Summary

Historical rebuild and maintenance recovery are much better than before, but still need a dedicated hardening pass.

Goal

Make learning recovery and rebuild behavior robust under corruption, interruption, and restart.

Scope

• stronger stale detection for historical index rebuilds
• better recovery from partial writes/corrupted history lines
• explicit maintenance resume/retry semantics after interruption
• consistency checks after rebuild/maintenance
• safer handling of partial state during daemon restart

Deliverables

• stronger rebuild invariants
• better recovery semantics for maintenance interruption
• consistency verification hooks
• reduced risk of silent learning-state drift

Non-goals

• do not redesign session history storage from scratch
• do not fold unrelated sandbox/reliability work into this issue

Related

• parent epic: [EPIC] Self-Learning Phase 2: Explainability, Validation, and Governance #225
• Follow up learning maintenance with index rebuild and downstream integration #215
• feat(learning): HistoricalLogIndex — inverted index over session transcripts #124
• feat(learning): Consolidation Scheduler — auto-trigger memory maintenance #127