Smoothed activity-only component of the preferred model of SM21, together with the Keplerian model components. Top: 100 random draws from the posterior describing the planet b Keplerian are also shown, to illustrate that this effect holds true across the posterior, and not simply for one point estimate. The light gray solid line shows the full activity-only model component, and the darker gray shows this model averaged over a (randomly chosen) 11.2 day timescale. (Note that the same pattern holds when choosing a slightly different smoothing timescale; i.e., this is not a result of aliasing.) Shaded gray regions indicate where there are observations. Bottom: same as the top, but with a zoomed-in y-axis. Takeaways: the activity-only component changes suddenly in windows of time where there are observations. When the activity-only component is averaged over shorter-timescale variations, the GP contributes to the fit on timescales similar to the Keplerians, even interfering destructively at some times. This casts doubt on the reality of the Keplerian signals reported in SM21, indicating that they may be favored because of overfitting.