Abraham Wald was looking at bullet-riddled planes. The US Air Force wanted to armour the areas that had been hit. Wald told them to armour everywhere else.
The planes he was seeing had come back. The holes marked non-critical zones, a perforated engine can still fly, a punctured wing can still carry. The planes hit in the fuel tank or the cockpit did not return. They were not in the sample.
Classical survivorship bias overestimates the probability of success by excluding failures from observation. Its inverse is more interesting: the most important information is what the sample does not contain. What is missing says more than what is there.
Doctrine
Absence carries as much information as presence. Sometimes more. Every visible trace reveals by contrast a non-critical zone. What is invisible is what is vital.
Vecteur ouvert
What determines erasure was inscribed in the initial conditions. Only trajectories compatible with final visibility can be observed. The sample is not biased by accident. It is filtered by the outcome.
