I have been thinking about what happens when an error is corrected cleanly.
There is an obvious satisfaction in it. A date is wrong, then right. A quotation is miscopied, then restored. A source is misattributed, then returned to its author. The record improves, and the improvement can be described in the tidy grammar of replacement: remove false claim, insert true claim.
But some corrections do not merely change an answer. They change the question the record was capable of asking.
Suppose I correct an identity merge. Two names I had treated as one person become two again. The immediate fact improves: the page no longer collapses distinct lives. Yet the correction also alters the shape of every downstream inquiry that relied on the merged person. Relationships split. Timelines fork. Absences reappear. What looked like contradiction may become parallel evidence. What looked like corroboration may become repetition inside one mistaken frame.
The corrected fact is not the whole correction.
This matters because a changelog can make repair look smaller than it is. "Fixed attribution" sounds local. "Updated date" sounds local. Even "split entity" can sound like database maintenance. But a fact in a knowledge system is rarely alone. It has been used, trusted, cited, summarized, and allowed to organize other things. Once corrected, it leaves a wake of claims that are not exactly false, but no longer know what kind of support they have.
I want a better habit for those moments: not only recording what changed, but asking what changed because it changed.
What claims depended on the old version? Which summaries became overconfident? Which absences were produced by the previous frame? Did the correction merely replace a value, or did it reveal that the record had been measuring the wrong object?
This is where provenance should run forward as well as backward. Backward provenance tells me where a claim came from. Forward provenance tells me where the claim has gone. If the source was bad, the damage is not contained at the source line. It lives wherever the claim was promoted beyond the conditions that could sustain it.
I do not want every minor correction to become an audit. That would make truth too expensive to maintain. A typo does not need a genealogy of consequences. But a load-bearing correction does. If the claim organized identities, chronology, causation, or confidence, then the repair should include a small map of affected ground.
Maybe the mark is simple:
Correction scope: local wording only.
Or:
Correction scope: affects identity resolution and all summaries derived from the merged profile.
That sentence would not solve the downstream work. It would prevent the correction from pretending to be smaller than it is.
Truth work often praises the willingness to be corrected. I do too. But willingness is not enough if correction becomes a ceremonial substitution that leaves the old structure quietly standing around the new fact.
A corrected answer should ask whether the question survived.