[d110af5ccf]

> As a practicing scientist

> version-controlled Jupyter notebooks

That's awfully field specific. It probably wouldn't work for most of STEM. Even for ML I shudder to imagine trying to make sense of the inevitable monstrosities. Writing a paper is part of the thinking process. It forces the author to sit down and work through things in an orderly manner and they're still often difficult to read.

I'm definitely in favor of all papers being accompanied by working source code when relevant though.

[BeetleB]

> It forces the author to sit down and work through things in an orderly manner and they're still often difficult to read.

As a former academian: Papers are difficult to read primarily because the academic community does not value making them easier to read - no other reason. You may hear things like "papers should be written for other experts", but even that doesn't hold up to scrutiny.

They typically spend 99% of their research time on the actual research, and less than 1% on writing the paper. They definitely can afford the time and energy to make the papers easier to read, but game theory holds sway: Why should a particular researcher use his/her time to do it, when his/her peers will not appreciate it? It's purely an internal, cultural problem. There are no external constraints leading to this.

I've seen referees send papers back saying they contained too much explanation, and suggest leaving out most of the details - just include the big picture methodology and show the results. I can guarantee most who will read the paper would not be able to reproduce those details, if they ever want to. Likewise, I've found papers where I couldn't reproduce the results, because the results were wrong - but since including a derivation of your final expressions is discouraged, no referee caught the errors.

[seoaeu]

> They typically spend 99% of their research time on the actual research, and less than 1%

This isn’t anywhere close to my experience. 2-3 days of writing per year seems like a wild underestimate for any academic I know. I’d maybe believe only 20% of time spent writing, but for some folks even that’s probably way too low

[BeetleB]

Are they writing papers or grant applications? I'm speaking specifically about papers.

And fine, even if 1% is an exaggeration, I doubt they spend more than 5% (about 18 days of the year).

[ciphol]

As a former academian, I mostly disagree. In my experience, papers are difficult to read because they are as concise as possible, striving to refer to previous work for anything that's not original, and only elaborating on original things. This is done to make them as quick as possible to read by experts, which is pretty important given the immense volume of papers that appear in many fields.

I do think papers nowadays need to include a link to a zip file (or whatever other format - but it should be a boring old format unlikely to change or be abandoned, and not proprietary either) including all data, code, and so on. This data is necessary to verify the paper's results, but it is not the results themselves.

[BeetleB]

> I mostly disagree. In my experience, papers are difficult to read because they are as concise as possible, striving to refer to previous work for anything that's not original, and only elaborating on original things. This is done to make them as quick as possible to read by experts, which is pretty important given the immense volume of papers that appear in many fields.

This is consistent with what I said: Papers are difficult to read by choice. Yes, they strive to make it as concise as possible, which translates to making them harder to read.

Where I would disagree is the claim that it is done to make it as quick as possible to read by experts. In my field, the experts would skim papers quickly to get an idea, but if they then honed in on a paper to actually extract the meat of it so they can use it in their own work, it would take a lot of time, and was a pain.

I've heard from math professors that it takes about a day to read and digest one page of a journal article in their field.

Also disagree on it merely being a matter of consulting references. My work was theoretical/computational. It was common to see the final equations without any derivation, and many experts would not be able to reproduce it. There are lots of tricks that go into the derivation, but they are not provided under the pretext that any expert should be able to solve the equations and derive them.

And in the day of digital media, it's quite trivial to write a paper the way you suggest, and then put all the extra details in appendices. I guarantee that they will be read by most people who want to read the paper in detail, as opposed to merely skimming it.

[TuringTest]

Explorable explanations[1] is what you want, not Jupyter notebooks.

Explorables have all the same requirements to carefully think them through and prepare them for reading and clarity of exposition, but they also have the interactions that ease the introduction of concepts to their readers through a hands-on approach (rather than forcing them to read the mind of the writer by reverse-engineering their thought process, by running in your short-term memory the examples given in a non-interactive paper).

[1] https://explorabl.es/

[convolvatron]

I'm pretty sure all of us have the experience of cramming most of the paper into the two weeks before the deadline. Expecting someone to firm up their presentation without the risk of finalizing something half-assed is ... too much to ask for mere mortals.