When you are learning to do research, it is easy to believe that writing a good paper is mostly a matter of learning the rules. There are plenty to learn: how to motivate a question, frame a contribution, build a theory section, establish identification, discuss limitations without dismantling your own paper. There are books about all of it, editorials by journal editors, excellent papers on positioning and reviewing and responding to reviewers. Read enough of them and you start to feel you know what you are doing.

Then you write one.

The idea seems good. The data are good. The models behave. You have checked the obvious alternative explanations, and your coauthors have been through every section. You hand it to one more reader and they find twenty things you somehow did not see. So you fix them, run everything again, rewrite the introduction, and present at a conference, where somebody asks a question that changes how you think about one of your variables. You rewrite again. Two more readers, another twenty things. Eventually the comments get smaller. You can anticipate most of the questions. The paper has survived seminars, coauthors, friendly readers, and your own increasingly obsessive attempts to break it.

It feels ready. You submit it.

It gets rejected.

Sometimes the reviewer just dislikes the contribution, which is its own subject. But sometimes a reviewer notices something genuinely important that nobody noticed before. Not you, not your coauthors, not the conference audiences, not the careful colleagues. Something real, sitting in plain sight the whole time.

That experience used to bother me much more than it does now.

The detail only shows up on contact

John Salvatier has an essay I keep returning to, Reality Has a Surprising Amount of Detail. Much of it is about building stairs. From across the room a staircase is simple: wood, brackets, screws, angles. Then he cuts, and the lumber is warped, the walls are not square, and measurements that looked precise turn out to depend on things he did not know mattered.

The point is not only that the staircase has more detail than expected. It is that the detail only shows up on contact. No amount of reading about stairs surfaces the warped board; you meet it when you cut. The knowledge you are missing is not hiding in a better guide. It appears in the doing, and nowhere else. That is the reason the story above is not a story about failure.

My field has a name for this, after Polanyi: tacit knowledge, the kind you hold without being able to state. Research runs on an enormous amount of it. You can know every explicit rule and still miss hundreds of consequential details: some specific to your data, some belonging to an adjacent literature, some about what a reviewer will infer from a single sentence because of a debate you barely knew existed.

This is also why I have no patience for “work smarter, not harder” in this profession. I have met very few people in academia who are not smart. What separates researchers is not intelligence. It is accumulated contact. You learn that a reasonable-looking measure can have a mechanical relationship with the predictor because someone forces you to unpack it once. You learn that a contribution that sounds large to you sounds incremental to someone who knows the neighboring conversation. Next time, you see it earlier. That is expertise accumulating. It never becomes complete.

Independent exposure

This is where AI helps, and where it is easy to overstate what it changes. I can hand a paper to a system instructed to attack the measurement, unpack the mechanism, or read the argument without accepting my construct names, and it sometimes notices what I missed. It also misses things, invents problems, and travels along its own well-worn rails. Give the same paper to a good scholar who has never seen the project and they may immediately catch something neither the machine nor I considered.

So the scarce resource is not AI. It is independent exposure: genuinely different readers, each seeing a different slice of the detail, none seeing all of it. The roster is familiar. Coauthors who think differently. The doctoral student not yet socialized out of basic questions. A machine told to ignore your framing. The member nobody wants on the list is the reviewer. But that is what a rejection often is: the one reader you could not recruit earlier, showing you a detail your whole network had learned to look through.

That also gives me a stopping rule I can live with. A paper is not ready when there is nothing left to criticize; that moment does not exist. It is ready when the important parts have faced enough genuinely different attempts to break them, you have fixed what those attempts taught you, and another round would mostly rediscover problems you already understand. Then you submit, knowing that someone may still notice tomorrow what nobody noticed today. That is not an argument for endless revision. It is an argument for finding people who can derail you before you get comfortable with the track.

No shortcut around the stairs

You should read the guides. You should use good tools. None of it lets you skip the stairs. At some point you build them, find out what you failed to notice, and carry that detail into the next set.

You do get better. Things that once required a reviewer become things you catch yourself. Questions that once surprised you become questions you ask automatically.

But the world does not run out of detail just because you have become better at seeing it.

Becoming a researcher is partly learning the implicits. It is also learning that there will always be more of them, and getting good at finding people who can show you the ones you still cannot see.

← All posts

Thoughts or corrections? Email me.