feat: verification panel always-visible flow + docs for v2.3.3

- Conversation Verification: show the "Select Correct Classification" +
comment panel by default for every exchange (hidden only once marked
Correct), so reviewers can give feedback without first clicking Incorrect.
Generate now also primes the panel for the first exchange. Fixes latent
output-arity mismatches in the generate functions' early-return paths.
- Docs: README -> v2.3.3, sdk_version 6.1.0, new "Latest Updates (v2.3.3)"
section; CHANGELOG [2.3.3]; in-interface Help updated (verification flow,
comment-only Provider Summary step, Clear Chat resets verification).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

CHANGELOG.md

@@ -2,6 +2,35 @@
 
 All notable changes to the Medical Assistant with Spiritual Support project.
 
+## [2.3.3] - 2026-05-27
+
+Conversation Verification usability fixes from Or's follow-up feedback.
+
+### Fixed
+- **Provider Summary save error**: Saving feedback on the final Provider Summary
+  step without selecting a flag raised an error and showed "Error" badges across
+  the whole tab. The classification Radio received an empty string, which Gradio 6
+  rejects (value not in choices). Now the flag is hidden on the Provider Summary
+  step and the Radio value is normalized to `None` across all verification handlers.
+- **Stale verification on Clear Chat**: "Clear Chat" cleared the chat but left the
+  Conversation Verification tab showing the previous scenario. It now also resets
+  the verification state and displayed components.
+- **Latent output-arity mismatches** in the verification handlers' empty-session and
+  early-return paths.
+
+### Changed
+- **Verification feedback panel**: The "Select Correct Classification" + comment
+  panel is now visible by default for every exchange (hidden only once marked
+  ✅ Correct), so reviewers can give feedback without first clicking "Incorrect".
+- **Help content**: Updated the Conversation Verification section to describe the
+  new flow, the comment-only Provider Summary step, and Clear Chat behavior.
+- **Gradio**: Pinned to 6.1.0 (README `sdk_version` aligned).
+
+### Added
+- **Latency diagnostics**: Lightweight per-stage `[PERF]` timing in the message
+  path (env `PERF_TIMING`, on by default) to diagnose slow (>10s) responses. No
+  model changes.
+
 ## [2.1.0] - 2025-12-18
 
 ### Added
@@ -47,6 +76,7 @@ All notable changes to the Medical Assistant with Spiritual Support project.
 
 ## Version History
 
+- **v2.3.3** (2026-05-27): Conversation Verification usability fixes (Or follow-up)
 - **v2.1.0** (2025-12-18): Spiritual Care Message customization features
 - **v2.0.0** (2024-12-18): Prompt optimization system and comprehensive testing
 - **v1.0.0** (2024-09-01): Initial release with core functionality

README.md

@@ -1,10 +1,10 @@
 ---
-title: Medical Assistant with Spiritual Support v.2.3.2
+title: Medical Assistant with Spiritual Support v.2.3.3
 emoji: 🩺
 colorFrom: blue
 colorTo: green
 sdk: gradio
-sdk_version: 6.0.2
+sdk_version: 6.1.0
 app_file: src/interface/simplified_gradio_app.py
 pinned: false
 ---
@@ -486,12 +486,38 @@ The Medical Assistant with Spiritual Support system is **fully functional and pr
 
 ---
 
-**Version:** 2.3  
-**Last Updated:** January 7, 2026  
+**Version:** 2.3.3  
+**Last Updated:** May 27, 2026  
 **Status:** ✅ Production Ready with Enhanced Provider Summary  
 **Test Coverage:** 65/65 tests passing
 
-### 🆕 Latest Updates (v2.3) - January 2026
+### 🆕 Latest Updates (v2.3.3) - May 2026
+
+Based on medical professional follow-up feedback from Or (Conversation Verification usability):
+
+**✅ Fixed "Error" when saving Provider Summary feedback:**
+- **Issue:** On the final Provider Summary step, saving a comment without picking a
+  green/yellow/red flag raised an error and showed "Error" badges on the whole tab.
+- **Root cause:** the classification Radio received an empty string, which Gradio 6
+  rejects (value not in choices), failing the entire event response.
+- **Solution:** the flag is now **hidden on the Provider Summary final step** (it was
+  already chosen earlier), and the Radio value is normalized to `None` across all
+  verification handlers. Text-only feedback now saves cleanly.
+
+**✅ "Clear Chat" now also clears the Conversation Verification tab:**
+- Starting a new scenario no longer leaves the previous scenario's verification
+  visible. Chat history and verification state/displays reset together.
+
+**✅ Classification panel always available while reviewing:**
+- The "Select Correct Classification" + comment panel is now shown by default for
+  every exchange (until it is marked ✅ Correct), so reviewers can give feedback
+  immediately after navigating — no need to click "Incorrect" first.
+
+**✅ Latency diagnostics:**
+- Added lightweight per-stage `[PERF]` timing in the message path (env `PERF_TIMING`,
+  on by default) to help diagnose occasional slow (>10s) responses. No model changes.
+
+### 🆕 Previous Updates (v2.3) - January 2026
 
 #### Provider Summary Improvements
 Based on medical professional feedback from `review/Or_3.txt`:

src/interface/gradio_app.py

@@ -660,7 +660,12 @@ def create_simplified_interface():
         generate_conv_verification_btn.click(
             verification_handlers._generate_conv_verification,
             inputs=[session_data],
-            outputs=[conv_verify_state, conv_verify_records, conv_verify_index, conv_verify_status, conv_verify_exchange, conv_position, conv_stats]
+            outputs=[
+                conv_verify_state, conv_verify_records, conv_verify_index,
+                conv_verify_status, conv_verify_exchange, conv_position, conv_stats,
+                # Show the classification/comment panel for the first exchange too
+                conv_incorrect_comment_row, conv_incorrect_comment, conv_correct_classification,
+            ]
         )
 
         conv_verify_download_btn.click(

src/interface/simplified_gradio_app.py

@@ -535,7 +535,12 @@ def create_simplified_interface():
         generate_conv_verification_btn.click(
             verification_handlers._generate_conv_verification_with_summary,
             inputs=[session_data],
-            outputs=[conv_verify_state, conv_verify_records, conv_verify_index, conv_verify_status, conv_verify_exchange, conv_position, conv_stats]
+            outputs=[
+                conv_verify_state, conv_verify_records, conv_verify_index,
+                conv_verify_status, conv_verify_exchange, conv_position, conv_stats,
+                # Show the classification/comment panel for the first exchange too
+                conv_incorrect_comment_row, conv_incorrect_comment, conv_correct_classification,
+            ]
         )
 
         conv_verify_download_csv_btn.click(

src/interface/simplified_help_content.py

@@ -43,18 +43,28 @@ Review and verify the accuracy of spiritual distress classifications.
 2. Review each exchange:
    - Read the user message and assistant response
    - Check the classification (GREEN/YELLOW/RED)
+   - The **"Select Correct Classification" + Comment** panel is shown by default
+     for every exchange, so you can give feedback right away — no need to click
+     "Incorrect" first. It hides only after you mark an exchange ✅ Correct.
    - Mark as ✅ Correct or ❌ Incorrect
 3. For incorrect classifications:
-   - Select the correct classification
+   - Select the correct classification (🟢 / 🟡 / 🔴)
    - Add a comment explaining why it's incorrect
    - Click "💾 Save comment"
-4. Navigate with ⬅️ Previous / Next ➡️ buttons
+4. Navigate with ⬅️ Previous / Next ➡️ buttons (the panel stays visible on the
+   next exchange unless it is already marked Correct)
 5. **Final step:** Review the Provider Summary
    - This is the message sent to the spiritual care team
-   - Verify it's complete and accurate
-   - Add feedback if needed
+   - **No flag selection here** — the classification was already decided in the
+     earlier exchanges, so this step shows a **comment box only**
+   - Verify the summary is complete and accurate; add a comment and click
+     "💾 Save comment" — text alone is enough
 6. Download results: "📄 Download CSV"
 
+> 💡 **Starting a new scenario?** Press "🗑️ Clear Chat" on the Chat tab — it now
+> also clears this Conversation Verification tab, so you never see a previous
+> scenario's exchanges by mistake.
+
 **Statistics shown:**
 - Reviewed: How many exchanges you've checked
 - ✅ Correct: Accurate classifications
@@ -166,9 +176,10 @@ The Provider Summary is the **most important** part to verify:
 - ✅ Clear, professional narrative style
 
 **If incorrect or incomplete:**
-- Mark as ❌ Incorrect
-- Add detailed comment about what's missing or wrong
+- Add a detailed comment about what's missing or wrong (no flag needed on this
+  final step — the comment box is all you need)
 - Specify what should be included
+- Click "💾 Save comment"
 
 ---
 
@@ -264,6 +275,14 @@ The CSV file includes:
 - Download CSV with your findings
 - Share with the development team
 
+**Q: Do I have to pick a flag to save feedback on the Provider Summary?**
+- No. On the final Provider Summary step there is no flag — just type your
+  comment and click "💾 Save comment". Text alone is saved.
+
+**Q: I started a new scenario but still see the old verification. Why?**
+- This is fixed: pressing "🗑️ Clear Chat" now also clears the Conversation
+  Verification tab. Clear the chat before starting a new scenario.
+
 ---
 
 ## 📞 Support

src/interface/verification_handlers.py

@@ -860,13 +860,19 @@ def _render_conv_exchange(records: list, index: int):
     return html, pos, stats
 
 def _comment_ui_state(records: list, idx: int):
-    """Return (row_update, note_value) based on current record state."""
+    """Return (row_update, note_value) based on current record state.
+
+    The classification + comment panel is shown unless the exchange is confirmed
+    Correct — i.e. it is visible for not-yet-reviewed (None) and Incorrect
+    exchanges, so a reviewer can give feedback right after navigating, without
+    first clicking "Incorrect".
+    """
     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:
+    is_correct = (r.get("is_correct") if isinstance(r, dict) else getattr(r, "is_correct", None))
+    if is_correct is True:
         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)
@@ -955,9 +961,9 @@ def _export_conv_records_to_csv(meta: dict, records: list):
 
 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", "", ""
+        return None, [], 0, "❌ No session/conversation found", "", "", "", gr.update(visible=False), "", gr.update(value=None)
     if not session.app_instance.conversation_logger.entries:
-        return None, [], 0, "⚠️ No exchanges to verify yet", "", ""
+        return None, [], 0, "⚠️ No exchanges to verify yet", "", "", "", gr.update(visible=False), "", gr.update(value=None)
 
     manager = ConversationVerificationManager()
     vs = manager.create_verification_session(session.app_instance.conversation_logger, "Medical Professional")
@@ -1003,7 +1009,12 @@ def _generate_conv_verification(session: SimplifiedSessionData):
         for r in vs.verification_records
     ]
     html, pos, stats = _render_conv_exchange(records_as_dicts, 0)
-    return meta, records_as_dicts, 0, f"✅ Generated session `{vs.session_id}`", html, pos, stats
+    row_upd, note_val = _comment_ui_state(records_as_dicts, 0)
+    radio_upd = _classification_update(records_as_dicts, 0)
+    return (
+        meta, records_as_dicts, 0, f"✅ Generated session `{vs.session_id}`",
+        html, pos, stats, row_upd, note_val, radio_upd,
+    )
 
 # Display text for each classification flag (shared by the verification handlers).
 _CLASSIFICATION_DISPLAY = {
@@ -1102,9 +1113,9 @@ def _generate_conv_verification_with_summary(session: SimplifiedSessionData):
     "Provider Summary to be the final exchange presented in that tab"
     """
     if session is None or not hasattr(session.app_instance, "conversation_logger"):
-        return None, [], 0, "❌ No session/conversation found", "", "", ""
+        return None, [], 0, "❌ No session/conversation found", "", "", "", gr.update(visible=False), "", gr.update(value=None)
     if not session.app_instance.conversation_logger.entries:
-        return None, [], 0, "⚠️ No exchanges to verify yet", "", "", ""
+        return None, [], 0, "⚠️ No exchanges to verify yet", "", "", "", gr.update(visible=False), "", gr.update(value=None)
 
     manager = ConversationVerificationManager()
     vs = manager.create_verification_session(session.app_instance.conversation_logger, "Medical Professional")
@@ -1197,9 +1208,15 @@ def _generate_conv_verification_with_summary(session: SimplifiedSessionData):
             "provider_summary_html": provider_summary_html,
         }
         records_as_dicts.append(final_exchange)
-    
+
     html, pos, stats = _render_conv_exchange(records_as_dicts, 0)
-    return meta, records_as_dicts, 0, f"✅ Generated session with {len(records_as_dicts)} exchanges (Provider Summary as final step)", html, pos, stats
+    row_upd, note_val = _comment_ui_state(records_as_dicts, 0)
+    radio_upd = _classification_update(records_as_dicts, 0)
+    return (
+        meta, records_as_dicts, 0,
+        f"✅ Generated session with {len(records_as_dicts)} exchanges (Provider Summary as final step)",
+        html, pos, stats, row_upd, note_val, radio_upd,
+    )
 
 
 def _auto_save_verification_report(meta: dict, records: list, session: SimplifiedSessionData):