Service channels and live stores

Purpose

Compile the non-editor references that mattered most to the overlay lane.

Strongest evidence

• VS Code decorations and comment threads are separate controller/channel surfaces.
• VS Code also keeps editor shell, widgets, marker diffs, and the view model in separate layers.
• TanStack DB uses normalized collections, live queries, and useSyncExternalStore-driven snapshots.

What this means

1. Typed channels beat one generic bucket

VS Code proves mature editors do not funnel:

• decorations
• comments
• commands
• other UI channels

through one generic surface.

That directly supports keeping Decoration, Annotation, and Widget split.

2. View-model separation is a real perf architecture move

VS Code does not ask one render layer to own:

• model state
• marker diffs
• content widgets
• overlay widgets
• visible-line projection

It splits them.

That matters because it is the strongest non-rich-text proof that typed lanes and a detached view model are performance architecture, not just API aesthetics.

3. Store/controller APIs beat rerender-time arrays

TanStack DB shows the cleaner React posture:

• explicit collection/store identity
• stable snapshot reads
• explicit subscription

That is a much better analogy for annotation metadata than array-registration hooks.

Take for Slate v2

• use typed overlay lanes
• prefer store/controller surfaces for annotation metadata
• keep the public API honest about ownership instead of hiding it behind hooks
• if Slate v2 ever wants a stronger field-best perf claim, a deeper view-model or invalidation-routing layer is the next serious move, not another generic React refactor