And a short 16 days after I submitted the gerbers to Seeed, the boards are back!
3 boards per 5x5cm panel, 14x50mm per board with 2mm of milling between each sub-board. The end slots are 10mm long and the mid slot is 18mm long. 10 0.6mm drills per tab tangential to the board with a 2mm tRestrict and bRestrict keeping the ground planes from bleeding. The middle slot was drawn on the dimension layer and it also exists on the milling layer, so I submitted the GML that their cam processor generates for whatever it was worth.
I did the basic board as a single unit, smashing all the components and deleting the part names and replacing them with manually laid up text. Once I was happy with the single board I opened a new blank board and imported the single board into it 3 times and drew up the perforations and mills by hand. If I didn't smash out the part names, R1 would become R7 on the 2nd board and R14 on the 3rd board - which isn't a game stopper, but it looks funny. Manually placed holes were funny after importing, so they were added after the panel was combined.
I drew the initial square-edged outline in the dimension layer and then hit all the internal corners with a 1mm radius to give the milling tool a break. The panels are sturdy enough that I used my bench vise to break them apart, but my angle cutters are strong enough to snip through the remaining perforations on the tabs. A cleanup with a sanding block and they're golden.
Judging by the density I've seen on matseng's panels I'm probably losing more space than I need to to the milling, but these seemed much easier to break apart soooooo. Eh.
The silkscreen is particularly misaligned on this run, but it's not going to kill me.
A quick pass with the continuity meter and a netlist tells me that at least the few boards I checked are all working as I designed, which is a relief. Now I just have to wait for the rest of the components to show up and I have a stack of 30 blinkenlights to make!