A fixed 167.43-day clock gives lower 2023–2026 MAXI errors for 4U 1820-30 than the published clock with a period decrease when early Fourier profiles are fixed

We test two published central clocks for the long X-ray brightness cycle of 4U 1820-30. One clock has a past period step and a later period of 167.43 days. The other has a steady period decrease. We fit separate Fourier profiles to public daily data through MJD 60157. We hold each clock and profile fixed for later data through 2026 October 19. On 937 later MAXI 2–20 keV measurements, the weighted root mean square errors are 0.30084 and 0.39307 photons per square centimetre per second. The fixed-clock error is 23.5% smaller, and 7.4% below a constant early-mean forecast. The fixed clock has smaller MAXI errors in all 18 sensitivity fits. Removing any one full test cycle keeps the mean squared error difference positive. Paired cycle-block resampling and joint early/later resampling give positive descriptive intervals. Swift BAT gives a small, unresolved difference and has strong early weight concentration. A free-period fit to the later MAXI data gives about 170 days, without a period uncertainty. Thus the result does not establish a true period of 167.43 days. It compares fixed central forecasts and does not propagate the published clock parameter errors. It does not select a physical cause or exclude either full model family. The package preserves input data, provenance, code, daily forecasts, checks, tables, and figures.

4U 1820-30 · MAXI · Swift BAT · superorbital cycle · X-ray binary · forecast test · public data