Tested consecutive 2027 schedules give fewer local period errors for GJ 887 than season schedules under two fixed noise models

We compare future observing schedules for the weak 2.21661-day radial velocity signal reported around GJ 887. We use 293 nightly times and formal errors from public HARPS and ESPRESSO tables. We add 24, 48, or 96 visits in 2027. Each schedule uses consecutive feasible nights or dates across one season. Two historical noise models remain fixed. New visits have assumed independent noise of 0.8 m/s and activity amplitude of 2.6 m/s. We inject a 0.37 m/s signal at 2.21661 or 2.21027 days. Each phase uses 2000 noise draws, with eight fixed phases. Every trial selects the largest likelihood score over 2.15 to 2.30 days. At 2.21661 days, 96 consecutive visits give mean period errors of 28.14% and 24.49% in the two models. Season schedules give 37.47% and 33.17%. The paired reductions are 9.33 and 8.68 percentage points, with Monte Carlo standard errors of 0.40 and 0.38 points. Consecutive schedules give lower mean errors at every tested visit count and both injected periods. An information calculation links this result to the sampling of smooth correlated noise. The experiment simulates old and new noise. It does not condition on measured velocity residuals or establish that a planet exists. The package preserves source bytes, hashes, dates, code, trial records, figures, and numerical checks.

GJ 887 · radial velocity · observing schedule · correlated noise · period recovery · public astronomical data · simulation