This record deposits two companion papers from a single measurement and theory-testing program on gravity at low accelerations, together with the complete repository snapshot that produced them. Paper 1 — "A direction-resolved analysis of Gaia EDR3 wide binaries: pair-level velocity errors and an upper limit on the low-acceleration velocity boost. " A forward-modeled joint likelihood over the normalized relative velocity and the velocity–separation angle of 14, 071 Gaia EDR3 wide pairs, marginalizing eccentricities, hidden companions, contamination, selection, and measurement noise. Three results: a gravity-independent calibration measurement showing pair-level velocity errors at least 2. 1 times the formal Gaia uncertainties at 0. 2–2 kAU, independent of RUWE; a methodological result that velocity-only fits of these data report a spurious boost which the direction data veto, rendering three conflicting published analyses arithmetically compatible; and an upper limit, with boost amplitude alpha >= 0. 5 excluded on the cleanest astrometric strata at every anchoring of the measured companion-rate prior, while alpha = 0 is not excluded. Paper 2 — "Coefficient-level tests of the radial acceleration relation: a measured zero-point coefficient, an acceleration scale at cH0/2pi, and a gas-dominated anomaly. " Coefficient-level tests on 2, 700 SPARC rotation-curve points, KiDS-1000 weak-lensing data, and per-galaxy environmental fields, treating the Cadoni hierarchical fits with measured distance and inclination priors return an acceleration scale on cH0/2pi for every function tried; and one anomaly is opened rather than closed, a sign-reversed leading coefficient in 38 gas-dominated galaxies surviving eight controls. The attached archive thermal-horizon-rar-snapshot-2026-08-21. zip is a git archive snapshot of the project repository (tracked files only). It carries the frozen internal record, the chronological lab notebook including every correction and retraction, the curated measurement ledger, the register of predictions signed before test, and every analysis script. Each analysis stage in the program was pre-registered: its verdict thresholds and kill conditions were committed to version control before the deciding computation ran, so the commit chain inside the archive timestamps every bar. The archive is attached as supplementary material for that evidentiary purpose. A third manuscript in the program, on the candidate mechanism, is present inside the archive but is not circulated as part of this deposit. Author: Filip Hájek (independent researcher). The computational analysis, literature verification, and manuscript drafting were performed in collaboration with Claude (Anthropic), as stated in each paper's acknowledgments. SHA-256 checksums of the deposited files: hajek2026awidebinaries. pdf — 5243ce9c66d74ee9361dc2708f9514053524324ad5a240cff4e5382c0915035fhajek2026bᵣarcoefficients. pdf — e137d05b9bcd34967196c278e32dd4d43eec4eff4e1facdde777817e5000f64cthermal-horizon-rar-snapshot-2026-08-21. zip — 598764b06bc758168c0b890ab83ef73d028e016f8db41464eb9155a94aa66733
Filip Hájek (2026) studied this question.