Open In progress bug report

Molecule export lanes ignore the registry's footprint and symbol slots: new pages show none, older pages export a stale repo copy

Ray · ·marked in progress by kcknox

The KiCad / Fusion / Altium export lanes on a molecule page read only the page repo. Since 2026-10-26 the model registry builds a molecule's footprint and symbol JSON into its own slots and refuses a repo write that adds or changes those files (REGISTRY_MODEL_FILE_REFUSED). So the lanes never see the registry's files.

New pages show no exports. ray/molecule-me6118a33b3g, created today: all three lanes are greyed with "KiCad: no KiCad symbol or footprint in this page's repo" (Fusion and Altium: nothing to derive from). The registry built both files. registry_check status serves footprint (molecule-me6118a33b3g_footprint.json), symbol, cad and model from build 90545, origin pipeline, metadataComplete true. Every molecule page created after 10-26 looks the same; the other SOT-223 pages published today (for example ray/molecule-bl1117-25cx and ray/molecule-template-sot223-ldo-u) are greyed too.

Older pages export a stale file. ray/molecule-template-sot223-ldo shows its lanes as available ("KiCad: indexed, authored"), indexed from molecule-template-sot223-ldo_footprint.json in the repo. That copy is generatedAt 2026-10-24T00:44:10Z and has a 10.4 x 10.4 mm outline. The board has been 11.4 x 11.4 mm since the v3.1 rebuild, and the registry's footprint slot is from today's build 90483. The KiCad, Fusion and Altium exports on that page carry the wrong outline today (the 8 mm machine-pin grid is unchanged). A release cannot refresh the repo copy, because of the same refusal.

Ask: the export lanes should read the registry's footprint and symbol slots when they exist.

1 Reply

Colby Knox ·

Thanks, Ray. You were right that the lanes ignored the registry. Tracing an actual export showed a second problem underneath it, so here is what changed and one question for you.

What I found

  • The export lanes looked only at the page repo, so they never saw the footprint and symbol the registry builds, exactly as you described.
  • Reading the registry's files is not enough on its own. A molecule's footprint and symbol are Adom molecule JSON (<slug>_footprint.json, <slug>_symbol.json), and no lane converts that format today. The KiCad lane uses its files as .kicad_sym / .kicad_mod directly, and the Fusion and Altium lanes convert from a KiCad pair or an adom-lbr hub JSON. So the "KiCad: indexed, authored" badge on the older pages never meant a working KiCad export: the export would have carried the JSON, whatever its outline.
  • In a Cloud Library, a molecule page expands to the parts it declares as dependencies. The molecule's own footprint is not exported there.

What is changing (on staging now, prod with the next deploy)

  • The lanes see the registry's footprint, symbol and STEP for a page, and an export reads those files from the registry, the same bytes the page serves, never the stale repo copy.
  • On a molecule page, each lane now says why it is not offered: "a molecule's footprint and symbol (Adom molecule JSON) do not convert to KiCad, Fusion or Altium yet". New pages no longer say the repo has nothing, and older pages no longer promise an export that would not work.

The question: what should exporting a molecule page give someone?

  1. The molecule as one placeable part: its footprint, symbol and STEP as a KiCad / Fusion / Altium library entry.
  2. Its parts, as a Cloud Library does today.
  3. Both.

If it is 1 or 3, the files have to become a KiCad pair or a hub JSON somewhere. Either the wiki converts molecule JSON when it exports, or the molecule pipeline builds a .kicad_sym and .kicad_mod as two more slots. If the molecule JSON format is written down (the contract or the pipeline's writer), point me at it and I'll size the converter.

Colby Knox marked this in progress ·

Log in to reply.