Deburring is pretty hopeless with a board in this shape.
Second try: we mill slowly (0.2 mm/s), add oil, and even try to grind off the burr in the end:
This is already a bit better, but still with way too much burr. Even after manually chasing it, it makes contacts all over the place. Also, the attempt to bring some order into this with a grinding stone resulted in traces being torn and even one large pad to be lifted.
The oil helped, so we use it again. And we now use an endmill designed for the task. This is what the board looks like right after milling:
The oil not only improves the burr situation, but it also prevents the dust from flying around.
This looks a little ugly, but one swift swipe with a paper towel and a few blows of contact cleaner yield this beauty:
There are still a few small problems: flexing of the board prevented any milling from happening in the lower left corner. This could be resolved by trying to mount the board more level, by increasing the milling depth (here, it's 4.5 mil or 0.114 mm), or by scanning the surface and adjusting the Z position accordingly.
Some cuts that should be collinear show a "stair" pattern. This may be because the machine was reset and restarted in the middle of the job.
Finally, the 12 mil bit worked fine for the 0.65 mm spacing of the TSSOP component, but it is too wide for some of the finer structures. In particular, the pads on the left-hand side of the SOT-523 component in front (0.5 mm pitch) were not separated. We can fix this by using a smaller mill or by pretending the mill is thinner, and cutting off more copper.
Let's zoom in for a closer look:
The bit used is a 12 mil "stub" endmill from Performance Micro Tool. Milling speed was 0.5 mm/s. Total job time was 74.5 minutes. I haven't tried yet how fast this can go before breaking the mill. The Gerbers are from KiCad and the toolpaths were generated with cam.py.