From 9d90d32370c33e7bcb52e54f4d960df4b4f60a93 Mon Sep 17 00:00:00 2001
From: Brad Stein <Brad.Stein@gmail.com>
Date: Tue, 3 Feb 2026 10:05:43 -0300
Subject: [PATCH] atlasbot: enforce keyword overlap for metrics

---
 atlasbot/engine/answerer.py | 34 ++++++++++++++++++++++++++++++++++
 1 file changed, 34 insertions(+)

diff --git a/atlasbot/engine/answerer.py b/atlasbot/engine/answerer.py
index a468884..2f48f2e 100644
--- a/atlasbot/engine/answerer.py
+++ b/atlasbot/engine/answerer.py
@@ -377,6 +377,10 @@ class AnswerEngine:
                             )
                         if not metric_facts and fallback_candidates:
                             metric_facts = fallback_candidates[: max(2, plan.max_subquestions)]
+                    if metric_facts and not _has_keyword_overlap(metric_facts, keyword_tokens):
+                        best_line = _best_keyword_line(summary_lines, keyword_tokens)
+                        if best_line:
+                            metric_facts = _merge_fact_lines([best_line], metric_facts)
                         if metric_facts:
                             key_facts = _merge_fact_lines(metric_facts, key_facts)
                 if self._settings.debug_pipeline:
@@ -1468,6 +1472,36 @@ def _global_facts(lines: list[str]) -> list[str]:
     return _dedupe_lines(facts, limit=6)
 
 
+def _has_keyword_overlap(lines: list[str], keywords: list[str]) -> bool:
+    if not lines or not keywords:
+        return False
+    tokens = [kw.lower() for kw in keywords if kw]
+    if not tokens:
+        return False
+    for line in lines:
+        lower = line.lower()
+        if any(tok in lower for tok in tokens):
+            return True
+    return False
+
+
+def _best_keyword_line(lines: list[str], keywords: list[str]) -> str | None:
+    if not lines or not keywords:
+        return None
+    tokens = [kw.lower() for kw in keywords if kw]
+    if not tokens:
+        return None
+    best = None
+    best_score = 0
+    for line in lines:
+        lower = line.lower()
+        score = sum(1 for tok in tokens if tok in lower)
+        if score > best_score:
+            best_score = score
+            best = line
+    return best if best_score > 0 else None
+
+
 def _lexicon_context(summary: dict[str, Any]) -> str:  # noqa: C901
     if not isinstance(summary, dict):
         return ""