I would like to congratulate the author on a generally well-written and intriguing article. I am convinced that IV methods are a valuable addition to the toolbox of ecology and evolution researchers. While I find no technical problems with the methods and examples presented, I believe that further consideration of how they are explained and discussed would result in a more useful and impactful article.
More specifically, it strikes me that the article is being pulled in two directions: a descriptive review of SEMIV and its use/applicability to ecology & evolution, and prescriptive, instructional guidelines of how to apply SEMIV. Though I think this could be a valuable combination, and that the article has the potential to accomplish both, I wonder if the current draft is limited somewhat by trying to accomplish too much in too little space.
For example, on the “review angle”, I was surprised that little space is devoted to discussing the relationship between RegIV and SEMIV (especially given how brief the Discussion is!). Specifically, the idea that SEMIV complements RegIV by providing an “integrative explanatory framework” (Line 475) seems underdeveloped. In the main text, references to Grace & Irvine 2020 (Lines 357, 363) seem to be doing much of the legwork in advancing this idea, with the distinctions between “explanatory and causal modeling” not being explained directly. If this and related challenges in the causal inference literature (Line 360) are critical to understanding the distinct benefit of SEMIV, I think that they should be presented in this article.
Regarding the “instructional angle”, certain sections (e.g., 307-332) seem too dense with SEM-specific terminology to carry along readers unfamiliar with the technique. If an intended goal is to encourage a wider audience of researchers in ecology & evolution to adopt the method, I think that demystifying these more technical sections by adding background for different steps would be helpful. I think that the author does an excellent job of this in some other procedural sections (e.g., Lines 389-410).
So, as the word limit permits, I think that strategic additions to clarify the features of SEMIV and stitch these directly to the stated benefit of building integrative explanatory models will result in a much stronger review.

Please find more specific comments below:
Line 3: At first glance, the title led me to expect that the inclusion of IVs in SEMs was a novel synthesis. If I understand correctly, this article is best characterized as an exposition of existing SEMIV methods for ecologists & evolutionary biologists.
Line 38: Would it be helpful to signal the distinction between realized and intended treatment here (e.g., “intent to treat”)? I think that some readers may assume that “treatment variable” and COI are synonymous.
Lines 70-73: Nice job setting up the background for RegIV
Line 74: As a relative newcomer to SEM, I’m left wondering how it relates to the general causal inference literature (i.e., backdoor-criterion methods etc.). Would it make sense to briefly situate SEM relative to this methodology?
Line 79: Possible typo: “independencies”?
Line 84: How exactly has SEMIV been applied implicitly in the natural sciences? This does not become obvious to me.
Lines 109-114: I think this is nicely done with diagrams for each case. However, I think that a brief written example for some of these cases could be helpful though (or maybe a citation to an example in the literature).
Line 136: Why is this assumption “asymptotic”?
Lines 180-192: If I understand correctly, fire severity (COI) would need to be measured here, separately from intent to treat for the IV method to apply. Intuitively, I imagine most basic ecological experiments do not distinguish treatment intent from COI. If so, it might be helpful to remind the reader that fire severity has been measured separately in this hypothetical example.
Line 195: At first glance, I would assume that a causal relationship is necessary between the IV and the COI (even a mediated one)? What sort of non-causal, “predictive” relationship would suffice?
Lines 206-212: Generally I think this is a very clearly written paragraph. I wonder if the phrase “exogenous alternative to the COI” could be explained further though? It would be helpful to have an intuitive explanation of this measure that does not rely on more method-specific language (i.e., “exogenous”).
Line 217: A small point: Could another letter be used for the intercept in Equation 8? The gamma looks similar to ‘y’ in Equation 9.
Lines 228-229: This might be an opportunity to further explain the key features of SEM. For example, while referring to Figure 1F is a great idea, it would be more helpful if the text included a brief description relating to specific terms in the diagram.
Line 260: Is “unsaturated” a commonly used term to describe models in Ecology & Evolutionary Biology? I am not familiar with it and wonder if a brief definition would be helpful.
Line 265: Similar in what way specifically? In any case, I think if the survey is informal it would be best to use qualitative terms rather than a percentage.
Lines 275-279: These sentences are identical to those in Grace & Irvine 2020. Surely this would need to be changed so as not to be plagiaristic?
Line 282: Subtitle could include reference to Whalen example? Otherwise seems like an odd shift in topic.
Lines 288-290: What is meant by Q3, separate from an association among treatment and covariates? The R script shows Q2 being addressed by linear regression, while Q3 is addressed by boxplots, though the variables being compared are the identical.
Line 291: Instead of “details relating to”, this statement might reflect that Appendix S2 more specifically contains the code for generating answers to these questions.
Line 294-305: It might be helpful to inject more biological interpretation here (i.e., carrying through the example).
Lines 307-322: As stated in the general comments, I think further explanation (maybe in stepwise format) of this section would be beneficial. From a structural standpoint, this section seems to be the critical procedural section of the article (i.e., how SEMIV is applied), so probably is deserving of some more text.
Line 308: Typo – “were” instead of “was”.
Line 318-319: What about Fig. 1 suggest this specifically?
Line 325: Why is this approach vulnerable to overfitting?
Line 333: Is there any reason that there is no direct comparison of interpretations from RegIV and SEMIV? That would be quite informative and interesting, I think.
Lines 344-346: As phrased, I wonder if it promotes too strong a view: surely there is no need to capitalize on strong IVs when there is complete control of the COI, and when there are no suspected sources of causal bias?
Lines 355-364: This Lesson seems like the best opportunity to delineate the broad advantages of implementing IV in a SEM framework. However, as it stands, there is too great a gap between the technical implementation of SEMIV and these theoretical points. I think that this section would be much stronger if the procedural approach (lines 307-332) were more explicitly linked to their purported improvement in explanatory potential. For example, what precisely is the reader meant to understand when comparing Figures 2 and 3?
Lines 358-360: This is a strong claim which would require some evidence or examples – how are the results from a SEM analysis better set up to build explanations?
Lines 362-364: I am not sure what is meant by interpretive content here…what specifically do users need to explain when applying IV in a SEM rather than regression framework?
Line 366: I am a little surprised to see the “Lessons Illustrated” paused here. Though I can imagine the benefit of attaching lessons directly to examples, I wonder if keeping all lessons together towards the end of the article would be more effective?
Lines 370-371: At least one IV for each causal effect one wishes to verify/refine seems quite demanding. Is there some sense in which IVs should be sought out when there is a reasonable expectation of bias (e.g., expectation of an unmeasured common-cause confound)?
Lines 390-392: This is a nice example of the more pedagogical text where the logic of the procedure is very clear.
Line 409: Which formal tests specifically?
Line 471: Is “human sciences” a recognized term? Otherwise I would state the specific fields (e.g. epidemiology, economics…).
Lines 473-476: Maybe I am being especially clueless, but this (important) message was not evident to me from the presentation of the RegIV and SEMIV results earlier in the text. In addition to expanding relevant sections in the text (e.g., Lesson 4), I would consider adding a section explicitly comparing the conclusions drawn from each method.
Line 534: Missing page number.
Line 560: I figure the very small values need to be converted to formal scientific notation?
Line 602-603: I think that brief explanation of the key interpretations/features would be helpful here rather than referring to the discussion. The caption for Table 1, for example, is excellent.