diff --git a/playlist_manager/playlist_manager/main.py b/playlist_manager/playlist_manager/main.py
index 23eec5e..cabf031 100644
--- a/playlist_manager/playlist_manager/main.py
+++ b/playlist_manager/playlist_manager/main.py
@@ -110,7 +110,11 @@ class PlaylistManager(object):
 				[tag.lower() for tag in row.tags],
 				row.first_event_id,
 				row.last_event_id,
-			) for row in query(conn, "SELECT playlist_id, tags, first_event_id, last_event_id FROM playlists WHERE tags IS NOT NULL")
+			) for row in query(conn, """
+				SELECT playlist_id, tags, first_event_id, last_event_id
+				FROM playlists
+				WHERE tags IS NOT NULL AND playlist_id IS NOT NULL
+			""")
 		}
 		self.dbmanager.put_conn(conn)
 		duplicates = set(playlists) & set(self.static_playlists)
diff --git a/postgres/setup.sh b/postgres/setup.sh
index 2495c50..254947e 100644
--- a/postgres/setup.sh
+++ b/postgres/setup.sh
@@ -141,16 +141,24 @@ CREATE TABLE editors (
 	name TEXT NOT NULL
 );
 
+-- A slight misnomer, this is all rows of the tags sheet.
+-- It includes tags that have been promoted to actual playlists, and ones that have not.
 -- Playlists are communicated to playlist manager via this table.
 CREATE TABLE playlists (
-	playlist_id TEXT PRIMARY KEY,
-	name TEXT NOT NULL,
+	id TEXT PRIMARY KEY,
+	-- These are sheet inputs, and aren't used directly by anything (except reverse sync)
+	name TEXT NOT NULL DEFAULT '',
+	description TEXT NOT NULL DEFAULT ''
 	-- When tags is NULL, it indicates tags have not been set and so the playlist should
 	-- match nothing. Conversely, when tags is empty, it indicates the playlist should match everything.
 	tags TEXT[],
-	first_event_id UUID REFERENCES events(id) ON DELETE SET NULL,
-	last_event_id UUID REFERENCES events(id) ON DELETE SET NULL,
-	show_in_description BOOLEAN NOT NULL
+	playlist_id TEXT,
+	show_in_description BOOLEAN NOT NULL DEFAULT FALSE,
+	-- These event ids are references into the events table, but they aren't foreign keys
+	-- because we don't want invalid input to cause integrity errors.
+	-- It's totally safe for these to point to non-existent events, it just does nothing.
+	first_event_id TEXT,
+	last_event_id TEXT
 );
 
 -- This table records time series data gleaned from the bus cam (right now, just the odometer).
diff --git a/sheetsync/sheetsync/main.py b/sheetsync/sheetsync/main.py
index 6786460..12e5e1f 100644
--- a/sheetsync/sheetsync/main.py
+++ b/sheetsync/sheetsync/main.py
@@ -312,9 +312,12 @@ class PlaylistsSync(SheetSync):
 
 	table = "playlists"
 	input_columns = {
-		"playlist_id",
-		"tags",
 		"name",
+		"description",
+		"tags",
+		"playlist_id",
+		"first_event_id",
+		"last_event_id",
 		"show_in_description",
 	}
 
diff --git a/sheetsync/sheetsync/streamlog.py b/sheetsync/sheetsync/streamlog.py
index 0055880..9db2eb7 100644
--- a/sheetsync/sheetsync/streamlog.py
+++ b/sheetsync/sheetsync/streamlog.py
@@ -65,9 +65,11 @@ class StreamLogPlaylistsMiddleware(Middleware):
 				# Special case for the "all everything" list, otherwise all playlists have a single tag.
 				"tags": [] if tag["tag"] == "<all>" else [tag["tag"]],
 				"playlist_id": tag["playlist"]
+				"description": tag["description"],
 				"name": "unknown", # TODO missing in StreamLog
 				"show_in_description": False, # TODO missing in StreamLog
-				"description": tag["description"],
+				"first_event_id": None, # TODO missing in StreamLog
+				"last_event_id": None, # TODO missing in StreamLog
 			})
 		return True, rows
 
diff --git a/thrimshim/thrimshim/main.py b/thrimshim/thrimshim/main.py
index 6129f9b..2099026 100644
--- a/thrimshim/thrimshim/main.py
+++ b/thrimshim/thrimshim/main.py
@@ -303,7 +303,7 @@ def update_row(ident, editor=None):
 		playlists = database.query(conn, """ 
 			SELECT playlist_id, name, tags
 			FROM playlists
-			WHERE show_in_description AND tags IS NOT NULL
+			WHERE show_in_description AND tags IS NOT NULL AND playlist_id IS NOT NULL
 		""")
 		# Filter for matching playlists for this video
 		playlists = [