Chesterton's fence has a maintenance schedule

The parable is a century old and still runs most modernization meetings from the back of the room. Chesterton's fence: do not remove a fence until you know why it was put up. In its original 1929 form it was an argument about reformers; in its engineering adoption it has become the standing defense of every inexplicable check, every 2 a.m. batch job, every firewall rule named TEMP_DO_NOT_DELETE. Somebody, the argument goes, put that there for a reason. Find the reason first.

The argument is correct. That is what makes its misuse expensive.

Notice what the parable actually demands: investigation, followed by a decision. Notice what its enterprise deployment usually delivers: neither. The fence question gets asked rhetorically, in a tone that means "this is risky and I do not wish to own it", the meeting moves on, and the fence acquires another year of tenure. Multiply by a few hundred fences and a decade, and you arrive at estates where a material fraction of the operational budget maintains artifacts nobody can explain, defended by a philosopher who was, on a fair reading, demanding explanations.

The failure is institutional, not intellectual, and it has a precise shape: the fence has an advocate and the investigation does not. Removing the fence risks an incident with a name attached; investigating it costs real hours with no deliverable; leaving it costs a small sum forever, billed to no one in particular. Every incentive points the same way. Organizations do not keep inexplicable fences because they are cautious. They keep them because caution is free at the point of use.

So: give the investigation an owner and the fence an expiry. The mechanics are not complicated, and I have watched versions of them work at three organizations of respectable size.

Every fence gets a card. When a check, job, rule or workaround resists explanation for more than a quarter, it enters a register: what it is, what it touches, who last claimed to understand it, and the date of that claim. This is an afternoon of clerical work per subsystem, and unlike most documentation it appreciates, because the register is a queue.

The queue gets a cadence. Some fixed, modest capacity, one investigation per team per month is enough, works the register oldest-first. An investigation ends in one of exactly three verdicts, written down. The reason exists and still applies: the fence stays, now with its reason attached, and the next decade of meetings gets shorter. The reason existed and has lapsed: the fence is scheduled for removal, behind a flag, with a rollback, during business hours, per all the ordinary hygiene. Or the reason cannot be determined at acceptable cost: the fence stays, but it is now listed as unexplained risk, which is a different thing from heritage, and it gets a monitoring hook so that the day it finally does something, someone learns what.

The third verdict matters most. It converts "we are afraid of this" from an ambient mood into a line item, and line items, unlike moods, can be budgeted against. A board that would never fund "explain old things" will fund the reduction of a numbered register with a dollar figure of annual carrying cost next to it. The framing is the product.

There is a deeper reason to institutionalize this, and it is about time rather than money. The fence's original reason decays in a specific, predictable way: it lives first in a person, then in a team's folklore, then in nothing. The window in which the investigation is cheap, one conversation with the right veteran, closes on the day that veteran leaves, after which the same answer costs a forensic reconstruction, if it can be had at all. An estate's unexplained artifacts are therefore cheapest to explain now, whatever now is, and the register's oldest-first discipline is really a race against retirements. I have watched an organization pay a contractor six weeks to reverse-engineer a clearing-file quirk that a recently departed employee could have explained over coffee. The fence had stood, unquestioned and politely defended, for the entire nine years she sat beside it.

None of this licenses the reformer Chesterton was actually writing about, the one who arrives sure that everything old is debris. Half the register's investigations end in verdict one, the fence earning its keep, and those are successes: a documented reason is an asset, and the check that survives its audit is stronger for it, politically as well as technically. The point was never that fences are bad. The point is that "someone had a reason" is the beginning of a process, and the industry has been using it as the end of one.

The gate is for understanding, Chesterton said, not for standing in front of forever. Put the understanding on a schedule, or the schedule will be set by the leaving dates of the people who remember, and their notice periods are shorter than your backlog.