Improve conversation verification UX

src/interface/simplified_gradio_app.py

@@ -2071,12 +2071,56 @@ To revert, use "Reset to Default" button.
             else:
                 rec = r
             html = vi._render_exchange_review(rec)
-            pos = f"### Exchange {index + 1} of {len(records)}"
-            # very small stats
-            checked = sum(1 for x in records if (x.get("is_correct") if isinstance(x, dict) else getattr(x, "is_correct", None)) is not None)
-            stats = f"<div><strong>Reviewed:</strong> {checked}/{len(records)}</div>"
+            # status badge
+            cur_is_correct = (r.get("is_correct") if isinstance(r, dict) else getattr(r, "is_correct", None))
+            if cur_is_correct is True:
+                badge = "✅"
+            elif cur_is_correct is False:
+                badge = "❌"
+            else:
+                badge = "⏳"
+            pos = f"### {badge} Exchange {index + 1} of {len(records)}"
+
+            # richer stats
+            reviewed = 0
+            correct = 0
+            incorrect = 0
+            incorrect_with_comment = 0
+            for x in records:
+                v = (x.get("is_correct") if isinstance(x, dict) else getattr(x, "is_correct", None))
+                if v is None:
+                    continue
+                reviewed += 1
+                if v is True:
+                    correct += 1
+                else:
+                    incorrect += 1
+                    note = (x.get("verifier_notes") if isinstance(x, dict) else getattr(x, "verifier_notes", None))
+                    if note and str(note).strip():
+                        incorrect_with_comment += 1
+
+            stats = (
+                "<div style='display:flex; gap:12px; flex-wrap:wrap;'>"
+                f"<div><strong>Reviewed:</strong> {reviewed}/{len(records)}</div>"
+                f"<div><strong>✅ Correct:</strong> {correct}</div>"
+                f"<div><strong>❌ Incorrect:</strong> {incorrect}</div>"
+                f"<div><strong>📝 Incorrect w/ comment:</strong> {incorrect_with_comment}</div>"
+                "</div>"
+            )
             return html, pos, stats
 
+        def _comment_ui_state(records: list, idx: int):
+            """Return (row_update, note_value) based on current record state."""
+            if not records:
+                return gr.update(visible=False), ""
+            idx = max(0, min(idx, len(records) - 1))
+            r = records[idx]
+            is_incorrect = (r.get("is_correct") is False) if isinstance(r, dict) else (getattr(r, "is_correct", None) is False)
+            if not is_incorrect:
+                return gr.update(visible=False), ""
+            note = (r.get("verifier_notes") or "") if isinstance(r, dict) else (getattr(r, "verifier_notes", "") or "")
+            return gr.update(visible=True), str(note)
+
         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
@@ -2189,38 +2233,41 @@ To revert, use "Reset to Default" button.
 
         def _mark_conv_correct(records: list, idx: int):
             if not records:
-                return records, idx, "", "", ""
+                return records, idx, "", "", "", gr.update(visible=False), ""
             idx = max(0, min(idx, len(records) - 1))
             if isinstance(records[idx], dict):
                 records[idx]["is_correct"] = True
+                # clear comment when marked correct (avoid stale notes)
+                records[idx]["verifier_notes"] = ""
             html, pos, stats = _render_conv_exchange(records, idx)
-            return records, idx, "✅ Marked correct", html, pos, stats
+            row_upd, note_val = _comment_ui_state(records, idx)
+            return records, idx, "✅ Marked correct", html, pos, stats, row_upd, note_val
 
         def _mark_conv_incorrect(records: list, idx: int):
             if not records:
-                return records, idx, "", "", ""
+                return records, idx, "", "", "", gr.update(visible=False), ""
             idx = max(0, min(idx, len(records) - 1))
             if isinstance(records[idx], dict):
                 records[idx]["is_correct"] = False
             html, pos, stats = _render_conv_exchange(records, idx)
-            return records, idx, "❌ Marked incorrect", html, pos, stats
+            row_upd, note_val = _comment_ui_state(records, idx)
+            return records, idx, "❌ Marked incorrect", html, pos, stats, row_upd, note_val
 
         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 ""
+            records, idx, status, html, pos, stats, _row, note = _mark_conv_incorrect(records, idx)
             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), ""
+                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)
+            row_upd, note_val = _comment_ui_state(records, idx)
+            # keep row visible after save (since still incorrect)
+            return records, idx, "💾 Comment saved", html, pos, stats, row_upd, note_val
 
         def _download_reviewed_json(meta: dict, records: list):
             return _export_conv_records_to_json(meta, records)
@@ -2230,10 +2277,11 @@ To revert, use "Reset to Default" button.
 
         def _nav_conv(records: list, idx: int, delta: int):
             if not records:
-                return idx, "", "", ""
+                return idx, "", "", "", gr.update(visible=False), ""
             idx = max(0, min(idx + delta, len(records) - 1))
             html, pos, stats = _render_conv_exchange(records, idx)
-            return idx, html, pos, stats
+            row_upd, note_val = _comment_ui_state(records, idx)
+            return idx, html, pos, stats, row_upd, note_val
 
         generate_conv_verification_btn.click(
             _generate_conv_verification,
@@ -2256,7 +2304,16 @@ To revert, use "Reset to Default" button.
         conv_correct_btn.click(
             _mark_conv_correct,
             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_incorrect_btn.click(
@@ -2285,19 +2342,20 @@ To revert, use "Reset to Default" button.
                 conv_position,
                 conv_stats,
                 conv_incorrect_comment_row,
+                conv_incorrect_comment,
             ]
         )
 
         conv_prev_btn.click(
             lambda records, idx: _nav_conv(records, idx, -1),
             inputs=[conv_verify_records, conv_verify_index],
-            outputs=[conv_verify_index, conv_verify_exchange, conv_position, conv_stats]
+            outputs=[conv_verify_index, conv_verify_exchange, conv_position, conv_stats, conv_incorrect_comment_row, conv_incorrect_comment]
         )
 
         conv_next_btn.click(
             lambda records, idx: _nav_conv(records, idx, 1),
             inputs=[conv_verify_records, conv_verify_index],
-            outputs=[conv_verify_index, conv_verify_exchange, conv_position, conv_stats]
+            outputs=[conv_verify_index, conv_verify_exchange, conv_position, conv_stats, conv_incorrect_comment_row, conv_incorrect_comment]
         )
         
         # Refresh conversation stats