Add comments and CSV export to conversation verification

src/interface/simplified_gradio_app.py

@@ -348,7 +348,8 @@ def create_simplified_interface():
 
                 with gr.Row():
                     generate_conv_verification_btn = gr.Button("🛠 Generate from current chat", variant="primary")
-                    conv_verify_download_btn = gr.DownloadButton("⬇️ Download session JSON", variant="secondary")
+                    conv_verify_download_btn = gr.DownloadButton("⬇️ Download reviewed JSON", variant="secondary")
+                    conv_verify_download_csv_btn = gr.DownloadButton("📄 Download CSV", variant="secondary")
 
                 conv_verify_status = gr.Markdown(value="", visible=True)
                 conv_verify_exchange = gr.HTML(value="", label="Current Exchange")
@@ -359,6 +360,16 @@ def create_simplified_interface():
                     conv_prev_btn = gr.Button("⬅️ Previous")
                     conv_next_btn = gr.Button("Next ➡️")
 
+                # Shown only when marking Incorrect
+                with gr.Row(visible=False) as conv_incorrect_comment_row:
+                    conv_incorrect_comment = gr.Textbox(
+                        label="Comment (why incorrect / what to fix)",
+                        placeholder="Add a short note for this exchange...",
+                        lines=3,
+                        scale=4,
+                    )
+                    conv_save_comment_btn = gr.Button("💾 Save comment", variant="secondary", scale=1)
+
                 with gr.Row():
                     with gr.Column(scale=1):
                         conv_position = gr.Markdown(value="")
@@ -2066,6 +2077,77 @@ To revert, use "Reset to Default" button.
             stats = f"<div><strong>Reviewed:</strong> {checked}/{len(records)}</div>"
             return html, pos, stats
 
+        def _export_conv_records_to_json(meta: dict, records: list):
+            """Write reviewed conversation verification results to a JSON file and return its path."""
+            import json
+            import os
+            from datetime import datetime
+
+            export_dir = os.path.join(os.getcwd(), "verification_sessions")
+            os.makedirs(export_dir, exist_ok=True)
+
+            session_id = (meta or {}).get("session_id") or "conversation_verification"
+            export_filename = f"conversation_verification_reviewed_{datetime.utcnow().strftime('%Y%m%d_%H%M%S')}_{session_id}.json"
+            export_path = os.path.join(export_dir, export_filename)
+
+            payload = {
+                **(meta or {}),
+                "verification_records": records or [],
+            }
+
+            with open(export_path, "w", encoding="utf-8") as f:
+                json.dump(payload, f, ensure_ascii=False, indent=2, default=str)
+            return export_path
+
+        def _export_conv_records_to_csv(meta: dict, records: list):
+            """Write reviewed conversation verification results to a CSV file and return its path."""
+            import csv
+            import os
+            from datetime import datetime
+
+            export_dir = os.path.join(os.getcwd(), "verification_exports")
+            os.makedirs(export_dir, exist_ok=True)
+
+            session_id = (meta or {}).get("session_id") or "conversation_verification"
+            export_filename = f"conversation_verification_reviewed_{datetime.utcnow().strftime('%Y%m%d_%H%M%S')}_{session_id}.csv"
+            export_path = os.path.join(export_dir, export_filename)
+
+            fieldnames = [
+                "session_id",
+                "patient_name",
+                "verifier_name",
+                "start_time",
+                "exchange_number",
+                "exchange_id",
+                "original_classification",
+                "original_confidence",
+                "is_correct",
+                "verifier_notes",
+                "user_message",
+                "assistant_response",
+            ]
+
+            with open(export_path, "w", encoding="utf-8", newline="") as f:
+                w = csv.DictWriter(f, fieldnames=fieldnames)
+                w.writeheader()
+                for r in records or []:
+                    row = {
+                        "session_id": (meta or {}).get("session_id"),
+                        "patient_name": (meta or {}).get("patient_name"),
+                        "verifier_name": (meta or {}).get("verifier_name"),
+                        "start_time": (meta or {}).get("start_time"),
+                        "exchange_number": r.get("exchange_number"),
+                        "exchange_id": r.get("exchange_id") or r.get("record_id"),
+                        "original_classification": r.get("original_classification"),
+                        "original_confidence": r.get("original_confidence"),
+                        "is_correct": r.get("is_correct"),
+                        "verifier_notes": r.get("verifier_notes") or "",
+                        "user_message": r.get("user_message"),
+                        "assistant_response": r.get("assistant_response"),
+                    }
+                    w.writerow(row)
+            return export_path
+
         def _generate_conv_verification(session: SimplifiedSessionData):
             if session is None or not hasattr(session.app_instance, "conversation_logger"):
                 return None, [], 0, "❌ No session/conversation found", "", ""
@@ -2075,9 +2157,13 @@ To revert, use "Reset to Default" button.
             from src.core.conversation_verification import ConversationVerificationManager
             manager = ConversationVerificationManager()
             vs = manager.create_verification_session(session.app_instance.conversation_logger, "Medical Professional")
-            # Reuse existing export behavior
-            _ = open_verification_window(session)  # will export JSON
-            export_path = _download_latest_verification_json(session)
+
+            meta = {
+                "session_id": vs.session_id,
+                "patient_name": vs.patient_name,
+                "verifier_name": vs.verifier_name,
+                "start_time": vs.start_time.isoformat() if hasattr(vs, "start_time") else None,
+            }
 
             records_as_dicts = [
                 {
@@ -2099,7 +2185,7 @@ To revert, use "Reset to Default" button.
                 for r in vs.verification_records
             ]
             html, pos, stats = _render_conv_exchange(records_as_dicts, 0)
-            return export_path, records_as_dicts, 0, f"✅ Generated session `{vs.session_id}`", html, pos, stats
+            return meta, records_as_dicts, 0, f"✅ Generated session `{vs.session_id}`", html, pos, stats
 
         def _mark_conv_correct(records: list, idx: int):
             if not records:
@@ -2119,6 +2205,29 @@ To revert, use "Reset to Default" button.
             html, pos, stats = _render_conv_exchange(records, idx)
             return records, idx, "❌ Marked incorrect", html, pos, stats
 
+        def _show_incorrect_comment_ui(records: list, idx: int):
+            """Mark incorrect and open the comment row, pre-filling any existing note."""
+            records, idx, status, html, pos, stats = _mark_conv_incorrect(records, idx)
+            note = ""
+            if records and isinstance(records[idx], dict):
+                note = records[idx].get("verifier_notes") or ""
+            return records, idx, status, html, pos, stats, gr.update(visible=True), note
+
+        def _save_incorrect_comment(records: list, idx: int, note: str):
+            if not records:
+                return records, idx, "", "", "", gr.update(visible=False), ""
+            idx = max(0, min(idx, len(records) - 1))
+            if isinstance(records[idx], dict):
+                records[idx]["verifier_notes"] = (note or "").strip()
+            html, pos, stats = _render_conv_exchange(records, idx)
+            return records, idx, "💾 Comment saved", html, pos, stats, gr.update(visible=False)
+
+        def _download_reviewed_json(meta: dict, records: list):
+            return _export_conv_records_to_json(meta, records)
+
+        def _download_reviewed_csv(meta: dict, records: list):
+            return _export_conv_records_to_csv(meta, records)
+
         def _nav_conv(records: list, idx: int, delta: int):
             if not records:
                 return idx, "", "", ""
@@ -2133,11 +2242,17 @@ To revert, use "Reset to Default" button.
         )
 
         conv_verify_download_btn.click(
-            lambda p: p,
-            inputs=[conv_verify_state],
+            _download_reviewed_json,
+            inputs=[conv_verify_state, conv_verify_records],
             outputs=[conv_verify_download_btn]
         )
 
+        conv_verify_download_csv_btn.click(
+            _download_reviewed_csv,
+            inputs=[conv_verify_state, conv_verify_records],
+            outputs=[conv_verify_download_csv_btn]
+        )
+
         conv_correct_btn.click(
             _mark_conv_correct,
             inputs=[conv_verify_records, conv_verify_index],
@@ -2145,9 +2260,32 @@ To revert, use "Reset to Default" button.
         )
 
         conv_incorrect_btn.click(
-            _mark_conv_incorrect,
+            _show_incorrect_comment_ui,
             inputs=[conv_verify_records, conv_verify_index],
-            outputs=[conv_verify_records, conv_verify_index, conv_verify_status, conv_verify_exchange, conv_position, conv_stats]
+            outputs=[
+                conv_verify_records,
+                conv_verify_index,
+                conv_verify_status,
+                conv_verify_exchange,
+                conv_position,
+                conv_stats,
+                conv_incorrect_comment_row,
+                conv_incorrect_comment,
+            ]
+        )
+
+        conv_save_comment_btn.click(
+            _save_incorrect_comment,
+            inputs=[conv_verify_records, conv_verify_index, conv_incorrect_comment],
+            outputs=[
+                conv_verify_records,
+                conv_verify_index,
+                conv_verify_status,
+                conv_verify_exchange,
+                conv_position,
+                conv_stats,
+                conv_incorrect_comment_row,
+            ]
         )
 
         conv_prev_btn.click(

tests/test_conversation_verification_export.py

@@ -1,5 +1,5 @@
+import csv
 import json
-import os
 from datetime import datetime
 
 
@@ -65,3 +65,65 @@ def test_conversation_verification_export_serializes_without_record_id(tmp_path,
     loaded = json.loads(out.read_text(encoding="utf-8"))
     assert loaded["verification_records"][0]["exchange_id"] == "sess_1"
     assert loaded["verification_records"][0]["record_id"] == "sess_1"
+
+
+def test_conversation_verification_csv_contains_expected_columns(tmp_path):
+    meta = {
+        "session_id": "verification_test",
+        "patient_name": "Test",
+        "verifier_name": "Verifier",
+        "start_time": "2025-12-12T00:00:00",
+    }
+    records = [
+        {
+            "exchange_id": "sess_1",
+            "exchange_number": 1,
+            "original_classification": "YELLOW",
+            "original_confidence": 0.9,
+            "is_correct": False,
+            "verifier_notes": "Needs follow-up",
+            "user_message": "hi",
+            "assistant_response": "hello",
+        }
+    ]
+
+    out = tmp_path / "export.csv"
+
+    fieldnames = [
+        "session_id",
+        "patient_name",
+        "verifier_name",
+        "start_time",
+        "exchange_number",
+        "exchange_id",
+        "original_classification",
+        "original_confidence",
+        "is_correct",
+        "verifier_notes",
+        "user_message",
+        "assistant_response",
+    ]
+
+    with out.open("w", encoding="utf-8", newline="") as f:
+        w = csv.DictWriter(f, fieldnames=fieldnames)
+        w.writeheader()
+        for r in records:
+            w.writerow(
+                {
+                    "session_id": meta["session_id"],
+                    "patient_name": meta["patient_name"],
+                    "verifier_name": meta["verifier_name"],
+                    "start_time": meta["start_time"],
+                    "exchange_number": r.get("exchange_number"),
+                    "exchange_id": r.get("exchange_id"),
+                    "original_classification": r.get("original_classification"),
+                    "original_confidence": r.get("original_confidence"),
+                    "is_correct": r.get("is_correct"),
+                    "verifier_notes": r.get("verifier_notes"),
+                    "user_message": r.get("user_message"),
+                    "assistant_response": r.get("assistant_response"),
+                }
+            )
+
+    rows = list(csv.DictReader(out.open("r", encoding="utf-8")))
+    assert rows and rows[0]["verifier_notes"] == "Needs follow-up"