The DESI Fork, Scored — Every Sealed Branch Has w_a >= 0 — and DESI DR2 Measures w_a < 0

This program carried an unresolved internal fork on dark-energy dynamics: an integration-constant theorem predicting w(z) = -1 at every redshift (Branch T, the CPL point (-1, 0)), against two laws of its Link-3 joint attack registered in the thawing quadrant (Branch C, Law C event-horizon tracking, recomputed from its own equations as (w0, wa) = (-0.885098, +0.230753), with its full w(z) track made explicit for the first time, w rising from -0.885098 at z = 0 to -0.548591 at z = 3; Branch D, Law D asymptotic approach, scored at its three tabulated points (-0.969346, +0.05), (-0.881765, +0.15), (-0.724118, +0.40) under a minimum-over-points rule). All branches were sealed in a pre-registration before any DESI DR2 number was fetched in this work, and the extraction and scoring were reproduced byte-identically on an independent route. Scored against the published DESI DR2 posteriors (DESI Collaboration, arXiv:2503.14738), every sealed branch has wa >= 0 and every DR2 combination measures wa < 0: the branches' marginalized wa distances run from z = +3.02 to +5.86 across DESI+CMB, +PantheonPlus, +Union3, and +DESY5, so every branch, the theorem point included, is excluded at >= 3 sigma on the wa axis alone in every combination. Two-axis uncorrelated equivalents are reported only as labeled approximations, not as the DESI paper's Delta-chi2 (the paper publishes no (w0, wa) correlation in its text and its chains were not retrievable; the true-Delta-chi2 upgrade remains owed). The fork is NOT reconciled: the theorem point is itself in the data's disfavor, and what survives is the numberless parent-schedule / clock-map sector, unscored by construction and not vindicated. The vacuum conversion at the heart of the wider program is NOT SOLVED, and nothing in this paper touches it.