That's interesting, it's good to see a latest version in the flesh., thanks for undoing those screws!
Just some observations...
- The board looks nice and clean, you used to get flux residues and fingerprints. I can't see any bodges... ah, it looks like they've bodged an extra (or bigger) capacitor next to U20.
- They've switched to a GOWIN FPGA. The originals were XILINX but I've no idea what the three intervening ones were. I hope, for Florent's sake, that they've stocked up on them this time!
- They're still using the same PCB for I and non-I versions. The isolated DC-DC converter footprint is different to the early versions though. I wonder what they use now (useful to know in case somebody wants to upgrade to the I version). They've also beefed up the optional (non-I) zero ohm resistors bridging the ground planes on the bottom of the board, maybe for improved EMC performance (or possibly, signal integrity).
- There's now a nice clearly marked JTAG header footprint. I notice there's also an unmarked header footprint on the opposite side of the FPGA.
- They've added an unpopulated push-button footprint, probably Reset for debug use?
- There's a new 3 pin through-hole connector footprint near the input regulators, maybe external power during development. [EDIT: No, duplicate serial port connector]
- It's interesting that, after all these versions, they've still retained the footprints for the serial interface. I wonder if they ever used it. [EDIT: Well, if they've added a duplicate connector, they must do during development]
- They've put the dedicated Video Sync decoder IC back, I thought they'd cost-reduced that out. Otherwise they analogue side looks the same, if it ain't broke...
- The Analogue -VE rail regulator is still a linear one. My early version has a switching regulator, with a lot of bodgery around it (I think it has lower overall USB supply current though).