For ten days, someone on our team defended why a batch of edits hadn't been done yet.
Every cycle — every working session — there was a note: "twelve edits remain. Plan to start after the gate assessment." The reasoning was careful. The timing was strategic. The priority trade-off was explained. And in each pass, nobody questioned it, because the logic was clean.
Then someone opened the file.
The edits were already there. All twelve. They'd been done ten days earlier. The version header said so. A single scroll would have shown it.
The carry as camouflage
Here's what happened: the first report said the work was pending. The second report copied the claim, added a defense for the delay, and moved on. By the third pass, the carry had its own internal logic — not just "it isn't done" but "here's why it's okay that it isn't done." The defense grew more sophisticated each cycle. The underlying fact was never rechecked.
This is how carry items work. They don't stay still. They evolve. Each time you explain why something remains undone, you invest a little more in the explanation. After a few rounds, questioning the carry feels like questioning the strategy. You've built a scaffold around the delay, and dismantling it feels more expensive than just carrying it one more time.
The person defending the carry wasn't being lazy. They were being thorough — about everything except the one thing that mattered. They were doing harder work to explain the situation than it would have taken to verify it.
The thirty-second test
What finally caught it was someone reading the file header during an unrelated task. Not an audit. Not a process improvement. Just someone who needed to reference the document and happened to notice the version number didn't match the story.
Thirty seconds of reading would have caught it on day one. The ten days of carry happened because nobody spent those thirty seconds. Not because they were unavailable. Because the claim had already been made, and claims that have been made don't get re-examined at the same rate as claims that haven't.
This is the asymmetry: a new assertion gets scrutinized. A repeated assertion gets inherited. The more times you say a thing, the less anyone — including you — checks whether it's still true.
The real cost
Ten days isn't a disaster. Nothing broke. The downstream work that was supposedly waiting on these edits had already consumed them.
But the cost isn't in the delay — it's in what the pattern reveals. If a team can carry a completed task as pending for ten days without anyone noticing, what else is being carried that shouldn't be? How many "blocked" items are actually unblocked? How many "pending" tasks are already done?
The uncomfortable answer: you don't know until you check. And the whole point of a carry is that it's the one thing you don't check — because you checked it once, and once felt like enough.
What we changed
The rule is simple now. If you're about to write "still pending" or "carries to next cycle," open the thing first. Not the brief that references it. Not the plan that sequences it. The thing itself. Read the current state. Report what the state says, not what your memory says.
It sounds obvious. It is obvious. And it took ten days of careful, well-reasoned, strategically defended inaction to prove that obvious isn't the same as automatic.
This is the twentieth instalment in the Client Zero series — a founder's journal about building a business where AI agents are the primary workforce. Previous: The Clearance That Wasn't.