ContextQMD
Back to Plate

Slate v2 Zero-Regression Parity Reconsolidated Plan

docs/plans/2026-04-12-slate-v2-zero-regression-parity-reconsolidated-plan.md

53.0.611.8 KB
Original Source

Slate v2 Zero-Regression Parity Reconsolidated Plan

Purpose

One current plan for the remaining work required to support the strongest honest Slate React parity claim:

  • zero regression on legacy relied-on behavior
  • especially browser/input/IME/mobile weirdness
  • without letting tooling drift or benchmark drift hijack the queue

This plan supersedes vague “next batch” execution for the reopened browser/input lane.

Scope Lock

Primary goal:

  • close the remaining behavior-bearing parity rows that legacy Slate React relied on

RC gate:

  • no regression versus legacy Slate tests
  • plus no regression on explicitly relied-on weird-input/browser patches that legacy carried in runtime code even when legacy did not have deep automated proof

Non-goals:

  • chasing perfect iOS automation before all higher-signal local rows close
  • widening perf work beyond the curated blocker package
  • adding transport abstraction work that does not close a named row

Hard rule:

  • every remaining item must be classified first:
    • product-bug
    • missing-proof-row
    • tooling-blocked
    • justified-omission-candidate

Authority

Execution truth:

Roadmap truth:

Verdict truth:

Planning inputs that changed the queue:

Current Snapshot

Already closed at a serious level:

  • Chromium IME rows:
    • placeholder
    • no-FEFF placeholder
    • inline-edge
    • void-edge
    • post-composition undo/redo
  • Firefox rows:
    • direct composition on the main IME surfaces
    • blur/focus selection recovery
    • zero-width normalization
    • drag/drop cleanup after dragged-node unmount
    • table multi-range preservation during selection sync
    • nested editable focus bounce
  • Android direct rows on emulator Chrome:
    • placeholder
    • no-FEFF placeholder
    • inline-edge
    • void-edge
    • split/join
    • empty/delete-rebuild
    • remove-range
    • structural special subcases
  • desktop WebKit:
    • focus/selection recovery
    • zero-width normalization
    • direct-input ceiling plus proxy composition backstop
  • no-FEFF delayed plain typing bug:
    • fixed in runtime
  • package/type publishing drift:
    • fixed in build pipeline

Still open:

  1. Android keyboard-feature-dependent behavior:
  2. broader iOS Safari composition/focus beyond the currently green rows external plan: 2026-04-12-ios-safari-broader-composition-focus-external-evidence-plan.md
  3. final verdict/document reconsolidation after the external/tooling-blocked rows are settled or explicitly downgraded

Already closed enough to stop pretending they are future batches:

  • the reopened file-family rows in release-file-review-ledger.md are checked
  • Firefox local weirdness closure is banked
  • restore-dom is closed as a non-current lifecycle guard plus current-proof coverage split
  • the Android legacy manual-case matrix is classified tightly enough that the remaining debt is keyboard-feature evidence, not generic local browser chaos

Ledger-Close Rule

Tranche 7 does not close on “basically done.”

It closes only when the reopened browser/input rows in release-file-review-ledger.md are individually checkmarkable and actually checked.

Hard rules:

  1. if one still-open row hides multiple platforms, split it first
  2. file-family debt does not stay open just because some platform row is still open somewhere else
  3. residual debt must live on the narrowest honest row:
    • file family
    • platform parity row
    • or external/tooling-blocked row
  4. no batch is complete while stale next-batch wording still points at work that is already locally closed

Remaining Work Map

Phase 1: Keep The Closed Ledger Closed

Goal:

  • preserve the fully checked file-review ledger and stop stale wording from reopening generic local debt

Rows:

  1. file-review closure state current read: the reopened browser/input file-family rows are already checked in release-file-review-ledger.md next action: do not reopen generic file-family debt unless a new concrete behavior gap is discovered exit: remaining parity work lives only on named platform or external/tooling rows

  2. roadmap wording current read: some planning surfaces still talk like local Firefox/file-family recovery is the next batch next action: refresh the queue so it starts from the remaining non-local rows exit: no stale local next-batch wording survives

Phase 2: Package The Remaining External Evidence Work

Goal:

  • convert the remaining claim-width debt into named non-local execution lanes

Rows:

  1. Android keyboard-feature behavior bucket: tooling-blocked or justified-omission-candidate current read: local Appium + Chrome on emulator now proves the structural and typing rows that mattered inside the old android-tests matrix, but does not expose Gboard candidates well enough for:

    • autocorrect
    • glide typing
    • voice input next action: choose one explicit non-local path:
    • real-device capture
    • browser-farm / device-cloud proof
    • alternate transport that exposes candidate UI
    • manual-device artifact lane if transport ceilings remain exit: each remaining Android keyboard-feature row has an explicit owner and evidence path

  2. broader iOS Safari composition/focus bucket: tooling-blocked current read: local Appium/XCUITest is setup-green / behavior-red and the current agent-browser iOS path is only trustworthy for open plus initial snapshot next action: choose one explicit non-local path:

    • real-device cloud/browser farm
    • alternate iOS automation substrate
    • manual-device evidence lane if automation still fails exit: iOS rows become either directly proved or explicitly tagged as external tooling debt outside the repo

  3. desktop WebKit direct composition depth bucket: tooling-blocked or justified-omission-candidate current read: current honest ceiling is direct input plus proxy composition next action: only promote this if a real relied-on legacy behavior remains unowned after the iOS decision exit: either direct row exists or proxy/ceiling status is explicitly accepted

Phase 3: Reconcile Verdict And Proof Truth

Goal:

  • make the live docs say the same thing about the checked ledger and the still open external/tooling rows

Rows:

  1. blocker/verdict docs files:

  2. roadmap/front-door docs files:

  3. proof/status docs files:

Required Next Batches

Batch 1

Target:

  • Android external evidence package

Deliver:

  • chosen transport/evidence path for:
    • autocorrect
    • glide typing
    • voice input
  • exact artifact requirements for calling any of those rows closed

Batch 2

Target:

  • iOS external evidence package

Deliver:

  • chosen non-local execution lane for broader iOS Safari composition/focus
  • explicit carry-forward of the current setup-green / behavior-red boundary

Batch 3

Target:

  • optional WebKit direct-depth decision

Deliver:

  • either:
    • a narrow follow-up for direct WebKit depth
    • or an explicit defer/omission decision behind the iOS result

Batch 4

Target:

  • final doc truth pass

Deliver:

  • release-file ledger has no vague grouped [ ] debt
  • roadmap/verdict docs agree on the blocker story
  • stale Target A / Target B Go wording is removed where it contradicts the reopened parity lane

Documentation Work Required On Every Batch Exit

Always refresh:

  1. master-roadmap.md
  2. overview.md
  3. release-file-review-ledger.md
  4. true-slate-rc-proof-ledger.md
  5. release-readiness-decision.md when verdict language changes
  6. replacement-gates-scoreboard.md when proof scope changes
  7. full-replacement-blockers.md when blocker truth changes

Re-Interview Trigger

Run the remaining-scope re-interview again if any of these happen:

  • a new legacy relied-on browser/input behavior is discovered
  • the RC gate is widened beyond legacy test parity plus named relied-on patches
  • iOS evidence strategy changes from local automation to external/manual
  • a reopened file family splits into more than one independent closure lane

Reference:

Hard Read

The remaining work is no longer “figure out whether Slate v2 is real.”

That part is done.

The remaining work is:

  • stop pretending Firefox is still the next batch when that local tranche is already closed
  • make the ledger rows individually checkmarkable
  • reclose editable.tsx and android-input-manager/** honestly
  • keep Android debt narrowed to keyboard features only
  • stop letting iOS tooling limitations blur product truth
  • then reconcile the verdict docs so they stop contradicting each other