I've checked the numbers for some DACs from Josh's line-up:
| DAC | Chip | Filter group delay | Delay @ 96 kHz |
| Matrix Sabre Pro MQA | ES9038PRO | 35/fs | 364.6 µs |
| Gustard X30 | ES9039SPRO | 33.8/fs | 352.1 µs |
| SMSL SU-1 | AK4493S | 29.2/fs | 304.2 µs |
All numbers are for the linear phase fast roll-off filter. For other DACs, the chip is either unclear (RME ADI-2 DAC FS, three possible versions) or custom and lacks the necessary datasheet (Topping D900, Schiit Yggsdrasil LIM). For the ES9039SPRO, the simulated filter latency is also listed in the datasheet and amounts to 771 µs @ 48 kHz, when group delay would only be 704.2 µs.
I was not sure at first if group delay would be the right metric in this case. However, at least AKM defines very clearly in the AK4493S datasheet that their group delay numbers refer to the delay between the arriving digital signal and the corresponding analog output. But I'm not sure if this is consistent with what ESS uses.
The typical order of magnitude for the temporal resolution of human hearing is
around 10 µs, as far as I am aware. However, the stimuli used in such tests will differ from what we are dealing with here.