Slate History Tranche 4 Execution

Goal

Close packages/slate-history honestly on top of the settled slate core claim:

• keep undo/redo and selection restore anchored to committed transaction truth
• preserve kept legacy + draft history behavior where it still earns its keep
• do not drag retired timing heuristics or wrapper-era assumptions back into core

Current Read

• slate core is now settled enough to unblock tranche 4:
  • 2026-04-19-slate-absolute-api-replan.md
• slate-history now has one direct proof owner landed:
  • packages/slate-history/test/history-contract.ts
• slate-history now also has the second direct proof owner landed:
  • packages/slate-history/test/integrity-contract.ts
• the package owner gate is green again:
  • bun test ./packages/slate-history/test/index.spec.ts
• the package closeout gate stack is green in the same turn:
  • bun test ./packages/slate-history/test/history-contract.ts
  • bun test ./packages/slate-history/test/integrity-contract.ts
  • bunx turbo build --filter=./packages/slate-history
  • bunx turbo typecheck --filter=./packages/slate-history
  • bun run lint:fix
  • bun run lint
• the missing history compare owner is now recovered:
  • bun run bench:history:compare:local
• the first red cluster was real and shared one cause:
  • transaction-owned delete / join / insert-break undo rows were missing history capture because capture still sat on the old write seam
• that owner is now fixed:
  • history capture is anchored to the committed publish seam instead of legacy per-op editor.apply(...) interception
• direct proof-owner coverage is now live for:
  • kept undo / redo parity rows
  • transaction-owned undo-unit capture
  • merge / save / split flags
  • stack-write override seam
  • commit-before-onChange history capture

Current Tactic

• treat slate-history as the first real proof that the settled core commit / transaction model survives contact with a support package
• keep history on the committed capture seam
• keep the explicit skips explicit
• do not reopen deleted legacy timing / fragment rows unless new evidence proves they belong in the kept claim

Rejected Tactics / Pivot History

• rejected: patching failing undo fixtures one by one while the same transaction-owned family stays red
• rejected: treating the missing history-contract.ts file as the main problem when the current package gate is already proving a deeper runtime bug
• rejected: reopening packages/slate/** unless a kept history row proves the bug really lives in core

Remaining Kept-Owner Ledger

None blocking this slate-history lane.

Deferred / carried forward:

• legacy timing-based contiguous / non-contiguous insert heuristics stay explicit-skip unless later evidence proves they earn their keep
• wider deleted delete / fragment rows stay explicit-skip unless later evidence proves they belong in the kept live claim

Perf Owner Status

• live history compare owner is now wired again:
  • bun run bench:history:compare:local
• latest history compare read:
  • typing undo: +29.35ms
  • typing redo: +20.04ms
  • fragment undo: +25.29ms
  • fragment redo: +31.77ms
• current read:
  • still slower than legacy
  • no longer the catastrophic multi-second regression shape
  • back in the low-tens-of-ms band for typing / fragment history ops
• current shared core regression floor still applies if any packages/slate/** file changes:
  • bun run bench:slate:6038:local
  • bun run bench:core:normalization:compare:local
  • bun run bench:core:observation:compare:local
  • bun run bench:core:huge-document:compare:local

Current Gates

• correctness owner:
  • bun test ./packages/slate-history/test/index.spec.ts
• direct proof owner:
  • bun test ./packages/slate-history/test/history-contract.ts
  • bun test ./packages/slate-history/test/integrity-contract.ts
• package closeout gates:
  • bunx turbo build --filter=./packages/slate-history
  • bunx turbo typecheck --filter=./packages/slate-history
  • bun run lint:fix
  • bun run lint
• perf owner:
  • bun run bench:history:compare:local

Next Move

1. treat slate-history as settled enough to stop blocking tranche 4
2. carry its explicit-skip and bounded-perf read forward
3. move to slate-hyperscript as the next package in order

Continue Checkpoint

• verdict:
  • replan
• latest landed API redesign owner:
  • slate-history now captures undo units from the committed publish seam
  • direct kept-row proof now exists in:
    • history-contract.ts
    • integrity-contract.ts
  • slate-history now also has a runnable compare owner again:
    • bench:history:compare:local
• latest hard-cut or demotion decision:
  • do not revive legacy timing heuristics or wider delete / fragment rows by default
• remaining unresolved API decisions:
  • no blocking slate-history package decisions remain in this lane
  • only carried-forward explicit-skip judgment remains
• latest current-vs-legacy compare read:
  • slate-history compare owner is live again
  • typing undo: +29.35ms
  • typing redo: +20.04ms
  • fragment undo: +25.29ms
  • fragment redo: +31.77ms
• drift read:
  • current work still points toward the better API
  • the main drift risk now is reopening already-classified history rows or stalling before slate-hyperscript
• next move after this checkpoint:
  • stop here for this execution owner
  • create or switch to the next tranche/package owner when the user wants

Repeated Continue Rule

• this execution owner is complete
• repeated continue calls against this same owner without a new scope or new contrary evidence should return:
  • replan
• reason:
  • the next honest move changes package ownership
  • more packages/slate-history churn here would be invented work, not progress
• repeat-count:
  • 1
• latest reaffirmation:
  • repeated continue was received against the same completed owner with no new scope, contrary evidence, or blocker change
  • verdict stays replan