Address PR feedback: revert unrelated changes in dashboard/app.py and scripts/quality_report.py

Author: gigistark-google

dashboard/app.py

@@ -1,9 +1,8 @@
-import re
-from google.cloud import bigquery
+import streamlit as st
 import pandas as pd
+from google.cloud import bigquery
 import plotly.express as px
-import streamlit as st
-
+import re
 
 # --- 1. Page Configuration ---
 st.set_page_config(page_title="Agent Analytics V2", layout="wide")
@@ -35,224 +34,172 @@
 
 # --- 3. Sidebar Configuration ---
 with st.sidebar:
-  st.header("Data Source")
-  project_id = st.text_input("GCP Project ID", "REPLACE WITH YOUR PROJECT ID")
-  dataset_id = st.text_input("Dataset ID", "REPLACE WITH YOUR DATASET ID")
-  table_id = st.text_input("Table ID", "REPLACE WITH YOUR TABLE ID")
-
-  # Identifier Validation
-  for val in [project_id, dataset_id, table_id]:
-    if not re.match(r"^[\w.-]+$", val):
-      st.error(f"Invalid identifier: {val}")
-      st.stop()
+    st.header("Data Source")
+    project_id = st.text_input("GCP Project ID", "REPLACE WITH YOUR PROJECT ID")
+    dataset_id = st.text_input("Dataset ID", "REPLACE WITH YOUR DATASET ID")
+    table_id = st.text_input("Table ID", "REPLACE WITH YOUR TABLE ID")
 
-  table_ref = f"`{project_id}.{dataset_id}.{table_id}`"
+    # Identifier Validation
+    for val in [project_id, dataset_id, table_id]:
+        if not re.match(r'^[\w.-]+$', val):
+            st.error(f"Invalid identifier: {val}")
+            st.stop()
 
-  st.divider()
-  st.header("Global Filters")
-  time_hours = st.slider("Time Range (Hours)", 1, 720, 72)
-  agent_name = st.text_input("Filter by Agent Name", "")
+    table_ref = f"`{project_id}.{dataset_id}.{table_id}`"
 
+    st.divider()
+    st.header("Global Filters")
+    time_hours = st.slider("Time Range (Hours)", 1, 720, 72)
+    agent_name = st.text_input("Filter by Agent Name", "")
 
 # --- 4. Data Connector ---
 @st.cache_resource
 def get_bq_client(p_id):
-  return bigquery.Client(project=p_id)
-
+    return bigquery.Client(project=p_id)
 
 client = get_bq_client(project_id)
 
-
 def safe_query(sql, params=None):
-  try:
-    job_config = (
-        bigquery.QueryJobConfig(query_parameters=params) if params else None
-    )
-    formatted_sql = sql.replace("{table_ref}", table_ref)
-    return client.query(formatted_sql, job_config=job_config).to_dataframe()
-  except Exception as e:
-    st.error(f"Query Failed: {e}")
-    return pd.DataFrame()
-
+    try:
+        job_config = bigquery.QueryJobConfig(query_parameters=params) if params else None
+        formatted_sql = sql.replace("{table_ref}", table_ref)
+        return client.query(formatted_sql, job_config=job_config).to_dataframe()
+    except Exception as e:
+        st.error(f"Query Failed: {e}")
+        return pd.DataFrame()
 
 # --- 5. Global Params (Resolved 400 Parameter Error) ---
 
 global_params = [
     bigquery.ScalarQueryParameter("time_hours", "INT64", time_hours),
-    bigquery.ScalarQueryParameter(
-        "agent_name", "STRING", agent_name if agent_name else None
-    ),
+    bigquery.ScalarQueryParameter("agent_name", "STRING", agent_name if agent_name else None)
 ]
 
 # WHERE clause for non-CTE queries (Executive Summary & Multimodal)
-where_clause = (
-    "timestamp >= TIMESTAMP_SUB(CURRENT_TIMESTAMP(), INTERVAL @time_hours HOUR)"
-)
+where_clause = "timestamp >= TIMESTAMP_SUB(CURRENT_TIMESTAMP(), INTERVAL @time_hours HOUR)"
 if agent_name:
-  where_clause += " AND agent = @agent_name"
+    where_clause += " AND agent = @agent_name"
 
 # --- 6. Executive Summary (Resolved PR Consistency Issue) ---
 st.header("📊 Executive Summary")
 
-kpi_data = safe_query(
-    f"""
+kpi_data = safe_query(f"""
     WITH enriched AS ({ENRICHED_EVENTS_CTE})
     SELECT
         COUNT(DISTINCT session_id) as sessions,
         COUNT(DISTINCT trace_id) as traces,
         COUNTIF(event_type = 'LLM_REQUEST') as llm_calls,
         SAFE_DIVIDE(COUNTIF(is_canonical_error = TRUE), COUNT(*)) as error_rate
     FROM enriched
-""",
-    params=global_params,
-)
+""", params=global_params)
 
 if not kpi_data.empty:
-  c1, c2, c3, c4 = st.columns(4)
-  c1.metric("Sessions", kpi_data["sessions"][0])
-  c2.metric("Traces", kpi_data["traces"][0])
-  c3.metric("LLM Calls", kpi_data["llm_calls"][0])
-  # The rate now includes TOOL_ERROR and LLM_ERROR automatically
-  c4.metric("Error Rate", f"{kpi_data['error_rate'][0]*100:.2f}%")
-
+    c1, c2, c3, c4 = st.columns(4)
+    c1.metric("Sessions", kpi_data['sessions'][0])
+    c2.metric("Traces", kpi_data['traces'][0])
+    c3.metric("LLM Calls", kpi_data['llm_calls'][0])
+    # The rate now includes TOOL_ERROR and LLM_ERROR automatically
+    c4.metric("Error Rate", f"{kpi_data['error_rate'][0]*100:.2f}%")
+    
 
 # --- 7. Detailed Tabs ---
-tabs = st.tabs([
-    "📈 Usage",
-    "⚡ Performance",
-    "🛡️ Reliability",
-    "🖼️ Multimodal",
-    "🤝 HITL",
-    "🔍 Trace Explorer",
-])
+tabs = st.tabs(["📈 Usage", "⚡ Performance", "🛡️ Reliability", "🖼️ Multimodal", "🤝 HITL", "🔍 Trace Explorer"])
 
 # TAB 0: USAGE & VOLUME
 with tabs[0]:
-  st.subheader("Token Consumption")
-  token_trend = safe_query(
-      f"""
+    st.subheader("Token Consumption")
+    token_trend = safe_query(f"""
         SELECT TIMESTAMP_TRUNC(timestamp, HOUR) as hour,
                SUM(CAST(JSON_VALUE(content, '$.usage.prompt') AS INT64)) as prompt,
                SUM(CAST(JSON_VALUE(content, '$.usage.completion') AS INT64)) as completion
         FROM {{table_ref}} WHERE event_type = 'LLM_RESPONSE' AND {where_clause}
         GROUP BY 1 ORDER BY 1
-    """,
-      params=global_params,
-  )
-  if not token_trend.empty:
-    fig = px.area(
-        token_trend,
-        x="hour",
-        y=["prompt", "completion"],
-        title="Tokens over Time",
-    )
-    st.plotly_chart(fig, use_container_width=True)
-  else:
-    st.info("No token usage data found.")
+    """, params=global_params)
+    if not token_trend.empty:
+        fig = px.area(token_trend, x="hour", y=["prompt", "completion"], title="Tokens over Time")
+        st.plotly_chart(fig, use_container_width=True)
+    else:
+        st.info("No token usage data found.")
 
 # TAB 1: PERFORMANCE
 with tabs[1]:
-  st.subheader("Latency Metrics")
-  lat_data = safe_query(
-      f"""
+    st.subheader("Latency Metrics")
+    lat_data = safe_query(f"""
         WITH enriched AS ({ENRICHED_EVENTS_CTE})
         SELECT
             APPROX_QUANTILES(CAST(JSON_VALUE(latency_ms, '$.total_ms') AS FLOAT64), 100)[OFFSET(50)] as p50,
             APPROX_QUANTILES(CAST(JSON_VALUE(latency_ms, '$.total_ms') AS FLOAT64), 100)[OFFSET(90)] as p90,
             AVG(CAST(JSON_VALUE(latency_ms, '$.time_to_first_token_ms') AS FLOAT64)) as avg_ttft
         FROM enriched WHERE event_family = 'llm'
-    """,
-      params=global_params,
-  )
-  if not lat_data.empty and pd.notnull(lat_data["p50"][0]):
-    p1, p2, p3 = st.columns(3)
-    p1.metric("p50 Latency", f"{lat_data['p50'][0]:.0f}ms")
-    p2.metric("p90 Latency", f"{lat_data['p90'][0]:.0f}ms")
-    p3.metric("Avg TTFT", f"{lat_data['avg_ttft'][0]:.0f}ms")
-  else:
-    st.warning("No performance data found.")
+    """, params=global_params)
+    if not lat_data.empty and pd.notnull(lat_data['p50'][0]):
+        p1, p2, p3 = st.columns(3)
+        p1.metric("p50 Latency", f"{lat_data['p50'][0]:.0f}ms")
+        p2.metric("p90 Latency", f"{lat_data['p90'][0]:.0f}ms")
+        p3.metric("Avg TTFT", f"{lat_data['avg_ttft'][0]:.0f}ms")
+    else:
+        st.warning("No performance data found.")
 
 # TAB 2: RELIABILITY
 with tabs[2]:
-  st.subheader("Error Distribution")
-  err_data = safe_query(
-      f"""
+    st.subheader("Error Distribution")
+    err_data = safe_query(f"""
         WITH enriched AS ({ENRICHED_EVENTS_CTE})
         SELECT event_family, event_type, COUNT(*) as errors
         FROM enriched WHERE is_canonical_error = TRUE
         GROUP BY 1, 2 ORDER BY errors DESC
-    """,
-      params=global_params,
-  )
-  if not err_data.empty:
-    fig = px.bar(err_data, x="event_type", y="errors", color="event_family")
-    st.plotly_chart(fig, use_container_width=True)
-  else:
-    st.success("No errors detected.")
+    """, params=global_params)
+    if not err_data.empty:
+        fig = px.bar(err_data, x="event_type", y="errors", color="event_family")
+        st.plotly_chart(fig, use_container_width=True)
+    else:
+        st.success("No errors detected.")
 
 # TAB 3: MULTIMODAL
 with tabs[3]:
-  st.subheader("Media Parts")
-  multi_data = safe_query(
-      f"""
+    st.subheader("Media Parts")
+    multi_data = safe_query(f"""
         SELECT cp.mime_type, COUNT(*) as count
         FROM {{table_ref}}, UNNEST(content_parts) as cp
         WHERE {where_clause} GROUP BY 1
-    """,
-      params=global_params,
-  )
-  if not multi_data.empty:
-    fig = px.pie(multi_data, values="count", names="mime_type")
-    st.plotly_chart(fig, use_container_width=True)
-  else:
-    st.info("No multimodal content found.")
+    """, params=global_params)
+    if not multi_data.empty:
+        fig = px.pie(multi_data, values='count', names='mime_type')
+        st.plotly_chart(fig, use_container_width=True)
+    else:
+        st.info("No multimodal content found.")
 
 # TAB 4: HITL
 with tabs[4]:
-  st.subheader("Human-in-the-Loop Activity")
-  hitl_data = safe_query(
-      f"""
+    st.subheader("Human-in-the-Loop Activity")
+    hitl_data = safe_query(f"""
         WITH enriched AS ({ENRICHED_EVENTS_CTE})
         SELECT event_type, COUNT(*) as count
         FROM enriched WHERE event_family = 'hitl'
         GROUP BY 1
-    """,
-      params=global_params,
-  )
-  if not hitl_data.empty:
-    st.table(hitl_data)
-  else:
-    st.info("No HITL events found.")
+    """, params=global_params)
+    if not hitl_data.empty:
+        st.table(hitl_data)
+    else:
+        st.info("No HITL events found.")
 
 # TAB 5: TRACE EXPLORER
 with tabs[5]:
-  st.subheader("Session Explorer")
-  sessions = safe_query(
-      f"""
+    st.subheader("Session Explorer")
+    sessions = safe_query(f"""
         SELECT session_id, agent, MIN(timestamp) as ts,
         FORMAT_TIMESTAMP('%m-%d %H:%M', MIN(timestamp)) as f_ts
         FROM {{table_ref}} WHERE {where_clause} GROUP BY 1, 2 ORDER BY ts DESC LIMIT 50
-    """,
-      params=global_params,
-  )
+    """, params=global_params)
 
-  if not sessions.empty:
-    sessions["label"] = (
-        sessions["f_ts"]
-        + " | "
-        + sessions["agent"].fillna("?")
-        + " | "
-        + sessions["session_id"].str[:8]
-    )
-    label_map = dict(zip(sessions["label"], sessions["session_id"]))
-    sel_label = st.selectbox("Select Session", sessions["label"])
-    sel_id = label_map[sel_label]
+    if not sessions.empty:
+        sessions['label'] = sessions['f_ts'] + " | " + sessions['agent'].fillna("?") + " | " + sessions['session_id'].str[:8]
+        label_map = dict(zip(sessions['label'], sessions['session_id']))
+        sel_label = st.selectbox("Select Session", sessions['label'])
+        sel_id = label_map[sel_label]
 
-    trace_detail = safe_query(
-        """
+        trace_detail = safe_query("""
             SELECT timestamp, event_type, agent, status, error_message
             FROM {table_ref} WHERE session_id = @sid ORDER BY timestamp
-        """,
-        params=[bigquery.ScalarQueryParameter("sid", "STRING", sel_id)],
-    )
-    st.dataframe(trace_detail, use_container_width=True)
+        """, params=[bigquery.ScalarQueryParameter("sid", "STRING", sel_id)])
+        st.dataframe(trace_detail, use_container_width=True)