spec: the object document's client-facing type is spec's to own — declare (or explicitly rule out) `titleFormat`, `list_views`/`listViews`, `icon`, which shipped objectui runtime reads today

[os-zhuang]

Carved by the triage seat from the maintainer ruling on objectstack-ai/objectui#5362 (2026-08-20, decision-inbox session, session session_01PjAP6vbcsg2yMtvySPv1Qo): of the three readings that card offered, the maintainer accepted reading 3 — the object document belongs to @objectstack/spec; objectui re-exports/aligns rather than maintaining a hand-written second copy (the same layer split objectui#3074 applied to PageNodeSchema). This is the contract-first slice that must land before the objectui half; ui#5362 is Blocked-by: this card.

The measured gap (from ui#5362, verified on objectui origin/main = bdf8cf76e)

objectui's ObjectSchemaMetadata hand-declares a subset of the object document. Three keys a real object definition carries are absent from it and from @objectstack/spec 17.0.0's dist/, while shipped objectui runtime reads all three:

key	read at (objectui)
list_views / listViews — both spellings accepted at every read site	ObjectView.tsx:1178,1194,1208 · InterfaceListPage.tsx:56 · AppHeader.tsx:501 · useElementDataSource.ts:104 · MetadataProvider.tsx:420
titleFormat	the ADR-0079 object-level display-name resolver (ObjectKanban.tsx:275,295 names the order objectDef.titleFormat → displayNameField → …)
icon	authored on objects; edited by the metadata-admin object inspector (ObjectDefaultInspector.tsx:94)

The gap is invisible at the boundary because DataSource.getObjectSchema() returns Promise<any>.

Scope of this card (spec side)

1. For each of the three keys: declare it on the spec's object schema with the semantics the runtime already implements, or explicitly rule it out (with the runtime read then owed a fix on the objectui side). "Declared by the type the consumer is told to use, silently read anyway" is the shape being ended — either direction is acceptable, silence is not.
2. list_views vs listViews: pick one canonical spelling in the declaration; the consumer-side dual-read becomes a deprecation question for the objectui follow-up, not a spec blessing of both.
3. Whatever ships, objectui's follow-up (ui#5362) derives its client type from the spec export instead of hand-copying — the derivation guard pattern applies.

Tier: contract accept-surface widening ⇒ Clause-②: yes, claude-fable-5.

Refs: objectstack-ai/objectui#5362 (the ruled card, with the full measurement) · objectui#3074 (the PageNodeSchema layer-split precedent) · ADR-0079 (titleFormat resolver order).