Merge branch 'main' into claude/issue-8541-framework-ceiling-raise

Author: os-sales

.changeset/8525-empty-container-override-friendly.md

@@ -0,0 +1,11 @@
+---
+'@object-ui/components': patch
+---
+
+fix(components): `Empty`'s base classes stop beating a caller's padding on desktop, and drop an inert `border-dashed`
+
+The shared `Empty` container carried `p-6 … md:p-12` in its base classes. A caller's unprefixed padding override (`px-3 py-8`, `py-10`, even a full `p-4`) lives in a different `tailwind-merge` variant from `md:p-12`, so `cn()` kept both and the `md:` rule won the cascade from 768px up — "tighten this panel" silently did nothing on any desktop viewport. The responsive default now travels in a custom property (`--empty-padding`: 24px, and 48px from `md`) read by ONE unprefixed `padding` utility, so a caller's plain padding wins at every viewport, while a site with no override renders exactly as before: 24px below `md`, 48px at and above it. Callers that are responsive themselves (`p-2 md:p-4`) keep working.
+
+Visible consequence: the three console sites that already wrote a tighter panel now get it on desktop — the AI chat conversations sidebar's empty state (`px-3 py-8`) and the metadata-admin audit panel's error and empty states (`py-10`).
+
+`border-dashed` is removed from the same string. It set only `border-style`; with preflight's zero `border-width` and no width supplied by any call site or ancestor, it drew nothing at any of the 44 `Empty` sites (measured on the console build), so removing it changes no pixel. No border width is added — a dashed frame around every empty state would be a product-wide visual decision that needs its own card.

.changeset/8530-filter-converter-array-comparand.md

@@ -0,0 +1,35 @@
+---
+'@object-ui/core': patch
+---
+
+`convertFiltersToAST` now refuses a bare ARRAY in comparand position —
+`{ tags: ['a', 'b'] }` — with a `FilterOperatorError` (`INVALID_FILTER` / 400)
+that names the field, prints the comparand and prescribes the spelling that
+works, instead of lowering it to `['tags', '=', ['a', 'b']]` (objectui#8530).
+
+That node was never answerable: the ObjectQL filter AST has no array-equality,
+so `@objectstack/driver-sql` refused it with `400 INVALID_FILTER` from the wire
+and every in-memory matcher (`@objectstack/formula`, `ValueDataSource` since
+objectui#8514) excluded every row. The author found out two layers away, as a
+failed list or an empty one. The refusal now lands at lowering time, where the
+field and the offending value are still in hand — the same treatment this file
+already gives `$regex` and `$not`, the two other shapes it cannot lower.
+
+It is deliberately NOT read as membership. `{ tags: [...] }` and
+`{ tags: { $in: [...] } }` are different statements and the second is already
+spellable; rewriting one into the other would guess at intent and silently
+change which rows a stored view returns — the lenient second contract
+objectui#8514 was resolved against on this same data shape one layer down. The
+error message says so, and names `$in` / `$nin` / `$between` as the spellings to
+use. `$in` / `$nin` / `$between` members, `$and` / `$or` groups and stored
+`ViewFilterRule` values (`in` / `between`) are legitimately arrays and keep
+lowering exactly as before.
+
+Every producer in this repository already spells a multi-value comparand
+`{ $in: [...] }` (measured across the `$filter` literals and record builders
+under `packages/*/src` and `apps/*/src`), so no shipped surface changes
+behaviour; only a hand-authored `{ field: [...] }` now fails at the producer
+instead of the consumer. No ruled contract ever answered that shape — the spec
+leaves an array outside `$in` / `$nin` / `$between` unruled, `driver-sql` and the
+in-memory matchers refuse it, and only a document store's native array-equality
+happened to read it — so nothing a backend was promised is taken away.