The setting nobody can explain

Every setting that is not self-evident was once a decision with a reason. The system keeps the result. It does not keep the reason. And without the reason, nobody can tell whether the setting is safe to change.

A checkbox with a history

A small online shop. Switzerland is switched off as a shipping destination. In the automation that sends invoices, one step waits two hours before anything happens. And the shop charges no VAT.

Someone set each of these on purpose. Maybe parcels to Switzerland came back with customs fees nobody had priced in. Maybe the payment provider confirmed some payments late, and invoices went out for orders that were never paid. Maybe the owner was under the small-business VAT threshold.

Maybe. That is the problem. A year later you are looking at three settings and not a single reason.

The reason goes first

A system stores the result of a decision on its own. It does not store the reason. That lives in the head of whoever made the call, and heads are poor storage: they forget, they change jobs, they are on holiday when something breaks.

The reason disappears fastest from exactly the settings that need it most. Nobody has to explain the obvious ones. The odd-looking ones almost always have a story, and after a year nobody remembers it.

Two ways to fail

Nobody dares

The first is quiet. Nobody knows what the two-hour wait is for, so nobody touches it. When something new is needed, it gets built around the old setting instead of through it. A second step, an exception, a rule that undoes the first rule.

That is how systems fossilise. Not because they were built badly, but because every unexplained setting is a small no-go zone, and eventually the system is more no-go zone than tool.

Somebody dares

The second is loud. Someone tidies up, decides the wait is pointless and deletes it. From the next day, invoices go out again for orders that will never be paid. It takes weeks to connect the two, because nothing sits between cause and effect to link them.

Both failures have the same cause. And both are reasonable, from where someone stands who does not know the reason.

Reasons expire

There is a third case, and it is the sneakiest: the setting was right, and it no longer is.

Take the German small-business rule. A business is exempt from charging VAT as long as its turnover did not exceed 25,000 euros the previous year and does not exceed 100,000 euros in the current one. The shop that charges no VAT knows nothing about any of this. It carries on regardless, including in the good year when the condition quietly stopped being true.

With the reason beside it, that looks different: “No VAT, small-business exemption. Check when annual turnover approaches 25,000 euros.” One sentence, and the setting carries its own condition. Whoever reads it knows not only why it is this way, but when it should stop being this way.

That holds for almost every decision. Switzerland stays off while shipping there costs more than it earns. The wait stays while the payment provider confirms late. A reason is not a justification forever. It describes the situation in which the decision was right.

One sentence, where the next person looks

What it takes is surprisingly little. One sentence per decision: what, why, and, if there is one, the condition under which it stops applying.

Where the sentence lives matters more than how it is worded. Nobody about to change a setting opens a document called “Shop settings, as of whenever”. The sentence belongs where you see it while you change things: the field’s description, the name of the automation step, the note next to it. A step called “Wait” explains nothing. A step called “Wait 2 h, payment provider confirms late” explains everything you need to know before you delete it.

The next person to look, by the way, is usually not a stranger. It is you, a year later, with about as much memory of it as a stranger would have.

The cheapest maintenance there is

At the moment you make a decision, its reason costs ten seconds. You know it already. A year later the same reason costs an afternoon of reconstruction, and often you do not get it back at all, only a plausible guess.

Hardly any other upkeep is this cheap. It needs no tool, no project and no budget, just the habit of writing one sentence while you tick the box. A system whose decisions you can read is a system you can change. One whose decisions you have to guess either never gets touched again or gets broken by accident.

Find the one setting in your shop or your most important automation that you would not dare to touch. Find out why it is the way it is, and write the reason right next to it. If nobody knows any more, that is an answer too: you now know where your system is fragile.