ui: warn when a record's table has cleaning artefacts (non-numeric cells, duplicate columns)

app.py

@@ -127,16 +127,34 @@ def render_document(pre_text, table, post_text):
     )
 
 
-def render_document_markdown(doc):
+def render_document_markdown(record):
     """Markdown view for the UI. Cleaner than the XML wrapping the model sees."""
+    doc = record["doc"]
+    feats = record.get("features", {})
     table, cols = doc["table"], list(doc["table"])
     rows = list(dict.fromkeys(r for col in table.values() for r in col))
     md = ["| | " + " | ".join(cols) + " |", "|---" * (len(cols) + 1) + "|"]
     for row in rows:
         cells = [str(table[c].get(row, "")) for c in cols]
         md.append(f"| {row} | " + " | ".join(cells) + " |")
+    notes = []
+    if feats.get("has_non_numeric_values"):
+        notes.append(
+            "non-numeric cells (the upstream cleaner sometimes folds first-row "
+            "values into the column-name string)"
+        )
+    if feats.get("has_duplicate_columns"):
+        notes.append("duplicate column headers not fully disambiguated by cleaning")
+    warning = ""
+    if notes:
+        warning = (
+            "_Note: this record's `features` flags it as having "
+            + " and ".join(notes)
+            + ". The table below is exactly what the model sees, artefacts and all._\n\n"
+        )
     return (
-        "**Pre-text:**  \n" + doc["pre_text"]
+        warning
+        + "**Pre-text:**  \n" + doc["pre_text"]
         + "\n\n**Table:**  \n" + "\n".join(md)
         + "\n\n**Post-text:**  \n" + doc["post_text"]
     )
@@ -260,7 +278,7 @@ def replay_record(record_id, progress=gr.Progress()):
     if not record_id:
         return "Pick a record from the dropdown.", None, "—"
     record = DEV_RECORDS[record_id]
-    doc_md = render_document_markdown(record["doc"])
+    doc_md = render_document_markdown(record)
     sys_msg = {
         "role": "system",
         "content": SYSTEM_PROMPT_PREFIX + render_document(
@@ -294,7 +312,7 @@ def chat_load_doc(record_id):
     if not record_id:
         return "Pick a record from the dropdown.", []
     record = DEV_RECORDS[record_id]
-    return render_document_markdown(record["doc"]), []
+    return render_document_markdown(record), []
 
 
 def chat_respond(user_msg, history, record_id):