Managed to extract from LLMs a human language version of what I was trying to say (a combo of several models). It's still too long for most, too dense, but it's cohesive, better than my lazy utterances and yet not math formulas:
...
A field can be locally competent and globally blind. Every anomalous datum gets explained away by a local mundane cause. Since the anomalies are not stored in a hypothesis-building way, no one ever integrates them into a higher-level model. This is the same failure mode as distributed systems logs where every machine reports “transient local error,” and no one reconstructs the global failure.
The disposal channels are specific and they are not exotic. An anomalous radiocarbon or OSL date that violates the expected chronology is assigned to contamination, sample mixing, or laboratory error and trimmed as an outlier—frequently before publication, so it never enters the synthesized record at all and dies in a lab report or an unpublished dataset. An artifact in the wrong stratum is assigned to bioturbation, root casts, krotovina, or post-depositional intrusion. A worked-looking object in deep sediment is assigned to natural geofact formation, frost fracture, or pseudo-retouch. A morphology that doesn't fit is assigned to pathology or individual variation. Each of these disposal categories is real, each is the correct explanation in the overwhelming majority of cases where it is invoked, and each is available at essentially no evidential cost: invoking it requires no positive demonstration, only the gesture of assignment.
The asymmetry is total. Accepting an anomaly costs the researcher their priors, their reputational exposure, the burden of an extraordinary defense, and a fight with reviewers; disposing of it costs a footnote. No one needs to coordinate this. Each individual, optimizing locally, reaches for the cheap channel, and the aggregate of a thousand independent cheap disposals is statistically indistinguishable from a policy of suppression while requiring no policy and no suppressor. This is the mechanism, it is sociological and behavioral rather than physical.
When Fable is back, you should discuss with it how much we actually do NOT know, and how any traces of a high tech civilization would be ignored, as they require operating in probabilities with such distances in time.