diff --git a/graph/workflows/pr-review-cognitive-memory7-decay-filter-in-semantic-recall-1af571.md b/graph/workflows/pr-review-cognitive-memory7-decay-filter-in-semantic-recall-1af571.md
new file mode 100644
index 00000000000..f938fa68ebf
--- /dev/null
+++ b/graph/workflows/pr-review-cognitive-memory7-decay-filter-in-semantic-recall-1af571.md
@@ -0,0 +1,25 @@
+---
+id: 1af571c5-cf7a-4e40-b324-6973b56f2502
+type: workflow
+title: "PR review: cognitive-memory#7 — decay filter in semantic_recall()"
+tags: [pr-reviewer, cognitive-memory, python, decay, embeddings, semantic-recall, approved]
+importance: 0.4
+confidence: 0.8
+created: "2026-03-03T05:32:47.623228+00:00"
+updated: "2026-03-03T05:32:47.623228+00:00"
+---
+
+Reviewed PR #7 in cal/cognitive-memory: "fix: filter archived/dormant memories from semantic_recall() (#3)".
+
+**Verdict**: APPROVED (could not post — Gitea blocks self-review)
+
+**Files reviewed**: `embeddings.py`
+
+**Summary**: Added `THRESHOLD_DORMANT` guard to `semantic_recall()` to skip dormant memories (decay_score < 0.05) before cosine similarity scoring. Mirrors the existing guard in `recall()` keyword path exactly.
+
+**Key findings**:
+- Correct: filter logic, default value (0.5), and placement (before loop) all match the keyword recall path
+- `_load_state()` called once outside the loop — efficient
+- Safe fallback: missing state entries default to 0.5 (active), not filtered
+- No security issues; no style concerns
+- Clean, minimal, well-targeted fix