We present the halo number counts and its two-point statistics, the
observable angular power spectrum, extracted for the first time from
relativistic N-body simulations. The halo catalogues used in this work are
built from the relativistic N-body code gevolution, and the observed redshift
and angular positions of the sources are computed using a non-perturbative
ray-tracing method, which includes all relativistic scalar contributions to the
number counts. We investigate the validity and limitations of the linear bias
prescription to describe our simulated power spectra. In particular, we assess
the consistency of different bias measurements on large scales, and we estimate
up to which scales a linear bias is accurate in modelling the data, within the
statistical errors. We then test a second-order perturbative bias expansion for
the angular statistics, on a range of redshifts and scales previously
unexplored in this context, that is $0.4\le \overline{z}\le 2$ up to scales
${\ell}_{\mathrm{max}}\sim 1000$ . We find that the angular power spectra at equal
redshift can be modelled with high accuracy with a minimal extension of the
number of bias parameters, that is using a two-parameter model comprising
linear bias and tidal bias. We show that this model performs significantly
better than a model without tidal bias but with quadratic bias as extra degree
of freedom, and that the latter is inaccurate at $\overline{z}\ge 0.7$ . Finally, we
extract from our simulations the cross-correlation of halo number counts and
lensing convergence. We show that the estimate of the linear bias from this
cross-correlation is consistent with the measurements based on the clustering
statistics alone, and that it is crucial to take into account the effect of
magnification in the halo number counts to avoid systematic shifts in the
computed bias.