What the Data May and May Not Do

The rule is already corpus precedent, and this section only names it. In AP28, structure proposed three candidate normalisations — 1/π, 1/2π, 1/4π — and measurement eliminated two. The direction of that traffic is the entire discipline: data may eliminate structural candidates; data may never generate them. A number that arrives from the measurement side, however clean it looks, is a fitted parameter wearing a costume, and the corpus’s zero-fitted-parameters claim dies at the first one admitted.

Applied here, the rule has teeth, because the whisper is real and it is fenced by being named. All figures in this section are in the paper’s structural convention, the quotient of §5. The commitment the recommended central value would demand, computed backwards, is 0.951 of one ε. Two clean fractions sit inside the window around it: 20/21 of ε lands about 12 ppm below the centre, and 17/18 of ε about 45 ppm above it. Both arrived from the data. Neither has a forcing argument. Both come pre-dressed in a story — an exempt channel here, a refunded share there — and the stories are exactly as good as the fractions, which is to say: generated by the answer. They are recorded in this section for one purpose. Seen. Refused. (In the discarded first-order shadow convention the backwards fraction reads 0.944 instead of 0.951 — and that the demanded fraction changes with a bookkeeping convention by an amount comparable to the gap between the whispers themselves (0.0066 of ε against their 0.0079) is one more proof that the digits cannot carry a decision.)

And the target itself is soft, which is the final proof that digit-matching is meaningless here. Measured against the lowest published determination the demanded fraction is 1.000 of ε — the paper’s own magnitude, a landing §9 already declines to lean on; against the highest, 0.925; the admissible window is the whole range between, seven and a half per cent of ε wide against a formal uncertainty of 22 ppm. Any fraction in the window “fits”. Structure alone can choose, and structure already chose before the window was drawn: the axiom writes one element, once. This paper states 1×ε, states nothing finer, and arms KS-CCC.1 over every future refinement — whoever writes it — that cannot show its forcing argument was written before its number.

And the fence holds the paper’s own history, because the selection pressure in this work did not run only on fractions — it ran on structure. Per-channel compounding and a channel-scaled cost were eliminated by measurement (legal under the rule, and disclosed); the count moved from ten-as-divisor to per-configuration inside the same measurement-laden opening analysis, and from per-configuration to the arena under purely structural pressure, with zero numerical motion after the first computation; and the 1×ε magnitude argument was articulated in the same breath as the numbers, against a residual that was already published canon — and the honest genre carries two clauses, both true and both claimed: in process, a structural correction with refused tuning — the commitment was developed over weeks aimed at the proton and the neutron, blind to what it would do to G, and moved here only when AP30’s precision closed the original target; in standing, postdictive — the 0.69% was published canon before this work began, and a landing on a known number is claimed as a landing, nothing more. The process is evidenced in PROV-CCC.1 through PROV-CCC.6; the standing is priced in §9. The dated sequence, with every elimination and its driver, is published as PROV-CCC.1 beside this draft, so that the claim of zero fitted parameters can be audited as a process rather than believed as a sentence.