diff --git a/graph/problems/cognitive-memory-clientpy-at-3260-lines-needs-modular-refact-8d3654.md b/graph/problems/cognitive-memory-clientpy-at-3260-lines-needs-modular-refact-8d3654.md
index 450c34fd7da..c2b72e0c0b9 100644
--- a/graph/problems/cognitive-memory-clientpy-at-3260-lines-needs-modular-refact-8d3654.md
+++ b/graph/problems/cognitive-memory-clientpy-at-3260-lines-needs-modular-refact-8d3654.md
@@ -1,29 +1,30 @@
 ---
 id: 8d365483-701f-49fe-a55e-4d524ae2eff3
 type: problem
-title: "cognitive-memory client.py at 3260 lines needs modular refactor"
+title: "cognitive-memory client.py at 3348 lines needs modular refactor"
 tags: [cognitive-memory, refactoring, architecture, tech-debt]
 importance: 0.7
-confidence: 0.8
+confidence: 0.9
 created: "2026-02-19T20:57:01.898902+00:00"
-updated: "2026-02-19T20:57:01.898902+00:00"
+updated: "2026-02-28T00:00:00+00:00"
 ---
 
-client.py has grown to 3260 lines with 47 methods in a single CognitiveMemoryClient class. Breakdown by concern:
+client.py has grown to 3348 lines with 66 methods (was 3260/47 on 2026-02-19). Breakdown by concern with accurate line counts:
 
-- Constants + module helpers: ~548 lines → common.py
-- Core CRUD (store/get/delete/search/recall): ~435 lines → stays in client.py
-- Edges (relate/edge_get/edge_search/edge_update/edge_delete): ~352 lines → edges.py
-- Embeddings (embed/semantic_recall/provider fallback): ~209 lines → embeddings.py
-- Decay/Core/Reflect (decay scoring, CORE.md gen, reflection): ~1097 lines → analysis.py (biggest chunk)
-- CLI parser + dispatch: ~463 lines → cli.py or __main__.py
-- Index/state/git plumbing: ~115 lines → internal helpers
+- **common.py** (~548 lines): Constants, imports, YAML/frontmatter parsing, slugify, decay calc, embedding helpers, cosine similarity, edge filename/frontmatter helpers
+- **client.py** (~1008 lines): CognitiveMemoryClient class — __init__, internal helpers (_ensure_dirs, _load/_save index/state, _resolve paths, _git_commit, _update_index_entry), store, recall, get, search, update, delete, related, stats, recent, episode, reindex, pin
+- **edges.py** (~274 lines): relate, edge_get, edge_search, edge_update, edge_delete + edge index helpers (_update_edge_index, _remove_edge_index, _scan_for_edge, _resolve_edge_path)
+- **embeddings.py** (~208 lines): _get_embedding_provider, _embed_texts_with_fallback, embed, semantic_recall
+- **analysis.py** (~839 lines): decay, core (CORE.md gen), tags_list/related/suggest, reflect, merge, reflection_summary — biggest chunk
+- **cli.py** (~471 lines): argparse parser + dispatch + main()
+
+Approach: mixin classes. Each module defines a Mixin class; CognitiveMemoryClient inherits from all mixins. This preserves `self` access and keeps the public API identical.
 
 Refactor considerations:
-- MCP server imports `from client import CognitiveMemoryClient` — needs updating
-- CLI wrapper `claude-memory` uses `exec uv run python client.py` — needs package structure
-- Systemd services depend on the wrapper — must test after
-- Could use mixin classes or composition to split the monolithic class
-- Similar to the Major Domo CLI modular refactor (that went well)
+- MCP server imports `from client import CognitiveMemoryClient` — needs re-export or updated import
+- CLI wrapper at `/home/cal/.local/bin/claude-memory` uses `exec uv run python client.py` — cli.py must be runnable
+- Systemd services (daily/weekly timers) depend on the wrapper — must test after
+- mcp_server.py also imports `_load_memory_config, MEMORY_DIR` from client — move to common.py
+- 4 module-level helper functions are used across concerns — belong in common.py
 
-Priority: near-future. Nothing is broken but finding anything requires line-number hunting.
+Priority: active. Line count growing, 66 methods in one class makes navigation painful.