diff --git a/atlasbot/engine/answerer.py b/atlasbot/engine/answerer.py
index 1568416..0cef0b9 100644
--- a/atlasbot/engine/answerer.py
+++ b/atlasbot/engine/answerer.py
@@ -166,6 +166,7 @@ class AnswerEngine:
         allowed_nodes = _allowed_nodes(summary)
         allowed_namespaces = _allowed_namespaces(summary)
         summary_lines = _summary_lines(snapshot_used)
+        metric_tokens = _metric_key_tokens(summary_lines)
         global_facts = _global_facts(summary_lines)
         kb_summary = self._kb.summary()
         runbooks = self._kb.runbook_titles(limit=6)
@@ -216,6 +217,8 @@ class AnswerEngine:
             classify.setdefault("follow_up", False)
             classify.setdefault("focus_entity", "unknown")
             classify.setdefault("focus_metric", "unknown")
+            if metric_tokens and keyword_tokens and any(token in metric_tokens for token in keyword_tokens):
+                classify["needs_snapshot"] = True
             _debug_log("route_parsed", {"classify": classify, "normalized": normalized})
             lowered_question = f"{question} {normalized}".lower()
             force_metric = bool(re.search(r"\bhow many\b|\bcount\b|\btotal\b", lowered_question))
@@ -1296,6 +1299,21 @@ def _hotspot_evidence(summary: dict[str, Any]) -> list[str]:
     return lines
 
 
+def _metric_key_tokens(summary_lines: list[str]) -> set[str]:
+    tokens: set[str] = set()
+    for line in summary_lines:
+        if not isinstance(line, str) or ":" not in line:
+            continue
+        key = line.split(":", 1)[0].strip().lower()
+        if not key:
+            continue
+        tokens.add(key)
+        for part in re.split(r"[_\\s]+", key):
+            if part:
+                tokens.add(part)
+    return tokens
+
+
 async def _select_best_candidate(
     call_llm: Callable[..., Any],
     question: str,