Audit execution and add frozen research phases

data/historical.py

@@ -8,6 +8,7 @@ from utils.helpers import feature_warmup_days
 from data.validation import DataFetchError, validate_ohlcv, positive_integer
 
 _YF_TIMEOUT_SECONDS = 30
+_RETENTION_GUARD_SECONDS = 300
 OHLCV_COLUMNS = ['Open', 'High', 'Low', 'Close', 'Volume']
 
 def _fetch_yf(symbol: str, interval: str, start: datetime, end: datetime) -> pd.DataFrame:
@@ -65,6 +66,16 @@ def _native_fetch(symbol: str, interval: str, days: int) -> pd.DataFrame:
     while remaining > 0:
         window_days = min(request_days, remaining)
         window_start = window_end - timedelta(days=window_days)
+        retention_guard = (
+            _RETENTION_GUARD_SECONDS
+            if max_days is not None and days == max_days and remaining == window_days
+            else 0
+        )
+        # A start at exactly now-minus-retention can age outside Yahoo's
+        # inclusive rolling bound while its request is in flight. Keep a
+        # small fixed provider guard, not a result-dependent shorter sample.
+        if retention_guard:
+            window_start += timedelta(seconds=retention_guard)
         frame = _download_window(symbol, interval, window_start, window_end)
         if not frame.empty:
             frames.append(frame)
@@ -81,6 +92,8 @@ def _native_fetch(symbol: str, interval: str, days: int) -> pd.DataFrame:
     df.attrs['effective_lookback_days'] = days
     df.attrs['source_interval'] = interval
     df.attrs['empty_request_windows'] = empty_windows
+    df.attrs['retention_guard_seconds'] = _RETENTION_GUARD_SECONDS if max_days is not None and days == max_days else 0
+    df.attrs['requested_earliest_start_utc'] = window_start.isoformat()
     return df
 
 def resample_ohlc(df: pd.DataFrame, rule: str, source_step=None) -> pd.DataFrame: