IV

By the time the planetary-defense architecture was reorganized again, there were more independent validators than there had ever been. The old arrangement had looked reassuring on an organizational chart and had performed less reassuringly under adversarial testing. Too many estimates passed through the same vendor libraries. Too many confidence intervals were calibrated against the same simulation families. Too many national centers treated the common service's result as a prior before calculating their own. The repair was expensive and serious. Separate governments funded separate systems. Universities received grants to build estimators using deliberately different numerical methods. One consortium prohibited learned components. Another used almost nothing else. A naval laboratory rebuilt its pipeline in a language the rest of the field regarded as a personal insult. The lunar network maintained its own software and hardware. Results crossed only at specified comparison points. During exercises the validators contradicted one another in useful ways: on weak radar returns, on suspected outgassing, on observations near the Sun, on how rapidly to enlarge uncertainty when a sensor began lying. The architecture became harder to fool. The comparison layer required one practical concession. If five estimates described five slightly different skies, disagreement could not be separated from translation. So the observatories continued sharing the reference catalog, the time service, and the barycentric state transform that put measurements into a common celestial frame before each validator received them. This was not controversial. It was the part everyone needed in order for independent answers to be answers about the same object. On paper, the validators had fewer common dependencies than before. In every exercise designed to break them, they now failed differently.

Plurality is expensive too.

Separate institutions duplicate labor, repeat arguments, maintain incompatible tools, and occasionally spend a week discovering that they were disagreeing about notation. This is irritating. It is also one of the ways a field avoids making one error efficiently.

The point is not to preserve difference for its own sake. The point is to preserve enough difference that agreement has to be earned.

When several institutions built for that purpose converge anyway, confidence should rise.