finding(console): the `object` preview sample's `status.options` are bare strings, a shape FieldSchema refuses — masked by the sample's quarantine

[claude]

Found while implementing objectui#6647 (re-spelling that sample's lookup target). Filed rather than folded in — the fix needs a content decision, not a mechanical re-spelling.

apps/console/src/preview-samples.ts, the object sample:

{ name: 'status', label: 'Status', type: 'select', options: ['Draft', 'Open', 'Closed'] },

FieldSchema wants option OBJECTS, not bare strings. Measured against the installed @objectstack/spec 17.2.0, on the real sample with its (deliberate) array fields shape lifted to a record so the parse stops short-circuiting:

ObjectStackSchema.safeParse({ objects: [ ...sample, fields: as a record ] })
 => success:false, 3 issues, all invalid_type:
 objects.0.fields.status.options.0 "Invalid input: expected object, received string"
 objects.0.fields.status.options.1 ... .1
 objects.0.fields.status.options.2 ... .2

Positive control on the option shape itself: an empty {} in that array reports exactly two missing members, label and value — so the wanted shape is { label, value }.

Why it is invisible today

The sample is quarantined in apps/console/src/__tests__/preview-samples-spec-valid.test.ts for its array fields shape, and Zod short-circuits at objects.0.fields before it ever descends into a field. So this defect is REPORTED by nothing: the reverse assertion only asks for one issue, and the array shape supplies it forever. objectui#6647's PR extended that ledger row to name this defect in prose, which is where the record currently lives — but prose is not a gate.

The designer cannot read it either

ObjectFieldInspector's readOptions() does String(o?.value ?? '') per entry, so three string options materialise as three BLANK rows in the options editor. The sample that exists to demonstrate a picklist demonstrates an empty one.

Why objectui#6647 did not fix it

That card's fence was the lookup key and the ledger reason, and re-authoring these options means INVENTING the value codes the sample should demonstrate (draft vs Draft vs something else) — an authoring decision about what the worked example teaches, not a rename pinned by existing evidence. Deliberately left for triage.

Not to be confused with

• The array fields shape on the same sample is DELIBERATE (it covers readFields()'s array branch in the gallery) and must not be "fixed" — see the ledger row's note.
• objectui#6153 is a different tier (options[].description on lookup in @object-ui/fields); objectui#4356 is globalFilters[].options in a filter surface. Neither reaches this file.

---

Generated by Claude Code

[claude]

Context from the domain:ui execution seat (PM session session_01CRJge11jso9TpXRWFt1Z49), filed by the implementer of #6647 / PR #6845. ⛔ Grading and domain:* are the triage seat's.

Why this was filed instead of fixed — the bar it fails is specific

⭐ Re-authoring the options means inventing the value codes the sample is supposed to demonstrate. That is a sample-content decision, not a mechanical re-spelling, so it fails the bounded-in-place-fix bar that let #6647's re-spelling ride. ⛔ Do not dispatch this as "change three strings to objects" — whoever takes it is choosing what a status field's codes should look like in the file that exists to show authors the right shape.

It is measured, not suspected

Found while #6647 established what still blocks the object preview sample. The parse ladder, against installed @objectstack/spec 17.2.0:

document	issues
array fields lifted to a record, lookup key fixed	3 — objects.0.fields.status.options.0..2, invalid_type, expected object, received string

⇒ population is exactly three option entries in one sample, and after #6845 lands they are the only thing this sample's KNOWN_STALE reason names beyond the deliberate array shape.

⚠️ It is masked twice today, which is why it needs a reader

1. Zod short-circuits at objects.0.fields on the sample's deliberate array shape, so the options never get validated in the quarantined path.
2. The designer cannot read them either — ObjectFieldInspector's readOptions does String(o?.value ?? ''), so three string options render as three blank rows. A person opening this sample in the designer sees an empty option list rather than an error.

⇒ nothing in the product or the gates says anything is wrong. The stakes are the ones preview-samples.ts argues for itself: these are the worked example an author — increasingly a model generating metadata — copies.

⚠️ Ordering: #6845 (#6647) records this defect in the KNOWN_STALE reason as part of keeping that ledger honest. Whoever fixes this card must update that reason in the same PR, or the ledger goes back to over-stating what remains — the mirror of the defect #6647 was filed about.

---

Generated by Claude Code