This was the first paper of the program I actually finished, and it was written out of frustration with a chicken-and-egg situation: a pre-theoretical pattern cannot attract theoretical work without empirical standing, and it cannot build empirical standing while it has no theory. Something has to break the loop, and the way you break a chicken-and-egg problem is by iterating: you put out a rough first version, expect it to be criticized, and let each pass sharpen the next. While drafting the Descartes–Soddy paper I kept hitting the same wall: is the strange-quark coincidence structure, or is it coincidence? And the uncomfortable realization was that nobody knows, not me, not a referee, not anyone, and nobody can know, possibly for decades. A pre-theoretical pattern sits in limbo: too speculative for the literature, too suggestive to throw away.
What I wanted was a repository: log the pattern, with its provenance and its numbers, and wait to see whether it converges or drifts as the data improve, without crowding the literature with pattern hunts in the meantime. That proposal is what this paper makes, as philosophy of science rather than as a physics footnote. Philosophy already has sophisticated machinery for evaluating theories, Lakatos's progressive and degenerating programmes, Mayo's severe testing, Bayesian confirmation, but it operates at the theory level. This paper extends that machinery downward, to pre-theoretical empirical patterns: explicit criteria organized around empirical convergence (does the pattern persist under improved measurement?) and theoretical traction (is theory moving toward explanation?), plus criteria for discovery provenance and independence of confirmation. Patterns that score well on both axes graduate; patterns with strong empirical records but theoretical stagnation stay honestly "stuck."
It changed how the rest of the program works. From the MPLA paper onward the policy has been: write the pre-registration first, wait for one full cycle of new PDG data, and only write the paper if the pattern survives it. Every paper in this program has survived at least one such cycle, the strange-quark lattice values have drifted around in recent FLAG updates, which makes the whole thing more exciting rather than less: we will find out.
There is also a deeper reason this question matters now. Particle physics has entered what one might call the post-new-particle era. Before the LHC even switched on, the community's stated fear was the "nightmare scenario: the Higgs and nothing else" (Adrian Cho, Science, 2007) — and that is, broadly, what the LHC has delivered: no new elementary particles found since the Higgs in 2012, leaving the Standard Model spectrum complete. There are exceptions and open anomalies, and I would not hang my hat on "no more particles." But the strategy of theory-first, find-the-data-later gets progressively harder when the undiscovered inventory thins out. For my own purposes the calculus was simpler: in that era, the frontier that matters to a solver-builder is calculating the particles we have. Which loops back to the same need: when progress comes from patterns in existing data rather than new bump-hunts, the community needs criteria for which patterns deserve the chase. That is the niche this paper occupies.
The review process sharpened this in a way I did not expect. One referee asked, simply: what even defines a theory? It is the right question, and it cuts deeper than it first appears. Where exactly is the line at which an empirical regularity becomes a theory? And what is a mechanism, for that matter? Is a mechanism, or a geometry, itself a philosophical commitment? My honest answer is that I am agnostic to the labels. The whole program started because I needed to compute fermion masses in a solver, and a formula you can evaluate does its job whether you file it under theory, geometry, or numerology. But the referee's question is precisely why the criteria in this paper are about observables: convergence under better data, traction toward explanation, provenance of the discovery. Those are checkable regardless of which label survives the philosophy.
The operational companion, what to do when machines generate patterns faster than humans can evaluate them, is the stratification framework in the London Journal of Physics, and the worked example is the pre-registered Monte Carlo discipline in the Koide–Soddy preprint.