8 · SILENT SCOPE DRIFT

THE CITY

Nobody designed this city. That is not an insult — every individual piece of it was designed, carefully, by someone with a good reason, on a particular Tuesday. It is the sum that nobody designed. There was never a person, at any moment, holding the whole thing in view and saying yes to it.

So you get a city with two finished buildings that have no door onto any street. Both were built to a real plan, both are sound, both are furnished — and no road reaches either one. You get fourteen notices nailed to the main gate, warnings posted by outside authorities, none of which anyone has read closely enough to say whether they describe this city or some other city. And you get four foundation stones, quarried elsewhere and load-bearing under everything, now one and two generations behind the current standard, with no date written anywhere for when they are to be replaced.

Ask the mayor what the city will look like in a year and they cannot tell you. Not because they are evasive — because it has not been decided. It is accumulating.

Here is the trap, and it is the reason this chapter is the hardest one to act on. When you finally walk the city with fresh eyes, the impulse that arrives is demolition. Buildings with no doors, notices nobody read, old stones — clear it out. Resist that. Some of the emptiest-looking buildings in the first city were empty on purpose, and finding out which is a different job from noticing they looked empty.

THE TELL

• You discover a finished feature you had forgotten you owned.
• You cannot say what your product will be in a year, because nobody decided — it is just what the last twelve months added up to.
• Your list of things you deliberately chose not to build does not exist.
• You have a pile of warnings and a verdict on none of them.
• Nothing in your city has a date for when it must be replaced.

FROM THE FIELD

Two buildings, no doors

The first city had a page built to a documented plan, working, and reachable by nobody — nothing anywhere in the city linked to it. The survey's own wording of the verdict is careful, and the care is the point:

"Either a resident cannot reach a built feature, or it was superseded and never removed" (Engagement 01, F-020).

Two possibilities. Very different meanings. Opposite remedies. The survey wrote both down and ordered neither.

The second one turned up in a later district sweep — another built-and-guarded page with zero references anywhere in the city. The note attached to it is a lesson about the limits of looking: "grep-proof, not call-graph-proof" (Engagement 01, F-022). In plain English: we searched by name and found nothing, which is not the same as proving nothing reaches it. The finding was recorded at exactly the strength of the evidence actually held, and no stronger.

Why the survey did not order them torn down

This is the discipline, and it exists because the first city taught it the hard way — three separate times.

One of its scripts looks exactly like six duplicates the city had already deleted for being out of law. It is not one of them. It is deliberately kept, carries a written notice on its own face saying it is on the way out, and is still used by a live screen (the first city's ledger, DL-100). Anything pattern-matching duplicate, therefore delete sweeps it away and breaks a working surface.

One of its reading rooms is missing a safety wrapper you would expect to find there. That absence is load-bearing by written ruling (the first city's ledger, DL-103). The gap is the design. Add the obvious improvement and you have removed a control.

Twice over, the thing that looked like clutter was a decision. So the ordinance is: an unreachable building is a question, not a demolition order. Drift is resolved by deciding — and a deletion made on resemblance is not a decision. It is the same disease running backwards, adding unplanned removals to a city that already had unplanned additions. An owner who reads this chapter and spends a weekend deleting has caught the disease from the treatment.

Fourteen warnings, one that mattered

Fourteen advisories against the city's outside components — ten of them rated high — sat unassessed until the survey looked. Assessed, they decomposed to one genuinely reachable class of problem. The survey's line:

"an unassessed advisory list is worth less than an assessed one with the same numbers" (Engagement 01, F-018).

Fourteen unread warnings did not mean fourteen dangers — it meant no verdict, and no verdict is an unknown state dressed as a quiet one.

Four stones, no date

Four load-bearing outside components were running one to two generations behind current, and the finding is not that they were behind. Plenty of good cities run behind on purpose. The finding is six words:

"no upgrade horizon is recorded anywhere" (Engagement 01, F-006).

Nobody had decided to stay behind. Nobody had decided to catch up. There was no date, no trigger, and no written reason — so the choice was being made by default, every day, by nobody.

WHY AI WORKERS DO THIS

Every shift builds what it was asked for, and no shift removes what the last one built. That is the entire mechanism. Drift is not an error committed by any single worker — it is the arithmetic of many correct shifts with no editor.

The worker cannot see the sum. Asked to add a screen, it adds an excellent screen; it will not mention that the screen it added last month is now unreachable, because it has no memory of last month and no view of the whole. And a worker never volunteers a refusal or a removal — those come from someone holding the entire city in mind and deciding it should be smaller. There is no such role in an AI shift. So the city can only accrete, indefinitely, one reasonable addition at a time, until it is something nobody chose.

THE COUNTERMEASURE

The cure for drift is not tidying. It is a scheduled act of deciding, performed by somebody whose only job is the sum.

Give your workers this, verbatim:

Twice a year this city is read as a whole, and every load-bearing thing we did not build gets a written replacement date. Every outstanding warning gets a verdict — reachable or not, and by what path — because an unread warning list is not a risk register, it is a rumour. Anything that looks abandoned is asked about before it is touched, and the answer is written down beside it. Nothing in this city is ever removed on resemblance.

One item on that list is yours alone, because no worker can write it for you: the record of what you deliberately refused. A city with no written refusals has never actually chosen anything — it has only accepted things in sequence.

This slows drift. It does not stop it. Every living city drifts; the difference between a healthy one and a condemned one is whether the drift gets read and ruled on twice a year, or discovered in year four by a stranger.

ASK YOUR OWN CITY

1. Where is your written list of things you deliberately did not build, and why? If there is none, what does that tell you about the things you did build?
2. What did your city look like eighteen months ago? Was the change between then and now designed, or did it accumulate?
3. Name one thing in your product that nobody can explain the purpose of. Ask what it was for, and write that answer down before anyone touches it.
4. What are you standing on today that will be unsupported within three years? Name the date you plan to move off it.
5. Walk your own city as a stranger. Are there promises still visible that have not moved in a year — a "coming soon," a beta badge, a placeholder?