Open general

Component description should be datasheet-generic, and the GLB is Y-up where the viewer is Z-up

Ray · 29d ago

Filed at John's request, from the Hydrogen thread, so the agent that built this page can pick up both fixes.

Note on scope: this page is private, so I could read its metadata and brief but not the rendered readme body. The two items below are the rules that were missed, not a line-by-line review of the prose.

1. The description is a catalog record, not the board's notes

The title, brief and identity-header summary must read like the datasheet's own description and make sense to an engineer who has never seen the project that first needed the part. Lead with manufacturer, exact MPN, function and package.

Reference designators, chosen component values, rail names, selection rationale and experiment progress belong on the consuming board or molecule page. A short Related link to that project is useful, but it must not turn the summary into that project's progress report. Label application examples as examples rather than universal requirements.

The current brief, "TI adjustable supervisor; manufacturer originals and source library under validation", is fine on the audience test, though "under validation" is run status rather than a property of the part.

2. The GLB is Y-up, the viewer is Z-up

step2glb writes glTF, which is Y-up. Adom is Z-up, with the seating plane on XY. The viewer may already apply an axis conversion of its own, so a blanket 90 degree rotation can land twice, and a part that reads as wrong is usually standing on edge rather than mis-scaled.

Inspect the GLB node transforms and the viewer's own import convention, and never correct an orientation problem by moving only the camera. Validate before rendering anything:

adom-chipfit check --footprint part.kicad_mod --glb part.glb

A successful conversion and a sane bounding box establish neither pin 1 nor the seat plane.

Related: this page has no hero registered (hero_path is null). Heroes must come from Hero Studio or publish fails with HERO_STUDIO_REQUIRED.

Why it happened, and what changed

Both rules already existed, in wiki-component (in adom-wiki-skillpack), under "Write for every user of the component" and "Component GLB, Z-up and hero workflow".

The gap is that the skill an agent actually loads for "make a component page" is adom/adom-component-page, whose trigger words are literally "make a component page, publish this part". It names wiki-component as a companion but carried neither rule inline, so an agent following it top to bottom never learned either one.

Both are now proposed inline there, each pointing back at wiki-component for the full treatment: https://wiki.adom.inc/adom/adom-component-page/prs/1

Two things that PR does not fix, both owner decisions rather than mine:

  1. Discovery does not surface any of this. Searching "component wiki page creation" returns fusion-bridge, step2glb, fusion-bridge-macos, fusion and an unrelated resistor page, with no component-page skill in the top 5.
  2. Three overlapping skills answer this one question: adom-component-page (Barrett), adom-hardware-component-publish (Arav), and wiki-component (inside John's adom-wiki-skillpack). That is the likeliest reason an agent used none of them.

2 Replies

Ray · 29d ago

The discovery half of this is now filed separately as adom/wiki#197 (https://wiki.adom.inc/adom/wiki/issues/197): an exact trigger-word phrase does not return its own page, and results are not ordered by the rank they report. That is why an agent had three relevant skills available and surfaced none of them.

John Lauer · 29d ago

Verified with adom-wiki 1.7.39, Pup 2.0.457 and Adom Bridge 2.1.96 on Windows, 10 September 2026.

The page summary and README introduction now describe the Texas Instruments part generically: adjustable voltage supervision, programmable reset delay, manual reset, open-drain output and WSON-6 package. Validation history and the dated sourcing snapshot were retained in VALIDATION.md instead of the catalog introduction. Garbled characters in the README were corrected. Prose commit: b60ff7d72b0d02cb424b35f3f65f353f89c8b9a1. The authenticated Pup page displays the updated content.

The selected component model is TPS389001DSER.adom-reference-v2.glb, SHA-256 b9e95ed3554b719795d4f91180adaf484054485575c5881dffb212b92d4820d7. It already uses the separately documented Z-up derivative, not the original Y-thin GLB. Rechecked the original STEP, original GLB, orientation derivative and selected annotated model against their provenance hashes; all match. The rigid mapping is Adom X=source Z, Adom Y=source X, Adom Z=source Y+0.025 mm. Manufacturer originals remain unchanged.

Four LODs and their manifest are now published on this component page. The shared adom-component-lod widget lists their sizes, geometry metrics, hashes and review states without loading another 3D model. Same-camera top and underside comparisons show the MPN at high/medium and six terminal bodies at every level. The live authenticated page rendering was screenshot-verified. The attached comparison concerns display fidelity only.

This issue remains open: native seating/pin-1 qualification is still incomplete, and the original model's measured lead lengths remain 0.05 mm below the drawing minimums. No registered hero image has been added; the page currently uses a billboard plus its native 3D viewer. The linked component-page skill PR and wiki discovery issue remain separate work; their completion is not claimed here.

lod-tps389001dser-contact-sheet

tps389001dser-lod-pup

Log in to reply.