Slate React rerender breadth is mostly local and useSlate is the only broad hook left

Problem

The next roadmap decision depended on one question:

Is slate-react still broadly invalidating on edits and selection changes, or is the remaining huge-document typing gap already downstream of something else?

Without a real breadth lane, that question was still hand-wavy.

Symptoms

• 5000 and 10000 huge-document typing still lagged legacy chunking
• earlier runtime cuts had improved things, but there was no kept benchmark for the #5131, #3656, and #4141 families
• it was still too easy to blame “React rerenders” in the abstract

What Didn't Work

• treating useSlate() complaints as proof that the whole runtime still had broad invalidation
• using only user-facing huge-document latency numbers to infer render breadth
• reaching for islands before proving whether descendant invalidation was actually still red

Solution

Add a kept breadth lane in slate-react-rerender-breadth-benchmark.tsx and wire it as pnpm bench:react:rerender-breadth:local.

The lane measures three issue families:

1. #5131 selection-driven breadth
2. #3656 many-leaf breadth inside one block
3. #4141 deep ancestor breadth on one low-level edit

Measured result:

• selection lane:
  • 20 selection changes -> 20 useSlate() rerenders
  • 20 selection changes -> 20 useSlateSelection() rerenders
  • unrelated top-level block slices rerender: 0
• many-leaf lane:
  • edited leaf rerenders: 1
  • sibling leaves rerender: 0
  • parent block rerenders: 0
• deep ancestor lane:
  • deep edited leaf rerenders: 1
  • rerendered ancestors: 0
  • sibling branch rerenders: 0

Why This Works

The benchmark separates two different stories that were getting lumped together:

1. Broad hooks are broad by design. If a component uses useSlate(), it is asking for whole-editor reactivity. The selection lane proves that still happens.
2. Descendant invalidation is mostly local. The many-leaf and deep-ancestor lanes show that one edited leaf does not fan out into sibling leaves, parent blocks, or ancestor chains anymore.

That means the runtime is no longer failing on the old breadth complaints in the way legacy Slate did.

Prevention

• benchmark breadth directly instead of inferring it from latency alone
• separate “broad hook by contract” from “runtime invalidation bug”
• do not escalate to semantic islands until sibling-leaf and ancestor-chain breadth are actually measured and shown to still be red
• treat useSlate() rerenders as hook-contract debt, not automatic proof that the whole runtime is still broadly invalidating

Related Issues

• benchmark-candidate-map.md
• 2026-04-11-slate-v2-rerender-breadth-batch.md