- Work RAM refresh timing seems to be a bit variable and the exact logic for the timing has been unclear
- The exact point when scroll values are latched from the internal scroll RAM when in full screen scroll mode is unknown
- Precise cause of a CPU lockup when configuring a 68K->VDP DMA and then performing a normal access instead is also unknown
- Certain details of sprite rendering that can show up in certain edge cases are also a bit unclear
- The function of some of the VDP debug registers is unknown
There's a lot more like that, but that hopefully gives you the gist. Nuked-MD-FPGA (and its C-based predecessor) will hopefully make a lot of those questions easier to answer since Verilog/C with a lot of unclear signal names is still a lot more easier to interpret than a die photo.