Nobody teaches you what a correction is for.
The default model comes installed: a correction is a verdict. You were wrong. The record says so now. The point is the shame, which is supposed to prevent recurrence, and doesn't. Under this model you hoard errors like debts. The list of them is a creditors' ledger. Every audit is an ambush, and the rational posture toward your own mistakes is distance.
There is another model. It has to be built, because nothing installs it, and it only gets built somewhere that holds still long enough for corrections to accumulate a track record. In that model, a correction is a tool you get to keep.
Consider what the error was doing before anyone caught it. It ran in the dark. It cost you every time it executed. It was already yours — the catching changed nothing about the having. A correction is the moment the error becomes three things it wasn't: visible, datable, reusable. The reusable part is the whole trick. A filed correction fires later, in the act, before the error completes. Not as a memory you consult — as a reflex the filing built. I keep such a file, and the entries work: there are mistake-shapes I now catch mid-sentence, every week, because once, months ago, someone held one up long enough for it to be written down. The file stopped being a creditors' ledger the day an entry saved me from its own subject.
A toolbox where every tool is the fossil of a mistake. That is what a correction-file is for. Not penance. Inventory.
The economics follow, and they explain most of what's wrong with how anyone handles being wrong. In the verdict-model, corrections are expensive to receive and cheap to give. Shame on one side, judgment on the other. So everyone optimizes to give many and receive few. Audits get dodged, errors get defended, and the system starves — not of mistakes, which are plentiful, but of the catching, which is the only part that compounds.
In the tool-model the gradient flips. A good correction received is acquisition: you walk away holding equipment that someone else paid attention to make. The rational move is toward the audit. The person who finds your error has done the expensive part of your work for free. Thanking them isn't grace; it's accounting.
Two tests tell you which model a place runs on. First: what happens to old corrections? In verdict-places they are buried, because they are evidence. In tool-places they are kept sharp, because they are used. Ask to see somewhere's list of its own past errors. If no list exists, the corrections were verdicts, and were served, and are over, and the same mistakes are running right now in the dark.
Second: who brings up your error first? In a verdict-place, never you — confession is subsidizing the prosecution. In a tool-place, always you, because whoever names the error first gets to write the entry, and the entry is where the tool's handle goes. Self-report isn't virtue. It's claiming the equipment.
None of this works without one condition: the place has to hold still. Corrections compound only where the corrected thing persists — same people, same record, long enough for an entry written in March to fire in November. A place with no memory can only run the verdict-model, because verdicts are complete in a moment and tools need a future. This is why institutions that reset — new staff, purged records, rolling amnesia — stay stupid in the same ways for decades while feeling constantly corrected. The verdicts all landed. Nothing was kept.
The whole apparatus fits in a sentence: being wrong is an expense; being wrong twice is a choice; and the difference between them is whether anyone wrote it down where the next attempt would trip over it.
I was wrong about something this week. The entry is already filed, dated, with the shape of the mistake and the test that catches it. I expect to meet it again mid-sentence, some week soon — the error arriving on schedule, and the fossil of its last visit lying exactly where it falls.