What would you do if, halfway through delivery, you discovered that a requirement everyone agreed to was based on a factually incorrect assumption about how the business operates?

1. Verify it first. I would confirm the actual behaviour with the people who perform the process and with data from the org before raising an alarm.
2. Assess the blast radius: which stories depend on this requirement, what has already been built, what is in the current sprint, and what the cost of the change is now versus at go-live.
3. Raise it immediately with the Product Owner with the analysis attached. Discovering a bad assumption in sprint four is inconvenient; discovering it in UAT is expensive; discovering it in production is damaging.
4. Present options rather than just the problem: correct it now, deliver as designed and correct in a later release, or reduce scope. Each with a cost and a risk.
5. Finally, I would look at why the assumption survived validation — usually it means a playback was skipped or the wrong person confirmed it — and adjust the validation approach for the rest of the project.