PRIMER & JADES reveal an abundance of massive quiescent galaxies at 2 < z < 5

We select a mass-complete sample of 225 quiescent galaxies at $z>2$ with M_* > 10¹⁰\ M_ from PRIMER and JADES photometry spanning a total area of 320 sq. arcmin. We restrict our analysis to only area with optical coverage in three $HST$ ACS filters, and provide evidence that this is important for selecting the most complete and clean samples of $z>2$ massive quiescent galaxy candidates. We investigate the contamination in our sample via $JWST$ NIRSpec spectroscopic validation, $Chandra$ X-ray imaging, and ALMA interferometry, calculating a modest total contamination fraction of 12.9-3.1+4.0 per cent. The removal of $HST$ data increases star-forming galaxy contamination by 10 per cent and results in a 20 per cent loss of candidates recovered from $HST$+$JWST$ data combined. We calculate massive quiescent galaxy number densities at $2<z<5$, finding values three times larger than pre-$JWST$ estimates, but generally in agreement with more-recent and larger-area $JWST$ studies. In comparison with galaxy evolution simulations, we find that most can now reproduce the observed massive quiescent galaxy number density at $2<z<3$, however they still increasingly fall short at $z>3$, with discrepancies of up to 1 dex. We place 14 of our $z>3$ massive quiescent galaxies on the BPT and WHaN diagrams using medium-resolution spectroscopic data from the EXCELS survey. We find a very high incidence of faint AGN in our sample, at a level of 50 per cent, consistent with recent results at cosmic noon. This is interesting in the context of maintenance-mode feedback, which is invoked in many simulations to prevent quenched galaxies from re-igniting star formation. To properly characterise the evolution of early massive quiescent galaxies, greater coverage in optical filters and significantly larger spectroscopic samples will be required.