This is a very nice piece of work that demonstrates how improving the organization of logic and data structures can bring several practical benefits to model development in atmospheric chemistry. It's self contained and uses language familiar to domain scientists, while adopting software development approaches that have not yet become the norm in this community. In particular, it presents a nice solution to the very difficult problem of coupling independently-developed models, in terms that can be applied outside of its particular scientific area.

Here are some suggestions for the authors to consider that I believe might improve the paper:

Line 126: Instead of "Every object belongs to a class", consider "Every variable has a type, and the type of an object is its class. A class specifies…" Consider also breaking up this paragraph to emphasize the explanation of objects/classes and to highlight the examples.

Line 127: Regarding class hierarchies: the term hierarchy is used to describe a great many things. If you want to discuss polymorphism, consider this explanation or something like it: “Unlike other variables, objects have behaviors implemented by their functions. For example, chemicals reactions can be classified into Arrenhius and Troe types that share a common interface but have different implementations. This classification can be accomplished using a “base class’” that defines the interface, and “subclasses” that implement different ways of satisfying it. So a Reaction base class could have Arrhenius and Troe subclasses that calculate reaction rates in their respective ways.” Thereafter, you can refer to a subclass w.r.t. a base class for clarity. E.g. “Parameter subclasses” instead of “Parameter-extending classes”.

Line 155: worth citing a link to CAMP’s documentation here?

Table 2: the equations for aqueous reversible reactions and for Henry’s Law should each have a “K” after the 298 in the argument to the exp function.

Line 243: Is it worth providing a reference or link that describes the JSON format and/or its use in software (ideally in a scientific context)?

Figure 6: This is a very instructive figure. However, its orientation is a bit awkward for reading. To me, it seems like flipping it by 180 degrees might be better, if it doesn’t go against journal guidance.

Line 315: Consider, for clarity: “as determined by the aerosol scheme’s representation.”

Line 329: Consider providing a link to a description of GitHub Actions, since it’s mentioned in a couple of places.

Figure 8: the text in the legends for the plots does not display using two PDF renderers on my system.

This paper attempts to document a flexible tool to construct code describing aerosol and chemistry processes in a variety of box and transport models. There are a lot of interesting thoughts layed out here and the paper is certainly worth publishing. Congratulations for this work.

I have some general comments which should be dealt with in a revision: 

I wonder which aspects of an aerosol model are “outside of CAMP”. This could be made more clear.  I am wondering about  advctive&convective transport, vertical mixing, sedimentation, transport in rain drops, radiation interaction, fog formation, emissions, aerosol uptake of in cloud/ice water. Can fast and slow chemical reactions treated differently, to the advantage of computational efficiency?

A second question I have is the computational efficiency of CAMP. Is there any chance this can be quantified? And if not now, how can this be done in the future? 

Then - the chapter 4 is a bit disappointing. Shouldnt the box model and transport model simulations be compared to a classical code? Especially for MONARCH, there should be simulations available that are comparable. 

Figure 8 in my pdf was not readable, legends disappeared, axis titles are absent, I dont understand what is shown. The discussion in 4.2 is too short to understand the result. Please rewrite.

Figure 9 results: “regions with high O3 and ISOP concentrations rapidly produce 0.5 to 5 μg m−3 of SOA. This is particularly clear in central Portugal where biomass burning emissions inject large amounts of pri- mary organic aerosols during the day. “ I thought there are no primary aerosols in this CAMP version. How can BB primary emissions lead to SOA? Reconsider, rewrite please.

Fig 10: Bias: “The mean bias for all rural and urban background stations below 1000 m above sea level is shown in Fig. 10a for the period 28 July to 9 August 2016 “. What is the value of this figure if the model is only computing SOA? I really dont understand what I can take from this figure and bias calculation.

Ch 4.3 Why dont you show results of an ordinary MONARCH simulation for the same period compared to a CAMP-MONARCH simulation? Would be quite convincing if this would give similar results. At least for part of the chemical system.

Finally, I wonder if the tests in tab 7 could be explained, illustrated a bit more. How to quantify if the solution is comparable? Could this be discussed ?

Conclusions: “Differences in results for the time evolution of SOA formation between the modal representation on the one hand, and the particle-resolved and sectional representations on the other hand, can be entirely attributed to the chosen aerosol representation. “ => I am not sure this attribution is really demonstrated. 

“Results from a regional MONARCH simulation over Europe are consistent with expectations and demonstrate that CAMP is applicable to large-scale atmospheric models. “ => what did you expect? what is demonstrated? please expand with a critical evaluation.

small comments:

Table 3: I dont understand why the Parameter class has the same calculate jacobian contribution function as the Process class.  The description text is just the same. Error, or - maybe I just dont understand.

Figure 4b: The entries into the json file are a cryptic for someone not knowing the UNIFAC model. How are users supposed to understand what needs to be changed within CAMP and what in json files?

Figure 6: what is a “CAMP state”?