A core banking migration is the closest thing in software to performing surgery on a marathon runner mid-race. The patient must keep moving; the surgeon cannot pause. Done well, nobody outside the engineering team notices anything happened. Done badly, you make the front page for the wrong reasons.
The strangler fig pattern, but realer
Everyone cites Martin Fowler's strangler fig pattern. Few teams actually run it on a live ledger. The reality: you build a routing layer in front of the monolith, gradually peel off bounded contexts, and run shadow traffic for weeks before you flip a single read. The interesting work is in the reconciliation jobs — the ones that compare old-system and new-system totals every five minutes and refuse to let you flip if they ever drift by even one cent.
What kills migrations
Not the tech. Politics. The team that built the legacy system has institutional power, and migration threatens that power. The teams that ship are the ones whose executive sponsor is willing to spend that political capital. We've now seen enough of these to predict success or failure within the first two weeks based on the kickoff meeting alone.
