Slate v2 migration must take transaction seams from real draft source, not reference proposals

Problem

The tranche-3 slate plan drifted toward a fake source of truth.

A historical reference doc in docs/slate-v2-draft/references/ was being treated like the draft implementation owner, which made the next batch assume a withBatch(...) seam that the real draft package did not even ship.

That is how planning turns into fan fiction.

Symptoms

• The plan said the next draft-backed seam was Editor.withBatch(...).
• The real draft in .tmp/slate-v2-draft/packages/slate/** only had:
  • getChildren
  • setChildren
  • Editor.withTransaction(...)
  • Transforms.applyBatch(...)
• The current slate fixture suite was also red for a separate reason: 108 same-path legacy JSX fixtures had drifted and no longer imported jsx, which made the harness scream about React/jsx being undefined before it could tell us anything useful about the actual wave.

What Didn't Work

• Using a reference/proposal doc as if it were code truth.
• Planning the public seam from memory instead of reading the actual draft package on disk.
• Treating the red fixture suite as proof the runtime seam was broken when the first failure was really fixture import drift.
• Writing a batch plan that tried to recover the wrong public entry point.

Solution

1. Re-ground the wave against the real draft source in .tmp/slate-v2-draft/packages/slate/**.
2. Rewrite the next wave around the seams the draft actually owns:
   • getChildren
   • setChildren
   • Editor.withTransaction(...)
   • Transforms.applyBatch(...)
3. Use focused draft-backed RED slices first instead of importing a whole draft suite and disappearing into horizontal code churn.
4. Restore the same-path legacy JSX fixture files that had drifted so the current index.spec.ts harness became honest again.
5. Recover the minimal public seam in current slate-v2:
   • accessors
   • snapshot/listener helpers
   • transaction seam
   • applyBatch(...)

Why This Works

There are three different kinds of evidence here, and they are not equal:

• reference docs explain ideas
• legacy source explains the old contract
• real draft code explains the new engine we are actually migrating from

When those disagree, the code wins.

The fixture repair mattered too: a red suite caused by missing jsx imports is not a runtime oracle. Fix the harness noise first, then trust the failures that remain.

Prevention

• For slate-v2 migration work, verify the seam in .tmp/slate-v2-draft before naming it in a plan. Do not promote docs/slate-v2-draft/references/** to implementation truth.
• If a current fixture suite explodes on import/pragma noise, fix that same-path drift before using the failures as runtime evidence.
• Pull focused draft RED slices first. Do not import every draft test and then go write all the code.
• If a legacy compatibility name is not in the real draft source, treat it as a separate explicit keep/cut decision, not as automatic draft recovery.