All facts are provisional. If a validating observation does not continue to hold, that is another observational fact. At some point of replication and sigmas, it becomes another Problem, to be fed back into the scientific meat grinder.
For example, maybe vague observations of Mercury were consistent with Newton. But eventually, telescopes improved, duration of accurate records extended, many good instruments were deployed around the world. Then it became a Problem. Also note that Einstein fitted GR to Mercury, so he could not predict it as independent confirmation. It took the Eddington observation of light bending near the sun, during a solar eclipse, to provide the first evidence (even that was weak, and fixed, but the fix was on the right side of history :)
For example, maybe vague observations of Mercury were consistent with Newton. But eventually, telescopes improved, duration of accurate records extended, many good instruments were deployed around the world. Then it became a Problem. Also note that Einstein fitted GR to Mercury, so he could not predict it as independent confirmation. It took the Eddington observation of light bending near the sun, during a solar eclipse, to provide the first evidence (even that was weak, and fixed, but the fix was on the right side of history :)
https://en.wikipedia.org/wiki/Eddington_experiment