diff --git a/src/persistence/mod.rs b/src/persistence/mod.rs
index b6823da..40a2c6c 100644
--- a/src/persistence/mod.rs
+++ b/src/persistence/mod.rs
@@ -155,7 +155,19 @@ impl Persistence {
     /// Write `data/<table>.csv` from a table snapshot. Atomic
     /// per file. Creates the `data/` directory if missing
     /// (tolerant of fresh projects).
+    ///
+    /// **Empty tables produce no CSV.** A header-only file
+    /// would carry no information beyond what `project.yaml`
+    /// already records, so an empty snapshot is treated
+    /// identically to "drop this table's data file": the CSV
+    /// is removed if it exists, no file is created if it
+    /// doesn't. This keeps the rule "data lives in CSV; no
+    /// data, no CSV" consistent and avoids surprising users
+    /// with files they didn't ask for.
     pub fn write_table_data(&self, table: &TableSnapshot) -> Result<(), PersistenceError> {
+        if table.rows.is_empty() {
+            return self.delete_table_data(&table.name);
+        }
         let data_dir = self.project_path.join(DATA_DIR);
         fs::create_dir_all(&data_dir).map_err(|source| PersistenceError::Io {
             operation: "create",
diff --git a/tests/iteration2_persistence.rs b/tests/iteration2_persistence.rs
index 3d1b7b6..b1a0535 100644
--- a/tests/iteration2_persistence.rs
+++ b/tests/iteration2_persistence.rs
@@ -240,11 +240,89 @@ fn delete_with_cascade_rewrites_both_csvs() {
         assert_eq!(result.rows_affected, 1);
     });
 
-    let customers_csv = read_csv(&path, "Customers").expect("Customers.csv");
-    let orders_csv = read_csv(&path, "Orders").expect("Orders.csv");
-    // Both CSVs should be header-only after cascade.
-    assert_eq!(customers_csv.lines().count(), 1, "got: {customers_csv}");
-    assert_eq!(orders_csv.lines().count(), 1, "got: {orders_csv}");
+    // Both CSVs should be gone after the cascade leaves both
+    // tables empty: empty table -> no CSV (the rule from
+    // Persistence::write_table_data; see ADR-0015 §4 commentary).
+    assert!(
+        read_csv(&path, "Customers").is_none(),
+        "Customers.csv should be gone after cascade leaves it empty",
+    );
+    assert!(
+        read_csv(&path, "Orders").is_none(),
+        "Orders.csv should be gone after cascade leaves it empty",
+    );
+}
+
+#[test]
+fn create_table_does_not_write_csv_for_empty_table() {
+    let data = tempdir();
+    let (_p, db, path) = open_project(&data);
+
+    rt().block_on(async {
+        db.create_table(
+            "Customers".to_string(),
+            vec![
+                ColumnSpec { name: "id".to_string(), ty: Type::Serial },
+                ColumnSpec { name: "Name".to_string(), ty: Type::Text },
+            ],
+            vec!["id".to_string()],
+            Some("create table Customers with pk id:serial".to_string()),
+        )
+        .await
+        .unwrap();
+    });
+
+    // Schema landed in YAML.
+    let yaml = read_yaml(&path);
+    assert!(yaml.contains("- name: Customers"), "yaml missing table:\n{yaml}");
+    // ...but no CSV until there's data.
+    assert!(
+        read_csv(&path, "Customers").is_none(),
+        "no CSV should exist for an empty table",
+    );
+}
+
+#[test]
+fn delete_all_rows_removes_csv() {
+    let data = tempdir();
+    let (_p, db, path) = open_project(&data);
+
+    rt().block_on(async {
+        db.create_table(
+            "Customers".to_string(),
+            vec![
+                ColumnSpec { name: "id".to_string(), ty: Type::Serial },
+                ColumnSpec { name: "Name".to_string(), ty: Type::Text },
+            ],
+            vec!["id".to_string()],
+            Some("create table Customers with pk id:serial".to_string()),
+        )
+        .await
+        .unwrap();
+        db.insert(
+            "Customers".to_string(),
+            None,
+            vec![Value::Text("Alice".to_string())],
+            Some("insert into Customers ('Alice')".to_string()),
+        )
+        .await
+        .unwrap();
+        // CSV exists once there's data.
+        assert!(read_csv(&path, "Customers").is_some());
+
+        db.delete(
+            "Customers".to_string(),
+            RowFilter::AllRows,
+            Some("delete from Customers --all-rows".to_string()),
+        )
+        .await
+        .unwrap();
+    });
+
+    assert!(
+        read_csv(&path, "Customers").is_none(),
+        "CSV should be removed when the table becomes empty",
+    );
 }
 
 #[test]