Slate v2 Mobile / IME Input Runtime Ralplan

1. Current Verdict

Pass 12 plus the cross-editor test-mining addendum are complete. The Slate Ralplan is done.

The closure pass found no contradiction between the scorecard, issue-sync text, PR reference, and final gates. All scheduled pass rows are complete, score is 0.95, no dimension is below 0.85, and the synced ledgers preserve the core truth: this plan accepts the shared input-runtime owner model, but it does not claim exact Android/iOS/IME fixes without matching proof.

The issue-sync accounting pass completed conservative durable sync without inflating claims. The issue coverage matrix now backfills #5711, #3634, and #4961 as related rows; the fork dossier aligns stale #6022, #5983, #5183, and #5088 statuses down to related; the matrix, fork dossier, live gitcrawl ledger, cluster ledger, and PR reference carry the Mobile/IME macro sync counts. No exact Fixes #... or Improves #... Mobile/IME claim changed. Score is now 0.95; closure and the requested cross-editor mining pass are complete.

The correct direction is not a new public API blitz and not one Android patch per issue. Slate v2 already has the right owner shape in live source: a shared slate-react input runtime, a slate-dom DOM bridge, explicit composition state, native beforeinput / input listeners, Android manager integration, selection import/export, repair, and kernel traces.

The related issue discovery pass confirms the shape: the frozen R7 input runtime surface is 149 rows, all cluster-synced, with 117 high-confidence rows and 32 medium-confidence rows. The live gitcrawl mirror adds 119 current keyword candidates with the discovery filter. That sweep is candidate generation, not claim proof.

The issue-ledger pass now gives the next accounting shape. Among the 123 frozen R7 rows missing from current dossier coverage, 44 need long-form dossier/proof-route rows, 59 stay matrix-only future proof rows, and 20 are matrix-only non-claim rows. Pass 11 backfilled the three existing dossier rows into the matrix and aligned the four matrix/dossier drift rows down to related.

The pass-5 refresh confirms the earlier architecture call. Current .tmp/slate-v2 source still owns Mobile/IME through the shared input runtime, not through one-off per-device patches: root runtime state and owners live in runtime-root-engine.ts:136, :188, :199, :246, :253, :283, :289, and :345; event-family routing lives in runtime-event-engine.ts:162, :194, :215, and :252; composition is a first-class runtime path in runtime-composition-events.ts:36, :71, and :106; Android is routed through runtime-android-engine.ts:8; native beforeinput / input listeners stay in input-router.ts:91; and slate-dom still exposes composing state plus Android DOM offset handling in dom-editor.ts:82, :166, :811, :1207, and :1511.

The pass-5 refresh also confirms the proof boundary. Current browser tests give desktop IME and synthetic mobile confidence, not exact device closure: rendering-strategy-runtime.test.ts:401 covers browser IME commit, :431 covers the generated composition gauntlet, and :477 labels mobile as mobile-synthetic-composition; dom-coverage-boundaries.test.ts:248 explicitly skips mobile for hidden-boundary IME; dom-coverage-boundaries.test.ts:283 covers mobile touch only; generated-editing.test.ts:1027 is synthetic selection-repair IME stress; and dom-coverage.ts:347 is unit-level composition boundary policy. The package.json scripts expose check, check:full, test:mobile-device-proof, and test:mobile-device-proof:raw, so exact Android/iOS claims have a route but still need matching artifacts.

The pass-4 boundary still stands: the 44 long-form rows are proof-route backlog, not claim backlog. The 20 matrix-only non-claim rows stay out of PR auto-close language. The former four drift rows now default to related until exact proof says otherwise. The existing v2-input-runtime Ralplan is done as an architecture bucket, and this macro plan has enough routing to drive follow-up ledger work. It still cannot claim "all Mobile/IME issues". Desktop IME, synthetic mobile, and Android helper tests are useful. They do not close Samsung Keyboard, Android Firefox predictive typing, Hangul placeholder, Chinese backspace, Android voice input, iOS Chinese input, or Windows Vietnamese rows without matching proof.

Pass 6 adds the hard pressure test: the current runtime shape is still good, but execution cannot start as a vague "make Mobile/IME better" effort. Performance claims must name cohorts and repeated-unit budgets; native-behavior claims must say which mode preserves browser find, selection, copy/paste, IME, mobile touch, undo/history, and collaboration; migration claims must stay at the substrate level; and tests must be vertical issue/proof rows, not a bulk dump of all 44 long-form rows. The current source is strong enough to keep the direction: root/runtime refs and memoized owners in runtime-root-engine.ts:136, :188, :214, :246, :266, and :345; root native listeners in input-router.ts:91; source-scoped listeners in public-state.ts:2175 and :2483; dirty runtime IDs and commit metadata in public-state.ts:367; and render-budget stress rows in generated-editing.test.ts:221, :947, and :1020. The missing part is release-grade budgets and exact device artifacts, not a new public API.

Pass 7 turns the likely maintainer pushback into plan edits. The strongest objections do not invalidate the direction; they tighten the stop rules. Raw device proof is still mandatory for exact mobile closure, synthetic rows must be labelled as synthetic in proof and PR text, performance claims need cohort plus repeated-unit evidence, and issue-sync work must happen in proof-heavy batches instead of trying to write all 44 long-form rows at once. The maintainer pass also accepts that the runtime already has Android-specific ownership, so the answer is to keep that ownership centralized in runtime-android-engine.ts:8 and runtime-root-engine.ts:246, not to scatter device policy through public APIs.

Pass 8 turns the high-risk review into release stop rules. The highest-risk failure is not "the plan is too cautious"; it is shipping an input change that duplicates text, drops composition text, moves the caret, or over-claims a real keyboard from synthetic proof. The plan keeps the runtime owner graph, but every execution slice now needs an explicit failure-mode table: data loss, selection corruption, public API lock-in, false mobile claim, React hot-path regression, collab/history nondeterminism, and native behavior regression. Any one of those failing in proof downgrades the claim or splits the slice before ledger sync.

Pass 9 locks the ecosystem story. Extension authors get typed state / tx namespaces, runtime registration, commit listeners, and narrow React hooks; they do not get raw public command slots or a Mobile/IME product namespace. Plate gets the primitive input runtime, composing/focus/selection hooks, schema/extension backbone, and proof gates it can wrap into richer UX. slate-yjs gets deterministic operation replay, commit metadata, and typed remote metadata; it does not get browser event timing encoded into the collaboration API. PR text must say this plainly, or maintainers will read the plan as a broad adapter promise.

Pass 10 folds the accepted pass-7, pass-8, and pass-9 revisions into one execution contract. The plan now has a single conservative rule set for batching, synthetic labels, raw-device proof, performance budgets, adapter ownership, high-risk failure modes, and ecosystem language. No contradiction was found: the shared input runtime remains the right owner, the public API stays primitive, and issue sync must preserve conservative Related / Needs repro language until matching proof exists.

Current owner: slate-ralplan, complete.

Next owner after user review: ralph execution planning.

2. Intent / Boundary Record

Intent:

• run a Slate Ralplan over all Mobile/IME issues, not just one local bug;
• turn the top-ranked issue macro theme into an execution-grade plan with exact claim boundaries;
• keep the current input-runtime architecture if the live source already earned it, and focus the plan on the missing proof/accounting.

Desired outcome:

• every Mobile/IME issue is classified as Fixes, Improves, Related, Not claimed, Needs repro, Stale, or Invalid;
• every claimed fix has a matching proof route;
• no Android/iOS/IME issue is auto-closed from desktop-only or semantic-mobile proof;
• raw Slate exposes runtime primitives, not product policy;
• the 44 long-form rows are prepared for proof and dossier work without implying they are fixed or improved;
• the 20 matrix-only non-claim rows remain visible as pressure, not execution work.

In scope:

• slate-react-v2 event/runtime ownership for beforeinput, input, keydown, composition, focus, selection import, Android integration, and repair;
• slate-dom-v2 DOM point/range translation, composing state, Android zero-width handling, and browser-facing input bridges;
• live gitcrawl Mobile/IME rows and frozen ledger Mobile/IME rows;
• test-candidate maps for Android, iOS, IME, composition, and native input;
• proof tiering: unit, browser, generated scenario, synthetic mobile, Appium/raw-device, and explicit non-claim rows;
• routing and stop rules for existing matrix/dossier drift.

Non-goals:

• no Slate v2 implementation patch in this Ralplan pass;
• no dossier or PR-reference file sync in this pass;
• no product autocomplete, slash-command, mention, toolbar, spellcheck, or text-limit policy in raw Slate;
• no React Native support promise;
• no exact mobile claim from Playwright mobile viewport alone;
• no docs/support issue closure unless the row is actually docs/support;
• no upward claim promotion for #6022, #5983, #5183, or #5088 from ledger drift alone.

Decision boundaries:

• this plan may decide issue classification, proof routes, package ownership, and pass order without another user question;
• this plan may not mark done until the scheduled passes are complete and the score reaches the Slate Ralplan threshold;
• device-specific claims require device-specific proof or stay non-claims;
• drift rows default downward to the more conservative claim unless exact proof promotes them;
• long-form dossier rows must state current proof level before any PR line is copied from them;
• unresolved browser/tooling proof stays pending, not blocked, while local issue discovery and plan hardening remain runnable;
• if a row asks for product behavior, raw Slate only owns the primitive that makes the behavior possible without DOM/model drift.

Unresolved user-decision points:

• none. The next work is evidence and classification, not a product choice.

Pass-4 pressure test:

No user question asked: the repo evidence gives enough authority to keep moving.

3. Decision Brief

Principles:

• Browser input is a runtime protocol.
• Composition is first-class state.
• Native-owned input and model-owned commands must not fight over selection.
• Mobile claims require matching mobile evidence.
• Raw Slate stays unopinionated; product behavior belongs above it.

Top drivers:

• Mobile/IME is the highest-priority macro theme in the frozen corpus: 129 issues, priority 21.37.
• The theme covers Android, iOS, Firefox Android, Windows IMEs, Hangul, Chinese, Vietnamese, predictive typing, placeholders, empty-state input, and suggestion replacement.
• Package ownership points to slate-react-v2 plus slate-dom-v2, with 124 of 129 rows in runtime ownership.
• Pass 3 routed 149 R7 input-runtime rows: 44 long-form proof rows, 59 future proof rows, 20 non-claim rows, 3 matrix backfills, and 4 drift rows.
• Live source already has input-runtime structure, so a second architecture would be waste.
• Existing proof is promising but not enough for per-device issue closure.

Viable options:

Chosen shape:

native browser event
  -> input controller intent
  -> runtime ownership decision
  -> selection import/export policy
  -> composition / Android / model / app handling
  -> repair and kernel trace
  -> proof artifact
  -> issue claim or non-claim

Consequences:

• Exact Android/iOS closures stay expensive.
• Some rows will stay Related or Needs repro even if the architecture is correct.
• Pass 6 tested the refreshed source/proof map against performance, DX, migration, regression, and simplicity pressure.
• Existing v2-input-runtime decisions should be inherited unless the Mobile/IME-specific issue pass invalidates them.
• Matrix/dossier drift is not a reason to promote claims; it is a reason to sync ledgers after proof ownership is clear.

Follow-ups:

• run closure score and final gates;
• append fork issue dossier sections only for rows that need long-form fork accounting;
• update live issue ledgers only when future classification changes are grounded;
• keep exact fixes separate from architecture coverage.

Decision locks from pass 4:

4. Confidence Scorecard

Total score: 0.95.

Why not higher:

• The closure pass still has to re-read final gates and decide done versus another concrete revision.
• 44 long-form dossier/proof rows are routed, but individual execution-batch sections still need exact repro/proof before any claim promotion.
• Ecosystem synthesis is refreshed for architecture direction, maintainer objections, high-risk failure modes, adoption language, and issue accounting, not for final implementation closure.
• Exact mobile/IME proof remains mixed and in places tooling-bound.

5. Source-Backed Architecture North Star

The north star is already visible in live source.

• Root runtime owns isComposing, input controller state, Android manager ref, selection reconciler, repair runtime, kernel trace runtime, event runtime, and event bindings in .tmp/slate-v2/packages/slate-react/src/editable/runtime-root-engine.ts:136, :188, :199, :246, :253, :283, :289, and :345.
• Event runtime fans out to beforeinput, input, clipboard, drag, composition, focus/mouse, and keyboard handlers through one runtime object in .tmp/slate-v2/packages/slate-react/src/editable/runtime-event-engine.ts:162 and :194.
• Input classification handles composition, paste/drop, delete, insert, format, and internal controls in .tmp/slate-v2/packages/slate-react/src/editable/input-controller.ts:121, :170, and :246.
• Composition events run through prepareEditableCompositionKernel, trace ownership, and apply start/update/end in .tmp/slate-v2/packages/slate-react/src/editable/runtime-composition-events.ts:36, :71, and :106.
• Android runtime is a first-class adapter to useAndroidInputManager in .tmp/slate-v2/packages/slate-react/src/editable/runtime-android-engine.ts:8.
• Native DOM beforeinput and input are attached outside React's polyfill in .tmp/slate-v2/packages/slate-react/src/editable/input-router.ts:91.
• slate-dom exposes composing state, Android flush hooks, and Android zero-width handling in .tmp/slate-v2/packages/slate-dom/src/plugin/dom-editor.ts:82, :166, :811, :1207, and :1511.

Decision: keep this owner model. The Ralplan work is to make it issue-grade and proof-grade, not to replace it.

6. Ecosystem Strategy Synthesis

Research/source refresh status: complete for pass 5. No docs/research write was required because the compiled layer already covers the current architecture question and the remaining gap is issue/device proof, not missing ecosystem research.

Pass-5 evidence ledger:

User-requested cross-editor IME/mobile mining addendum:

Pass-9 ecosystem maintainer ledger:

Pass-9 decision:

• keep the current ecosystem story because the live extension and commit substrate already gives extension, Plate, and collaboration owners something real to build on;
• revise PR and adoption language so it says "substrate and proof gates", not "Plate/yjs adapters are done";
• keep Mobile/IME product behavior out of raw Slate;
• route any missing primitive found during implementation to the revision pass before it becomes public API.

7. Public API Target

Keep:

• editor.dom.isComposing(): boolean;
• useEditorComposing(): boolean;
• onDOMBeforeInput?: (event: InputEvent) => boolean | void;
• narrow input rules / app hooks where already source-backed.

Do not add yet:

• a new Mobile/IME public API namespace;
• product autocomplete or text suggestion APIs;
• mobile-device feature flags as public Slate policy;
• issue-number-specific APIs.

Decision: the public API is not the current blocker. The issue pass may propose new narrow extension points only if a concrete row proves existing hooks cannot express the behavior without DOM/model drift.

8. Internal Runtime Target

Target:

• input controller classifies native intent;
• event runtime centralizes handler families;
• composition engine owns start/update/end state and trace rows;
• Android engine owns Android-specific pending diffs and selection timing;
• selection runtime imports browser selection only under explicit ownership;
• repair runtime patches DOM/model drift after the kernel decision;
• trace runtime records enough detail to prove issue claims.

Already done in live source:

• the owner graph exists across runtime-root-engine.ts, runtime-event-engine.ts, input-controller.ts, runtime-composition-events.ts, runtime-android-engine.ts, runtime-input-events.ts, and input-router.ts.

Gap:

• issue-specific proof rows must show the graph works for each Mobile/IME class.

9. Hook / Component / Render DX Target

Target:

• editor hooks expose composition/focus/selection state narrowly;
• renderers do not need to manually patch Mobile/IME DOM behavior;
• app renderers can remain ordinary Slate renderers;
• product controls inside the editor are classified as internal controls and do not hijack composition/input.

Evidence:

• internal control classification lives in .tmp/slate-v2/packages/slate-react/src/editable/input-controller.ts:89;
• useEditorComposing is public docs surface in .tmp/slate-v2/docs/libraries/slate-react/hooks.md:7.

Gap:

• inline void and interactive internal-control Mobile/IME issues need exact issue rows before any renderer contract change is proposed.

10. Plate Migration-Backbone Target

Plate should inherit:

• a stable input-runtime substrate;
• composition/focus/selection state hooks;
• input rule hooks that do not desync DOM and model;
• browser proof gates it can reuse without owning raw Slate internals.

Plate should not get:

• raw Slate product policy hardcoded for mentions, autocomplete, slash commands, or toolbar UX;
• current-version adapter promises from this Ralplan.

Pass-9 status: backbone and adoption wording reviewed; adapters still out of scope.

Evidence:

• .tmp/slate-v2/packages/slate/test/migration-backbone-contract.ts:33 proves extension state / tx namespaces and schema specs without adapter-shaped api, tf, plate, yjs, or feature fields on the editor surface.
• .tmp/slate-v2/packages/slate/src/core/editor-runtime.ts:97 exposes snapshot, runtime-id, selection, subscribe, read, and update runtime groups behind one internal runtime.
• .tmp/slate-v2/packages/slate/test/commit-metadata-contract.ts:18 and :109 prove update tags, selection before/after, dirty runtime IDs, and typed metadata.

Decision: Plate migration pressure reinforces the current primitive substrate. Do not add Plate-shaped editor APIs to raw Slate from this Mobile/IME plan. PR/adoption language should say Plate can wrap the substrate, not that this plan ships current-version Plate integration.

11. slate-yjs Migration-Backbone Target

Yjs/collab cares about deterministic local commits, selection metadata, and composition lifecycle boundaries.

Target:

• composition start/update/end and DOM-selection policy should produce deterministic local commits or explicit browser-owned non-commit spans;
• remote apply should not infer browser event timing;
• issue fixes should not introduce hidden non-deterministic DOM-derived writes.

Status: ecosystem wording reviewed; current-version yjs adapter work remains out of scope.

Evidence:

• .tmp/slate-v2/packages/slate/test/migration-backbone-contract.ts:171 replays deterministic operations with commit tags and local-only runtime targets.
• .tmp/slate-v2/packages/slate/test/collab-history-runtime-contract.ts:30 publishes one commit truth for collab subscribers, extension listeners, and history.
• .tmp/slate-v2/packages/slate/test/collab-history-runtime-contract.ts:155 uses typed remote collaboration metadata to skip local undo history.

Decision: yjs/collab needs deterministic operations, commit metadata, and local-only target semantics. It does not need Mobile/IME product policy or browser-event timing encoded into raw collaboration APIs.

12. Full Issue-Ledger Accounting

Current corpus facts:

• Frozen issue corpus: 682 rows in docs/slate-issues/open-issues-ledger.md:18.
• Live gitcrawl overlay: 630 open issues in docs/slate-issues/gitcrawl-live-open-ledger.md:12.
• Mobile/IME macro theme: 129 issues, cluster-synced, no per-device closure without device proof in docs/slate-issues/open-issues-ledger.md:84.
• Mobile/IME is the top-ranked macro theme with priority 21.37 in docs/slate-issues/issue-clusters.md:46.
• Requirements R7 assigns input/composition/IME semantics to slate-react-v2 plus slate-dom-v2 in docs/slate-issues/requirements-from-issues.md:247.
• Package matrix assigns Mobile/IME to slate-react-v2 + slate-dom-v2, with 124 runtime rows in docs/slate-issues/package-impact-matrix.md:75.

ClawSweeper related-issue pass:

• status: complete for discovery and pass-11 durable macro sync;
• trigger: this plan touches public runtime behavior, browser behavior, issue claims, and PR narrative;
• freshness: gitcrawl doctor --json reports 659 open Slate threads, 617 clusters, last_sync_at: 2026-05-04T14:58:11.123944Z, api_supported: false, and no GitHub token;
• frozen exact R7 input-runtime set: 149 rows, all cluster-synced;
• R7 confidence split: 117 high, 32 medium;
• R7 readiness split: 80 ready-with-minor-setup, 45 ready-now, 24 not-a-test-candidate;
• live gitcrawl candidate sweep: 119 current keyword rows with the pass-2 filter; treat this as candidate generation because keyword matching includes adjacent selection, placeholder, spellcheck, and false-positive rows;
• issue coverage matrix: 28 current input-runtime/input-related rows found;
• fork issue dossier: 26 existing v2-input-runtime sections found;
• claim result: no exact Fixes or Improves additions;
• discovery result: existing related rows stay related/not-claimed; remaining R7 rows are related-discovered and wait for the issue-ledger pass.

Dossier output:

• existing dossier sections found for #6022, #5983, #4400, #5883, #6051, #3777, #5095, #5096, #5603, #5669, #4994, #5026, #4001, #5989, #5984, #5931, #5830, #5643, #5130, #5050, #5014, #4348, #4223, #4067, #3568, and #3470;
• pass 11 appended a Mobile/IME macro sync section and aligned the four stale drift statuses down to related;
• pass 3 decides that 44 of the 123 R7 rows missing from current dossier coverage need long-form dossier/proof-route sections, while 79 stay matrix-only for now.

Issue-ledger pass:

• status: complete for routing and pass-11 durable file sync;
• frozen R7 rows routed: 149;
• existing matrix and dossier coverage: 23 R7 rows;
• existing dossier rows needing matrix backfill: 3;
• new long-form dossier/proof-route rows needed: 44;
• matrix-only future-proof rows: 59;
• matrix-only non-claim rows: 20;
• claim result: no exact Fixes or Improves additions from this pass.

Pass-3 routing matrix:

Claim-level drift found:

Pass-11 drift result: all four dossier statuses now align down to related. No row was promoted.

Research/source refresh effect on issue-ledger accounting:

• The 44 long-form rows remain long-form proof-route rows. Pass 5 proves routing and owner fit; it does not prove issue closure.
• The strongest first batch should be rows with both clear dossier text and public test rows: Android empty-state / placeholder composition (#5989, #5983, #5891, #5883), Android Chinese delete / duplicate insertion (#5984, #5083, #5078, #5034, #5023), Android autocorrect / suggestion replacement (#4531, #5653), Android heading Enter / block type rows (#4521, #5175), inline-boundary IME rows (#4693, #4136), and empty-editor IME rows (#4030, #3943, #3882).
• The former four drift rows stay conservative at related until exact proof decides whether to promote.
• Live source did not produce a new public API requirement in pass 5.

Fixed issues:

• none from this Ralplan pass.

Materially improved issues:

• none from this Ralplan pass.

Related but not fixed initial seed:

Primary R7 families discovered in pass 2:

Two-row R7 families:

• Android readOnly: #5034, #4994;
• suppressed input: #5152, #5010;
• focus initialization: #5213, #4696;
• focus events: #5004, #4560;
• input interception: #5050, #4795;
• mobile autocorrect suggestions: #4531, #4354;
• mobile backspace: #5984, #4959;
• placeholder composition: #5989, #4136;
• iOS placeholder selection: #5481, #4640;
• placeholder timing: #5420, #4673;
• plugin/render composition: #3222, #3177;
• decoration/selection stability: #3354, #3162;
• render during composition: #5433, #5398;
• marks/decorations/inlines: #2564, #2465;
• SSR/hydration: #5430, #5380.

One-row R7 families:

• 57 one-row families remain. Do not explode them into claims in discovery pass. Pass 3 routed them as matrix rows, dossier rows, or explicit non-claims.

Current live high-signal rows not fully claimable yet:

Live gitcrawl ledger sync:

• updated in pass 11 with an append-only V2 sync note. The live table remains the raw gitcrawl mirror; the planning overlay records 149 routed rows, 44 long-form proof-route rows, 59 matrix-only future-proof rows, 20 matrix-only non-claim rows, 3 matrix backfills, and 4 drift rows aligned down to related.

Fork issue dossier sync:

• updated in pass 11. A Mobile/IME macro sync section records the reviewed surface, and #6022, #5983, #5183, and #5088 now align with the conservative related matrix claim.

Issue coverage matrix sync:

• updated in pass 11. #5711, #3634, and #4961 were added as related matrix rows; a Mobile/IME macro planning sync table records the `23 + 3 + 44
  • 59 + 20` route accounting without adding closure language.

PR description:

• updated in pass 11. The summary now reports 154 related issue matrix rows, the Mobile/IME macro counts, and 0 new exact fixed/improved Mobile/IME claims from this plan.

13. Legacy Regression Proof Matrix

14. Browser Stress / Parity Strategy

Proof tiers:

1. Unit/runtime contracts for input classification, composition state, Android manager decisions, DOM bridge offsets, and repair policy.
2. Generated browser scenario rows for native beforeinput, input, composition, selection import/export, and kernel trace legality.
3. Desktop native IME rows for Chromium, Firefox, and WebKit where supported.
4. Synthetic mobile rows for architecture confidence only.
5. Appium/raw-device rows for Android/iOS claims.
6. Manual or external device-lab rows for Samsung Keyboard, Android Firefox predictive typing, Gboard/autocorrect/voice, iOS Safari Chinese, and Windows IME rows when automation cannot drive the real input method.

No exact issue closure may skip the tier matching the reported environment.

Pass-5 proof inventory conclusion:

• Desktop Chromium IME proof exists for the rendering-strategy runtime.
• Synthetic composition proof exists for non-Chromium and mobile projects.
• Mobile touch proof exists around hidden DOM coverage boundaries.
• Unit proof exists for composition-boundary materialization policy.
• Raw Android/iOS proof scripts exist, but no fresh artifact from this Ralplan pass can promote a device-specific claim.

Cross-editor tests to steal before execution:

• Port Lexical's CDP staged IME pattern from ../lexical/packages/lexical-playground/__tests__/e2e/Composition.spec.mjs into Slate browser scenarios where the current editor.ime.compose helper is too coarse.
• Add translated cases for composition across inline voids/mentions, formatting boundaries, emojis/decorations, hidden DOM coverage boundaries, cancellation, delete-after-composition, and Korean replacement across multiple formatted text nodes.
• Add Lexical-style history rows: composed text forms the right undo group, canceled composition is not pushed to history, and undo restores selection.
• Keep Safari post-composition delete-selection rows from Lexical regression 8153-safari-ime-delete-selection.spec.mjs as WebKit regression gates; both translated richtext rows landed in slice 24.
• Add ProseMirror-style DOM-change rows: composed DOM text is read without repainting the active text node, markup-bound text mutations are preserved, ambiguous replacement is resolved, and composition changes carry explicit commit metadata.
• Add Android/iOS raw-device or manual rows for the cases local desktop tests cannot prove: Android virtual-keyboard backspace/autocorrect, iOS word prediction, iOS Korean backspace, Samsung Keyboard, Android Firefox predictive typing, voice input, and iOS Safari Chinese.

Pass-6 performance and native-behavior contract:

Pass-10 execution contract for issue sync:

15. Applicable Implementation-Skill Review Matrix

16. High-Risk Deliberate-Mode Pre-Mortem

Status: complete for pass 8.

Trigger: yes. This plan touches selection, focus, IME, DOM repair, browser runtime behavior, React runtime strategy, migration substrate, collaboration metadata, public proof claims, issue ledgers, and release gates.

Current high-risk source owners refreshed in pass 8:

• root runtime state, input controller, Android manager, selection import, repair, kernel trace, and event runtime: .tmp/slate-v2/packages/slate-react/src/editable/runtime-root-engine.ts:136, :188, :199, :214, :246, :253, :283, :289, and :345;
• input handling and DOM input repair: .tmp/slate-v2/packages/slate-react/src/editable/runtime-input-events.ts:52, :79, :89, :106, and :135;
• composition start/update/end ownership: .tmp/slate-v2/packages/slate-react/src/editable/runtime-composition-events.ts:36, :71, and :106;
• internal-control, keyboard, beforeinput, composition, and composing-state classification: .tmp/slate-v2/packages/slate-react/src/editable/input-controller.ts:75, :121, :170, :230, and :246;
• DOM bridge composing state, Android pending diff hooks, Android zero-width handling, point/range translation, and public DOM capability: .tmp/slate-v2/packages/slate-dom/src/plugin/dom-editor.ts:82, :161, :811, :1207, :1227, :1440, and :1511;
• current proof scripts and proof split: .tmp/slate-v2/package.json:58, :59, :60, and :63;
• browser proof labels: .tmp/slate-v2/playwright/integration/examples/rendering-strategy-runtime.test.ts:401, :431, and :477;
• hidden-boundary split between desktop IME and mobile touch: .tmp/slate-v2/playwright/integration/examples/dom-coverage-boundaries.test.ts:248 and :283;
• commit/collab substrate: .tmp/slate-v2/packages/slate/test/commit-metadata-contract.ts:18, :84, :109, and :144; .tmp/slate-v2/packages/slate/test/migration-backbone-contract.ts:33 and :171; .tmp/slate-v2/packages/slate/test/collab-history-runtime-contract.ts:30, :114, and :155.

Blast radius:

Pre-mortem:

Expanded proof plan:

Rollback / hard-cut / remediation answer:

• If a proof row fails with data loss, duplicate input, or caret corruption, stop the implementation slice and keep the issue Related / Needs repro; do not sync optimistic claim text.
• If the failure requires a broad public API, route it back through the revision pass before accepting it; do not ship a Mobile/IME namespace by drift.
• If synthetic proof passes but raw-device proof is missing, downgrade the claim to architecture confidence and keep exact device rows non-claims.
• If a performance row fails, split the fix into owner-local runtime work before any ledger promotion.
• If collab/history determinism fails, block adapter-facing migration language and keep the plan at substrate repair.

High-risk verdict:

• keep the shared input-runtime owner graph;
• revise execution acceptance criteria to require a failure-mode table per implementation slice;
• split work by proof-heavy issue families, not by broad Mobile/IME category;
• drop any exact issue claim whose proof tier does not match the report;
• next owner is closure score and final gates, using the synced pass-11 claim text instead of reopening broad review.

17. Hard Cuts And Rejected Alternatives

Hard cuts:

• no per-device closure without per-device proof;
• no legacy monolithic Editable restoration;
• no product input policy in raw Slate;
• no new public API until the issue pass proves existing hooks cannot carry the behavior;
• no "mobile" claim from viewport-only Playwright rows.
• no public command slots from this Mobile/IME plan;
• no current-version Plate or slate-yjs adapter promise from raw Slate;
• no issue-sync promotion before proof tier and claim text agree.

Rejected:

• broad browser special-case tables without kernel traces;
• core-only fixes for browser event timing;
• app-level workarounds as Slate v2 architecture;
• issue-number decoration in PR text without claim classification.

18. Slate Maintainer Objection Ledger

Pass-7 revisions accepted:

• add a batching stop rule: proof-heavy rows first, not all 44 long-form rows before code;
• add a synthetic label stop rule: viewport/mobile-synthetic rows never become exact mobile claim text;
• add a performance claim stop rule: large/stress claims require cohort, repeated-unit, DOM/component/listener/memory, and native-behavior evidence;
• add an adapter ownership stop rule: Plate/yjs adapter proof is later work; this plan owns raw Slate substrate only;
• add a review-loop stop rule: finish the scheduled gates, then hand execution to ralph unless a concrete contradiction appears.

Status: complete for pass 7. No objection invalidated the shared input-runtime owner model, but several objections tightened claim labels, batching, performance proof, and adapter ownership.

19. Pass Schedule And Pass-State Ledger

20. Plan Deltas From Review

Added in pass 1:

• created Mobile/IME macro Ralplan instead of reusing only the existing v2-input-runtime bucket plan;
• explicitly inherited the live input-runtime owner graph;
• separated architecture coverage from exact issue closure;
• set initial score to 0.77;
• set ClawSweeper as next owner.

Dropped in pass 1:

• no implementation changes;
• no fixed issue claims;
• no PR-reference edits;
• no live-ledger status changes.

No-change defense:

• current public composition API stays unchanged because no pass-1 through pass-6 evidence proves it is too weak.

Added in pass 2:

• completed ClawSweeper-style related discovery for the Mobile/IME/input runtime surface;
• recorded gitcrawl doctor freshness and the local live keyword candidate count;
• extracted the frozen R7 input-runtime set: 149 rows, all cluster-synced;
• mapped primary, two-row, and one-row R7 families;
• found existing ledger coverage: 28 input-runtime matrix rows and 26 dossier sections;
• raised score to 0.80.

Dropped in pass 2:

• no implementation changes;
• no fixed issue claims;
• no live-ledger status changes;
• no new dossier sections until the issue-ledger pass decides which rows need them.

Added in pass 3:

• routed all 149 frozen R7 input-runtime rows;
• decided 44 missing-dossier rows need long-form dossier/proof-route sections;
• decided 59 rows stay matrix-only future proof rows;
• decided 20 rows stay matrix-only non-claim rows;
• found 3 existing dossier rows that need matrix backfill;
• found 4 matrix/dossier claim-level drift rows;
• raised score to 0.84.

Dropped in pass 3:

• no implementation changes;
• no new exact Fixes or Improves claims;
• no immediate file sync for matrix/dossier/PR reference because this pass only owns routing, not the later issue-sync accounting pass.

Added in pass 4:

• hardened the intent/boundary record against the pass-3 routing;
• made the 44 long-form rows proof-route backlog, not claim backlog;
• made the 20 matrix-only non-claim rows explicit non-execution pressure;
• added stop rules for drift rows, synthetic mobile proof, device proof, and product-policy leakage;
• added decision locks for public API, dossier batching, drift handling, device proof, and implementation timing;
• raised score to 0.87.

Dropped in pass 4:

• no implementation changes;
• no dossier, matrix, live-ledger, or PR-reference sync;
• no user question because the repo evidence is enough to continue.

Added in pass 5:

• re-read current Slate v2 source owners for root runtime, event runtime, input controller, composition runtime, Android runtime, native input listeners, and slate-dom composing/Android offset handling;
• re-read current browser/unit/stress proof rows and confirmed the mobile proof split between synthetic composition, touch usability, and raw-device gates;
• rechecked compiled ProseMirror, Lexical, Tiptap, EditContext/IME, and editing recovery research and kept the current runtime owner model;
• tied representative issue dossiers and public test rows back to the 44 long-form proof backlog;
• raised score to 0.89.

Dropped in pass 5:

• no implementation changes;
• no research-layer write, because the compiled research already answered the architecture question for this pass;
• no dossier, matrix, live-ledger, or PR-reference sync;
• no fixed/improved claim promotion from source evidence alone.

Added in pass 6:

• applied Vercel React, performance, performance-oracle, tdd, regression-lock, and simplicity lenses to the refreshed owner/proof map;
• added cohort, repeated-unit, React runtime, render-budget, native-behavior, degradation, and migration contracts;
• confirmed Plate/yjs migration pressure is substrate-level through state / tx namespaces, deterministic operation replay, commit metadata, and local-only runtime targets;
• locked the test strategy to vertical issue/proof rows rather than bulk-writing all 44 long-form rows;
• raised score to 0.91.

Dropped in pass 6:

• no implementation changes;
• no new Mobile/IME public namespace;
• no virtualization or shell mode as default proof for Mobile/IME correctness;
• no current-version Plate/yjs adapter promise from raw Slate;
• no fixed/improved claim promotion from performance or migration evidence.

Added in pass 7:

• expanded the maintainer objection ledger from 3 seed rows to 13 reviewed decisions;
• kept the shared input-runtime owner model after fair objections;
• accepted revisions for proof-heavy batching, synthetic mobile labels, performance evidence, adapter ownership, issue-sync ordering, and scheduled review-loop stop;
• raised score to 0.92.

Dropped in pass 7:

• no exact Mobile/IME issue promotion from objection review;
• no new public Mobile/IME namespace;
• no upstream issue-comment plan from the fork dossier;
• no extra broad review pass beyond the scheduled gates.

Added in pass 8:

• refreshed current high-risk owners in .tmp/slate-v2 source and proof files;
• expanded the high-risk section into trigger, blast radius, seven-scenario pre-mortem, proof-tier table, rollback/remediation answer, and keep/revise verdict;
• added execution acceptance rules for data loss, selection corruption, public API lock-in, false mobile claims, React hot-path regression, collab/history nondeterminism, and native behavior regression;
• strengthened regression-proof and research-evidence score dimensions while keeping total score at 0.92.

Dropped in pass 8:

• no exact Mobile/IME issue promotion from high-risk review;
• no implementation changes;
• no device/browser claim from synthetic proof;
• no adapter-shaped migration promise from raw Slate.

Added in pass 9:

• added ecosystem maintainer rows for raw Slate extension authors, Plate, slate-yjs/collab, React shell authors, release/PR reviewers, and product plugin authors;
• locked adoption language to substrate and proof gates rather than current Plate/yjs adapter completion;
• confirmed raw Slate offers typed state / tx groups, runtime registration, commit listeners, commit metadata, selector hooks, and composition primitives without public command slots or Mobile/IME product policy;
• raised score to 0.93.

Dropped in pass 9:

• no new public command slots;
• no raw Slate Mobile/IME product namespace;
• no current-version Plate or slate-yjs adapter promise;
• no exact issue claim promotion from ecosystem wording.

Added in pass 10:

• consolidated pass-7, pass-8, and pass-9 accepted revisions into one execution contract for issue-sync text;
• updated decision locks, hard cuts, scorecard, phase order, and open questions so the next pass can write durable claim rows without reopening review;
• raised score to 0.94.

Dropped in pass 10:

• no new issue discovery;
• no implementation changes;
• no ledger/dossier/PR-reference sync yet;
• no exact Fixes or Improves claim promotion from revision cleanup.

Added in pass 11:

• updated the issue coverage matrix with related backfill rows for #5711, #3634, and #4961;
• added a Mobile/IME macro planning sync table to the issue coverage matrix;
• appended a Mobile/IME macro sync section to the fork issue dossier;
• aligned stale dossier statuses for #6022, #5983, #5183, and #5088 down to related;
• added V2 sync notes to the live gitcrawl ledger and gitcrawl cluster file;
• updated the PR reference with matrix row count, Mobile/IME macro counts, and 0 new exact fixed/improved Mobile/IME claims;
• raised score to 0.95.

Dropped in pass 11:

• no implementation changes;
• no new exact Fixes #... or Improves #... claims;
• no all-at-once expansion of 44 proof-backlog rows into fake closure text;
• no raw-device or browser claim without matching proof.

Added in pass 12:

• re-read the completion file, continuation prompt, final gates, scorecard, pass-state ledger, PR reference, and synced issue artifacts;
• found no contradiction between the conservative issue-sync text and final gates;
• updated stale plan wording from the previous next owner to the final review / later ralph owner;
• marked the closure pass complete.

Dropped in pass 12:

• no implementation changes;
• no new review pass;
• no Mobile/IME issue claim promotion from closure.

Added in pass 13:

• read local ../lexical IME/mobile composition E2E, history, Safari regression, Android/iOS timing, autocorrect, and Korean backspace evidence;
• read local ../prosemirror transaction metadata, DOM-change, composition viewdesc, and DOM-change webtest evidence;
• compared that evidence to current .tmp/slate-v2 composition proof and added a stolen-test execution backlog for missing rows.

Dropped in pass 13:

• no implementation changes;
• no new exact claim promotion;
• no claim that ProseMirror/Lexical architecture should be copied wholesale.

21. Open Questions And What Would Change The Decision

Open questions:

• Which of the 44 long-form rows can be closed with existing browser/unit proof after exact repro mapping, and which need new raw-device proof?
• Which of the 4 drift rows should be promoted in a later execution batch if exact proof appears?
• Which issue classes need raw Android/iOS proof versus generated browser proof?
• Does any issue require a new narrow public API, or can runtime/proof hardening handle it internally?
• Which proof-heavy batch should ralph execute first after user acceptance?

What would change the decision:

• a future execution batch finds a concrete contradiction between routed issue rows, proof tiers, conservative claim text, and runtime behavior;
• live source shows a current owner different from the runtime graph captured in pass 1 and refreshed in pass 5, pass 6, and pass 8;
• raw-device proof exposes a runtime architecture fault that cannot be handled by the current input controller / Android manager / DOM bridge split.

22. Implementation Phases With Owners

No implementation starts from this pass.

Draft phase order after plan closure:

1. User review of this Ralplan.
2. ralph execution planning after acceptance.
3. Proof inventory, missing red/proof rows, runtime fixes only where proof fails, then ledger/PR sync.

Owners:

• plan owner: slate-ralplan;
• issue discovery owner: clawsweeper for pass 2, complete;
• execution owner after user approval: later ralph continuation;
• implementation owner after plan acceptance: slate-react-v2 and slate-dom-v2 packages in .tmp/slate-v2.

23. Fast Driver Gates

Use focused gates after implementation slices, not during this planning pass:

cd /Users/zbeyens/git/slate-v2
bun test ./packages/slate-dom/test/dom-coverage.ts
bun test:stress
bun test:integration-local --project=chromium --grep "composition|IME|input"
bun test:mobile-device-proof

Closure gates may require:

cd /Users/zbeyens/git/slate-v2
bun check
bun check:full
bun test:mobile-device-proof:raw

Use bun test:mobile-device-proof:raw only on a machine/device lane that can provide real Android/iOS artifacts.

24. Final User-Review Handoff

Accepted runtime owner model:

• Keep Slate v2's shared slate-react input runtime plus slate-dom bridge as the owner for Mobile/IME/input behavior.
• Keep Android-specific ownership centralized in runtime-android-engine.ts through the root runtime.
• Keep composition state, native listeners, selection import/export, repair, kernel traces, commit metadata, and selector hooks as primitives.

Public API:

• Kept: existing primitive runtime hooks and typed state / tx extension groups.
• Cut/rejected: broad public command slots, a raw Slate Mobile/IME product namespace, current-version Plate/yjs adapter promises, and product input policy in raw Slate.
• Added by this plan: no implementation API; this is planning/review only.

Issue accounting:

• Fixed Mobile/IME claims from this macro plan: 0.
• Improved Mobile/IME claims from this macro plan: 0.
• Frozen R7 input-runtime rows reviewed: 149.
• Existing matrix/dossier anchors: 23.
• Matrix backfills completed: 3 (#5711, #3634, #4961).
• Long-form proof-route backlog: 44.
• Matrix-only future-proof rows: 59.
• Matrix-only non-claim rows: 20.
• Drift rows aligned down to related: 4 (#6022, #5983, #5183, #5088).

Proof gates:

• Unit/runtime proof can guard primitives.
• Browser proof is required for browser-specific closure.
• Synthetic mobile proof stays architecture confidence only.
• Raw-device or matching manual artifacts are required for Android/iOS/Samsung Keyboard/Firefox Android/voice input/Windows IME closure.
• Performance claims need cohort, repeated-unit, listener, DOM/component, memory, and native-behavior evidence.

Synced artifacts:

• docs/slate-v2/ledgers/issue-coverage-matrix.md.
• docs/slate-v2/ledgers/fork-issue-dossier.md.
• docs/slate-issues/gitcrawl-live-open-ledger.md.
• docs/slate-issues/gitcrawl-clusters.md.
• docs/slate-v2/references/pr-description.md.

Implementation handoff:

• After user acceptance, run ralph.
• Before changing runtime code for a Mobile/IME issue, check whether the matching Lexical/ProseMirror stolen-test row above is missing in Slate v2; add or translate that proof first when the route exists.
• Execute proof-heavy batches first; do not bulk-write all 44 long-form rows before code.
• Fix runtime only where proof fails.
• Sync ledgers and PR text after each proof-backed execution slice.

25. Final Completion Gates

Slate Ralplan closure status: done.

Closure gate results:

• total score is at least 0.92: pass, score 0.95;
• no dimension is below 0.85: pass, lowest dimension 0.92;
• full issue-ledger pass is complete: pass;
• live gitcrawl ledger sync is complete for touched rows/clusters: pass, macro sync note appended;
• fork issue dossier sync is complete for reviewed rows: pass, macro section appended and four stale statuses aligned down;
• issue coverage matrix and PR reference are updated or explicitly unchanged: pass;
• ecosystem synthesis is complete: pass;
• high-risk deliberate mode is complete: pass;
• implementation-skill review matrix is complete: pass;
• all scheduled pass rows are complete: pass, including the user-requested cross-editor IME/mobile test-mining addendum;
• no exact Mobile/IME issue is claimed without matching proof: pass.

26. Ralph Execution State

Status: pending.

Activation:

• ralph accepted the completed Slate Ralplan for execution on 2026-05-07T12:45:37+08:00.
• Completion state moved back to pending because runnable proof and execution backlog remains.
• Current slice: slice-57-mobile-ime-next-proof-route-selection.

Slice 1 owner:

• .tmp/slate-v2/playwright/integration/examples/mentions.test.ts.
• .tmp/slate-v2/packages/slate-browser/src/playwright/ime.ts only if the existing IME helper cannot express the translated Lexical row.

Slice 1 target:

• Add one browser proof row translated from Lexical's staged CDP IME tests: compose Japanese text next to a markable inline mention and assert model text, selection, and kernel trace ownership.
• Keep this as proof/characterization first. Runtime code changes are allowed only if the new proof exposes a real failure.

Slice 1 claim policy:

• No exact Fixes #... or Improves #... claim from this slice by default.
• Related pressure rows: inline-boundary IME (#4693, #4136) and broader Mobile/IME proof backlog.
• Keep synthetic/mobile labels honest; desktop Chromium IME proof is not Android or iOS proof.

Slice 1 required evidence:

• Focused Playwright row for the mentions example, ideally Chromium first.
• If the proof changes maintainer-facing claim text, sync docs/slate-v2/references/pr-description.md, docs/slate-v2/ledgers/issue-coverage-matrix.md, docs/slate-v2/ledgers/fork-issue-dossier.md, docs/slate-issues/gitcrawl-live-open-ledger.md, and docs/slate-issues/gitcrawl-clusters.md.
• If no claim text changes, record reference docs: no change here with the reason.

Slice 1 result:

• Added commits staged IME composition before a markable inline mention to .tmp/slate-v2/playwright/integration/examples/mentions.test.ts.
• The row translates Lexical's CDP staged IME pattern into Slate v2's mentions example and asserts inserted model text, preserved inline mention, final selection, and allowed compositionend kernel trace.
• Runtime code changes: none.
• Issue claims changed: none.
• Reference docs: no change, because this is new proof coverage and does not promote any Fixes #... or Improves #... claim.
• Focused proof passed: bun playwright test playwright/integration/examples/mentions.test.ts --project=chromium --grep "commits staged IME composition before a markable inline mention".
• Lint/format focused gate passed: bun run lint:fix playwright/integration/examples/mentions.test.ts.
• Completion gate result: bun run completion-check reports status: pending, which is expected because more Mobile/IME proof batches remain.

Next slice:

• slice-2-lexical-ime-history-proof.
• Candidate owner: .tmp/slate-v2/playwright/integration/examples/rendering-strategy-runtime.test.ts.
• Target: translate Lexical's IME history grouping rows so composed text forms the right undo group and canceled composition does not enter history.

Slice 2 result:

• Added undoes committed IME composition as one history step to .tmp/slate-v2/playwright/integration/examples/rendering-strategy-runtime.test.ts.
• The row translates Lexical's IME history proof into Slate v2's runtime example: staged native composition commits text, one undo removes the committed text, selection returns to the original caret, and compositionend remains an allowed kernel transition.
• Runtime code changes: none.
• Issue claims changed: none.
• Reference docs: no change, because this is proof coverage and does not promote any Fixes #... or Improves #... claim.
• Focused proof passed: bun playwright test playwright/integration/examples/rendering-strategy-runtime.test.ts --project=chromium --grep "undoes committed IME composition as one history step".
• Lint/format focused gate passed: bun run lint:fix playwright/integration/examples/rendering-strategy-runtime.test.ts.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 3 is still runnable.

Next slice:

• slice-3-lexical-ime-cancel-history-proof.
• Candidate owner: .tmp/slate-v2/playwright/integration/examples/rendering-strategy-runtime.test.ts.
• Target: translate Lexical's canceled-composition history row so canceled composition does not enter undo history.

Slice 3 result:

• Added does not push canceled IME composition onto history to .tmp/slate-v2/playwright/integration/examples/rendering-strategy-runtime.test.ts.
• Added local cancelNativeComposition CDP helper because the shared editor.ime.compose helper only expresses committed composition.
• Initial proof attempt used a as a sentinel and failed because the full model text already contains a; corrected the sentinel to ! and reran the same focused proof.
• Runtime code changes: none.
• Issue claims changed: none.
• Reference docs: no change, because this is proof coverage and does not promote any Fixes #... or Improves #... claim.
• Focused proof passed: bun playwright test playwright/integration/examples/rendering-strategy-runtime.test.ts --project=chromium --grep "does not push canceled IME composition onto history".
• Lint/format focused gate passed: bun run lint:fix playwright/integration/examples/rendering-strategy-runtime.test.ts.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 4 is still runnable.

Next slice:

• slice-4-lexical-safari-ime-delete-selection-proof.
• Candidate owner: .tmp/slate-v2/playwright/integration/examples/rendering-strategy-runtime.test.ts or a narrower WebKit/Safari-capable example test.
• Target: translate Lexical regression 8153-safari-ime-delete-selection so a selection can be deleted after composition ends without extra deletion or stale selection behavior.

Slice 4 result:

• Added deletes shell-backed selection after WebKit compositionend to .tmp/slate-v2/playwright/integration/examples/rendering-strategy-runtime.test.ts.
• Added local dispatchCompositionEnd helper to mirror Lexical regression 8153-safari-ime-delete-selection: WebKit fires compositionend, browser select-all creates shell-backed selection, then Backspace deletes the selected content and leaves a collapsed Slate selection at the document start.
• Runtime code changes: none.
• Issue claims changed: none.
• Reference docs: no change, because this is proof coverage and does not promote any Fixes #... or Improves #... claim.
• Focused WebKit proof passed: bun playwright test playwright/integration/examples/rendering-strategy-runtime.test.ts --project=webkit --grep "deletes shell-backed selection after WebKit compositionend".
• Lint/format focused gate passed: bun run lint:fix playwright/integration/examples/rendering-strategy-runtime.test.ts.
• TypeScript root gate passed: bun typecheck:root.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 5 is still runnable.

Next slice:

• slice-5-prosemirror-dom-change-markup-proof.
• Candidate owner: .tmp/slate-v2/playwright/integration/examples/richtext.test.ts or a narrower low-level Slate DOM/browser test.
• Target: translate ProseMirror DOM-change tests for text deletion/typing inside markup and active text-node preservation.

Slice 5 result:

• Added syncs browser text mutations inside bold markup to .tmp/slate-v2/playwright/integration/examples/richtext.test.ts.
• The row translates ProseMirror's markup DOM-change rows into Slate's browser contract: typing inside bold rich produces riZch, preserves the bold markup, advances Slate selection inside the marked text, then Backspace returns to rich with selection restored.
• Initial proof tried to require ProseMirror-style same text-node identity and failed. That exact mechanism is not Slate's contract for marked input; the final row keeps the behavior proof and rejects text-node identity as a required invariant here.
• Runtime code changes: none.
• Issue claims changed: none.
• Reference docs: no change, because this is proof coverage and does not promote any Fixes #... or Improves #... claim.
• Focused proof passed after narrowing the invariant: bun playwright test playwright/integration/examples/richtext.test.ts --project=chromium --grep "syncs browser text mutations inside bold markup".
• Lint/format focused gate passed: bun run lint:fix playwright/integration/examples/richtext.test.ts.
• TypeScript root gate passed: bun typecheck:root.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 6 is still runnable.

Next slice:

• slice-6-prosemirror-ambiguous-replacement-proof.
• Candidate owner: .tmp/slate-v2/playwright/integration/examples/richtext.test.ts.
• Target: translate ProseMirror's ambiguous text replacement row around marked text, proving typed-over text lands in the intended marked node instead of duplicating adjacent content.

Slice 6 result:

• Added resolves ambiguous browser insertion at a mark boundary to .tmp/slate-v2/playwright/integration/examples/richtext.test.ts.
• The row translates ProseMirror's ambiguous replacement pressure into Slate's marked-boundary contract: inserting a space at the end of bold rich before the adjacent unmarked space keeps the inserted character inside the bold leaf, advances selection inside that marked leaf, and Backspace restores the original marked text.
• Runtime code changes: none.
• Issue claims changed: none.
• Reference docs: no change, because this is proof coverage and does not promote any Fixes #... or Improves #... claim.
• Focused proof passed: bun playwright test playwright/integration/examples/richtext.test.ts --project=chromium --grep "resolves ambiguous browser insertion at a mark boundary".
• Lint/format focused gate passed: bun run lint:fix playwright/integration/examples/richtext.test.ts.
• TypeScript root gate passed: bun typecheck:root.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 7 is still runnable.

Next slice:

• slice-7-lexical-ios-autocorrect-plain-text-proof.
• Candidate owner: .tmp/slate-v2/playwright/integration/examples/richtext.test.ts or a narrower clipboard/native-input test.
• Target: translate Lexical's iOS word prediction/autocorrect plain-text row so identical text/html and text/plain prediction payloads do not inject HTML or break selection formatting.

Slice 7 result:

• Added treats iOS prediction payload as plain text inside formatted selection to .tmp/slate-v2/playwright/integration/examples/paste-html.test.ts.
• Updated .tmp/slate-v2/site/examples/ts/paste-html.tsx so the HTML-paste example detects real identical text/html + text/plain payloads and routes them through text insertion, preserving current selection formatting instead of parsing plain prediction text as an HTML fragment.
• The guard checks the actual text/plain data type before comparing payloads, so HTML-only paste events still use the rich HTML path.
• Initial proof failed with TypeError: Cannot read properties of undefined (reading 'length'), proving the old handler misrouted the prediction-shaped payload.
• Runtime code changes: paste-html example extension only; core runtime unchanged.
• Issue claims changed: none.
• Reference docs: no change, because this is proof coverage and example behavior, not an exact Mobile/IME issue closure.
• Focused proof passed: bun playwright test playwright/integration/examples/paste-html.test.ts --project=chromium --grep "treats iOS prediction payload as plain text inside formatted selection".
• Regression proof for the rich HTML path also passed: bun playwright test playwright/integration/examples/paste-html.test.ts --project=chromium --grep "pasted bold text uses|treats iOS prediction payload".
• Lint/format focused gate passed: bun run lint:fix playwright/integration/examples/paste-html.test.ts site/examples/ts/paste-html.tsx.
• TypeScript root gate passed: bun typecheck:root.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 8 is still runnable.

Next slice:

• slice-8-prosemirror-composition-commit-metadata-proof.
• Candidate owner: .tmp/slate-v2/playwright/integration/examples/rendering-strategy-runtime.test.ts or the narrowest existing commit/kernel metadata contract.
• Target: translate ProseMirror's composition transaction metadata pressure into a Slate v2 proof that composition commits carry explicit runtime metadata / trace ownership without promoting exact Android/iOS claims.

Slice 8 result:

• Added records runtime metadata for committed IME composition to .tmp/slate-v2/playwright/integration/examples/rendering-strategy-runtime.test.ts.
• The row translates ProseMirror's composition transaction metadata pressure into Slate v2's runtime proof shape: native Chromium IME composition must produce explicit compositionstart, compositionupdate, and compositionend kernel trace entries with intent: composition, ownership: native-allowed, nativeAllowed: true, editor target ownership, frame IDs, and allowed transitions.
• Runtime code changes: none.
• Issue claims changed: none.
• Reference docs: no change, because this is proof coverage and does not promote any Fixes #... or Improves #... claim.
• Focused proof passed: bun playwright test playwright/integration/examples/rendering-strategy-runtime.test.ts --project=chromium --grep "records runtime metadata for committed IME composition".
• Lint/format focused gate passed: bun run lint:fix playwright/integration/examples/rendering-strategy-runtime.test.ts.
• TypeScript root gate passed: bun typecheck:root.

Next slice:

• slice-9-prosemirror-android-virtual-keyboard-fallback-proof.
• Candidate owner: .tmp/slate-v2/packages/slate-react/test/model-input-strategy-contract.test.ts, .tmp/slate-v2/packages/slate-react/test/editing-kernel-contract.ts, or the narrowest existing Android input-manager contract.
• Target: translate ProseMirror's Android virtual-keyboard fallback pressure into a unit/browser-contract proof for Android insertReplacementText / deleteContentBackward routing through the centralized Android/input runtime. This is architecture proof only, not exact Android device closure.

Slice 9 result:

• Added Android-style replacement/backspace beforeinput coverage to .tmp/slate-v2/packages/slate-react/test/model-input-strategy-contract.test.ts.
• The row translates ProseMirror's Android virtual-keyboard fallback pressure into Slate v2's model-owned input contract: insertReplacementText replaces the current selection through the beforeinput command path, and deleteContentBackward routes through a model-owned delete with explicit caret repair.
• Runtime code changes: none.
• Issue claims changed: none.
• Reference docs: no change, because this is architecture proof and does not promote any exact Android device claim.
• Initial bun test packages/slate-react/test/model-input-strategy-contract.test.ts attempt was the wrong runner because root Bun ignores *.test.*; the package Vitest runner is the correct focused gate.
• Focused proof passed: bun test:vitest test/model-input-strategy-contract.test.ts from .tmp/slate-v2/packages/slate-react.
• Lint/format focused gate passed: bun run lint:fix packages/slate-react/test/model-input-strategy-contract.test.ts.
• TypeScript root gate passed: bun typecheck:root.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 10 is still runnable.

Next slice:

• slice-10-lexical-ime-formatting-boundary-proof.
• Candidate owner: .tmp/slate-v2/playwright/integration/examples/richtext.test.ts.
• Target: translate Lexical's staged CDP IME formatting-boundary rows so native composition inside marked text preserves the mark, advances selection inside the marked leaf, and records composition trace ownership without exact mobile device closure.

Slice 10 result:

• Added commits IME composition inside bold markup to .tmp/slate-v2/playwright/integration/examples/richtext.test.ts.
• The row translates ProseMirror's handles composition inside marks pressure into Slate's marked-text contract: DOM composition inside bold rich at offset 2 produces riすしch, preserves the bold mark, advances selection inside the marked leaf, records composition trace ownership, and two Backspaces restore the original marked text and caret.
• The initial native CDP proof failed when both model selection and DOM selection inserted at the end of the bold leaf (richすし) instead of inside the marked text. A follow-up replacementStart / replacementEnd experiment also failed by targeting the wrong document offset. That route is not an honest proof for mid-mark composition in this harness.
• Runtime code changes: none.
• Issue claims changed: none.
• Reference docs: no change, because this is proof coverage and does not promote any exact Mobile/IME issue claim or native-CDP/mobile-device claim.
• Focused proof passed: bun playwright test playwright/integration/examples/richtext.test.ts --project=chromium --grep "commits IME composition inside bold markup" --retries=0.
• Lint/format focused gate passed: bun run lint:fix playwright/integration/examples/richtext.test.ts.
• TypeScript root gate passed: bun typecheck:root.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 11 is still runnable.
• ce-compound note added: docs/solutions/developer-experience/2026-05-07-slate-browser-ime-proof-rows-need-honest-dom-composition.md.

Next slice:

• slice-11-prosemirror-ime-decoration-boundary-proof.
• Candidate owner: .tmp/slate-v2/playwright/integration/examples/highlighted-text.test.ts.
• Target: translate ProseMirror's composition-with-decorations rows so IME composition inside decorated/highlighted text is not interrupted or overwritten by decoration rendering. This is browser proof only, not exact Android/iOS device closure.

Slice 11 result:

• Added commits IME composition inside decorated text to .tmp/slate-v2/playwright/integration/examples/highlighted-text.test.ts.
• The row translates ProseMirror's composition-with-decorations pressure into Slate's highlighted-text contract: DOM composition inside decorated lph inserts すし at the selected offset, preserves data-tone="warm", advances selection after the committed text, records allowed composition trace ownership, and two Backspaces restore the original text and caret.
• Runtime code changes: none.
• Issue claims changed: none.
• Reference docs: no change, because this is proof coverage and does not promote any exact Mobile/IME issue claim or native-CDP/mobile-device claim.
• Focused proof passed: bun playwright test playwright/integration/examples/highlighted-text.test.ts --project=chromium --grep "commits IME composition inside decorated text" --retries=0.
• Lint/format focused gate passed: bun run lint:fix playwright/integration/examples/highlighted-text.test.ts.
• TypeScript root gate passed: bun typecheck:root.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 12 is still runnable.

Next slice:

• slice-12-prosemirror-ime-spanning-nodes-proof.
• Candidate owner: .tmp/slate-v2/playwright/integration/examples/highlighted-text.test.ts or the narrowest existing browser example that exposes adjacent decorated text nodes.
• Target: translate ProseMirror's composition-spanning-multiple-nodes row so a browser-owned composition that crosses adjacent decorated DOM nodes imports the final text without overwriting nearby rendered decoration content. This is browser proof only, not exact Android/iOS device closure.

Slice 12 result:

• Added commits IME composition spanning decorated text nodes to .tmp/slate-v2/playwright/integration/examples/highlighted-text.test.ts.
• Factored the local DOM composition helper in the highlighted-text suite so the inside-decoration and spanning-decoration rows share the same event/mutation proof shape.
• The first proof failed with visible duplicate DOM text: expected alすしbeta, received alすしすしbeta.
• Debug evidence showed the Slate model was already correct at alすしbeta; the duplicate was a stale unmanaged browser text node outside [data-slate-string="true"] in the projected text host.
• Fixed .tmp/slate-v2/packages/slate-react/src/editable/composition-state.ts so the Chrome composition-end fallback removes unmanaged composition text nodes after writing the committed text into the model.
• Runtime code changes: Chrome composition fallback DOM cleanup for unmanaged text outside Slate string wrappers.
• Issue claims changed: none.
• Reference docs: no change, because this is proof/runtime coverage and does not promote any exact Mobile/IME issue claim or native-CDP/mobile-device claim.
• Focused proof passed after the runtime fix: bun playwright test playwright/integration/examples/highlighted-text.test.ts --project=chromium --grep "commits IME composition spanning decorated text nodes" --retries=0.
• Adjacent decorated IME regression proof passed: bun playwright test playwright/integration/examples/highlighted-text.test.ts --project=chromium --grep "commits IME composition inside decorated text|commits IME composition spanning decorated text nodes" --retries=0.
• Existing richtext mid-mark IME regression proof passed: bun playwright test playwright/integration/examples/richtext.test.ts --project=chromium --grep "commits IME composition inside bold markup" --retries=0.
• Lint/format focused gate passed: bun run lint:fix packages/slate-react/src/editable/composition-state.ts playwright/integration/examples/highlighted-text.test.ts.
• TypeScript root gate passed: bun typecheck:root.
• ce-compound note added: docs/solutions/ui-bugs/2026-05-07-slate-react-chrome-composition-fallback-must-clean-unmanaged-projection-dom-text.md.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 13 is still runnable.

Next slice:

• slice-13-prosemirror-ime-widget-adjacent-proof.
• Candidate owner: .tmp/slate-v2/playwright/integration/examples/mentions.test.ts or the narrowest existing inline-void/widget browser example.
• Target: translate ProseMirror's doesn't overwrite widgets next to the composition row so IME composition next to an inline widget/void preserves the widget and committed text without exact Android/iOS device closure.

Slice 13 result:

• Added commits IME composition between inline mentions without overwriting them to .tmp/slate-v2/playwright/integration/examples/mentions.test.ts.
• The row translates ProseMirror's widget-adjacent composition pressure into Slate's markable-inline mention contract: DOM composition inside the text node between two inline mentions commits すし, preserves both mention widgets, advances selection in the text node, and records allowed composition trace ownership.
• Initial native CDP proof failed honestly: it committed at the end of the between-mentions text (or すし!) instead of the selected offset. The row was switched to the DOM-composition proof shape instead of treating CDP offsets as equivalent.
• Runtime code changes: none.
• Issue claims changed: none.
• Reference docs: no change, because this is proof coverage and does not promote any exact Mobile/IME issue claim or native-CDP/mobile-device claim.
• Focused proof passed: bun playwright test playwright/integration/examples/mentions.test.ts --project=chromium --grep "commits IME composition between inline mentions without overwriting them" --retries=0.
• Existing mention IME regression proof passed: bun playwright test playwright/integration/examples/mentions.test.ts --project=chromium --grep "commits staged IME composition before a markable inline mention|commits IME composition between inline mentions without overwriting them" --retries=0.
• Lint/format focused gate passed: bun run lint:fix playwright/integration/examples/mentions.test.ts packages/slate-react/src/editable/composition-state.ts playwright/integration/examples/highlighted-text.test.ts.
• TypeScript root gate passed: bun typecheck:root.
• ce-compound evaluation: skipped; the CDP-offset limitation is already covered by the slice-10 DOM composition proof note, and this slice added no distinct reusable runtime lesson.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 14 is still runnable.

Next slice:

• slice-14-prosemirror-ime-overlap-cancel-proof.
• Candidate owner: .tmp/slate-v2/playwright/integration/examples/rendering-strategy-runtime.test.ts or the narrowest existing browser/runtime test that can apply a model change during composition.
• Target: translate ProseMirror's composition-cancel rows for overlapping model changes so Slate does not duplicate, lose, or over-claim composition text when a model change overlaps the active composition range. This is browser proof only, not exact Android/iOS device closure.

Slice 14 result:

• Added drops active IME composition when a model change overlaps it to .tmp/slate-v2/playwright/integration/examples/rendering-strategy-runtime.test.ts.
• The row translates ProseMirror's cancels composition when a change fully overlaps with it pressure into Slate's input-runtime contract: a real DOM composition starts inside default block 1, a model-owned replacement overlaps that active composition range, and the later compositionend must not append stale committed composition text.
• The first proof failed with ---すし, proving Chrome's composition-end fallback still wrote the stale composition payload after the overlapping model change had already replaced the range.
• Fixed .tmp/slate-v2/packages/slate-react/src/editable/composition-state.ts and .tmp/slate-v2/packages/slate-react/src/editable/runtime-composition-events.ts so the Chrome composition-end fallback skips model insertion when the active composition has been superseded by a model-owned command, while still cleaning unmanaged composition DOM text.
• A too-tight first fix skipped legitimate DOM-current composition updates and broke the spanning decorated text row by producing albeta; narrowed the cancellation test to the actual model-owned command signal.
• Runtime code changes: Chrome composition fallback cancellation for model-owned overlap, plus the event runtime now passes inputController to composition-end handling.
• Issue claims changed: none.
• Reference docs: no change, because this is proof/runtime coverage and does not promote any exact Mobile/IME issue claim or native-CDP/mobile-device claim.
• Focused proof passed after the runtime fix: bun playwright test playwright/integration/examples/rendering-strategy-runtime.test.ts --project=chromium --grep "drops active IME composition when a model change overlaps it" --retries=0.
• Existing rendering-strategy IME regression proof passed: bun playwright test playwright/integration/examples/rendering-strategy-runtime.test.ts --project=chromium --grep "commits IME composition through the browser editing path|records runtime metadata for committed IME composition|undoes committed IME composition as one history step|does not push canceled IME composition onto history|drops active IME composition when a model change overlaps it" --retries=0.
• Existing richtext mid-mark IME regression proof passed: bun playwright test playwright/integration/examples/richtext.test.ts --project=chromium --grep "commits IME composition inside bold markup" --retries=0.
• Existing decorated IME regression proof passed: bun playwright test playwright/integration/examples/highlighted-text.test.ts --project=chromium --grep "commits IME composition inside decorated text|commits IME composition spanning decorated text nodes" --retries=0.
• Existing mention IME regression proof passed: bun playwright test playwright/integration/examples/mentions.test.ts --project=chromium --grep "commits staged IME composition before a markable inline mention|commits IME composition between inline mentions without overwriting them" --retries=0.
• Lint/format focused gate passed: bun run lint:fix packages/slate-react/src/editable/composition-state.ts packages/slate-react/src/editable/runtime-composition-events.ts playwright/integration/examples/rendering-strategy-runtime.test.ts playwright/integration/examples/highlighted-text.test.ts playwright/integration/examples/mentions.test.ts.
• TypeScript root gate passed: bun typecheck:root.
• ce-compound note added: docs/solutions/ui-bugs/2026-05-07-slate-react-chrome-composition-fallback-must-cancel-model-owned-overlap.md.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 15 is still runnable.

Next slice:

• slice-15-prosemirror-ime-non-overlap-change-proof.
• Candidate owner: .tmp/slate-v2/playwright/integration/examples/rendering-strategy-runtime.test.ts.
• Target: translate ProseMirror's doesn't cancel composition when a change happens elsewhere row so a non-overlapping model change during composition does not cancel, duplicate, or misplace the committed composition text. This is browser proof only, not exact Android/iOS device closure.

Slice 15 result:

• Added keeps active IME composition when a model change happens elsewhere to .tmp/slate-v2/playwright/integration/examples/rendering-strategy-runtime.test.ts.
• The row translates ProseMirror's doesn't cancel composition when a change happens elsewhere pressure into Slate's runtime contract: a real DOM composition starts inside default block 1, a non-overlapping model operation inserts ! into default block 2, and compositionend still commits すし at the original composition point.
• Runtime code changes: none for this slice; it verifies that the slice-14 model-owned cancellation guard does not over-cancel non-overlapping model operations.
• Issue claims changed: none.
• Reference docs: no change, because this is proof/runtime coverage and does not promote any exact Mobile/IME issue claim or native-CDP/mobile-device claim.
• Focused proof passed: bun playwright test playwright/integration/examples/rendering-strategy-runtime.test.ts --project=chromium --grep "keeps active IME composition when a model change happens elsewhere" --retries=0.
• Existing rendering-strategy IME regression proof passed: bun playwright test playwright/integration/examples/rendering-strategy-runtime.test.ts --project=chromium --grep "commits IME composition through the browser editing path|records runtime metadata for committed IME composition|undoes committed IME composition as one history step|does not push canceled IME composition onto history|drops active IME composition when a model change overlaps it|keeps active IME composition when a model change happens elsewhere" --retries=0.
• Lint/format focused gate passed: bun run lint:fix packages/slate-react/src/editable/composition-state.ts packages/slate-react/src/editable/runtime-composition-events.ts playwright/integration/examples/rendering-strategy-runtime.test.ts.
• TypeScript root gate passed: bun typecheck:root.
• Post-lint focused overlap/non-overlap proof passed: bun playwright test playwright/integration/examples/rendering-strategy-runtime.test.ts --project=chromium --grep "drops active IME composition when a model change overlaps it|keeps active IME composition when a model change happens elsewhere" --retries=0.
• ce-compound evaluation: skipped; slice 15 adds a guardrail proof for the slice-14 runtime lesson and does not add a distinct reusable fix pattern.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 16 is still runnable.

Next slice:

• slice-16-prosemirror-ime-partial-overlap-cancel-proof.
• Candidate owner: .tmp/slate-v2/playwright/integration/examples/rendering-strategy-runtime.test.ts.
• Target: translate ProseMirror's partial-overlap / inside-composition cancel rows so a model-owned change that intersects only part of the active composition still cancels the stale Chrome compositionend payload without exact Android/iOS device closure.

Slice 16 result:

• Added drops active IME composition when a model change partially overlaps it to .tmp/slate-v2/playwright/integration/examples/rendering-strategy-runtime.test.ts.
• The row translates ProseMirror's partial-overlap composition-cancel pressure: a real DOM composition starts inside default block 1, a nearby model-owned replacement intersects only part of the active composition neighborhood, and the later compositionend must not append stale すし.
• The first draft used an expanded DOM composition selection; Slate correctly deleted that range on compositionstart, so the row was tightened to a collapsed composition point plus a partial model-owned replacement.
• Runtime code changes: none for this slice; it verifies that the slice-14 fallback cancellation handles narrower overlap pressure.
• Issue claims changed: none.
• Reference docs: no change, because this is proof/runtime coverage and does not promote any exact Mobile/IME issue claim or native-CDP/mobile-device claim.
• Focused proof passed: bun playwright test playwright/integration/examples/rendering-strategy-runtime.test.ts --project=chromium --grep "drops active IME composition when a model change partially overlaps it" --retries=0.
• Focused overlap/non-overlap contract proof passed: bun playwright test playwright/integration/examples/rendering-strategy-runtime.test.ts --project=chromium --grep "drops active IME composition when a model change overlaps it|drops active IME composition when a model change partially overlaps it|keeps active IME composition when a model change happens elsewhere" --retries=0.
• Existing rendering-strategy IME regression proof passed: bun playwright test playwright/integration/examples/rendering-strategy-runtime.test.ts --project=chromium --grep "commits IME composition through the browser editing path|records runtime metadata for committed IME composition|undoes committed IME composition as one history step|does not push canceled IME composition onto history|drops active IME composition when a model change overlaps it|drops active IME composition when a model change partially overlaps it|keeps active IME composition when a model change happens elsewhere" --retries=0.
• Lint/format focused gate passed: bun run lint:fix packages/slate-react/src/editable/composition-state.ts packages/slate-react/src/editable/runtime-composition-events.ts playwright/integration/examples/rendering-strategy-runtime.test.ts.
• TypeScript root gate passed: bun typecheck:root.
• ce-compound evaluation: skipped; slice 16 extends the slice-14 fallback cancellation lesson and does not add a distinct reusable fix pattern.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 17 is still runnable.

Next slice:

• slice-17-prosemirror-ime-inside-change-cancel-proof.
• Candidate owner: .tmp/slate-v2/playwright/integration/examples/rendering-strategy-runtime.test.ts.
• Target: translate ProseMirror's cancels composition when a change happens inside of it row so an inside-range model-owned change cancels the stale Chrome compositionend payload without exact Android/iOS device closure.

Slice 17 result:

• Added drops active IME composition when a model change happens at its insertion point to .tmp/slate-v2/playwright/integration/examples/rendering-strategy-runtime.test.ts.
• The row translates ProseMirror's inside-composition cancel pressure into a Slate runtime row: a real DOM composition starts at the default block 1 insertion point, a model-owned insert happens at that same point, and the later compositionend must not append stale すし.
• Runtime code changes: none for this slice; it verifies that the slice-14 fallback cancellation also covers collapsed inside-point model changes.
• Issue claims changed: none.
• Reference docs: no change, because this is proof/runtime coverage and does not promote any exact Mobile/IME issue claim or native-CDP/mobile-device claim.
• Focused proof passed: bun playwright test playwright/integration/examples/rendering-strategy-runtime.test.ts --project=chromium --grep "drops active IME composition when a model change happens at its insertion point" --retries=0.
• Focused overlap/non-overlap contract proof passed: bun playwright test playwright/integration/examples/rendering-strategy-runtime.test.ts --project=chromium --grep "drops active IME composition when a model change overlaps it|drops active IME composition when a model change partially overlaps it|drops active IME composition when a model change happens at its insertion point|keeps active IME composition when a model change happens elsewhere" --retries=0.
• Existing rendering-strategy IME regression proof passed: bun playwright test playwright/integration/examples/rendering-strategy-runtime.test.ts --project=chromium --grep "commits IME composition through the browser editing path|records runtime metadata for committed IME composition|undoes committed IME composition as one history step|does not push canceled IME composition onto history|drops active IME composition when a model change overlaps it|drops active IME composition when a model change partially overlaps it|drops active IME composition when a model change happens at its insertion point|keeps active IME composition when a model change happens elsewhere" --retries=0.
• Lint/format focused gate passed: bun run lint:fix packages/slate-react/src/editable/composition-state.ts packages/slate-react/src/editable/runtime-composition-events.ts playwright/integration/examples/rendering-strategy-runtime.test.ts.
• TypeScript root gate passed: bun typecheck:root.
• ce-compound evaluation: skipped; slice 17 extends the slice-14 fallback cancellation lesson and does not add a distinct reusable fix pattern.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 18 is still runnable.

Next slice:

• slice-18-prosemirror-ime-rapid-following-proof.
• Candidate owner: .tmp/slate-v2/playwright/integration/examples/rendering-strategy-runtime.test.ts.
• Target: translate ProseMirror's handles compositions rapidly following each other row so two back-to-back DOM compositions in separate text blocks commit without stale composing state, text loss, or caret drift. This is browser proof only, not exact Android/iOS device closure.

Slice 18 result:

• Added commits rapidly following IME compositions in separate text blocks to .tmp/slate-v2/playwright/integration/examples/rendering-strategy-runtime.test.ts.
• The row translates ProseMirror's rapidly-following composition pressure into Slate's runtime contract: native Chromium IME composition commits at the end of default block 1, immediately followed by another native composition at the end of default block 2, with both model text and final caret preserved.
• The first assertion expected all six blocks from assert.blockTexts; the harness only reports the mounted visible block list for this surface, so the row now asserts the two visible edited blocks and separately checks model text still contains default block 6.
• Runtime code changes: none.
• Issue claims changed: none.
• Reference docs: no change, because this is proof coverage and does not promote any exact Mobile/IME issue claim or native-CDP/mobile-device claim.
• Focused proof passed: bun playwright test playwright/integration/examples/rendering-strategy-runtime.test.ts --project=chromium --grep "commits rapidly following IME compositions in separate text blocks" --retries=0.
• Existing rendering-strategy IME regression proof passed: bun playwright test playwright/integration/examples/rendering-strategy-runtime.test.ts --project=chromium --grep "commits IME composition through the browser editing path|records runtime metadata for committed IME composition|undoes committed IME composition as one history step|does not push canceled IME composition onto history|drops active IME composition when a model change overlaps it|drops active IME composition when a model change partially overlaps it|drops active IME composition when a model change happens at its insertion point|keeps active IME composition when a model change happens elsewhere|commits rapidly following IME compositions in separate text blocks" --retries=0.
• Lint/format focused gate passed: bun run lint:fix packages/slate-react/src/editable/composition-state.ts packages/slate-react/src/editable/runtime-composition-events.ts playwright/integration/examples/rendering-strategy-runtime.test.ts.
• TypeScript root gate passed: bun typecheck:root.
• ce-compound evaluation: skipped; slice 18 adds proof coverage but no distinct reusable runtime fix pattern.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 19 is still runnable.

Next slice:

• slice-19-prosemirror-ime-cross-paragraph-proof.
• Candidate owner: .tmp/slate-v2/playwright/integration/examples/rendering-strategy-runtime.test.ts.
• Target: translate ProseMirror's can handle cross-paragraph compositions row into a Slate browser proof for composition spanning block boundaries without text loss, duplicated committed text, or exact Android/iOS device closure.

Slice 19 result:

• Added commits cross-paragraph IME composition as one replacement plus a local selectDOMTextRange helper to .tmp/slate-v2/playwright/integration/examples/rendering-strategy-runtime.test.ts.
• The row translates ProseMirror's cross-paragraph composition pressure into a Slate runtime row: a browser DOM composition spanning from default block 1 into default block 2 replaces the selected block range, merges the remaining suffix into the first block, preserves later model text, and leaves both model selection and DOM caret after the committed composition text.
• The first setup targeted default block 3, but that default example region is wrapped by the document-section control and the generic DOM selection assertion returned null. The final proof uses the first two editable text blocks and an explicit DOM range while syncing Slate's model selection through the browser handle.
• Runtime code changes: none.
• Issue claims changed: none.
• Reference docs: no change, because this is proof coverage and does not promote any exact Mobile/IME issue claim or native-CDP/mobile-device claim.
• Focused proof passed: bun playwright test playwright/integration/examples/rendering-strategy-runtime.test.ts --project=chromium --grep "commits cross-paragraph IME composition as one replacement" --retries=0.
• Rendering-strategy IME regression proof passed: bun playwright test playwright/integration/examples/rendering-strategy-runtime.test.ts --project=chromium --grep "commits IME composition through|records runtime metadata for committed IME composition|undoes committed IME composition as one history step|does not push canceled IME composition onto history|drops active IME composition when a model change overlaps it|drops active IME composition when a model change partially overlaps it|drops active IME composition when a model change happens at its insertion point|keeps active IME composition when a model change happens elsewhere|commits rapidly following IME compositions in separate text blocks|commits cross-paragraph IME composition as one replacement" --retries=0.
• Lint/format focused gate passed: bun run lint:fix playwright/integration/examples/rendering-strategy-runtime.test.ts.
• TypeScript root gate passed: bun typecheck:root.
• ce-compound evaluation: skipped; slice 19 extends the existing honest DOM composition proof-row lesson and did not add a distinct runtime fix pattern.

Next slice:

• slice-20-lexical-ios-korean-backspace-native-proof.
• Candidate owner: .tmp/slate-v2/packages/slate-react/src/editable/keyboard-input-strategy.ts.
• Target: translate Lexical's iOS Korean Backspace exception into a narrow Slate runtime contract: iOS + ko-KR + Backspace should defer to native input instead of becoming model-owned from keydown. This is runtime proof only, not exact raw-device iOS closure.

Slice 20 result:

• Added shouldDeferBackspaceToNativeInput to .tmp/slate-v2/packages/slate-react/src/editable/keyboard-input-strategy.ts.
• Added .tmp/slate-v2/packages/slate-react/test/keyboard-input-strategy-contract.test.ts.
• The row translates Lexical's iOS Korean Backspace guard from ../lexical/packages/lexical-plain-text/src/index.ts:286 and ../lexical/packages/lexical-rich-text/src/index.ts:893 into Slate's runtime contract.
• The first proof failed with TypeError: shouldDeferBackspaceToNativeInput is not a function.
• Runtime code now returns keyDownUnhandled() before destructive Backspace is model-owned when the narrow predicate is true: iOS, navigator.language === 'ko-KR', and backward delete.
• Runtime code still keeps non-Korean iOS Backspace and non-Backspace keys out of this exception.
• Issue claims changed: none.
• Reference docs: no change, because this is a runtime contract proof and does not promote any exact iOS keyboard or raw-device claim.
• Focused proof passed: bun test:vitest test/keyboard-input-strategy-contract.test.ts.
• Adjacent package contract proof passed: bun test:vitest test/keyboard-input-strategy-contract.test.ts test/model-input-strategy-contract.test.ts test/editing-kernel-contract.test.ts.
• Lint/format focused gate passed: bun run lint:fix packages/slate-react/src/editable/keyboard-input-strategy.ts packages/slate-react/test/keyboard-input-strategy-contract.test.ts.
• Post-lint adjacent package contract proof passed: bun test:vitest test/keyboard-input-strategy-contract.test.ts test/model-input-strategy-contract.test.ts test/editing-kernel-contract.test.ts.
• TypeScript root gate passed: bun typecheck:root.
• ce-compound note added: docs/solutions/ui-bugs/2026-05-07-slate-react-ios-korean-backspace-must-stay-native-owned.md.

Next slice:

• slice-21-lexical-ime-typeahead-menu-proof.
• Candidate owner: .tmp/slate-v2/playwright/integration/examples/mentions.test.ts.
• Target: translate Lexical's Typeahead menu should not close during IME composition row so the mentions portal stays visible while Chromium IME composition is active. This is desktop Chromium overlay proof only, not exact mobile-device closure.

Slice 21 result:

• Added keeps mention portal open during IME composition to .tmp/slate-v2/playwright/integration/examples/mentions.test.ts.
• The row translates Lexical's typeahead/IME overlay pressure into Slate's mentions example: after @ma opens the portal, active Chromium IME composition must not close it while composition text is still in progress.
• Runtime code changes: none.
• Issue claims changed: none.
• Reference docs: no change, because this is proof coverage and does not promote any exact Mobile/IME issue claim or native-CDP/mobile-device claim.
• Focused proof passed: bun playwright test playwright/integration/examples/mentions.test.ts --project=chromium --grep "keeps mention portal open during IME composition" --retries=0.
• Existing mention IME regression proof passed: bun playwright test playwright/integration/examples/mentions.test.ts --project=chromium --grep "keeps mention portal open during IME composition|commits staged IME composition before a markable inline mention|commits IME composition between inline mentions without overwriting them" --retries=0.
• Lint/format focused gate passed: bun run lint:fix playwright/integration/examples/mentions.test.ts.
• TypeScript root gate passed: bun typecheck:root.
• ce-compound evaluation: skipped; this is a proof-only port of a Lexical overlay row and did not add a distinct reusable runtime fix pattern.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 22 is still runnable.

Slice 22 result:

• Added replaces multiple formatted text nodes with Korean IME composition to .tmp/slate-v2/playwright/integration/examples/richtext.test.ts.
• The row translates Lexical's Can replace multiple formatted text nodes with IME composition (Korean) into a Slate rich-text proof using Chromium CDP Input.imeSetComposition.
• The first synthetic proof was intentionally red but too weak: it changed DOM text while Slate model text still contained the old content.
• The CDP proof exposed the real runtime issue: trusted native composition needs model-selection pre-delete, visible mark capture, model-owned insert while composing, and cleanup of stale browser-owned composition text in managed Slate strings.
• Runtime changes: .tmp/slate-v2/packages/slate-react/src/editable/composition-state.ts now gates expanded-selection pre-delete on trusted native composition, captures composition marks synchronously, tracks latest composition text, and repairs stale composition text left inside managed DOM strings.
• Runtime changes: .tmp/slate-v2/packages/slate-react/src/editable/mutation-controller.ts now requests a render after model-owned text insertion during composition.
• The adjacent bold IME row now asserts semantic mark preservation across split bold leaves instead of requiring React to merge adjacent <strong> output.
• Issue claims changed: none.
• Reference docs: no change, because this is desktop Chromium CDP proof and does not promote exact Mobile/IME issue closure or raw-device claims.
• ce-compound note added: docs/solutions/ui-bugs/2026-05-07-slate-react-ime-formatted-selection-needs-native-owned-cleanup.md.
• Focused proof passed: bun playwright test playwright/integration/examples/richtext.test.ts --project=chromium --grep "replaces multiple formatted text nodes with Korean IME composition" --retries=0.
• Richtext regression proof passed: bun playwright test playwright/integration/examples/richtext.test.ts --project=chromium --grep "syncs browser text mutations inside bold markup|commits IME composition inside bold markup|replaces multiple formatted text nodes with Korean IME composition" --retries=0.
• Rendering-strategy IME regression proof passed: bun playwright test playwright/integration/examples/rendering-strategy-runtime.test.ts --project=chromium --grep "commits IME composition through|records runtime metadata for committed IME composition|undoes committed IME composition as one history step|does not push canceled IME composition onto history|drops active IME composition when a model change overlaps it|drops active IME composition when a model change partially overlaps it|drops active IME composition when a model change happens at its insertion point|keeps active IME composition when a model change happens elsewhere|commits rapidly following IME compositions in separate text blocks|commits cross-paragraph IME composition as one replacement" --retries=0.
• Lint/format focused gate passed: bun run lint:fix packages/slate-react/src/editable/composition-state.ts packages/slate-react/src/editable/mutation-controller.ts playwright/integration/examples/richtext.test.ts.
• TypeScript root gate passed: bun typecheck:root.

Next slice:

• slice-23-mobile-ime-next-proof-route-selection.
• Candidate owner: docs/plans/2026-05-07-slate-v2-mobile-ime-input-runtime-ralplan.md plus the narrowest matching .tmp/slate-v2 source/test owner selected from the cross-editor backlog.
• Target: choose the next missing Mobile/IME proof route after the ProseMirror composition rows, Lexical iOS Korean native-backspace contract, Lexical typeahead overlay row, and Lexical Korean multi-format composition row, preserving raw-device boundaries.

Slice 23 result:

• Selected the Lexical Safari/WebKit delete-after-composition regression route.
• Local evidence read: ../lexical/packages/lexical-playground/__tests__/regression/8153-safari-ime-delete-selection.spec.mjs.
• The route owns two WebKit-only rows: select-all + Backspace after compositionend, and Shift+ArrowUp range deletion after compositionend.
• Decision: translate both rows into richtext WebKit browser proof because they exercise real browser selection/deletion after Safari-style composition timing. This is WebKit browser proof, not exact iOS/raw-device closure.

Slice 24 result:

• Added dispatchWebKitCompositionEnd to .tmp/slate-v2/playwright/integration/examples/richtext.test.ts.
• Added deletes rich text selection after WebKit compositionend.
• Added deletes rich text line selection after WebKit compositionend.
• The rows translate Lexical regression #8153 into Slate richtext browser proof: Safari-style compositionend must not leave stale composition state that prevents select-all deletion or native line-selection deletion.
• Initial focused proof was red on assertion shape, not runtime behavior: WebKit serializes block selection text with line breaks, and the richtext empty document renders Enter some rich text… as visible placeholder text while Slate model text is empty.
• Runtime code changes: none.
• Issue claims changed: none.
• Reference docs: no change, because this is WebKit browser proof and does not promote exact Mobile/IME issue closure or raw-device claims.
• Focused proof passed: bun playwright test playwright/integration/examples/richtext.test.ts --project=webkit --grep "deletes rich text (selection|line selection) after WebKit compositionend" --retries=0.
• Richtext WebKit deletion regression proof passed: bun playwright test playwright/integration/examples/richtext.test.ts --project=webkit --grep "deletes rich text (selection|line selection) after WebKit compositionend|keeps caret editable after browser Backspace deletes selected range|removes the current block after browser triple click and Backspace" --retries=0.
• Existing rendering-strategy WebKit composition regression proof passed: bun playwright test playwright/integration/examples/rendering-strategy-runtime.test.ts --project=webkit --grep "deletes shell-backed selection after WebKit compositionend" --retries=0.
• Focused lint/format passed: bun run lint:fix playwright/integration/examples/richtext.test.ts.
• TypeScript root gate passed: bun typecheck:root.
• Post-lint focused WebKit proof passed: bun playwright test playwright/integration/examples/richtext.test.ts --project=webkit --grep "deletes rich text (selection|line selection) after WebKit compositionend" --retries=0.
• ce-compound evaluation: skipped; this reused existing IME/WebKit proof lessons and did not add a distinct reusable runtime fix pattern.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 25 is still runnable.

Next slice:

• slice-25-mobile-ime-next-proof-route-selection.
• Candidate owner: docs/plans/2026-05-07-slate-v2-mobile-ime-input-runtime-ralplan.md plus the narrowest matching .tmp/slate-v2 source/test owner selected from the remaining Lexical/ProseMirror Mobile/IME backlog.
• Target: choose the next missing Mobile/IME proof route after the ProseMirror composition rows, Lexical iOS Korean native-backspace contract, Lexical typeahead overlay row, Lexical Korean multi-format composition row, and Lexical Safari delete-after-composition rows, preserving raw-device boundaries.

Slice 25 result:

• Selected the Lexical history row RangeSelection should be retained when undo IME.
• Local evidence read: ../lexical/packages/lexical-playground/__tests__/e2e/History.spec.mjs.
• Decision: translate the row into Slate's rendering-strategy runtime example using Chromium CDP IME over an expanded backward DOM selection. This is desktop Chromium IME/history proof, not exact mobile-device closure.

Slice 26 result:

• Added restores expanded selection after undoing IME replacement to .tmp/slate-v2/playwright/integration/examples/rendering-strategy-runtime.test.ts.
• The row translates Lexical's undo-selection pressure into Slate's runtime contract: replacing the selected b in ab with IME text yields aす, and one undo restores both ab and the original backward selection.
• The first proof failed: undo left a, proving trusted native composition pre-delete and the Chrome composition-end fallback insert were separate history batches.
• Runtime change: .tmp/slate-v2/packages/slate-react/src/editable/composition-state.ts now tracks trusted native expanded-selection pre-delete and merges the following Chrome composition-end fallback insert into that history batch.
• Issue claims changed: none.
• Reference docs: no change, because this is desktop Chromium CDP proof and does not promote exact Mobile/IME issue closure or raw-device claims.
• ce-compound note added: docs/solutions/ui-bugs/2026-05-07-slate-react-ime-replacement-undo-must-merge-native-predelete.md.
• Rendering-strategy IME regression proof passed: bun playwright test playwright/integration/examples/rendering-strategy-runtime.test.ts --project=chromium --grep "commits IME composition through|records runtime metadata for committed IME composition|undoes committed IME composition as one history step|restores expanded selection after undoing IME replacement|does not push canceled IME composition onto history|drops active IME composition when a model change overlaps it|drops active IME composition when a model change partially overlaps it|drops active IME composition when a model change happens at its insertion point|keeps active IME composition when a model change happens elsewhere|commits rapidly following IME compositions in separate text blocks|commits cross-paragraph IME composition as one replacement" --retries=0.
• Richtext Chromium IME regression proof passed: bun playwright test playwright/integration/examples/richtext.test.ts --project=chromium --grep "syncs browser text mutations inside bold markup|commits IME composition inside bold markup|replaces multiple formatted text nodes with Korean IME composition" --retries=0.
• WebKit rendering-strategy regression proof passed: bun playwright test playwright/integration/examples/rendering-strategy-runtime.test.ts --project=webkit --grep "deletes shell-backed selection after WebKit compositionend" --retries=0.
• WebKit richtext regression proof passed: bun playwright test playwright/integration/examples/richtext.test.ts --project=webkit --grep "deletes rich text (selection|line selection) after WebKit compositionend" --retries=0.
• Lint/format focused gate passed: bun run lint:fix packages/slate-react/src/editable/composition-state.ts playwright/integration/examples/rendering-strategy-runtime.test.ts playwright/integration/examples/richtext.test.ts.
• TypeScript root gate passed: bun typecheck:root.

Next slice:

• slice-27-lexical-floating-toolbar-ime-proof.
• Candidate owner: .tmp/slate-v2/playwright/integration/examples/hovering-toolbar.test.ts.
• Target: translate Lexical's Floating toolbar should not be displayed when using IME row so the hovering toolbar stays hidden while Chromium IME composition is active, then appears for a real expanded selection after commit. This is desktop Chromium overlay proof only, not exact mobile-device closure.

Slice 27 result:

• Selected the Lexical floating-toolbar IME row from ../lexical/packages/lexical-playground/__tests__/e2e/Composition.spec.mjs.
• Decision: use Slate's hovering-toolbar example because it has the same selection-driven overlay behavior and an existing Playwright owner.
• No runtime or product-code edit was planned before proof.

Slice 28 result:

• Added keeps hovering toolbar hidden during IME composition to .tmp/slate-v2/playwright/integration/examples/hovering-toolbar.test.ts.
• The row proves active Chromium IME composition does not display the hovering toolbar, while a real expanded DOM selection after commit still displays it.
• Runtime code changes: none.
• Issue claims changed: none.
• Reference docs: no change, because this is desktop Chromium overlay proof and does not promote any exact Mobile/IME issue claim or raw-device claim.
• Focused proof passed: bun playwright test playwright/integration/examples/hovering-toolbar.test.ts --project=chromium --grep "keeps hovering toolbar hidden during IME composition" --retries=0.
• Hovering-toolbar regression proof passed: bun playwright test playwright/integration/examples/hovering-toolbar.test.ts --project=chromium --grep "hovering toolbar" --retries=0.
• Lint/format focused gate passed: bun run lint:fix playwright/integration/examples/hovering-toolbar.test.ts.
• TypeScript root gate passed: bun typecheck:root.
• Accidental broad examples command was not used as slice evidence: bun playwright test playwright/integration/examples/hovering-toolbar.test.ts --project=chromium --retries=0 expanded to the broader Chromium examples suite and failed on unrelated check-list, persistent-annotation, table, and huge-document rows. The targeted hovering-toolbar grep above is the relevant proof.
• ce-compound evaluation: skipped; this is a proof-only port of a Lexical overlay row and did not add a distinct reusable runtime fix pattern.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 29 is still runnable.

Next slice:

• slice-29-mobile-ime-next-proof-route-selection.
• Candidate owner: docs/plans/2026-05-07-slate-v2-mobile-ime-input-runtime-ralplan.md plus the narrowest matching .tmp/slate-v2 source/test owner selected from the remaining Lexical/ProseMirror Mobile/IME backlog.
• Target: choose the next missing Mobile/IME proof route after the covered ProseMirror composition rows, Lexical iOS Korean native-backspace contract, Lexical typeahead overlay row, Lexical floating-toolbar overlay row, Lexical Korean multi-format composition row, Lexical IME undo-selection row, and Lexical Safari delete-after-composition rows, preserving raw-device boundaries.

Slice 29 result:

• Selected the Lexical history row Merge IME input when less delay from ../lexical/packages/lexical-playground/__tests__/e2e/History.spec.mjs.
• Decision: translate the Slate-owned part of the row, not Lexical's literal millisecond history policy. Slate history currently merges by operation shape and explicit metadata, so the honest proof is immediate native text plus immediately following IME composition undoing as one user action.
• No runtime code edit was planned before proof.

Slice 30 result:

• Added undoes native text and immediately following IME composition together to .tmp/slate-v2/playwright/integration/examples/rendering-strategy-runtime.test.ts.
• The row proves that browser-native Input.insertText("a") followed by Chromium IME composition す becomes one undoable action, restoring the original text and caret after a single undo.
• Runtime code changes: none.
• Issue claims changed: none.
• Reference docs: no change, because this is desktop Chromium IME/history proof and does not promote exact Mobile/IME issue closure or raw-device claims.
• Focused proof passed: bun playwright test playwright/integration/examples/rendering-strategy-runtime.test.ts --project=chromium --grep "undoes native text and immediately following IME composition together" --retries=0.
• Rendering-strategy IME regression proof passed: bun playwright test playwright/integration/examples/rendering-strategy-runtime.test.ts --project=chromium --grep "commits IME composition through|records runtime metadata for committed IME composition|undoes committed IME composition as one history step|undoes native text and immediately following IME composition together|restores expanded selection after undoing IME replacement|does not push canceled IME composition onto history|drops active IME composition when a model change overlaps it|drops active IME composition when a model change partially overlaps it|drops active IME composition when a model change happens at its insertion point|keeps active IME composition when a model change happens elsewhere|commits rapidly following IME compositions in separate text blocks|commits cross-paragraph IME composition as one replacement" --retries=0.
• Lint/format focused gate passed: bun run lint:fix playwright/integration/examples/rendering-strategy-runtime.test.ts.
• TypeScript root gate passed: bun typecheck:root.
• ce-compound evaluation: skipped; this is a proof-only port of a Lexical history row and did not add a distinct reusable runtime fix pattern.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 31 is still runnable.

Next slice:

• slice-31-mobile-ime-next-proof-route-selection.
• Candidate owner: docs/plans/2026-05-07-slate-v2-mobile-ime-input-runtime-ralplan.md plus the narrowest matching .tmp/slate-v2 source/test owner selected from the remaining Lexical/ProseMirror Mobile/IME backlog.
• Target: choose the next missing Mobile/IME proof route after the covered ProseMirror composition rows, Lexical iOS Korean native-backspace contract, Lexical typeahead and floating-toolbar overlay rows, Lexical Korean multi-format composition row, Lexical IME undo-selection and immediate text-plus-IME history rows, and Lexical Safari delete-after-composition rows, preserving raw-device boundaries.

Slice 31 result:

• Considered ProseMirror's can deal with Android-style newline-after-composition row from ../prosemirror/view/test/webtest-composition.ts.
• Deferred that row because the source mutates the editor DOM into a new block shape and collapses into the new line, which is a larger DOM-change owner and easy to over-claim without Android proof.
• Selected Lexical's Can type, delete and cancel Hiragana via IME row from ../lexical/packages/lexical-playground/__tests__/e2e/Composition.spec.mjs instead.
• Decision: translate the cancellation stability contract into Slate's rendering-strategy runtime example using Chromium CDP IME. This is desktop Chromium IME proof, not exact mobile-device closure.

Slice 32 result:

• Added keeps text stable after type-delete-cancel IME composition to .tmp/slate-v2/playwright/integration/examples/rendering-strategy-runtime.test.ts.
• The row proves a staged Hiragana composition that types, deletes back to a shorter candidate, then cancels leaves the model text stable and keeps the selection stable both at the end of text and before a following space.
• Runtime code changes: none.
• Issue claims changed: none.
• Reference docs: no change, because this is desktop Chromium IME proof and does not promote exact Mobile/IME issue closure or raw-device claims.
• Focused proof passed: bun playwright test playwright/integration/examples/rendering-strategy-runtime.test.ts --project=chromium --grep "keeps text stable after type-delete-cancel IME composition" --retries=0.
• Rendering-strategy IME regression proof passed: bun playwright test playwright/integration/examples/rendering-strategy-runtime.test.ts --project=chromium --grep "commits IME composition through|records runtime metadata for committed IME composition|undoes committed IME composition as one history step|undoes native text and immediately following IME composition together|restores expanded selection after undoing IME replacement|does not push canceled IME composition onto history|keeps text stable after type-delete-cancel IME composition|drops active IME composition when a model change overlaps it|drops active IME composition when a model change partially overlaps it|drops active IME composition when a model change happens at its insertion point|keeps active IME composition when a model change happens elsewhere|commits rapidly following IME compositions in separate text blocks|commits cross-paragraph IME composition as one replacement" --retries=0.
• Lint/format focused gate passed: bun run lint:fix playwright/integration/examples/rendering-strategy-runtime.test.ts.
• TypeScript root gate passed: bun typecheck:root.
• ce-compound evaluation: skipped; this is a proof-only port of a Lexical cancellation row and did not add a distinct reusable runtime fix pattern.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 33 is still runnable.

Next slice:

• slice-33-mobile-ime-next-proof-route-selection.
• Candidate owner: docs/plans/2026-05-07-slate-v2-mobile-ime-input-runtime-ralplan.md plus the narrowest matching .tmp/slate-v2 source/test owner selected from the remaining Lexical/ProseMirror Mobile/IME backlog.
• Target: choose the next missing Mobile/IME proof route after the covered ProseMirror composition rows, Lexical iOS Korean native-backspace contract, Lexical typeahead and floating-toolbar overlay rows, Lexical Korean multi-format composition row, Lexical IME undo-selection and immediate text-plus-IME history rows, Lexical type/delete/cancel row, and Lexical Safari delete-after-composition rows, preserving raw-device boundaries.

Slice 33 result:

• Selected ProseMirror's handles replacement of existing words row from ../prosemirror/view/test/webtest-composition.ts.
• Decision: translate it into Slate's rendering-strategy runtime example as a plain existing-word IME replacement row. This gives a simpler baseline than the existing formatted-selection and multi-format replacement rows.

Slice 34 result:

• Added replaces an existing word with IME composition to .tmp/slate-v2/playwright/integration/examples/rendering-strategy-runtime.test.ts.
• The row proves selecting two in one two three and committing IME text replaces only that word, preserves surrounding text, advances selection after the committed text, and records an allowed compositionend trace.
• Runtime code changes: none.
• Issue claims changed: none.
• Reference docs: no change, because this is desktop Chromium IME proof and does not promote exact Mobile/IME issue closure or raw-device claims.
• Focused proof passed: bun playwright test playwright/integration/examples/rendering-strategy-runtime.test.ts --project=chromium --grep "replaces an existing word with IME composition" --retries=0.
• Rendering-strategy IME regression proof passed: bun playwright test playwright/integration/examples/rendering-strategy-runtime.test.ts --project=chromium --grep "commits IME composition through|records runtime metadata for committed IME composition|undoes committed IME composition as one history step|undoes native text and immediately following IME composition together|restores expanded selection after undoing IME replacement|replaces an existing word with IME composition|does not push canceled IME composition onto history|keeps text stable after type-delete-cancel IME composition|drops active IME composition when a model change overlaps it|drops active IME composition when a model change partially overlaps it|drops active IME composition when a model change happens at its insertion point|keeps active IME composition when a model change happens elsewhere|commits rapidly following IME compositions in separate text blocks|commits cross-paragraph IME composition as one replacement" --retries=0.
• Lint/format focused gate passed: bun run lint:fix playwright/integration/examples/rendering-strategy-runtime.test.ts.
• TypeScript root gate passed: bun typecheck:root.
• ce-compound evaluation: skipped; this is a proof-only port of a ProseMirror replacement row and did not add a distinct reusable runtime fix pattern.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 35 is still runnable.

Next slice:

• slice-35-mobile-ime-next-proof-route-selection.
• Candidate owner: docs/plans/2026-05-07-slate-v2-mobile-ime-input-runtime-ralplan.md plus the narrowest matching .tmp/slate-v2 source/test owner selected from the remaining Lexical/ProseMirror Mobile/IME backlog.
• Target: choose the next missing Mobile/IME proof route after the covered ProseMirror composition rows, ProseMirror existing-word replacement row, Lexical iOS Korean native-backspace contract, Lexical typeahead and floating-toolbar overlay rows, Lexical Korean multi-format composition row, Lexical IME undo-selection and immediate text-plus-IME history rows, Lexical type/delete/cancel row, and Lexical Safari delete-after-composition rows, preserving raw-device boundaries.

Slice 35 result:

• Selected ProseMirror's supports composition in a cursor wrapper row from ../prosemirror/view/test/webtest-composition.ts.
• Decision: translate ProseMirror stored-mark cursor composition into Slate's richtext example as active bold mark plus an empty block. This is desktop Chromium IME proof, not exact mobile-device closure.

Slice 36 result:

• Added commits IME composition through an active mark in an empty block to .tmp/slate-v2/playwright/integration/examples/richtext.test.ts.
• The row proves an empty richtext block with active bold mark accepts Chromium IME composition, inserts abc, renders the committed text as <strong>, advances selection after the committed text, and records an allowed compositionend trace.
• Runtime code changes: none.
• Issue claims changed: none.
• Reference docs: no change, because this is desktop Chromium IME proof and does not promote exact Mobile/IME issue closure or raw-device claims.
• Focused proof passed: bun playwright test playwright/integration/examples/richtext.test.ts --project=chromium --grep "commits IME composition through an active mark in an empty block" --retries=0.
• Richtext Chromium IME regression proof passed: bun playwright test playwright/integration/examples/richtext.test.ts --project=chromium --grep "syncs browser text mutations inside bold markup|commits IME composition inside bold markup|commits IME composition through an active mark in an empty block|replaces multiple formatted text nodes with Korean IME composition" --retries=0.
• Lint/format focused gate passed: bun run lint:fix playwright/integration/examples/richtext.test.ts.
• TypeScript root gate passed: bun typecheck:root.
• ce-compound evaluation: skipped; this is a proof-only port of a ProseMirror cursor-wrapper row and did not add a distinct reusable runtime fix pattern.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 37 is still runnable.

Next slice:

• slice-37-mobile-ime-next-proof-route-selection.
• Candidate owner: docs/plans/2026-05-07-slate-v2-mobile-ime-input-runtime-ralplan.md plus the narrowest matching .tmp/slate-v2 source/test owner selected from the remaining Lexical/ProseMirror Mobile/IME backlog.
• Target: choose the next missing Mobile/IME proof route after the covered ProseMirror composition rows, ProseMirror existing-word replacement row, ProseMirror cursor-wrapper active-mark row, Lexical iOS Korean native-backspace contract, Lexical typeahead and floating-toolbar overlay rows, Lexical Korean multi-format composition row, Lexical IME undo-selection and immediate text-plus-IME history rows, Lexical type/delete/cancel row, and Lexical Safari delete-after-composition rows, preserving raw-device boundaries.

Slice 37 result:

• Selected ProseMirror's handles composition in a multi-child mark with a cursor wrapper row from ../prosemirror/view/test/webtest-composition.ts.
• Decision: translate ProseMirror's nested mark context into Slate's flat mark model as explicit active italic+code marks at the cursor before a bold-italic sibling.

Slice 38 result:

• Added commits IME composition through an active mark before a formatted sibling to .tmp/slate-v2/playwright/integration/examples/richtext.test.ts.
• First attempt failed because the collapsed boundary did not implicitly carry italic in Slate's flat mark model; the browser inserted standalone code. The test was corrected to make the active italic+code mark set explicit before composition.
• The row proves Chromium IME composition inserts oow with active italic+code marks before the bold-italic three sibling, preserves the formatted sibling, advances selection after the committed text, and records an allowed compositionend trace.
• Runtime code changes: none.
• Issue claims changed: none.
• Reference docs: no change, because this is desktop Chromium IME proof and does not promote exact Mobile/IME issue closure or raw-device claims.
• Focused proof passed: bun playwright test playwright/integration/examples/richtext.test.ts --project=chromium --grep "commits IME composition through an active mark before a formatted sibling" --retries=0.
• Richtext Chromium IME regression proof passed: bun playwright test playwright/integration/examples/richtext.test.ts --project=chromium --grep "syncs browser text mutations inside bold markup|commits IME composition inside bold markup|commits IME composition through an active mark in an empty block|commits IME composition through an active mark before a formatted sibling|replaces multiple formatted text nodes with Korean IME composition" --retries=0.
• Lint/format focused gate passed: bun run lint:fix playwright/integration/examples/richtext.test.ts.
• TypeScript root gate passed: bun typecheck:root.
• ce-compound evaluation: skipped; this is a proof-only port of a ProseMirror cursor-wrapper row and did not add a distinct reusable runtime fix pattern.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 39 is still runnable.

Next slice:

• slice-39-mobile-ime-next-proof-route-selection.
• Candidate owner: docs/plans/2026-05-07-slate-v2-mobile-ime-input-runtime-ralplan.md plus the narrowest matching .tmp/slate-v2 source/test owner selected from the remaining Lexical/ProseMirror Mobile/IME backlog.
• Target: choose the next missing Mobile/IME proof route after the covered ProseMirror composition rows, ProseMirror existing-word replacement row, ProseMirror cursor-wrapper active-mark rows, Lexical iOS Korean native-backspace contract, Lexical typeahead and floating-toolbar overlay rows, Lexical Korean multi-format composition row, Lexical IME undo-selection and immediate text-plus-IME history rows, Lexical type/delete/cancel row, and Lexical Safari delete-after-composition rows, preserving raw-device boundaries.

Slice 39 result:

• Selected ProseMirror's supports composition at start of block in a new node row from ../prosemirror/view/test/webtest-composition.ts.
• Decision: translate it with direct DOM composition in Slate's rendering-strategy runtime example so the browser inserts a composed text node before existing text.

Slice 40 result:

• Added commits IME composition at the start of a text block to .tmp/slate-v2/playwright/integration/examples/rendering-strategy-runtime.test.ts.
• The row proves direct DOM composition at offset 0 inserts !? before foo, preserves the following text, advances both model and DOM caret after the committed prefix, and records an allowed compositionend trace.
• Runtime code changes: none.
• Issue claims changed: none.
• Reference docs: no change, because this is desktop Chromium DOM composition proof and does not promote exact Mobile/IME issue closure or raw-device claims.
• Focused proof passed: bun playwright test playwright/integration/examples/rendering-strategy-runtime.test.ts --project=chromium --grep "commits IME composition at the start of a text block" --retries=0.
• Rendering-strategy IME regression proof passed: bun playwright test playwright/integration/examples/rendering-strategy-runtime.test.ts --project=chromium --grep "commits IME composition through|records runtime metadata for committed IME composition|undoes committed IME composition as one history step|undoes native text and immediately following IME composition together|restores expanded selection after undoing IME replacement|replaces an existing word with IME composition|commits IME composition at the start of a text block|does not push canceled IME composition onto history|keeps text stable after type-delete-cancel IME composition|drops active IME composition when a model change overlaps it|drops active IME composition when a model change partially overlaps it|drops active IME composition when a model change happens at its insertion point|keeps active IME composition when a model change happens elsewhere|commits rapidly following IME compositions in separate text blocks|commits cross-paragraph IME composition as one replacement" --retries=0.
• Lint/format focused gate passed: bun run lint:fix playwright/integration/examples/richtext.test.ts playwright/integration/examples/rendering-strategy-runtime.test.ts.
• TypeScript root gate passed: bun typecheck:root.
• ce-compound evaluation: skipped; this is a proof-only port of a ProseMirror start-of-block row and did not add a distinct reusable runtime fix pattern.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 41 is still runnable.

Next slice:

• slice-41-mobile-ime-next-proof-route-selection.
• Candidate owner: docs/plans/2026-05-07-slate-v2-mobile-ime-input-runtime-ralplan.md plus the narrowest matching .tmp/slate-v2 source/test owner selected from the remaining Lexical/ProseMirror Mobile/IME backlog.
• Target: choose the next missing Mobile/IME proof route after the covered ProseMirror composition rows, ProseMirror existing-word replacement row, ProseMirror cursor-wrapper active-mark rows, ProseMirror start-of-block DOM composition row, Lexical iOS Korean native-backspace contract, Lexical typeahead and floating-toolbar overlay rows, Lexical Korean multi-format composition row, Lexical IME undo-selection and immediate text-plus-IME history rows, Lexical type/delete/cancel row, and Lexical Safari delete-after-composition rows, preserving raw-device boundaries.

Slice 41 result:

• Selected ProseMirror's supports composition inside existing text row from ../prosemirror/view/test/webtest-composition.ts.
• Decision: translate it with direct DOM composition at a mid-text collapsed selection in Slate's rendering-strategy runtime example.

Slice 42 result:

• Added commits IME composition inside existing text to .tmp/slate-v2/playwright/integration/examples/rendering-strategy-runtime.test.ts.
• The row proves direct DOM composition at offset 1 in foo inserts xyz, yields fxyzoo, advances both model and DOM caret after the committed composition, and records an allowed compositionend trace.
• Runtime code changes: none.
• Issue claims changed: none.
• Reference docs: no change, because this is desktop Chromium DOM composition proof and does not promote exact Mobile/IME issue closure or raw-device claims.
• Focused proof passed: bun playwright test playwright/integration/examples/rendering-strategy-runtime.test.ts --project=chromium --grep "commits IME composition inside existing text" --retries=0.
• Rendering-strategy IME regression proof passed: bun playwright test playwright/integration/examples/rendering-strategy-runtime.test.ts --project=chromium --grep "commits IME composition through|records runtime metadata for committed IME composition|undoes committed IME composition as one history step|undoes native text and immediately following IME composition together|restores expanded selection after undoing IME replacement|replaces an existing word with IME composition|commits IME composition at the start of a text block|commits IME composition inside existing text|does not push canceled IME composition onto history|keeps text stable after type-delete-cancel IME composition|drops active IME composition when a model change overlaps it|drops active IME composition when a model change partially overlaps it|drops active IME composition when a model change happens at its insertion point|keeps active IME composition when a model change happens elsewhere|commits rapidly following IME compositions in separate text blocks|commits cross-paragraph IME composition as one replacement" --retries=0.
• Lint/format focused gate passed: bun run lint:fix playwright/integration/examples/rendering-strategy-runtime.test.ts.
• TypeScript root gate passed: bun typecheck:root.
• ce-compound evaluation: skipped; this is a proof-only port of a ProseMirror inside-existing-text row and did not add a distinct reusable runtime fix pattern.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 43 is still runnable.

Next slice:

• slice-43-mobile-ime-next-proof-route-selection.
• Candidate owner: docs/plans/2026-05-07-slate-v2-mobile-ime-input-runtime-ralplan.md plus the narrowest matching .tmp/slate-v2 source/test owner selected from the remaining Lexical/ProseMirror Mobile/IME backlog.
• Target: choose the next missing Mobile/IME proof route after the covered ProseMirror composition rows, ProseMirror existing-word replacement row, ProseMirror cursor-wrapper active-mark rows, ProseMirror start-of-block and inside-existing-text DOM composition rows, Lexical iOS Korean native-backspace contract, Lexical typeahead and floating-toolbar overlay rows, Lexical Korean multi-format composition row, Lexical IME undo-selection and immediate text-plus-IME history rows, Lexical type/delete/cancel row, and Lexical Safari delete-after-composition rows, preserving raw-device boundaries.

Slice 43 result:

• Selected ProseMirror's supports composition at end of block in a new node row from ../prosemirror/view/test/webtest-composition.ts.
• Decision: translate it with direct DOM composition at the end of an existing text block in Slate's rendering-strategy runtime example, separate from the existing native-CDP end-of-block row.

Slice 44 result:

• Added commits IME composition at the end of a text block to .tmp/slate-v2/playwright/integration/examples/rendering-strategy-runtime.test.ts.
• The row proves direct DOM composition at the end of foo inserts !?, yields foo!?, advances both model and DOM caret after the committed text, and records an allowed compositionend trace.
• Runtime code changes: none.
• Issue claims changed: none.
• Reference docs: no change, because this is desktop Chromium DOM composition proof and does not promote exact Mobile/IME issue closure or raw-device claims.
• Focused proof passed: bun playwright test playwright/integration/examples/rendering-strategy-runtime.test.ts --project=chromium --grep "commits IME composition at the end of a text block" --retries=0.
• Rendering-strategy IME regression proof passed: bun playwright test playwright/integration/examples/rendering-strategy-runtime.test.ts --project=chromium --grep "commits IME composition through|records runtime metadata for committed IME composition|undoes committed IME composition as one history step|undoes native text and immediately following IME composition together|restores expanded selection after undoing IME replacement|replaces an existing word with IME composition|commits IME composition at the start of a text block|commits IME composition inside existing text|commits IME composition at the end of a text block|does not push canceled IME composition onto history|keeps text stable after type-delete-cancel IME composition|drops active IME composition when a model change overlaps it|drops active IME composition when a model change partially overlaps it|drops active IME composition when a model change happens at its insertion point|keeps active IME composition when a model change happens elsewhere|commits rapidly following IME compositions in separate text blocks|commits cross-paragraph IME composition as one replacement" --retries=0.
• Lint/format focused gate passed: bun run lint:fix playwright/integration/examples/rendering-strategy-runtime.test.ts.
• TypeScript root gate passed: bun typecheck:root.
• ce-compound evaluation: skipped; this is a proof-only port of a ProseMirror end-of-block row and did not add a distinct reusable runtime fix pattern.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 45 is still runnable.

Next slice:

• slice-45-mobile-ime-next-proof-route-selection.
• Candidate owner: docs/plans/2026-05-07-slate-v2-mobile-ime-input-runtime-ralplan.md plus the narrowest matching .tmp/slate-v2 source/test owner selected from the remaining Lexical/ProseMirror Mobile/IME backlog.
• Target: choose the next missing Mobile/IME proof route after the covered ProseMirror composition rows, ProseMirror existing-word replacement row, ProseMirror cursor-wrapper active-mark rows, ProseMirror start/end-of-block and inside-existing-text DOM composition rows, Lexical iOS Korean native-backspace contract, Lexical typeahead and floating-toolbar overlay rows, Lexical Korean multi-format composition row, Lexical IME undo-selection and immediate text-plus-IME history rows, Lexical type/delete/cancel row, and Lexical Safari delete-after-composition rows, preserving raw-device boundaries.

Slice 45 result:

• Selected ProseMirror's supports composition in an empty block row from ../prosemirror/view/test/webtest-composition.ts.
• Decision: translate it through Slate's custom-placeholder example so an empty placeholder-backed editor accepts native Chromium IME composition and hides the placeholder after commit.

Slice 46 result:

• Added commits IME composition from the custom placeholder empty state to .tmp/slate-v2/playwright/integration/examples/placeholder.test.ts.
• The row proves a collapsed empty placeholder selection commits abc, yields model text abc, advances selection to offset 3, hides the placeholder, and records an allowed compositionend trace.
• Runtime code changes: none.
• Issue claims changed: none.
• Reference docs: no change, because this is desktop Chromium IME proof and does not promote exact Mobile/IME issue closure or raw-device claims.
• Focused proof passed: bun playwright test playwright/integration/examples/placeholder.test.ts --project=chromium --grep "commits IME composition from the custom placeholder empty state" --retries=0.
• Lint/format focused gate passed: bun run lint:fix playwright/integration/examples/placeholder.test.ts.
• TypeScript root gate passed: bun typecheck:root.
• Accidental broad command note: bun playwright test playwright/integration/examples/placeholder.test.ts --project=chromium --retries=0 unexpectedly expanded to the full examples/stress suite; it failed unrelated checklist, annotation, table, and huge-document rows after 218 passing tests. This broad run is not used as slice evidence.
• ce-compound evaluation: skipped; this is a proof-only port of a ProseMirror empty-block row and did not add a distinct reusable runtime fix pattern.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 47 is still runnable.

Next slice:

• slice-47-mobile-ime-next-proof-route-selection.
• Candidate owner: docs/plans/2026-05-07-slate-v2-mobile-ime-input-runtime-ralplan.md plus the narrowest matching .tmp/slate-v2 source/test owner selected from the remaining Lexical/ProseMirror Mobile/IME backlog.
• Target: choose the next missing Mobile/IME proof route after the covered ProseMirror composition rows, ProseMirror existing-word replacement row, ProseMirror cursor-wrapper active-mark rows, ProseMirror start/end-of-block, inside-existing-text, and custom-placeholder empty-state DOM composition rows, Lexical iOS Korean native-backspace contract, Lexical typeahead and floating-toolbar overlay rows, Lexical Korean multi-format composition row, Lexical IME undo-selection and immediate text-plus-IME history rows, Lexical type/delete/cancel row, and Lexical Safari delete-after-composition rows, preserving raw-device boundaries.

Slice 47 result:

• Selected Lexical's Can undo composed Hirigana via IME after composition ends (#2479) row from ../lexical/packages/lexical-playground/__tests__/e2e/History.spec.mjs.
• Decision: translate the row into Slate's rendering-strategy runtime as delayed Hiragana native composition, delayed native space, delayed second composition, then undo each logical user step.

Slice 48 result:

• Added undoes delayed Hiragana IME compositions as separate history steps to .tmp/slate-v2/playwright/integration/examples/rendering-strategy-runtime.test.ts.
• First proof failed usefully: after すし もじあ, one undo returned to the initial document instead of stopping at すし . Root cause: browser-native text imports reached slate-history only as adjacent insert_text operations, so the generic adjacent-text merge heuristic had no user-pause boundary.
• Runtime fix added .tmp/slate-v2/packages/slate-react/src/editable/input-history.ts and applies delayed native-text history: { mode: 'push' } metadata from composition-state.ts and dom-repair-queue.ts. Immediate native text plus immediate IME composition still merges.
• Issue claims changed: none.
• Reference docs: no maintainer-facing claim file changed, because this is desktop Chromium IME/history proof and does not promote exact Mobile/IME issue closure or raw-device claims.
• Focused proof passed after the runtime fix and post-lint rerun: bun playwright test playwright/integration/examples/rendering-strategy-runtime.test.ts --project=chromium --grep "undoes delayed Hiragana IME compositions as separate history steps" --retries=0.
• Rendering-strategy IME regression proof passed: bun playwright test playwright/integration/examples/rendering-strategy-runtime.test.ts --project=chromium --grep "commits IME composition through|records runtime metadata for committed IME composition|undoes committed IME composition as one history step|undoes native text and immediately following IME composition together|undoes delayed Hiragana IME compositions as separate history steps|restores expanded selection after undoing IME replacement|replaces an existing word with IME composition|commits IME composition at the start of a text block|commits IME composition inside existing text|commits IME composition at the end of a text block|does not push canceled IME composition onto history|keeps text stable after type-delete-cancel IME composition|drops active IME composition when a model change overlaps it|drops active IME composition when a model change partially overlaps it|drops active IME composition when a model change happens at its insertion point|keeps active IME composition when a model change happens elsewhere|commits rapidly following IME compositions in separate text blocks|commits cross-paragraph IME composition as one replacement" --retries=0.
• Lint/format focused gate passed: bun run lint:fix packages/slate-react/src/editable/input-history.ts packages/slate-react/src/editable/composition-state.ts packages/slate-react/src/editable/dom-repair-queue.ts playwright/integration/examples/rendering-strategy-runtime.test.ts.
• TypeScript root gate passed: bun typecheck:root.
• ce-compound note added: docs/solutions/ui-bugs/2026-05-07-slate-react-native-ime-history-boundaries-need-explicit-push-metadata.md.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 49 is still runnable.

Next slice:

• slice-49-mobile-ime-next-proof-route-selection.
• Candidate owner: docs/plans/2026-05-07-slate-v2-mobile-ime-input-runtime-ralplan.md plus the narrowest matching .tmp/slate-v2 source/test owner selected from the remaining Lexical/ProseMirror Mobile/IME backlog.
• Target: choose the next missing Mobile/IME proof route after the covered ProseMirror composition rows, ProseMirror existing-word replacement row, ProseMirror cursor-wrapper active-mark rows, ProseMirror start/end-of-block, inside-existing-text, and custom-placeholder empty-state DOM composition rows, Lexical iOS Korean native-backspace contract, Lexical typeahead and floating-toolbar overlay rows, Lexical Korean multi-format composition row, Lexical IME undo-selection, immediate text-plus-IME, and delayed Hiragana IME history rows, Lexical type/delete/cancel row, and Lexical Safari delete-after-composition rows, preserving raw-device boundaries.

Slice 49 result:

• Selected Lexical's Can type Hiragana via IME between emojis row from ../lexical/packages/lexical-playground/__tests__/e2e/Composition.spec.mjs.
• Decision: translate the useful Slate-owned part as direct DOM composition at a UTF-16 surrogate-pair boundary, without importing Lexical's emoji autocomplete surface.

Slice 50 result:

• Added commits IME composition between emoji text to .tmp/slate-v2/playwright/integration/examples/rendering-strategy-runtime.test.ts.
• The row proves composing すし between two 🙂 characters yields 🙂すし🙂, preserves the second emoji, advances the selection after the committed text at the correct UTF-16 offset, and records an allowed compositionend trace.
• Runtime code changes: none.
• Issue claims changed: none.
• Reference docs: no change, because this is desktop Chromium DOM composition proof and does not promote exact Mobile/IME issue closure or raw-device claims.
• Focused proof passed: bun playwright test playwright/integration/examples/rendering-strategy-runtime.test.ts --project=chromium --grep "commits IME composition between emoji text" --retries=0.
• Rendering-strategy IME regression proof passed: bun playwright test playwright/integration/examples/rendering-strategy-runtime.test.ts --project=chromium --grep "commits IME composition through|records runtime metadata for committed IME composition|undoes committed IME composition as one history step|undoes native text and immediately following IME composition together|undoes delayed Hiragana IME compositions as separate history steps|restores expanded selection after undoing IME replacement|replaces an existing word with IME composition|commits IME composition at the start of a text block|commits IME composition inside existing text|commits IME composition at the end of a text block|commits IME composition between emoji text|does not push canceled IME composition onto history|keeps text stable after type-delete-cancel IME composition|drops active IME composition when a model change overlaps it|drops active IME composition when a model change partially overlaps it|drops active IME composition when a model change happens at its insertion point|keeps active IME composition when a model change happens elsewhere|commits rapidly following IME compositions in separate text blocks|commits cross-paragraph IME composition as one replacement" --retries=0.
• Lint/format focused gate passed: bun run lint:fix playwright/integration/examples/rendering-strategy-runtime.test.ts.
• TypeScript root gate passed: bun typecheck:root.
• ce-compound evaluation: skipped; this is a proof-only port of a Lexical emoji-boundary row and did not add a distinct reusable runtime fix pattern.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 51 is still runnable.

Next slice:

• slice-51-mobile-ime-next-proof-route-selection.
• Candidate owner: docs/plans/2026-05-07-slate-v2-mobile-ime-input-runtime-ralplan.md plus the narrowest matching .tmp/slate-v2 source/test owner selected from the remaining Lexical/ProseMirror Mobile/IME backlog.
• Target: choose the next missing Mobile/IME proof route after the covered ProseMirror composition rows, ProseMirror existing-word replacement row, ProseMirror cursor-wrapper active-mark rows, ProseMirror start/end-of-block, inside-existing-text, and custom-placeholder empty-state DOM composition rows, Lexical iOS Korean native-backspace contract, Lexical typeahead and floating-toolbar overlay rows, Lexical Korean multi-format composition row, Lexical IME undo-selection, immediate text-plus-IME, delayed Hiragana IME history, and emoji-boundary rows, Lexical type/delete/cancel row, and Lexical Safari delete-after-composition rows, preserving raw-device boundaries.

Slice 51 result:

• Selected Lexical's mention-boundary IME rows from ../lexical/packages/lexical-playground/__tests__/e2e/Composition.spec.mjs, focusing on the still-missing "at the end of a mention" shape.
• Decision: Slate already had before-mention, between-mentions, and typeahead-open proof; add a direct DOM composition row at the start of the text node immediately after an inline mention.

Slice 52 result:

• Added commits IME composition immediately after an inline mention to .tmp/slate-v2/playwright/integration/examples/mentions.test.ts.
• The row proves composing すし immediately after the first mention preserves both inline mentions, inserts into the following text node, advances selection after the committed text, and records an allowed compositionend trace.
• Runtime code changes: none.
• Issue claims changed: none.
• Reference docs: no change, because this is desktop Chromium DOM composition proof and does not promote exact Mobile/IME issue closure or raw-device claims.
• Focused proof passed: bun playwright test playwright/integration/examples/mentions.test.ts --project=chromium --grep "commits IME composition immediately after an inline mention" --retries=0.
• Mentions IME regression proof passed: bun playwright test playwright/integration/examples/mentions.test.ts --project=chromium --grep "keeps mention portal open during IME composition|commits staged IME composition before a markable inline mention|commits IME composition between inline mentions without overwriting them|commits IME composition immediately after an inline mention" --retries=0.
• Lint/format focused gate passed: bun run lint:fix playwright/integration/examples/mentions.test.ts.
• TypeScript root gate passed: bun typecheck:root.
• ce-compound evaluation: skipped; this is a proof-only port of a Lexical inline-mention boundary row and did not add a distinct reusable runtime fix pattern.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 53 is still runnable.

Next slice:

• slice-53-mobile-ime-next-proof-route-selection.
• Candidate owner: docs/plans/2026-05-07-slate-v2-mobile-ime-input-runtime-ralplan.md plus the narrowest matching .tmp/slate-v2 source/test owner selected from the remaining Lexical/ProseMirror Mobile/IME backlog.
• Target: choose the next missing Mobile/IME proof route after the covered ProseMirror composition rows, ProseMirror existing-word replacement row, ProseMirror cursor-wrapper active-mark rows, ProseMirror start/end-of-block, inside-existing-text, and custom-placeholder empty-state DOM composition rows, Lexical iOS Korean native-backspace contract, Lexical typeahead, floating-toolbar, and inline-mention boundary rows, Lexical Korean multi-format composition row, Lexical IME undo-selection, immediate text-plus-IME, delayed Hiragana IME history, and emoji-boundary rows, Lexical type/delete/cancel row, and Lexical Safari delete-after-composition rows, preserving raw-device boundaries.

Slice 53 result:

• Classified the remaining Lexical ../lexical/packages/lexical-playground/__tests__/e2e/Composition.spec.mjs rows after the new emoji and inline-mention proof.
• Accounting: base Hiragana typing is covered by slice 48's final text plus undo-boundary proof; new-bold-format is covered by active-mark cursor rows; between-emojis is covered by slice 50; type/delete/cancel is covered by slice 32; floating toolbar, typeahead, Korean multi-format, and inline mention boundaries are covered by earlier/new rows.
• Explicit deferrals: Lexical's soft line-break row is not directly portable while Slate v2 insertSoftBreak still aliases block split; hashtag rows are product/plugin-specific and need a Slate hashtag surface before they can become raw Slate proof. These remain backlog pressure, not exact claims.

Slice 54 result:

• Runtime code changes: none.
• Issue claims changed: none.
• Reference docs: no change, because this is proof-route accounting and does not promote exact Mobile/IME issue closure or raw-device claims.
• ce-compound evaluation: skipped; this is accounting against already read Lexical rows and did not add a new runtime fix pattern.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 55 is still runnable.

Next slice:

• slice-55-mobile-ime-next-proof-route-selection.
• Candidate owner: docs/plans/2026-05-07-slate-v2-mobile-ime-input-runtime-ralplan.md plus the narrowest matching .tmp/slate-v2 source/test owner selected from the remaining Lexical/ProseMirror Mobile/IME backlog.
• Target: choose the next missing Mobile/IME proof route after the translated ProseMirror composition rows, the covered or classified Lexical Composition.spec.mjs rows, Lexical iOS Korean native-backspace contract, Lexical history rows, Lexical type/delete/cancel row, and Lexical Safari delete-after-composition rows, preserving raw-device and product-plugin boundaries.

Slice 55 result:

• Classified the remaining Lexical ../lexical/packages/lexical-playground/__tests__/e2e/History.spec.mjs IME rows after the delayed Hiragana history fix.
• Accounting: Cancel composition not push undo stack is covered by does not push canceled IME composition onto history; Merge IME input when less delay is covered by undoes native text and immediately following IME composition together plus the delayed split row; RangeSelection should be retained when undo IME is covered by restores expanded selection after undoing IME replacement.
• Decision: no new code/test row was added because each remaining Lexical history row has an existing focused Slate proof with matching behavior. Re-running duplicate rows would add noise, not confidence.

Slice 56 result:

• Runtime code changes: none.
• Issue claims changed: none.
• Reference docs: no change, because this is proof-route accounting and does not promote exact Mobile/IME issue closure or raw-device claims.
• ce-compound evaluation: skipped; this is accounting against already read Lexical rows and did not add a new runtime fix pattern.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 57 is still runnable.

Slice 57 result:

• Classified Lexical's Safari regression source at ../lexical/packages/lexical-playground/__tests__/regression/8153-safari-ime-delete-selection.spec.mjs.
• Accounting: Can delete all text selected with Cmd+A after IME composition end on Safari is covered by Slate's WebKit deletes rich text selection after WebKit compositionend row and the shell-backed deletes shell-backed selection after WebKit compositionend row; Can delete multi-paragraph selection with Shift+ArrowUp after IME composition end on Safari is covered by deletes rich text line selection after WebKit compositionend.
• Decision: no new code/test row was added because the exact WebKit stale-compositionend selection/delete behaviors already have matching Slate proof rows. Re-running duplicate rows would be noise.

Slice 58 result:

• Runtime code changes: none.
• Issue claims changed: none.
• Reference docs: no change, because this is proof-route accounting and does not promote exact Mobile/IME issue closure or raw-device claims.
• ce-compound evaluation: skipped; this is accounting against already covered Safari/WebKit rows and did not add a new runtime fix pattern.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 59 is still runnable.

Slice 59 result:

• Selected issue #5891, [Android] The autocorrect doesn't work when creating the first line, from docs/slate-issues/open-issues-dossiers/5912-5771.md and docs/slate-issues/test-candidate-map/5912-5771.md.
• Decision: translate the proof route as empty-state insertReplacementText / autocorrect architecture coverage first, not exact raw Android keyboard closure.

Slice 60 result:

• Added routes Android-style first-line autocorrect through empty-state replacement text to .tmp/slate-v2/packages/slate-react/test/model-input-strategy-contract.test.ts.
• The row proves insertReplacementText at an empty first-line selection inserts the replacement text, advances model selection after the inserted text, and returns no repair request.
• Runtime code changes: none.
• Issue claims changed: none; this is architecture proof for the Android/autocorrect route, not an exact #5891 raw-device closure.
• Reference docs: no change, because no exact Mobile/IME issue claim, public API text, release gate, or PR narrative was promoted.
• Failed-command note: root bun test packages/slate-react/test/model-input-strategy-contract.test.ts... is wrong for this file because bunfig.toml ignores *.test.*; the correct focused runner is package Vitest.
• Focused proof passed: bun --filter slate-react test:vitest -- model-input-strategy-contract.test.ts -t "routes Android-style first-line autocorrect through empty-state replacement text".
• Lint/format focused gate passed: bun run lint:fix packages/slate-react/test/model-input-strategy-contract.test.ts.
• Package typecheck passed: bun --filter slate-react typecheck.
• ce-compound evaluation: defer until the lane closeout; the useful reusable finding is the Vitest-owned *.test.* runner boundary.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 61 is still runnable.

Slice 61 result:

• Selected issue #5099, Android empty placeholder backspace behavior, from docs/slate-issues/open-issues-dossiers/5129-5066.md and docs/slate-issues/test-candidate-map/5129-5066.md.
• Decision: translate the proof route as empty-state deleteContentBackward architecture coverage first, not exact raw Android Chrome keyboard closure.

Slice 62 result:

• Added keeps Android-style empty-state backspace from mutating placeholder text to .tmp/slate-v2/packages/slate-react/test/model-input-strategy-contract.test.ts.
• The row proves model-owned empty-state deleteContentBackward leaves editor text empty, keeps selection at [0,0]/0, and requests only caret repair.
• Runtime code changes: none.
• Issue claims changed: none; this is architecture proof for the Android/backspace route, not an exact #5099 raw-device closure.
• Reference docs: no change, because no exact Mobile/IME issue claim, public API text, release gate, or PR narrative was promoted.
• Focused proof passed: bun --filter slate-react test:vitest -- model-input-strategy-contract.test.ts -t "keeps Android-style empty-state backspace from mutating placeholder text".
• Focused regression proof passed: bun --filter slate-react test:vitest -- model-input-strategy-contract.test.ts -t "routes Android-style first-line autocorrect through empty-state replacement text|keeps Android-style empty-state backspace from mutating placeholder text".
• Lint/format focused gate passed: bun run lint:fix packages/slate-react/test/model-input-strategy-contract.test.ts.
• Package typecheck passed: bun --filter slate-react typecheck.
• ce-compound evaluation: defer until the lane closeout; this reuses the same Vitest-owned test-file runner boundary as slice 60.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 63 is still runnable.

Slice 63 result:

• Selected the Android/CJK duplicate-delete proof family from the strong first batch, especially issue #5984 from docs/slate-issues/open-issues-dossiers/5994-5918.md, issue #5083 from docs/slate-issues/open-issues-dossiers/5129-5066.md, and issue #5023 from docs/slate-issues/open-issues-dossiers/5064-4971.md.
• Decision: translate the proof route as model-owned expanded CJK composition insertion/deletion coverage first, not exact Sogou/Android device closure.

Slice 64 result:

• Added replaces expanded CJK composition selection once and deletes expanded CJK composition selection once to .tmp/slate-v2/packages/slate-react/test/model-input-strategy-contract.test.ts.
• The rows prove insertFromComposition over an expanded CJK selection inserts one replacement string without duplicate text, and deleteByComposition over an expanded CJK selection deletes one selected unit without requiring a second delete.
• Runtime code changes: none.
• Issue claims changed: none; this is architecture proof for the Android/CJK route, not exact #5984, #5083, or #5023 device/browser closure.
• Reference docs: no change, because no exact Mobile/IME issue claim, public API text, release gate, or PR narrative was promoted.
• Focused proof passed: bun --filter slate-react test:vitest -- model-input-strategy-contract.test.ts -t "replaces expanded CJK composition selection once|deletes expanded CJK composition selection once".
• Focused regression proof passed: bun --filter slate-react test:vitest -- model-input-strategy-contract.test.ts -t "Android-style first-line autocorrect|Android-style empty-state backspace|expanded CJK composition selection once".
• Lint/format focused gate passed: bun run lint:fix packages/slate-react/test/model-input-strategy-contract.test.ts.
• Package typecheck passed: bun --filter slate-react typecheck.
• ce-compound evaluation: defer until the lane closeout; this adds no new reusable runtime-fix pattern beyond the existing issue-proof row pattern.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 65 is still runnable.

Slice 65 result:

• Selected the remaining Android/CJK onChange/readOnly route around #5078, #5034, and #5026.
• Finding: #5078 is not an inline-boundary CJK row; it is Android composition/onChange freshness. #5034 and #5026 are Android readOnly lifecycle rows. None are honestly closed by the model-input unit rows.

Slice 66 result:

• Decision: defer these to a React integration, Android manager, or raw-device/browser route that can observe onChange operations, readOnly transition behavior, and device/browser event timing.
• Runtime code changes: none.
• Issue claims changed: none.
• Reference docs: no change, because this is conservative proof-route accounting and does not promote exact Mobile/IME issue closure.
• ce-compound evaluation: skipped; this is classification against existing dossier/test-map rows and did not add a reusable runtime fix pattern.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 67 is still runnable.

Slice 67 result:

• Selected the autocomplete/suggestion replacement route and corrected the issue mapping from disk: #5653 is Microsoft IME blur duplication, while Android autocomplete prefix replacement is #5643; #4531 remains a broader Android suggestion-flicker row.
• Checked sibling coverage before adding proof. Lexical's closest stealable row is ../lexical/packages/lexical-playground/__tests__/e2e/Events.spec.mjs, which dispatches insertText followed by insertReplacementText over a target range. The local ProseMirror sibling check did not expose a matching autocomplete/replacement row.
• Decision: translate the proof route as model-owned replacement text over the intended prefix range, not exact Android suggestion UI closure and not #5653 focus/blur lifecycle closure.

Slice 68 result:

• Added replaces an autocorrect prefix without appending after it to .tmp/slate-v2/packages/slate-react/test/model-input-strategy-contract.test.ts.
• The row proves a native-style character insert followed by replacement text over the intended prefix range yields IS, not an appended duplicate, and leaves model selection after the replacement prefix.
• Runtime code changes: none.
• Issue claims changed: none; this is architecture proof for the #5643 / #4531 replacement-pressure route only.
• Reference docs: no change, because no exact Mobile/IME issue claim, public API text, release gate, or PR narrative was promoted.
• Focused proof passed: bun --filter slate-react test:vitest -- model-input-strategy-contract.test.ts -t "autocorrect prefix|first-line autocorrect|empty-state backspace|expanded CJK".
• Lint/format focused gate passed: bun run lint:fix packages/slate-react/test/model-input-strategy-contract.test.ts.
• Package typecheck passed: bun --filter slate-react typecheck.
• ce-compound evaluation: defer until the lane closeout; this reuses the same Vitest-owned test-file runner boundary and the already documented honest-proof boundary.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 69 is still runnable.

Slice 69 result:

• Selected #5653, Microsoft IME blur duplication, because its dossier names a concrete event sequence: compositionend, then beforeinput deleteContentBackward, then beforeinput insertText.
• Checked sibling coverage. The local Lexical and ProseMirror checkouts did not expose an exact Microsoft IME blur row. The useful current-source owner was Slate's own beforeinput selection reconciliation.
• Added red proof imports expanded delete target ranges from blur-time IME cleanup events to .tmp/slate-v2/packages/slate-react/test/selection-reconciler-contract.ts.
• Red result: the focused proof failed because syncSelectionForBeforeInput ignored expanded target ranges for deleteContentBackward and kept the old collapsed model selection.

Slice 70 result:

• Updated .tmp/slate-v2/packages/slate-react/src/editable/selection-reconciler.ts so expanded beforeinput delete target ranges are imported even for forward/backward delete cleanup events. Collapsed delete target ranges still fall back to command-derived behavior.
• Runtime code changes: yes, targeted selection reconciliation fix.
• Issue claims changed: no exact #5653 Windows/Microsoft IME closure was promoted. The fix covers the event-level runtime owner when the browser supplies an expanded delete target range during blur cleanup; exact Windows/IME closure still needs platform proof.
• Reference docs: no change, because no exact Mobile/IME issue claim, public API text, release gate, or PR narrative was promoted.
• Red proof failed before the fix: bun test ./packages/slate-react/test/selection-reconciler-contract.ts -t "imports expanded delete target ranges".
• Focused proof passed after the fix: bun test ./packages/slate-react/test/selection-reconciler-contract.ts -t "imports expanded delete target ranges".
• Focused regression proof passed: bun test ./packages/slate-react/test/selection-reconciler-contract.ts.
• Adjacent regression proof passed: bun --filter slate-react test:vitest -- model-input-strategy-contract.test.ts -t "autocorrect prefix|first-line autocorrect|empty-state backspace|expanded CJK".
• Lint/format focused gate passed: bun run lint:fix packages/slate-react/src/editable/selection-reconciler.ts packages/slate-react/test/selection-reconciler-contract.ts packages/slate-react/test/model-input-strategy-contract.test.ts.
• Package typecheck passed: bun --filter slate-react typecheck.
• ce-compound evaluation: defer until lane closeout; the reusable technical lesson is likely that beforeinput delete target ranges are not optional for IME/focus cleanup.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 71 is still runnable.

Slice 71 result:

• Selected the Android heading Enter / block-type route around #4521 and #5175.
• Found matching external shape in ProseMirror's ../prosemirror-commands/test/test-commands.ts splitBlock heading rows and Lexical's ../lexical/packages/lexical-playground/__tests__/e2e/Headings/HeadingsEnterAtEnd.spec.mjs / HeadingsEnterInMiddle.spec.mjs.
• Red attempt: the stolen paragraph-after-heading expectation failed against raw Slate. Decision: do not implement that expectation in raw Slate because it hardcodes product block policy. Keep paragraph-after-heading behavior in example/plugin/device proof.

Slice 72 result:

• Added splits a custom block on Enter without dropping follow-up text to .tmp/slate-v2/packages/slate-react/test/model-input-strategy-contract.test.ts.
• The row proves raw model-owned insertParagraph splits a custom block, moves selection into the inserted block, and preserves the first follow-up inserted character. It intentionally does not force heading-to-paragraph product semantics.
• Runtime code changes: none.
• Issue claims changed: none; this is architecture proof for the #4521 / #5175 raw runtime route, not exact Android heading DOM or last inserted element closure.
• Reference docs: no change, because no exact Mobile/IME issue claim, public API text, release gate, or PR narrative was promoted.
• Focused proof passed: bun --filter slate-react test:vitest -- model-input-strategy-contract.test.ts -t "splits a custom block".
• Focused regression proof passed: bun --filter slate-react test:vitest -- model-input-strategy-contract.test.ts -t "custom block|autocorrect prefix|first-line autocorrect|empty-state backspace|expanded CJK".
• Adjacent regression proof passed: bun test ./packages/slate-react/test/selection-reconciler-contract.ts.
• Lint/format focused gate passed: bun run lint:fix packages/slate-react/src/editable/selection-reconciler.ts packages/slate-react/test/selection-reconciler-contract.ts packages/slate-react/test/model-input-strategy-contract.test.ts.
• Package typecheck passed: bun --filter slate-react typecheck.
• ce-compound evaluation: defer until lane closeout; the useful reusable lesson is the raw Slate primitive/product block-policy boundary.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 73 is still runnable.

Slice 73 result:

• Selected the inline-boundary / multibyte IME route around #4693 and #4136.
• Issue source check: docs/slate-issues/open-issues-dossiers/4741-4643.md keeps #4693 as Android Korean/Japanese composition architecture pressure, while docs/slate-issues/open-issues-dossiers/4160-4074.md keeps #4136 as multibyte composition after clearing placeholder or near inline boundaries.
• Test-map check: docs/slate-issues/test-candidate-map/4741-4643.md asks for Android Korean composition in the richtext example, and docs/slate-issues/test-candidate-map/4160-4074.md asks for multibyte IME input after clearing placeholder or after an inline boundary.
• Sibling-source check found the relevant external rows: ProseMirror's webtest-composition.ts covers composition inside marks, multi-child marks, cursor wrappers, highlighted/decorated text, multi-node composition, and widgets next to composition; Lexical's Composition.spec.mjs covers Hiragana near mentions, IME after mentions, hashtags, typeahead stability, and Korean replacement across multiple formatted nodes.

Slice 74 result:

• Decision: do not add a new implementation row for #4693 / #4136 in this slice. The architecture route is already represented by existing Slate v2 browser rows: .tmp/slate-v2/playwright/integration/examples/richtext.test.ts covers active mark cursor-wrapper composition and Korean replacement across formatted nodes; .tmp/slate-v2/playwright/integration/examples/mentions.test.ts covers composition before, between, and after inline mentions; .tmp/slate-v2/playwright/integration/examples/highlighted-text.test.ts covers decorated-text and spanning-decorator composition; and .tmp/slate-v2/playwright/integration/examples/placeholder.test.ts covers custom-placeholder empty-state composition.
• Runtime code changes: none.
• Issue claims changed: none. This is architecture coverage only; exact Android Korean/Japanese, Sogou, or multibyte crash closure still needs raw-device or issue-specific browser proof.
• Reference docs: no change, because no exact Mobile/IME issue claim, public API text, release gate, or PR narrative was promoted.
• Focused verification: no code changed. Existing proof owners were re-read on disk instead of rerun.
• ce-compound evaluation: skipped; this slice only classified existing proof coverage and did not introduce a new reusable runtime fix pattern.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 75 is still runnable.

Slice 75 result:

• Selected the empty-editor IME route around #4030, #3943, and #3882.
• Issue source check: docs/slate-issues/open-issues-dossiers/3948-3881.md keeps #3943 as a Safari Chinese composition caret-reset row and #3882 as an empty richtext IME crash row. docs/slate-issues/open-issues-dossiers/4067-3949.md keeps #4030 as a Safari list-item IME caret-reset row.
• Test-map check: docs/slate-issues/test-candidate-map/3948-3881.md asks for Safari Chinese composition caret stability and empty richtext IME typing without a crash; docs/slate-issues/test-candidate-map/4067-3949.md asks for Safari list item IME composition to keep caret position.
• Sibling-source check found ProseMirror's empty-block composition row in ../prosemirror/view/test/webtest-composition.ts. The existing Slate v2 custom-placeholder row was close, but it did not prove the richtext example empty-block route named by #3882.
• WebKit proof boundary check: docs/solutions/logic-errors/2026-04-11-webkit-playwright-inserttext-is-a-direct-input-ceiling-not-a-direct-composition-lane.md still applies. Desktop WebKit Playwright insertText is useful direct-input evidence, not honest composition evidence, so #4030 and #3943 stay deferred to Safari composition proof.

Slice 76 result:

• Added commits IME composition in an empty rich text block to .tmp/slate-v2/playwright/integration/examples/richtext.test.ts.
• The row clears the richtext example, proves the placeholder-backed empty block accepts a native Chromium IME commit, asserts final model text すし, selection after the committed text, and an allowed compositionend trace.
• Runtime code changes: none.
• Issue claims changed: none; this is architecture proof for the #3882 empty-richtext route, not exact old Korean/Chinese crash closure. #4030 and #3943 remain Safari-specific deferred rows.
• Reference docs: no change, because no exact Mobile/IME issue claim, public API text, release gate, or PR narrative was promoted.
• Focused proof passed: bun playwright test playwright/integration/examples/richtext.test.ts --project=chromium --grep "commits IME composition in an empty rich text block" --retries=0.
• Focused regression proof passed: bun playwright test playwright/integration/examples/richtext.test.ts --project=chromium --grep "empty rich text block|active mark in an empty block|replaces multiple formatted text nodes with Korean" --retries=0.
• Lint/format focused gate passed: bun run lint:fix playwright/integration/examples/richtext.test.ts.
• Root TypeScript check passed: bun run typecheck:root.
• ce-compound evaluation: skipped; this reused the already documented honest browser IME proof boundary and did not introduce a new runtime fix pattern.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 77 is still runnable.

Slice 77 result:

• Selected the select-all / first-entry composition route around #4067, #4031, and #4353.
• Issue source check: docs/slate-issues/open-issues-dossiers/4067-3949.md keeps #4067 as select-all then composition crash/desync, and #4031 as empty-editor Japanese composition confirm crash. docs/slate-issues/open-issues-dossiers/4390-4269.md keeps #4353 as select-all then Japanese composition wedging after the first character.
• Test-map check: docs/slate-issues/test-candidate-map/4067-3949.md asks for select-all then composition and empty-editor Japanese confirm rows; docs/slate-issues/test-candidate-map/4390-4269.md asks for select-all then start Japanese IME input.
• Sibling-source check found no exact Lexical or ProseMirror select-all IME row in the local checkouts. The closest external mechanism remains ProseMirror's empty-block composition row plus Slate's own expanded-selection and replacement IME rows.

Slice 78 result:

• Added replaces select-all rich text with IME composition to .tmp/slate-v2/playwright/integration/examples/richtext.test.ts.
• The row selects all in the richtext example, starts native Chromium IME composition, commits すし, and asserts the full selection is replaced by one text node with selection after the committed text and an allowed compositionend trace.
• Runtime code changes: none.
• Issue claims changed: none; this is architecture proof for the #4067/#4353 select-all composition route and complements the #4031 empty-editor route, not exact old Japanese/Chinese crash closure.
• Reference docs: no change, because no exact Mobile/IME issue claim, public API text, release gate, or PR narrative was promoted.
• Focused proof passed: bun playwright test playwright/integration/examples/richtext.test.ts --project=chromium --grep "replaces select-all rich text with IME composition" --retries=0.
• Focused regression proof passed: bun playwright test playwright/integration/examples/richtext.test.ts --project=chromium --grep "empty rich text block|select-all rich text|active mark in an empty block|replaces multiple formatted text nodes with Korean" --retries=0.
• Lint/format focused gate passed: bun run lint:fix playwright/integration/examples/richtext.test.ts.
• Root TypeScript check passed: bun run typecheck:root.
• ce-compound evaluation: skipped; this is another browser proof row using the existing IME harness boundary and did not introduce a reusable runtime fix pattern.
• Completion gate result: bun run completion-check reports status: pending, which is expected because slice 79 is still runnable.

Slice 79 result:

• Selected the composition focus-lifecycle route around #5830 and #4232.
• Issue source check: docs/slate-issues/open-issues-dossiers/5912-5771.md keeps #5830 as onBlur not firing while Japanese/Korean composition is active in an empty editor. docs/slate-issues/open-issues-dossiers/4268-4162.md keeps #4232 as select-all/delete plus composition corrupting text or pasting page text on blur.
• Test-map check: docs/slate-issues/test-candidate-map/5912-5771.md asks for composition onBlur on an empty editor when focus moves away mid-composition. docs/slate-issues/test-candidate-map/4268-4162.md asks for a broader Cmd+A/delete/composition/blur corruption row.
• Sibling-source check found ProseMirror keeps focus/blur lifecycle independent from composition state in ../prosemirror/view/src/input.ts. Lexical registers blur as a root event in ../lexical/packages/lexical/src/LexicalEvents.ts; the only local Lexical blur special case found was plugin-specific focus restoration in LexicalDraggableBlockPlugin, not a composition-blur proof row.

Slice 80 result:

• Added fires blur when focus leaves during placeholder IME composition to .tmp/slate-v2/playwright/integration/examples/placeholder.test.ts.
• Added a non-visual public Editable.onBlur counter to .tmp/slate-v2/site/examples/ts/custom-placeholder.tsx so the browser row proves the user callback route, not only a native DOM blur listener.
• The row starts Chromium CDP IME composition from the empty custom-placeholder editor, moves focus to an outside button mid-composition, asserts the outside target is focused, asserts the public onBlur callback fired exactly once, and asserts the Slate kernel recorded an allowed blur event.
• Runtime code changes: none.
• Issue claims changed: #5830 is upgraded from "no exact composition-blur proof" to related architecture proof for Chromium empty-placeholder composition blur. No exact Japanese/Korean raw IME closure is claimed. #4232 remains only partially represented by adjacent select-all and blur lifecycle rows; exact reverse-input/page-text-paste closure is not claimed.
• Reference docs: updated docs/slate-v2/ledgers/issue-coverage-matrix.md for the #5830 related row. PR fixed/improved counts stay unchanged because this is not a Fixes #... or Improves #... claim.
• Focused proof passed: bun playwright test playwright/integration/examples/placeholder.test.ts --project=chromium --grep "fires blur when focus leaves during placeholder IME composition" --retries=0.
• Focused regression proof passed: bun playwright test playwright/integration/examples/placeholder.test.ts --project=chromium --grep "custom placeholder empty state|focus leaves during placeholder IME|undoes typing from the custom placeholder" --retries=0.
• Lint/format focused gate passed: bun run lint:fix site/examples/ts/custom-placeholder.tsx playwright/integration/examples/placeholder.test.ts.
• Root TypeScript check passed: bun run typecheck:root.
• ce-compound evaluation: skipped; this is a proof row and conservative ledger sync, not a new reusable runtime fix pattern.

Slice 81 result:

• Classified the remaining Mobile/IME rows after the translated Lexical/ProseMirror rows and the new Slate v2 proof slices.
• No new desktop/browser proof row was selected. The remaining exact-closure rows need one of four owners: raw-device/manual proof, Safari/WebKit-specific composition/autocorrect proof, Android readOnly/onChange integration proof, or product/plugin policy.
• Raw-device/manual exact-closure bucket: #6051, #6022, #5983, #5989, #5883, #4400, #5680, #5493, #5130, #5836, #5805, #5666, #5291, #5371, #5167, #4959, #4861, #4602, #4354, #4372, #5183, #5391. Existing Slate v2 architecture covers the shared runtime pressure; exact closure still needs the reported Android/iOS/Samsung/Firefox Android/voice environment.
• Safari/WebKit/browser-specific exact-closure bucket: #4030, #3943, #4640, #4543, #4085. The local WebKit direct-input rows and Chromium IME rows do not honestly prove these Safari composition or autocorrect reports.
• Integration/readOnly/onChange bucket: #5078, #5034, #5026, #4994. The earlier slice preserved these as Android manager/readOnly integration rows because exact proof must observe emitted operations, onChange freshness, and readOnly transitions under the relevant Android path.
• Product/plugin/policy bucket: #5928, #4532, #4223, #3573, #4621, and the plugin/render composition rows such as #3222/#3177. These inform runtime/event policy but are not raw Slate exact bug closures.
• Already represented architecture rows without exact-issue promotion: #3873 is covered by the existing keyboard-composition kernel contract; #5173, #4232, and #4269 are represented by the focus-lifecycle, select-all, composition-cancel, and composition-keydown proof families, but their exact historical repros remain non-claims; #3695, #5375, #4770, and #3611 are represented by active-mark, inline-boundary, emoji-boundary, and void/zero-width architecture proof families, not exact old browser/device closures.
• Reference docs: no additional fixed/improved claim sync required. The #5830 related row was already synced in slice 80, and PR fixed/improved counts remain unchanged.
• Verification: no code changed in this slice. The relevant same-turn proof for the only new code/test surface remains the slice 80 focused Playwright row, focused placeholder regression run, lint:fix, and root TypeScript check.
• ce-compound evaluation: skipped. The reusable rules used here already exist in the plan and solution docs: synthetic/mobile proof labels stay explicit, raw-device claims need matching artifacts, and product/plugin rows do not become raw Slate closures.

Closure:

• Status: done for the requested Mobile/IME slate-ralplan execution.
• Exact fixed/improved Mobile/IME issue claims from this macro plan: none.
• The lane added or confirmed architecture proof rows, synced #5830 related proof text, and explicitly preserved raw-device, Safari/WebKit, integration, and product/plugin non-claims.
• Remaining exact closures are future lanes, not autonomous desktop-proof work left in this run.