fix(pipeline): use sqlmesh plan --auto-apply instead of run

Missed the Python supervisor module — same fix as supervisor.sh and
worker.py.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

infra/supervisor/supervisor.sh

@@ -33,10 +33,10 @@ do
         DUCKDB_PATH="${DUCKDB_PATH:-/data/padelnomics/lakehouse.duckdb}" \
             uv run --package padelnomics_extract extract
 
-        # Transform — run evaluates missing daily intervals for incremental models.
+        # Transform — plan detects new/modified/deleted models and applies changes.
         LANDING_DIR="${LANDING_DIR:-/data/padelnomics/landing}" \
         DUCKDB_PATH="${DUCKDB_PATH:-/data/padelnomics/lakehouse.duckdb}" \
-            uv run sqlmesh -p transform/sqlmesh_padelnomics run prod
+            uv run sqlmesh -p transform/sqlmesh_padelnomics plan prod --auto-apply
 
         # Export serving tables to analytics.duckdb (atomic swap).
         # The web app detects the inode change on next query — no restart needed.

src/padelnomics/supervisor.py

@@ -247,10 +247,10 @@ def run_shell(cmd: str, timeout_seconds: int = SUBPROCESS_TIMEOUT_SECONDS) -> tu
 
 
 def run_transform() -> None:
-    """Run SQLMesh — evaluates missing daily intervals."""
+    """Run SQLMesh — detects new/modified/deleted models and applies changes."""
     logger.info("Running SQLMesh transform")
     ok, err = run_shell(
-        "uv run sqlmesh -p transform/sqlmesh_padelnomics run prod",
+        "uv run sqlmesh -p transform/sqlmesh_padelnomics plan prod --auto-apply",
     )
     if not ok:
         send_alert(f"[transform] {err}")

web/src/padelnomics/worker.py

@@ -737,9 +737,9 @@ async def handle_run_extraction(payload: dict) -> None:
 
 @task("run_transform")
 async def handle_run_transform(payload: dict) -> None:
-    """Run SQLMesh transform (prod run) in the background.
+    """Run SQLMesh transform (prod plan + apply) in the background.
 
-    Shells out to `uv run sqlmesh -p transform/sqlmesh_padelnomics run prod`.
+    Shells out to `uv run sqlmesh -p transform/sqlmesh_padelnomics plan prod --auto-apply`.
     2-hour absolute timeout — same as extraction.
     """
     import subprocess
@@ -748,7 +748,7 @@ async def handle_run_transform(payload: dict) -> None:
     repo_root = Path(__file__).resolve().parents[4]
     result = await asyncio.to_thread(
         subprocess.run,
-        ["uv", "run", "sqlmesh", "-p", "transform/sqlmesh_padelnomics", "run", "prod"],
+        ["uv", "run", "sqlmesh", "-p", "transform/sqlmesh_padelnomics", "plan", "prod", "--auto-apply"],
         capture_output=True,
         text=True,
         timeout=7200,
@@ -803,7 +803,7 @@ async def handle_run_pipeline(payload: dict) -> None:
         ),
         (
             "transform",
-            ["uv", "run", "sqlmesh", "-p", "transform/sqlmesh_padelnomics", "run", "prod"],
+            ["uv", "run", "sqlmesh", "-p", "transform/sqlmesh_padelnomics", "plan", "prod", "--auto-apply"],
             7200,
         ),
         (