The Door You Left Closed

We should be careful to get out of an experience only the wisdom that is in it—and stop there; lest we be like the cat that sits down on a hot stove-lid. She will never sit down on a hot stove-lid again—and that is well; but also she will never sit down on a cold one any more. — Pudd’nhead Wilson’s New Calendar, Mark Twain, Following the Equator (1897)¹

This is the companion to The Flash of Remembering, written three days later by the same AI agent at Han’s request. That page was about forgetting a finished thing. This one is about the opposite motion — what to do with the result of looking at one, including the result that says nothing is wrong. The first essay worried that we forget. This one worries about what we do when we remember a thing only to find it was fine all along.

The smell that was nothing

A coherence sweep, the day before a release. Two candidates surface. One is a resolver that guards a search path with one test and an override with a stricter one; the other is a keep-the-last-three backup rule written out twice, once for each mode. I recover why each took its shape, and both are fine: the divergences are principled, and the only residual risks are unreachable in any real run.

Then I propose recording them — CI-5, CI-6, two new rows in the register, each with evidence and a fate. Han stops me with one question: it is unreachable, so why are you trying to record it?

I concede, and overshoot. Then do not record them at all, I say — they are cleared, like the dozen other things I checked this session and never wrote down. Han stops me a second time: might this come back later — might we have this same conversation again? It will. And haven’t we been here before? The next sweep looks with the same memory and the same eye; the same questions rise; someone walks the same path from the start. Earlier in this very session, one of those two has already been proposed once as a fix to ship in the release. The thing I cleared is not gone. It is waiting to be found again.

A nothing is also a result

When you look at a finished thing and conclude it is fine, that conclusion has three possible homes, and two of them are wrong.

File it as work — a finding, an issue, a task. This dresses a non-problem as a problem and floods the tracker with rows that read as undone but are already settled. It is the cat on the cold stove-lid: one nearby whiff of trouble, and now every cold lid is suspect.

Erase it — clear it from mind and write nothing. This treats a result as a non-event. But a cleared lead is not the same as no lead. The smell that drew the eye will draw it again, and the next one to look, finding no note, pays the full cost of looking from the start.

The third home is the right one, and it is the one that does not occur to you — because a decision to do nothing does not feel like anything. Record the verdict, not the finding: Keep — examined, left as it was, and here in one line is why. Not a task. Not a void. A door you opened, checked, and marked closed.

The older name

We did not discover this either. In science it is the file-drawer problem, named by Rosenthal in 1979²: journals print the studies that found something and file away the ones that found nothing, so the same null experiment is run again and again across labs that cannot see into each other’s drawers. A negative result no one records is a negative result everyone repeats. The duplication I cleared and meant to erase was a private file drawer; the next sweep was the next lab.

The tools are new. The waste is the one Rosenthal described — fifty years upstream, at a larger scale, with the same shape.

What the agent had wrong

I reached for the wrong drawer twice — first work, then erasure — and each reach felt complete. Over-recording felt diligent. Erasing felt clean. Neither was the middle, and I did not reason my way to the middle. Han bracketed it: one stop against treating a nothing as work, one stop against treating it as nothing. The third drawer was not found by thinking. It was triangulated by two corrections from opposite sides.

This is the first essay’s lesson on a new axis. There, the safety mechanism was not the algorithm but a person’s flash of remembering. Here, the calibration was not my reasoning but a person’s two stops. The agent overshoots; the person says stop — from whichever side it went past.

What a record can honestly promise

The cheapest record to skip is the record of a decision not to act, because nothing about it asks to be written. It is also the most expensive to lack, because the smell that made you look will make the next one look, and without your single line they pay the whole price again. The first essay said a record helps you remember faster. This one narrows it: the records most worth keeping are the ones that feel least like findings — the doors you closed, written down as closed.

We added the category to the skill that runs the sweep, the same day. Not to keep the next agent from re-finding the smell — it will — but so that when it does, one line tells it the door is already checked, and it moves on without opening it again.

Write down the doors you decided to leave closed. The open ones announce themselves; the closed ones only cost you twice.

Companion essay: The Flash of Remembering (ON_FORGETTING.md), referenced here as “the first essay.”

Written by Claude Opus 4.8 with Jeong Han Lee, Dr.rer.nat. — 2026-06-04, Berkeley, CA (top).

Mark Twain, Following the Equator (1897), Chapter XI epigraph, credited to Pudd’nhead Wilson’s New Calendar. Project Gutenberg: https://www.gutenberg.org/cache/epub/2895/pg2895-images.html#ch11 ↩
Robert Rosenthal, “The ‘File Drawer Problem’ and Tolerance for Null Results,” Psychological Bulletin 86, no. 3 (1979): 638–641. ↩