Loop Walking: A Field Guide to Knowing Whether You Found Anything - How to tell a signal from your own instrument breathing

ABSTRACT This guide exists because the method kept evaporating. The instrument is portable — that was the point, and it turned out to be the problem. Anything reconstructible from a short spec gets reconstructed, and every reconstruction is a fresh chance to lose a correction that cost a week to find. A tweak nobody wrote down is not a method. It is a habit, and habits do not survive a rebuild. What follows is the written form: eight rules for reading a quantised topological invariant out of noisy data, each stated with what it cost to learn. The rules come from three worked programs — a shelved seismic precursor campaign, a sky survey on Gaia DR3 and the quasar reference frame, and a nucleic-acid topology port. Two of the three produced no publishable positive. All three produced the method. The organising idea is the competence surface: not “the instrument works” or “it does not,” but a map, over the axes that actually govern performance, of where a tool reads clean and where it fights the substrate. We show one built before the claims instead of after the funeral, and what the chapters on either side of it cost for lacking one. The worked failures include four experiments run on a channel whose effect was one eleventh of its own noise; a control that flipped dirty when the station polygon was swapped for a worse-conditioned one; a classifier claim written before its floor was measured; and an accession error that survived every internal check the program had and was caught by a single query to a reference the pipeline did not author. Two appendices follow: the mistakes by category, and a cross-domain vocabulary — sixteen fields’ names for one swept area, eleven false friends, the mature prior tools this instrument is not competing with and must not be confused for, and a five-question diagnostic for deciding whether the quantity in your own field is this one under a local name. A negative with a floor under it transfers. A positive without one does not. Keywords: research methodology · noise floor · competence surface · pre-registration · null results · negative results · reproducibility · instrument characterization · detection limit · data provenance · cross-domain instrument · holonomy · anholonomy · parallel transport · geometric phase · Berry phase · Wilson loop · Thomas precession · deficit angle · winding number · linking number · Burgers vector · topological invariant · topological data analysis · protein topology · knotted proteins · nucleic acid topology · writhe · NMR ensemble · conformational variability · GNSS · integer ambiguity resolution · phase unwrapping · astrometry · Gaia DR3 · quasar reference frame · secular aberration drift · quadrature · orbital angular momentum · crystal dislocation · effect size · signal-to-noise ratio · AUC · out-of-sample validation · permutation test