Two hundred failed trials can mean the drugs do not work, or that the failures were never measured. Those are different worlds, and the literature does not distinguish them.
“The trial failed” and “the trial stopped” are not the same sentence.
A trial can end because the intervention did not work, or because the money ran out, the recruitment stalled, or the sponsor lost interest. Only the first is evidence about the disease. The register records the distinction inconsistently, and the literature mostly does not record it at all.
So the census is of every interventional trial rather than a sample of them, classified by the reason it ended — because the claim that a hypothesis has been exhausted rests entirely on which of those endings you are permitted to count.
The negative controls are run first, and they are run on us.
At genome scale an apparatus will produce a confident answer whether or not one exists. So the controls that should return nothing are run, and seen to return nothing, before the analysis that might return something is allowed to run at all.
Instruments built here are tested against the other programmes rather than only against their own domain. An instrument that only works where it was built has not been shown to work at all.
“A field that cannot tell abandonment from refutation cannot declare a hypothesis dead.”
