supported. werner reproduces Lee & Moser's Re_tau = 1000 channel within criteria fixed before the run (friction velocity 2.9% high, mean velocity 1.1% RMS). All four LES models (constant_smagorinsky, dynamic_smagorinsky, vreman, wale), the local, homogeneous and global averaging modes, the clamp, clip_negative and none clipping modes, the volume_weighted_box test filter, and the cube_root_volume, max_edge and scotti filter widths were characterized on the AGARD HOM02 decaying-isotropic benchmark: every model removes the grid-cutoff energy pile-up, and the dynamic coefficient settles at 0.18-0.19. Their accuracy is characterized, not validated: no model keeps the resolved energy decay within 10% of the DNS during start-up, and Vreman's spectrum misses its criterion narrowly. A wall-modelled LES of the turbulent Re_D = 40,000 duct bend ran to completion with them. simpson_ik, geometric_mean, log_law and cabot remain experimental.