Three tomographic redshift-dipole estimators barely respond to the observer's velocity in DESI-like quasar samples cut on observed redshift

The Doppler dipole in spectroscopic redshifts has been used to test, independently of source counts, whether the excess quasar and radio-source number-count dipole reflects our motion. Recent measurements with the tomographic redshift-dipole estimator report 196 km/s (combined eBOSS tracers), 443.8 km/s and 357.95 km/s (DESI DR1 quasars). Injection-recovery on 16 DESI DR1 EZmock quasar realisations and on the shuffled real DR1 catalogue shows that when the sample is cut on observed redshift from a parent that extends beyond the cut, the estimator's response to the injected velocity is consistent with zero: 0.014 (95% interval −0.057 to 0.085) for FM24's eBOSS-quasar configuration, −0.031 for the public code, 0.043 for our transcription of Wu & Xia's published equations (which does not reproduce their real-data solution). Cut on true redshift before the shift, the same estimators respond at 0.994, 0.996 and 0.975; a light-cone ending at the analysis limits gives 0.415 (F) and 0.420 (B). A three-dimensional grid search on 4 mocks does not restore the response. Under true selection a response near one is expected to first order for any estimator of this kind whose shells are assigned on de-Dopplered redshift, and in these tests the precision of the estimate appears to come from the sharp edges of the redshift window. The quasar mocks behind the published validations are described as covering the analysis range, so they could not have been re-cut on observed redshift from a parent extending beyond the cut; they were either not re-cut (our true-selection case) or re-cut as light-cones (a partial response in our versions of the estimators). For FM24's quasar configuration and our transcription of Wu & Xia's equations applied to DESI-like quasar samples, such validations therefore do not show calibration in a real catalogue; in that sense neither the quasar contribution to the 196 km/s value nor the 357.95 km/s value has been shown to be a calibrated measurement of our motion. eBOSS itself, FM24's galaxy configurations (whose mocks extend beyond FM24's upper cuts), Wu & Xia's own code and the Chen et al. configuration were not tested. A whole-sample regression keeps a response of 0.258 in mocks, but on the real catalogue it is dominated by a weighted-redshift offset between imaging regions. Note on versions. Version 2 supersedes version 1, which stays available. It follows an independent post-publication review. Estimator B divided the catalogue WEIGHT by the mean completeness weight per tile count, which that column already includes; all B runs were redone (observed-selection response 0.039 → 0.043, true 0.956 → 0.975, light-cone 0.446 → 0.420; no conclusion changed). The 1480 km/s runs of the shell estimators other than the edge-removal test were rescanned on a wider grid (no minimum of the unchanged estimators moved). Added: the objective as an equation, estimator F in the light-cone ordering, a central-shell control for the edge mechanism, three-dimensional light-cone runs, and the first-order argument that true selection is calibrated by construction. The title is now limited to three estimators, and the window edges are said to set the precision rather than the calibration. Withdrawn: version 1's statement that the null scatter applies to all three orderings, its conclusion that a match of mock amplitudes is "not diagnostic either way", its unqualified conclusion about the 196 and 357.95 km/s velocities, and its claim that the checker tests every result-dependent statement. Full list in the paper's Note on versions. The deposit contains all scripts, per-run outputs (and version 1's superseded outputs), the aggregation producing every number, and a checker that tests the result-dependent statements it lists and the quotations of other work against their sources. Computations and drafting were assisted by an AI system (Claude).