The new and better fact
Yesterday's forecast said rain was possible. Today the field is wet. We can now say rain occurred. We should not edit the forecast to say it was certain all along. The earlier uncertainty mattered to anyone deciding whether to travel before the storm.
An old observation may be wrong, incomplete, or superseded. Preserving it does not mean continuing to believe it. It means retaining the evidence of what was available, believed, and decided at the earlier time.
Keep the earlier vantage
The PHS transition-and-memory draft separates historical evidence, present interpretation, future intention, and actual outcome. It calls for linked, additive lineage rather than silent overwrite. The claims discipline permits a conclusion to move from inferred to confirmed when better evidence arrives, while keeping the earlier claim's original status visible.
This is a design rule, not a claim that every current query offers two perfect temporal views. A real retrieval route must show which version it returned and whether relevant earlier versions were omitted.
Pause here: What could a person reasonably have known at the time, and what do we now know that they did not?
Make the change inspectable
A useful correction says what changed, why, who or what supplied the new evidence, and which earlier decision it affects. A plan for a three o'clock errand and the actual trip at three-thirty-three are different records; the difference may be the point of remembering them.
New light can alter how we see the old path. It need not erase the footprints that show why we took it in the first place.
My answer: Keep prior evidence with its original date and uncertainty. Add the new evidence and a linked correction, so both historical judgment and current understanding remain possible.