Slate v2 overlay architecture beats legacy Slate and aligns with the best parts of the field

Question

Why is the current Slate v2 overlay architecture better than legacy Slate, and how should it be judged against the editor-architecture candidate field?

Decision

Accept the following read:

• Slate v2 is clearly superior to legacy Slate on overlays.
• It is not “better than every editor” in the field.
• It is better described as:
  • escaping legacy Slate’s callback soup
  • converging on the strongest overlay ideas from ProseMirror, Lexical, Tiptap, VS Code, and TanStack DB
  • keeping a more React-native and app-owned integration story than those systems where that is useful

Why it beats legacy Slate

Legacy Slate made one decorate callback pretend it could own:

• transient inline highlighting
• durable range-backed comments
• anchored widget UI
• browser-facing composition decoration
• large-document invalidation

That mixed incompatible ownership models into one surface.

Slate v2 fixes that by splitting the system into:

• Decoration transient, overlap-friendly, mapped or externally indexed
• Annotation durable, id-bearing, bookmark-backed
• Widget anchored UI, geometry-derived, narrower public surface

That win is not cosmetic. It gives the system:

• explicit ownership instead of one callback owning everything
• Bookmark as the public durable-anchor story instead of RangeRef leakage
• store/controller-style annotation integration instead of render-time arrays
• widget UI that does not abuse text-decoration semantics
• corridor-scoped large-document overlay invalidation
• explicit browser-truth hardening on selection, clipboard, and IME boundaries

Relative to the field

ProseMirror

Slate v2 now aligns with ProseMirror on the important overlay ideas:

• mapped overlay discipline
• durable anchor/bookmark semantics
• explicit separation between document meaning and UI chrome

It still does not beat ProseMirror as the most disciplined general document engine. That is not the right claim.

Lexical

Slate v2 now aligns with Lexical on:

• split ownership between text overlays, durable anchor state, and decorator UI
• narrow invalidation and runtime-local subscription thinking

Lexical is still the stronger end-to-end runtime challenger. Slate v2 is better read as “finally using the same good ideas” than “beating Lexical outright.”

Tiptap

Slate v2 overlay architecture still beats legacy Slate's decorate story. Do not read that as a current large-document typing superiority claim: fresh 5000/10000-block shell-island comparisons lose steady typing/select lanes to legacy chunking-on. Tiptap is still the productization and packaging benchmark.

The useful lesson from Tiptap is:

• comments, suggestions, and floating UI are distinct product lanes

not:

• copy the entire Tiptap stack

Premirror + Pretext

Slate v2 does not solve page-layout composition the way Premirror + Pretext are trying to. That is a different lane.

What Slate v2 does get right is:

• keep overlay/UI semantics separate from layout composition

VS Code + TanStack DB

Slate v2 now agrees with the strongest non-editor lessons too:

• VS Code: typed visual channels beat one generic overlay bucket
• TanStack DB: annotation metadata wants stable store/controller semantics

What this means in practice

The honest superiority claim is:

• better than legacy Slate for overlay architecture by a wide margin
• aligned with the best overlay semantics in the field
• not currently proven faster than legacy chunking-on for large-document typing
• still not the single best engine on every axis

That is a stronger and more defensible claim than pretending the field does not exist.