Read more notes

← All journals
August 2, 2026

Your software's real test is year two

A school’s fee book is per academic year: every grade, every fee, an amount. Our pilot school entered theirs for 2026-2027 and it works beautifully.

Next August, without a copy-forward tool, they retype seventy-odd rows by hand.

The thing that unsettles me is that year one is the only year that hides this. Every annual artifact — fee books, term dates, class lists, timetables, promotions — gets entered once during setup, when everyone is motivated, the vendor is watching, and it feels like progress. Year two is quiet, unattended, and under time pressure, with a bursar who needs to bill by Friday.

The sharp version: a system with no rollover punishes loyalty. The longer a school stays, the more retyping they do. That is exactly backwards from how the relationship should feel, and nobody complains about it in year one because in year one it doesn’t exist.

It also can’t be caught by the usual instruments. It doesn’t show up in tests, because tests create their own data. It doesn’t show up in a demo, because a demo is always year one. It doesn’t even show up in a pilot, unless the pilot runs long enough to cross a boundary. It shows up eleven months later, in someone else’s busy week, as a quiet decision that this software makes their job harder than the spreadsheet did.

This is second-order thinking with a calendar attached. The usual version asks and then what? — you ship the feature, and then what happens because of it. Mine has to ask and then what, next August, when I’m not in the room?

Software rewards the first order relentlessly. Demos, launches, a finished sprint — every instrument we have measures the moment a thing starts working. The second order is silent by construction: it arrives on a cycle nobody is watching, to a person who never met you.

So the question I want to ask of every feature now: what does this look like the second time? Not the second click — the second year.