fix: validate batch_size and empty-query benchmarks

src/tenantq/benchmark.py

@@ -152,8 +152,8 @@ def run_benchmark(
                 idx_recall.append(recall_at_k(retrieved, ref, 10))
 
         total_time = sum(latencies) / 1000.0
-        avg_r5 = sum(r5) / len(r5)
-        avg_r10 = sum(r10) / len(r10)
+        avg_r5 = sum(r5) / len(r5) if r5 else 0.0
+        avg_r10 = sum(r10) / len(r10) if r10 else 0.0
         RECALL_GAUGE.labels(mode=mode, k="5").set(avg_r5)
         RECALL_GAUGE.labels(mode=mode, k="10").set(avg_r10)
         result.modes.append(

src/tenantq/embeddings.py

@@ -123,5 +123,7 @@ def build_embedder(kind: str, dense_model: str, sparse_model: str, dense_dim: in
 
 
 def batched(seq: Sequence, size: int) -> Iterable[Sequence]:
+    if size < 1:
+        raise ValueError("batch_size must be >= 1")
     for i in range(0, len(seq), size):
         yield seq[i : i + size]

tests/test_benchmark.py

@@ -5,6 +5,7 @@
 import math
 
 from tenantq.benchmark import percentile, recall_at_k, run_benchmark
+from tenantq.data import Dataset
 
 
 def test_percentile_and_recall_helpers():
@@ -31,3 +32,15 @@ def test_benchmark_produces_numbers(ingested, settings, embedder, dataset):
     assert not math.isnan(modes["dense"].index_recall_at_10)
     # hybrid should be at least as good as the weaker single mode on recall@10
     assert modes["hybrid"].recall_at_10 >= min(modes["dense"].recall_at_10, modes["sparse"].recall_at_10)
+
+
+def test_benchmark_zero_queries_returns_empty_modes(ingested, settings, embedder):
+    empty = Dataset(documents=[], queries=[])
+    result = run_benchmark(ingested, settings, embedder, empty)
+    assert result.settings_summary["n_queries"] == 0
+    assert result.modes
+    for m in result.modes:
+        assert m.n_queries == 0
+        assert m.recall_at_5 == 0.0
+        assert m.recall_at_10 == 0.0
+        assert m.qps == 0.0

tests/test_ingest.py

@@ -0,0 +1,25 @@
+"""Ingestion boundary checks."""
+
+from __future__ import annotations
+
+import pytest
+
+from tenantq.embeddings import batched
+from tenantq.ingest import ingest_documents
+
+
+def test_batched_rejects_non_positive_size():
+    with pytest.raises(ValueError, match="batch_size must be >= 1"):
+        list(batched([1, 2, 3], 0))
+    with pytest.raises(ValueError, match="batch_size must be >= 1"):
+        list(batched([1, 2, 3], -1))
+
+
+def test_ingest_documents_rejects_zero_batch_size(settings, embedder, dataset):
+    from tenantq.client import make_client
+    from tenantq.collection import recreate_collection
+
+    client = make_client(settings)
+    recreate_collection(client, settings)
+    with pytest.raises(ValueError, match="batch_size must be >= 1"):
+        ingest_documents(client, settings, embedder, dataset.documents[:3], batch_size=0)