What the C60 Geometry Forces: An Honest Accounting of the Forced Count, with One Booked Prediction Loss

Version 2 (27 October 2026) answers the question version 1 registered as open, and corrects the count. Version 1 asked whether the C60 adjacency forces the B1A5 solar-angle assignment sin2θ12 = 1/(2+φ), stating it as open and explicitly not pre-judged. It is now computed, and the registration stands timestamped ahead of the computation. It does not force it. The kernel's rule obtains the solar angle as the squared middle component of the normalized five-fold axis (0, 1, φ). Enumerating every squared component of every icosahedral symmetry axis — 6 five-fold, 10 three-fold, 15 two-fold — yields exactly eight distinct candidate values in (0,1): 1/(4φ2) = 0.095492, 0.127322, 1/4, 1/(2+φ) = 0.276393, 1/3, φ2/4, φ2/(2+φ), and 0.872678. Reaching 1/(2+φ) requires two choices the geometry does not make: which axis class, and which component of it. By the framework's own rule that a free choice makes a result FIT rather than FORCED, the solar angle is conditional. The count is corrected to 34 forced · 11 conditional · 0 lost, of 45 distinct targets, superseding the 35 · 10 · 0 of version 1. This is booked as a framework-level reading failure, not a prediction loss; the prediction loss on GR1 stands separately at one. The stronger finding: the rule is closed. None of the eight candidates reaches the measurement. Zero land inside 0.3092 ± 0.0087, and the measured value falls in a gap between the two nearest, 1/(2+φ) = 0.276393 below at 3.77σ and 1/3 below—above at 2.77σ, which contains no icosahedral axis component. Every candidate is excluded at 2.77σ or worse against global data and 4.25σ or worse against JUNO alone. So the failure is not that the framework chose the wrong one of eight: the rule "sin2θ12 is the squared component of an icosahedral symmetry axis" cannot produce the measured solar angle under any choice available to it. That closes the route rather than one assignment, which is the more useful statement. Which candidate looks closest is also dataset-dependent — 1/3 wins against global data, 1/(2+φ) against JUNO alone — a further reason the selection cannot be made after the fact. The 34 remaining forced results are untouched; none routes through the axis-component rule. Both the original paper and the addendum are included, with verification scripts for each. An earlier presentation of this framework carried the headline "zero free parameters — 206 constants verified." That headline does not survive honest accounting and is retracted here. What the C60 adjacency, read under Ih, actually forces is a definite but smaller thing: of 45 distinct physical targets, 35 are forced, 10 are conditional on named premises, and 0 are lost, where "lost" means a forced value contradicted by data in the same frame it was forced in. Separately, on the forward-prediction ledger — values staked before the measurement, which must never be summed with the first count — the framework has booked one loss: the solar mixing angle. The framework forces sin2θ12 = 1/(2+φ) = 0.27639. This is the golden-ratio solar mixing prediction, first stated by Kajiyama, Raidal and Strumia (2007) and realised in A5 family-symmetry models by Everett and Stuart (2009) and Feruglio and Paris (2011); in the current classification it is case B1A5. Since Ih ≅ A5 × Z2, the geometric route here and the flavour-symmetry route reach the same group, so we claim the independent rederivation and not the prediction. JUNO now measures 0.3036 ± 0.0064, and Petcov and Titov book B1A5 as disfavoured at 3.3σ from a full likelihood fit. The loss is shared with the entire A5 golden-ratio programme, which is the more informative statement: the evidence is not that this framework erred where others succeeded, but that A5 appears not to be the lepton flavour symmetry, and any construction forcing B1A5 inherits that. The paper gives the corrected count, the provenance of its denominator, the passenger test that decides whether a target belongs in the count at all, the results that are firmly forced, the ten conditional targets with their premises named, and the seven claims cut. It also records a near-miss the authors' own method rejected during revision: the two golden-ratio forms in the A5 literature bracket the measured angle and their geometric mean is exactly 1/(2φ) = sin 18° = 0.309017, which sits 0.02σ from the global value — and a base-rate enumeration shows 26 of 1299 simple φ-expressions land inside the same error bar, a 2.0% base rate, so the match carries no evidential weight and receives no label. It is an accounting, not a victory lap.