A BAT point on a notice date reduces conditional cycle prediction gain for 2S 0114+650 in 2023–2026

We tested a known X-ray cycle in 2S 0114+650 with public monitor data saved on 2026-10-21. We fitted its period and shape to Swift/BAT data before 2023. We then held the cycle fixed. Annual means and the binary orbit model could refit in the later data. One point on 2019-03-30 passes the monitor flag rule. It has 46.83% of all accepted inverse-variance weight. Its date appears in a BAT communication-loss notice. With this point, the two-harmonic cycle reduces 2023–2026 weighted squared error by 2.77%. Removing this point alone increases the gain to 4.50%. Removing both dates in the notice also gives 4.50%, with a training period of 30.755 day. Conditional 95% block intervals are 0.81–8.29%, 0.49–8.40%, and 0.06–8.06% for blocks of 31, 62, and 124 days. The one-harmonic score changes from -0.66% to +2.96% when the low-error point is removed. MAXI data give smaller gains. Their intervals include zero. The BAT gain is positive for 2023–2025 and negative for the 49 accepted days in 2026. These results give a new test of cycle prediction and a check on the current public file. The test is retrospective. Its intervals hold the fitted model fixed. It does not establish the physical cause of the known cycle.

2S 0114+650 · 4U 0114+650 · superorbital modulation · Swift BAT · MAXI · time series · data quality · conditional prediction