Editor Node DX Research

Status

Done.

Goal

Produce a full research-wiki pass for the absolute-best editor node/text/mark DX and runtime architecture, unconstrained by legacy Slate closeness, while preserving React/shadcn flexibility and avoiding perf regressions.

Scope

Primary repo: /Users/zbeyens/git/plate-2.
Local comparison repos/raw:
- /Users/zbeyens/git/prosemirror
- /Users/zbeyens/git/lexical
- /Users/zbeyens/git/tiptap
- /Users/zbeyens/git/raw/prosemirror
- /Users/zbeyens/git/raw/lexical
- /Users/zbeyens/git/raw/tiptap
Output: updated docs/research compiled layer plus concise architecture verdict.

Plan

Read research-wiki baseline and existing editor architecture pages.
Inspect official/local source entrypoints for ProseMirror, Lexical, and Tiptap.
Read strongest raw/source hits for node specs, text/mark APIs, render surfaces, NodeViews/decorator nodes, extension APIs, and React perf guidance.
Create a per-corpus evidence ledger.
Promote durable synthesis into source/system/decision research pages.
Update docs/research/index.md, docs/research/log.md, and completion state.

Early Findings

Existing research already accepts the broad direction: Lexical lifecycle and dirtiness, ProseMirror transaction/DOM discipline, Tiptap extension DX.
This pass narrows the question to node/text/mark/render DX and should avoid a fake "copy Tiptap" or "copy Lexical" answer.
ProseMirror has the best declarative node/mark schema vocabulary and mapped decoration model, but NodeViews are too imperative as a default React API.
Lexical has the best runtime dirtiness and text-mode model, but class nodes and subclass replacement are too expensive as public DX.
Tiptap has the best extension/product packaging and React selector posture, but React NodeViews are explicitly wrapper-heavy and potentially expensive.
The accepted architecture is spec-first nodes/marks/text behavior with runtime-owned DOM shells and app-owned visible React renderers.

Verification

bun run completion-check