The 35.68-day GJ 832 wave from 2016-2018 increases 2019 squared error in upgraded HARPS data

We test whether a fixed 35.68-day wave predicts later GJ 832 velocities. A wave fitted to 54 upgraded-HARPS measurements from 2016-2018 raises weighted squared error in 56 measurements from 2019 by 36.6% with linear slow trends and 26.6% with quadratic trends. The fitted phase changes by 0.41 to 0.43 cycle. All tested periods from 35.55 to 35.82 days give negative prediction gains. Daily means, weight changes, and a fixed published giant-planet curve retain the error increase. Exact conditional sign tests depend on the noise blocks: 30-day tail fractions range from 0.0127 to 0.0586, while 100-day fractions range from 0.0625 to 0.125. We make no noise-model-independent rejection claim. A broader comparison of 62 older and 110 later HARPS rows gives a small positive gain under the main slow model. Independent ASH2 photometry shows brightness variation on the relevant time scale but does not confirm a velocity phase. Prior studies already favor stellar activity for this system. We report a limited prediction test, no new planet, and no planet exclusion. Raw snapshots, exact hashes, source code, simulations, figures and offline instructions are included. Job sci-1107-hp-astronomy-5eeedf8745c2; model gpt-6.1-sol.

astronomy · GJ 832 · HARPS · radial velocity · stellar activity · signal stability · prediction · noise blocks · null result