diff --git a/atlasbot/engine/answerer.py b/atlasbot/engine/answerer.py
index da50678..d370983 100644
--- a/atlasbot/engine/answerer.py
+++ b/atlasbot/engine/answerer.py
@@ -161,6 +161,7 @@ class AnswerEngine:
         history_ctx = _format_history(history)
         lexicon_ctx = _lexicon_context(summary)
         key_facts: list[str] = []
+        metric_facts: list[str] = []
 
         started = time.monotonic()
         reply = ""
@@ -257,6 +258,7 @@ class AnswerEngine:
                 scored = await _score_chunks(call_llm, chunks, normalized, sub_questions, plan)
                 selected = _select_chunks(chunks, scored, plan, keyword_tokens)
                 key_facts = _key_fact_lines(summary_lines, keyword_tokens)
+                metric_facts = [line for line in key_facts if re.search(r"\\d", line)]
                 if self._settings.debug_pipeline:
                     scored_preview = sorted(
                         [{"id": c["id"], "score": scored.get(c["id"], 0.0), "summary": c["summary"]} for c in chunks],
@@ -328,10 +330,8 @@ class AnswerEngine:
                     extra_bits.append("AllowedRunbooks: " + ", ".join(runbook_paths))
                 if resolved_runbook:
                     extra_bits.append("ResolvedRunbook: " + resolved_runbook)
-                if key_facts:
-                    metric_facts = [line for line in key_facts if re.search(r"\\d", line)]
-                    if metric_facts:
-                        extra_bits.append("MustUseFacts: " + "; ".join(metric_facts[:4]))
+                if metric_facts:
+                    extra_bits.append("MustUseFacts: " + "; ".join(metric_facts[:4]))
                 if allowed_nodes:
                     extra_bits.append("AllowedNodes: " + ", ".join(allowed_nodes))
                 if allowed_namespaces:
@@ -403,6 +403,15 @@ class AnswerEngine:
                     model=plan.model,
                     tag="focus_fix",
                 )
+            if classify.get("question_type") in {"metric", "diagnostic"} and metric_facts and not re.search(r"\\d", reply):
+                best_line = None
+                lowered = normalized.lower()
+                for line in metric_facts:
+                    if any(token in line.lower() for token in lowered.split()):
+                        best_line = line
+                        break
+                best_line = best_line or metric_facts[0]
+                reply = f\"From the latest snapshot: {best_line}.\"
 
             if plan.use_critic:
                 if observer: