Slate React v2 projection proof must split range semantics from the React overlay store

What happened

The first honest decorations and annotation-anchor proof for slate-react-v2 looked like it wanted one big new subsystem.

That was the trap.

There are two different jobs here:

• convert a logical Slate Range into local text slices keyed by runtime id
• subscribe React components only to the local overlay slices they care about

Trying to do both in one package boundary would have recreated the old blur:

• slate-v2 would start owning render-time overlay state
• or slate-react-v2 would start guessing document semantics

What fixed it

The proof became clean only after the seam was split explicitly:

• slate-v2 owns pure range semantics:
  • Editor.projectRange(editor, range)
• slate-react-v2 owns the overlay projection runtime:
  • createSlateProjectionStore(editor, source)
  • useSlateProjections(runtimeId)

The source function stays logical:

• it derives ranges from committed editor snapshot state
• or from external decoration state plus an explicit refresh()

The store then projects those logical ranges into local per-runtime-id slices.

Why this mattered

This cut fixes the actual slate-react pain instead of hiding it:

• local overlay updates stop forcing broad rerenders
• selection-derived annotation overlays can track committed selection changes
• core stays document-first
• React stays subscription-first

It also keeps the next step obvious:

• persistent comment anchors need range refs or bookmarks later
• they do not belong in the first projection proof

Reusable rule

For Slate overlay architecture:

• core owns logical range meaning
• React owns overlay caching and subscription breadth
• DOM owns geometry later

If a decorations or annotations design cannot say those three lines cleanly, it is not done.