diff --git a/docs/OFFLOADING_DECISION_EVENTS.md b/docs/OFFLOADING_DECISION_EVENTS.md
index e53ed14..af2669b 100644
--- a/docs/OFFLOADING_DECISION_EVENTS.md
+++ b/docs/OFFLOADING_DECISION_EVENTS.md
@@ -111,6 +111,7 @@ python scripts/analysis/offloading_decision_summary.py \
 The summary includes selected layer, strategy, estimated total milliseconds,
 observed total milliseconds, and error milliseconds.
 
-> Note: with segment splitting enabled the base `offloading_decisions.csv` is not
-> written. Point the script at a specific segment, or disable splitting for a
-> single-file run.
+The script reads a whole run via `iter_run_offloading_decisions`, so pointing
+it at the base name (e.g. `offloading_decisions.csv`) covers every split
+segment for that run, even though the base file itself is never written once
+splitting is enabled.
diff --git a/scripts/analysis/offloading_decision_summary.py b/scripts/analysis/offloading_decision_summary.py
index d19ee8b..48f534f 100644
--- a/scripts/analysis/offloading_decision_summary.py
+++ b/scripts/analysis/offloading_decision_summary.py
@@ -7,7 +7,7 @@
 
 from server.commons import EvaluationFiles
 from server.telemetry.offloading_decisions import (
-    iter_offloading_decisions,
+    iter_run_offloading_decisions,
     summarize_prediction_error,
 )
 
@@ -27,7 +27,16 @@
 
 
 def summarize_file(path: str | Path) -> list[dict]:
-    return [summarize_prediction_error(event) for event in iter_offloading_decisions(path)]
+    """Summarize one run's offloading decisions, across all split segments.
+
+    ``path`` may point at a single legacy file or at any segment's base name
+    (e.g. ``offloading_decisions.csv``); ``iter_run_offloading_decisions``
+    discovers and reads every ``.segNNNN_*`` segment for that run, in order.
+    """
+    return [
+        summarize_prediction_error(event)
+        for event in iter_run_offloading_decisions(path)
+    ]
 
 
 def main(argv: list[str] | None = None) -> int:
@@ -38,7 +47,10 @@ def main(argv: list[str] | None = None) -> int:
         "input",
         nargs="?",
         default=EvaluationFiles.offloading_decisions_file_path,
-        help="Path to offloading_decisions.csv or a legacy JSONL file.",
+        help=(
+            "Path to offloading_decisions.csv (or a legacy JSONL file). "
+            "All split segments for the run are read automatically."
+        ),
     )
     parser.add_argument(
         "--output",
diff --git a/tests/unit/test_offloading_decision_summary.py b/tests/unit/test_offloading_decision_summary.py
index ef6c3df..1fab42a 100644
--- a/tests/unit/test_offloading_decision_summary.py
+++ b/tests/unit/test_offloading_decision_summary.py
@@ -42,3 +42,24 @@ def test_offloading_decision_summary_reads_csv(tmp_path):
             "error_ms": 5.0,
         }
     ]
+
+
+def test_offloading_decision_summary_reads_all_split_segments(tmp_path):
+    base = tmp_path / "offloading_decisions.csv"
+    for i in range(3):
+        event = build_offloading_decision_event(
+            device_id="device-1",
+            model_dir="model",
+            request_id=f"REQ{i}",
+            selected_layer=i,
+            selection_reason="lowest_estimated_cost",
+            avg_speed_bytes_per_second=1,
+        )
+        append_offloading_decision(base, event, max_rows=1)
+
+    # Splitting keeps segments and never writes the base file directly.
+    assert not base.exists()
+
+    rows = summarize_file(base)
+
+    assert [row["request_id"] for row in rows] == ["REQ0", "REQ1", "REQ2"]