diff --git a/app/services/universe_service.py b/app/services/universe_service.py
index dc3e979..881f8e4 100644
--- a/app/services/universe_service.py
+++ b/app/services/universe_service.py
@@ -88,7 +88,12 @@ class UniverseService:
         No sector/industry (those remain NULL in registry).
         No market_cap filter (filter happens at screening time).
         """
-        _US_EXCHANGES = {"NYSE", "NASDAQ", "AMEX", "ARCA", "BATS", "NYSEArca", "OTC"}
+        # Case-insensitive match; normalize to canonical uppercase form
+        _US_EXCHANGE_MAP = {
+            "nyse": "NYSE", "nasdaq": "NASDAQ", "amex": "AMEX",
+            "nysemkt": "AMEX", "arca": "ARCA", "nysearca": "ARCA",
+            "bats": "BATS", "otc": "OTC",
+        }
         try:
             data = await self._http.fetch_json(
                 "https://www.sec.gov/files/company_tickers_exchange.json"
@@ -110,11 +115,12 @@ class UniverseService:
         for row in rows_raw:
             try:
                 ticker = str(row[ticker_idx]).upper().strip()
-                exchange = str(row[exchange_idx]).strip()
+                exchange_raw = str(row[exchange_idx]).strip()
+                canonical_exchange = _US_EXCHANGE_MAP.get(exchange_raw.lower())
                 if not ticker or len(ticker) > 10:
                     continue
-                # Keep only major US exchanges
-                if exchange not in _US_EXCHANGES:
+                # Keep only major US exchanges (case-insensitive)
+                if not canonical_exchange:
                     continue
                 # Skip preferred stocks / warrants / rights / units (contain - or end in W/R/U/Z)
                 if "-" in ticker:
@@ -125,7 +131,7 @@ class UniverseService:
                     "symbol": ticker,
                     "shortName": str(row[name_idx]) if row[name_idx] else None,
                     "cik_override": str(row[cik_idx]).zfill(10),
-                    "exchange": exchange,
+                    "exchange": canonical_exchange,
                     "sector": None,
                     "industry": None,
                     "quoteType": "EQUITY",
@@ -320,6 +326,7 @@ class UniverseService:
         total_snapshots = 0
         total_failed = 0
         total_batches = (len(ticker_list) + _PRICE_BATCH - 1) // _PRICE_BATCH
+        _CACHE_CLEAR_EVERY = 20   # clear SEC in-memory cache every N batches to prevent OOM
 
         # ---- Process in batches — each batch uses its own DB session ----
         for batch_start in range(0, len(ticker_list), _PRICE_BATCH):
@@ -425,6 +432,15 @@ class UniverseService:
                 f"Universe: batch {batch_num}/{total_batches} — "
                 f"{len(batch)} tickers, {len(batch_rows)} snapshot rows"
             )
+            # Periodically clear SEC in-memory JSON cache to prevent OOM.
+            # Disk cache is retained — subsequent lookups re-load from disk.
+            if batch_num % _CACHE_CLEAR_EVERY == 0:
+                import gc
+                self._http._json_cache.clear()
+                self._http._text_cache.clear()
+                gc.collect()
+                logger.info(f"Universe: cleared SEC JSON cache at batch {batch_num}")
+
             # Yield to event loop between batches to keep API responsive
             await asyncio.sleep(0)