fix: skip silent daily fallback for unknown freq

src/sktime_mcp/runtime/executor.py

@@ -1430,7 +1430,10 @@ def format_data_handle(
                         elif most_common_diff.days >= 28 and most_common_diff.days <= 31:
                             freq = "MS"
                         else:
-                            freq = "D"
+                            changes_made["frequency_warning"] = (
+                                f"Could not determine frequency from most common interval "
+                                f"({most_common_diff}). Reindexing skipped."
+                            )
 
                 # Create complete date range
                 if freq:

src/sktime_mcp/tools/transform_data.py

@@ -145,6 +145,8 @@ def _action_format(
         changes_applied.append(
             f"Filled {changes['missing_filled']} missing values (forward/backward fill)"
         )
+    if changes.get("frequency_warning"):
+        changes_applied.append(changes["frequency_warning"])
     if not changes_applied:
         changes_applied.append("No changes needed — data was already clean")
 

tests/test_format_unrecognized_freq.py

@@ -0,0 +1,145 @@
+"""Unrecognized intervals must not be reindexed to daily (#313)."""
+
+import pandas as pd
+
+from sktime_mcp.runtime.executor import get_executor
+from sktime_mcp.tools.fit_predict import fit_tool
+from sktime_mcp.tools.inspect_data import inspect_data_tool
+from sktime_mcp.tools.instantiate import instantiate_tool
+from sktime_mcp.tools.transform_data import transform_data_tool
+
+_BIWEEKLY = [
+    "2023-01-02",
+    "2023-01-16",
+    "2023-01-30",
+    "2023-02-13",
+    "2023-02-28",
+    "2023-03-13",
+    "2023-03-27",
+    "2023-04-10",
+]
+
+
+def _register(ex, handle, index, values=None):
+    y = pd.Series(
+        values if values is not None else range(len(index)),
+        index=pd.DatetimeIndex(index),
+        dtype=float,
+        name="value",
+    )
+    ex._data_handles[handle] = {
+        "y": y,
+        "X": None,
+        "metadata": {"rows": len(y), "frequency": None},
+        "validation": {},
+        "config": {},
+    }
+    return handle
+
+
+def _pop(ex, *handles):
+    for h in handles:
+        if h:
+            ex._data_handles.pop(h, None)
+
+
+def test_irregular_biweekly_not_expanded_to_daily():
+    ex = get_executor()
+    src = _register(ex, "fmt_biweekly", _BIWEEKLY, [10, 12, 11, 14, 13, 15, 12, 16])
+    res = None
+    try:
+        res = ex.format_data_handle(src, release_original=False)
+        assert res["success"], res
+        assert res["metadata"]["rows"] == 8
+        assert not res["changes_made"].get("frequency_set")
+        assert "frequency_warning" in res["changes_made"]
+        assert "14 days" in res["changes_made"]["frequency_warning"]
+        y = ex._data_handles[res["data_handle"]]["y"]
+        assert len(y) == 8
+        assert list(y.index) == list(pd.to_datetime(_BIWEEKLY))
+    finally:
+        _pop(ex, src, res.get("data_handle") if res else None)
+
+
+def test_gapped_15min_not_collapsed_to_one_day():
+    ex = get_executor()
+    idx = pd.date_range("2023-01-01 00:00", periods=20, freq="15min").delete(7)
+    src = _register(ex, "fmt_15min", idx)
+    res = None
+    try:
+        res = ex.format_data_handle(src, release_original=False)
+        assert res["success"], res
+        assert res["metadata"]["rows"] == 19
+        assert not res["changes_made"].get("frequency_set")
+        assert "frequency_warning" in res["changes_made"]
+        assert len(ex._data_handles[res["data_handle"]]["y"]) == 19
+    finally:
+        _pop(ex, src, res.get("data_handle") if res else None)
+
+
+def test_daily_gap_still_filled():
+    ex = get_executor()
+    idx = pd.date_range("2023-01-01", periods=10, freq="D").delete(4)
+    src = _register(ex, "fmt_daily_gap", idx)
+    res = None
+    try:
+        res = ex.format_data_handle(src, release_original=False)
+        assert res["success"], res
+        assert res["changes_made"]["frequency_set"]
+        assert res["changes_made"]["frequency"] == "D"
+        assert res["metadata"]["rows"] == 10
+        assert res["changes_made"]["gaps_filled"] == 1
+    finally:
+        _pop(ex, src, res.get("data_handle") if res else None)
+
+
+def test_transform_data_surfaces_frequency_warning():
+    ex = get_executor()
+    src = _register(ex, "fmt_td_biweekly", _BIWEEKLY)
+    res = None
+    try:
+        res = transform_data_tool(data_handle=src, action="format")
+        assert res["success"], res
+        assert res["metadata"]["rows"] == 8
+        applied = " ".join(res["changes_applied"])
+        assert "Reindexing skipped" in applied
+        assert "set frequency to 'D'" not in applied
+        assert "already clean" not in applied
+    finally:
+        _pop(ex, src, res.get("data_handle") if res else None)
+
+
+def test_irregular_biweekly_load_inspect_fit_keeps_real_rows():
+    ex = get_executor()
+    load = ex.load_data_source(
+        {
+            "type": "pandas",
+            "data": {
+                "date": _BIWEEKLY,
+                "value": [10.0, 12, 11, 14, 13, 15, 12, 16],
+                "temp": [1.0, 2, 3, 4, 5, 6, 7, 8],
+            },
+            "time_column": "date",
+            "target_column": "value",
+        }
+    )
+    inst = instantiate_tool(spec="NaiveForecaster(strategy='last')")
+    handle = inst["handle"]
+    try:
+        assert load["success"], load
+        assert not load["changes_made"]["frequency_set"]
+        assert "frequency_warning" in load["changes_made"]
+        dh = load["data_handle"]
+        y = ex._data_handles[dh]["y"]
+        assert len(y) == 8
+        assert isinstance(y.index, pd.DatetimeIndex)
+        X = ex._data_handles[dh]["X"]
+        assert X is not None and len(X) == 8
+        inspected = inspect_data_tool(dh)
+        assert inspected["success"], inspected
+        assert inspected["shape"][0] == 8
+        fit = fit_tool(estimator_handle=handle, y_handle=dh)
+        assert fit["success"], fit
+    finally:
+        _pop(ex, load.get("data_handle"))
+        ex._handle_manager.release_handle(handle)