let's try and mkae the limitatons not too extensive. LLMs (when reviewing) actually pick up on these and present them as independent evidence that something's wrong with the paper.

nice idea! But keep for ourselves for now.

well done. I assume energy has such a big impact bc the agent no longer can tell when it runs dry or is full of energy? so basically doing 'stupid' mistakes? It would be good to give a concrete reasoning like this, even if speculative.

it would be better ot compare against one of our own models (otherwise too confusing who said what and why)

maybe clear from my cokments above but no need ot argue here against Ott et al. This is a standalone paper. Rather derive directly from our motivation.

before going in these results, we should motivate this, i.e. saying what we'll learn from this analysis. Otherise the following description will just appear confusing and will draw comments by reviewers.

I'm not partiiculary a friend of the terms 'extreme' and 'intermediate', so we might drop them if not required

this will be too confusing for 1st time readers. I'd structure it like this: We just define our own models (even if they are the same in Ott et al), e.g. planning-only, preference-only, full model. Think of names that make sense for us. Then analyse all of them. And only mention then the hybrid model.

This would be the cleanest way of presenting the results. And we also refer to Ott et al 2022 so no one will accuse us of not mentioning prior work.

put this before 2.5 RTs small changes to the first paragraph: go quickly for the kill, no reference to Ott necessary here. As we will say in the intro that task structure may be used to form cheap priors, so ask there the question what are those in the present task. List them (briefly motivate each one of them why they give prior info). And then first say in a sentence or two what kind of analyses we're proposing and only then go through each analysis (as you do already now)

better: formulate hypotheses in the intro, and refer back to them. ('Predictions' sound too grand. )

there won't be the need to explain in detail what Ott found before. It should be in there but only 1 or 2 sentences and say that there was already some evidence found.

The RT analysis would be a cornerstore, and we need to refer back briefly to our narrative of the intro to remind readers of the big picture

I assume this would be the appendices in the paper?

add motivation: bc we wanted to compare models

good. it'd be even better to have some prior motivation/hypothesis/conjecture that we can build on. And this motivation would be laid out in the introduction and only referred here to.

replace 'observed' by 'found' and now is a good time to cite Ott et al 2022

nice.... if this result has been used before though, it's better for a smooth review to say that we followed their analysis.

the question is also why not model all 240 (?) trials, why we shortened (suboptimally) to 13 trials. Explain that it was our assumption that this is sufficient.

Before going here, we also need to say how many trials there were and that the task was 'continuous' so doesnt end after 20 trials or so but goes much further.

to make sure we both share the same understanding: A paper should usually be standalone so ideally you would put here all the info about the experiment and not assume that readers would read Ott et al 2022 to find out.

no need to say this here again. But if the figure was taken from Florians paper, we must get approval of the journal and say that we did, usually at the end of the figure caption. if we changed somehting, the text at the end would be 'Figure adapted from Ott et al...' (or similar)

we should talk about making Florian Ott a co-author. He provided the data, wrote the application with me, and one might even argue that the previous paper gave us the idea to think harder about what preferences subjects use. I'm neutral on this question of co-authorship but interested in an objective assessment of his contribution.

write ethics approval number

We don't need to say this here. From the point of the reader, it's quite irrelevant where the data came from.