Slate v2 Release-Readiness Decision

Purpose

Canonical live verdict for /Users/zbeyens/git/slate-v2.

It answers:

• what is honest to recommend now?
• what is still not honest to claim?

Use it with:

• replacement-family-ledger.md
• replacement-gates-scoreboard.md
• release-file-review-ledger.md
• true-slate-rc-proof-ledger.md
• replacement-candidate.md
• live-shape-register.md

Decision

Live Verdict Terms

• Target A
  • release-grade confidence for the proved default surface:
    • Slate
    • EditableBlocks
    • withHistory(createEditor())
• Target B
  • honest blanket replacement for legacy Slate as a whole

Roadmap Mapping

• completed POC RC
  • roadmap name for the work that now supports Target A
• True Slate RC
  • roadmap name for the broader destination that the live verdict still calls Target B

Verdict

• Target A: Go
• Target B: Reopened under challenge

Post-RC Browser / Input Follow-up Lane

As of 2026-04-15, browser/input parity is deferred to post-RC follow-up.

Reason:

• legacy Slate React previously relied on behavior-bearing weird-input patches in Android input-manager, restore-dom, focus/selection recovery, and browser-specific composition handling
• current slate-v2 has real Chromium proof on several IME rows, but that is still not enough to honestly support the broader zero-regression claim
• the remaining unresolved slice is already mostly external evidence: Android keyboard features plus broader iOS Safari composition/focus
• that follow-up work is no longer treated as a blocker for the current RC recommendation surface
• the approved proof authority for this lane is now true-slate-rc-proof-ledger.md, with release-file-review-ledger.md owning the legacy-file closure read
• execution order for the post-RC follow-up lane now lives in 2026-04-12-slate-v2-zero-regression-parity-reconsolidated-plan.md

Hard rule:

• do not claim full zero-regression versus legacy Slate React on every relied-on IME/mobile/browser weird-input surface until the behavior/parity ledger closes or the broader claim is explicitly narrowed

Current narrow follow-up read:

• local Firefox/browser parity is exhausted
• local Android structural/browser parity is exhausted
• the remaining unresolved slice is external Android keyboard-feature behavior plus broader iOS Safari composition/focus
• current execution plans for those rows:
  • 2026-04-12-android-keyboard-feature-external-evidence-plan.md
  • 2026-04-12-ios-safari-broader-composition-focus-external-evidence-plan.md

Broader Replacement Claim Blockers

Current broader blocker read:

• exhaustive API / public-surface contract-width audit is now unresolved for the stronger replacement claim
• direct legacy fixture proof now closes Editor.nodes/** and the non-Editor interfaces/** namespaces more honestly than the earlier family buckets did
• the direct transform audit is green again, so the transform lane is no longer a blanket reopened family
• the transform matrix still has real non-green rows:
  • explicit-skip rows
• browser/input zero-regression parity is deferred to post-RC follow-up for the broader blanket replacement claim
• overlay architecture is no longer the blocker: the repo now has a final first-class design for decoration sources, Bookmark-backed durable anchors, annotation stores, and widget lanes
• example parity is now explicitly reopened: same-path current example files and same-path browser tests are not enough to close the broader legacy parity story
• no current perf blocker remains under the curated package
• local file-family/browser rows no longer block by themselves:
  • Firefox local weirdness rows are closed
  • reopened browser/input file-family rows stay open until same-path parity or explicit engine-rewrite exceptions are recorded under repair-drift.mdc
  • deleted restore-dom and android-input-manager/** families do not count as closed just because specialist browser lanes are green; the surviving shared surface still has to prove the relied-on behavior
• remaining blocker rows are:
  • exhaustive API / public-surface audit:
    • transform explicit-skip rows
  • example parity:
    • legacy/current example matrix across all examples
    • explicit mirrored vs extended classification
• external collaboration adapters remain external by design

Post-RC follow-up rows:

• external Android keyboard-feature evidence:
  • autocorrect
  • glide typing
  • voice input
• broader iOS Safari composition/focus evidence

What Is Safe To Recommend Now

Default recommendation:

• Slate
• EditableBlocks
• withHistory(createEditor())

That surface is still the completed POC RC surface for the already-proved runtime/perf stack.

What is not honest to say right now:

• that the repo has already closed full legacy API/public-surface parity
• that the RC has already proved zero regression versus legacy Slate React on every relied-on IME/mobile/browser weird-input surface
• that the repo has already reached maximum parity with legacy for all examples
• that browser/input parity and the exhaustive API/public-surface audit are actually closed under the repair-drift.mdc rule, not just softened into specialist-proof follow-up
• declaration-merging CustomTypes is no longer a hard-cut non-claim on the current type surface; the package seam has been recovered and stale skip docs need refresh

What changed:

• the functionality/proof stack is still real
• but the curated perf gate package is now active
• the mainstream richtext perf lane is now green again on the current blocker-facing benchmark surface
• placeholder and huge-document blocker lanes are also green enough under the current curated package
• the default withHistory(createEditor()) surface is still covered on the headless lane, but headless/history substrate proof does not close shared browser hotkey/input integration in Editable
• core normalization benches remain ugly, but stay diagnostic until mapped into user-facing blocker lanes

See:

• replacement-family-ledger.md
• replacement-gates-scoreboard.md
• true-slate-rc-proof-ledger.md

Why The Earlier Go Verdict Was Reopened

The curated perf blocker package still owns part of the live truth. It is no longer the whole truth.

Browser/input parity no longer blocks the current RC gate. It still limits the broader blanket replacement claim.

Current blocker-facing read:

1. mounted-runtime placeholder baseline is faster than legacy
2. huge-document user flows are green enough on the measured gate lane
3. mainstream richtext formatting latency is now faster than legacy on the current blocker rerun
4. slate-history undo / redo on the default recommendation surface is now measured directly; it stays slower than legacy, but the fresh same-turn 5000-block compare still lands in a tolerable band: typing undo 20.27ms, typing redo 17.7ms, fragment undo 31.77ms, fragment redo 29.11ms, so it does not justify reopening RC
5. there is no honest browser huge-document history compare row to keep: legacy Slate never exposed a history-backed huge-document surface there, so the headless history compare lane remains the sole owner
6. core normalization benches are still substantially slower than legacy, but stay diagnostic until mapped into user-facing blocker lanes

Current deletion-review read:

• major file/test deletion review is now explicitly closed
• it no longer blocks Target B by itself

Current extension-model read:

• extension model / behavior interception is now explicitly closed
• it no longer blocks Target B by itself

Current normalization read:

• schema / normalization extensibility is now explicitly closed on the current default-vs-explicit model
• safe live invariants are real by default
• heavier canonicalization lives on explicit/load/import/app-owned boundaries
• broader always-on live coercion remains explicitly outside the default claim unless a future design survives the full proof stack

Current headless read:

• non-React / headless core usability is now explicitly closed on the current package-split model
• direct composition across slate, slate-history, and slate-hyperscript is directly proved without React
• the docs now show a real headless entry path instead of implying React is the only serious starting point

Current collaboration read:

• operation-history-collaboration integrity is now explicitly closed on the local substrate
• operation and history records are proved as collaboration-safe inputs
• external slate-yjs still owns CRDT/provider/cursor integration

Current public-surface read:

• the package-level matrix across stable surfaces, secondary surfaces, and explicit non-claims is still useful
• but broad API / public surface reconciliation is reopened under challenge until exhaustive per-API contract-width audit closes
• the first exposed miss family, legacy Editor.before(...) / Editor.after(...) voids: true rows and nonSelectable traversal rows, is now recovered in code and proof
• docs, ledgers, and proof rows must now be re-audited for accepted-argument, option-bag, and behavior-width parity instead of name-only helper presence

The old lane-by-lane-only perf posture is no longer the live read.

The curated perf gate package now still matters to both:

• Target A, because mainstream richtext flows belong to the default recommendation surface
• Target B, because the same lane belongs to the broader replacement claim

Final Gate

Target A

Verdict:

• Go

Gate read:

1. current core oracle proof exists
2. runtime and browser proof exist for the default envelope
3. contributor-facing recovery for the current recommendation is real
4. blocker-facing perf gates now exist
5. current blocker-facing perf lanes clear under the curated package, including the explicit history compare lane
6. the remaining browser/input rows are explicit post-RC follow-up, not current RC blockers

Target B

Verdict:

• Reopened under challenge

Gate read:

1. contract-recovery lanes are closed
2. broader proof depth is now wide enough to support the package-level claim
3. the curated blocker-facing perf package is currently green enough for the broader replacement claim
4. browser/input zero-regression parity is deferred to post-RC follow-up and is still unresolved for the broader blanket replacement claim
5. exhaustive API / public-surface contract-width audit is reopened and still unresolved

Reclose Rule

Target B may return to Go only after:

1. the exhaustive API / public-surface contract-width audit closes the reopened rows or marks every surviving cut explicitly with owner sign-off
2. the example-parity lane closes or marks every surviving extension/cut explicitly with owner sign-off
3. the behavior/parity ledger closes the IME/mobile/browser rows or marks a row as justified omission with explicit owner sign-off
4. the RC docs are reconciled back to the behavior/parity, example-parity, and API-truth authorities
5. the perf blocker package stays green enough on the already-proved lanes

Target A stays Go unless a blocker reopens inside the proved default recommendation surface or the current claim is widened back to full zero-regression browser/input parity.

Current Read

This file is not the roadmap.

It owns the live verdict only:

• completed POC RC still names the default recommendation surface
• the current curated perf gate package still matters
• example parity is now an explicit broader-claim blocker, not a side-effect of file-review closure
• browser/input parity is now explicit post-RC follow-up work: it does not block the current RC recommendation, but it still blocks the broader blanket replacement claim until closed or explicitly narrowed