Slate v2 Lexical Harvest Test Processing Ralplan

status: done score: 0.92 date: 2026-05-09 skill: slate-ralplan input harvest: docs/editor-test-harvester/lexical/report.md target repo: /Users/zbeyens/git/slate-v2

Current Verdict

The Lexical harvest is done, but the broader Lexical test processing is not.

The harvest report closed at 0.93 with 271 inventory rows, 137 portable or portable-mixed runnable files, and 1996 indexed test names (docs/editor-test-harvester/lexical/report.md:3, docs/editor-test-harvester/lexical/inventory.md:15, docs/editor-test-harvester/lexical/test-index.md:7). It also says the selected implementation lane only covered IME/history, clipboard corpus, browser transport, and table containment; everything else is backlog until a new apply pass names it (docs/editor-test-harvester/lexical/report.md:18).

So the correct next move is not "copy all Lexical tests." That would be noisy and wrong. The correct move is to process every Lexical source row into a Slate owner ledger, then let ralph implement the rows that survive as portable Slate behavior.

The first processing ledger now exists at docs/editor-test-harvester/lexical/slate-processing-ledger.md. It accounts for every portable and portable-mixed source file at file granularity. Related issue discovery, issue-ledger sync, live-source refresh, pressure pass, maintainer objection pass, high-risk pass, revision pass, final issue-accounting review, and closure scoring are now complete. This ralplan is ready for ralph execution.

Ralph execution update on 2026-05-09: the first package-only slice is complete. Lexical #7163 now has explicit Slate text-unit distances plus real forward and backward unit: "character" deletion proof in .tmp/slate-v2/packages/slate/test/text-units-contract.ts. The implementation also removed the stale complex-script reverse-delete restoration in .tmp/slate-v2/packages/slate/src/transforms-text/delete-text.ts, so deletion now uses the same text-unit boundary law as getCharacterDistance. The Thai unit-character fixtures were updated to match that law. No browser, raw mobile, table, collaboration, or issue claim changed.

Ralph execution update on 2026-05-09: clipboard HTML source-drill is complete. The row-level decisions are recorded in docs/editor-test-harvester/lexical/slate-processing-ledger.md under Clipboard HTML Row-Level Source Drill. The accepted next code slice is text-format HTML only: Google Docs-style BIU spans plus safe semantic <b> handling if the first red row proves it. Links, lists, and table rows are mapped but stay out of the next slice. No .tmp/slate-v2 code, browser, raw mobile, table-model, collaboration, or issue claim changed during source-drill.

Ralph execution update on 2026-05-09: paste-html text-format residual is complete. .tmp/slate-v2/playwright/integration/examples/paste-html.test.ts now proves Google Docs BIU spans and standard semantic <b> paste. The parser fix in .tmp/slate-v2/site/examples/ts/paste-html-import.ts keeps Google Docs <b style="font-weight:normal"> wrappers non-bold, maps normal semantic <b> to bold, and converts body-level inline strings into text nodes before wrapping them in paragraphs. Focused Playwright proof and .tmp/slate-v2 bun check passed. No link, list, table, raw mobile, collaboration, table-model, or issue claim changed.

Ralph execution update on 2026-05-09: paste-html simple anchor import residual is complete. .tmp/slate-v2/playwright/integration/examples/paste-html.test.ts now proves <a href="https://facebook.com">Facebook!</a> imports as one safe link in the paste-html example. Focused Playwright proof and .tmp/slate-v2 bun check passed. No parser change, list, table, paste-into-existing-link, raw mobile, collaboration, table-model, or issue claim changed.

Ralph execution update on 2026-05-09: paste-html noisy link-in-list residual is complete. .tmp/slate-v2/playwright/integration/examples/paste-html.test.ts now proves a minimized Lexical #3193-style fixture keeps the leading block, two list items, and a safe link inside the first list item. Focused Playwright proof and .tmp/slate-v2 bun check passed. No parser change, generic list claim, table, paste-into-existing-link, raw mobile, collaboration, table-model, or issue claim changed.

Ralph execution update on 2026-05-09: paste-html basic unordered-list import residual is complete. .tmp/slate-v2/playwright/integration/examples/paste-html.test.ts now proves <ul><li>Hello</li><li>world!</li></ul> imports as one list with two items. Focused Playwright proof and .tmp/slate-v2 bun check passed. No parser change, nested-list claim, table, paste-into-existing-link, raw mobile, collaboration, table-model, or issue claim changed.

Ralph execution update on 2026-05-09: paste-html compact nested-list import residual is complete. .tmp/slate-v2/playwright/integration/examples/paste-html.test.ts now proves the three Lexical nested-list variants keep visible text and nested list structure. The test resets the example between variants so proof does not accidentally stack lists inside the previous paste. Focused Playwright proof and .tmp/slate-v2 bun check passed. No parser change, nested-div list claim, table, paste-into-existing-link, raw mobile, collaboration, table-model, or issue claim changed.

Ralph execution update on 2026-05-09: paste-html nested div list-boundary residual is complete. .tmp/slate-v2/playwright/integration/examples/paste-html.test.ts now proves <ol><li>1<div>2</div>3</li><li>A<div>B</div>C</li></ol> keeps one ordered list, two list items, and Slate paragraph boundaries for the nested block content. .tmp/slate-v2/site/examples/ts/paste-html-import.ts preserves sibling-bounded DIV children inside LI as paragraph children. Focused Playwright proof and .tmp/slate-v2 bun check passed. No table, paste-into-existing-link, raw mobile, collaboration, table-model, or issue claim changed.

Ralph execution update on 2026-05-09: paste-html Google Docs table residual is complete. .tmp/slate-v2/playwright/integration/examples/paste-html.test.ts now proves Google Docs-style table HTML imports one 2x3 table, keeps the multi-paragraph second cell, and preserves 11pt text styling. Existing .tmp/slate-v2/site/examples/ts/paste-html-import.ts already handled the parser shape, so no importer code changed. Focused Playwright proof and .tmp/slate-v2 bun check passed. No column-width, table-layout, table-model, paste-into-existing-link, raw mobile, collaboration, or issue claim changed.

Ralph execution update on 2026-05-09: paste-html Quip table residual is complete. .tmp/slate-v2/playwright/integration/examples/paste-html.test.ts now proves Quip-style 2x3 table HTML imports rows/cells and preserves b b as newline text content inside the second cell. Existing .tmp/slate-v2/site/examples/ts/paste-html-import.ts already handled the parser shape, so no importer code changed. Focused Playwright proof and .tmp/slate-v2 bun check passed. No column-width, table-layout, table-model, paste-into-existing-link, raw mobile, collaboration, or issue claim changed.

Ralph execution update on 2026-05-09: DOM mutation repair residual is complete for the accepted Slate owners. Source-read Lexical Mutations.spec.mjs: its portable rows are native text-node mutation with selection clamp, and structural DOM rollback. Current richtext.test.ts already covers browser text mutation, selection, and repair traces. .tmp/slate-v2/packages/slate-react/test/dom-repair-policy-contract.ts now adds the missing structural restore-manager policy proof: child-list mutations are rolled back, while character-data mutations stay available for text sync. Focused package proof passed; .tmp/slate-v2 bun check passed. No desktop structural repair, raw Android device, raw mobile, collaboration, table-model, or issue claim changed.

Ralph execution update on 2026-05-09: focus/autoscroll residual is complete for accepted raw Slate behavior. Source-read Lexical Focus.spec.mjs and AutoScroll.spec.mjs. .tmp/slate-v2/playwright/integration/examples/richtext.test.ts now proves Slate keeps model selection when focus moves outside the editor and keeps the caret visible while typing through line breaks in both a scrollable editor root and scrollable parent. The first autoscroll proof failed because Chrome returned a zero rect for an empty-line caret and defaultScrollSelectionIntoView no-oped; .tmp/slate-v2/packages/slate-react/src/components/editable.tsx now falls back to scrolling the leaf rect. Focused Playwright proof passed. No Lexical playground tab-out, raw mobile, collaboration, table-model, or issue claim changed.

Ralph execution update on 2026-05-09: inline paste/mention/link/HR boundary source-read is complete. .tmp/slate-v2/playwright/integration/examples/inlines.test.ts now proves plain-text and rich-HTML paste at inline-link boundaries stays outside the link, and .tmp/slate-v2/packages/slate/test/clipboard-contract.ts now proves partial link fragment copy/paste preserves link attributes. Lexical Links.spec.mjs, Mentions.spec.mjs, HorizontalRule.spec.mjs, and deferred LinksHTMLCopyAndPaste.spec.mjs rows were split by owner: accepted raw link boundary rows are applied, Slate markable-inline-void mention behavior is already covered by existing mention/clipboard owners, editable Lexical mention text and toolbar/autolink/link-image behavior are rejected as product policy, paste-inside-link splitting is deferred to an explicit link-plugin policy, and HR rows are deferred because this checkout has generic block-void proof but no HR owner. Focused package proof, focused Playwright proof, bun run lint:fix, and .tmp/slate-v2 bun check passed. No raw mobile, collaboration, table-model, or issue claim changed.

Ralph execution update on 2026-05-09: delete-character/selection-boundary source-read is complete for the selected regression files. Lexical #1258 and #1730 are already covered by Slate's unit-character delete fixtures. Lexical #6974 is already covered by the existing inline-void reverse point-delete fixture. Lexical #7246 is now locked by .tmp/slate-v2/packages/slate/test/delete-contract.ts, proving Backspace at a following paragraph start merges into the previous list item. Lexical #7319 is split: HR styling and Lexical NodeSelection classes stay rejected, while the portable block-void behavior exposed a real gap. .tmp/slate-v2/packages/slate/src/transforms-text/delete-text.ts now treats a collapsed delete target inside an adjacent void/read-only element as a path delete and keeps the fallback caret point through operation transforming, so adjacent block voids delete one at a time. The regression lock lives in .tmp/slate-v2/packages/slate/test/transforms/delete/voids-false/block-after-multiple-reverse.tsx. Focused delete contract, text-unit contract, full delete fixture sweep, lint, and .tmp/slate-v2 bun check passed. The reusable rule is captured in docs/solutions/logic-errors/2026-05-09-slate-v2-adjacent-block-void-delete-must-path-delete-one-target.md. No browser, raw mobile, collaboration, table-model, HR owner, or issue claim changed.

Ralph execution update on 2026-05-09: Navigation/Selection/LexicalSelection source-read is complete for the selected files. Lexical Navigation.spec.mjs maps mostly to existing Slate query contracts: offset, character, word, block, inline-fragmentation, atomic void, non-selectable, and top-level boundary traversal are already covered in .tmp/slate-v2/packages/slate/test/query-contract.ts. Lexical Selection.spec.mjs is split: existing Slate browser/package owners already cover no focus-on-load, triple-click block selection, table containment, inline-boundary insertion, blurred-editor selection preservation, and package line/unit delete behavior; whole-table range expansion, RTL decorator movement, collapsible/date-time product nodes, and mark/style persistence stay deferred to their own owners. The useful package gap was the selection-rebase invariant from Lexical's selection helper/unit tests: when a selected leaf disappears, the selection must land on the nearest surviving inline point. That proof now lives in .tmp/slate-v2/packages/slate/test/selection-rebase-contract.ts. Focused package proof, bun run lint:fix, and .tmp/slate-v2 bun check passed. No browser, raw mobile, collaboration, table-model, product-decorator, or issue claim changed.

Ralph execution update on 2026-05-09: LexicalCaret source-read is complete. Lexical LexicalCaret.test.ts splits into five groups: caret object construction/splice/remove internals, selection-to-caret-range conversion, destructive range deletion, point ordering/common ancestor helpers, and split-at-caret helpers. Slate already covers traversal/order behavior through .tmp/slate-v2/packages/slate/test/query-contract.ts, raw split operation rows in .tmp/slate-v2/packages/slate/test/operations-contract.ts, transform split rows in .tmp/slate-v2/packages/slate/test/transforms-contract.ts, and text-unit movement in .tmp/slate-v2/packages/slate/test/text-units-contract.ts. The useful portable gap was the destructive caret-range edge matrix, so .tmp/slate-v2/packages/slate/test/delete-contract.ts now proves expanded range deletion trims both sibling-leaf edges and cross-block edges before collapsing at the anchor. Focused package proof, bun run lint:fix, and .tmp/slate-v2 bun check passed. The reusable harvesting rule is captured in docs/solutions/best-practices/2026-05-09-lexical-caret-harvest-rows-need-range-edge-contracts.md.

Ralph execution update on 2026-05-09: docs-traversals source-read is complete. Lexical docs-traversals.test.ts splits into sibling iteration, root child iteration, depth-first caret order, and "wholly included" range traversal. Lexical caret enter/exit helpers are not Slate API, but the portable behavior belongs in Slate's query contract. .tmp/slate-v2/packages/slate/test/query-contract.ts now proves sibling order, reverse child order, descendant depth-first paths, lowest leaf range traversal, and lowest element range traversal for the same nested inline document shape. Focused package proof, bun run lint:fix, and .tmp/slate-v2 bun check passed. No browser, raw mobile, collaboration, table-model, or issue claim changed. Lexical caret class internals, node-key mechanics, decorator payloads, token/segmented modes, browser, raw mobile, collaboration, table-model, and issue claims stayed out. The reusable harvesting rule is captured in docs/solutions/best-practices/2026-05-09-lexical-docs-traversal-harvest-rows-need-query-shape-contracts.md.

Ralph execution update on 2026-05-09: LexicalElementNode source-read is complete. The file splits into Lexical JSON schema, child/text query helpers, splice/selection behavior, transform scheduling, DOM slot behavior, and DOM indexPath helpers. The portable gap was query/text traversal, not Lexical node lifecycle. .tmp/slate-v2/packages/slate/test/query-contract.ts now proves nested text-leaf traversal with Node.texts, first/last leaf lookup, and Node.string text content for the same nested element family. Splice/selection behavior routes to existing operation and selection owners; DOM slot, indexPath, Lexical JSON schema, node-key, lifecycle, browser, raw mobile, collaboration, table-model, and issue claims stayed out. Focused package proof, bun run lint:fix, and .tmp/slate-v2 bun check passed. The reusable harvesting rule is captured in docs/solutions/best-practices/2026-05-09-lexical-element-node-harvest-rows-need-api-shape-splitting.md.

Ralph execution update on 2026-05-09: LexicalLineBreakNode source-read is complete. The file only has one portable behavior row: a line break contributes newline text and renders as a break. Slate has no equivalent LineBreakNode class, so Lexical type/schema/update/type-guard rows stay out. The accepted behavior is already covered by the Quip paste-html browser row preserving b b as newline text, ZeroWidthString line-break placeholder proof, structural insertBreak contracts, and Android newline-to-soft-break input handling. No .tmp/slate-v2 code or tests changed, and no browser, raw mobile, collaboration, table-model, or issue claim changed.

Ralph execution update on 2026-05-09: LexicalParagraphNode source-read is complete. The file splits into Lexical node class/schema/DOM helper rows, paragraph split/insert behavior, and paragraph DOM import alignment. Slate's existing insert-break, split/merge, and normalization owners already cover the accepted split/merge behavior, so the useful gap was paste-html import: .tmp/slate-v2/playwright/integration/examples/paste-html.test.ts now proves legacy align, CSS text-align precedence, and invalid align rejection. .tmp/slate-v2/site/examples/ts/paste-html-import.ts imports valid paragraph alignment and .tmp/slate-v2/site/examples/ts/paste-html.tsx renders the existing align field. The red browser proof first failed on missing alignment, then passed after the importer/rendering change. Focused Playwright proof, .tmp/slate-v2 bun run lint:fix, and .tmp/slate-v2 bun check passed. No raw mobile, collaboration, table-model, or issue claim changed. The reusable harvesting rule is captured in docs/solutions/best-practices/2026-05-09-lexical-paragraph-node-harvest-rows-need-dom-import-splitting.md.

Ralph execution update on 2026-05-09: LexicalRootNode source-read is complete. The file splits into Lexical root node API/schema/DOM helper rows, cached root text implementation rows, root normalization/string behavior, and root-child selection repair. Slate has no root selection point and Editor.string is explicitly block-separator-free, so Lexical cached-text/decorator/root API rows stay out. Existing root normalization/top-level cleanup and string owners cover the Slate root behavior. The useful gap was selected top-level child removal: .tmp/slate-v2/packages/slate/test/selection-rebase-contract.ts now proves removing the selected first block rebases selection to the next surviving block, and removing the selected only block clears selection. Focused package proof passed, followed by .tmp/slate-v2 bun run lint:fix, .tmp/slate-v2 bun check, and plate-2 pnpm lint:fix. No browser, raw mobile, collaboration, table-model, or issue claim changed.

Ralph execution update on 2026-05-09: LexicalTextNode source-read is complete. The file splits into Lexical text node schema/mode/detail/style internals, format helper flags, selection helpers, split/merge behavior, DOM create/update/export rendering policy, and composition-key/node-state internals. Slate already has owners for text-unit deletion, compatible text normalization, mark add/remove/toggle, and rich HTML mark import. The useful gap was raw text operation selection repair: .tmp/slate-v2/packages/slate/test/operations-contract.ts now proves split_node rebases an expanded selection across split text branches, and merge_node rebases selection into the surviving text branch. Focused package proof passed, followed by .tmp/slate-v2 bun run lint:fix, .tmp/slate-v2 bun check, and plate-2 pnpm lint:fix. Lexical JSON schema, node-key, mode/detail/style, DOM rendering, node-state, and composition-key rows stay out. No browser, raw mobile, collaboration, table-model, or issue claim changed.

Intent And Boundary

Intent: turn the completed Lexical harvest into an execution-grade Slate v2 test-processing plan that accounts for every harvested source row.

Desired outcome:

• every portable and portable-mixed Lexical source file has a Slate processing status;
• every copied/refactored/created row points at a current Slate v2 owner;
• every rejected/deferred row has a concrete reason;
• selected implementation slices are small enough for repeated ralph runs;
• issue claims stay honest.

In scope:

• Lexical harvest artifacts under docs/editor-test-harvester/lexical/;
• current Slate v2 package, browser, and Playwright proof owners;
• portable editor behavior: IME, clipboard, input transport, DOM selection, table containment, code/list/markdown/html units, history, undo, focus, scroll, grapheme/delete semantics, inline atoms, and collaboration substrate;
• issue accounting when a planned row touches an existing Slate issue cluster.

Non-goals:

• porting Lexical node classes, command registry, Composer/plugin identity, private MIME format, playground product chrome, ESLint tooling, release fixtures, or framework shell behavior;
• pretending desktop Playwright mobile viewports prove raw mobile/IME device behavior;
• silently turning raw Slate into Lexical's whole-table selection model;
• creating 1996 one-off tests.

Decision boundaries:

• Group upstream tests by invariant family, not by upstream file shape.
• Prefer strengthening existing Slate tests when they already own the behavior.
• Create new tests only when the invariant has no honest Slate proof.
• Defer raw-device, whole-table selection, and yjs browser rows unless a later plan explicitly accepts those capabilities.

Decision Brief

Principles:

1. Preserve portable behavior, not Lexical architecture.
2. Keep Slate raw and unopinionated.
3. Proof has to match the behavior: browser rows for browser transport, unit rows for model invariants, raw device rows for raw mobile claims.
4. DRY wins over test-count vanity.
5. Issue claims require exact repro proof, not vibes.

Top drivers:

• The harvest has enough breadth to drown implementation if copied literally.
• Slate v2 already absorbed the highest-value rows in selected apply slices.
• Remaining rows span different owners and need separate proof gates.

Options:

Chosen shape: create a Lexical-to-Slate processing ledger, then execute by behavior family.

Consequences:

• This plan stays pending until the ledger exists and follow-up passes finish.
• Some Lexical rows will be rejected forever. That is healthy.
• Whole-table selection remains a table-model decision, not a hidden side effect of DOM-selection work.

Current Source Read

Harvest facts:

• docs/editor-test-harvester/lexical/report.md:3 says the harvest is done.
• docs/editor-test-harvester/lexical/report.md:4 scores it 0.93.
• docs/editor-test-harvester/lexical/inventory.md:15 records 271 total rows, 196 runnable rows, 124 portable rows, 13 portable-mixed rows, 33 product-shell rows, 12 harness rows, 89 skipped rows, and 0 uncertain rows.
• docs/editor-test-harvester/lexical/test-index.md:7 records 137 indexed portable/portable-mixed runnable files.
• docs/editor-test-harvester/lexical/test-index.md:8 records 1996 extracted test/describe/it names.
• docs/editor-test-harvester/lexical/report.md:159 starts the selected apply ledger; this is not global Lexical closure.

Current Slate v2 proof owners found in live source:

Live-source refresh added on 2026-05-09:

Issue ledger facts:

• docs/slate-v2/ledgers/issue-coverage-matrix.md:229 keeps #2558 Not claimed because whole-table/multi-cell selection needs a table selection model.
• docs/slate-v2/ledgers/fork-issue-dossier.md:6182 has the #2558 dossier section, also not claimed.

Source-Backed North Star

Slate v2 should use Lexical as a test corpus, not as an architecture donor.

The architecture target is a raw editor with small behavior contracts:

• model invariants in package tests;
• browser transport in Playwright examples/stress rows;
• native IME/paste/selection proof where the browser is the bug;
• issue claims only where the original user-facing bug is actually proven.

Ecosystem Strategy Synthesis

Lexical Inventory Processing Plan

Process the full inventory into docs/editor-test-harvester/lexical/slate-processing-ledger.md.

Statuses allowed in that ledger:

• already-applied: selected apply lane already added or strengthened Slate proof;
• covered: current Slate v2 already has a good proof row;
• refactor-existing: strengthen, rename, or split an existing Slate row;
• create-new: add a new Slate unit/browser row;
• defer: valid behavior, but needs raw device, yjs browser, or table-model infrastructure;
• reject: non-portable, product-shell, harness-only, or Lexical-internal row.

Family accounting from the harvest inventory:

Pressure Pass Results

The pressure pass keeps the plan conservative. It does not add issue claims and does not edit .tmp/slate-v2.

Performance-oracle pressure:

• The first text-unit slice is package-only and has no hot browser/render path; no benchmark is required before a focused unit ralph.
• Clipboard large-paste, autoscroll, huge-document, and many-inline rows need measured browser or benchmark proof before any performance claim.
• Browser Playwright rows that build the Slate v2 site must run sequentially; do not parallelize site-backed Playwright commands.

DX and migration pressure:

• No public API change is accepted. If implementation finds an API gap, stop and update this Ralplan first.
• Keep raw Slate behavior as the target. Plate and slate-yjs benefit from the substrate; they do not justify importing Lexical plugin architecture.

Regression pressure:

• Every accepted create-new or refactor-existing row needs a focused failing row first when feasible.
• Browser-only behavior needs browser proof. Unit passing is not enough for clipboard transport, focus/autoscroll, DOM mutation import, IME, raw mobile, or table selection claims.

Maintainer Objection Ledger

Accepted maintainer revisions:

• Clipboard residual cannot be implemented as one broad grab bag. It needs a source-drill slice before code.
• Focus/autoscroll must prove raw editor behavior, not product shell layout.
• DOM mutation repair stays refactor-only, but only if the next owner first checks whether richtext or generated stress already covers the invariant.
• Table, raw mobile, and collaboration browser deferrals are explicit non-claims.

High-Risk Gate

High-risk trigger: the plan routes external editor tests into Slate v2 behavior work across browser transport, destructive delete, clipboard, table selection, mobile/IME, collaboration, issue claims, and release gates. The main risk is false confidence: a unit row, ledger row, or planning command could be mistaken for browser, device, or issue-fix proof.

Blast radius:

Three-scenario pre-mortem:

Expanded proof plan:

High-risk verdicts:

Keep/revise/drop verdict: keep the plan, with proof-type guards. The plan stays pending until closure score and final gates are recorded.

Revision Pass Results

Revision verdict: the plan is execution-grade for the first ralph slice, but not globally done. The revision pass kept the high-risk guards and moved them into the implementation-facing parts of the plan so the future handoff is not just "read the whole essay and infer the contract."

Ralph-ready slices:

Stale-wording audit:

Final issue-accounting review can be a no-op if the next pass confirms:

• the revision pass changed no fixed/improved issue claim;
• no PR reference count changed after revision;
• no related issue classification changed after revision;
• #2558, raw mobile rows, and collaboration browser rows remain non-claims.

Score dimensions prepared for closure:

Prepared closure score: 0.92. No dimension is below 0.85.

Final Issue-Accounting Review

Verdict: no-op sync. The revision pass changed execution wording and proof gates only. It did not change issue claims, PR reference counts, or related classifications.

Evidence checked:

No matrix, dossier, PR reference, or live/open ledger edit is needed from this pass.

Closure Score And Final Gates

Closure verdict: done. This is a completed planning/accounting lane, not a Slate v2 implementation claim.

Final score: 0.92.

Next execution handoff:

• owner: ralph;
• target: first Ralph-ready slice, text-unit/delete package parity;
• primary Slate v2 owner: .tmp/slate-v2/packages/slate/test/text-units-contract.ts;
• Lexical source: ../lexical/packages/lexical-playground/__tests__/regression/7163-graphemes.spec.mjs;
• first proof: cd .tmp/slate-v2 && bun test ./packages/slate/test/text-units-contract.ts;
• closure proof after implementation: cd .tmp/slate-v2 && bun check;
• scope lock: no browser, raw mobile, table, collaboration, or issue claim from the first slice unless exact proof is added and the plan is reopened.

Implementation Phases

Phase 0: processing ledger only.

• Create docs/editor-test-harvester/lexical/slate-processing-ledger.md.
• Account for every portable and portable-mixed source file.
• Do not touch .tmp/slate-v2.

Phase 1: text-unit and destructive-delete package parity.

• Start with Lexical #7163 grapheme matrix.
• Refactor text-units-contract.ts and existing delete/unit-character fixtures instead of creating duplicate rows.
• Keep this first slice package-only unless a browser-specific gap appears.
• Do not claim issue closure from this slice unless an exact issue repro and matching Slate v2 proof are added.
• First required command: cd .tmp/slate-v2 && bun test ./packages/slate/test/text-units-contract.ts.
• Closure command after implementation: cd .tmp/slate-v2 && bun check.

Phase 2: clipboard HTML source-drill.

• Open exact Lexical link/list/text-format/table source rows.
• For each row, name expected Slate value/DOM output and whether it is raw parser behavior or example app policy.
• Only after this slice may a later ralph add focused browser/package rows.
• Do not edit .tmp/slate-v2 in this phase.
• Do not run ClawSweeper for this phase unless the source-drill changes an issue claim or classification.
• Status: complete on 2026-05-09. Exact accepted/rejected/deferred rows are in docs/editor-test-harvester/lexical/slate-processing-ledger.md.

Phase 3: clipboard HTML corpus residual.

• First slice: text-format HTML only. Complete on 2026-05-09. Google Docs-style BIU spans and safe semantic <b> handling are covered in .tmp/slate-v2/playwright/integration/examples/paste-html.test.ts and .tmp/slate-v2/site/examples/ts/paste-html-import.ts.
• Next slice: simple anchor import only. Complete on 2026-05-09. A browser row proves <a href="https://facebook.com">Facebook!</a> imports as a safe link in the paste-html example.
• Next slice: minimized noisy social HTML with link-in-list from Lexical #3193. Complete on 2026-05-09. The browser row keeps the leading block, two list items, and one safe link inside a list item.
• Next slice: basic unordered list import only. Complete on 2026-05-09.
• Next slice: compact nested-list import corpus only. Complete on 2026-05-09.
• Next slice: nested div text boundaries inside list items only. Complete on 2026-05-09.
• Next slice: Google Docs table import only: rows/cells, multi-paragraph cell, and 11pt text. Complete on 2026-05-09. No column-width, layout, or table-model claim.
• Next slice: Quip table import row with inside a cell. Complete on 2026-05-09. No width, layout, or table-model claim.
• Selected clipboard HTML source rows for this phase are now closed. Link paste-into-existing rows stay with inline paste; custom table widths, merge grids, rowspan/colspan, empty-row synthesis, HR-in-list splitting, raw autolink plugin behavior, and table-cell block normalization stay deferred.
• Deferred rows: custom table widths, table grid merge, rowspan/colspan, empty-row synthesis, HR-in-list splitting, raw autolink plugin behavior, and table-cell block normalization until their owners are accepted.
• Run package and browser proof by owner; do not close browser paste behavior from package-only tests.

Phase 4: DOM mutation repair residual.

• Treat Lexical Mutations.spec.mjs as refactor-existing.
• Strengthen richtext.test.ts, generated stress, or the policy unit owner only if a missing invariant remains after pressure review.
• Package policy tests can support this phase, but browser mutation behavior needs Playwright proof.
• Status: complete on 2026-05-09 for accepted Slate owners. Existing richtext browser rows cover text mutation/selection repair; dom-repair-policy-contract.ts covers structural restore-manager policy. No desktop structural repair or raw Android device claim.

Phase 5: focus, scroll, and native browser input.

• TextEntry, Events, Keyboard, Focus, AutoScroll, and remaining Mutations rows.
• Use browser proof where native transport or DOM authority is the behavior.
• Status: Focus and AutoScroll rows are complete on 2026-05-09 for accepted raw Slate behavior. Plain-text tab-out and broader TextEntry/Events/Keyboard rows remain separate owners unless a later slice accepts them.
• Do not claim raw mobile or raw IME device closure from desktop viewport rows.

Phase 6: selection and inline/void atom residual.

• Mention delete/paste boundaries.
• Link/hashtag/horizontal-rule navigation only when Slate owns the equivalent raw invariant.
• Split raw inline behavior from product toolbar, autolink, or app shell policy.

Phase 7: table model decision.

• Keep current containment and triple-click proof.
• Do not claim multi-cell/whole-table selection until Slate accepts a table selection model.
• Route raw mobile table rows to raw-device proof only.
• This plan does not accept a table selection model.

Phase 8: collaboration/browser.

• Keep current collaboration/history substrate rows.
• Add yjs/browser rows only when a real collab owner exists.
• This plan does not claim collaboration browser parity.

Public API Target

No public API change is accepted by this pass.

Any later ralph slice that discovers an API gap must update this plan first, including adoption story, migration pressure, and issue accounting.

Internal Runtime Target

Use the existing owner split:

• slate package tests for model/operation/history-like invariants;
• slate-dom for clipboard and DOM conversion;
• slate-react unit tests for runtime policy;
• Playwright examples/stress rows for browser transport.

Do not put browser-only correctness into model tests and call it done.

Hook, Component, And Render DX Target

No React hook/component API change is planned in this pass. If a Lexical row only tests Composer/plugin identity, reject it as product/framework shell.

Plate And slate-yjs Migration Backbone

Plate migration pressure: test rows should prove raw behavior that Plate can compose, not opinionated Plate features.

slate-yjs migration pressure: current substrate/history rows can be used as the base. Browser yjs behavior is deferred until a real collab browser owner exists.

Issue-Ledger Accounting

Current issue pass status: related discovery, issue-ledger sync, live-source refresh, pressure pass, maintainer objection pass, high-risk pass, revision pass, and final issue-accounting review complete. No issue-ledger files changed in the final review because no claims, counts, or related classifications changed.

Discovery inputs:

• gitcrawl doctor --json reports local archive readiness with version 0.2.1, 630 live-open issue rows in the committed live ledger, 617 local clusters, and no GitHub token. Broad local gitcrawl search queries for IME, clipboard, DOM selection, and text-unit families returned no hits, so this pass used the committed ledgers and candidate maps as the durable source.
• docs/editor-test-harvester/lexical/slate-processing-ledger.md accounts for 137 portable or portable-mixed Lexical source files: 5 already-applied, 1 covered, 12 create-new, 83 refactor-existing, 21 defer, and 15 reject.
• docs/slate-v2/ledgers/issue-coverage-matrix.md:28 keeps the proof bar explicit: browser, IME, mobile, clipboard, selection, and DOM bridge issues need matching browser or device proof.
• docs/slate-issues/package-impact-matrix.md:77 through docs/slate-issues/package-impact-matrix.md:83 assign Mobile/IME, Selection/Focus/DOM bridge, Clipboard, Core, and React runtime to their current v2 owners.

Related classifications from the Lexical processing ledger:

No new Fixes #... or Improves #... claim is made by this planning pass.

Issue sync results from pass 4:

• #5826 and #5806 already had matrix and dossier coverage; no row change.
• #5355 already had a matrix row and dossier section; the dossier status was aligned to not-claimed to match the actual decision.
• #4716 already had matrix and dossier coverage; the dossier status was aligned down from improves-claimed to cluster-synced because the decision and PR text are related-only.
• #5894, #4730, #3387, and #3872 now have conservative matrix rows and fork-dossier sections.
• docs/slate-v2/references/pr-description.md keeps fixed issue claims at 31, updates related matrix row count to 158, and records 0 new exact fixed/improved claims from the Lexical harvest sync.
• docs/slate-issues/gitcrawl-live-open-ledger.md remains unchanged because it is a generated live-field mirror and no live issue state changed.

Legacy Regression Proof Matrix

Browser Stress Strategy

Browser rows must be sequential when they build or share the Slate v2 site. The local solution note says not to run multiple site-backed Playwright commands in parallel unless the server path is explicitly shared (docs/solutions/workflow-issues/2026-05-08-slate-v2-playwright-webserver-checks-should-run-sequentially.md:43).

IME rows must use honest DOM composition boundaries; synthetic or model-only selection setup is not enough for native browser behavior (docs/solutions/developer-experience/2026-05-07-slate-browser-ime-proof-rows-need-honest-dom-composition.md:77).

DOM selection plans must rank external test sources by owner; Lexical is strong for browser-regression shape and skip discipline, not for every DOM selection architecture choice (docs/solutions/workflow-issues/2026-05-08-dom-selection-ralplans-should-rank-editor-test-sources-by-owner.md:47).

Applicable Implementation-Skill Review Matrix

High-Risk Pre-Mortem

Pass Schedule

Plan Deltas

• Created a new plan because the existing Lexical API steal plan is about API shape, not the test harvest.
• Reopened the harvest report's remaining backlog under a new processing lane.
• Kept the previous selected apply ledger intact; it remains useful but not comprehensive.
• Made slate-processing-ledger.md the next durable artifact.
• Completed the related issue discovery pass with conservative routing for table, clipboard, DOM selection, IME/mobile, focus/scroll, text-unit/delete, collaboration, and performance-adjacent Lexical rows.
• Kept all new Lexical issue relationships as related/proof-route or not-claimed; no new fixed or improved claim is added by this pass.
• Completed the issue-ledger sync pass for the named backfill candidates: #5826, #5806, #5894, #5355, #4730, #3387, #3872, and #4716.
• Added matrix/dossier coverage for #5894, #4730, #3387, and #3872; corrected stale dossier claim levels for #5355 and #4716; updated the PR reference related-row count without changing fixed claims.
• Completed live-source refresh against current .tmp/slate-v2 owners.
• Resolved the ordering conflict between the broad implementation phases and slate-processing-ledger.md: first apply slice is now text-unit/delete package parity, clipboard HTML source-drill comes before clipboard code, and DOM mutation repair is refactor-existing.
• Kept issue claims unchanged after source refresh: 0 new exact fixed or improved claims from the Lexical harvest plan.
• Completed the pressure pass. Text-unit/delete package parity survives as the first ralph slice; clipboard HTML residual stays second; DOM mutation repair is still refactor-existing; focus/autoscroll and inline atom rows stay later browser slices; table, raw mobile, and collaboration browser rows stay deferred.
• Updated docs/editor-test-harvester/lexical/slate-processing-ledger.md so its priority queue matches the plan and Mutations.spec.mjs no longer says create-new.
• Completed the maintainer objection ledger. The pass kept text-unit/delete as first implementation slice, inserted a clipboard source-drill phase before clipboard code, kept DOM mutation repair as refactor-only, and made table, raw mobile, and collaboration browser rows explicit non-claims.
• Completed the high-risk pass. The plan now gates package, browser, raw mobile, table-model, collaboration, issue-claim, and planning verification proof separately; it also records that clipboard source-drill does not need ClawSweeper before code when claims remain unchanged.
• Raised score to 0.92, but kept completion pending because revision, final issue-accounting review, and closure remain runnable.
• Completed the revision pass. High-risk guards are now reflected in Ralph-ready slices and each implementation phase. Prepared score dimensions keep the plan at 0.92 with no dimension below 0.85.
• Confirmed the next issue-accounting pass can be a no-op if it verifies that the revision changed no claims, counts, or related classifications.
• Completed final issue-accounting review as a no-op: fixed/improved claims, PR counts, related classifications, fork dossier statuses, and non-claim surfaces stayed unchanged.
• Completed closure scoring at 0.92 with no dimension below 0.85.
• Marked this ralplan done and prepared active goal state for the next ralph execution slice: text-unit/delete package parity only.
• Completed the first ralph execution slice. Lexical #7163 is now applied in Slate's text-unit contract, reverse complex-script deletion follows the same text-unit law as forward deletion, Thai unit-character fixtures are aligned, and .tmp/slate-v2 bun check passes.
• Completed the lexical-history LexicalHistory source-read slice. .tmp/slate-v2/packages/slate-history/test/history-contract.ts now proves redo-stack clearing after undo plus a new edit, selected block property undo/redo, and node-property history capture while ignoring empty updates. Lexical command notification APIs, clear-history command shape, React harness details, shared nested-editor history, raw mobile, collaboration, table-model, and issue claims stayed out. Focused package proof and .tmp/slate-v2 bun check passed. The reusable harvest boundary is captured in docs/solutions/best-practices/2026-05-09-lexical-history-harvest-rows-need-stack-law-contracts.md.
• Kept completion pending for the broader execution lane because lexical-html LexicalHtml source-read is the next runnable slice.
• Completed the lexical-html LexicalHtml source-read slice with no Slate v2 code/test change. Selected-fragment export is already covered by .tmp/slate-v2/packages/slate-dom/test/clipboard-boundary.ts, and paragraph alignment import is already covered by the paste-html browser row and importer. Standalone full-editor HTML serialization and custom node DocumentFragment export stay rejected/deferred until a generic Slate HTML serializer owner exists. Focused existing proofs passed: bun test ./packages/slate-dom/test/clipboard-boundary.ts and PLAYWRIGHT_RETRIES=0 bunx playwright test playwright/integration/examples/paste-html.test.ts --project=chromium -g "paragraph alignment".
• Kept completion pending for the broader execution lane because lexical-list utils source-read is the next runnable slice.
• Completed the lexical-list utils source-read slice. The accepted Slate row is public path/query behavior, not Lexical helper API shape: .tmp/slate-v2/packages/slate/test/query-contract.ts now proves nested list depth, non-root top list ancestry, descendant list ancestors, and terminal versus non-terminal list-item paths through Path and Node APIs. Focused query proof, bun run lint:fix, and .tmp/slate-v2 bun check passed. The reusable harvest rule is captured in docs/solutions/best-practices/2026-05-09-lexical-list-utils-harvest-rows-need-path-ancestry-contracts.md.
• Kept completion pending for the broader execution lane because Lexical regression #231 empty-text-nodes source-read is the next runnable slice.

Open Questions

No user decision is needed for the next pass.

No open planning questions remain. The next action is execution under ralph, continuing with Lexical regression #231 empty-text-nodes source-read before any new normalization, delete, rendered-DOM, or browser row changes.

Fast Driver Gates

Planning repo gates:

cd /Users/zbeyens/git/plate-2
pnpm lint:fix
bun run completion-check -- --file .tmp/completion-checks/slate-v2-lexical-harvest-test-processing-ralplan.md

Slate v2 gates for later implementation slices:

cd /Users/zbeyens/git/slate-v2
bun test ./packages/slate/test/text-units-contract.ts
bun test ./packages/slate-dom/test/clipboard-boundary.ts
bun test ./packages/slate-history/test/history-contract.ts
PLAYWRIGHT_RETRIES=0 bunx playwright test playwright/integration/examples/paste-html.test.ts --project=chromium
PLAYWRIGHT_RETRIES=0 bunx playwright test playwright/integration/examples/tables.test.ts --project=chromium
PLAYWRIGHT_RETRIES=0 bunx playwright test playwright/stress/generated-editing.test.ts --project=chromium -g '<accepted-family>'
bun check

Release/browser closure gate:

cd /Users/zbeyens/git/slate-v2
bun check:full

Raw device gate:

cd /Users/zbeyens/git/slate-v2
SLATE_BROWSER_RAW_MOBILE_REQUIRED=1 bun test:mobile-device-proof:raw

Final Completion Gates

This plan can become done only when:

• docs/editor-test-harvester/lexical/slate-processing-ledger.md exists;
• every portable and portable-mixed source file has a processing status;
• every create/refactor row has a Slate owner and command;
• every defer/reject row has a concrete reason;
• ClawSweeper issue accounting is complete for every issue-facing row;
• no raw mobile, whole-table selection, or yjs browser claim is made without the matching proof owner;
• any implementation slice executed under this plan has passed its relevant .tmp/slate-v2 gates;
• final score is at least 0.92 with no dimension below 0.85.

Closure status: all gates above are satisfied for the planning lane. Executed implementation slices have passed their focused proof, lint where relevant, and .tmp/slate-v2 bun check. Broader Lexical apply work remains pending because later runnable slices still exist.

Ralph execution update on 2026-05-09: LexicalTabNode source-read is complete. .tmp/slate-v2/packages/slate-dom/test/clipboard-boundary.ts now proves multiline plain-text fallback preserves literal tabs while splitting blocks, and .tmp/slate-v2/playwright/integration/examples/paste-html.test.ts now proves Google Docs Apple-tab-span HTML preserves two visible Hello\tworld lines from the paragraph plus loose-line source shape. A trial switch from tx.nodes.insert to tx.fragment.insert in the paste-html example removed the empty paragraph but broke existing nested-list import proof, so the importer path stayed unchanged and the new browser assertion targets the visible tab/newline invariant. Focused package proof, full paste-html Chromium proof, post-format focused Playwright proof, lint, and .tmp/slate-v2 bun check passed. The reusable rule is captured in docs/solutions/best-practices/2026-05-09-lexical-tab-node-harvest-rows-need-clipboard-browser-proof-boundaries.md. Lexical TabNode class/schema/serialization/command API rows stay out; tab-indent and adjacent literal-tab selection rows are deferred to later owners. No raw mobile, collaboration, table-model, or issue claim changed.

Ralph execution update on 2026-05-09: LexicalEditor source-read is complete. .tmp/slate-v2/packages/slate-react/test/rendered-dom-shape-contract.tsx now proves that an empty inline element inside a non-empty block does not render a visual line break placeholder. Source-read classified the rest of LexicalEditor.test.tsx conservatively: Slate's read-update-contract, transaction-contract, commit-metadata-contract, normalization-contract, operations-contract, and snapshot-contract already cover the accepted read/update, batching, command, normalization, move, and state invariants. Lexical editor API shape, node-key maps, listener/mutation registration, Lexical update tags, decorator rendering, root mounting, DOM reconciliation, and product-shell rows stay out or remain separate runtime owners. Focused React test proof, bun run lint:fix, and .tmp/slate-v2 bun check passed. No raw mobile, collaboration, table-model, or issue claim changed.

Ralph execution update on 2026-05-09: LexicalEditorState source-read is complete. .tmp/slate-v2/packages/slate/test/snapshot-contract.ts now proves that removing a node purges the removed block and text runtime IDs from both current idToPath and pathToId. Source-read classified the rest of LexicalEditorState.test.ts: snapshot immutability and read/update boundaries are covered by existing Slate owners; Lexical EditorState constructor, frozen node-map identity, JSON schema, node keys, and parser/editor-context API rows stay out. The first focused Bun command used a name-filter path and failed before execution; rerunning with ./packages/slate/test/snapshot-contract.ts passed all 196 tests. bun run lint:fix && bun check passed from .tmp/slate-v2, the focused snapshot test passed again after formatting, and pnpm lint:fix passed in plate-2. No browser, raw mobile, collaboration, table-model, or issue claim changed.

Ralph execution update on 2026-05-09: LexicalElementHelpers source-read is complete with no Slate v2 code/test change. The file only tests addClassNamesToElement and removeClassNamesFromElement for basic, empty, multiple, space-separated, and whitespace-padded DOM class strings. That is a Lexical DOM utility helper, not portable editor behavior for Slate v2, so the ledger now rejects the row instead of keeping the stale refactor-existing classification. No browser, raw mobile, collaboration, table-model, or issue claim changed.

Ralph execution update on 2026-05-09: LexicalNode source-read is complete. .tmp/slate-v2/packages/slate/test/query-contract.ts now proves the portable node-shape part of LexicalNode: ancestry order, reverse ancestry, sibling path order, common ancestor resolution, previous/next path helpers, and parent/ancestor predicates. Source-read classified the rest conservatively: text content, empty-block normalization, selected removal/rebase, text split/merge, and structural move/insert/replace rows are already owned by existing Slate query, normalization, operation, transform, and selection contracts. Lexical class APIs, node keys, clone/config/JSON, live-node read throws, DOM rendering, token/segmented/directionless text modes, and selectNext helper shape stay out or route to later keyboard/selection owners. Focused query proof, bun run lint:fix, and .tmp/slate-v2 bun check passed. No browser, raw mobile, collaboration, table-model, or issue claim changed.

Ralph execution update on 2026-05-09: LexicalNormalization source-read is complete. The file only tests Lexical $normalizeSelection, so the accepted Slate behavior is query/location edge resolution rather than tree repair. .tmp/slate-v2/packages/slate/test/query-contract.ts now proves that element paths, nested element paths, and backward ranges resolve to stable text points through Editor.range, Editor.edges, and Editor.point. Lexical decorator endpoint semantics and element-point API shape stay out or route to a future decorator/void/browser selection owner. Focused query proof, bun run lint:fix, and .tmp/slate-v2 bun check passed. The reusable rule is captured in docs/solutions/best-practices/2026-05-09-lexical-normalization-harvest-rows-need-selection-query-boundaries.md. No browser, raw mobile, collaboration, table-model, or issue claim changed.

Ralph execution update on 2026-05-09: LexicalSerialization source-read is complete. The file is one large Lexical editor-state JSON snapshot plus a parse/stringify round-trip. The accepted Slate behavior is not Lexical's editor-state schema; it is raw document value JSON portability through the public state API. .tmp/slate-v2/packages/slate/test/state-tx-public-api-contract.ts now proves rich raw document values round-trip through JSON, can be rehydrated as initialValue, and do not serialize runtime index metadata such as pathToId or idToPath. Lexical node keys, parser API shape, class schema, and code/list/table metadata stay out. Focused public API proof, bun run lint:fix, and .tmp/slate-v2 bun check passed. The reusable rule is captured in docs/solutions/best-practices/2026-05-09-lexical-serialization-harvest-rows-need-public-value-roundtrip-tests.md. No browser, raw mobile, collaboration, table-model, or issue claim changed.

Ralph execution update on 2026-05-09: LexicalUtils source-read is complete. The useful missing row was generic shortcut matching, not Lexical node utility internals. .tmp/slate-v2/packages/slate-dom/test/hotkeys.ts now proves direct isHotkey calls match semantic letter keys before physical-code fallback, handle uppercase keys, reject ASCII remapped-layout physical keys, and still fallback to event.code for non-English single-letter layouts. Existing Slate owners cover accepted query/runtime-id, state/read/update, commit metadata, focus/autoscroll, and browser selection behavior. Lexical microtask helpers, random keys, token/segmented text modes, node-key maps, cached type maps, node replacement/copy class APIs, listener-order internals, and generic DOM helpers stay out or route to later dedicated owners. The first focused command hit Bun's ignored .test.ts wrapper filter; rerunning the real source file ./packages/slate-dom/test/hotkeys.ts passed. bun run lint:fix and .tmp/slate-v2 bun check passed. No browser, raw mobile, collaboration, table-model, or issue claim changed.

Ralph execution update on 2026-05-09: LexicalUtilsInsertNodeToNearestRoot source-read is complete. The portable row maps to Slate's existing insertNodes mode: 'highest' transform behavior, not to Lexical's helper API. .tmp/slate-v2/packages/slate/test/transforms-contract.ts now proves highest-mode block insertion splits the highest selected block and inserts the block at the root level. Existing insertNodes fixtures already cover paragraph middle/start/end, empty paragraph, root start/child/end, and default selection insertion. Lexical command dispatch, node-key maps, decorator class shape, shadow-root wording, and list HTML specifics stay out or route to later list/plugin owners. Focused package proof, bun run lint:fix, and .tmp/slate-v2 bun check passed. No browser, raw mobile, collaboration, table-model, or issue claim changed.

Ralph execution update on 2026-05-09: LexicalUtilsSplitNode source-read is complete. The useful missing row was public root split rejection. Slate already covered normal split behavior through splitNodes fixtures and operation/snapshot contracts: paragraph middle/start/end, multi-text element split, nested block split, highest-mode ancestor split, mark/property preservation, selection rebase, and raw root operation rejection. .tmp/slate-v2/packages/slate/src/transforms-node/split-nodes.ts now rejects Editor.splitNodes(..., { at: [] }) with a direct editor-root error, and .tmp/slate-v2/packages/slate/test/transforms-contract.ts proves that public contract. Focused package proof, bun run lint:fix, and .tmp/slate-v2 bun check passed. No browser, raw mobile, collaboration, table-model, or issue claim changed.

Ralph execution update on 2026-05-09: lexical-code-shiki and lexical-code LexicalCodeNodeTabs source-read is complete. The portable Slate-owned row is not Lexical's full indent/outdent command matrix; it is the code-highlighting example's raw Tab behavior. .tmp/slate-v2/playwright/integration/examples/code-highlighting.test.ts now proves Tab inside a code line inserts the configured spaces and advances the caret. Lexical INDENT_CONTENT_COMMAND/OUTDENT_CONTENT_COMMAND, backward selection preservation, Shiki tokenizer/theme loading, node class APIs, and product command dispatch stay out or wait for a future accepted code-block indent owner. Focused browser proof, bun run lint:fix, and .tmp/slate-v2 bun check passed. No raw mobile, collaboration, table-model, or issue claim changed.

Ralph execution update on 2026-05-09: lexical-code LexicalCodeNode source-read is complete. The useful missing row was not Lexical's code-node class lifecycle; it was multi-line code indent/outdent behavior. .tmp/slate-v2/playwright/integration/examples/code-highlighting.test.ts now proves Tab and Shift+Tab indent/outdent every selected code line, and .tmp/slate-v2/site/examples/ts/code-highlighting.tsx routes expanded code-line selections through line-start edits while keeping collapsed Tab as configured space insertion at the caret. Lexical constructor/createDOM/updateDOM/exportJSON, language metadata, tokenizer/highlighter transform policy, command dispatch, Alt+Arrow line shifting, Home/End visual-caret policy, node keys, raw mobile, collaboration, table-model, and issue claims stay out or wait for separate accepted owners. Focused browser proof, full code-highlighting browser proof, bun run lint:fix, and .tmp/slate-v2 bun check passed.

Ralph execution update on 2026-05-09: Lexical regression #231 empty-text-nodes source-read is complete. .tmp/slate-v2/packages/slate/test/delete-contract.ts now proves repeated Backspace through token-like marked text collapses to a canonical empty block with selection at [0, 0] offset 0. Existing normalization, leaf lifecycle, and rendered-DOM owners cover empty block anchors and invalid empty leaf cleanup. Focused delete proof, bun run lint:fix, and .tmp/slate-v2 bun check passed. No hashtag plugin, Playground CSS, collaboration harness, raw mobile, table-model, or issue claim changed.

The completion checkpoint stays pending because Lexical regression #3433 merge-markdown-lists source-read is the next runnable slice. Continue there before adding markdown/list model or browser rows.

Ralph execution update on 2026-05-09: Lexical regression #3433 merge-markdown-lists source-read is complete. .tmp/slate-v2/playwright/integration/examples/markdown-shortcuts.test.ts now proves that a markdown-created list immediately before an existing list renders as one ul with two before one. The first focused proof failed red with two adjacent lists. .tmp/slate-v2/site/examples/ts/markdown-shortcuts.tsx now merges adjacent bulleted-list wrappers after wrapping the current block as a list item. Focused browser proof, full markdown-shortcuts browser file, bun run lint:fix, and .tmp/slate-v2 bun check passed. No generic core wrapNodes/mergeNodes behavior, raw mobile, collaboration, table-model, or issue claim changed.

The completion checkpoint stays pending because LexicalHeadingNode.test.ts source-read is the next runnable non-table slice.

Ralph execution update on 2026-05-09: LexicalHeadingNode source-read is complete. .tmp/slate-v2/playwright/integration/examples/richtext.test.ts now proves heading middle split, heading-end paragraph insertion, and empty-heading paragraph insertion from the browser. The first focused proof failed red with two h1 elements at heading end. .tmp/slate-v2/site/examples/ts/richtext.tsx now routes Enter at heading end or in an empty heading to a paragraph after the heading, while middle Enter stays on the generic core split path. Focused heading proof, full heading grep in the richtext browser file, bun run lint:fix, and .tmp/slate-v2 bun check passed. No generic core insertBreak/splitNodes behavior, raw mobile, collaboration, table-model, or issue claim changed.

The completion checkpoint stays pending because LexicalQuoteNode.test.ts is the next runnable non-table slice.

Ralph execution update on 2026-05-09: LexicalQuoteNode source-read is complete. .tmp/slate-v2/playwright/integration/examples/richtext.test.ts now proves Enter in an empty blockquote keeps one blockquote and creates a paragraph after it. The first focused proof failed red with two blockquote elements. .tmp/slate-v2/site/examples/ts/richtext.tsx now shares one exit-on-Enter policy for heading-one, heading-two, and block-quote. Focused quote proof, combined heading/quote browser grep, bun run lint:fix, and .tmp/slate-v2 bun check passed. No generic core insertBreak/splitNodes behavior, raw mobile, collaboration, table-model, or issue claim changed.

The completion checkpoint stays pending because $sliceSelectedTextNodeContent.test.ts is the next runnable non-table slice.

Ralph execution update on 2026-05-09: Lexical selection $sliceSelectedTextNodeContent source-read is complete. .tmp/slate-v2/packages/slate/test/clipboard-contract.ts now proves selected fragment extraction across unmarked and bold text leaves slices both edges and does not mutate source text when the returned fragment is changed. Focused package proof, full clipboard contract, bun run lint:fix, and .tmp/slate-v2 bun check passed. No browser, raw mobile, collaboration, table-model, token/segmented mode, or issue claim changed.

The completion checkpoint stays pending because LexicalListItemNode.test.ts is the next runnable non-table slice.

Ralph execution update on 2026-05-09: LexicalListItemNode source-read is complete. .tmp/slate-v2/packages/slate/test/snapshot-contract.ts now proves insertBreak inside a non-empty list item splits the item while preserving the bulleted-list wrapper and moving selection into the new item. Focused package proof, full snapshot contract, bun run lint:fix, and .tmp/slate-v2 bun check passed. Lexical class/DOM/create/type-guard/export rows stay rejected. Replace/remove nested-list normalization, indent/style inheritance, and ordered-list restart policy stay deferred to explicit list-plugin/model owners. No raw mobile, collaboration, table-model, or issue claim changed.

The completion checkpoint stays pending because LexicalListNode.test.ts is the next runnable non-table slice.

Ralph execution update on 2026-05-09: LexicalListNode source-read is complete. .tmp/slate-v2/packages/slate/test/snapshot-contract.ts now proves selected top-level blocks can be converted to list-item children and wrapped in a numbered-list wrapper, complementing the existing bulleted-list proof. Focused package proof, full snapshot contract, bun run lint:fix, and .tmp/slate-v2 bun check passed. Lexical class/DOM/depth-class/update/create, subclass transform wiring, checklist role cleanup, and private list-value rows stay rejected or deferred to explicit list-plugin/model owners. No raw mobile, collaboration, table-model, or issue claim changed.

Ralph execution update on 2026-05-09: lexical-list formatList.test.ts source-read is complete. No .tmp/slate-v2 code or tests changed. Existing Slate owners already cover root list formatting, numbered/bulleted wrapper creation, and non-empty list continuation through the snapshot contract plus markdown and richtext browser rows. Lexical command/extension setup, table-cell shadow-root insertion, element-anchored LineBreakNode selection, decorator-node list continuation, subclass preservation, indent/outdent helper APIs, empty/whitespace list-item exit, middle-empty-item list splitting, and ordered-list metadata stay deferred or rejected until explicit list-plugin/model/table/product owners accept them. No browser, raw mobile, collaboration, table-model, or issue claim changed.

The completion checkpoint stays pending because registerListStrictIndentTransform.test.ts is the next runnable non-table slice.

Ralph execution update on 2026-05-09: lexical-list registerListStrictIndentTransform.test.ts source-read is complete. No .tmp/slate-v2 code or tests changed. Current paste-html browser coverage already proves compact malformed nested-list import variants without flattening visible list content. Lexical's exact strict indentation rewrite, $generateHtmlFromNodes HTML output, list value attributes, and transform registration stay deferred or rejected until Slate accepts a list-normalization or generic HTML serializer owner. No browser, raw mobile, collaboration, table-model, or issue claim changed.

The completion checkpoint stays pending because LexicalMarkdown.test.ts is the next runnable non-table slice.

Ralph execution update on 2026-05-09: lexical-markdown LexicalMarkdown.test.ts source-read is complete. No .tmp/slate-v2 code or tests changed. Current Slate v2 owns markdown shortcuts and markdown preview examples, not a markdown import/export package. Existing browser rows already cover blockquote, list, heading, list continuation, and adjacent markdown-list merge shortcuts. The broad import/export matrix, markdown normalizer, nested fence round-trips, whitespace and hard-break import, link/mark escape matrix, MDX/custom transformer rows, list-marker/checklist metadata, selection-after-import semantics, and code-fence Enter shortcut stay deferred or rejected until explicit markdown serialization, markdown transformer, list/checklist, or code-block shortcut owners accept them. No browser, raw mobile, collaboration, table-model, or issue claim changed.

The completion checkpoint stays pending because MarkdownTransformers.test.ts is the next runnable non-table slice.

Ralph execution update on 2026-05-09: lexical-markdown MarkdownTransformers.test.ts source-read is complete. No .tmp/slate-v2 code or tests changed. The Lexical rows are markdown-link transformer behavior: preserving text or marks before a parsed markdown link, avoiding greedy matches, and rejecting markdown link creation inside an existing link. Current Slate v2 already has raw inline-link paste/copy boundary coverage, but no markdown-link automd or serializer API. These rows stay deferred or rejected until an explicit markdown-link plugin/serializer owner exists. No browser, raw mobile, collaboration, table-model, or issue claim changed.

The completion checkpoint stays pending because ClearFormatting.spec.mjs is the next runnable non-table slice.

Ralph execution update on 2026-05-09: lexical-playground ClearFormatting.spec.mjs source-read is complete. The portable accepted rows are selected mark removal, block alignment reset, semantic block preservation, and partial multi-paragraph clearing. .tmp/slate-v2/playwright/integration/examples/richtext.test.ts now proves clearing a selected range from the middle of one paragraph through the middle of the next removes marks only from the selected text, clears block alignment, and preserves blockquote structure. The first focused browser proof failed red because the richtext toolbar had no clear-formatting control. .tmp/slate-v2/site/examples/ts/richtext.tsx now exposes a clear-formatting button that removes the richtext example's text marks and resets selected block alignment without changing semantic block types. Link, mention, hashtag, indent/outdent, theme markup, raw mobile, collaboration, table-model, and issue rows stay out or wait for explicit future owners. Focused browser proof, bun run lint:fix, rerun focused browser proof, and .tmp/slate-v2 bun check passed.

The completion checkpoint stays pending because CodeBlock.spec.mjs is the next runnable non-table slice.

Ralph execution update on 2026-05-09: lexical-playground CodeBlock.spec.mjs source-read is complete. .tmp/slate-v2/playwright/integration/examples/code-highlighting.test.ts now proves selecting paragraph text and clicking the code-block control creates a code block with code-line content and keeps the existing code blocks intact. Existing code-highlighting browser rows cover Enter inside code lines, collapsed Tab insertion, and selected multi-line Tab/Shift+Tab indentation. Markdown code fences, language parser matrix, auto-indent-on-newline, line movement, visual navigation, theme DOM, raw mobile, collaboration, table-model, and issue rows stay deferred or rejected under explicit future owners. Focused browser proof passed.

The completion checkpoint stays pending because ElementFormat.spec.mjs is the next runnable non-table slice.

Ralph execution update on 2026-05-09: lexical-playground ElementFormat.spec.mjs source-read is complete. The accepted portable row is fresh empty-paragraph center alignment from the toolbar. .tmp/slate-v2/playwright/integration/examples/richtext.test.ts now proves a fresh empty paragraph can be center-aligned while the Slate selection stays collapsed on that empty block. .tmp/slate-v2/site/examples/ts/richtext.tsx now explicitly targets the containing block path for collapsed alignment, rather than relying on implicit selected-block expansion. The first browser attempts failed because full-document delete hit placeholder text and then retained centered alignment in the model; the final proof creates a fresh empty paragraph from an unaligned block. The Lexical caret-within-link plus indent row stays deferred because this checkout has separate richtext alignment and inline-link owners, but no combined link/block-format owner and no indent API. Toolbar shell/theme rows stay rejected. Focused browser proof passed. No raw mobile, collaboration, table-model, link-plugin policy, indent policy, or issue claim changed. Follow-up verification passed in .tmp/slate-v2: bun run lint:fix, the richtext Playwright grep for empty paragraph alignment, existing toolbar alignment, and clear formatting, then bun check.

The completion checkpoint stays pending because Headings/HeadingsBackspaceAtStart.spec.mjs is the next runnable non-table slice.

Ralph execution update on 2026-05-09: lexical-playground Headings/HeadingsBackspaceAtStart.spec.mjs source-read is complete. The portable row is Backspace at offset 0 of the first heading with no previous sibling. .tmp/slate-v2/playwright/integration/examples/richtext.test.ts now proves that path is a no-op: the h1 text remains Welcome to the playground, no paragraph is created, and the collapsed selection stays at [0, 0] offset 0. Runtime already behaved correctly; this pass added the missing proof. Lexical Playground setup, dropdown/theme DOM, raw mobile, collaboration, table-model, and issue rows stay out. Focused browser proof passed. Follow-up verification passed in .tmp/slate-v2: bun run lint:fix, the richtext Playwright grep for first-heading Backspace and heading Enter behavior, then bun check.

The completion checkpoint stays pending because Headings/HeadingsEnterAtEnd.spec.mjs is the next runnable non-table slice.

Ralph execution update on 2026-05-09: lexical-playground Headings/HeadingsEnterAtEnd.spec.mjs source-read is complete. No .tmp/slate-v2 code or test changed. Existing richtext browser proof already covers the accepted row: Enter at the end of a heading leaves the heading in place and creates a following paragraph. The example policy lives in handleExitBlockEnter. Lexical Playground setup, dropdown/theme DOM, raw mobile, collaboration, table-model, and issue rows stay out. Same-turn heading browser verification already passed while closing the preceding heading slice.

The completion checkpoint stays pending because Headings/HeadingsEnterInMiddle.spec.mjs is the next runnable non-table slice.

Ralph execution update on 2026-05-09: lexical-playground Headings/HeadingsEnterInMiddle.spec.mjs source-read is complete. No .tmp/slate-v2 code or test changed. Existing richtext browser proof already covers the accepted row: Enter in the middle of a heading splits it into two headings. Lexical movement helpers, Playground setup, dropdown/theme DOM, raw mobile, collaboration, table-model, and issue rows stay out. Same-turn heading browser verification already passed while closing the preceding heading slice.

The completion checkpoint stays pending because Indentation.spec.mjs is the next runnable non-table slice.

Ralph execution update on 2026-05-09: lexical-playground Indentation.spec.mjs source-read is complete. No .tmp/slate-v2 code or test changed. The file is a mixed product-owner suite, not one raw Slate invariant: existing code-highlighting browser rows already cover selected code-line Tab/Shift+Tab indentation, and snapshot-contract.ts already covers a basic list outdent model flow by lifting selected list items to paragraphs. Paragraph padding indent, table-cell paragraph indent, max-depth caps, nested list indent limits, negative indent cleanup after HTML padding import, and exact toolbar DOM output stay deferred or rejected until Slate accepts explicit paragraph-indent, list-indent, table-indent, or HTML-indent owners. No raw mobile, collaboration, table-model, or issue claim changed.

The completion checkpoint stays pending because List.spec.mjs is the next runnable non-table slice.

Ralph execution update on 2026-05-09: lexical-playground List.spec.mjs source-read is complete. .tmp/slate-v2/playwright/integration/examples/richtext.test.ts now proves selected richtext blocks can be turned into a bulleted list, toggled back to paragraphs, turned into a numbered list, and converted back to a bulleted list from the toolbar. Existing snapshot/browser owners already cover basic list wrapper creation, numbered/bulleted list formatting, non-empty list continuation, and markdown-created adjacent list merge. Lexical checklist focus/keyboard behavior, indent/outdent depth, empty-list-item exit, collapse at start, ordered start metadata, paragraph markdown suppression inside lists, format-menu list splitting, autolink/list interactions, raw mobile, collaboration, table-model, and exact theme DOM output stay deferred or rejected until explicit future owners accept them. Focused browser proof passed.

The completion checkpoint stays pending because Markdown.spec.mjs is the next runnable non-table slice.

Ralph execution update on 2026-05-09: lexical-playground Markdown.spec.mjs source-read is complete. .tmp/slate-v2/playwright/integration/examples/markdown-shortcuts.test.ts now proves a numeric ordered-list markdown shortcut with a non-1 start number: typing 25. creates a numbered list rendered as ol start="25" and accepts item text. .tmp/slate-v2/site/examples/ts/markdown-shortcuts.tsx now parses ordered-list shortcuts and unwraps both bulleted and numbered lists on markdown Backspace exit. Existing markdown-shortcuts rows already cover headings, blockquote, bulleted lists, list continuation, and adjacent markdown-created bulleted-list merge. Markdown import/export cycles, inline text transformers, emoji/image/equation decorators, code-formatted text exclusion, selection after link text-match transforms, list-marker export/copy behavior, HR/code-fence shortcuts, raw mobile, collaboration, table-model, and exact theme DOM output stay deferred or rejected until explicit future owners accept them. Focused browser proof passed.

The completion checkpoint stays pending because TextFormatting.spec.mjs is the next runnable non-table slice.

Ralph execution update on 2026-05-09: lexical-playground TextFormatting.spec.mjs source-read is complete. .tmp/slate-v2/packages/slate/test/snapshot-contract.ts now proves Editor.toggleMark and tx.marks.toggle clear inherited collapsed marks before the next inserted text. .tmp/slate-v2/packages/slate/src/core/public-state.ts now defaults tx.marks.toggle to true instead of a fresh object, and .tmp/slate-v2/site/examples/ts/richtext.tsx now prevents default browser handling for editor-owned mark hotkeys. .tmp/slate-v2/playwright/integration/examples/richtext.test.ts now proves bold, italic, underline, and code hotkeys create formatted inserted text and clear active marks before later plain text. The first browser proof failed red because inserted text stayed plain; after default prevention, it failed again because Plain stayed bold, exposing the transaction toggle default/inherited-mark bug. Capitalization, font-size/family, strikethrough, date-time/decorator formatting, toolbar active-state CSS, multiline toolbar state, raw mobile, collaboration, table-model, and exact theme DOM output stay deferred or rejected until explicit future owners accept them. Focused package and browser proofs passed.

The completion checkpoint stays pending because ../lexical/packages/lexical/src/__tests__/unit/CodeBlock.test.ts is the next runnable non-table slice.

Ralph execution update on 2026-05-09: lexical core CodeBlock.test.ts source-read is complete. The portable row is code-source HTML import through a DataTransfer-style rich HTML insertion path, not Lexical node class behavior. .tmp/slate-v2/playwright/integration/examples/paste-html.test.ts now proves Quip-style <pre>, VS Code-style white-space: pre line <div>s, and GitHub-style code tables import as one <pre><code> block without source gutters. .tmp/slate-v2/site/examples/ts/paste-html-import.ts now extracts code-source text before generic table/paragraph import, preserves line breaks, normalizes non-breaking spaces, and strips trailing source newlines. A first browser-native clipboard version failed red because Chromium flattened <pre> before the app importer saw it; the final proof uses the same app DataTransfer insertion boundary as Lexical's unit test so the existing kernel-trace gauntlet stays intact. Source-token colors/bold keyword styling, broad Google Docs title inference, sub/sup typography, native clipboard transport, raw mobile, collaboration, table-model, and issue rows stay deferred or rejected. Focused and full paste-html browser proofs passed, and .tmp/slate-v2 bun run lint:fix && bun check passed. Added docs/solutions/best-practices/2026-05-09-lexical-codeblock-harvest-rows-need-data-transfer-boundaries.md to preserve the DataTransfer-vs-native-clipboard boundary lesson.

The completion checkpoint stays pending because ../lexical/packages/lexical/src/__tests__/unit/LexicalEditorListener.test.ts is the next runnable source-read slice.

Ralph execution update on 2026-05-09: lexical core LexicalEditorListener.test.ts source-read is complete. Current Slate already owns the portable subscriber lifecycle through Editor.subscribe / editor.subscribe unsubscribe, Editor.subscribeSource source routing, Editor.registerCommitListener cleanup, and the public commit/subscriber contract tests in snapshot-contract.ts, transaction-contract.ts, collab-history-runtime-contract.ts, and apply-onchange-hard-cut-contract.ts. Lexical registerRootListener, registerEditableListener, private _listeners, setRootElement, and setEditable callback cleanup rows are editor/root-shell API policy, not raw Slate behavior. No .tmp/slate-v2 code or test changed, and no browser, raw mobile, collaboration, table-model, or issue claim changed.

The completion checkpoint stays pending because ../lexical/packages/lexical/src/__tests__/unit/LexicalListPlugin.test.tsx is the next runnable source-read slice.

Ralph execution update on 2026-05-09: lexical core LexicalListPlugin.test.tsx source-read is complete. The accepted portable row is empty paragraph toolbar list toggle, not Lexical React plugin wiring. .tmp/slate-v2/playwright/integration/examples/richtext.test.ts now proves a fresh empty paragraph toggles into one bulleted-list item and back to one empty paragraph while preserving collapsed selection. Current Slate already covers selected block list toggle/conversion, list wrapper creation, and basic list outdent. Empty-list indent/outdent and regression #7036 block-type-after-nested list behavior stay deferred or rejected until Slate accepts explicit list-indent, list-normalization, or list-plugin policy owners. The first focused proof failed because the empty paragraph DOM includes rendered placeholder text; the final proof asserts model text and selection instead. Focused list browser proof, bun run lint:fix, and bun check passed. No runtime code changed, and no raw mobile, collaboration, table-model, or issue claim changed.

The completion checkpoint stays pending because ../lexical/packages/lexical-playground/__tests__/e2e/CopyAndPaste/html/HTMLCopyAndPaste.spec.mjs is the next runnable source-read slice.

Ralph execution update on 2026-05-09: lexical-playground HTMLCopyAndPaste.spec.mjs source-read is complete. The accepted rows were multiline paragraph HTML with extra raw newlines and code-source HTML in a <code> element with line breaks. .tmp/slate-v2/playwright/integration/examples/paste-html.test.ts now proves multiline pasted HTML normalizes to four paragraphs, preserves the intended spaces and inline <b> / <i> formatting, and imports <code> with as a code block through the existing source-code HTML corpus. .tmp/slate-v2/site/examples/ts/paste-html-import.ts now strips raw newlines from normal HTML text nodes, drops whitespace-only inter-block text, preserves explicit code/pre whitespace, and removes the empty paste target created by native block-fragment paste. Existing source-code rows still cover <pre>, white-space-preserving editor divs, and GitHub code tables. HR insertion between two rules and top-level HR paste into the middle of a paragraph stay deferred to a future HR/block-void plus block-fragment owner. Focused paste-html browser proof, the full paste-html browser corpus, bun run lint:fix, and bun check passed. Added docs/solutions/best-practices/2026-05-09-lexical-htmlcopy-harvest-rows-need-whitespace-and-empty-target-boundaries.md to preserve the whitespace/empty-target boundary lesson.

The completion checkpoint stays pending because ../lexical/packages/lexical-playground/__tests__/e2e/CopyAndPaste/html/ListsHTMLCopyAndPaste.spec.mjs is the next unresolved clipboard/list source-read candidate unless the prior row-level drill proves it is exhausted.

Ralph execution update on 2026-05-09: lexical-playground ListsHTMLCopyAndPaste.spec.mjs source-read is complete. Exact source-read confirmed the earlier row-level drill exhausted the accepted HTML list parser rows: basic unordered-list import, compact nested ul variants, and nested <div> boundaries inside list items are already applied in .tmp/slate-v2/playwright/integration/examples/paste-html.test.ts. The structural post-fragment selection behavior is covered by .tmp/slate-v2/packages/slate/test/clipboard-contract.ts, not duplicated in the HTML browser row. Lexical checklist private attributes stay rejected as app/list-plugin policy. Top-level HR paste into the middle of a list stays deferred to a future HR/block-void plus list-split owner. Toolbar indent/outdent assertions stay deferred or rejected under explicit list-indent policy. No .tmp/slate-v2 code or test file changed, so no new Slate v2 command was required for this classification-only slice.

The completion checkpoint stays pending because ../lexical/packages/lexical-playground/__tests__/e2e/CopyAndPaste/lexical/ListsCopyAndPaste.spec.mjs is the next unresolved clipboard/list source-read candidate.

Ralph execution update on 2026-05-09: lexical-playground ListsCopyAndPaste.spec.mjs source-read is complete. Exact source-read split the seven native list copy/paste rows from Lexical playground shell and DOM theme assertions. .tmp/slate-v2/packages/slate/test/clipboard-contract.ts adds package proof for the accepted portable rows: a partial list item plus following paragraph copied into an empty editor; the same fragment inserted into an empty list item by splitting the surrounding list; a copied list inserted into selected paragraph text while preserving surrounding text; two copied paragraphs inserted inside a list item with the first paragraph kept in the list and the tail paragraph promoted after the list; and two copied paragraphs inserted at the end of a list as one list item plus a following block. .tmp/slate-v2/packages/slate/src/transforms-text/insert-fragment.ts handles those rows with explicit structural fragment replacement before the generic inline/block unwrapping path. Exact Lexical DOM classes, list value attributes, browser/OS flake tags, native clipboard transport, raw mobile, collaboration, table-model, and issue claims stay out of this package row. Focused list-fragment proof and the full clipboard contract passed after bun run lint:fix, and bun check passed after fixing a TypeScript narrowing bug in the compatibility helper. Added docs/solutions/logic-errors/2026-05-09-native-list-fragment-paste-must-run-before-generic-fragment-unwrapping.md to preserve the structural-fragment-before-generic-unwrapping lesson.

The completion checkpoint stays pending because ../lexical/packages/lexical-playground/__tests__/unit/ImageHTML.test.ts is the next unresolved clipboard/image source-read candidate.

Ralph execution update on 2026-05-09: lexical-playground ImageHTML.test.ts source-read is complete. The portable rows are image HTML import from Lexical's exported shapes: no-caption <p></p> and plain-text caption <figure><figcaption>caption text</figcaption></figure>. .tmp/slate-v2/playwright/integration/examples/paste-html.test.ts now proves both shapes through browser paste/import. The first focused proof failed red because Slate rendered the no-caption image as a block void nested under a paragraph. .tmp/slate-v2/site/examples/ts/paste-html-import.ts now unwraps paragraph wrappers that only contain block elements, so block images stay block images. Caption text imports as a following paragraph. Focused ImageHTML browser proof, the full paste-html browser corpus, bun run lint:fix, and bun check passed. Added docs/solutions/logic-errors/2026-05-09-block-only-html-paragraphs-must-not-wrap-block-voids.md to preserve the block-only paragraph wrapper rule. Exact Lexical alt, height, width, serializer API, caption editor internals, native clipboard transport, raw mobile, collaboration, table-model, and export claims stay out.

The completion checkpoint stays pending because ../lexical/packages/lexical/src/__tests__/unit/HTMLCopyAndPaste.test.ts is the next unresolved core clipboard source-read candidate.

Ralph execution update on 2026-05-09: lexical core HTMLCopyAndPaste.test.ts source-read is complete. The accepted portable rows are core HTML block-shape paste behavior: plain DOM text, malformed paragraph pair, single div, nested spans/divs, nested span in div, and nested div in span. .tmp/slate-v2/playwright/integration/examples/paste-html.test.ts now proves those shapes through the browser paste/import path, and the existing rich HTML paste row now asserts typed text after a strong paste remains inside <strong>. The first focused proof failed red because Slate flattened generic div block boundaries into one paragraph; .tmp/slate-v2/site/examples/ts/paste-html-import.ts now imports generic div nodes as block-boundary fragments while preserving list-item div boundary handling. Google Docs, GitHub, and Joplin checklist rows stay deferred/rejected until a checklist/list-plugin owner accepts checklist schema, ARIA/theme output, and checkbox normalization. The focused core proof passed after the importer fix. Full paste-html browser proof, bun run lint:fix, and bun check passed. Added docs/solutions/logic-errors/2026-05-09-html-div-boundaries-must-survive-inline-wrappers.md to preserve the generic div boundary import rule.

The completion checkpoint stays pending because ../lexical/packages/lexical-playground/__tests__/e2e/CopyAndPaste/lexical/CopyAndPaste.spec.mjs is the next unresolved native copy/paste residual audit.

Ralph execution update on 2026-05-09: lexical-playground native CopyAndPaste.spec.mjs residual audit is complete. The accepted missing row was paragraph fragment paste into an empty quote: the first pasted text block should fill the target quote and the tail paragraph should promote after it. .tmp/slate-v2/packages/slate-dom/test/clipboard-boundary.ts now proves that through model-backed DOM clipboard payloads. The first focused proof failed red because Slate replaced the empty quote with two paragraphs. .tmp/slate-v2/packages/slate/src/transforms-text/insert-fragment.ts now keeps the target empty text block wrapper for the first pasted text block and keeps later text blocks source-shaped. Current owners already cover multi-block copy/paste, full-document replacement, inline link fragments, multiline plain-text splitting, font-size HTML paste, and collapsed-copy/cut clipboard preservation. Copied-heading source-wrapper preservation into an empty paragraph stays deferred to future block-format paste policy because Slate deliberately keeps target wrappers for single text-block fragments. Lexical hashtag/theme output, YouTube decorator embeds, exact plaintext rendering, native transport flake tags, raw mobile, collaboration, table-model, and issue claims stay out. Focused DOM clipboard proof, full DOM clipboard boundary, core clipboard contract, bun run lint:fix, and .tmp/slate-v2 bun check passed. Added docs/solutions/logic-errors/2026-05-09-empty-target-fragment-paste-keeps-first-block-wrapper.md to preserve the target-first-block paste rule.

The completion checkpoint stays pending because ../lexical/packages/lexical-playground/__tests__/e2e/Events.spec.mjs is the next unresolved input-runtime source-read candidate.

Ralph execution update on 2026-05-09: lexical-playground Events.spec.mjs input-runtime source-read is complete. The accepted rows were Mac-style beforeinput substitutions: autocapitalization via native insertText followed by insertReplacementText, and double-space period replacement after emoji via an expanded insertText target range. .tmp/slate-v2/playwright/integration/examples/plaintext.test.ts now proves both through browser-dispatched beforeinput/input events. The first focused proof failed red with only I, then with iSI, then with 🙂 . , exposing three owners. .tmp/slate-v2/packages/slate-react/src/editable/runtime-before-input-events.ts now flushes queued native text repair before the next model-owned beforeinput. .tmp/slate-v2/packages/slate-react/src/editable/mutation-controller.ts now applies provided replacement target ranges even after native repair moves the current selection. .tmp/slate-v2/packages/slate-react/src/editable/selection-reconciler.ts now honors expanded insertText target ranges even when model selection is preferred. Product shell, theme spans, page.pause, OS labels, raw mobile, collaboration, table-model, and issue claims stay out. The focused plaintext browser proof passed after the runtime fixes.

The completion checkpoint stays pending because ../lexical/packages/lexical-playground/__tests__/e2e/Keyboard.spec.mjs is the next unresolved keyboard/input-runtime source-read candidate.

Ralph execution update on 2026-05-09: lexical-playground Keyboard.spec.mjs input-runtime source-read is complete. The file contains one portable row: Mac-style insertTranspose from Control+T swaps adjacent characters. Slate already exposed the hotkey matcher but dropped the beforeinput command. The first focused package proof failed red with abc still unchanged. Slate React now has a first-class transpose-character command parsed from insertTranspose, applied through model-input-strategy.ts and mutation-controller.ts, and covered by both package and plaintext browser proof. Repeated transpose from abc at offset 1 now produces bca with the selection after the swapped pair. Lexical Playground setup, OS/browser flake tags, raw mobile, collaboration, table-model, and issue claims stay out.

The completion checkpoint stays pending because ../lexical/packages/lexical-playground/__tests__/e2e/KeyboardShortcuts.spec.mjs is the next unresolved keyboard-shortcut source-read candidate.

Ralph execution update on 2026-05-09: lexical-playground KeyboardShortcuts.spec.mjs source-read is complete. The accepted portable rows were shortcut access to editor commands already owned by Slate examples: paragraph/heading block conversion, block alignment, clear formatting, and code-block conversion. .tmp/slate-v2/playwright/integration/examples/richtext.test.ts first failed red because heading block shortcuts did nothing, then passed after .tmp/slate-v2/site/examples/ts/richtext.tsx routed block/alignment/clear shortcuts through the same owners as the toolbar commands. .tmp/slate-v2/playwright/integration/examples/code-highlighting.test.ts first failed red because the code-block shortcut did nothing, then passed after .tmp/slate-v2/site/examples/ts/code-highlighting.tsx reused the existing selected-paragraph code-block conversion command for mod+shift+c and mod+alt+c. Existing mark-hotkey rows from the TextFormatting slice cover bold/italic/underline/code marks. Heading three, checklist, product typography, font-size controls, strikethrough/subscript/superscript, paragraph/list/table indent shortcuts, active-state DOM, browser/OS labels, raw mobile, collaboration, table-model, and issue claims stay out. Focused richtext and code-highlighting browser rows passed after the implementation, bun run lint:fix had no fixes, and .tmp/slate-v2 bun check passed.

The completion checkpoint stays pending because ../lexical/packages/lexical-playground/__tests__/e2e/TextEntry.spec.mjs is the next unresolved text-entry/input-runtime source-read candidate.

Ralph execution update on 2026-05-09: lexical-playground TextEntry.spec.mjs source-read is complete. The accepted missing rows were heading-start Enter after markdown heading creation and native Enter at the boundary between plain and marked text. .tmp/slate-v2/playwright/integration/examples/markdown-shortcuts.test.ts now proves a heading created by # gets an empty paragraph before it when Enter is pressed at the heading start, with selection staying at the heading start. .tmp/slate-v2/site/examples/ts/markdown-shortcuts.tsx handles that heading-start split by turning the split empty heading into a paragraph. .tmp/slate-v2/playwright/integration/examples/richtext.test.ts now proves native Enter between plain text and bold text creates a plain paragraph followed by a bold paragraph, with selection at the start of the bold text. Existing plaintext/richtext input and generated stress owners cover basic typing, select-all replacement, partial replacement, character delete, word delete, and selection synchronization. Lexical emoji/theme output, trimmable-space whitespace policy, soft-break/newline-node navigation, browser/OS labels, raw mobile, collaboration, table-model, and issue claims stay rejected or deferred. The first markdown-shortcuts focused proof failed red with no paragraph before the heading; after the example keydown fix it passed. The richtext marked-boundary proof passed without runtime changes. bun run lint:fix, the focused richtext proof, the focused markdown-shortcuts proof, and bun check passed. One parallel focused-proof rerun hit a Next build lock; the same markdown proof passed when rerun sequentially. Added docs/solutions/logic-errors/2026-05-09-heading-start-enter-must-normalize-split-block-type.md and validated its YAML frontmatter.

The completion checkpoint stays pending because ../lexical/packages/lexical-playground/__tests__/e2e/SpecialTexts.spec.mjs is the next unresolved product-adjacent source-read candidate.

Ralph execution update on 2026-05-09: lexical-playground SpecialTexts.spec.mjs source-read is complete. The bracket-triggered highlighting rows are product transform/theme policy: typing [MLH Fellowship] creates themed special-text spans, strips brackets, and inserts Lexical Playground DOM output. Slate should not copy that plugin behavior into core coverage. The disabled-option row keeps [MLH Fellowship] as literal text, which is ordinary text insertion already covered by existing Slate input owners. No .tmp/slate-v2 code/test edit was needed.

The completion checkpoint stays pending because ../lexical/packages/lexical-playground/__tests__/e2e/Placeholder.spec.mjs is the next unresolved placeholder/browser source-read candidate.

Ralph execution update on 2026-05-09: lexical-playground Placeholder.spec.mjs source-read is complete. The accepted portable row is empty-editor placeholder behavior: a placeholder is visible, the model remains empty, and focus lands selection at the start. .tmp/slate-v2/playwright/integration/examples/placeholder.test.ts now proves the Slate equivalent on the custom-placeholder example: placeholder visible, modelText === "", and selection at [0, 0] offset 0. Existing .tmp/slate-v2/packages/slate-react/test/rendered-dom-shape-contract.tsx owns the empty-block line-break placeholder DOM shape. Lexical's rich/plain/collab placeholder strings are product shell text and stay rejected. The focused placeholder browser proof passed, bun run lint:fix had no fixes, and .tmp/slate-v2 bun check passed.

The completion checkpoint stays pending because ../lexical/packages/lexical-playground/__tests__/e2e/Images.spec.mjs is the next unresolved image void/paste source-read candidate.

Ralph execution update on 2026-05-09: lexical-playground Images.spec.mjs source-read is complete. The accepted missing row was keyboard deletion of a clicked selected image void. .tmp/slate-v2/playwright/integration/examples/images.test.ts now proves a clicked image selects [1, 0], Backspace removes that image, and the editor keeps the remaining image. Existing image browser rows already cover horizontal movement into/out of image voids, vertical movement, shifted selection into images, selected-image Enter, and void spacer rendering. Existing paste-html rows cover Lexical image HTML with optional captions. URL/upload dialogs, image dimensions, drag/drop product policy, caption editor UI, multi-node NodeSelection replacement, raw mobile, collaboration, table-model, and issue claims stay rejected or deferred. The focused image browser proof passed, bun run lint:fix had no fixes, and .tmp/slate-v2 bun check passed.

The completion checkpoint stays pending because ../lexical/packages/lexical-playground/__tests__/e2e/Tab.spec.mjs is the next unresolved tab/code/list/clipboard source-read candidate.

Ralph execution update on 2026-05-09: lexical-playground Tab.spec.mjs source-read is complete. The accepted portable rows are already owned: code block Tab behavior is covered by the current code-highlighting browser proof, and literal tab preservation is covered by the earlier plain-text fallback and Google Docs Apple-tab-span paste proofs. Lexical's paragraph Tab plus Chromium CDP IME row mixes product paragraph-indent policy, TabNode DOM output, and IME runtime behavior, so it stays deferred to explicit keyboard/IME/paragraph-indent owners. Ctrl/Meta+ArrowLeft after a typed Tab stays deferred to a future text-boundary/keyboard owner if Slate accepts literal-tab keyboard insertion as editor-owned behavior. No .tmp/slate-v2 code/test edit was needed.

The completion checkpoint stays pending because ../lexical/packages/lexical-playground/__tests__/regression/1055-fast-typing-undo.spec.mjs is the next unresolved history/input source-read candidate.

Ralph execution update on 2026-05-09: lexical-playground regression 1055-fast-typing-undo.spec.mjs source-read is complete. The portable row is history stack law: after undo, the next user edit must create a fresh undoable batch and clear redo. .tmp/slate-v2/packages/slate-history/test/history-contract.ts already owned redo invalidation after a new edit follows undo; that row is now strengthened to undo the follow-up edit back to the pre-edit state as well. Lexical Playground HTML output and helper setup stay rejected as harness details. The focused history-contract grep passed, and the full history-contract.ts file passed. bun run lint:fix had no fixes, and .tmp/slate-v2 bun check passed.

The completion checkpoint stays pending because ../lexical/packages/lexical-playground/__tests__/regression/1083-backspace-with-element-at-front.spec.mjs is the next unresolved delete/selection source-read candidate.

Ralph execution update on 2026-05-09: lexical-playground regression 1083-backspace-with-element-at-front.spec.mjs source-read is complete. The portable row is expanded delete over inline content at the front of the selected range. Slate normalizes a spacer before leading inline content, so the full selection proof selects from that spacer through the trailing text. A second proof selects from an inline link after preceding text through the trailing text. .tmp/slate-v2/packages/slate/src/transforms-text/delete-text.ts now collects fully-selected inline ancestors and removes them as nodes, so expanded delete no longer leaves empty inline shells or invalid inline child points. .tmp/slate-v2/packages/slate/test/delete-contract.ts covers both rows. The focused inline-element delete grep passed, and the full delete-contract.ts file passed. The first full bun check exposed an over-broad version of the fix that removed inline shells for selections wholly inside an inline; the final fix removes a fully-selected inline ancestor only when the expanded selection crosses outside that inline. The delete fixture suite, bun run lint:fix, and .tmp/slate-v2 bun check passed. Added docs/solutions/logic-errors/2026-05-09-expanded-delete-across-inline-boundaries-must-remove-crossed-inline-ancestors.md and validated its YAML frontmatter.

The completion checkpoint stays pending because ../lexical/packages/lexical-playground/__tests__/regression/1113-link-newline-at-end.spec.mjs is the next unresolved link/end-boundary source-read candidate.

Ralph execution update on 2026-05-09: lexical-playground regression 1113-link-newline-at-end.spec.mjs source-read is complete. The portable row is link end-boundary selection after inserting a new line. A new browser proof in .tmp/slate-v2/playwright/integration/examples/inlines.test.ts failed red because Enter after a typed URL inline link produced a second empty link at the start of the next paragraph. .tmp/slate-v2/site/examples/ts/inlines.tsx now moves one offset out of a collapsed inserted link, matching the inline boundary policy used by the mentions example. The proof now locks exactly one typed URL link, selection in the next block, and follow-up text outside the link. Focused proof, full inlines browser proof, bun run lint:fix, and .tmp/slate-v2 bun check passed. Added docs/solutions/logic-errors/2026-05-09-collapsed-inline-link-insertion-must-move-selection-outside.md and validated its YAML frontmatter. No generic autolink API, raw mobile, collaboration, table-model, or issue claim changed.

The completion checkpoint stays pending because ../lexical/packages/lexical-playground/__tests__/regression/1384-insert-nodes.spec.mjs is the next unresolved insert-nodes source-read candidate.

Ralph execution update on 2026-05-09: lexical-playground regression 1384-insert-nodes.spec.mjs source-read is complete. The filename is misleading: the portable row is code-block clipboard insertion, where copying a nested code-line fragment and pasting inside an active code line must merge into that line instead of inserting a sibling or top-level code block. A new package contract in .tmp/slate-v2/packages/slate/test/clipboard-contract.ts failed red, then .tmp/slate-v2/packages/slate/src/transforms-text/insert-fragment.ts was updated to fit a single nested text-block child from the same structural container into the active nested text block. A new browser proof in .tmp/slate-v2/playwright/integration/examples/code-highlighting.test.ts locks the native copy/paste behavior. Focused package proof, full clipboard contract, focused browser proof, full code-highlighting browser proof, bun run lint:fix, and .tmp/slate-v2 bun check passed. Added docs/solutions/logic-errors/2026-05-09-nested-clipboard-fragments-must-merge-into-active-text-block.md and validated its YAML frontmatter. No Lexical Playground setup, exact gutter/theme output, raw mobile, collaboration, table-model, or issue claim changed.

The completion checkpoint stays pending because ../lexical/packages/lexical-playground/__tests__/regression/221-editing-hashtags.spec.mjs is the next unresolved inline-token editing source-read candidate.

Ralph execution update on 2026-05-09: lexical-playground regression 221-editing-hashtags.spec.mjs source-read is complete. The portable row is dynamic decorated text-token editing, not mention, mark, inline-void, hashtag plugin, or Lexical theme behavior. .tmp/slate-v2/site/examples/ts/highlighted-text.tsx now adds hashtag-style projections over normal text, and .tmp/slate-v2/playwright/integration/examples/highlighted-text.test.ts proves Space splitting #yolo to decorated #yo, Delete removing the plain trailing space at the boundary, and Backspace from the plain tail preserving decorated #yol. The first focused proof failed red because no hashtag projection rendered. A full-document Backspace setup also exposed a separate projected-document replacement edge, so the final proof uses model-owned token setup and browser keyboard behavior for the accepted invariant. Focused hashtag proof, full highlighted-text browser proof, bun run lint:fix, and .tmp/slate-v2 bun check passed. Added docs/solutions/best-practices/2026-05-09-hashtag-like-text-tokens-should-use-dynamic-decorations.md and validated its YAML frontmatter. No raw mobile, collaboration, table-model, or issue claim changed.

The completion checkpoint stays pending because ../lexical/packages/lexical-playground/__tests__/regression/230-navigation-around-hashtags.spec.mjs is the next unresolved inline-token navigation source-read candidate.

Ralph execution update on 2026-05-09: lexical-playground regression 230-navigation-around-hashtags.spec.mjs source-read is complete. The portable row is dynamic decorated text-token navigation after transient prefix insertion and deletion, not a Lexical hashtag plugin or DOM theme contract. .tmp/slate-v2/playwright/integration/examples/highlighted-text.test.ts now proves that after replacing the document with #foo, inserting a before it, deleting that prefix, then pressing ArrowRight moves the collapsed selection into the decorated hashtag-style token at offset 1 through the model-owned movement path. Focused browser proof, full highlighted-text browser proof, bun run lint:fix, and .tmp/slate-v2 bun check passed. No runtime code, raw mobile, collaboration, table-model, or issue claim changed.

The completion checkpoint stays pending because ../lexical/packages/lexical-playground/__tests__/regression/3136-insert-nodes-adjacent-to-inline.spec.mjs is the next unresolved inline-adjacent rich paste/replacement source-read candidate.

Ralph execution update on 2026-05-09: lexical-playground regression 3136-insert-nodes-adjacent-to-inline.spec.mjs source-read is complete. The portable rows are rich paste/replacement of selected plain text immediately before and immediately after an inline link. .tmp/slate-v2/packages/slate/test/clipboard-contract.ts now proves both core fragment replacement directions, and .tmp/slate-v2/playwright/integration/examples/inlines.test.ts proves rich HTML paste replaces adjacent selected text without expanding or swallowing the link. Focused package proof, full clipboard contract, focused browser proof, full inlines browser proof, bun run lint:fix, and .tmp/slate-v2 bun check passed. No runtime code, raw mobile, collaboration, table-model, or issue claim changed.

The completion checkpoint stays pending because ../lexical/packages/lexical-playground/__tests__/regression/379-backspace-with-mentions.spec.mjs is the next unresolved mention/text-token Backspace source-read candidate.

Ralph execution update on 2026-05-09: lexical-playground regression 379-backspace-with-mentions.spec.mjs source-read is complete. The portable row is Backspace at a line boundary before a mention/text token, not Lexical Playground typeahead setup, mention styling, exact DOM/theme output, raw mobile, collaboration, table-model, or issue behavior. .tmp/slate-v2/playwright/integration/examples/mentions.test.ts now proves that splitting immediately before a leading inline mention and pressing Backspace at that new line boundary preserves both mention atoms and restores the boundary selection. Focused browser proof, full mentions Chromium proof, bun run lint:fix, and .tmp/slate-v2 bun check passed. No runtime code, raw mobile, collaboration, table-model, or issue claim changed.

../lexical/packages/lexical-playground/__tests__/regression/399-open-line.spec.mjs is source-read/applied. The portable row is Mac Ctrl+O open-line behavior: open an empty line before the current line without moving selection past the following text. Slate now has a dedicated open-line insert-break variant wired through slate-dom hotkeys and the slate-react keyboard/mutation runtime. The runtime inserts an empty paragraph before the current block directly, instead of reusing normal split-block and repairing selection afterward. Browser proof asserts foo, empty line, bar, selection on the empty line, and open-line kernel command metadata.

Verification for the 399 slice:

• cd .tmp/slate-v2 && PLAYWRIGHT_RETRIES=0 bunx playwright test playwright/integration/examples/richtext.test.ts --project=chromium -g "opens a line with Mac Ctrl\\+O" passed.
• cd .tmp/slate-v2 && bun test ./packages/slate-dom/test/hotkeys.ts passed.
• cd .tmp/slate-v2 && bun test ./packages/slate-react/test/editing-kernel-contract.ts passed.
• cd .tmp/slate-v2 && bun run lint:fix && bun check passed.

The full richtext Chromium owner sweep was attempted and showed one deterministic unrelated IME mark expectation mismatch in commits IME composition through an active mark before a formatted sibling: the rendered DOM has sibling em, code, and em strong nodes, while the test expects em code. That is not part of the 399 open-line owner and stays out of this slice.

The completion checkpoint stays pending because ../lexical/packages/lexical-playground/__tests__/regression/429-swapping-emoji.spec.mjs is the next unresolved emoji-line Enter/Backspace source-read candidate.

Ralph execution update on 2026-05-09: lexical-playground regression 429-swapping-emoji.spec.mjs source-read is complete. The portable row is emoji-line Enter/Backspace behavior: Enter at the start of an emoji line opens an empty block before it, keeps selection at the emoji-line start, and Backspace rejoins without corrupting emoji text or selection. Lexical Playground emoji substitution UI, exact emoji span/theme DOM, flaky tagging, raw mobile, collaboration, table-model, and issue behavior stay out. .tmp/slate-v2/playwright/integration/examples/richtext.test.ts now owns the plain Unicode emoji browser proof. Focused browser proof and bun run lint:fix && bun check passed from .tmp/slate-v2.

The completion checkpoint stays pending because the next runnable non-table row is ../lexical/packages/lexical-playground/__tests__/regression/5251-paste-into-inline-element.spec.mjs.

Ralph execution update on 2026-05-09: lexical-playground regression 5251-paste-into-inline-element.spec.mjs source-read is complete. The portable row is selected inline-link text replacement by a rich clipboard payload: pasted content must stay outside the surviving link tail. Slate now has package proof for rich fragment insertion, DOM clipboard fallback proof for HTML-with-plain-text payloads, and an inlines browser proof for user-visible paste behavior. insertFragment now handles selected one-level inline text with a single replacement operation, and DOM plain-text fallback routes expanded same-inline selections through fragment insertion instead of inheriting the inline. Focused red browser proof initially failed with replaced inside the link; the final focused browser proof, full inlines Chromium file, clipboard-contract, clipboard-boundary, bun run lint:fix, and .tmp/slate-v2 bun check passed. Added docs/solutions/logic-errors/2026-05-09-html-clipboard-fallback-must-not-inherit-selected-inline-text.md and validated its YAML frontmatter. Lexical Playground link toolbar setup, exact DOM/theme output, raw mobile, collaboration, table-model, and issue behavior stay out.

The completion checkpoint stays pending because the next runnable non-table row is ../lexical/packages/lexical-playground/__tests__/regression/5583-select-list-followed-by-element-node.spec.mjs.

Ralph execution update on 2026-05-09: lexical-playground regression 5583-select-list-followed-by-element-node.spec.mjs source-read is complete. The portable row is native selection import over list content followed by a non-text/boundary element without crashing. Slate already had adjacent block-void delete proof and a generic native drag-selection boundary row, but not the list + boundary combination, so .tmp/slate-v2/playwright/integration/examples/dom-coverage-boundaries.test.ts now proves a native drag from a materialized list item to a boundary placeholder imports into a valid Slate range with no page errors. Focused browser proof, full dom-coverage Chromium proof, bun run lint:fix, and .tmp/slate-v2 bun check passed. Lexical Playground drag helpers, horizontal-rule product shell, exact DOM/theme output, raw mobile/device, collaboration, table-model, and issue behavior stay out.

The completion checkpoint stays pending because the next runnable non-table row is ../lexical/packages/lexical-playground/__tests__/regression/7635-SELECTION_INSERT_CLIPBOARD_NODES_COMMAND.spec.mjs.

Ralph execution update on 2026-05-09: lexical-playground regression 7635-SELECTION_INSERT_CLIPBOARD_NODES_COMMAND.spec.mjs source-read is complete. The portable row is rich HTML clipboard insertion inside a nested editable island plus restoring parent-editor use afterward. Slate now proves that rich HTML paste inside the nested rich-text editor preserves bold text, does not leak into the parent value, preserves the parent selection, and still allows parent-editor typing after nested editing. site/examples/ts/richtext.tsx now owns a narrow HTML paste capability that reuses the shared deserializer but normalizes imported content to the rich-text example's supported blocks and marks. The selection runtime now rejects DOM selections and beforeinput target ranges whose endpoints cross from a parent editor into a nested editor before they can replace parent content.

Verification for the 7635 slice:

• cd .tmp/slate-v2 && PLAYWRIGHT_RETRIES=0 bunx playwright test playwright/integration/examples/editable-voids.test.ts --project=chromium -g "rich HTML inside nested editor" first failed red because the nested editor inserted plain text and dropped <strong>.
• cd .tmp/slate-v2 && PLAYWRIGHT_RETRIES=0 bunx playwright test playwright/integration/examples/editable-voids.test.ts --project=chromium -g "parent selection that crosses|rich HTML inside nested editor" passed after tightening parent/nested selection import.
• cd .tmp/slate-v2 && PLAYWRIGHT_RETRIES=0 bunx playwright test playwright/integration/examples/editable-voids.test.ts --project=chromium passed: 12 tests.
• cd .tmp/slate-v2 && bun run lint:fix && bun check passed: Biome formatted touched files, package/site/root typecheck passed, Bun tests passed, and slate-react Vitest passed.
• ruby -e 'require "yaml"; YAML.load_file("docs/solutions/logic-errors/2026-05-09-parent-nested-dom-selections-must-not-import-as-parent-ranges.md")' passed for the solution note.

Lexical image caption UI, SELECTION_INSERT_CLIPBOARD_NODES_COMMAND API, Playground command harness, exact DOM/theme output, raw mobile/device, collaboration, table-model, and issue behavior stay out.

The completion checkpoint stays pending because the next runnable non-table row is ../lexical/packages/lexical-playground/__tests__/e2e/AutoLinks.spec.mjs.

Ralph execution update on 2026-05-09: lexical-playground e2e AutoLinks.spec.mjs source-read is complete. The portable Slate-owned row is not Lexical's full autolink plugin. Slate now proves the narrow accepted behavior in .tmp/slate-v2/playwright/integration/examples/inlines.test.ts: pasting a single URL at a collapsed selection creates one safe inline link, places the caret outside the link, and keeps follow-up typing outside the link. Focused pasted-URL browser proof, full inlines Chromium proof, bun run lint:fix, and .tmp/slate-v2 bun check passed. Lexical's broad URL/email parser matrix, delimiter tokenization, invalid URL grammar, unlink/relink UI, unlinked-autolink preservation, emoji shortcode destruction, styling/font rows, toolbar/product shell, exact DOM/theme output, raw mobile, collaboration, table-model, and issue behavior stay deferred or rejected under future link/autolink plugin owners.

The completion checkpoint stays pending because the next runnable non-table row is ../lexical/packages/lexical-playground/__tests__/e2e/Hashtags.spec.mjs.

Ralph execution update on 2026-05-09: lexical-playground e2e Hashtags.spec.mjs source-read is complete. The portable Slate-owned rows are dynamic decorated text-token editing, not Lexical's hashtag plugin grammar. Existing highlighted-text rows already covered hashtag-style decoration, Space splitting, Delete/Backspace boundaries, transient-prefix navigation, decorated-range copy/cut, and decorated-text IME/editing. The new proof in .tmp/slate-v2/playwright/integration/examples/highlighted-text.test.ts covers the missing delimiter/marker update: deleting the delimiter between #hello and world expands the decorated token, restoring the delimiter shrinks it back, and deleting the leading # drops the decoration with stable caret placement. Focused delimiter proof, full highlighted-text Chromium proof, bun run lint:fix, and .tmp/slate-v2 bun check passed. Lexical's hashtag plugin grammar, invalid-match matrix, markdown import/export shell, format inheritance styling, exact DOM/theme output, raw mobile, collaboration, table-model, and issue behavior stay deferred or rejected.

The completion checkpoint stays pending because the next runnable non-table row is ../lexical/packages/lexical-playground/__tests__/e2e/Keywords.spec.mjs.

Ralph execution update on 2026-05-09: lexical-playground e2e Keywords.spec.mjs source-read is complete with no Slate code change. The raw portable rows are already covered by the highlighted-text owner: projected text selection/movement/editing, mark-click typing through projections, decorated IME, copy/cut semantics, hashtag-style delimiter merge/split, and deleting a decorated selected range. Lexical's exact congrats grammar, bracket/team token policy, keyword styling, product plugin setup, exact DOM/theme output, raw mobile, collaboration, table-model, and issue behavior stay deferred or rejected. No new .tmp/slate-v2 verification was needed for this slice because no code or test file changed.

The completion checkpoint stays pending because the next runnable non-table row is ../lexical/packages/lexical-playground/__tests__/e2e/SelectionAlwaysOnDisplay.spec.mjs.

Ralph execution update on 2026-05-09: lexical-playground e2e SelectionAlwaysOnDisplay.spec.mjs source-read is complete with no Slate code change. The only raw Slate-owned invariant is blurred-editor model-selection persistence, and current .tmp/slate-v2/playwright/integration/examples/richtext.test.ts already proves it in keeps model selection when focus moves outside the editor. Lexical's always-on fake-selection overlay, highlight alignment tolerance, plain-text skip policy, exact DOM/theme output, raw mobile/device, collaboration, table-model, and issue behavior stay deferred or rejected as product visualization or future dedicated owners. The remaining raw mobile and collaboration buckets are explicit future-owner deferrals, so the selected Lexical non-table execution lane is closed.