Time averaged diagnostics issue

[yopology]

 I have being using time-averaged diagnostics to reduce the noise. The PICMI snippet is below: 

ave_field_diag = picmi.TimeAveragedFieldDiagnostic(
name = 'time_ave_field_diag',
write_dir = 'diags',
grid = self.grid,
period = self.diag_steps,
warpx_file_min_digits = 8,
warpx_time_average_mode = 'dynamic_start',
warpx_average_period_steps = self.diag_ave_steps,
# no need to output rho, as rho = rho_ions + rho_electrons
data_list = ['B', 'J', 'E',
'rho_electrons', 'rho_ions',
'rho_electrons_bg', 'rho_ions_bg'],
warpx_particle_fields_species = ['electrons', 'ions',
'electrons_bg', 'ions_bg'],
warpx_particle_fields_to_plot = field_plot_list,
warpx_format = 'openpmd',
warpx_openpmd_backend = 'h5',
)

I noticed that the second output frame (the one after the initial condition) has a systematic error of being a few percent too large. It looks like an inconsistency in the number of snapshots that are included in averaging. All other frames look fine to me.

I can correct it by using the total number of particles as a reference, as it should be conserved in my simulations. I was wondering if anyone has noticed this before?