Slate v2 Editing Epoch Legacy Timing Recovery Audit

Verdict

Recover the legacy timing discipline, not the legacy Editable monolith.

The legacy file is valuable because event ordering and browser caveats lived in one coherent control flow. Slate v2 keeps that knowledge, but assigns it to explicit owners:

txt

input-router
  -> EditableEditingEpochKernel
  -> selection-controller
  -> mutation-controller
  -> dom-repair-queue
  -> slate-browser generated proof

No legacy public state, Transforms.* teaching path, or browser-native structural editing comes back.

Recovery Matrix

Legacy timing rule	Legacy location	V2 owner	Status	Evidence
Attach native `beforeinput`, not React's polyfilled event.	`../slate/packages/slate-react/src/components/editable.tsx` native listener comment around React issue 11211	`packages/slate-react/src/editable/input-router.ts` and `components/editable.tsx`	Recovered	Native `beforeinput` and `input` listeners are attached directly; richtext generated destructive rows pass cross-browser.
Flush queued `selectionchange` before `beforeinput` because IMEs/extensions may set selection immediately before input.	legacy `scheduleOnDOMSelectionChange.flush()` / `onDOMSelectionChange.flush()` before target sync	`components/editable.tsx` beforeinput path	Recovered	Current beforeinput still flushes both queues before resolving native/model ownership.
Do not import target ranges for ordinary delete before delete commands decide the deletion range.	legacy delete-target-range comment	`selection-reconciler`, `model-input-strategy`, `editing-epoch-kernel`	Recovered and strengthened	Destructive keydown is model-owned; duplicate beforeinput is skipped inside the epoch; generated destructive gauntlet covers delete after paste and repeated word-delete.
Expanded selection plus delete should delete the expanded selection even when the input type looks directional.	legacy expanded delete branch	`getEditableCommandFromBeforeInputType`, `mutation-controller`	Recovered	Backspace/Delete selected-range rows pass cross-browser in richtext proof.
Word delete uses editor model commands, not browser DOM mutation as final truth.	legacy `deleteWordBackward` / `deleteWordForward` branches	`keyboard-input-strategy`, `model-input-strategy`, `editing-epoch-kernel`	Recovered and hardened	Persistent native word-delete row, generated destructive gauntlet, and persistent soak are green.
Native single-character insert is allowed only for narrow safe text cases.	legacy `native` insertText branch with marks/node-map/link/pre exclusions	`native-input-strategy`, `model-input-strategy`, direct DOM text sync capability	Recovered with narrower capability boundary	Native text remains capability-based; model-owned repair takes over when DOM/model text diverge.
Safari composition end can arrive after `beforeinput insertFromComposition`; clear composition before commit.	legacy Safari composition comment	`composition-state.ts`	Recovered	`commitInsertFromComposition` keeps the Safari ordering comment and behavior.
Chrome composition `beforeinput` can be wrong; use compositionend fallback.	legacy Chrome composition fallback	`composition-state.ts`	Recovered	`commitChromeCompositionEndFallback` owns the rule and writes inside `editor.update`.
Android beforeinput is not cancelable; Android gets its own manager.	legacy Android beforeinput branch	`android-input-manager`	Recovered, scoped	Android manager owns noncancelable beforeinput, pending selection, and mobile paste-like insertText comments; raw mobile device proof remains a separate claim.
Selectionchange must ignore internal repair/programmatic selection updates.	legacy `state.isUpdatingSelection` guard	`selection-controller`, `editing-epoch-kernel`	Recovered and strengthened	Repair-induced/programmatic selectionchange stays model-owned and closes the destructive epoch.
Chrome can fire selectionchange when inputs/textareas are appended.	legacy INPUT/TEXTAREA filter	`selection-controller` model-selection preference and browser generated proof	Rejected as a literal filter, replaced by ownership gating	The v2 filter is provenance-based rather than tag-based. It refuses repair-induced imports and imports only native-user DOM selection that belongs to the editor. Current generated/internal-control rows cover the real risk.
Firefox nested editable focus must be prevented for keyboard navigation.	legacy Firefox focus branch	`input-router`, `selection-controller`, focus owner assertions	Recovered by owner split	Internal-control rows and focus-owner assertions cover focus leakage without centralizing all focus code in `Editable`.
Paste without formatting and Safari paste may not produce useful beforeinput data.	legacy React `onPaste` fallback comments	`clipboard-input-strategy`, `mutation-controller`	Recovered and fixed	Plain text paste row is green; DataTransfer insert now enters `editor.update`; generated destructive paste gauntlet and release proof are green.
Drag/drop needs global dragend/drop cleanup because Firefox may unmount the dragged element.	legacy global drag lifecycle comment	`input-router`, `clipboard-input-strategy`	Recovered	Global drag listeners remain split from `Editable`; drop command stays model-owned.
Selection export must tolerate bridge failures during commits.	legacy defensive selection sync comments	`selection-controller`, `dom-repair-queue`	Recovered	`syncEditableDOMSelectionToEditor` catches transient DOM bridge failures and leaves browser selection unchanged.

Rejections

Do not restore legacy editor.selection as public truth. V2 uses live reads, transaction target resolution, and selection provenance.
Do not restore Transforms.* as the public editing story. Primitive editor methods inside editor.update are the runtime contract.
Do not restore the monolithic Editable. The timing rules are assigned to explicit owners and enforced by generated browser proof.
Do not trust browser-native structural delete as final truth. Native delete remains model-owned.

Scoped Defers

Raw Android/iOS keyboard, clipboard, and IME proof remains outside this local environment. Playwright mobile viewport and semantic handles do not satisfy raw-device claims.
Full composition stress on real mobile keyboards remains device-lab work.
Browser engine proof is strong for the current richtext destructive/paste generated rows, but full test:integration-local remains the closure sweep, not an inner-loop command.

Required Gates

Current legacy timing recovery is accepted only with these gates:

bash

bun test ./packages/slate-react/test/editing-epoch-kernel-contract.ts ./packages/slate-react/test/editing-kernel-contract.ts ./packages/slate-react/test/selection-controller-contract.ts ./packages/slate-react/test/dom-repair-policy-contract.ts ./packages/slate-react/test/target-runtime-contract.tsx ./packages/slate-react/test/dom-text-sync-contract.ts --bail 1
bun run --cwd packages/slate-browser test:core --bail 1
bun --filter slate-browser test:proof
SLATE_BROWSER_SOAK_BASE_URL=http://localhost:3100 SLATE_BROWSER_SOAK_ITERATIONS=5 bun ./scripts/proof/persistent-browser-soak.mjs
PLAYWRIGHT_BASE_URL=http://localhost:3100 bunx playwright test ./playwright/integration/examples/richtext.test.ts --project=chromium --project=firefox --project=webkit --project=mobile --grep "generated destructive paste|generated mixed editing conformance|persistent native word-delete" --workers=4 --retries=0
bun test:release-proof