We are almost there!

""We found that all three types of magnitude comparison necessitated WM resources, albeit in distinct ways”

The "in distinct ways" get the unjustified claim of a difference back in, so should be removed. The following sentences in the discussion clarify appropriately.

"“Also, we observed different interference effects in processing and translating numerical representations for smaller (1 - 4) and larger (5 – 9) numerosities, which were examined separately.”

I see how you have emphasized the statement is based on separate tests; and thus the "different interference effects" claim is best read as literally a different pattern of significance and non-significance, which is true. However, an intuiive reading still remains that it is a claim that the effects were different, and this may occur because so many researchers do draw that illegitmate conclusion in their own work (understandably - the rules of asserting conclusions in hypothesis testing are paradoxical). So you could add at the end "so differences in the effects cannot be claimed"(unless you tested the differences in interfernce effects and found them significant).

best

Zoltan

The two reviewers are positive about the Stage 2, but make some important points. Reviewer 2 asks for some more clarity about the central executive; note that you cannot add the suggested extra sentences to the introduction, but there might be some way of addresing the point in the discussion itself. Reviewer 1 points out that the pre-registered protocol was not followed exactly with respect to the conditions you would use non-parametric tests. It is important to stick to the letter of the pre-registration. I checked your reported analyses against the Design Table and found other discerepancies:

p 28
"Finally, comparing the two interference conditions (PL and VSSP), we found no difference either for accuracy, t(80) = 0.58, p = .57, d = 0.06, or median RT, t (78) = 0.09, p = .93, d = .01."
and also
". Finally, performance in the two non-symbolic interference conditions (PL and VSSP) significantly differed, t(79) = 2.44, p = .02, d = 0.27, with participants performing better under PL than VSSP interference. Median RTs did not differ between these two conditions, t(78) = 1.48, p = .14, d = 0.17."
These are not in the Design Table; place in a separate exploratory section.

Two things to bear in mind in drawing conclusions: Only conclude there was a difference between conditions if the difference was tested (and only draw such conclusions i nthe abstract or give them importance in the discussion if they were pre-registered); and also do not assert there was no effect because of a non-significant result unless you calculated power with a justified minimal effect for precisely that test.
p 34 "We also found that mechanisms for processing and translating numerical representations can differ for smaller (1 - 4) and larger (5 – 9) numerosities."
As the difference between numerosities was not tested this conclusion does not follow.

p 35
"However, these speed-accuracy correlation coefficients did not differ significantly, which suggests that there was no significant change in participants’ speed-accuracy trade-off across the conditions. Thus, a speed-accuracy trade-off on its own cannot explain the finding of improved accuracy in interference conditions."
A non-significant result without a properly justified power analysis for precisely that test does not justify asserting no effect.

abstract
"albeit involving different components of WM, to a different extent."
Testing differences between dot and digit tasks, or between their respective dual task conditions, were not pre-registered; so claims about such differences should not appear in the abstract.

One further point of clarity|:

"Surprisingly, in this task, accuracy improved"

I look forward to receiving your revision!