Slot elements round-trip as plain containers

[nathanacurtis]

A slot comes back typed container, and the children binding that made it a slot is dropped.

Evidence

FormGroupHeading:

path	baseline	round-trip
anatomy.children.type	"slot"	"container"
default.elements.children.children	{$binding: "#/props/children"}	(absent)

AvatarGroup:

path	baseline	round-trip
anatomy.items.type	"slot"	"container"

Note props.items does survive with type: slot and maxChildren: 4 intact, so the slot property itself is being created — it is the anatomy type and the element binding that are lost. That internal inconsistency (a spec where props.items.type is slot but anatomy.items.type is container) is part of what to fix.

Explicitly out of scope

props.*.anyOf → preferredValues is a documented deferral in SlotElement.slotSettingsOptions (it needs published-library keys, unavailable for local components). Leave it deferred; do not widen this issue to cover it.

Where

packages/figma-from-specs/src/Elements/SlotElement.ts (promote, bind, configure) and Component.ts (materializeSlots). Whether the loss is on the render or the read side is undetermined — establish that first, since props.items surviving suggests promotion partly works.

Fixtures

• FormGroupHeading — slot plus dropped children binding
• AvatarGroup — slot with surviving props.items, the internal-inconsistency case

Verification

Round-trip both. anatomy.children.type / anatomy.items.type come back as slot, and anatomy agrees with props on every slot in the spec.

Part of #281.

[nathanacurtis]

Fixed in specs-from-figma 3e4afe1 on feat/figma-from-specs.

Cause

Two independent defects, both on the render side.

Component.materializeSlots promoted only the first variant's placeholder frame to a real Figma SLOT node. Sibling variants kept plain frames carrying a slotContentId reference. Since the reader classifies anatomy from node.type === 'SLOT' and takes the first variant it reads, a slot came back as container whenever that variant was one of the frames — and Children.evaluate requires a real SLOT node, so the children binding was dropped and the frame's literal child layers were emitted instead. That is where anatomy.egdsAvatar came from.

Separately, PropConfigurations.apply passed every configured value to setProperties in one call. That call is atomic, so icon — an INSTANCE_SWAP property configured by name rather than by component key — threw and took every other property with it. Avatars rendered at their component defaults, which masked the per-variant slot content as identical.

Fix

• Every variant's placeholder is promoted to a real SLOT node and bound to the property the first one minted, with the surplus property each extra createSlot() mints deleted from the set.
• Configured values are checked against their own definition type before batching, so an unusable value is skipped on its own rather than voiding the instance.

Verification

Round-tripped both fixtures against their current baselines.

AvatarGroup:

section	before	after
api	2 differences	0
variants	5 differences	0
examples	7 differences	6

FormGroupHeading: anatomy.children.type returns as slot and the children binding survives.

Both criteria in this issue hold: slot anatomy types round-trip, and anatomy agrees with props on every slot.

Remaining, not part of this issue

• propConfigurations.image is dropped — no matching property on the instance.
• egdsAvatar.styles gains explicit width/height tokens absent from the baseline. The rendered sizes are correct; the readback adds them.
• props.*.anyOf → preferredValues remains deferred as this issue specified.