The Kill Quota
Every sweep is expected to discard most of what it generates before anything survives. The point of a lab is not to produce ideas. It is to find the few that do not die.
The Yard runs on a simple, unfashionable rule: every research sweep is expected to discard most of what it generates before anything is allowed to survive. Not because the discarded ideas are stupid. Because the point of a lab is not to produce ideas. Ideas are cheap and the supply is effectively infinite. The point is to find the small number that do not die when attacked.
A sweep begins by casting wide: candidate mechanisms, candidate hypotheses, candidate reinterpretations of an old dataset. The first pass is generous. The second pass is not. Each surviving candidate is assigned the single axis most likely to destroy it, and it is attacked there first, not last. Selection bias, an order-of-magnitude argument, a missing control, a dependency that has not been stabilized. If it falls, it falls early and cheaply.
The discipline has a number attached because numbers resist self-deception. In one sweep of a disputed corpus the first pass produced a comfortable count of survivors. The second pass, run by someone whose job was to break them rather than defend them, cut that count to a fraction. The corrected number was not a disappointment. It was the actual result. The inflated number had been an artifact of not yet having tried hard enough to be wrong.
This is the part outsiders find strange. A lab that kills most of its own work looks, from a distance, unproductive. From inside, it is the opposite. The kill quota is what makes a survivor mean something. A claim that has been through it is not merely interesting; it is a claim that a hostile reader, working from inside the building, could not take down on the axis where it was weakest.
The rule protects against the failure mode that ends most ambitious research programs, which is not fraud but affection. You come to like an idea. You defend it. You stop attacking it precisely when it most needs attacking. The quota removes the choice. Something has to die this round, and the only question is whether you find the flaw or someone outside does, later, more expensively.
A survivor is only a survivor if the round was real.