Error weights change fitted eclipse times for EXO 1722−363 in 2005–2026 Swift/BAT data

We test eclipse timing in public 15–50 keV Swift/BAT data for EXO 1722−363, also called IGR J17252−3616. The frozen data cover 2005-02-14 to 2026-10-20. Of 84,702 accepted pointings, 20 on two monitor notice dates carry 48.76% of the supplied inverse-error-squared weight. Removing both dates shifts the fitted 2017–2019 transition center by 19.4 hours. None of these 20 rates is exactly zero. A separate 2013 point carries 32.12% of the weight in its fitted span after notice removal. Deleting this point shifts that center by 4.18 hours. Its origin is unknown. A reference model removes the notice dates and adds an error floor of 0.001 count cm⁻² s⁻¹. Its fitted fractional period rate is −7.51 × 10⁻⁶ per year. The central 95% orbit-resample range is −22.46 to +4.03 × 10⁻⁶ per year. This test does not resolve decay. Profile bounds, source changes, and retrospective choices limit the physical interpretation. We also find 4,776 integer-date errors in this source table. We supply the source bytes, row ledgers, code, figures, timing results, and reproduction instructions. The main result is a measured sensitivity of flux-based eclipse times to accepted-point weights.

EXO 1722-363 · IGR J17252-3616 · Swift BAT · eclipse timing · high-mass X-ray binary · orbital decay · data quality · error weights