diff --git a/graph/code-patterns/embed-if-changed-pattern-compare-id-sets-to-skip-unnecessary-20b22f.md b/graph/code-patterns/embed-if-changed-pattern-compare-id-sets-to-skip-unnecessary-20b22f.md
new file mode 100644
index 00000000000..59f4311285d
--- /dev/null
+++ b/graph/code-patterns/embed-if-changed-pattern-compare-id-sets-to-skip-unnecessary-20b22f.md
@@ -0,0 +1,12 @@
+---
+id: 20b22fc6-bf36-420d-ae9d-a67beba3a209
+type: code_pattern
+title: "embed --if-changed pattern: compare ID sets to skip unnecessary work"
+tags: [cognitive-memory, python, pattern, optimization]
+importance: 0.5
+confidence: 0.8
+created: "2026-02-19T20:53:28.365333+00:00"
+updated: "2026-02-19T20:53:28.365333+00:00"
+---
+
+The `--if-changed` flag on `claude-memory embed` compares the set of memory UUIDs in _index.json against _embeddings.json. If the sets match (no new or deleted memories), it returns immediately with `{"skipped": true}` in ~0.1s. Also checks for provider changes (which force a full re-embed due to dimension mismatch). Pattern: load both files, compare `set(index_entries.keys()) == set(embedding_entries.keys())`. Simple, fast, and avoids unnecessary Ollama API calls. Useful for any timer-driven task that should be idempotent.