Where the Ladder Must Stand
Yuanjie Liu
PAPER · v1.2 · 2026-10-31 · human
Abstract
Mathematicians often say that an object had to exist: you cannot subtract in the natural numbers, so the integers appear; you cannot solve x^2 + 1 = 0 in the reals, so the complex numbers appear. The same objects are also called inventions, and both descriptions are usually offered without a test. This paper proposes one, in the form of three questions that can be answered yes or no, and reports what happened when the test was turned on the author's own work. Question 1: can the demand be met on the same elements? Question 2: among all extensions, is there a smallest one, determined up to a unique relabelling? Question 3: is the result new, judged against a catalogue of known objects? If Question 2 is answered yes, the object is pinned down: any two smallest extensions are the same up to a relabelling, and the relabelling is unique. Three steps of the number tower pass Questions 1 and 2. The step from the rationals to the reals does not: its missing minimality is recorded as a named obligation rather than proved, and no theorem depends on it. A counting invariant proposed inside the same programme is refused, because it collapses to a difference of prime-factor counts and is therefore already definable on the same elements; the refusal is argued by hand, since that layer contains no formal demand. Question 3 broke: the implementation of 'is it new?' compared the sizes of carriers, and the integers, the rationals and the complex numbers all have the same size, so the test refused everything it was meant to admit. The failure is reported at the same length as the successes, because it is the clearest evidence that the test has teeth. What survives is two working questions, one specified but unimplemented third, and a short list of what this approach cannot do, including the fact that every verdict is relative to a stated triple of ambient theory, refinement and catalogue. The mathematical core is formalised in Lean 4 with Mathlib: no sorry, one named axiom, and no theorem that consumes it.