memory-consolidationlisted

# Memory Consolidation **Source**: the Redis architecture guide names this explicitly as pipeline stage four — "consolidation decides what stays as raw episodes and what gets promoted into more durable knowledge... without it, your memory store grows indefinitely and retrieval quality degrades over time." ML Mastery's five-pattern guide independently makes the same point: "all patterns require growth management... mandatory operational components, not optional refinements." ## Two distinct consolidation moves **1. Checkpoint → episode.** Many small, frequent checkpoints from `session-checkpoint` accumulate fast. Periodically, fold a run of checkpoints from one session into a single coherent episode entry in `.memory/episodes/` — the checkpoints did their job (safety net during the session); once the session's actually over, the individual snapshots are less useful than one narrative summary of what happened. The raw checkpoints can then be archived (see `memory-hygiene`). **2. Episode → semantic.** When a lesson from a specific episode generalizes past its original context — "we learned that X approach doesn't work for this codebase" becomes a fact worth knowing regardless of which task surfaces it again — promote that specific, generalizable lesson into `.memory/semantic/project-facts.md`, stripped of the episode-specific narrative detail. The original episode stays in `.memory/episodes/` as the historical record of *how* that lesson was learned; the promoted version in