diff --git a/divref/divref/duckdb_index.py b/divref/divref/duckdb_index.py
index 4a3f916..bf16c3c 100644
--- a/divref/divref/duckdb_index.py
+++ b/divref/divref/duckdb_index.py
@@ -62,8 +62,12 @@ def read_and_validate_pops_legends(table_pairs: list[TablePair]) -> tuple[list[s
         A tuple of `(gnomad_pops_legend, hgdp_pops_legend)`.
 
     Raises:
-        ValueError: If any contig's gnomAD or HGDP legend disagrees with the bootstrapped legend.
+        ValueError: If `table_pairs` is empty, or if any contig's gnomAD or HGDP legend disagrees
+            with the bootstrapped legend.
     """
+    if not table_pairs:
+        raise ValueError("read_and_validate_pops_legends requires at least one table pair.")
+
     first_with_hap: TablePair | None = next(
         (tp for tp in table_pairs if tp.haplotype_table_path is not None), None
     )
@@ -134,17 +138,25 @@ def write_metadata_tables(
         joint_pops_legend: Joint population codes stored as JSON in `joint_pops_legend`.
         version: Version identifier stored in the `VERSION` table.
     """
-    conn.execute("CREATE TABLE window_size AS SELECT ? AS window_size", [window_size])
-    conn.execute(
-        "CREATE TABLE hgdp_haplotype_pops_legend AS SELECT ? AS pops_legend",
-        [json.dumps(hgdp_pops_legend)],
-    )
-    conn.execute(
-        "CREATE TABLE gnomad_variant_pops_legend AS SELECT ? AS pops_legend",
-        [json.dumps(gnomad_pops_legend)],
-    )
-    conn.execute(
-        "CREATE TABLE joint_pops_legend AS SELECT ? AS pops_legend",
-        [json.dumps(joint_pops_legend)],
-    )
-    conn.execute("CREATE TABLE VERSION AS SELECT ? AS version", [version])
+    # Write the five tables in one transaction so an interrupted init leaves no partially
+    # populated index (which a later append/finalize would then read as corrupt metadata).
+    conn.execute("BEGIN TRANSACTION")
+    try:
+        conn.execute("CREATE TABLE window_size AS SELECT ? AS window_size", [window_size])
+        conn.execute(
+            "CREATE TABLE hgdp_haplotype_pops_legend AS SELECT ? AS pops_legend",
+            [json.dumps(hgdp_pops_legend)],
+        )
+        conn.execute(
+            "CREATE TABLE gnomad_variant_pops_legend AS SELECT ? AS pops_legend",
+            [json.dumps(gnomad_pops_legend)],
+        )
+        conn.execute(
+            "CREATE TABLE joint_pops_legend AS SELECT ? AS pops_legend",
+            [json.dumps(joint_pops_legend)],
+        )
+        conn.execute("CREATE TABLE VERSION AS SELECT ? AS version", [version])
+        conn.execute("COMMIT")
+    except BaseException:
+        conn.execute("ROLLBACK")
+        raise
diff --git a/divref/divref/hail.py b/divref/divref/hail.py
index 009978f..e5aa026 100644
--- a/divref/divref/hail.py
+++ b/divref/divref/hail.py
@@ -5,6 +5,28 @@
 import pyspark
 
 
+def _export_gcs_credentials(gcs_credentials_path: Path | None) -> None:
+    """
+    Export GOOGLE_APPLICATION_CREDENTIALS for the JVM subprocess (GCS mode only).
+
+    Args:
+        gcs_credentials_path: Path to the ADC JSON file; must be provided and must exist.
+
+    Raises:
+        ValueError: If `gcs_credentials_path` is None.
+        FileNotFoundError: If the file does not exist.
+    """
+    if gcs_credentials_path is None:
+        raise ValueError("gcs_credentials_path is required when use_s3 is False.")
+    if not gcs_credentials_path.exists():
+        raise FileNotFoundError(
+            f"GCS credentials file not found at {gcs_credentials_path}. Run "
+            "`gcloud auth application-default login` or pass a valid --gcs-credentials-path."
+        )
+    if "GOOGLE_APPLICATION_CREDENTIALS" not in os.environ:
+        os.environ["GOOGLE_APPLICATION_CREDENTIALS"] = str(gcs_credentials_path)
+
+
 def hail_init(
     gcs_credentials_path: Path | None = None,
     spark_driver_memory_gb: int = 1,
@@ -25,9 +47,9 @@ def hail_init(
 
     Args:
         gcs_credentials_path: Absolute path to a GCP Application Default Credentials
-            JSON file. Required when `use_s3` is `False`; ignored otherwise. If the
-            file exists and `GOOGLE_APPLICATION_CREDENTIALS` is not already set, it
-            is exported to the environment before Hail starts.
+            JSON file. Required (and must exist) when `use_s3` is `False`; ignored
+            otherwise. When `GOOGLE_APPLICATION_CREDENTIALS` is not already set, it is
+            exported to the environment before Hail starts.
         spark_driver_memory_gb: Memory in GB to allocate to the Spark driver.
         spark_executor_memory_gb: Memory in GB to allocate to the Spark executor.
         use_s3: If `True`, validate and load the S3A connector JARs and configure
@@ -38,9 +60,9 @@ def hail_init(
         ValueError: If `spark_driver_memory_gb` or `spark_executor_memory_gb`
             is less than 1, or if `use_s3` is `False` and `gcs_credentials_path`
             is `None`.
-        FileNotFoundError: If `use_s3` is `False` and the GCS connector JAR is
-            missing, or if `use_s3` is `True` and either S3A connector JAR is
-            missing.
+        FileNotFoundError: If `use_s3` is `False` and either the credentials file at
+            `gcs_credentials_path` or the GCS connector JAR is missing, or if `use_s3`
+            is `True` and either S3A connector JAR is missing.
     """
     if spark_driver_memory_gb < 1:
         raise ValueError(
@@ -58,10 +80,7 @@ def hail_init(
     )
 
     if not use_s3:
-        if gcs_credentials_path is None:
-            raise ValueError("gcs_credentials_path is required when use_s3 is False.")
-        if gcs_credentials_path.exists() and "GOOGLE_APPLICATION_CREDENTIALS" not in os.environ:
-            os.environ["GOOGLE_APPLICATION_CREDENTIALS"] = str(gcs_credentials_path)
+        _export_gcs_credentials(gcs_credentials_path)
 
     jars_dir = os.path.join(pyspark.__path__[0], "jars")
     cloud_jars: list[str] = []
diff --git a/divref/divref/tools/finalize_duckdb_index.py b/divref/divref/tools/finalize_duckdb_index.py
index bb9483e..9ba4a78 100644
--- a/divref/divref/tools/finalize_duckdb_index.py
+++ b/divref/divref/tools/finalize_duckdb_index.py
@@ -37,6 +37,12 @@ def finalize_duckdb_index(
                 f"DuckDB index {out_duckdb_file} has no 'sequences' table; "
                 f"run append_contig_to_duckdb_index before finalizing."
             )
+        row_count_row = conn.execute("SELECT COUNT(*) FROM sequences").fetchone()
+        if row_count_row is not None and row_count_row[0] == 0:
+            logger.warning(
+                f"DuckDB index {out_duckdb_file} has an empty 'sequences' table; "
+                f"finalizing an index with no rows."
+            )
         # IF NOT EXISTS keeps finalize idempotent: a re-triggered Snakemake rule or a manual retry
         # over an already-finalized index is a no-op rather than a raw DuckDB "index already exists"
         # catalog error.
diff --git a/divref/tests/test_hail.py b/divref/tests/test_hail.py
new file mode 100644
index 0000000..879425c
--- /dev/null
+++ b/divref/tests/test_hail.py
@@ -0,0 +1,19 @@
+"""Tests for divref.hail.hail_init input validation (no Hail context required)."""
+
+from pathlib import Path
+
+import pytest
+
+from divref.hail import hail_init
+
+
+def test_hail_init_missing_credentials_file_raises(tmp_path: Path) -> None:
+    """GCS mode with a provided-but-missing credentials file fails loudly before starting Hail."""
+    with pytest.raises(FileNotFoundError, match="credentials file not found"):
+        hail_init(gcs_credentials_path=tmp_path / "does_not_exist.json", use_s3=False)
+
+
+def test_hail_init_requires_credentials_for_gcs() -> None:
+    """GCS mode with no credentials path raises ValueError."""
+    with pytest.raises(ValueError, match="gcs_credentials_path is required"):
+        hail_init(gcs_credentials_path=None, use_s3=False)
diff --git a/divref/tests/tools/test_finalize_duckdb_index.py b/divref/tests/tools/test_finalize_duckdb_index.py
index 16e1fd4..df4e1e2 100644
--- a/divref/tests/tools/test_finalize_duckdb_index.py
+++ b/divref/tests/tools/test_finalize_duckdb_index.py
@@ -1,5 +1,6 @@
 """Tests for the finalize_duckdb_index tool."""
 
+import logging
 from pathlib import Path
 
 import duckdb
@@ -10,6 +11,21 @@
 from divref.tools.init_duckdb_index import init_duckdb_index
 
 
+def test_finalize_warns_on_empty_sequences(
+    tmp_path: Path,
+    caplog: pytest.LogCaptureFixture,
+) -> None:
+    """Finalizing an index whose `sequences` table has no rows logs a warning (pure DuckDB)."""
+    output_base = tmp_path / "idx"
+    with duckdb.connect(str(Path(f"{output_base}.haplotypes_gnomad_merge.index.duckdb"))) as conn:
+        conn.execute("CREATE TABLE sequences (sequence_id VARCHAR)")
+
+    with caplog.at_level(logging.WARNING):
+        finalize_duckdb_index(output_base=output_base)
+
+    assert "empty 'sequences' table" in caplog.text
+
+
 def _write_table_pairs_tsv(path: Path, rows: list[tuple[str, str, str]]) -> Path:
     """Write a table-pairs TSV with a (contig, haplotype_table_path, sites_table_path) per row."""
     lines = ["contig\thaplotype_table_path\tsites_table_path"]
diff --git a/divref/tests/tools/test_init_duckdb_index.py b/divref/tests/tools/test_init_duckdb_index.py
index 90de15d..308f7e6 100644
--- a/divref/tests/tools/test_init_duckdb_index.py
+++ b/divref/tests/tools/test_init_duckdb_index.py
@@ -6,11 +6,18 @@
 import pytest
 
 from divref.duckdb_index import compute_joint_legend
+from divref.duckdb_index import read_and_validate_pops_legends
 from divref.tools.append_contig_to_duckdb_index import read_legend
 from divref.tools.append_contig_to_duckdb_index import read_window_size
 from divref.tools.init_duckdb_index import init_duckdb_index
 
 
+def test_read_and_validate_pops_legends_empty_raises() -> None:
+    """An empty table-pair list is rejected with a clear error rather than an IndexError."""
+    with pytest.raises(ValueError, match="at least one table pair"):
+        read_and_validate_pops_legends([])
+
+
 def _write_table_pairs_tsv(path: Path, rows: list[tuple[str, str, str]]) -> Path:
     """Write a table-pairs TSV with a (contig, haplotype_table_path, sites_table_path) per row."""
     lines = ["contig\thaplotype_table_path\tsites_table_path"]