Importing Ten Years of Registers

The old files are messier than anyone remembers. Plan the import around that instead of pretending otherwise.

Every centre's archive has the same three problems: the same student appears under two spellings, a column changed meaning halfway through 2019, and one year is missing entirely because it was kept by somebody who has since left. None of these are discovered by looking at the first twenty rows.

Import the newest year first. It is the year people remember well enough to check, and the errors you find in it are the ones that will repeat backwards through the archive. Fix the mapping until that year is clean, then move one year earlier at a time.

Keep the original files untouched, side by side with the imported data, until a full cycle has run on the new system. Migration is the one operation where having two sources of truth is correct, and the moment to collapse them is after a real result day has passed, not before.