A result can be true, well-formed, and change nothing. This programme is built around that gap rather than around the result.
A screen that finds everything has found nothing.
Put every expressed gene against survival and a large fraction of them will separate the curves. Not because they matter, but because there are far more measurements than there are patients, and the arithmetic does the rest.
That is not a defect to be corrected afterwards. It is what the method does, and the apparatus exists so that the outcome cannot reach publication by accident — including when the result is one we would like to be true.
A correlation may not be described in the language of treatment.
Words carry claims. A result that separated two groups on a survival curve has not shown that acting on it helps anyone, and may not be written as though it had.
Every claim is graded against the experiment that produced it and refused if its vocabulary outranks its evidence. The instrument is pointed at this programme’s own output first.
“A result can be true, well-formed, and still change nothing.”
