rebuild the library on adom-lbr 2.32.0. The published copy was built by 2.18.0, which imported every KiCad roundrect and oval pad as a SQUARE EAGLE smd; this carries the real pad shape across (0 -> 6 rounded pads) and stops dropping geometry drawn on KiCad User.N layers. Attributes, connects, pads and the pin-1 marker are gated against decreasing and are unchanged.
3D: replace the model to match the corrected SOT-23-6 footprint. The page shipped a 3-lead SOT-23 body for a 6-lead part, which no longer depicted the part after the footprint was fixed from its datasheet. Model is the exact one the new footprint references in the KiCad official library (CC-BY-SA 4.0), the same provenance page.json already records; .glb regenerated from it via service-step2glb.
footprint: the part is a 6-lead SOT-23 (datasheet: 6 pads at 0.95mm pitch, and the symbol has 6 pins) sitting on a 3-pad SOT-23 footprint. Chosen from this part's OWN datasheet, not inferred from a sibling - the Richtek SOP-8-EP parts alone carry three different exposed-pad sizes, so family inference would have been wrong. Rebuilds the .lbr, re-derives pin 1 and pads.json.
footprint: the part is a 6-lead SOT-23 (datasheet: 6 pads at 0.95mm pitch, and the symbol has 6 pins) sitting on a 3-pad SOT-23 footprint. Chosen from this part's OWN datasheet, not inferred from a sibling - the Richtek SOP-8-EP parts alone carry three different exposed-pad sizes, so family inference would have been wrong. Rebuilds the .lbr, re-derives pin 1 and pads.json.
distributor part numbers (digikey=296-47238-2-ND, lcsc=C116592). page.json only: this page's .lbr could not be rebuilt because the symbol pin numbers do not resolve against the footprint pad names (connect-pad-unresolved), and publishing a library that fails validate would be worse than leaving the old one. Once the footprint is fixed a rebuild will carry these into <technology> automatically.