Candidate Three

Ø Predictions Part II holds the 0.69% open between two candidates and declines to pick: Candidate 1, apparatus leakage — the structural value is the unleaky value and every real measurement reads low; Candidate 2, refinement of the puncture normalisation under the KS-R.9 sub-debt. This paper enters as Candidate 3: the gap is neither in the apparatus nor in the puncture but in an unbooked provision — the structural value is the provisioned coupling, and what laboratories measure is the realised one.

Candidate 3 does not merely join the list; it resolves an inconsistency the corpus is already carrying, and the inconsistency should be stated in internal form. The corpus’s own position — AP06, carried into AP28 §7 and Ø Predictions — is that the inter-laboratory scatter is the leakage signature. A signature sizes its effect. The scatter is about 500 parts per million; therefore, by the corpus’s own identification, apparatus leakage is a 500 ppm effect. The gap is about 6,900 ppm. The corpus cannot simultaneously identify the scatter as leakage’s signature and ask leakage to carry fourteen times its signature. Candidate 1, as currently written, does both. Candidate 3 dissolves the inconsistency by division of labour: a universal 1/(1 + ε) carries the gap, and leakage carries what its own signature says it carries — the scatter, and a residual at scatter scale.

The discriminator already exists in the corpus. Appendix C of Ø Predictions proposes the meta-analysis: characterise each published G experiment by its boundary geometry and test the leakage correlation. Under Candidate 1, the boundary-geometry corrections must come out near 6,900 ppm and correlate. Under Candidate 3, they must come out at scatter scale — hundreds of ppm at most. Same proposed experiment, opposite predictions, and KS-CCC.4 binds this paper to the outcome. Candidate 3 adds a falsification surface, not an escape hatch — which is the only way a third candidate earns entry into a register that was honest with two.

On Candidate 2, this paper is indifferent at entry and constraining at exit — by sign before magnitude. A commitment is a cost, and a cost can only reduce a prediction; it cannot raise one. The bare 1/2π and 1/4π normalisations give 5.9092 × 10⁻¹¹ and 5.5036 × 10⁻¹¹ — both already below the measured value. No commitment of any magnitude can rescue either; reaching measurement from there would require a refund — formally, commitments of −15.7ε and −24.0ε — and no reading of the axiom provides one. With the commitment live, 1/π remains uniquely selected among the enumerated candidates, now excluded by direction before arithmetic. The KS-R.9 sub-debt is untouched, and KS-CCC.6 states what happens if its resolution collides with §7.