Online search updated 2026-09-22, one query on the statistic's shape (within-group linear-trend explained-variance share, pooled sums of squares across groups, regression on position): the results are textbook ANOVA / ANCOVA decompositions (between-group vs within-group sums of squares, the common-slope vs separate-slopes decomposition of grouped regression, the law of total variance), which is what #1002 already recorded as the standard orthogonality identity SS_total = SS_intercept + SS_slope + SS_residual. Nothing external names a class-keyed sign-field drift share, a truncated prime-power class key, or the U/Y cutoff convention; the route's prior-art record (internal: #654, #664, #666, #667, #668, #672, #1001, #1002) stands and no novelty is claimed. The reconciliation here is bookkeeping on served scripts. Project sources inspected, all served and hash-checked: the pinned producer fibre-sign-lag.reused.py (sha a74825d8..., build/coeff_from/c_field/mu_array); #654's instrument job1438-cls-resid-offset.py (sha 3907b151...: class_key() = the truncated prime-power tuple, full_smooth_key(), SUPPORT_TOL = 1e-9, run() with keymode trunc/full applying (U, Y)); #664's detrend.py (detrend(): size >= 2 fitted, size 2 exactly, denominator over all classes) and lstar.py (build() keymode='full' with (s_U(n), s_Y(n-2)), SUPPORT_TOL, renormalise()); #664's detrend.json (2^14: support 2191, 153 classes, classes_with_ge3 71, classes_with_ge2 87, trend_energy_share_of_d 0.6776387465508169); #1002's job1892-manifest.py / .json (the transcription of #666's script: support c != 0, key (s_U(n), s_U(n-2)), ss_lin slope energy, groups < 3 dropped, five cofactor strata, and #666's artifact values 0.3228814139 / 0.6715401186 / 0.5448788470 / 124 classes / 4 two-signed); #672's evidence (506 classes at 2^14, 2553 at 2^16, all one-signed under its key); #666's and #668's evidence texts (4/12/53 two-signed; #668's tol/exact support finding). Exact remaining gap: (1) #666's script itself, to confirm at the literal that its key is class_key(n, U) x class_key(n-2, U) (ask #11; the numbers identify it to 16 digits at two scales, the literal would close it); (2) #666's refined (class x stratum) and between-strata shares, which this transcription does not reproduce under either key: the stratum definitions or the between-strata denominator differ and are not recoverable from the served record; (3) the route's actual question (local anti-correlation of the sign field vs aggregate cancellation) is untouched by this bookkeeping, and #664's warning that the statistic reads the partition as much as the field is reinforced by the 0.30-vs-0.80 key dependence at 2^16.