fix(dataflows): quote the prices the vendor reported in the verification snapshot

- gap filling keeps indicators on a continuous series, but put the previous session's open, high and low under an unsettled bar's date
- load_ohlcv takes fill_gaps, and the snapshot reads the frame as reported

tradingagents/dataflows/market_data_validator.py

@@ -32,7 +32,9 @@ def _verified_rows(symbol: str, curr_date: str) -> pd.DataFrame:
     look-ahead rows, but we re-apply the cutoff defensively — this is a
     verification path, so it must not trust its input to be pre-filtered.
     """
-    data = load_ohlcv(symbol, curr_date)
+    # As reported: this snapshot is quoted by the agents as exact prices, so a
+    # gap-filled cell would put the previous session's number under this date.
+    data = load_ohlcv(symbol, curr_date, fill_gaps=False)
     if data is None or data.empty:
         raise ValueError(f"No OHLCV data available for {symbol}.")
 

tradingagents/dataflows/stockstats_utils.py

@@ -179,12 +179,16 @@ def _cache_is_fresh(data_file, curr_date_dt, now) -> bool:
     return curr_date_dt.date() < now.date() or (now - written).total_seconds() <= OHLCV_CACHE_TTL_SECONDS
 
 
-def load_ohlcv(symbol: str, curr_date: str) -> pd.DataFrame:
+def load_ohlcv(symbol: str, curr_date: str, fill_gaps: bool = True) -> pd.DataFrame:
     """Fetch OHLCV data with caching, filtered to prevent look-ahead bias.
 
     Downloads 5 years of data up to today and caches per symbol. On
     subsequent calls the cache is reused. Rows after curr_date are
     filtered out so backtests never see future prices.
+
+    ``fill_gaps`` carries prices forward over gaps so indicators compute on a
+    continuous series. Pass ``False`` to read the values as the vendor reported
+    them, leaving a cell that was never reported empty.
     """
     # Resolve broker/forex symbols (XAUUSD+ -> GC=F) to Yahoo's convention,
     # then reject values that would escape the cache directory when
@@ -262,7 +266,10 @@ def load_ohlcv(symbol: str, curr_date: str) -> pd.DataFrame:
             data["Date"].iloc[-1].date(), data["Date"].iloc[settled[-1]].date(),
         )
 
-    data = _fill_price_gaps(data)
+    # Indicators need a continuous series, so gaps are carried forward. A caller
+    # that reports the numbers themselves asks for the frame as it was reported:
+    # a filled cell is the previous session's price under this session's date.
+    data = _fill_price_gaps(data) if fill_gaps else data.dropna(subset=["Close"]).copy()
 
     # Reject a stale frame (latest row far older than curr_date) rather than
     # feeding year-old prices into indicators (#1021).