V

The object entered the catalog as 2042 FM27 and remained boring for thirty-six hours. It was dark, slow against the stars, and farther away than the automated surveys usually became excited about. The first orbit left Earth comfortably outside the uncertainty volume. Follow-up shortened the arc and moved the volume. A radar window shortened it again. Then the corridor touched Earth. The player began purchasing observation without purchasing anything: a weather satellite rolled six degrees during a dead interval; a university telescope abandoned half a night of exoplanet photometry; two commercial spacecraft took images through star trackers that had never been sold as astronomy instruments. By the fifth day, the object had ceased to belong to astronomy. Insurance models ingested it. Launch providers moved manifests. Civil-defense agencies requested regional probabilities no responsible orbit determination could yet provide. The validators remained separated and received the same growing measurement set. One small instrument on a mountain in northern Chile arrived late. A landslide had cut its fiber the previous winter, and the modernization work was still incomplete; its reduction computer therefore remained on the old local stack, the one scheduled to disappear when the trench was rebuilt. Its three positions lay consistently outside the common ellipse. The raw images were clean. The stars were clean. The timestamps passed internal checks. The player took the centroids before the local reductions touched them and passed them through the modern transform. Two points moved inside the ellipse. The third sat on its edge. The observation had not been discarded; it had been rescued from an outdated interpretation. The local solution, including its residuals, went into the incident archive beside the raw frames. The Chilean instrument was assigned another night. Its next measurements agreed. Over the following week the independent validators converged toward an impact probability above sixty percent. Their uncertainty shapes differed. Their preferred perturbation models differed. Their central answer did not. The old station's first three points remained in the archive as a solved discrepancy.

Correction often looks exactly like this.

A report arrives from the edge. The evidence is checked. The raw observation survives. The interpretation does not. 

We do not protect error by calling it local knowledge. We do not preserve a bad map because somebody drew it independently. A corrigible institution has to be capable of hearing an objection and concluding, after contact with the world, that the objection was wrong.

Otherwise "correction" has become a ceremonial right to be ignored.