The manuscript analyses the relevance of seed limitation on community richness and biomass under a combination of three different treatments (herbivory exclusion, warming and fertilization) and two species’ origins (local or novel species). Through this approach, the authors aim to disentangle part of the context dependency affecting seed limitation effects on community composition and function that is not frequently addressed in similar studies. I guess that considering the effect of biotic and abiotic conditions on seed limitation is a particularly relevant question and the authors implemented an accurate and well-developed experiment for this purpose. Because of the interest of the topic and the accuracy of the experiment I think that it might be of interest for the audience of Ecology Letters. Nevertheless, I have a few concerns about content and results that should be addressed before recommending its publication.
Major comments:
Introduction. Lines 84-94. Throughout this paragraph and part of the rest of the introduction, it emerges that community niches saturation and saturation of diversity-productivity curves are the main underlaying drivers determining the effect of seed limitation on community richness and biomass. Despite it is already well explained I would suggest using these saturations as a common thread throughout the paragraph (and other parts of the introduction if needed) to make the content even clearer for the readers.
Methods. The experiment started in august 2009 and the seed addition was carried out in 2009 and 2010. Do the authors think that the fact of adding the seeds short after the beginning of the experiment could affect the results? I’m wondering if by adding the seeds later, once the local grass communities were adapted to the relaxation of herbivory, temperature and nutrient limitations, the seed establishment and biomass change would have been lower or different than already seen. In fact, adding seeds later might be more coherent according to the rational of introduction and hypothesis 1.
Results. In general, I’ve found the results a bit difficult to follow. There are multiple treatment and interactions between them and the way the results are presented does not simplify too much the understanding. Although showing the results of ANOVAs in the supplementary material is useful, I think that results will improve the readability in case of including post hoc contrast tests, so in addition of knowing the significance of each single variable the readers could read in a supplementary table if the effect of a treatment is positive or negative and with which magnitude. Contrast function from R package emmeans (Lenth 2021) is particularly useful for that. In addition, it allows to compare a given treatment in relation to other treatments, which could be particularly helpful in this study.
Discussion. In general, I missed a better linking of the results with the objectives and hypotheses. Particularly as it seems that results contradict hypothesis 1 in some cases, as higher relevance of seed limitation was expected under control conditions than under rest of treatments.

Minor comments:
Lines 88. It could be deduced that environmental limiting conditions always led to unsaturated communities with higher niche availability. However, harsh environmental conditions might, in fact, limit seed establishment because of the reduced number of available niches in these conditions (e.g., drought conditions in Myers & Harms 2011). If possible, you I suggest mentioning briefly the effect of environmental filtering on limiting seed establishment.
Lines 116. Species invasions do not necessarily lead to reduction in local species richness.
Lines 142. If possible, also add information on total annual precipitation. In addition, it is a good practice to reference the time period to which the climate is referred, particularly in a context of ongoing climate change.
Lines 146. How far are plots from each other?
Lines 148. Specify if control plots are included within the total number of treatment combinations.
Lines 154-157. It would be useful to add some pictures of the experiment in the supplementary material.
Lines 163. It doesn’t seem really useful to know how much the OTC increased the temperature during August since according to the previous explanation the chamber was removed to allow grazing between July and August.
Lines 169-171. Maybe the absence of a consistent decrease in richness in the warming experiment is not enough to justify the lack of effect of chambers sides on natural colonization. It might happen that chambers’ sides reduce the natural colonization, but this lower colonization is compensated with a higher establishment of those arriving seeds due to the warmer conditions. Thus, in absence of the chambers effect on colonization the warming experiment would show higher richness values than already observed. If you consider this is possible and natural colonization is important, it might worth mentioning that limitation in the discussion.
Lines 177-180. One thing that could potentially affect the richness results depending on species origin is the novel:local seed ratio used to sow the plots. For example, if novel species are clearly more abundant in the seed mix it would be logical to find higher proportion of novel species among the new established. Globally this ratio seems to be fairly even as it is 14:11 (or 9:11 in case of discarding novel species which never established). However, it is not clear if this ratio keeps similar across plots. If local seeds are redundant with plot species, it would be expected that these seeds would not contribute to the increase of richness. I would suggest adding a table to supplementary material with the exact local:novel ratio in case you find it is highly variable across plots.
Lines 184. Do all the species have similar viability rates? if the proportion of non-viable seeds is different across species this may imply differences in the likelihood of establishment not due to the suitability of treatments.
Lines 206. Did the authors missed to clarify that after summing the values they divide the value by the number of species to obtain the mean?
Lines 210. Please specify that each of the explanatory variables (fertilization, warming and herbivory treatment) was binary.
Lines 214. Explain before why and how you applied a PCA. If possible, also show the PCA axes and species shorting in the PCA space (colored by species origin: local, resident, novel) in the supplementary material.
Lines 214. Supplementary Table S1 does not show subplots nested within plots as random effects.
Lines 215. Please specify if transformed variables were always response variables.
Lines 216. Supplementary Table S3 does not show response variables transformations.
Lines 221-223. I’ve found this sentence a bit incomplete. Do the authors mean that they used the change measurement instead of using other measure because of novel species? Which other measurement did the authors discard?
Lines 225. Supplementary Material Table S4 shows glm with poisson distribution instead of negative binomial. In addition, the code stile line summarizing the model use “lm” instead of “glm”.
Lines 352. According to Figure 1 warming_x_herbivory treatment did not increase the abundance particularly in case of novel species. It couldn’t be deduced that warming conditions contribute to compensate the growth of novel species after grazing.
Figure 1. Why do the authors show panel b and d by separating herbivory and fence treatment instead of showing herbivory treatments (H, HxW, HxF and HxFxW) as four independent bars. I think that this last option would be clearer for this first figure.
Figure 2 and 3. Why do the authors follow different representation styles for similar figures? I understand that Figure 1 requires using bars to show the different proportion of each species origin. But in Fig 2 and 3, why not always using bars or point range styles?

Bibliography
Lenth, R. V. (2021). emmeans: Estimated Marginal Means, aka Least-Squares Means. R package version 1.7.1-1.
Myers, J.A. & Harms, K.E. (2011). Seed arrival and ecological filters interact to assemble high-diversity plant communities. Ecology, 92, 676–686.

The authors have satisfactorily solved all my major concerns. I sincerely appreciate the effort the authors have made addressing all the comments and I think that they have improved the clarity and cohesion throughout the manuscript. It is also a well written piece and carefully designed experiment.
However, after reading again the paper I have one more comment. Given the hypotheses, it is not only important to know that seed and unseed communities have different richness and biomass, but also to know under which treatments these differences are higher. However, Figure 1b-d and 2, do not allow to easily compare the magnitude of the differences between seed and unseed plots across treatments (eg. It is not easy to check if difference in biomass between seed-unseed in WxF treatment is higher than control one, and this information is relevant to respond to the hypotheses, as we expect this difference to be higher in control). If possible, I would suggest adding an extra figure equivalent to figure 1 or 3 with difference in biomass/richness in the “y” axis and “x” axis containing each treatment and control for all species. It can be a Supplementary Material figure. Anyway, I’m suggesting that for extra clarifications, and I won’t be really strict on it, in case authors do not consider it necessary. Post-hoc tests would add similar information by adding the difference in estimates between seed-unseed within treatment, but I agree with the authors that the statistical approach is not identical as ANOVA, so I agree with not using them if they lead to results differences.
In addition, I have one extra doubt. If I'm not wrong, Figure 1 shows that the 95% confidence intervals of seed and unseed communities for Warming x Fertilization x Exclosure treatment do not overlap, suggesting that WxFxE treatment might be significant. However, according to Table S3b this triple interaction (FxWxExS) does not emerges as significant. Maybe is something related to mixed models or differences in dataset used for representation, but it will be good to check it.
After checking these comments, I’m happy to recommend the paper for publication.