diff --git a/plugins/data-designer-retrieval-sdg/src/data_designer_retrieval_sdg/postprocess.py b/plugins/data-designer-retrieval-sdg/src/data_designer_retrieval_sdg/postprocess.py
index 1bbff7f..e31085f 100644
--- a/plugins/data-designer-retrieval-sdg/src/data_designer_retrieval_sdg/postprocess.py
+++ b/plugins/data-designer-retrieval-sdg/src/data_designer_retrieval_sdg/postprocess.py
@@ -190,6 +190,8 @@ def load_positive_docs_with_modality(
         Tuple of ``(positive_docs_df, doc_to_modality_final)``.
     """
     qrels_df = pd.read_csv(test_tsv_path, sep="\t")
+    if "corpus-id" not in qrels_df.columns and "corpus-id " in qrels_df.columns:
+        qrels_df = qrels_df.rename(columns={"corpus-id ": "corpus-id"})
 
     with open(split_json_path, encoding="utf-8") as f:
         splits = json.load(f)
@@ -202,7 +204,7 @@ def load_positive_docs_with_modality(
     doc_to_modality: dict[str, set[str]] = defaultdict(set)
     for _, row in qrels_df.iterrows():
         query_id = row["query-id"]
-        corpus_id = row["corpus-id "]  # trailing space in column name
+        corpus_id = row["corpus-id"]
         if query_id in query_to_modality:
             doc_to_modality[corpus_id].add(query_to_modality[query_id])
 
@@ -212,7 +214,7 @@ def load_positive_docs_with_modality(
             doc_to_modality_final[doc_id] = next(iter(modalities))
         else:
             modality_counts: dict[str, int] = defaultdict(int)
-            for _, r in qrels_df[qrels_df["corpus-id "] == doc_id].iterrows():
+            for _, r in qrels_df[qrels_df["corpus-id"] == doc_id].iterrows():
                 qid = r["query-id"]
                 if qid in query_to_modality:
                     modality_counts[query_to_modality[qid]] += 1
diff --git a/plugins/data-designer-retrieval-sdg/tests/test_postprocess.py b/plugins/data-designer-retrieval-sdg/tests/test_postprocess.py
index f73b24d..8cc9a25 100644
--- a/plugins/data-designer-retrieval-sdg/tests/test_postprocess.py
+++ b/plugins/data-designer-retrieval-sdg/tests/test_postprocess.py
@@ -1,9 +1,17 @@
 # SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 # SPDX-License-Identifier: Apache-2.0
 
+import json
+from pathlib import Path
+
 import pandas as pd
+import pytest
 
-from data_designer_retrieval_sdg.postprocess import filter_qa_pairs_by_quality, postprocess_retriever_data
+from data_designer_retrieval_sdg.postprocess import (
+    filter_qa_pairs_by_quality,
+    load_positive_docs_with_modality,
+    postprocess_retriever_data,
+)
 
 # ---------------------------------------------------------------------------
 # postprocess_retriever_data
@@ -85,3 +93,45 @@ def test_filter_skips_mismatched() -> None:
     filtered_df, skipped = filter_qa_pairs_by_quality(df, quality_threshold=5.0)
     assert len(filtered_df) == 0
     assert len(skipped) == 1
+
+
+# ---------------------------------------------------------------------------
+# load_positive_docs_with_modality
+# ---------------------------------------------------------------------------
+
+
+@pytest.mark.parametrize("corpus_id_header", ["corpus-id", "corpus-id "])
+def test_load_positive_docs_with_modality_accepts_corpus_id_headers(tmp_path: Path, corpus_id_header: str) -> None:
+    """Canonical and legacy corpus ID headers produce the same result."""
+    qrels_path = tmp_path / "test.tsv"
+    qrels_path.write_text(f"query-id\t{corpus_id_header}\tscore\nq1\td1\t1\n", encoding="utf-8")
+
+    corpus_path = tmp_path / "corpus.jsonl"
+    corpus_path.write_text(json.dumps({"_id": "d1", "text": "hello world", "title": "T"}) + "\n", encoding="utf-8")
+
+    split_path = tmp_path / "split.json"
+    split_path.write_text(json.dumps({"text": ["q1"]}), encoding="utf-8")
+
+    docs_df, modality_map = load_positive_docs_with_modality(qrels_path, corpus_path, split_path)
+
+    assert len(docs_df) == 1
+    assert docs_df.iloc[0]["doc_id"] == "d1"
+    assert docs_df.iloc[0]["modality"] == "text"
+    assert modality_map == {"d1": "text"}
+
+
+def test_load_positive_docs_with_modality_uses_most_common_modality(tmp_path: Path) -> None:
+    """A document shared across modalities uses its most frequent modality."""
+    qrels_path = tmp_path / "test.tsv"
+    qrels_path.write_text("query-id\tcorpus-id\tscore\nq1\td1\t1\nq2\td1\t1\nq3\td1\t1\n", encoding="utf-8")
+
+    corpus_path = tmp_path / "corpus.jsonl"
+    corpus_path.write_text(json.dumps({"_id": "d1", "text": "hello world", "title": "T"}) + "\n", encoding="utf-8")
+
+    split_path = tmp_path / "split.json"
+    split_path.write_text(json.dumps({"text": ["q1", "q2"], "image": ["q3"]}), encoding="utf-8")
+
+    docs_df, modality_map = load_positive_docs_with_modality(qrels_path, corpus_path, split_path)
+
+    assert docs_df.iloc[0]["modality"] == "text"
+    assert modality_map == {"d1": "text"}