diff --git a/lea/conductor.py b/lea/conductor.py
index e2e1f98..6c8ea43 100644
--- a/lea/conductor.py
+++ b/lea/conductor.py
@@ -425,7 +425,7 @@ def name_user_dataset(self) -> str:
     def list_materialized_audit_table_refs(
         self, database_client: DatabaseClient, dataset: str
     ) -> set[TableRef]:
-        existing_tables = database_client.list_tables(dataset)
+        existing_tables = database_client.list_table_stats(dataset)
         existing_audit_tables = {
             table_ref: stats
             for table_ref, stats in existing_tables.items()
diff --git a/lea/databases.py b/lea/databases.py
index 30ed106..1d2fff8 100644
--- a/lea/databases.py
+++ b/lea/databases.py
@@ -39,6 +39,9 @@ class DatabaseClient(typing.Protocol):
     def create_dataset(self, dataset_name: str):
         pass
 
+    def delete_dataset(self, dataset_name: str):
+        pass
+
     def materialize_script(self, script: scripts.Script) -> DatabaseJob:
         pass
 
@@ -58,7 +61,10 @@ def delete_and_insert(
     def delete_table(self, table_ref: scripts.TableRef) -> DatabaseJob:
         pass
 
-    def list_tables(self, dataset_name: str) -> dict[scripts.TableRef, TableStats]:
+    def list_table_stats(self, dataset_name: str) -> dict[scripts.TableRef, TableStats]:
+        pass
+
+    def list_table_fields(self, dataset_name: str) -> dict[scripts.TableRef, list[scripts.Field]]:
         pass
 
 
@@ -102,7 +108,7 @@ def exception(self) -> Exception:
         return self.query_job.exception()
 
 
-@dataclasses.dataclass
+@dataclasses.dataclass(frozen=True)
 class TableStats:
     n_rows: int
     n_bytes: int
@@ -129,8 +135,6 @@ def __init__(
         self.dry_run = dry_run
 
     def create_dataset(self, dataset_name: str):
-        from google.cloud import bigquery
-
         dataset_ref = bigquery.DatasetReference(
             project=self.write_project_id, dataset_id=dataset_name
         )
@@ -236,7 +240,7 @@ def delete_table(self, table_ref: scripts.TableRef) -> BigQueryJob:
             client=self, query_job=self.client.query(delete_code, job_config=job_config)
         )
 
-    def list_tables(self, dataset_name: str) -> dict[scripts.TableRef, TableStats]:
+    def list_table_stats(self, dataset_name: str) -> dict[scripts.TableRef, TableStats]:
         query = f"""
         SELECT table_id, row_count, size_bytes
         FROM `{self.write_project_id}.{dataset_name}.__TABLES__`
@@ -249,6 +253,22 @@ def list_tables(self, dataset_name: str) -> dict[scripts.TableRef, TableStats]:
             for row in job.result()
         }
 
+    def list_table_fields(self, dataset_name: str) -> dict[scripts.TableRef, set[scripts.Field]]:
+        query = f"""
+        SELECT table_name, column_name
+        FROM `{self.write_project_id}.{dataset_name}.INFORMATION_SCHEMA.COLUMNS`
+        """
+        job = self.client.query(query)
+        return {
+            BigQueryDialect.parse_table_ref(f"{dataset_name}.{table_name}"): [
+                scripts.Field(name=row["column_name"]) for _, row in rows.iterrows()
+            ]
+            for table_name, rows in job.result()
+            .to_dataframe()
+            .sort_values(["table_name", "column_name"])
+            .groupby("table_name")
+        }
+
     def make_job_config(
         self, script: scripts.SQLScript | None = None, **kwargs
     ) -> bigquery.QueryJobConfig:
diff --git a/lea/field.py b/lea/field.py
index 1e425cc..0bae9b3 100644
--- a/lea/field.py
+++ b/lea/field.py
@@ -4,11 +4,11 @@
 import enum
 
 
-@dataclasses.dataclass
+@dataclasses.dataclass(frozen=True)
 class Field:
     name: str
-    tags: set[FieldTag]
-    description: str
+    tags: set[FieldTag] = dataclasses.field(default_factory=set)
+    description: str | None = None
 
     @property
     def is_unique(self):