diff --git a/CLAUDE.md b/CLAUDE.md
index 81270d8..96eb49e 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -170,7 +170,7 @@ Auth providers in `app/auth/` (FastAPI-based):
 ## Key Implementation Details
 
 ### DuckDB Schema (src/db.py)
-- Schema v3 with auto-migration from v1→v2→v3
+- Schema v4 with auto-migration from v1→v2→v3→v4
 - `table_registry`: id, name, source_type, bucket, source_table, query_mode, sync_schedule, etc.
 - `sync_state`, `sync_history`: track extraction progress
 - `users`, `dataset_permissions`, `audit_log`: auth + RBAC
diff --git a/app/api/metrics.py b/app/api/metrics.py
index a608559..ec32911 100644
--- a/app/api/metrics.py
+++ b/app/api/metrics.py
@@ -1,7 +1,5 @@
 """Metrics API endpoints — CRUD for metric definitions stored in DuckDB."""
 
-import os
-import tempfile
 from typing import List, Optional
 
 import duckdb
@@ -119,14 +117,60 @@ async def import_metrics(
 ):
     """Import metrics from uploaded YAML file."""
     content = await file.read()
-
-    with tempfile.NamedTemporaryFile(suffix=".yml", delete=False, mode="wb") as tmp:
-        tmp.write(content)
-        tmp_path = tmp.name
-
     try:
-        repo = MetricRepository(conn)
-        count = repo.import_from_yaml(tmp_path)
-        return {"status": "imported", "count": count}
-    finally:
-        os.unlink(tmp_path)
+        data = yaml.safe_load(content)
+    except yaml.YAMLError as e:
+        raise HTTPException(status_code=400, detail=f"Invalid YAML: {e}")
+
+    if not data:
+        raise HTTPException(status_code=400, detail="Empty YAML file")
+
+    metric_list = data if isinstance(data, list) else [data]
+    repo = MetricRepository(conn)
+    count = 0
+
+    for metric in metric_list:
+        if not isinstance(metric, dict):
+            continue
+        name = metric.get("name")
+        category = metric.get("category")
+        if not name or not category:
+            raise HTTPException(
+                status_code=400,
+                detail="Each metric must have 'name' and 'category' fields",
+            )
+
+        metric_id = f"{category}/{name}"
+        table_name = metric.pop("table", None) or metric.get("table_name")
+
+        # Collect sql_by_* variants
+        sql_variants = {}
+        for key in list(metric.keys()):
+            if key.startswith("sql_by_"):
+                sql_variants[key[4:]] = metric.pop(key)
+
+        repo.create(
+            id=metric_id,
+            name=name,
+            display_name=metric.get("display_name", name),
+            category=category,
+            description=metric.get("description"),
+            type=metric.get("type", "sum"),
+            unit=metric.get("unit"),
+            grain=metric.get("grain", "monthly"),
+            table_name=table_name,
+            tables=metric.get("tables"),
+            expression=metric.get("expression"),
+            time_column=metric.get("time_column"),
+            dimensions=metric.get("dimensions"),
+            filters=metric.get("filters"),
+            synonyms=metric.get("synonyms"),
+            notes=metric.get("notes"),
+            sql=metric.get("sql", ""),
+            sql_variants=sql_variants if sql_variants else None,
+            validation=metric.get("validation"),
+            source="yaml_import",
+        )
+        count += 1
+
+    return {"status": "imported", "count": count}
diff --git a/cli/commands/analyst.py b/cli/commands/analyst.py
index 4534148..b9f9f32 100644
--- a/cli/commands/analyst.py
+++ b/cli/commands/analyst.py
@@ -197,6 +197,8 @@ def _download_data(workspace: Path, server_url: str, token: str) -> int:
             typer.echo(f"  Downloaded {table_id}")
         except Exception as e:
             typer.echo(f"  Warning: could not download {table_id}: {e}", err=True)
+            if target.exists():
+                target.unlink()
 
     return downloaded
 
@@ -231,10 +233,11 @@ def _initialize_duckdb(workspace: Path) -> int:
             typer.echo(f"  Warning: Skipping {pq_file.name}: unsafe view name", err=True)
             continue
         abs_path = str(pq_resolved)
+        safe_path = abs_path.replace("'", "''")
         try:
             conn.execute(f'DROP VIEW IF EXISTS "{view_name}"')
             conn.execute(
-                f"CREATE VIEW \"{view_name}\" AS SELECT * FROM read_parquet('{abs_path}')"
+                f"CREATE VIEW \"{view_name}\" AS SELECT * FROM read_parquet('{safe_path}')"
             )
             count = conn.execute(f'SELECT count(*) FROM "{view_name}"').fetchone()[0]
             total_rows += count