diff --git a/goodnews/api.py b/goodnews/api.py
index 4b55324..f35a0cf 100644
--- a/goodnews/api.py
+++ b/goodnews/api.py
@@ -57,6 +57,18 @@ def get_conn():
         conn.close()
 
 
+def _prefs_sql_kw(fp, now) -> dict:
+    """Categorical prefs → queries.feed keyword filters (avoid-terms stay Python)."""
+    return dict(
+        include_topics=fp.include_topics or None,
+        include_flavors=fp.include_flavors or None,
+        mute_topics=list(fp.muted_topics(now)) or None,
+        mute_flavors=list(fp.muted_flavors(now)) or None,
+        max_cortisol=fp.max_cortisol,
+        max_ragebait=fp.max_ragebait,
+    )
+
+
 def _pick_lead(items: list[dict]) -> list[dict]:
     """Lead with a gentle, readable, ideally illustrated story.
 
@@ -303,15 +315,28 @@ def create_app() -> FastAPI:
         prefs: str | None = Query(None),
     ) -> BriefResponse:
         fp = prefs_from_json(prefs)
+        now = datetime.now(timezone.utc)
         with get_conn() as conn:
             data = queries.brief(conn, brief_date=date, limit=limit)
-        items = data["items"]
-        if not fp.is_empty():
-            # MVP: filter the stored brief DOWN; no refill from outside the brief.
-            # Runs before hero selection, so personal avoid-terms take precedence.
-            items = filter_articles(items, fp, datetime.now(timezone.utc))
+            items = data["items"]
+            if not fp.is_empty():
+                # Apply personal boundaries (avoid-terms take precedence over curation).
+                items = filter_articles(items, fp, now)
+                # Keep the highlights full: if a boundary hid a story, top up with
+                # other readable, boundary-respecting good news rather than show fewer.
+                if len(items) < limit:
+                    have = {a["id"] for a in items}
+                    pool = queries.feed(
+                        conn, accepted_only=True, limit=limit * 5 + 20, offset=0, **_prefs_sql_kw(fp, now)
+                    )
+                    for a in filter_articles(pool, fp, now):
+                        if len(items) >= limit:
+                            break
+                        if a["id"] not in have:
+                            items.append(a)
+                            have.add(a["id"])
         # Lead with a gentle, readable story (charged or paywalled stories stay
-        # in the five, just not as the first thing seen).
+        # in the set, just not as the first thing seen).
         items = _pick_lead(items)
         return BriefResponse(
             brief_date=data["brief_date"],
diff --git a/tests/test_brief_refill.py b/tests/test_brief_refill.py
new file mode 100644
index 0000000..124ce03
--- /dev/null
+++ b/tests/test_brief_refill.py
@@ -0,0 +1,50 @@
+import json
+import pytest
+from fastapi.testclient import TestClient
+from goodnews.db import connect, init_db
+
+
+@pytest.fixture
+def client(tmp_path, monkeypatch):
+    db = tmp_path / "t.sqlite3"
+    monkeypatch.setenv("GOODNEWS_DB", str(db))
+    conn = connect(db); init_db(conn)
+    conn.execute("INSERT INTO sources (id,name,feed_url,trust_score) VALUES (1,'S','http://s/f',6)")
+
+    def add(aid, topic, in_brief_rank=None, title=None):
+        conn.execute("INSERT INTO articles (id,source_id,canonical_url,title,published_at,url_hash) "
+                     "VALUES (?,1,?,?, '2026-05-31T10:00:00+00:00', ?)",
+                     (aid, f"http://s/{aid}", title or f"t{aid}", f"h{aid}"))
+        conn.execute("INSERT INTO article_scores (article_id,constructive_score,agency_score,human_benefit_score,"
+                     "cortisol_score,ragebait_score,pr_risk_score,accepted,topic,flavor) "
+                     "VALUES (?,7,2,2,1,0,2,1,?,'solution')", (aid, topic))
+        if in_brief_rank:
+            conn.execute("INSERT INTO daily_brief_items (brief_id,article_id,rank) VALUES (1,?,?)", (aid, in_brief_rank))
+
+    conn.execute("INSERT INTO daily_briefs (id,brief_date,title) VALUES (1,'2026-05-31','B')")
+    add(1, "health", 1)        # will be muted
+    add(2, "science", 2)
+    add(3, "environment", 3)
+    add(4, "community")        # refill candidates (accepted, not in brief)
+    add(5, "animals")
+    add(6, "culture")
+    conn.commit(); conn.close()
+    from goodnews.api import create_app
+    return TestClient(create_app())
+
+
+def test_brief_refills_to_full_count_under_boundary(client):
+    full = client.get("/api/brief", params={"limit": 3}).json()
+    assert len(full["items"]) == 3
+
+    muted = client.get("/api/brief", params={"limit": 3, "prefs": json.dumps({"mute_topics": ["health"]})}).json()
+    topics = [i["topic"] for i in muted["items"]]
+    assert len(muted["items"]) == 3            # stayed full (refilled)
+    assert "health" not in topics             # boundary respected
+
+
+def test_brief_refill_respects_avoid_terms(client):
+    # avoid a word in the health item's title
+    client_resp = client.get("/api/brief", params={"limit": 3, "prefs": json.dumps({"avoid_terms": ["t1"]})}).json()
+    assert len(client_resp["items"]) == 3
+    assert all(i["id"] != 1 for i in client_resp["items"])