finding(data-objectstack): 2 of 14 adapter meta call sites address the type segment in the plural, working only because the server folds plural to singular

[yinlianghui]

Observation-class, filed unassigned per Prime Directive #10. Tripped over while tracing the real app-list endpoint for #4887 (PR #4937); not fixed there — out of that card's scope, which was six comments only.

Measurement (worktree off objectui origin/main @ 25c8007, framework origin/main @ 49a49a739)

packages/data-objectstack/src/index.ts has 14 client.meta.* call sites. Twelve pass the metadata type in the singular, one passes a caller-supplied category variable, and two pass a plural collection spelling:

line	method	type argument
3258, 3421	getItems	'view'
3298, 3633	getItem	'view'
3378, 3556, 3655	saveItem	'view'
3506	getItems	'mapping'
3846	getItem	'app' — probeAppAccess
3891	saveItem	'dashboard'
4272	getItem	category (dynamic)
3786	getItem	'apps' — getApp
3864	getItem	'pages' — getPage

The two plural sites resolve today only because the server normalizes: RestServer.metaTypeSingular (packages/rest/src/rest-server.ts:1500-1503) maps through PLURAL_TO_SINGULAR from @objectstack/spec/shared, pinned in packages/spec/src/shared/metadata-collection.test.ts:283 (expect(PLURAL_TO_SINGULAR['apps']).toBe('app')). So nothing a user hits is broken.

Why it is still worth a row

getApp (:3786, plural) and probeAppAccess (:3846, singular) address the same metadata type sixty lines apart, and probeAppAccess's own comment argues the singular spelling is deliberate ("the address objectstack#8013 pinned its cases against, and the same one MetadataProvider reads items by"). MetadataProvider does read the singular (TYPE_BY_STATE_KEY = { apps: 'app', … }, providers/MetadataProvider.tsx:72). One file, one type, two spellings, and the reasoning for one of them is written down while the other is silent — which is how #4887's six comments came to name a spelling nothing calls.

The drift is unexercised rather than latent-broken, but it is load-bearing on a server-side fold, so it is only as durable as that fold: the two sites are the ones that would fail if the normalization were ever narrowed, and nothing in this repo asserts either spelling for them (appAccessProbe.test.ts:94 pins the singular for probeAppAccess only).

Not proposing a direction here — canonicalizing to the singular is the obvious cheap move, but whether the adapter should also grow a pin (so a new call site cannot pick the plural) is a maintainer call about gate strength, same family as #4894 / #4902.