五年分の記録を移すのは一つの判断です。

Every agency that changes system arrives at the same week. Five years of students sit in spreadsheets, folders, an old database and several people's inboxes, and somebody has to decide what comes across. It gets treated as a technical exercise with a deadline, and it is not. It is the moment the agency decides what it will be able to say about its own past.

The decisions are unavoidable, and they are made either deliberately in a meeting or accidentally by whoever writes the import. The only real choice is which.

What is actually being decided.

• Which students come across at all, and whether closed cases from four years ago are among them.
• What happens to a record that is incomplete: dropped, imported with gaps, or imported with a flag saying what is missing.
• Which of three conflicting spreadsheets is treated as correct when they disagree about a date or a fee.
• Whether documents come with the records, or stay in a folder nobody will open again.
• What the new system says a record's history is, when the old system recorded no history at all.

The third one is where most of the pain is. Conflicting sources are normal, and the instinct is to pick the newest file. The newest file is frequently the one somebody edited for a report, and the oldest is the one written at the time.

Do not silently improve the data.

The strongest temptation during an import is to tidy. Fill in the missing dates with something plausible. Set every case with no outcome to closed. Round a fee that looks wrong. Every one of those turns an honest gap into a confident falsehood, and the falsehood is indistinguishable from real data six months later.

Import the gap. A field that says nothing is a true statement about what the agency knows. A field filled with a guess is a false one, and it is the sort of thing that goes badly when a record is produced in support of something that matters.

Run it three times.

A migration that is run once, on the day, is a migration with no way back. The pattern that works is three passes, and the first two are the ones that find everything.

1. 01A sample pass of perhaps fifty records, chosen deliberately to include the messy ones, checked field by field by somebody who knows the cases.
2. 02A full dry run into a copy, followed by a count: how many students, how many applications, how many documents, compared against the counts in the old system.
3. 03The real run, with the old system kept readable rather than deleted, for at least a full cycle.

The counts in the second pass are the single most valuable thing in the process. A migration that loses four per cent of applications does not announce itself. It is discovered a year later by somebody looking for one student.

The parts that never import cleanly.

Notes.

Free text is where the actual history lives and it is the least structured thing in the old system. It is worth importing even when it lands in one long field, because a counsellor reading a case in two years needs what was said, not only what was recorded.

Money.

Fees, payments and commissions in a spreadsheet are usually missing a currency, a date, or a distinction between what was invoiced and what was received. Decide the rule before the import, not per record during it.

People.

Cases belong to counsellors who have left. Assign them to somebody real rather than to a blank, because an unassigned case is a case nobody will ever look at again.

What to write down.

One document, written during the import rather than after: what was brought across, what was excluded and on what basis, which source won where sources disagreed, and what is known to be missing. Two pages is enough.

It is the document that answers the question every migration eventually produces, which is why the old figure and the new figure are different. Without it, nobody can tell whether the answer is a bug or a decision.