Bias in the ROQ geocent_time posterior

Running the fiducial BBH test for a 128s ROQ basis, there seems to be a bias in the geocent_time:

Full corner plot:

Further more, a pp plot also suggests a bias

where the individual p-values are given by

09:37 bilby INFO    : Key: KS-test p-value
09:37 bilby INFO    : a_1: 0.004353009481190579
09:37 bilby INFO    : a_2: 0.03985168897683587
09:37 bilby INFO    : chirp_mass: 0.6807714838241979
09:37 bilby INFO    : dec: 0.10544070562244892
09:37 bilby INFO    : geocent_time: 0.0032471782742264045
09:37 bilby INFO    : luminosity_distance: 0.8240397996477468
09:37 bilby INFO    : mass_2: 0.048316085706487676
09:37 bilby INFO    : mass_ratio: 0.05841381950543859
09:37 bilby INFO    : phase: 0.8647187964800038
09:37 bilby INFO    : phi_12: 0.13550324406264547
09:37 bilby INFO    : phi_jl: 0.059429838161255764
09:37 bilby INFO    : psi: 0.5781828834354825
09:37 bilby INFO    : ra: 0.13952060698874352
09:37 bilby INFO    : theta_jn: 0.002548388019280191
09:37 bilby INFO    : tilt_1: 0.5544800028886142
09:37 bilby INFO    : tilt_2: 0.16335729310487174
09:37 bilby INFO    : Combined p-value: 1.0659765250158535e-05

Notably, the geocent time (along with spins) is small.