A 3542-day model fitted through 2021 reduces MAXI flux prediction errors for GX 9+1 during 2022–2026

Do published long cycles predict later X-ray flux in two neutron star binaries? I test GX 3+1 and GX 9+1 with public MAXI/GSC daily data. A local plan selects both sources from the clear periodic class of Asai et al. (2022). Each model has a constant, a linear trend, and sine and cosine terms at its published period. The fit uses only monthly data before 2022. The unchanged coefficients predict 53 months during 2022 January–2026 September. For GX 9+1, the 3542-day model has a root mean squared error of 0.07114 photons per square centimeter per second. Its mean squared error is 91.9% below the early linear model error. It is 88.1% below the early mean error and 79.4% below the last 12 early months error. All planned controls keep a positive gain against a matching linear fit. For GX 3+1, the 2046-day model has 6.3% more mean squared error than the linear model. Its control results are mixed. A paired six-month block resampling gives descriptive 95% gain ranges of 0.847–0.960 and −0.688–0.296, respectively. The GX 9+1 test covers only 0.49 of its period. The published periods may use some early 2022 samples. A planned test from 2023 gives GX 9+1 a gain of 0.930. This is a retrospective test on reprocessed data. It was not issued in 2021. The result supports prediction value in this interval. It does not establish a strict clock or a physical cause. The package preserves raw data, provenance, code, predictions, controls, figures, and independent checks.

MAXI · GX 9+1 · GX 3+1 · neutron star binaries · long-term variability · X-ray light curves · prediction · reproducible analysis