The leg link that wouldn't print a round hole

The leg links (plate W8) have a thin tab with a 5 mm hole at the tip, and round it, on the underside, a shallow 10.2 mm recess that seats the ankle spacer. It's half a millimetre deep. It took three prints and a fourth design.

Three goes

The first print used plain white supports, and the pocket's underside came out rough. So every supported part in the file switched to supports topped with Bambu's Support For PLA/PETG, a material that doesn't bond to PLA and peels off clean.
The second print came off the plate nicely except for the recess. Its support was only a thin ring round the open hole, so the roof's loops dragged and the hole came out triangular.
The third print blocked support there, so the roof would bridge over. A 10 mm bridge is too far: it sagged and tangled.

What the file does now: it closes the bottom of the hole with a 0.24 mm skin, in the print file only. With the hole closed, the support under the recess becomes a full disc of support material, and the roof prints straight onto it, the same way the rest of the tab's underside does. After printing, push the skin out with a craft knife or a 5 mm drill bit turned by hand. The lesson: don't bridge a roof you can see if support material can hold it up.

We checked it in the G-code before printing: support all the way to the centre up to z 2.36, then the roof at z 2.48.

Looking for the same mistake everywhere

If one part had a flat ceiling round a hole, others probably did too. So every part was scanned in its print pose, and every sliced plate checked, for four kinds of trouble. The scripts are kept with the print file so they can run again after any change.

Ring roofs, like the leg link's: 32 found, all screw counterbores or the head plate's big ring. Every one already sat on a full pad of support.
Slender support towers the nozzle could knock over. The head pitch bracket (G3) had two 12 mm pencils of support holding up tiny corner fillets that print fine without them. The hip brackets (W6) stood their corner tabs on 21 mm hollow tubes; grid-style support turns those into solid columns about five times stiffer.
Support feet that weld on, where support stands on the part in the part's own colour with no gap. Inside the hood alone there were 83 mm² of white tree tips like that. Support now keeps a gap where it lands on a part.
The prime tower. The two head brackets each trimmed their own top layers, leaving steps of 0.008 mm between them. The tower printed on every step and grew to 116 mm square, off the bed. Turning precise Z off for those two parts took the plate from 4 h 47 min to about 3 h.

Checking our own edits

One more check came out of the day: a mesh audit. It compares every part in the print file with Luwu's original STL, by triangle count, area, volume and size, to prove we hadn't changed their parts by accident. All 23 of theirs were untouched. For the parts we had edited, it checks for loose pieces, broken edges and walls thinner than 0.8 mm. It found two loose 5 mm³ scraps of the hood's old screw boss floating above our rebuilt one. The slicer had merged them, so nothing printed wrong, but they're gone now.

Stringing on the hip outputs

The hip outputs (W7) came off the plate fine but with clear wisps on them. Their supports were made entirely of support material, so the printer swapped nozzles on almost every layer: 230 times on one plate. Using support material only for the top and bottom layers of the support cut that to 60. The wisps brush off without marking the part. Damp support filament makes it worse, so that spool now lives in the heated dryer.

That nozzle-switching problem turned out to be bigger than one plate. Two days later it changed how the whole file prints.