A paper is done when the reviewers are satisfied. A system is done when it survives the customer who does the one thing you told them never to do. Those are very different finish lines.
The gap nobody warns you about
Research optimises for novelty and defensibility. Production optimises for the boring 80% — retries, migrations, on-call, invoices that must reconcile to the cent. CYCLOPS was a clean idea in a paper; turning it into a billing system meant confronting every unglamorous edge the paper got to ignore.
Draft — to finish
*(Outline: the three things that didn't survive contact with real usage; why "metering" is harder than "billing"; the moment I realised the research was the easy part.)*