Testing gravity at low accelerations: two papers on Gaia EDR3 wide binaries and coefficient-level tests of the radial acceleration relation

This deposit contains two companion research papers examining gravitational behavior at low accelerations, along with a complete archived project repository. Paper 1 analyzes 14,071 Gaia EDR3 wide binary pairs using forward-modeled likelihood analysis over both the relative velocity and the velocity–separation angle. Key findings: velocity-only fits of these data report a spurious boost of about half the galactic calibration, which the direction data veto; after stratifying by astrometric quality, boost amplitudes alpha ≥ 0.5 are excluded on the cleanest strata at every anchoring of the measured companion prior, with alpha = 0 not excluded; the pair-level error model is audited in full, with the joint fits demanding effective errors near twice the formal values (origin unresolved, decidable with Gaia DR4). Paper 2 tests the radial acceleration relation using 2,700 SPARC rotation-curve points and KiDS-1000 weak-lensing data. Results show "the leading low-acceleration coefficient is measured at 0.26–0.45" with zero excluded across treatments, "an acceleration scale on cH0/2pi," and an unexpected anomaly in gas-dominated galaxies. The supplementary archive (thermal-horizon-rar-snapshot-2026-08-21.zip, in the access-restricted v1.0 record) provides the frozen project repository, laboratory notebook, measurement ledger, pre-registered predictions, and analysis scripts for evidentiary purposes. The live repository is now public at github.com/TheCake/thermal-horizon-rar. Computational collaboration with Claude (Anthropic) is acknowledged in both papers. Version notes v1.2 (2026-09-04): correction #21 — earlier versions claimed a calibration measurement of pair-level velocity errors at 2.1 times formal or more on a boost-free separation bin. A null-injection audit by the program's own referee process showed the measuring instrument prior-dominated, and the claim is withdrawn throughout; see Paper 1, Section 4.1 and Appendix A. Both papers were additionally hardened by two further adversarial review rounds; no scientific conclusion changed. SHA-256 Checksums (v1.2) hajek2026a_wide_binaries.pdf: 0f596d20a65d26d1020f3e828fbc5439da6b5db63079402c49b69f8f124ea19f hajek2026b_rar_coefficients.pdf: 92d466247111b25851d95c28529e2c1fe2e75348b1892f030de37ef3c724cb13