationship, as the bias falls outside its own limits of agreement. This indicates either a reporting error in the source paper or a data extraction error in this review. The authors must investigate the source, clearly state the likely cause, and consider whether data from Asaeda et al. can be retained in the synthesis.

5. Reference 16 is Inappropriate

Reference 16 (Meeus M., Health Literacy from Reference to Review, 2016) is cited as the source for definitions of "accuracy" and "validity" in a biomechanical measurement context. This is a health literacy textbook and is not an appropriate citation here. It should be replaced with a peer-reviewed publication from the measurement science or biomechanics literature.

6. Unsupported Claims Regarding the Noise Floor of Marker-Based Systems

Lines 330–331 state that OpenCap error values "remain within the practical noise floor of marker-based systems" and lines 355–356 similarly claim that low MAE values "are comparable to the typical variability attributed to marker-based systems themselves". Marker-based systems such as produced by Vicon achieve sub-millimetre spatial accuracy under static conditions; their angular error under dynamic conditions depends on inter-marker distance, segment velocity, and acquisition frequency and is not self-evidently of the order of several degrees. These claims require explicit supporting citations or authors' own reference measurements.

7. Methodological Heterogeneity Undermines Comparative Conclusions

The authors themselves acknowledge "substantial methodological heterogeneity across studies" in Section 5, citing variation in movement tasks, participant characteristics, camera configurations, and biomechanical modelling approaches. This heterogeneity is not merely a limitation of interpretation — it raises a more fundamental question about whether the included studies are sufficiently comparable to support cross-study synthesis at all. When technologies, tasks, planes of motion, outcome metrics, and reference systems all vary simultaneously, it becomes difficult to attribute observed differences in performance to the markerless technology itself rather than to experimental design. The authors should address this issue directly: either demonstrate that meaningful comparisons can still be drawn despite this heterogeneity, or acknowledge that the review's primary contribution is the aggregation of existing evidence rather than the derivation of generalisable conclusions about specific technologies.

8. Missing Citation Numbers Throughout the Text

Numerous passages refer to included studies by author name without providing the corresponding reference number (e.g., lines 104, 136-138, 142, 164-166, 169, 171, 173, 213, 220, 224, 239, 241, 243, 247, 249, 261, 263-264, 267, 270, 272, 284-285, 287, 315-316, 319, 321, 326, 333, 341, 346, 353, 358, 375, 380, 388, 406, 409, 412, 416, 419, 436, 442, 446, 448, 452, 467, 472, 496, and Table 1). While repeated citation of the same source is not always necessary, each first mention of a study within a section must carry a reference number. This appears to be a systematic editorial omission.

9. iPhone Nomenclature in Table 1

Entries for Turner et al., Barzyk et al., and Aleksic et al. list hardware as "X iPhone(s) (iPhone YY)". All other entries in the same column specify only the device model or product name without a preceding category label. The redundant use of "iPhone" before and inside the parenthesis is inconsistent. The authors should either remove the category label and list only the model (consistent with other entries), or introduce a device category label (e.g., "Smartphone") for all applicable rows.

10. Table 1 Orientation

The current layout, with studies as columns and characteristics as rows, results in a table that is difficult to read, particularly for the movement task row which contains the most text. Transposing the table — with studies as rows and characteristics as columns — would improve legibility and align the structure with Tables 2–7. Shared row headers (e.g., "Number of participants" spanning sub-rows for All/Male/Female) could further reduce redundancy.

11. Inconsistent Reference Formatting

Several references with complete bibliographic data (journal, volume, pages) are formatted as internet articles with "Available from: [URL]" appended (e.g., References 3, 4, 5, 10, 14, 17). This format should be reserved for sources without standard bibliographic identifiers. All references with full journal data should be formatted consistently according to MDPI citation guidelines.

12. Mathematical Notation

Statistical symbols and variable names (e.g., r, p, n, SD) should be typeset in italics throughout, while abbreviations (RMSE, MAE, ICC, CMC) should appear in roman type. This requirement applies consistently throughout the text and tables.

13. Missing Subsection Heading

The paragraph beginning at line 482 ("Across studies, several markerless motion-capture systems demonstrated…") constitutes a cross-metric synthesis of technology-level performance and is structurally distinct from the preceding subsection 4.2.4. It should be given its own heading, e.g., "4.3 Comparative Performance Across Technologies."

14. Recurring Typographical Error

The word "vertical" is misspelled as "verticle" in lines 195 and 329.

15. Inconsistent Plane Terminology

"Coronal plane" and "frontal plane" refer to the same anatomical plane but are used interchangeably. One term should be selected and applied consistently throughout.

16. Negative ICC Values Not Discussed

Table 5 contains several negative ICC(2,1) values from Mauntel et al. (e.g., −0.19, −0.4, −1.52). Negative values under this model indicate that measurement error exceeds between-subjects variance, implying severe disagreement. These values are not addressed in the text; their implications should be explicitly discussed.

17. Interpretive Thresholds Should Be Pre-specified in the Methods

The thresholds used in the Discussion to classify accuracy (e.g., RMSE ≤ 5° as acceptable) are introduced without prior specification in the Methods. These criteria, with supporting references, should be stated in the Methods section to avoid the appearance of post-hoc interpretation.

Author Response

1. PROSPERO Registration

The PROSPERO record (CRD420251011431) lists the review as incomplete despite the manuscript being submitted for publication. Authors must update the registration status to "Completed." and take into account works discovered throughout entire study. Additionally, the author list in the PROSPERO record differs from that in the manuscript; this discrepancy should be explained.

I will get in touch with the person in charge of the PROSPERO and get the registration status changed. I have added the additional two authors onto the article. 

2. Absence of Methodological Quality Assessment

No formal quality appraisal of included studies is reported. For a systematic review evaluating measurement validity, a tool such as COSMIN or QAREL should be applied. Without this, all 14 studies are implicitly treated as methodologically equivalent, which is not warranted given the wide variation in sample sizes and designs. The authors should either apply a validated appraisal tool or provide a clear justification for its omission, and reflect any quality concerns in the conclusions.

I have produced a COSMIN table and included this within the methods section. 

3. Single-Participant Studies

Two included studies (Giblin et al. and McGroarty et al.) involved only a single participant. The authors acknowledge excluding these from the results synthesis, but their inclusion in the review itself is questionable. The authors should either exclude these studies entirely and update the PRISMA diagram, or explicitly justify their retention and discuss the implications for the review's conclusions.

I have removed giblin, mcgroarty and karatzas from the paper and prisma diagram. 

4. Statistical Incoherence in Bland–Altman Data (Asaeda et al.)

By definition, the Bland–Altman bias must lie centrally between the limits of agreement, since LoA = bias ± 1.96 × SD. The values reported for Asaeda et al. (bias: −19.28°; LoA: −12.91° to +25.66°) violate this relationship, as the bias falls outside its own limits of agreement. This indicates either a reporting error in the source paper or a data extraction error in this review. The authors must investigate the source, clearly state the likely cause, and consider whether data from Asaeda et al. can be retained in the synthesis.

I have commented on this on line 620 onwards. 

5. Reference 16 is Inappropriate

Reference 16 (Meeus M., Health Literacy from Reference to Review, 2016) is cited as the source for definitions of "accuracy" and "validity" in a biomechanical measurement context. This is a health literacy textbook and is not an appropriate citation here. It should be replaced with a peer-reviewed publication from the measurement science or biomechanics literature.

This is an academic textbook and has been quoted by other peer-reviewed papers for definitions, hence why I have deployed it here. 

6. Unsupported Claims Regarding the Noise Floor of Marker-Based Systems

Lines 330–331 state that OpenCap error values "remain within the practical noise floor of marker-based systems" and lines 355–356 similarly claim that low MAE values "are comparable to the typical variability attributed to marker-based systems themselves". Marker-based systems such as produced by Vicon achieve sub-millimetre spatial accuracy under static conditions; their angular error under dynamic conditions depends on inter-marker distance, segment velocity, and acquisition frequency and is not self-evidently of the order of several degrees. These claims require explicit supporting citations or authors' own reference measurements.

I have provided references for these claims.

7. Methodological Heterogeneity Undermines Comparative Conclusions

The authors themselves acknowledge "substantial methodological heterogeneity across studies" in Section 5, citing variation in movement tasks, participant characteristics, camera configurations, and biomechanical modelling approaches. This heterogeneity is not merely a limitation of interpretation — it raises a more fundamental question about whether the included studies are sufficiently comparable to support cross-study synthesis at all. When technologies, tasks, planes of motion, outcome metrics, and reference systems all vary simultaneously, it becomes difficult to attribute observed differences in performance to the markerless technology itself rather than to experimental design. The authors should address this issue directly: either demonstrate that meaningful comparisons can still be drawn despite this heterogeneity, or acknowledge that the review's primary contribution is the aggregation of existing evidence rather than the derivation of generalisable conclusions about specific technologies.

I have commented upon heterogeneity on line 794 onwards. 

8. Missing Citation Numbers Throughout the Text

Numerous passages refer to included studies by author name without providing the corresponding reference number (e.g., lines 104, 136-138, 142, 164-166, 169, 171, 173, 213, 220, 224, 239, 241, 243, 247, 249, 261, 263-264, 267, 270, 272, 284-285, 287, 315-316, 319, 321, 326, 333, 341, 346, 353, 358, 375, 380, 388, 406, 409, 412, 416, 419, 436, 442, 446, 448, 452, 467, 472, 496, and Table 1). While repeated citation of the same source is not always necessary, each first mention of a study within a section must carry a reference number. This appears to be a systematic editorial omission.

The appropriate citations have been added.

9. iPhone Nomenclature in Table 1

Entries for Turner et al., Barzyk et al., and Aleksic et al. list hardware as "X iPhone(s) (iPhone YY)". All other entries in the same column specify only the device model or product name without a preceding category label. The redundant use of "iPhone" before and inside the parenthesis is inconsistent. The authors should either remove the category label and list only the model (consistent with other entries), or introduce a device category label (e.g., "Smartphone") for all applicable rows.

The nomenclature has been changed in the table. 

10. Table 1 Orientation

The current layout, with studies as columns and characteristics as rows, results in a table that is difficult to read, particularly for the movement task row which contains the most text. Transposing the table — with studies as rows and characteristics as columns — would improve legibility and align the structure with Tables 2–7. Shared row headers (e.g., "Number of participants" spanning sub-rows for All/Male/Female) could further reduce redundancy.

The table has been adjusted to meet these suggestions. 

11. Inconsistent Reference Formatting

Several references with complete bibliographic data (journal, volume, pages) are formatted as internet articles with "Available from: [URL]" appended (e.g., References 3, 4, 5, 10, 14, 17). This format should be reserved for sources without standard bibliographic identifiers. All references with full journal data should be formatted consistently according to MDPI citation guidelines.

Reference formatting has been amended. 

12. Mathematical Notation

Statistical symbols and variable names (e.g., r, p, n, SD) should be typeset in italics throughout, while abbreviations (RMSE, MAE, ICC, CMC) should appear in roman type. This requirement applies consistently throughout the text and tables.

Appropriate mathematical notation has been employed. 

13. Missing Subsection Heading

The paragraph beginning at line 482 ("Across studies, several markerless motion-capture systems demonstrated…") constitutes a cross-metric synthesis of technology-level performance and is structurally distinct from the preceding subsection 4.2.4. It should be given its own heading, e.g., "4.3 Comparative Performance Across Technologies."

The subsection heading has been added.

14. Recurring Typographical Error

The word "vertical" is misspelled as "verticle" in lines 195 and 329.

This has been amended. 

15. Inconsistent Plane Terminology

"Coronal plane" and "frontal plane" refer to the same anatomical plane but are used interchangeably. One term should be selected and applied consistently throughout.

This has been amended. Coronal used throughout. 

16. Negative ICC Values Not Discussed

Table 5 contains several negative ICC(2,1) values from Mauntel et al. (e.g., −0.19, −0.4, −1.52). Negative values under this model indicate that measurement error exceeds between-subjects variance, implying severe disagreement. These values are not addressed in the text; their implications should be explicitly discussed.

These values have now been discussed on line 667 onwards. 

17. Interpretive Thresholds Should Be Pre-specified in the Methods

The thresholds used in the Discussion to classify accuracy (e.g., RMSE ≤ 5° as acceptable) are introduced without prior specification in the Methods. These criteria, with supporting references, should be stated in the Methods section to avoid the appearance of post-hoc interpretation.

Interpretative thresholds have been moved to methods. 

Reviewer 2 Report

Comments and Suggestions for Authors

Review report – 4330137– Sensors

 

A brief summary

 

The aim of this systematic review was to evaluate the accuracy and validity (and extent of applicability) of markerless motion capture systems for measuring lower-limb kinematics during jump-landing tasks commonly used in ACL injury screening. Authors findings suggest that markerless motion capture shows potential for biomechanical assessment in ACL injury screening, with further validation.

 

General comments

 

Introducing overview was covered with relevant earlier research and state-of-the-art of the domain and supported by methodology used to fulfil aim of the analysis. The variables assessed in this systematic review are in line with basic biomechanics approaches. Experimental design was proper. PRISMA flow diagram, characteristics of the studies included in the review, and general methodology supported following discussion and conclusions. In limitations of the study authors correctly delineated the scope of generalisations. Therefore, this is valuable contribution to broader scientific public.

 

Specific comments

 

·       Ln33-35:” Approximately 90% of ACL injuries occur through non-contact or indirect contact mechanisms, demonstrating that most ACL injuries arise from intrinsic biomechanical loading patterns rather than external collision forces [4]” – although cited corectly, external collision forces never occure without intrisic loadings absolutely separately, and  its not the case vice versa. To be more precise – ‘context’ has same power as collision – decision tree obstruction etc…pointed later as ‘sport demands’. Clarify and extend please.

·       Ln 60-61: “ Alternatively, depth cameras such as Kinect can be used which provide depth maps directly [13].” – it is true, however, with limited validity for all ACL issues with regard to gold standard. It differs to general group biomechanics, and should be more clear to reader in introduction and at all points in manuscript.

·       Ln136: “ all studies involved a maximum of 20 participants.” – it is more relevant that most studies (10 out of 14) were between 11 and 20 participants, 2 were with 1, and 2 were above 20.

·       Ln316-318:”Aleksic et al. found the smallest RMSE for ankle angles and the largest for hip angles when comparing MMPose to Qualisys; the right lower limb joints also consistently demonstrated higher RMSE than the left.” – this seem to be one of two, a ‘glitch’ or very valuable and applicable insight

·       In conclusions, It would be valuable to point out more directly lower accuracy with regard to distal joints in all plains, and e.g. ‘substantial’ systematic error and variability for knee valgus

Author Response

Ln33-35:” Approximately 90% of ACL injuries occur through non-contact or indirect contact mechanisms, demonstrating that most ACL injuries arise from intrinsic biomechanical loading patterns rather than external collision forces [4]” – although cited corectly, external collision forces never occure without intrisic loadings absolutely separately, and  its not the case vice versa. To be more precise – ‘context’ has same power as collision – decision tree obstruction etc…pointed later as ‘sport demands’. Clarify and extend please.

This has been amended (line 34). 

Ln 60-61: “ Alternatively, depth cameras such as Kinect can be used which provide depth maps directly [13].” – it is true, however, with limited validity for all ACL issues with regard to gold standard. It differs to general group biomechanics, and should be more clear to reader in introduction and at all points in manuscript.

Authors agree. 

Ln136: “ all studies involved a maximum of 20 participants.” – it is more relevant that most studies (10 out of 14) were between 11 and 20 participants, 2 were with 1, and 2 were above 20.

This has been amended (line 200). 

·       Ln316-318:”Aleksic et al. found the smallest RMSE for ankle angles and the largest for hip angles when comparing MMPose to Qualisys; the right lower limb joints also consistently demonstrated higher RMSE than the left.” – this seem to be one of two, a ‘glitch’ or very valuable and applicable insight

Authors agree.

In conclusions, It would be valuable to point out more directly lower accuracy with regard to distal joints in all plains, and e.g. ‘substantial’ systematic error and variability for knee valgus

This has been added to conclusion.

 

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The authors have substantially revised the manuscript and addressed the majority of concerns raised in the first round, including the addition of a formal COSMIN-based quality appraisal, removal of single-participant studies, resolution of the Bland–Altman incoherence, pre-specification of interpretive thresholds, and numerous editorial corrections. The manuscript is considerably stronger as a result. A small number of points remain, detailed below, alongside one issue concerning the change in authorship that requires clarification before the manuscript can proceed.

Comment 1 (PROSPERO Registration / Authorship)
The PROSPERO record has been reconciled with the manuscript by adding two authors (A. Machesney and T. Mather), now credited with "methodology" in the Author Contributions statement — a contribution category that did not appear at all in the original submission. This raises a question that the revision does not address: why were these two contributors, apparently responsible for the methodology underpinning the entire review, omitted from the original author list? Two possibilities exist, and both are concerning. Either their contribution was substantive and they were wrongly excluded from the original submission (which would itself be an authorship rights issue), or they have been added at this stage primarily to resolve the PROSPERO discrepancy without a corresponding substantive contribution. The authors are asked to provide a clear explanation of the circumstances. Depending on the explanation, this may also need to be documented separately with the handling editor in line with the journal's authorship change policy.

Comment 2 (Methodological Quality Assessment)
Addressed. The addition of a COSMIN Risk of Bias assessment (Table 1), with per-study ratings across all eight standards and a clearly justified worst-score-counts classification, is a substantial improvement and directly resolves the original concern.
One organisational suggestion: Table 1 presents the results of applying the COSMIN tool to each included study and, as such, may sit more naturally in Section 3 (Results) alongside Table 2, rather than in Section 2 (Methods), where only the description of the tool itself would remain. Relatedly, whichever table first introduces the included studies should carry the corresponding reference numbers (see Comment 8). This is a suggestion for clarity rather than a requirement.

Comment 3 (Single-Participant Studies)
Fully addressed. Three studies have been removed from the review, the PRISMA diagram updated accordingly, and the narrative adjusted throughout.

Comment 4 (Statistical Incoherence in Bland–Altman Data, Asaeda et al.)
Addressed. The authors have identified that Asaeda et al. likely reported incorrect limit of agreement. The explanation in lines 432–436 together with the COSMIN Table 1 footnote (S6*) provides a coherent and well-reasoned resolution. No further action required.

Comment 5 ("Health Literacy from Reference to Review" Reference)
The citation (now Reference 34) remains "Meeus M. Health Literacy from Reference to Review. 2016" with no page numbers or chapter information. If the authors wish to retain this source for the definitions of "accuracy" and "validity" in Section 3.3/3.4, two corrections are required: first, full bibliographic entry should be provided; second, given the work spans over 200 pages, the specific chapter and page range from which these definitions are drawn must be provided. Alternatively, the authors are encouraged to reconsider a biomechanics- or measurement-science-specific source, as suggested in Round 1.

Comment 6 (Noise Floor of Marker-Based Systems)
Partially addressed, but the added citation does not fully support the claim as currently worded. Reference 35 (Song et al., 2023) is now cited in support of statements that markerless errors "remain within the practical noise floor of marker-based systems" and are "comparable to the typical variability attributed to marker-based systems themselves".
However, Song et al. frame marker-based motion capture as having "inherent inaccuracy," a term in the biomechanics literature that is generally attributed to soft-tissue artefact, marker placement relative to anatomical landmarks, and inter-protocol differences in segment/joint-centre definitions — not to the instrumental precision (noise floor) of the optoelectronic hardware itself, which is sub-degree under controlled conditions. The manuscript's framing risks conflating these two distinct sources of variability: between-protocol variability driven by STA and modelling conventions (which Reference 35 documents) versus device-level measurement noise (which the manuscript's wording implies).
The authors are asked to either reframe these statements to refer specifically to between-protocol variability/inherent inaccuracy (matching the terminology and findings of their own cited source) rather than "noise floor," or to provide a citation that directly characterises the instrumental noise floor of motion capture systems under the dynamic conditions relevant here.

Comment 7 (Methodological Heterogeneity)
Partially addressed. The new justification in Section 5 (lines 565–569) — that most studies use Vicon as a common reference and that all tasks are variations of jumping and landing — provides a general rationale for cross-study synthesis, but does not engage with the specific case raised in Round 1. For example, in Table 3/Section 4.1.1, RMSE values Aleksic et al. (countermovement jump) and Drazan et al. (vertical jump) are discussed alongside one another, yet these represent different movement tasks as well as different technologies. The general justification offered does not establish whether RMSE differences between these two studies can be attributed to the markerless technology rather than the task. The authors are encouraged to provide insight into the relations between values reported by different studies rather than focus on aggregating those values.

Comment 8 (Missing Citation Numbers)
Partially addressed. Citation numbers have been added to most rows of Table 2 and to the prose in Section 3.1. However, four entries in Table 2 (Turner et al., Tipton et al., Templin et al., and Asaeda et al.) still lack reference numbers, despite these numbers being available elsewhere ([23]–[26]). Additionally, throughout Sections 3.3–4.3, where each study is discussed in detail (often at greater length than in Section 3.1), studies continue to be referred to by author name only, with no reference number at first mention within these sections. The authors should complete the citation numbers in Table 2 and add a reference number at the first mention of each study within the Results and Discussion sections.

Comment 9 (iPhone Nomenclature, Table 2)
Fully addressed.

Comment 10 (Table Orientation)
Fully addressed.

Comment 11 (Inconsistent Reference Formatting)
Fully addressed.

Comment 12 (Mathematical Notation)
Fully ddressed.

Comment 13 (Missing Subsection Heading)
Fully addressed.

Comment 14 (Recurring Typographical Error, "verticle")
Fully addressed.

Comment 15 (Inconsistent Plane Terminology)
Partially addressed. The running text has been consistently updated from "frontal" to "coronal" plane throughout Sections 4.2–6. However, Table 6 (Stone et al. entries, "Frontal plane knee angle (varus/valgus alignment)," appearing twice) was not updated and now represents the sole remaining instance of "frontal plane" terminology in the manuscript. This should be corrected to "coronal plane" for consistency with the rest of the text.

Comment 16 (Negative ICC Values)
Fully addressed.

Comment 17 (Interpretive Thresholds Pre-specified in Methods)
Fully addressed.

Author Response

Comment 1 (PROSPERO Registration / Authorship)
The PROSPERO record has been reconciled with the manuscript by adding two authors (A. Machesney and T. Mather), now credited with "methodology" in the Author Contributions statement — a contribution category that did not appear at all in the original submission. This raises a question that the revision does not address: why were these two contributors, apparently responsible for the methodology underpinning the entire review, omitted from the original author list? Two possibilities exist, and both are concerning. Either their contribution was substantive and they were wrongly excluded from the original submission (which would itself be an authorship rights issue), or they have been added at this stage primarily to resolve the PROSPERO discrepancy without a corresponding substantive contribution. The authors are asked to provide a clear explanation of the circumstances. Depending on the explanation, this may also need to be documented separately with the handling editor in line with the journal's authorship change policy.

The authors were involved in curating the methodology and literature search. I forgot to add them initially by mistake - they should have been on there in the beginning. This has been sorted with the authors and the editorial office. 

Comment 2 (Methodological Quality Assessment)
Addressed. The addition of a COSMIN Risk of Bias assessment (Table 1), with per-study ratings across all eight standards and a clearly justified worst-score-counts classification, is a substantial improvement and directly resolves the original concern.
One organisational suggestion: Table 1 presents the results of applying the COSMIN tool to each included study and, as such, may sit more naturally in Section 3 (Results) alongside Table 2, rather than in Section 2 (Methods), where only the description of the tool itself would remain. Relatedly, whichever table first introduces the included studies should carry the corresponding reference numbers (see Comment 8). This is a suggestion for clarity rather than a requirement.

COSMIN table has been moved to results section underneath study characteristics table. 

Comment 5 ("Health Literacy from Reference to Review" Reference)
The citation (now Reference 34) remains "Meeus M. Health Literacy from Reference to Review. 2016" with no page numbers or chapter information. If the authors wish to retain this source for the definitions of "accuracy" and "validity" in Section 3.3/3.4, two corrections are required: first, full bibliographic entry should be provided; second, given the work spans over 200 pages, the specific chapter and page range from which these definitions are drawn must be provided. Alternatively, the authors are encouraged to reconsider a biomechanics- or measurement-science-specific source, as suggested in Round 1.

A difference source is used for the definitions of accuracy (agreement) and validity. 

Comment 6 (Noise Floor of Marker-Based Systems)
Partially addressed, but the added citation does not fully support the claim as currently worded. Reference 35 (Song et al., 2023) is now cited in support of statements that markerless errors "remain within the practical noise floor of marker-based systems" and are "comparable to the typical variability attributed to marker-based systems themselves".
However, Song et al. frame marker-based motion capture as having "inherent inaccuracy," a term in the biomechanics literature that is generally attributed to soft-tissue artefact, marker placement relative to anatomical landmarks, and inter-protocol differences in segment/joint-centre definitions — not to the instrumental precision (noise floor) of the optoelectronic hardware itself, which is sub-degree under controlled conditions. The manuscript's framing risks conflating these two distinct sources of variability: between-protocol variability driven by STA and modelling conventions (which Reference 35 documents) versus device-level measurement noise (which the manuscript's wording implies).
The authors are asked to either reframe these statements to refer specifically to between-protocol variability/inherent inaccuracy (matching the terminology and findings of their own cited source) rather than "noise floor," or to provide a citation that directly characterises the instrumental noise floor of motion capture systems under the dynamic conditions relevant here.

Terminology has been changed to clarify between the two sources of variability.

Comment 7 (Methodological Heterogeneity)
Partially addressed. The new justification in Section 5 (lines 565–569) — that most studies use Vicon as a common reference and that all tasks are variations of jumping and landing — provides a general rationale for cross-study synthesis, but does not engage with the specific case raised in Round 1. For example, in Table 3/Section 4.1.1, RMSE values Aleksic et al. (countermovement jump) and Drazan et al. (vertical jump) are discussed alongside one another, yet these represent different movement tasks as well as different technologies. The general justification offered does not establish whether RMSE differences between these two studies can be attributed to the markerless technology rather than the task. The authors are encouraged to provide insight into the relations between values reported by different studies rather than focus on aggregating those values.

General justification has been replaced with a focused insight into the specific example raised. 

Comment 8 (Missing Citation Numbers)
Partially addressed. Citation numbers have been added to most rows of Table 2 and to the prose in Section 3.1. However, four entries in Table 2 (Turner et al., Tipton et al., Templin et al., and Asaeda et al.) still lack reference numbers, despite these numbers being available elsewhere ([23]–[26]). Additionally, throughout Sections 3.3–4.3, where each study is discussed in detail (often at greater length than in Section 3.1), studies continue to be referred to by author name only, with no reference number at first mention within these sections. The authors should complete the citation numbers in Table 2 and add a reference number at the first mention of each study within the Results and Discussion sections.

Citation numbers added to table and first mention of studies in results and discussion section. 

Comment 15 (Inconsistent Plane Terminology)
Partially addressed. The running text has been consistently updated from "frontal" to "coronal" plane throughout Sections 4.2–6. However, Table 6 (Stone et al. entries, "Frontal plane knee angle (varus/valgus alignment)," appearing twice) was not updated and now represents the sole remaining instance of "frontal plane" terminology in the manuscript. This should be corrected to "coronal plane" for consistency with the rest of the text.

Addressed.