Persistent range-ref projections belong in a React hook, not the headless store

What happened

After the slate-v2 range-ref proof landed, the next obvious step was wiring durable anchors into slate-react-v2 projections.

The first attempt shoved RangeRef support directly into the headless projection store.

That looked small. It was the wrong boundary.

Why it was wrong

The headless store is intentionally simple:

• raw Range projections in
• local runtime-id slices out

Once it started accepting RangeRefs directly, it quietly inherited React-side ownership problems:

• who refreshes when anchor props change?
• who removes stale UI after an anchor disappears?
• who owns invalidation semantics?

That is not core work. That is React integration work.

What fixed it

The honest split was:

• keep projection-store.ts raw-Range only
• add a React hook that adapts RangeRef props into raw Range projections

The hook owns:

• prop-driven refresh
• cleanup on unmount/editor change
• the React-shaped contract that callers actually use

The headless store stays boring. That is good.

Reusable rule

For Slate v2 projection architecture:

• core owns durable anchor semantics
• the headless projection store owns raw range-to-slice projection
• React hooks own prop-driven RangeRef integration

If a headless store starts reading live refs directly, it is probably stealing React work and lying about ownership.