Fixed the bad classification issue.

src/fgai/profile_suggestions.py

@@ -164,6 +164,7 @@ IGNORED_DISCOVERY_FIELDS = {
     "fgai_stream_id",
     "fgai_stream_name",
 }
+EMPTY_DISCOVERY_VALUES = {"", "-", "--", "unknown", "n/a", "none", "null", "nil", "undefined", "[]", "{}"}
 
 
 def _stream_id(event: LogEvent) -> str:
@@ -182,6 +183,11 @@ def _is_number(value: str) -> bool:
         return False
 
 
+def _has_discovery_value(value: object) -> bool:
+    text = str(value).strip()
+    return text.lower() not in EMPTY_DISCOVERY_VALUES
+
+
 def _pick_present(priority: tuple[str, ...], coverage: Counter[str], total: int, *, min_ratio: float = 0.05, limit: int = 6) -> list[str]:
     selected = []
     for field in priority:
@@ -436,7 +442,7 @@ def suggest_stream_profiles(events: list[LogEvent], *, existing_profiles: dict[s
         detector_counts: Counter[str] = Counter()
         for event in stream_events:
             for field, value in event.fields.items():
-                if not value:
+                if not _has_discovery_value(value):
                     continue
                 coverage[field] += 1
                 if len(unique_values[field]) < 200: