ISA #3: answer confidence/star badge (mobile-safe markdown)

app.py

@@ -59,6 +59,24 @@ except Exception:
 
 CONTEXT_TURNS = 5  # matches the Streamlit app's last-5-messages window
 
+_TOOL_LABELS = {
+    "cdms_label": "CDMS label", "cdms": "CDMS label", "pesticide_label": "CDMS label",
+    "rag": "CDMS label", "documentation": "CDMS label",
+    "weather": "weather", "soil": "soil", "agriculture_web": "agriculture web", "ag_web": "agriculture web",
+}
+
+
+def _confidence_badge(tool: str, confidence: float) -> str:
+    """Answer confidence/star badge (ISA feedback #3, ported from the Streamlit UI).
+
+    Plain-markdown so it renders cleanly on mobile — the Streamlit version's badge
+    was CSS-clipped on phones. `confidence` is the tool-router's match score (0-1).
+    """
+    stars = max(1, min(5, round((confidence or 0.0) * 5)))
+    bar = "★" * stars + "☆" * (5 - stars)
+    label = _TOOL_LABELS.get(tool, tool)
+    return f"\n\n---\n`{bar}` · **{confidence:.0%} confidence** · answered via *{label}*"
+
 
 def answer(question: str, history: list[dict]) -> str:
     """Run one question through the tool pipeline. Mirrors the Streamlit flow."""
@@ -72,9 +90,11 @@ def answer(question: str, history: list[dict]) -> str:
     except Exception:
         keywords = []
 
+    confidence = 0.3
     try:
         match = tool_matcher.match_tool(keywords, question, conversation_context=context)
         tool = match["tool_name"]
+        confidence = float(match.get("confidence", 0.3) or 0.3)
     except Exception:
         logger.exception("Tool matching failed; falling back to the label tool")
         tool = "cdms_label"
@@ -92,7 +112,12 @@ def answer(question: str, history: list[dict]) -> str:
         return result.get(
             "llm_response", "I couldn't find that in my label set."
         )
-    return result.get("llm_response", "I couldn't process that request.")
+    reply = result.get("llm_response", "I couldn't process that request.")
+    # Show a confidence/star rating with real answers. Not on an abstention — a
+    # confidence score on "I don't have that label" would be misleading.
+    if "don't have the label" not in reply:
+        reply += _confidence_badge(tool, confidence)
+    return reply
 
 
 def on_submit(question: str, history: list[dict], user: dict | None):