Performance Docs

This directory is the human-facing source of truth for Plate editor performance work.

Read Order

1. editor-performance-master-plan.md
   • the main program document
   • current state, key wins, remaining red lanes, and sequencing
2. plate-vs-slate-benchmarks.md
   • benchmark narrative and forensic history
   • use this when you need to understand why a cut mattered
3. editor-performance-next-phase-consensus.md
   • planning artifact for the earlier “what next” decision
   • useful for rationale, not for the latest numbers
4. performance-benchmark-spec.md
   • benchmark contract for the public Performance guide
   • defines scope, workloads, and claim boundaries
5. 2026-04-04-standalone-benchmark-gap-analysis.md
   • why the standalone rich-markdown lab exposed a Plate mount gap
   • current diagnosis and next optimization targets
6. 2026-04-03-affinity-redesign-plan.md
   • plan for the hard-affinity code-mark redesign
7. 2026-04-03-remaining-basic-nodes-benchmark.md
   • plan for the remaining @platejs/basic-nodes census extension

Artifact Policy

• Durable markdown lives here.
• Raw benchmark JSON lives in .tmp/.
• Older docs may still mention docs/plans/*.json from before the raw artifact move. When that happens, prefer the matching file in .tmp/.
• Internal one-off planning/proof notes belong under docs/plans/, not here, unless they are directly useful to understanding the performance story.

Release Snapshot

As of 2026-04-03:

• plain core is good enough versus Slate
• insert-text perf is good enough
• nodeId is no longer the cliff
• the dedicated nodeId paste/import lane now exists, and it shows:
  • raw import is basically cheap
  • duplicate-id paste is the only remaining meaningful withNodeId hotspot
• code-affinity redesign landed and materially improved the worst core-plugin lane
• newly benchmarked remaining basic-nodes plugins split cleanly:
  • green enough: KbdPlugin, SubscriptPlugin, SuperscriptPlugin
  • still red: HighlightPlugin, StrikethroughPlugin

What Is Left After Release

If we ship now, the meaningful remaining performance work is:

1. HighlightPlugin
   • currently still well behind Slate on the activated lane
2. StrikethroughPlugin
   • same story; cheap mark path still too expensive
3. Table selection
   • still a separate real hotspot, not solved by core editor work
4. Performance-doc cleanup
   • some historical JSON artifact links still need cleanup/consolidation

What Is Not Worth Chasing Right Now

• more cheap-mark surgery on bold/italic/underline
• insert-text perf
• more init-time nodeId surgery without a new duplicate-paste result
• chunking rhetoric without a new workload
• more code-affinity tuning unless behavior bugs appear