Slate v2 Clipboard / Fragment Serialization Ralplan

Verdict

Do not rewrite the clipboard architecture.

The current split is the right long-term shape:

slate:
  fragment extraction, fragment insertion, selection/resulting model shape

slate-dom:
  DataTransfer, internal MIME keys, embedded HTML fragment fallback,
  malformed payload rejection, plain-text fallback, DOM coverage policy

slate-react:
  copy/cut/paste/drop event routing, DOM coverage materialization,
  model-owned caret repair

apps:
  product HTML/image/table/link paste policy through typed clipboard handlers

What changed in the Ralph execution: #4806 is now justified as fixed by browser clipboard proof. #4802 still stays Improves because the proof covers an external contenteditable target, not a named third-party editor. #4056 still stays Improves, but its benchmark owner is repaired and fresh again.

Intent And Boundary

Intent:

• confirm the best architecture for clipboard and Slate fragment transport;
• keep rich text standard support app/extensibility-owned instead of baking a product HTML parser into raw Slate;
• turn the remaining proof gaps into a Ralph-ready execution lane;
• avoid promoting issue claims from old artifacts or headless tests alone.

In scope:

• .tmp/slate-v2/packages/slate/src/transforms-text/insert-fragment.ts
• .tmp/slate-v2/packages/slate/test/clipboard-contract.ts
• .tmp/slate-v2/packages/slate-dom/src/plugin/dom-clipboard-runtime.ts
• .tmp/slate-v2/packages/slate-dom/test/clipboard-boundary.ts
• .tmp/slate-v2/packages/slate-react/src/editable/clipboard-input-strategy.ts
• focused React/Vitest DOM coverage tests when DOM strategy affects copy/paste
• focused Playwright/browser proof only for browser-visible clipboard claims
• issue ledgers and PR claim text only when proof changes

Out of scope:

• no generic Markdown/HTML/table serializer in raw slate;
• no exact inter-editor closure for #4802 without inspecting browser clipboard output;
• no exact full-book/browser closure for #4056 from a headless benchmark;
• no public raw editor.clipboard namespace;
• no broad fragment non-merging policy claim for #3155.

Current State

Source-backed facts:

• slate owns the model side. insert-fragment.ts fits single text-block fragments into active text blocks, preserves multi-block separation in the middle of paragraphs, handles top-level block replacement, nested text-block replacement, and structural block insertion through one logical operation where the fast paths apply.
• slate-dom owns the trust boundary. dom-clipboard-runtime.ts writes application/${clipboardFormatKey}, text/html, and text/plain; it also accepts keyed embedded HTML fragments, rejects malformed internal payloads, and falls back to plain text.
• slate-react is correctly thin. clipboard-input-strategy.ts prevents default where needed, materializes DOM coverage boundaries, calls editor.api.clipboard.writeSelection / insertData, and requests caret repair after model-owned cut/paste/drop.
• The public API shape is acceptable: low-level clipboard remains editor.api.clipboard, and app-specific paste behavior uses extension capability clipboard.insertData(data, { editor, next }).

Fresh verification from this pass:

cd .tmp/slate-v2
bun test ./packages/slate/test/clipboard-contract.ts \
  ./packages/slate-dom/test/clipboard-boundary.ts \
  ./packages/slate-react/test/dom-coverage-native-bridge-contract.test.ts
# 56 pass, 0 fail

cd .tmp/slate-v2/packages/slate-react
bun test:vitest test/dom-strategy-and-scroll.test.tsx \
  test/dom-coverage-native-bridge-contract.test.ts
# 2 files passed, 47 tests passed

Fresh Ralph execution proof:

cd .tmp/slate-v2
bun run bench:core:clipboard-large-payload:local
# passed with issueTargetsEnabled: false

SLATE_CLIPBOARD_BENCH_ISSUE_TARGETS=1 bun run bench:core:clipboard-large-payload:local
# passed with issueTargetsEnabled: true

PLAYWRIGHT_RETRIES=0 bunx playwright test \
  playwright/integration/examples/mentions.test.ts \
  --project=chromium --workers=1 --grep "mention"
# 13 passed

cd packages/slate-react
bun test:vitest test/dom-strategy-and-scroll.test.tsx \
  test/dom-coverage-native-bridge-contract.test.ts
# 2 files passed, 47 tests passed

The refreshed issue-target benchmark records 49.35ms for copying 10,000 populated blocks and 235.22ms for pasting 10,000 plaintext lines into a 10,000-block populated editor, each through one logical operation.

Issue Accounting

Ledger note: docs/slate-issues/gitcrawl-recluster-map.md still says this singleton sweep needs exact proof. This pass confirms that is still true for #4802, #4806, and #4056; #5089 already has a later fixed row in the fork dossier and coverage matrix.

Architecture Decision

Keep these boundaries:

• Editor.getFragment / Editor.insertFragment are model contracts.
• editor.api.clipboard is a DOM-host capability installed by dom() / withReact, not a raw core namespace.
• clipboard.insertData extension handlers are input hooks. They can own product HTML/image/table policy, but they do not own internal Slate fragment serialization.
• DOM coverage copy/paste is part of the clipboard boundary because staged, hidden, and future virtualized regions still need model-backed clipboard data.
• Plain text full-document or empty-block paste can keep optimized fast paths, but rich fragments must stay on the normal Slate fragment path.

Rejected:

• raw Slate importing arbitrary text/html;
• DOM clipboard code deciding schema-specific rich text policy;
• React event handlers containing serialization semantics;
• issue closure from old benchmark JSON when the benchmark command no longer runs against current source.

Scorecard

Total: 0.90.

This is high enough to call the lane closed without redesigning the architecture.

Ralph Execution Result

Completed:

• repaired .tmp/slate-v2/scripts/benchmarks/core/current/clipboard-large-payload.mjs to use current dom() extension installation;
• added Chromium browser proof in .tmp/slate-v2/playwright/integration/examples/mentions.test.ts;
• verified selected inline void native clipboard text/html, text/plain, and embedded Slate fragment;
• verified no FEFF or neighboring text leakage;
• verified external contenteditable paste visibility for #4802 evidence;
• verified Slate round-trip paste plus cut with one mention remove_node and model-owned caret repair for #4806;
• refreshed the #4056 issue-target benchmark artifact;
• synced coverage matrix, fork dossier, PR reference, gitcrawl sync ledger, and recluster map.

Pass State

Completion Gates

This lane is done because:

• the benchmark command runs against current dom() source;
• inline void browser clipboard payload proof exists for #4802 / #4806;
• focused package, DOM, React, Chromium, lint, and root typecheck gates pass;
• #4806 promotion and #4056 evidence refresh are synced across coverage matrix, fork dossier, PR reference, gitcrawl sync ledger, and recluster map.