Add fact-graph nudge test coverage, document graph in quickstart, fix layer count

- test_retrieval.py: cover search_all()'s fact-side graph nudge, previously
  untested (only the chunk-side nudge had a regression test).
- quickstart.md: add a knowledge-graph section — the doc had zero mention
  of it despite being a headline feature.
- README.md: "four memory layers" heading was stale after Graph became a
  fifth row in that table.
- .gitignore: ignore blog.md/newwork.md (working notes) and .DS_Store/
  .coverage (local artifacts).

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

Author: CodebyKumar

.gitignore

@@ -5,3 +5,7 @@ dist/
 *.egg-info/
 .venv/
 uv.lock
+blog.md
+newwork.md
+.DS_Store
+.coverage

README.md

@@ -137,7 +137,7 @@ graph expansion at FULL tier, 1-hop at STANDARD, off entirely at LITE (zero
 extra queries, zero extra writes). Full design and the reasoning behind it:
 [governor.md's Knowledge graph section](docs/governor.md#knowledge-graph).
 
-## The four memory layers
+## The five memory layers
 
 | Layer                | What it holds                                                  | Where                                       |
 | -------------------- | -------------------------------------------------------------- | ------------------------------------------- |

docs/quickstart.md

@@ -104,6 +104,28 @@ If you don't pass an embedder, Elastimem activates its own small built-in
 one automatically (see [governor.md](governor.md)) — semantic recall works
 out of the box, no setup required.
 
+## The knowledge graph (with an LLM configured)
+
+With `llm=` set, `record_turn` also extracts entities and relationships in
+the same background pass that captures facts — no extra model call, no NER
+library. They connect memories that share no vocabulary:
+
+```python
+mem.record_turn("I'm building a robot called Tuffy that runs on a Jetson",
+                 "Sounds like a fun project!")
+mem.end_session()   # graph maintenance runs here: decay, dedup, clustering
+
+hits = mem.recall("what do I know about my Jetson")   # finds the Tuffy turn too
+
+for cluster in mem.clusters():        # entities auto-grouped into topics
+    print(cluster["label"], cluster["members"])
+```
+
+This is one more retrieval signal, not a separate database — off entirely
+at LITE tier, deeper at FULL. See [governor.md](governor.md#knowledge-graph)
+for the full design, and [api.md](api.md) for `explain()`/`timeline()`, the
+two Experimental query methods built on top of it.
+
 ## Where to go next
 
 - [installation.md](installation.md) — optional extras, supported Python versions

tests/test_retrieval.py

@@ -177,6 +177,34 @@ def test_graph_nudge_breaks_tie(tmp_path):
     s.close()
 
 
+def test_graph_nudge_breaks_tie_for_facts(tmp_path):
+    """Same as test_graph_nudge_breaks_tie, but for facts returned through
+    recall() (search_all's fact leg, retrieval.py's search_all graph_nudge
+    branch) rather than chunks — a separate code path that reuses
+    graph_relevance's expanded names but applies the nudge to fact rows
+    instead of chunk rows."""
+    from elastimem import graph
+    from elastimem.governor import Tier
+
+    s = make_store(tmp_path / "gf.db", embed=None, tier_override=Tier.FULL)
+    s.remember("device_one", "Tuffy is a device that needs a new battery pack")
+    s.remember("device_two", "the thermostat is a device that needs new batteries")
+
+    conn = s._conn
+    jetson_id = graph.upsert_node(conn, "thing", "Jetson", confidence=1.0)
+    tuffy_id = graph.upsert_node(conn, "thing", "Tuffy", confidence=1.0)
+    graph.upsert_edge(conn, tuffy_id, jetson_id, "runs_on", confidence=1.0)
+
+    hits = s.recall("tell me about the device and my Jetson")
+    fact_hits = [h for h in hits if h.kind == "fact"]
+    assert fact_hits
+    tuffy_facts = [h for h in fact_hits if "Tuffy" in h.text]
+    other_facts = [h for h in fact_hits if "thermostat" in h.text]
+    assert tuffy_facts and other_facts
+    assert tuffy_facts[0].score > other_facts[0].score
+    s.close()
+
+
 def test_graph_hops_zero_matches_graph_absent_ranking(tmp_path):
     """LITE tier (graph_hops=0): ranking must be identical to a store with
     no graph data at all — the graph leg must not be reachable."""