There seems to be only handful of videos showing regressions actually, i) Shaky_Fireworks_3840x2160_2997fps.y4m (UGC, some motion + shakes + ISO noises) ~4% across metrics.
ii) other UGC videos showing mixed results with 1-2% loss on many metrics, and some metrics
iii) SolLevanteDragons_1920x1080p24_hdr10.y4m (HDR and SDR versions, SCC NTFLX sequence, high motion dragons, hard to encode), ~4% loss
iv) SparksWelding (HDR, small sparks motion, hard to encode again), 2-3% 
v) NocturneRoom_3840x2160_60fps_hdr10.y4m (HDR again, bit-dark more bluish sequence), 5-6% loss, DE2k is outliner at low_qps giving 43% gains. 
vi) Slides1_1920x1080_30fps_8bit_420.y4m (SCC, slides animation white screens with transition animations), 3-4% loss
vii) Debugging_1920x1080_30fps_8bit_420.y4m (VSCode SCC) 0.5-3%
vii)RedKayak_360_2997.y4m (high motion kayaking), ~2-3% PSNR,
viii) DinnerSceneCropped_1920x1080_2997fps_10bit_420.y4m, 3-8% across sequecnes, worth noting that at high QPS (118..188) have losses ~<1%
ix) Tango_3840x2160_5994fps_10bit_420.y4m, low Qps (78..118) showing ~3% loss. 
x) Neon1224_3840x2160_2997fps.y4m (bit dark video with neon lights and dancing with some level of motion), 0.7-3%. 
Rest of the clips looks very well, 
Notably images are lossless. 
PS: Numbers are quick glance over the fat spreadsheets. 
Quick previews can be seen at https://codecwar.com/gallery 
AWCY: https://beta.arewecompressedyet.com/?job=rav1e-analysis-refactor%402023-06-10T09%3A57%3A02.006Z&job=rav1e-analysis-refactor-minmax%402023-06-10T16%3A57%3A31.024Z