The Obstruction Census: A Method for Killing Your Own Ideas in Public

A research program grades its own work, and it grades leniently. That is not a character flaw. It is what happens when the grading is invented after the computing, by the same hands that did the computing, in the light of a result they wanted to keep. This paper states a method that moves the grading to before the computing and then prints the grades where strangers can check them. I call it the obstruction census. It has five parts, and a researcher in any field can run each one with no equipment beyond a notebook and the discipline to write the hard sentence first: close the candidate set before any member is graded; register every test's kill condition in numbers before the test runs; count how many places you looked and put that count inside every significance you quote; audit each claimed sensitivity by asking how the conclusion moves when each input moves; and trace every number in the public text back to its computation, by machine, so that one untraced digit fails the whole text. I then run the method on the only program I can grade without mercy and without guessing, my own: the vacuum catastrophe campaign in Einstein-Cartan gravity, told as what it was, a sequence of obstructions. Nine waves of computation ran in one day and the kills were printed as results. Two locks still stand. The headline of that campaign is NOT SOLVED / PARTIAL, with producing-count 0 at derivation grade, and it stayed there through the largest single-day advance in the program's record. That unmoved headline is the method's best exhibit. An instrument that cannot spare its own program is an instrument a stranger can trust. The five named parts of the instrument: The Closed Set; The Pre-Registered Bar; The Look-Elsewhere Correction; The Elasticity Audit; The Numbers-Diff Certificate. Worked-example headline, unmoved: NOT SOLVED / PARTIAL, with producing-count 0 at derivation grade.