The Offset Level: A Sample-Wide +0.0140 Beneath the External-Field Fits, Rebuilt as a Level with Its Dispersion Stated

The external-field effect is the sharpest near-field test my program's gravity record engages: in the published SPARC analysis (Chae et al.), each galaxy's fitted external field e should agree, galaxy by galaxy, with the catalogue's environmental estimate e_env — and on the *corrected* published table (the erratum's Table 2, 147 rows, parsed in full and validated against the paper's own figures), that per-galaxy agreement fails its own data: identity containment 0.34 against the ~0.68 the intervals promise, rank correlation −0.155 (p = 0.10). What survives the failure is a level. Rebuilt around the offset form — fitted e = e_env + e_0 — the offset is e_0 = +0.0510, bootstrap 68% interval [+0.046, +0.058], shared by all 112 galaxies of the paper's environment subset, with fitted e exceeding e_env for every one of them as a level statement. This paper executes the rebuild and grades the rebuilt species test by test, at full size. The level reproduces exactly. The constancy fails: e_0's per-galaxy scatter (standard deviation 0.1051 against a median printed error of 0.0395; chi-squared per degree of freedom 8.44) carries an excess dispersion of 0.0788 beyond the errors — e_0 is a floor with dispersion, not a constant. The running test that is clean — against the environment proxy ⟨x_0⟩, the one regressor that is not a component of e_0 itself — returns none: Spearman −0.040 (p = 0.67), tercile medians 0.0535 / 0.0600 / 0.0500 with overlapping intervals; the floor is environment-depth-blind at this resolution. The per-galaxy containment is not repaired: the offset form places 0.37 of the galaxies inside their fitted 68% intervals, barely above the identity form's 0.34 and far from the 0.68 expectation — the level is repaired, the containment is not, and named galaxies violate the floor in both directions (NGC 5033 sits on it at +0.054; NGC 5055 below it at +0.014; UGC 06917 negative at −0.037). Read at the source paper's address, the corrected catalogue's e_env undershoots the fitted field's level by about 0.051 across the subset — or the fits carry a sample-wide positive systematic of that size; the corrected table cannot separate the two readings, and the dispersion says neither is the whole story per galaxy. The separating instrument is named: the SPARC mass-model outside check, which also carries the mass arm the corrected table cannot run (it has no mass column). A limitation is printed on this paper's face, in §5: the corrected table's parse file was held in temporary storage by the campaign and no longer exists, so the sample statistics are carried at the campaign chair gate's independently re-derived grade (23 of 23 figures re-derived) rather than re-executed here; every closed-form relation among the quoted statistics is verified on this paper's bench. Grade, printed at full size: REBUILT AS A LEVEL, AWAITING ITS NEXT TURN — the level stands; the constancy and the per-galaxy containment do not. The programme headline this paper serves stands unmoved: at the gravity fork's address my side owns one robust offset object and a live species that can read it; at the particle fork's address it owns nothing, and this paper claims nothing there.