Retire deferred scope and preserve future performance experiments

docs/roadmap/future.md

@@ -2,8 +2,9 @@
 
 This document lists potential enhancements for consideration in future ConfigKit
 releases. Entries are not accepted contracts, implementation plans, or release
-commitments. [Deferred scope](deferred.md) remains the owner of current exclusions;
-adopting a candidate requires an explicit design decision and a separate plan.
+commitments. Adopting a candidate requires an explicit design decision and a
+separate plan. Significant architectural exclusions and their rationale belong
+in ADRs; this document does not maintain a list of rejected proposals.
 
 ## 1. Identified record sequences
 
@@ -130,3 +131,31 @@ compatible version or replace final configuration validation.
   reader, byte, and mapping inputs, glob matches, and repeated imports? Define
   behavior for absent optional inputs and preserve source-local diagnostics,
   inspection evidence, and existing sensitivity handling.
+
+## 3. Performance experiments
+
+Status: deferred pending evidence. The completed performance audit identified
+four internal optimization candidates for separate investigation:
+
+- Branch-local explanation replay instead of the current whole-tree replay.
+- Redesigning the private metadata-copy handoff while preserving detached public
+  records.
+- Caching schema descriptors across loads.
+- Indexing complex contributor-trace membership while preserving its equality
+  rules and compact linked traces.
+
+These are experiments, not committed implementation work or promised speedups.
+Reconsider them with evidence that the benefit justifies the ownership,
+retained-memory, or equivalence risks. The completed audit and implementation
+evidence are preserved in Git history at `8e91a2b` under `docs/roadmap/`.
+
+### Questions to resolve before implementation
+
+- Which representative workloads show enough remaining cost to justify each
+  optimization, and what measurements would establish a useful improvement?
+- Can narrower replay and membership indexes preserve explanation results,
+  ordering, provenance, and sensitivity behavior?
+- Can metadata handoff changes preserve detached public records and ownership
+  across loads without introducing aliasing?
+- What cache lifetime and concurrency design would make schema reuse worthwhile
+  without excessive retained memory?