Update app.py

app.py

@@ -1,7 +1,7 @@
-# ===========================
-# FINAL app.py
-# Live Logs + Tool Tracking
-# ===========================
+# ==========================================================
+# UPDATED app.py
+# Observability + Model Logs + Adaptive Recursion
+# ==========================================================
 
 import os
 import sys
@@ -18,10 +18,11 @@ from agent import (
     extract_final_answer,
     extract_tools_used,
     classify_question,
+    get_last_trace,
 )
 
 # ==========================================================
-# LIVE LOGS
+# LIVE LOGGING
 # ==========================================================
 
 sys.stdout.reconfigure(line_buffering=True)
@@ -35,13 +36,19 @@ DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
 TIMEOUTS = {
     "simple": 20,
     "medium": 35,
-    "hard": 55
+    "hard": 55,
 }
 
 PAUSE = {
     "simple": 0.5,
-    "medium": 1,
-    "hard": 1.5
+    "medium": 1.0,
+    "hard": 1.5,
+}
+
+RECURSION_LIMITS = {
+    "simple": 6,
+    "medium": 10,
+    "hard": 14,
 }
 
 # ==========================================================
@@ -52,7 +59,7 @@ def log(msg):
     print(msg, flush=True)
 
 # ==========================================================
-# TIMEOUT THREAD
+# THREAD TIMEOUT
 # ==========================================================
 
 def run_with_timeout(fn, timeout_seconds):
@@ -93,18 +100,22 @@ class BenchmarkAgent:
         log("Agent ready.")
 
     def __call__(self, question: str):
+        difficulty = classify_question(question)
+        recursion_limit = RECURSION_LIMITS[difficulty]
+
         result = self.graph.invoke(
             {"messages": [HumanMessage(content=question)]},
-            {"recursion_limit": 4}
+            {"recursion_limit": recursion_limit}
         )
 
         answer = extract_final_answer(result)
         tools = extract_tools_used(result)
+        trace = get_last_trace()
 
         if not answer:
             answer = "N/A"
 
-        return answer, tools
+        return answer, tools, trace
 
 # ==========================================================
 # HELPERS
@@ -140,7 +151,7 @@ def sort_questions(items):
     )
 
 # ==========================================================
-# MAIN RUN
+# MAIN RUNNER
 # ==========================================================
 
 def run_and_submit_all(profile: gr.OAuthProfile | None):
@@ -162,10 +173,10 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
         f"{space_id}/tree/main"
     )
 
-    log("=" * 70)
+    log("=" * 80)
     log(f"User: {username}")
-    log("Starting benchmark...")
-    log("=" * 70)
+    log("Starting benchmark run...")
+    log("=" * 80)
 
     # ------------------------------------------------------
     # INIT
@@ -177,7 +188,7 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
         return f"Agent init failed: {e}", None
 
     # ------------------------------------------------------
-    # QUESTIONS
+    # FETCH QUESTIONS
     # ------------------------------------------------------
 
     try:
@@ -206,19 +217,22 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
 
         difficulty = classify_question(question)
         timeout = TIMEOUTS[difficulty]
+        recursion_limit = RECURSION_LIMITS[difficulty]
 
-        log("\n" + "-" * 70)
-        log(f"Question {idx}/{total}")
-        log(f"Task ID    : {task_id}")
-        log(f"Difficulty : {difficulty.upper()}")
-        log(f"Timeout    : {timeout}s")
-        log(f"Q: {question[:160]}")
-        log("-" * 70)
+        log("\n" + "-" * 80)
+        log(f"Question      : {idx}/{total}")
+        log(f"Task ID       : {task_id}")
+        log(f"Difficulty    : {difficulty.upper()}")
+        log(f"Timeout       : {timeout}s")
+        log(f"Recursion     : {recursion_limit}")
+        log(f"Question Text : {question[:180]}")
+        log("-" * 80)
 
         start = time.time()
 
         submitted_answer = "N/A"
         tools_used = []
+        trace = {}
         status = "UNKNOWN"
 
         try:
@@ -236,15 +250,12 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
             )
 
             if did_timeout:
-                status = f"TIMEOUT ({elapsed}s)"
                 timeout_count += 1
+                status = f"TIMEOUT ({elapsed}s)"
                 submitted_answer = "N/A"
 
             else:
-                submitted_answer, tools_used = result
-
-                if not submitted_answer:
-                    submitted_answer = "N/A"
+                submitted_answer, tools_used, trace = result
 
                 if submitted_answer != "N/A":
                     answered += 1
@@ -261,11 +272,26 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
             status = f"ERROR ({elapsed}s)"
             submitted_answer = "N/A"
 
-            log(f"Error: {e}")
+            trace = {
+                "model": "N/A",
+                "fallback": "N/A",
+                "model_error": str(e)
+            }
 
-        log(f"Tools Used : {', '.join(tools_used) if tools_used else 'None'}")
-        log(f"Answer     : {submitted_answer}")
-        log(f"Status     : {status}")
+        # --------------------------------------------------
+        # LOG DETAILS
+        # --------------------------------------------------
+
+        log(f"Model Used    : {trace.get('model','N/A')}")
+        log(f"Fallback Used : {trace.get('fallback','N/A')}")
+        log(f"Model Error   : {trace.get('model_error','None')}")
+        log(f"Tools Used    : {', '.join(tools_used) if tools_used else 'None'}")
+        log(f"Answer        : {submitted_answer}")
+        log(f"Status        : {status}")
+
+        # --------------------------------------------------
+        # STORE
+        # --------------------------------------------------
 
         answers.append({
             "task_id": task_id,
@@ -276,10 +302,13 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
             "No": idx,
             "Task ID": task_id,
             "Difficulty": difficulty.upper(),
+            "Model Used": trace.get("model", "N/A"),
+            "Fallback": trace.get("fallback", "N/A"),
+            "Model Error": trace.get("model_error", "None"),
             "Tools Used": ", ".join(tools_used),
             "Submitted Answer": submitted_answer,
             "Status": status,
-            "Question": question[:120]
+            "Question": question[:140]
         })
 
         time.sleep(PAUSE[difficulty])
@@ -291,11 +320,8 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
     df = pd.DataFrame(logs)
 
     try:
-        df.to_csv(
-            "last_run_results.csv",
-            index=False
-        )
-        log("Saved CSV results.")
+        df.to_csv("last_run_results.csv", index=False)
+        log("Saved results CSV.")
     except Exception:
         pass
 
@@ -328,9 +354,9 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
             f"{result.get('message','')}"
         )
 
-        log("=" * 70)
+        log("=" * 80)
         log(final_status)
-        log("=" * 70)
+        log("=" * 80)
 
         return final_status, df
 
@@ -349,18 +375,19 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
 # ==========================================================
 
 with gr.Blocks() as demo:
-    gr.Markdown("# Benchmark Agent")
+    gr.Markdown("# Benchmark Agent - Observability Mode")
 
     gr.Markdown(
         """
 ### Features
 
-- Exact-match benchmark mode
-- Live progress logs
-- Tool tracking per question
-- Fast timeout strategy
+- Adaptive recursion limits
+- Logs selected model per question
+- Logs fallback model usage
+- Logs rate/model errors
+- Logs tools used
+- CSV export
 - Auto submit
-- CSV saved
         """
     )
 
@@ -378,7 +405,7 @@ with gr.Blocks() as demo:
     )
 
     table = gr.DataFrame(
-        label="Per Question Analysis",
+        label="Per Question Analytics",
         wrap=True
     )