There's a particular kind of afternoon every writer of research-heavy fiction knows well: you sit down to write a single paragraph about how a Roman quartermaster might record a grain shipment, and four hours later you've read an exceedingly specific journal article, made half a page of notes you'll never use, and written no paragraph at all.
I don't think this is wasted time, exactly. But I've learned to distrust it while it's happening, because it feels productive in a way that actual drafting often doesn't. Research has a beginning and clear little victories — you find the fact, you understand the system, you close the tab with a small hit of satisfaction. A scene doesn't work like that. A scene resists you for days sometimes before it's right, and research is a very comfortable place to hide from that resistance.
The test I use now, imperfectly, is this: would this fact change a decision a character makes, or the way a reader understands the stakes? If yes, it's worth chasing down, however deep the hole goes — this is exactly how the administrative detail behind Roman land surveys ended up as the entire engine of Measured Ground, not a footnote to it. If the honest answer is "no, I just find this interesting," I write it on a separate note for later — sometimes for a blog post, as it happens — and get back to the page.
The uncomfortable truth is that readers will forgive an author who lightly under-researches a period far more readily than they'll forgive a scene that drags because the author fell in love with their own research. Authenticity earns trust. Showing your working doesn't. The goal was never to prove I know things — it was to build a world solid enough that a reader stops noticing it's built at all.