During World War II, the US Army Air Forces sought to reinforce their aircraft. Those that returned bore visible impacts: wings, fuselage, tail. The intuitive response was to armor the most hit zones. Wald inverted the map. If these planes returned with these impacts, then the hit zones were not those where impact destroys the aircraft. The zones almost intact on returned planes needed reinforcement.
The absence is the data.
Visible impacts belong to survivors. They do not map aircraft vulnerability. They map damage compatible with return. A perforated wing can still carry. A hit fuselage can still hold. But a plane hit in a truly critical zone does not always return to show its damage.
Absence enters the reasoning.
Survivorship bias is often described as a simple error: we observe successes and forget failures. But its hardest form lies elsewhere. The sample is not merely incomplete. It is produced by a terminal condition. To belong to the sample, the plane had to return.
Each visible hole is therefore double. It indicates that a projectile struck this zone. It also indicates that this strike did not prevent the plane from being observed. The trace does not only say what occurred. It says that what occurred remained compatible with inscription in the archive.
Absent planes do not provide their impacts. They provide the reading rule for present impacts.
In the series of notes he wrote for the Statistical Research Group in 1943, Wald does not correct data by adding what is missing. He changes the question posed to available data. The map of holes is not a map of zones to protect. It is a map of zones where the plane can be hit without disappearing from the sample.
What is visible designates the non-fatal.
What is missing perhaps designates the vital.
The sample is filtered by its result. Destroyed trajectories do not enter the collection that serves to decide. They existed, flew, were hit, fell. But they did not cross the documentary threshold of return. Return is not only a military outcome. It is a condition of statistical appearance.
The archive does not lie. It speaks from cases that survived the entry condition.
Wald's lesson therefore concerns not only planes. It concerns any collection produced by a visibility threshold. A database contains cases that were recorded. An archive contains documents that survived conservation. Scientific literature contains results that were published. A success story contains those who crossed the success threshold.
In each case, reading the sample requires reading the condition that produced it.
Doctrine
Absence carries information when it results from a filter.
A visible trace does not only say what occurred. It also says that what occurred remained compatible with its inscription in the sample. The survivor does not directly show system vulnerability. It shows damage the system can absorb without disappearing.
The sample is not biased by accident. It is constituted by its appearance threshold.
Open vector
Every sample is produced by an entry condition.
Companies still active write the history of strategy. Patients returned to hospital write the history of treatment. Models trained on available data learn the world that left traces. Archives preserve what crossed their conservation thresholds.
The question is therefore not only: what does the sample contain? But what had to be survived to enter it?
