Spaces:

Divyonko
/

LivePulse

Running

DivYonko

Add warning: switch to Stats tab before clicking export button

d96515a 16 days ago

104 kB

	# -- coding: utf-8 --
	import streamlit as st
	import redis
	import json
	import pandas as pd
	import plotly.graph_objects as go
	import plotly.express as px
	import time
	import re
	import sys
	import os
	import subprocess
	from datetime import datetime, timedelta
	from collections import defaultdict

	sys.path.insert(0, os.path.join(os.path.dirname(__file__), '..', 'backend'))
	from config import REDIS_HOST, REDIS_PORT, REDIS_DB

	st.set_page_config(
	page_title="LivePulse",
	layout="wide",
	page_icon="📡",
	initial_sidebar_state="expanded"
	)

	r = redis.Redis(host=REDIS_HOST, port=REDIS_PORT, db=REDIS_DB, decode_responses=True)

	TOPIC_LABELS = ["Appreciation", "Question", "Request/Feedback", "Promo", "Spam", "General", "MCQ Answer"]
	TOPIC_COLOR = {
	"Appreciation": "#f59e0b", "Question": "#3b82f6",
	"Request/Feedback": "#8b5cf6",
	"Promo": "#ec4899", "Spam": "#ef4444", "General": "#6b7280",
	"MCQ Answer": "#10b981"
	}
	SENT_COLORS = {"Positive": "#22c55e", "Neutral": "#eab308", "Negative": "#ef4444"}

	# ── JS: detect Streamlit's live theme and set data-livepulse attribute ──
	THEME_JS = """<script>
	(function() {
	function applyTheme() {
	const html = window.parent.document.documentElement;
	const style = window.parent.getComputedStyle(html);
	const bg = style.getPropertyValue('--background-color').trim();
	let isDark = true;
	const m = bg.match(/rgb$(\d+),\s(\d+),\s(\d+)$/);
	if (m) { isDark = (0.299m[1] + 0.587m[2] + 0.114*m[3]) < 128; }
	else {
	const bodyBg = window.parent.getComputedStyle(window.parent.document.body).backgroundColor;
	const m2 = bodyBg.match(/rgb$(\d+),\s(\d+),\s(\d+)$/);
	if (m2) { isDark = (0.299m2[1] + 0.587m2[2] + 0.114*m2[3]) < 128; }
	}
	html.setAttribute('data-livepulse', isDark ? 'dark' : 'light');
	}
	applyTheme();
	const obs = new MutationObserver(applyTheme);
	obs.observe(window.parent.document.documentElement, { attributes: true, attributeFilter: ['style','class'] });
	obs.observe(window.parent.document.body, { attributes: true, attributeFilter: ['style','class'] });
	})();
	</script>"""

	CSS = """<style>
	@import url('https://fonts.googleapis.com/css2?family=Space+Grotesk:wght@400;500;600;700;800&display=swap');

	:root, [data-livepulse="dark"] {
	--bg:#07070f; --bg-card:#0f0f1e; --border:rgba(255,255,255,0.07);
	--text-1:#f1f5f9; --text-2:#94a3b8; --text-3:#475569;
	--accent:#7c3aed; --accent2:#4f46e5; --accent-text:#a78bfa;
	--live:#22c55e; --input-bg:rgba(255,255,255,0.04); --input-border:rgba(255,255,255,0.1);
	--divider:rgba(255,255,255,0.06); --badge-bg:rgba(255,255,255,0.05);
	--shadow:0 4px 24px rgba(0,0,0,0.4); --shadow-sm:0 2px 8px rgba(0,0,0,0.3);
	--pill-bg:rgba(124,58,237,0.15); --pill-border:rgba(124,58,237,0.3); --pill-text:#a78bfa;
	--plotly-paper:rgba(0,0,0,0); --plotly-plot:rgba(255,255,255,0.015); --plotly-grid:rgba(255,255,255,0.05); --plotly-text:#94a3b8;
	--alert-bg:rgba(239,68,68,0.1); --alert-border:rgba(239,68,68,0.3);
	--pin-bg:rgba(234,179,8,0.1); --pin-border:rgba(234,179,8,0.35);
	}
	[data-livepulse="light"] {
	--bg:#f4f6ff; --bg-card:#ffffff; --border:rgba(99,102,241,0.12);
	--text-1:#0f172a; --text-2:#475569; --text-3:#94a3b8;
	--accent:#6d28d9; --accent2:#4338ca; --accent-text:#6d28d9;
	--live:#16a34a; --input-bg:#ffffff; --input-border:rgba(99,102,241,0.2);
	--divider:rgba(99,102,241,0.1); --badge-bg:rgba(99,102,241,0.06);
	--shadow:0 4px 24px rgba(99,102,241,0.12); --shadow-sm:0 2px 8px rgba(99,102,241,0.08);
	--pill-bg:rgba(109,40,217,0.08); --pill-border:rgba(109,40,217,0.2); --pill-text:#6d28d9;
	--plotly-paper:rgba(0,0,0,0); --plotly-plot:rgba(255,255,255,0.7); --plotly-grid:rgba(0,0,0,0.06); --plotly-text:#475569;
	--alert-bg:rgba(239,68,68,0.07); --alert-border:rgba(239,68,68,0.25);
	--pin-bg:rgba(234,179,8,0.08); --pin-border:rgba(234,179,8,0.3);
	}

	html,body,[data-testid="stAppViewContainer"],[data-testid="stMain"],.main .block-container {
	background:var(--bg)!important; color:var(--text-1)!important;
	font-family:'Space Grotesk',sans-serif!important; transition:background 0.3s,color 0.3s;
	}
	[data-testid="stSidebar"] { background:var(--bg-card)!important; border-right:1px solid var(--border)!important; transition:background 0.3s; }
	[data-testid="stHeader"] { background:transparent!important; }
	::-webkit-scrollbar{width:4px;} ::-webkit-scrollbar-track{background:var(--bg);}
	::-webkit-scrollbar-thumb{background:linear-gradient(var(--accent),var(--accent2));border-radius:4px;}

	[data-testid="metric-container"] {
	background:var(--bg-card)!important; border:1px solid var(--border)!important;
	border-radius:16px!important; padding:18px!important; box-shadow:var(--shadow-sm)!important; transition:background 0.3s;
	}
	[data-testid="stMetricLabel"]{color:var(--text-2)!important;font-size:0.8rem!important;}
	[data-testid="stMetricValue"]{color:var(--text-1)!important;font-weight:700!important;}
	[data-testid="stMetricDelta"]{color:var(--accent-text)!important;}

	.stTextInput input { background:var(--input-bg)!important; border:1px solid var(--input-border)!important; border-radius:10px!important; color:var(--text-1)!important; }
	.stTextInput input::placeholder { color:var(--text-3)!important; opacity:1!important; }
	[data-testid="stSidebar"] .stTextInput input { background:#1a1a2e!important; border:1px solid rgba(124,58,237,0.4)!important; color:#f1f5f9!important; font-weight:500!important; }
	[data-testid="stSidebar"] .stTextInput input::placeholder { color:#64748b!important; }
	[data-testid="stSidebar"] .stTextInput input:focus { border-color:var(--accent)!important; box-shadow:0 0 0 2px rgba(124,58,237,0.2)!important; outline:none!important; }
	[data-testid="stSidebar"] label { color:var(--text-2)!important; }
	[data-baseweb="select"]>div { background:var(--input-bg)!important; border:1px solid var(--input-border)!important; border-radius:10px!important; color:var(--text-1)!important; }
	.stButton>button { background:linear-gradient(135deg,var(--accent),var(--accent2))!important; color:#fff!important; border:none!important; border-radius:10px!important; font-weight:600!important; font-family:'Space Grotesk',sans-serif!important; box-shadow:0 4px 16px rgba(124,58,237,0.3)!important; transition:all 0.2s!important; }
	.stButton>button:hover{transform:translateY(-2px)!important;}
	hr{border:none!important;border-top:1px solid var(--divider)!important;margin:1.2rem 0!important;}
	[data-testid="stSidebar"] label,[data-testid="stSidebar"] .stMarkdown p{color:var(--text-2)!important;font-size:0.83rem!important;}

	[data-testid="stDownloadButton"]>button { background:var(--bg-card)!important; color:var(--text-2)!important; border:1px solid var(--border)!important; border-radius:8px!important; font-size:0.75rem!important; box-shadow:none!important; }
	[data-testid="stDownloadButton"]>button:hover { background:var(--pill-bg)!important; color:var(--accent-text)!important; border-color:var(--pill-border)!important; }

	[data-testid="stCheckbox"] label, [data-testid="stCheckbox"] span { color:var(--text-2)!important; font-size:0.82rem!important; }
	[data-testid="stCheckbox"] [data-testid="stWidgetLabel"] { color:var(--text-2)!important; }

	@keyframes pulse{0%{box-shadow:0 0 0 0 rgba(34,197,94,0.7);}70%{box-shadow:0 0 0 10px rgba(34,197,94,0);}100%{box-shadow:0 0 0 0 rgba(34,197,94,0);}}
	.live-dot{display:inline-block;width:9px;height:9px;background:var(--live);border-radius:50%;animation:pulse 1.8s infinite;margin-right:6px;vertical-align:middle;}

	@keyframes alertPulse{0%{opacity:1;}50%{opacity:0.7;}100%{opacity:1;}}
	.alert-banner{background:var(--alert-bg);border:1px solid var(--alert-border);border-radius:14px;padding:14px 18px;margin:12px 0;display:flex;align-items:center;gap:12px;animation:alertPulse 2s infinite;}
	.alert-icon{font-size:1.4rem;}
	.alert-text{font-size:0.88rem;font-weight:600;color:#ef4444;}
	.alert-sub{font-size:0.75rem;color:var(--text-3);margin-top:2px;}

	.stat-grid{display:flex;gap:12px;margin:10px 0 18px;flex-wrap:wrap;}
	.stat-card{flex:1;min-width:130px;background:var(--bg-card);border:1px solid var(--border);border-radius:20px;padding:22px 18px;text-align:center;transition:transform 0.2s,box-shadow 0.2s,background 0.3s;position:relative;overflow:hidden;box-shadow:var(--shadow-sm);}
	.stat-card:hover{transform:translateY(-4px);box-shadow:var(--shadow);}
	.stat-accent{position:absolute;top:0;left:0;right:0;height:3px;border-radius:20px 20px 0 0;}
	.stat-number{font-size:2.6rem;font-weight:800;line-height:1;margin-bottom:6px;letter-spacing:-0.03em;}
	.stat-label{font-size:0.82rem;color:var(--text-2);font-weight:600;text-transform:uppercase;letter-spacing:0.06em;}
	.stat-sub{font-size:0.7rem;color:var(--text-3);margin-top:4px;}

	.velocity-card{background:var(--bg-card);border:1px solid var(--border);border-radius:20px;padding:18px 22px;box-shadow:var(--shadow-sm);display:flex;align-items:center;gap:16px;}
	.velocity-arrow{font-size:2rem;line-height:1;}
	.velocity-val{font-size:1.6rem;font-weight:800;letter-spacing:-0.03em;}
	.velocity-label{font-size:0.75rem;color:var(--text-3);font-weight:600;text-transform:uppercase;letter-spacing:0.06em;margin-top:2px;}

	.sec-hdr{display:flex;align-items:center;gap:10px;margin:6px 0 14px;}
	.sec-ttl{font-size:1rem;font-weight:700;color:var(--text-1);letter-spacing:-0.01em;}
	.sec-pill{background:var(--pill-bg);border:1px solid var(--pill-border);border-radius:20px;padding:2px 10px;font-size:0.68rem;color:var(--pill-text);font-weight:700;text-transform:uppercase;letter-spacing:0.08em;}

	.chart-wrap{background:var(--bg-card);border:1px solid var(--border);border-radius:20px;padding:14px 14px 6px;box-shadow:var(--shadow-sm);transition:background 0.3s,border 0.3s;}
	.chart-title{font-size:0.88rem;font-weight:700;color:var(--text-1);margin-bottom:2px;}
	.chart-sub{font-size:0.72rem;color:var(--text-3);margin-bottom:10px;}

	.topic-grid{display:flex;gap:10px;flex-wrap:wrap;margin-bottom:18px;}
	.topic-pill{background:var(--bg-card);border-radius:16px;padding:14px 20px;text-align:center;min-width:110px;box-shadow:var(--shadow-sm);transition:transform 0.2s,box-shadow 0.2s;}
	.topic-pill:hover{transform:translateY(-3px);box-shadow:var(--shadow);}
	.topic-count{font-size:1.4rem;font-weight:800;letter-spacing:-0.02em;}
	.topic-name{font-size:0.7rem;color:var(--text-3);margin-top:3px;font-weight:600;text-transform:uppercase;letter-spacing:0.06em;}

	@keyframes slideIn{from{opacity:0;transform:translateY(6px);}to{opacity:1;transform:translateY(0);}}
	.chat-card{background:var(--bg-card);border:1px solid var(--border);border-radius:16px;padding:14px 16px;margin-bottom:10px;border-left:3px solid transparent;animation:slideIn 0.2s ease;transition:background 0.2s,transform 0.15s,box-shadow 0.2s;box-shadow:var(--shadow-sm);}
	.chat-card:hover{transform:translateX(4px);box-shadow:var(--shadow);}
	.chat-positive{border-left-color:#22c55e;} .chat-negative{border-left-color:#ef4444;} .chat-neutral{border-left-color:#eab308;}
	.chat-pinned{border-left-color:#eab308!important;background:var(--pin-bg)!important;border-color:var(--pin-border)!important;}
	.chat-author{font-weight:700;font-size:0.83rem;color:var(--accent-text);margin-bottom:5px;}
	.chat-text{font-size:0.92rem;color:var(--text-2);line-height:1.55;margin-bottom:9px;}
	.chat-badges{display:flex;gap:6px;flex-wrap:wrap;}
	.badge{display:inline-flex;align-items:center;background:var(--badge-bg);border:1px solid var(--border);border-radius:20px;padding:3px 10px;font-size:0.7rem;font-weight:600;color:var(--text-2);}
	.pin-badge{background:rgba(234,179,8,0.15);border-color:rgba(234,179,8,0.4);color:#eab308;}

	.compare-label{font-size:0.72rem;font-weight:700;text-transform:uppercase;letter-spacing:0.08em;padding:3px 10px;border-radius:20px;display:inline-block;margin-bottom:8px;}

	.engage-card{background:var(--bg-card);border:1px solid var(--border);border-radius:20px;padding:20px 24px;box-shadow:var(--shadow-sm);position:relative;overflow:hidden;}
	.engage-score{font-size:3rem;font-weight:800;letter-spacing:-0.04em;line-height:1;}
	.engage-label{font-size:0.75rem;color:var(--text-3);font-weight:600;text-transform:uppercase;letter-spacing:0.08em;margin-top:4px;}
	.engage-bar-bg{background:var(--border);border-radius:99px;height:6px;margin-top:12px;overflow:hidden;}
	.engage-bar-fill{height:6px;border-radius:99px;transition:width 0.6s ease;}
	.engage-breakdown{display:flex;gap:16px;margin-top:10px;flex-wrap:wrap;}
	.engage-item{font-size:0.72rem;color:var(--text-3);}
	.engage-item span{font-weight:700;color:var(--text-2);}

	.leaderboard-row{display:flex;align-items:center;gap:12px;padding:10px 14px;background:var(--bg-card);border:1px solid var(--border);border-radius:14px;margin-bottom:8px;transition:transform 0.15s,box-shadow 0.15s;}
	.leaderboard-row:hover{transform:translateX(4px);box-shadow:var(--shadow);}
	.lb-rank{font-size:1rem;font-weight:800;color:var(--text-3);min-width:28px;}
	.lb-rank.gold{color:#f59e0b;} .lb-rank.silver{color:#94a3b8;} .lb-rank.bronze{color:#b45309;}
	.lb-author{font-size:0.85rem;font-weight:700;color:var(--text-1);flex:1;overflow:hidden;text-overflow:ellipsis;white-space:nowrap;}
	.lb-count{font-size:0.78rem;color:var(--text-3);min-width:40px;text-align:right;}
	.lb-bar{flex:2;height:5px;background:var(--border);border-radius:99px;overflow:hidden;}
	.lb-bar-fill{height:5px;border-radius:99px;}
	.lb-sent{display:flex;gap:4px;min-width:80px;justify-content:flex-end;}
	.lb-dot{width:8px;height:8px;border-radius:50%;display:inline-block;}

	.spam-alert{background:rgba(239,68,68,0.08);border:1px solid rgba(239,68,68,0.25);border-radius:14px;padding:14px 18px;margin:12px 0;display:flex;align-items:center;gap:12px;}
	.spam-alert-text{font-size:0.88rem;font-weight:600;color:#ef4444;}
	.spam-alert-sub{font-size:0.75rem;color:var(--text-3);margin-top:2px;}

	.empty-state{text-align:center;padding:80px 20px;background:var(--bg-card);border:1px solid var(--border);border-radius:24px;margin:40px 0;box-shadow:var(--shadow-sm);}
	.empty-icon{font-size:3.5rem;margin-bottom:16px;}
	.empty-title{font-size:1.1rem;color:var(--text-2);font-weight:700;}
	.empty-sub{font-size:0.84rem;color:var(--text-3);margin-top:6px;}

	[data-testid="stSidebar"] [role="radiogroup"] { display:flex; flex-direction:row; flex-wrap:nowrap; gap:4px; }
	[data-testid="stSidebar"] [role="radiogroup"] label { flex:1; display:flex; align-items:center; justify-content:center; background:var(--bg-card); border:1px solid var(--pill-border); border-radius:8px; padding:6px 2px; cursor:pointer; transition:background 0.15s,border 0.15s; }
	[data-testid="stSidebar"] [role="radiogroup"] label:hover { background:var(--pill-bg); border-color:var(--accent); }
	[data-testid="stSidebar"] [role="radiogroup"] label[data-checked="true"],
	[data-testid="stSidebar"] [role="radiogroup"] label:has(input:checked) { background:linear-gradient(135deg,var(--accent),var(--accent2)); border-color:var(--accent); }
	[data-testid="stSidebar"] [role="radiogroup"] label p,
	[data-testid="stSidebar"] [role="radiogroup"] label span { font-size:0.82rem !important; font-weight:700 !important; color:var(--text-1) !important; white-space:nowrap !important; }
	[data-testid="stSidebar"] [role="radiogroup"] label:has(input:checked) p,
	[data-testid="stSidebar"] [role="radiogroup"] label:has(input:checked) span { color:#fff !important; }
	[data-testid="stSidebar"] [role="radiogroup"] input[type="radio"] { display:none !important; }
	[data-testid="stSidebar"] [data-testid="stWidgetLabel"]:has(+ [role="radiogroup"]) { color:var(--text-2) !important; font-size:0.75rem !important; margin-bottom:4px; }
	</style>"""

	st.markdown(THEME_JS, unsafe_allow_html=True)
	st.markdown(CSS, unsafe_allow_html=True)


	# ── HELPERS ──────────────────────────────────────────────────
	def extract_video_id(url_or_id):
	url_or_id = url_or_id.strip()
	match = re.search(r"(?:v=\|/live/\|youtu\.be/)([A-Za-z0-9_-]{11})", url_or_id)
	if match:
	return match.group(1)
	if re.match(r"^[A-Za-z0-9_-]{11}$", url_or_id):
	return url_or_id
	return url_or_id

	def update_config_video_id(video_id):
	config_path = os.path.join(os.path.dirname(__file__), '..', 'backend', 'config.py')
	with open(config_path, 'r') as f:
	content = f.read()
	content = re.sub(r'VIDEO_ID\s=\s".*?"', f'VIDEO_ID = "{video_id}"', content)
	with open(config_path, 'w') as f:
	f.write(content)

	def fetch_video_title(video_id):
	try:
	import urllib.request
	url = f"https://www.youtube.com/oembed?url=https://www.youtube.com/watch?v={video_id}&format=json"
	with urllib.request.urlopen(url, timeout=5) as resp:
	return json.loads(resp.read())["title"]
	except Exception:
	return None

	def clean_topic(val):
	if pd.isna(val) or str(val).strip() == "" or str(val).strip().lower() == "nan":
	return "General"
	return str(val).strip()

	def clean_sentiment(val):
	if str(val).strip() in ("Positive", "Negative", "Neutral"):
	return str(val).strip()
	return "Neutral"

	def plotly_layout(height=280):
	return dict(
	paper_bgcolor="rgba(0,0,0,0)",
	plot_bgcolor="rgba(0,0,0,0)",
	height=height,
	margin=dict(l=10, r=10, t=10, b=10),
	font=dict(family="Space Grotesk"),
	xaxis=dict(showgrid=False, zeroline=False, showline=False,
	tickfont=dict(size=11), title=None),
	yaxis=dict(showgrid=True, gridcolor="rgba(128,128,128,0.12)",
	zeroline=False, showline=False, tickfont=dict(size=11), title=None),
	showlegend=False,
	hoverlabel=dict(font_family="Space Grotesk", font_size=12),
	)

	def csv_download(df_export, label, filename):
	csv = df_export.to_csv(index=False).encode("utf-8")
	st.download_button(label=f"⬇ {label}", data=csv,
	file_name=filename, mime="text/csv", key=filename)

	@st.cache_data(ttl=5, show_spinner=False)
	def load_stream_data(redis_key: str, limit: int \| None = None):
	"""Load and parse messages from a Redis key. Cached for 5s to avoid redundant reads."""
	if limit:
	raws = r.lrange(redis_key, -limit, -1)
	else:
	raws = r.lrange(redis_key, 0, -1)
	data = []
	for raw in raws:
	try:
	data.append(json.loads(raw))
	except Exception:
	pass
	return data

	@st.cache_data(ttl=10, show_spinner=False)
	def compute_velocity(df_all_json: str, window: int = 20) -> dict:
	"""
	Compute sentiment velocity. Accepts JSON string for cache key compatibility.
	"""
	import json as _json
	sentiments = [m.get("sentiment", "Neutral") for m in _json.loads(df_all_json)]
	n = len(sentiments)
	if n < window * 2:
	return {"direction": "→", "delta": 0.0, "label": "Stable", "color": "#eab308"}
	recent = sentiments[-window:]
	prev = sentiments[-window*2:-window]
	r_pos = sum(1 for s in recent if s == "Positive") / window
	p_pos = sum(1 for s in prev if s == "Positive") / window
	delta = r_pos - p_pos
	if delta > 0.08:
	return {"direction": "↑", "delta": delta, "label": "Rising", "color": "#22c55e"}
	elif delta < -0.08:
	return {"direction": "↓", "delta": delta, "label": "Falling", "color": "#ef4444"}
	return {"direction": "→", "delta": delta, "label": "Stable", "color": "#eab308"}


	@st.cache_data(ttl=10, show_spinner=False)
	def build_heatmap_data(df_all_json: str, bucket_minutes: int = 1) -> pd.DataFrame:
	"""
	Bucket messages into time intervals. Accepts JSON string for cache key compatibility.
	"""
	import json as _json
	records = _json.loads(df_all_json)
	if not records:
	return pd.DataFrame()
	df_t = pd.DataFrame(records)
	if "time" not in df_t.columns:
	return pd.DataFrame()
	df_t["time"] = pd.to_datetime(df_t["time"], errors="coerce")
	df_t = df_t.dropna(subset=["time"])
	if df_t.empty:
	return pd.DataFrame()
	df_t["bucket"] = df_t["time"].dt.floor(f"{bucket_minutes}min")
	grouped = df_t.groupby(["bucket", "sentiment"]).size().unstack(fill_value=0)
	for col in ["Positive", "Neutral", "Negative"]:
	if col not in grouped.columns:
	grouped[col] = 0
	grouped = grouped.reset_index()
	grouped.columns.name = None
	return grouped[["bucket", "Positive", "Neutral", "Negative"]]

	def check_alert(df_all: pd.DataFrame, threshold: float = 0.4, window: int = 15) -> dict \| None:
	"""Return alert info if negative ratio in last `window` messages exceeds threshold."""
	if len(df_all) < window:
	return None
	recent = df_all.iloc[-window:]
	neg_ratio = (recent["sentiment"] == "Negative").mean()
	if neg_ratio >= threshold:
	return {
	"neg_ratio": neg_ratio,
	"count": int((recent["sentiment"] == "Negative").sum()),
	"window": window,
	}
	return None


	@st.cache_data(ttl=10, show_spinner=False)
	def compute_engagement(all_data_json: str, window: int = 50) -> dict:
	"""
	Engagement score (0–100) = weighted combo of:
	- message rate (msgs per minute, last window)
	- positive ratio (last window)
	- question density (last window)
	"""
	import json as _j
	msgs = _j.loads(all_data_json)
	if not msgs:
	return {"score": 0, "rate": 0.0, "pos_ratio": 0.0, "q_density": 0.0, "grade": "—"}

	recent = msgs[-window:]
	n = len(recent)

	# Message rate: msgs per minute using timestamps
	rate = 0.0
	try:
	t0 = datetime.fromisoformat(recent[0]["time"])
	t1 = datetime.fromisoformat(recent[-1]["time"])
	elapsed = max((t1 - t0).total_seconds() / 60, 0.1)
	rate = round(n / elapsed, 1)
	except Exception:
	rate = float(n)

	pos_ratio = sum(1 for m in recent if m.get("sentiment") == "Positive") / max(n, 1)
	q_density = sum(1 for m in recent if m.get("topic") == "Question") / max(n, 1)

	# Normalise rate: cap at 60 msgs/min = 100%
	rate_norm = min(rate / 60, 1.0)
	score = round((rate_norm * 0.4 + pos_ratio * 0.4 + q_density * 0.2) * 100)

	if score >= 70: grade = "🔥 High"
	elif score >= 40: grade = "⚡ Medium"
	else: grade = "💤 Low"

	return {"score": score, "rate": rate, "pos_ratio": pos_ratio, "q_density": q_density, "grade": grade}


	@st.cache_data(ttl=10, show_spinner=False)
	def compute_top_contributors(all_data_json: str, top_n: int = 10) -> list[dict]:
	"""Return top N authors by message count with sentiment + topic breakdown."""
	import json as _j
	from collections import Counter
	msgs = _j.loads(all_data_json)
	if not msgs:
	return []

	TOPICS = ["Appreciation", "Question", "Request/Feedback", "Promo", "Spam", "General", "MCQ Answer"]
	author_data: dict[str, dict] = {}
	for m in msgs:
	a = m.get("author", "Unknown")
	if a not in author_data:
	author_data[a] = {
	"count": 0,
	"Positive": 0, "Neutral": 0, "Negative": 0,
	**{t: 0 for t in TOPICS},
	}
	author_data[a]["count"] += 1
	s = m.get("sentiment", "Neutral")
	if s in ("Positive", "Neutral", "Negative"):
	author_data[a][s] += 1
	t = m.get("topic", "General")
	if t not in TOPICS:
	t = "General"
	author_data[a][t] += 1

	sorted_authors = sorted(author_data.items(), key=lambda x: x[1]["count"], reverse=True)[:top_n]
	result = []
	for author, d in sorted_authors:
	total = max(d["count"], 1)
	result.append({
	"author": author,
	"count": d["count"],
	"pos_pct": round(d["Positive"] / total * 100),
	"neu_pct": round(d["Neutral"] / total * 100),
	"neg_pct": round(d["Negative"] / total * 100),
	"t_appr": round(d["Appreciation"] / total * 100),
	"t_ques": round(d["Question"] / total * 100),
	"t_rf": round(d["Request/Feedback"] / total * 100),
	"t_promo": round(d["Promo"] / total * 100),
	"t_spam": round(d["Spam"] / total * 100),
	"t_gen": round(d["General"] / total * 100),
	"t_mcq": round(d["MCQ Answer"] / total * 100),
	})
	return result


	@st.cache_data(ttl=10, show_spinner=False)
	def compute_word_freq(all_data_json: str, sentiment_filter: str = "All",
	topic_filter: str = "All", top_n: int = 60) -> list[tuple[str, int]]:
	"""Return top N (word, count) pairs after filtering stopwords."""
	import json as _j
	from collections import Counter

	STOPWORDS = {
	"the","a","an","is","it","in","on","at","to","of","and","or","but","for",
	"with","this","that","are","was","be","as","by","from","have","has","had",
	"not","no","so","if","do","did","will","can","just","i","you","he","she",
	"we","they","my","your","his","her","our","their","me","him","us","them",
	"what","how","why","when","where","who","which","there","here","been",
	"would","could","should","may","might","shall","than","then","now","also",
	"more","very","too","up","out","about","into","over","after","before",
	"yaar","bhi","hai","hain","ho","kar","ke","ki","ka","ko","se","ne","ye",
	"vo","woh","aur","nahi","nhi","toh","toh","koi","kuch","ab","ek","hi",
	}

	msgs = _j.loads(all_data_json)
	words: list[str] = []
	for m in msgs:
	if sentiment_filter != "All" and m.get("sentiment") != sentiment_filter:
	continue
	if topic_filter != "All" and m.get("topic") != topic_filter:
	continue
	text = re.sub(r"[^\w\s]", " ", m.get("text", "").lower())
	for w in text.split():
	if len(w) > 2 and w not in STOPWORDS and not w.isdigit():
	words.append(w)

	return Counter(words).most_common(top_n)


	def check_spam_alert(df_all: pd.DataFrame, threshold: float = 0.3, window: int = 20) -> dict \| None:
	"""Return alert if spam ratio in last `window` messages exceeds threshold."""
	if "topic" not in df_all.columns or len(df_all) < window:
	return None
	recent = df_all.iloc[-window:]
	spam_ratio = (recent["topic"] == "Spam").mean()
	if spam_ratio >= threshold:
	return {
	"spam_ratio": spam_ratio,
	"count": int((recent["topic"] == "Spam").sum()),
	"window": window,
	}
	return None


	@st.cache_data(ttl=10, show_spinner=False)
	def detect_repeat_spammers(all_data_json: str, window_sec: int = 15, min_repeats: int = 2) -> list[dict]:
	"""
	Detect users who send the same (or near-identical) message multiple times
	within `window_sec` seconds. Returns list of spam burst dicts sorted by
	repeat count descending.
	"""
	import json as _j
	import re as _re

	msgs = _j.loads(all_data_json)
	if not msgs:
	return []

	def _normalize(t: str) -> str:
	return _re.sub(r"[^\w]", "", t.lower().strip())

	bursts: dict[tuple, dict] = {}
	for m in msgs:
	author = m.get("author", "Unknown")
	text = m.get("text", "").strip()
	if not text:
	continue
	norm = _normalize(text)
	if len(norm) < 4:
	continue
	ts_str = m.get("time", "")
	try:
	ts = datetime.fromisoformat(ts_str)
	except Exception:
	continue
	key = (author, norm)
	if key not in bursts:
	bursts[key] = {
	"author": author,
	"text": text,
	"topic": m.get("topic", "General"),
	"sentiment": m.get("sentiment", "Neutral"),
	"timestamps": [],
	}
	bursts[key]["timestamps"].append(ts)

	results = []
	for key, burst in bursts.items():
	times = sorted(burst["timestamps"])
	max_in_window = 1
	for i in range(len(times)):
	count_in_window = sum(
	1 for t in times[i:]
	if (t - times[i]).total_seconds() <= window_sec
	)
	max_in_window = max(max_in_window, count_in_window)

	if max_in_window >= min_repeats:
	results.append({
	"author": burst["author"],
	"text": burst["text"],
	"topic": burst["topic"],
	"sentiment": burst["sentiment"],
	"count": len(times),
	"max_burst": max_in_window,
	"first_seen": times[0].strftime("%H:%M:%S"),
	"last_seen": times[-1].strftime("%H:%M:%S"),
	})

	return sorted(results, key=lambda x: x["max_burst"], reverse=True)


	# ── SESSION STATE INIT ────────────────────────────────────────
	MAX_STREAMS = 5
	STREAM_COLORS = ["#7c3aed", "#10b981", "#f59e0b", "#3b82f6", "#ec4899"]
	STREAM_NAMES = ["A", "B", "C", "D", "E"]

	if "pinned_messages" not in st.session_state:
	st.session_state.pinned_messages = []
	if "alert_dismissed" not in st.session_state:
	st.session_state.alert_dismissed = False
	if "last_alert_count" not in st.session_state:
	st.session_state.last_alert_count = 0
	if "last_view" not in st.session_state:
	st.session_state.last_view = "💬 Comments"
	# Multi-stream: list of dicts {video_id, redis_key, label, proc}
	if "streams" not in st.session_state:
	st.session_state.streams = [
	{"video_id": "", "redis_key": "chat_messages", "label": "Stream A", "proc": None}
	]

	# ── SIDEBAR ──────────────────────────────────────────────────
	with st.sidebar:
	st.markdown(
	'<div style="padding:12px 0 20px;">'
	'<div style="font-size:1.35rem;font-weight:800;color:var(--text-1);letter-spacing:-0.02em;">📡 LivePulse</div>'
	'<div style="font-size:0.75rem;color:var(--text-3);margin-top:2px;">YouTube Chat Analytics</div>'
	'</div>', unsafe_allow_html=True
	)
	st.divider()

	# ── Display Settings ──
	st.markdown('<p style="font-size:0.68rem;font-weight:700;color:var(--accent);text-transform:uppercase;letter-spacing:0.1em;margin-bottom:8px;">Display Settings</p>', unsafe_allow_html=True)
	active_view = st.radio(
	"View",
	options=["💬 Comments", "📊 Stats & Info"],
	index=0,
	horizontal=True,
	key="active_view",
	)
	refresh_rate = st.radio(
	"Refresh interval (s)",
	options=[10, 20, 30, 40, 50, 60],
	index=0,
	horizontal=True,
	key="refresh_rate",
	)
	msg_limit = st.slider("Message window", 10, 400, 50, step=10)
	auto_refresh = st.toggle("Live auto-refresh", value=True)
	st.divider()

	# ── Alert Settings ──
	st.markdown('<p style="font-size:0.68rem;font-weight:700;color:var(--accent);text-transform:uppercase;letter-spacing:0.1em;margin-bottom:8px;">Alert Settings</p>', unsafe_allow_html=True)
	alert_enabled = st.toggle("Negative spike alerts", value=True)
	alert_threshold = st.slider("Neg alert threshold (%)", 20, 80, 40) / 100
	alert_window = st.slider("Alert window (msgs)", 5, 30, 15)
	spam_alert_on = st.toggle("Spam rate alerts", value=True)
	spam_threshold = st.slider("Spam alert threshold (%)", 10, 60, 30) / 100
	st.divider()

	# ── Multi-Stream Scraper Control ──
	st.markdown('<p style="font-size:0.68rem;font-weight:700;color:var(--accent);text-transform:uppercase;letter-spacing:0.1em;margin-bottom:8px;">Stream Control</p>', unsafe_allow_html=True)

	import importlib
	import config as _cfg
	importlib.reload(_cfg)

	# Pre-fill Stream A video_id from config on first load
	if st.session_state.streams[0]["video_id"] == "":
	st.session_state.streams[0]["video_id"] = _cfg.VIDEO_ID

	for idx, stream in enumerate(st.session_state.streams):
	color = STREAM_COLORS[idx]
	label = STREAM_NAMES[idx]
	st.markdown(
	f'<div style="font-size:0.72rem;font-weight:700;color:{color};text-transform:uppercase;'
	f'letter-spacing:0.08em;margin:10px 0 4px;border-left:3px solid {color};padding-left:8px;">'
	f'Stream {label}</div>',
	unsafe_allow_html=True
	)
	# Use widget key as the source of truth — never override with value= after first set
	vid_skey = f"vid_{idx}"
	rkey_skey = f"rkey_{idx}"
	if vid_skey not in st.session_state:
	st.session_state[vid_skey] = stream["video_id"]
	if rkey_skey not in st.session_state:
	st.session_state[rkey_skey] = stream["redis_key"]

	st.text_input("Video ID / URL", placeholder="e.g. eFSK2-QRB0A", key=vid_skey)
	st.text_input("Redis key", placeholder=f"chat_messages_{label.lower()}", key=rkey_skey)

	sc1, sc2 = st.columns(2)
	with sc1:
	if st.button("▶ Start", key=f"start_{idx}", width='stretch'):
	vid = extract_video_id(st.session_state[vid_skey])
	rkey = st.session_state[rkey_skey].strip() or f"chat_messages_{label.lower()}"
	if vid:
	# Stop existing proc for this slot
	old_proc = st.session_state.streams[idx].get("proc")
	if old_proc and old_proc.poll() is None:
	old_proc.terminate()
	proc = subprocess.Popen(
	[sys.executable, "-m", "backend.scraper",
	"--video_id", vid, "--redis_key", rkey],
	cwd=os.path.abspath(os.path.join(os.path.dirname(__file__), "..")),
	stdout=subprocess.DEVNULL,
	stderr=subprocess.DEVNULL,
	)
	st.session_state.streams[idx]["proc"] = proc
	st.session_state.streams[idx]["video_id"] = vid
	st.session_state.streams[idx]["redis_key"] = rkey
	# Fetch and store title for ALL streams (used in header pills)
	_title = fetch_video_title(vid)
	st.session_state.streams[idx]["video_title"] = _title or vid
	if idx == 0:
	update_config_video_id(vid)
	r.set("video_title", _title) if _title else r.delete("video_title")
	st.session_state.alert_dismissed = False
	st.success(f"Stream {label} started → `{rkey}`")
	else:
	st.error("Invalid video ID")
	with sc2:
	if st.button("⏹ Stop", key=f"stop_{idx}", width='stretch'):
	proc = st.session_state.streams[idx].get("proc")
	if proc and proc.poll() is None:
	proc.terminate()
	st.session_state.streams[idx]["proc"] = None
	st.success(f"Stream {label} stopped")
	else:
	st.warning("Not running")

	proc = st.session_state.streams[idx].get("proc")
	running = proc is not None and proc.poll() is None
	dot_color = "#22c55e" if running else "#ef4444"
	status = "running" if running else "stopped"
	st.markdown(f'<div style="font-size:0.72rem;color:{dot_color};margin-bottom:4px;">● {status}</div>', unsafe_allow_html=True)

	st.divider()

	# ── Add / Remove stream slots ──
	add_col, rem_col = st.columns(2)
	with add_col:
	if len(st.session_state.streams) < MAX_STREAMS:
	if st.button("＋ Add stream", width='stretch'):
	n = len(st.session_state.streams)
	st.session_state.streams.append({
	"video_id": "",
	"redis_key": f"chat_messages_{STREAM_NAMES[n].lower()}",
	"label": f"Stream {STREAM_NAMES[n]}",
	"proc": None,
	})
	st.rerun()
	with rem_col:
	if len(st.session_state.streams) > 1:
	if st.button("－ Remove last", width='stretch'):
	removed = st.session_state.streams.pop()
	proc = removed.get("proc")
	if proc and proc.poll() is None:
	proc.terminate()
	st.rerun()

	st.divider()

	# ── Pinned Messages ──
	st.markdown('<p style="font-size:0.68rem;font-weight:700;color:var(--accent);text-transform:uppercase;letter-spacing:0.1em;margin-bottom:8px;">Pinned Messages</p>', unsafe_allow_html=True)
	pin_count = len(st.session_state.pinned_messages)
	st.markdown(f'<div style="font-size:0.78rem;color:var(--text-3);">{pin_count} message{"s" if pin_count != 1 else ""} pinned</div>', unsafe_allow_html=True)
	if pin_count > 0 and st.button("🗑 Clear pins", width='stretch'):
	st.session_state.pinned_messages = []
	st.rerun()
	st.divider()

	# ── Download Data ──
	st.markdown('<p style="font-size:0.68rem;font-weight:700;color:var(--accent);text-transform:uppercase;letter-spacing:0.1em;margin-bottom:8px;">Download Data</p>', unsafe_allow_html=True)
	_active_streams = [s for s in st.session_state.streams if s.get("redis_key")]
	if _active_streams:
	for _s in _active_streams:
	_rkey = _s["redis_key"]
	_slabel = _s["label"]
	_all_raws = r.lrange(_rkey, 0, -1)
	_dl_rows = []
	for _raw in _all_raws:
	try:
	_dl_rows.append(json.loads(_raw))
	except Exception:
	pass
	if _dl_rows:
	_dl_df = pd.DataFrame(_dl_rows)
	_ts = datetime.now().strftime("%Y%m%d_%H%M%S")
	_fname = f"livepulse_{_rkey}_{_ts}.csv"
	_csv_bytes = _dl_df.to_csv(index=False).encode("utf-8")
	st.download_button(
	label=f"⬇ {_slabel} ({len(_dl_rows)} msgs)",
	data=_csv_bytes,
	file_name=_fname,
	mime="text/csv",
	key=f"dl_{_rkey}",
	)
	# PDF button removed — use the Export button on the Stats page instead
	else:
	st.markdown(f'<div style="font-size:0.72rem;color:var(--text-3);">{_slabel}: no data yet</div>', unsafe_allow_html=True)
	else:
	st.markdown('<div style="font-size:0.72rem;color:var(--text-3);">No active streams</div>', unsafe_allow_html=True)
	st.divider()

	# ── Export ──
	st.markdown('<p style="font-size:0.68rem;font-weight:700;color:var(--accent);text-transform:uppercase;letter-spacing:0.1em;margin-bottom:8px;">Export</p>', unsafe_allow_html=True)
	st.markdown(
	'<div style="font-size:0.7rem;color:var(--text-3);margin-bottom:6px;">'
	'\u26a0\ufe0f Go to <b style="color:var(--accent-text);">Stats & Info</b> tab first, then click.</div>',
	unsafe_allow_html=True
	)
	import streamlit.components.v1 as _comp2
	_comp2.html("""
	<div style="padding:2px 0;">
	<button id="sidebarScreenshotBtn2" style="
	width:100%; background:linear-gradient(135deg,#7c3aed,#4f46e5);
	color:#fff; border:none; border-radius:10px; padding:8px 12px;
	font-size:13px; font-weight:600; cursor:pointer;
	box-shadow:0 4px 16px rgba(124,58,237,0.3); transition:transform 0.2s;"
	onmouseover="this.style.transform='translateY(-2px)'"
	onmouseout="this.style.transform='translateY(0)'"
	onclick="sidebarCapture2()">
	📷 Download Stats as PDF
	</button>
	<div id="sidebarMsg2" style="margin-top:6px;font-size:11px;color:#94a3b8;text-align:center;"></div>
	</div>
	<script>
	async function sidebarCapture2() {
	const btn = document.getElementById('sidebarScreenshotBtn2');
	const msg = document.getElementById('sidebarMsg2');
	btn.disabled = true; btn.textContent = 'Capturing...';
	msg.textContent = 'Please wait...';
	try {
	const target = window.parent.document.querySelector('[data-testid="stMain"]')
	\|\| window.parent.document.querySelector('.main')
	\|\| window.parent.document.body;
	const canvas = await window.parent.html2canvas(target, {
	scale:1.5, useCORS:true, allowTaint:true,
	backgroundColor:'#07070f', logging:false,
	windowWidth:target.scrollWidth, windowHeight:target.scrollHeight,
	scrollX:0, scrollY:0,
	});
	const imgData = canvas.toDataURL('image/png', 0.95);
	const { jsPDF } = window.parent.jspdf;
	const pdf = new jsPDF({
	orientation: canvas.width > canvas.height ? 'l' : 'p',
	unit:'px', format:[canvas.width, canvas.height], compress:true,
	});
	pdf.addImage(imgData, 'PNG', 0, 0, canvas.width, canvas.height);
	const ts = new Date().toISOString().slice(0,16).replace('T','_').replace(':','-');
	pdf.save('livepulse_stats_' + ts + '.pdf');
	btn.textContent = 'Download Stats as PDF'; btn.disabled = false;
	msg.textContent = 'Done!';
	setTimeout(() => { msg.textContent = ''; }, 3000);
	} catch(e) {
	btn.textContent = 'Download Stats as PDF'; btn.disabled = false;
	msg.textContent = 'Error: ' + e.message;
	}
	}
	function loadScript2(src, name) {
	return new Promise(r => {
	if (window.parent[name]) { r(); return; }
	const s = window.parent.document.createElement('script');
	s.src = src; s.onload = r;
	window.parent.document.head.appendChild(s);
	});
	}
	(async () => {
	await loadScript2('https://cdnjs.cloudflare.com/ajax/libs/html2canvas/1.4.1/html2canvas.min.js','html2canvas');
	await loadScript2('https://cdnjs.cloudflare.com/ajax/libs/jspdf/2.5.1/jspdf.umd.min.js','jspdf');
	})();
	</script>
	""", height=75)
	st.divider()

	# ── Danger Zone ──
	st.markdown('<p style="font-size:0.68rem;font-weight:700;color:#ef4444;text-transform:uppercase;letter-spacing:0.1em;margin-bottom:8px;">Danger Zone</p>', unsafe_allow_html=True)
	if st.button("🗑 Clear all data", width='stretch'):
	for s in st.session_state.streams:
	r.delete(s["redis_key"])
	st.session_state.pinned_messages = []
	st.session_state.alert_dismissed = False
	st.success("All stream data cleared.")
	st.divider()
	st.markdown(
	'<div style="font-size:0.72rem;color:var(--text-3);text-align:center;line-height:1.6;">'
	'Theme follows Streamlit settings<br>'
	'<span style="font-size:0.65rem;">☰ → Settings → Theme</span>'
	'</div>', unsafe_allow_html=True
	)


	# ── PAGE HEADER ───────────────────────────────────────────────
	_video_title = r.get("video_title")

	# Build subtitle showing ALL active stream titles
	_all_titles = []
	for _si, _ss in enumerate(st.session_state.streams):
	_st = _ss.get("video_title") or _ss.get("video_id")
	_sk = _ss.get("redis_key", "")
	_sp = _ss.get("proc")
	_sr = _sp is not None and _sp.poll() is None
	if _st and (r.llen(_sk) > 0 or _sr):
	_all_titles.append(f"▶ {_st}")
	if _all_titles:
	_subtitle = " · ".join(_all_titles)
	else:
	_subtitle = "Real-time sentiment · topic classification · engagement insights"

	# Build active stream pills for header
	_active_stream_pills = ""
	for _hi, _hs in enumerate(st.session_state.streams):
	_hkey = _hs.get("redis_key", "")
	_hproc = _hs.get("proc")
	_hrunning = _hproc is not None and _hproc.poll() is None
	if r.llen(_hkey) > 0 or _hrunning:
	_hcolor = STREAM_COLORS[_hi]
	_hlabel = STREAM_NAMES[_hi]
	_htitle = (
	_hs.get("video_title")
	or _hs.get("video_id")
	or _hkey
	or f"Stream {_hlabel}"
	)
	_hdot = f'<span style="display:inline-block;width:7px;height:7px;background:{"#22c55e" if _hrunning else "#ef4444"};border-radius:50%;margin-right:5px;vertical-align:middle;"></span>'
	_active_stream_pills += (
	f'<span style="display:inline-flex;align-items:center;background:{_hcolor}18;'
	f'border:1px solid {_hcolor}44;border-radius:20px;padding:3px 12px;'
	f'font-size:0.75rem;font-weight:700;color:{_hcolor};margin-right:8px;">'
	f'{_hdot}Stream {_hlabel} · {str(_htitle)[:22]}</span>'
	)

	col_title, col_live = st.columns([7, 1])
	with col_title:
	st.markdown(
	'<div style="padding:8px 0 4px;">'
	'<div style="font-size:2rem;font-weight:800;color:var(--text-1);letter-spacing:-0.04em;">YouTube Live Chat Analytics</div>'
	f'<div style="font-size:1.25rem;color:var(--accent-text);font-weight:600;margin-top:6px;">{_subtitle}</div>'
	+ (f'<div style="margin-top:10px;">{_active_stream_pills}</div>' if _active_stream_pills else '') +
	'</div>', unsafe_allow_html=True
	)
	with col_live:
	st.markdown(
	'<div style="text-align:right;padding-top:22px;">'
	'<span class="live-dot"></span>'
	'<span style="font-size:0.78rem;color:var(--live);font-weight:700;letter-spacing:0.05em;">LIVE</span>'
	'</div>', unsafe_allow_html=True
	)

	st.divider()

	# ── PRIMARY STREAM SELECTOR ───────────────────────────────────
	_streams_with_data = [
	s for s in st.session_state.streams
	if r.llen(s.get("redis_key", "")) > 0 or (s.get("proc") is not None and s.get("proc").poll() is None)
	]
	if len(_streams_with_data) > 1:
	_ps_options = {}
	for _pss in _streams_with_data:
	_psi_real = st.session_state.streams.index(_pss)
	_pst = _pss.get("video_title") or _pss.get("video_id") or _pss.get("redis_key")
	_psl = f"Stream {STREAM_NAMES[_psi_real]} — {str(_pst)[:35]}"
	_ps_options[_psl] = _pss["redis_key"]
	_ps_col, _ = st.columns([2, 3])
	with _ps_col:
	_selected_primary_label = st.selectbox(
	"📊 Dashboard data source",
	list(_ps_options.keys()),
	key="primary_stream_select",
	help="Switch which stream's data powers the main dashboard stats and charts"
	)
	_primary_key = _ps_options[_selected_primary_label]
	else:
	_primary_key = st.session_state.streams[0]["redis_key"]

	# ── DATA LOAD ─────────────────────────────────────────────────
	_current_len = r.llen(_primary_key)
	# Cap cumulative load at 50k — enough for accurate stats, avoids 100k+ slowdowns
	_CUMULATIVE_CAP = 50_000
	all_data = load_stream_data(_primary_key, limit=_CUMULATIVE_CAP if _current_len > _CUMULATIVE_CAP else None)
	data = all_data[-msg_limit:] if len(all_data) > msg_limit else all_data

	if not all_data:
	st.markdown(
	'<div class="empty-state">'
	'<div class="empty-icon">📭</div>'
	'<div class="empty-title">No messages yet</div>'
	'<div class="empty-sub">Set a video ID in the sidebar, then click ▶ Start</div>'
	'</div>', unsafe_allow_html=True
	)
	if auto_refresh:
	time.sleep(refresh_rate)
	st.rerun()
	st.stop()

	df = pd.DataFrame(data)
	all_df = pd.DataFrame(all_data)

	df["sentiment"] = df["sentiment"].apply(clean_sentiment)
	df["topic"] = df["topic"].apply(clean_topic) if "topic" in df.columns else "General"
	all_df["sentiment"] = all_df["sentiment"].apply(clean_sentiment)
	all_df["topic"] = all_df["topic"].apply(clean_topic) if "topic" in all_df.columns else "General"

	# ── VIEW ROUTING ──────────────────────────────────────────────
	# Read directly from session state to get the current widget value
	_active_view = st.session_state.get("active_view", "💬 Comments")
	_show_stats = _active_view == "📊 Stats & Info"
	_show_comments = _active_view == "💬 Comments"

	if _show_comments:
	st.markdown('<div class="sec-hdr"><span class="sec-ttl">Live Chat Feed</span></div>', unsafe_allow_html=True)

	# ── PINNED MESSAGES (shown above the feed) ────────────────
	if st.session_state.pinned_messages:
	st.markdown(
	'<div class="sec-hdr"><span class="sec-ttl">📌 Pinned Messages</span>'
	f'<span class="sec-pill">{len(st.session_state.pinned_messages)} pinned</span></div>',
	unsafe_allow_html=True
	)
	for _pidx, _pmsg in enumerate(st.session_state.pinned_messages):
	_ps = _pmsg.get("sentiment", "Neutral")
	_ps_color = SENT_COLORS.get(_ps, "#6b7280")
	_pt_color = TOPIC_COLOR.get(_pmsg.get("topic", "General"), "#6b7280")
	_pcol1, _pcol2 = st.columns([10, 1])
	with _pcol1:
	st.markdown(
	f'<div class="chat-card chat-pinned">'
	f'<div class="chat-author">📌 {_pmsg.get("author", "Unknown")}</div>'
	f'<div class="chat-text">{_pmsg.get("text", "")}</div>'
	f'<div class="chat-badges">'
	f'<span class="badge pin-badge">Pinned</span>'
	f'<span class="badge" style="color:{_ps_color};">{_ps}</span>'
	f'<span class="badge" style="color:{_pt_color};">{_pmsg.get("topic","General")}</span>'
	f'<span class="badge">{_pmsg.get("time","")[:19]}</span>'
	f'</div></div>',
	unsafe_allow_html=True
	)
	with _pcol2:
	if st.button("\u2715", key=f"unpin_top_{_pidx}", width='stretch'):
	st.session_state.pinned_messages.pop(_pidx)
	st.rerun()
	st.divider()

	# Build stream options
	_feed_stream_options = {}
	for _fs in st.session_state.streams:
	_fkey = _fs.get("redis_key", "")
	_flen = r.llen(_fkey)
	if _flen > 0:
	_fidx = st.session_state.streams.index(_fs)
	_flabel = f"Stream {STREAM_NAMES[_fidx]} — {_fs.get('video_id', _fkey)[:20]}"
	_feed_stream_options[_flabel] = _fkey

	_cf0, _cf1, _cf2, _cf3, _cf4 = st.columns([1, 1, 1, 1, 2])
	with _cf0:
	if len(_feed_stream_options) > 1:
	_selected_stream_label = st.selectbox(
	"Stream", list(_feed_stream_options.keys()), key="feed_stream_select"
	)
	_feed_key = _feed_stream_options[_selected_stream_label]
	else:
	_feed_key = st.session_state.streams[0]["redis_key"]
	if _feed_stream_options:
	st.markdown(
	f'<div style="font-size:0.75rem;color:var(--text-2);padding-top:28px;">'
	f'{list(_feed_stream_options.keys())[0]}</div>',
	unsafe_allow_html=True
	)

	if _feed_key == st.session_state.streams[0]["redis_key"]:
	_feed_df = df.copy()
	else:
	_feed_raw = load_stream_data(_feed_key, limit=msg_limit)
	_feed_df = pd.DataFrame(_feed_raw) if _feed_raw else pd.DataFrame()
	if not _feed_df.empty:
	_feed_df["sentiment"] = _feed_df["sentiment"].apply(clean_sentiment)
	_feed_df["topic"] = _feed_df["topic"].apply(clean_topic) if "topic" in _feed_df.columns else "General"

	with _cf1:
	_sentiment_filter = st.selectbox("Sentiment", ["All", "Positive", "Neutral", "Negative"])
	with _cf2:
	_topic_filter = st.selectbox("Topic", ["All"] + TOPIC_LABELS)
	with _cf3:
	_all_action_types = [
	"General Appreciation", "Testimonials", "Faculty Request", "Faculty Feedback",
	"Content requests", "Content Feedback", "Academic / Lecture / Concept Doubts",
	"Academic requests", "Study Materials, Deliverables & Learning Resources",
	"Access & Support", "Batch details / structure / offerings (incl faculty)",
	"Schedule & logistics (Batch)", "Information- Exam", "Information- Post Exam",
	"Eligibility & audience fit - Can I take this?", "Suitability & Sufficiency (Is this enough?)",
	"Guidance- What should I take/do?", "Language Request", "Language medium",
	"Pricing, discounts, scholarships, offer validity", "Fees + Financial Queries",
	"Product/feature requests (non-content)", "Offline expansion & event-city requests",
	"Offers + Events", "General Feedback", "Others", "N/A",
	]
	_action_type_filter = st.selectbox("Action Type", ["All"] + _all_action_types)
	with _cf4:
	_search_term = st.text_input("Search messages", placeholder="Filter by keyword...")

	_filtered = _feed_df.copy() if not _feed_df.empty else pd.DataFrame()
	_any_filter = (
	_sentiment_filter != "All"
	or _topic_filter != "All"
	or _action_type_filter != "All"
	or bool(_search_term)
	)

	if _any_filter:
	_full_raw = load_stream_data(_feed_key)
	if _full_raw:
	_full_df = pd.DataFrame(_full_raw)
	_full_df["sentiment"] = _full_df["sentiment"].apply(clean_sentiment)
	_full_df["topic"] = _full_df["topic"].apply(clean_topic) if "topic" in _full_df.columns else "General"
	_filtered = _full_df.copy()
	if _sentiment_filter != "All":
	_filtered = _filtered[_filtered["sentiment"] == _sentiment_filter]
	if _topic_filter != "All":
	_filtered = _filtered[_filtered["topic"] == _topic_filter]
	if _action_type_filter != "All":
	if "action_type" in _filtered.columns:
	_filtered = _filtered[_filtered["action_type"] == _action_type_filter]
	if _search_term:
	_filtered = _filtered[_filtered["text"].str.contains(_search_term, case=False, na=False)]
	if len(_filtered) > msg_limit:
	_filtered = _filtered.iloc[-msg_limit:]
	else:
	_filtered = pd.DataFrame()
	_total_scanned = len(_full_raw) if _full_raw else 0
	else:
	if not _filtered.empty:
	if _sentiment_filter != "All":
	_filtered = _filtered[_filtered["sentiment"] == _sentiment_filter]
	if _topic_filter != "All":
	_filtered = _filtered[_filtered["topic"] == _topic_filter]
	if _action_type_filter != "All":
	if "action_type" in _filtered.columns:
	_filtered = _filtered[_filtered["action_type"] == _action_type_filter]
	if _search_term:
	_filtered = _filtered[_filtered["text"].str.contains(_search_term, case=False, na=False)]
	_total_scanned = len(_feed_df)

	_feed_hdr, _feed_dl = st.columns([3, 1])
	with _feed_hdr:
	if _any_filter:
	st.markdown(
	f'<div style="font-size:0.78rem;color:var(--text-3);margin-bottom:12px;">'
	f'Showing {len(_filtered)} matching messages (scanned all {_total_scanned}, capped at {msg_limit})</div>',
	unsafe_allow_html=True
	)
	else:
	st.markdown(
	f'<div style="font-size:0.78rem;color:var(--text-3);margin-bottom:12px;">'
	f'Showing {len(_filtered)} of {len(_feed_df)} messages</div>',
	unsafe_allow_html=True
	)
	with _feed_dl:
	if not _filtered.empty:
	_export_cols = [c for c in ["author", "text", "sentiment", "confidence", "topic", "time"] if c in _filtered.columns]
	csv_download(_filtered[_export_cols], "Download Feed CSV", "chat_feed.csv")

	_SENT_ICON = {"Positive": "🟢", "Negative": "🔴", "Neutral": "🟡"}
	_pinned_texts = {m.get("text", "") for m in st.session_state.pinned_messages}

	for _i, (_, _row) in enumerate(_filtered.iloc[::-1].iterrows()):
	_s = _row.get("sentiment", "Neutral")
	_conf_pct = int(_row.get("confidence", 0) * 100)
	_topic = clean_topic(_row.get("topic", "General"))
	_t_color = TOPIC_COLOR.get(_topic, "#6b7280")
	_s_color = SENT_COLORS.get(_s, "#6b7280")
	_s_icon = _SENT_ICON.get(_s, "⚪")
	_conf_color = "#22c55e" if _conf_pct >= 70 else "#eab308" if _conf_pct >= 40 else "#ef4444"
	_msg_text = _row.get("text", "")
	import re as _re2
	_display_text = _re2.sub(r":[a-zA-Z0-9_\-]+:", "", _msg_text).strip() or _msg_text
	_is_pinned = _msg_text in _pinned_texts
	_action_type = _row.get("action_type", "N/A") or "N/A"
	_card_class = f"chat-card chat-{_s.lower()}" + (" chat-pinned" if _is_pinned else "")

	_msg_col, _pin_col = st.columns([11, 1])
	with _msg_col:
	_ab = (
	f'<span class="badge" style="color:#a78bfa;border-color:#a78bfa33;">🏷 {_action_type}</span>'
	if _action_type not in ("N/A", "", None) else ""
	)
	st.markdown(
	f'<div class="{_card_class}">'
	f'<div class="chat-author">{_s_icon} {_row.get("author", "Unknown")}'
	+ (' <span style="font-size:0.7rem;color:#eab308;">📌</span>' if _is_pinned else '') +
	f'</div>'
	f'<div class="chat-text">{_display_text}</div>'
	f'<div class="chat-badges">'
	f'<span class="badge" style="color:{_s_color};border-color:{_s_color}33;">{_s}</span>'
	f'<span class="badge" style="color:{_conf_color};">Confidence: {_conf_pct}%</span>'
	f'<span class="badge" style="color:{_t_color};border-color:{_t_color}33;">{_topic}</span>'
	f'{_ab}'
	f'</div></div>',
	unsafe_allow_html=True
	)
	with _pin_col:
	if _is_pinned:
	if st.button("📌", key=f"unpin_feed_{_i}", help="Unpin this message"):
	st.session_state.pinned_messages = [
	m for m in st.session_state.pinned_messages if m.get("text") != _msg_text
	]
	st.rerun()
	else:
	if st.button("📍", key=f"pin_{_i}", help="Pin this message"):
	_msg_dict = _row.to_dict()
	if _msg_dict not in st.session_state.pinned_messages:
	st.session_state.pinned_messages.append(_msg_dict)
	st.rerun()

	if auto_refresh:
	time.sleep(refresh_rate)
	st.rerun()
	st.stop()

	# ── ALERT BANNERS (Stats view only) ───────────────────────────
	if alert_enabled:
	alert = check_alert(all_df, threshold=alert_threshold, window=alert_window)
	total_now = len(all_df)
	if total_now != st.session_state.last_alert_count:
	st.session_state.last_alert_count = total_now
	if alert:
	st.session_state.alert_dismissed = False

	if alert and not st.session_state.alert_dismissed:
	a1, a2 = st.columns([8, 1])
	with a1:
	st.markdown(
	f'<div class="alert-banner">'
	f'<span class="alert-icon">🚨</span>'
	f'<div>'
	f'<div class="alert-text">Negative sentiment spike — {alert["neg_ratio"]*100:.0f}% negative in last {alert["window"]} messages</div>'
	f'<div class="alert-sub">{alert["count"]} of {alert["window"]} messages are negative. Consider moderating.</div>'
	f'</div></div>',
	unsafe_allow_html=True
	)
	with a2:
	if st.button("✕ Dismiss", key="dismiss_alert"):
	st.session_state.alert_dismissed = True
	st.rerun()

	if spam_alert_on:
	spam_alert = check_spam_alert(all_df, threshold=spam_threshold, window=alert_window)
	if spam_alert and not st.session_state.get("spam_dismissed", False):
	s1, s2 = st.columns([8, 1])
	with s1:
	st.markdown(
	f'<div class="spam-alert">'
	f'<span class="alert-icon">🛡️</span>'
	f'<div>'
	f'<div class="spam-alert-text">Spam surge detected — {spam_alert["spam_ratio"]*100:.0f}% spam in last {spam_alert["window"]} messages</div>'
	f'<div class="spam-alert-sub">{spam_alert["count"]} spam messages detected. Chat may be under flood attack.</div>'
	f'</div></div>',
	unsafe_allow_html=True
	)
	with s2:
	if st.button("✕", key="dismiss_spam"):
	st.session_state.spam_dismissed = True
	st.rerun()
	elif not spam_alert:
	st.session_state.spam_dismissed = False

	# ── CUMULATIVE STATS ──────────────────────────────────────────
	all_counts = all_df["sentiment"].value_counts().to_dict()
	c_pos = all_counts.get("Positive", 0)
	c_neu = all_counts.get("Neutral", 0)
	c_neg = all_counts.get("Negative", 0)
	c_total = max(c_pos + c_neu + c_neg, 1)

	# Sentiment velocity
	velocity = compute_velocity(json.dumps([{"sentiment": m.get("sentiment","Neutral")} for m in all_data]))

	st.markdown(
	'<div class="sec-hdr"><span class="sec-ttl">Cumulative Sentiment</span><span class="sec-pill">All Time</span></div>',
	unsafe_allow_html=True
	)

	v1, v2, v3, v4, v5 = st.columns([1, 1, 1, 1, 1])
	with v1:
	st.markdown(
	f'<div class="stat-card"><div class="stat-accent" style="background:linear-gradient(90deg,#22c55e,#16a34a);"></div>'
	f'<div class="stat-number" style="color:#22c55e;">{c_pos}</div><div class="stat-label">Positive</div><div class="stat-sub">{c_pos/c_total*100:.1f}% of total</div></div>',
	unsafe_allow_html=True
	)
	with v2:
	st.markdown(
	f'<div class="stat-card"><div class="stat-accent" style="background:linear-gradient(90deg,#eab308,#ca8a04);"></div>'
	f'<div class="stat-number" style="color:#eab308;">{c_neu}</div><div class="stat-label">Neutral</div><div class="stat-sub">{c_neu/c_total*100:.1f}% of total</div></div>',
	unsafe_allow_html=True
	)
	with v3:
	st.markdown(
	f'<div class="stat-card"><div class="stat-accent" style="background:linear-gradient(90deg,#ef4444,#dc2626);"></div>'
	f'<div class="stat-number" style="color:#ef4444;">{c_neg}</div><div class="stat-label">Negative</div><div class="stat-sub">{c_neg/c_total*100:.1f}% of total</div></div>',
	unsafe_allow_html=True
	)
	with v4:
	st.markdown(
	f'<div class="stat-card"><div class="stat-accent" style="background:linear-gradient(90deg,#7c3aed,#4f46e5);"></div>'
	f'<div class="stat-number" style="color:var(--accent-text);">{c_total}</div><div class="stat-label">Total</div><div class="stat-sub">all time</div></div>',
	unsafe_allow_html=True
	)
	with v5:
	# Sentiment velocity card
	vc = velocity["color"]
	st.markdown(
	f'<div class="velocity-card" style="border-color:{vc}44;">'
	f'<div class="velocity-arrow" style="color:{vc};">{velocity["direction"]}</div>'
	f'<div>'
	f'<div class="velocity-val" style="color:{vc};">{velocity["label"]}</div>'
	f'<div class="velocity-label">Sentiment Velocity<br>'
	f'<span style="color:{vc};">{velocity["delta"]:+.0%} pos shift</span></div>'
	f'</div></div>',
	unsafe_allow_html=True
	)

	# ── CUMULATIVE TOPIC ──────────────────────────────────────────
	st.divider()
	st.markdown(
	'<div class="sec-hdr"><span class="sec-ttl">Cumulative Topic</span><span class="sec-pill">All Time</span></div>',
	unsafe_allow_html=True
	)
	_topic_colors_list = ["#f59e0b", "#3b82f6", "#ec4899", "#ef4444", "#6b7280", "#10b981"]
	_ct_cols = st.columns(len(TOPIC_LABELS))
	for _ci, (_lbl, _clr) in enumerate(zip(TOPIC_LABELS, _topic_colors_list)):
	_cnt = int((all_df["topic"] == _lbl).sum()) if "topic" in all_df.columns else 0
	_pct = _cnt / max(c_total, 1) * 100
	with _ct_cols[_ci]:
	st.markdown(
	f'<div class="stat-card"><div class="stat-accent" style="background:{_clr};"></div>'
	f'<div class="stat-number" style="color:{_clr};font-size:1.8rem;">{_cnt}</div>'
	f'<div class="stat-label">{_lbl}</div>'
	f'<div class="stat-sub">{_pct:.1f}% of msgs</div></div>',
	unsafe_allow_html=True
	)


	# ── ENGAGEMENT SCORE (moved here — after topic, before window) ──
	_eng_json = json.dumps([{"sentiment": m.get("sentiment","Neutral"), "topic": m.get("topic","General"), "time": m.get("time","")} for m in all_data])
	eng = compute_engagement(_eng_json)

	st.divider()
	st.markdown(
	'<div class="sec-hdr"><span class="sec-ttl">Engagement Score</span><span class="sec-pill">Live</span></div>',
	unsafe_allow_html=True
	)
	ec1, ec2, ec3, ec4 = st.columns([2, 1, 1, 1])
	with ec1:
	score_color = "#22c55e" if eng["score"] >= 70 else "#eab308" if eng["score"] >= 40 else "#ef4444"
	bar_w = eng["score"]
	st.markdown(
	f'<div class="engage-card" style="border-color:{score_color}44;">'
	f'<div class="engage-score" style="color:{score_color};">{eng["score"]}</div>'
	f'<div class="engage-label">Engagement Score / 100 \u2014 {eng["grade"]}</div>'
	f'<div class="engage-bar-bg"><div class="engage-bar-fill" style="width:{bar_w}%;background:{score_color};"></div></div>'
	f'<div class="engage-breakdown">'
	f'<div class="engage-item">Msg rate <span>{eng["rate"]}/min</span></div>'
	f'<div class="engage-item">Positive <span>{eng["pos_ratio"]*100:.0f}%</span></div>'
	f'<div class="engage-item">Questions <span>{eng["q_density"]*100:.0f}%</span></div>'
	f'</div></div>',
	unsafe_allow_html=True
	)
	with ec2:
	st.metric("Msgs/min", f"{eng['rate']:.1f}")
	with ec3:
	st.metric("Positive ratio", f"{eng['pos_ratio']*100:.0f}%")
	with ec4:
	st.metric("Question density", f"{eng['q_density']*100:.0f}%")

	# ── WINDOW METRICS ────────────────────────────────────────────
	st.divider()
	counts = df["sentiment"].value_counts().to_dict()
	pos = counts.get("Positive", 0)
	neu = counts.get("Neutral", 0)
	neg = counts.get("Negative", 0)
	total = max(pos + neu + neg, 1)

	st.markdown(
	f'<div class="sec-hdr"><span class="sec-ttl">Window Snapshot</span><span class="sec-pill">Last {msg_limit} msgs</span></div>',
	unsafe_allow_html=True
	)
	c1, c2, c3, c4 = st.columns(4)
	c1.metric("Messages", total)
	c2.metric("Positive", pos, f"{pos/total*100:.1f}%")
	c3.metric("Neutral", neu, f"{neu/total*100:.1f}%")
	c4.metric("Negative", neg, f"{neg/total*100:.1f}%")

	# ── SENTIMENT + TOPIC CHARTS (ALL TIME) ──────────────────────
	st.divider()
	col_s1, col_s2, col_t1, col_t2 = st.columns(4)

	with col_s1:
	st.markdown('<div class="chart-wrap">', unsafe_allow_html=True)
	st.markdown('<div class="chart-title">Sentiment Distribution</div><div class="chart-sub">All-time message count by sentiment class</div>', unsafe_allow_html=True)
	fig_bar = go.Figure(go.Bar(
	x=["Positive", "Neutral", "Negative"],
	y=[c_pos, c_neu, c_neg],
	marker_color=["#22c55e", "#eab308", "#ef4444"],
	marker_line_width=0,
	text=[c_pos, c_neu, c_neg],
	textposition="outside",
	textfont=dict(size=12),
	hovertemplate="<b>%{x}</b><br>Count: %{y}<extra></extra>",
	))
	fig_bar.update_layout(**plotly_layout(260))
	st.plotly_chart(fig_bar, width='stretch', config={"displayModeBar": False})
	bar_hdr, bar_dl = st.columns([1, 1])
	with bar_hdr:
	show_bar_data = st.checkbox("View data", key="show_bar")
	with bar_dl:
	bar_df = pd.DataFrame({"Sentiment": ["Positive", "Neutral", "Negative"], "Count": [c_pos, c_neu, c_neg]})
	csv_download(bar_df, "Download CSV", "sentiment_distribution.csv")
	if show_bar_data:
	st.dataframe(bar_df, width='stretch', hide_index=True)
	st.markdown('</div>', unsafe_allow_html=True)

	with col_s2:
	st.markdown('<div class="chart-wrap">', unsafe_allow_html=True)
	st.markdown('<div class="chart-title">Sentiment Donut</div><div class="chart-sub">All-time proportional share per class</div>', unsafe_allow_html=True)
	fig_pie = go.Figure(go.Pie(
	labels=["Positive", "Neutral", "Negative"],
	values=[c_pos, c_neu, c_neg],
	marker_colors=["#22c55e", "#eab308", "#ef4444"],
	hole=0.58,
	textinfo="percent",
	hovertemplate="<b>%{label}</b><br>%{value} messages (%{percent})<extra></extra>",
	))
	fig_pie.update_layout(
	{plotly_layout(260),
	"showlegend": True,
	"legend": dict(orientation="h", y=-0.08, font=dict(size=11, color="#f1f5f9"))}
	)
	st.plotly_chart(fig_pie, width='stretch', config={"displayModeBar": False})
	pie_hdr, pie_dl = st.columns([1, 1])
	with pie_hdr:
	show_pie_data = st.checkbox("View data", key="show_pie")
	with pie_dl:
	pie_df = pd.DataFrame({
	"Sentiment": ["Positive", "Neutral", "Negative"],
	"Count": [c_pos, c_neu, c_neg],
	"Percentage": [f"{c_pos/c_total100:.1f}%", f"{c_neu/c_total100:.1f}%", f"{c_neg/c_total*100:.1f}%"]
	})
	csv_download(pie_df, "Download CSV", "sentiment_breakdown.csv")
	if show_pie_data:
	st.dataframe(pie_df, width='stretch', hide_index=True)
	st.markdown('</div>', unsafe_allow_html=True)

	with col_t1:
	st.markdown('<div class="chart-wrap">', unsafe_allow_html=True)
	st.markdown('<div class="chart-title">Topic Distribution</div><div class="chart-sub">Message count by topic class</div>', unsafe_allow_html=True)
	_tc_vals = [int((all_df["topic"] == l).sum()) if "topic" in all_df.columns else 0 for l in TOPIC_LABELS]
	_tc_colors = ["#f59e0b", "#3b82f6", "#ec4899", "#ef4444", "#6b7280", "#10b981"]
	fig_tbar = go.Figure(go.Bar(
	x=TOPIC_LABELS,
	y=_tc_vals,
	marker_color=_tc_colors,
	marker_line_width=0,
	text=_tc_vals,
	textposition="outside",
	textfont=dict(size=11),
	hovertemplate="<b>%{x}</b><br>Count: %{y}<extra></extra>",
	))
	_tbar_layout = plotly_layout(260)
	_tbar_layout["xaxis"]["tickfont"] = dict(size=9)
	fig_tbar.update_layout(**_tbar_layout)
	st.plotly_chart(fig_tbar, width='stretch', config={"displayModeBar": False})
	st.markdown('</div>', unsafe_allow_html=True)

	with col_t2:
	st.markdown('<div class="chart-wrap">', unsafe_allow_html=True)
	st.markdown('<div class="chart-title">Topic Donut</div><div class="chart-sub">Proportional share per topic</div>', unsafe_allow_html=True)
	fig_tpie = go.Figure(go.Pie(
	labels=TOPIC_LABELS,
	values=_tc_vals,
	marker_colors=_tc_colors,
	hole=0.58,
	textinfo="percent",
	hovertemplate="<b>%{label}</b><br>%{value} messages (%{percent})<extra></extra>",
	))
	fig_tpie.update_layout(
	{plotly_layout(260),
	"showlegend": True,
	"legend": dict(orientation="h", y=-0.08, font=dict(size=10, color="#f1f5f9"))}
	)
	st.plotly_chart(fig_tpie, width='stretch', config={"displayModeBar": False})
	st.markdown('</div>', unsafe_allow_html=True)

	# ── SENTIMENT HEATMAP OVER TIME ───────────────────────────────
	st.divider()
	st.markdown(
	'<div class="sec-hdr"><span class="sec-ttl">Sentiment Heatmap</span><span class="sec-pill">Over Time</span></div>',
	unsafe_allow_html=True
	)
	heatmap_data = build_heatmap_data(json.dumps([{"time": m.get("time",""), "sentiment": m.get("sentiment","Neutral")} for m in all_data]), bucket_minutes=1)

	if not heatmap_data.empty:
	st.markdown('<div class="chart-wrap">', unsafe_allow_html=True)
	st.markdown('<div class="chart-title">Sentiment Over Time</div><div class="chart-sub">Message volume per sentiment per minute bucket</div>', unsafe_allow_html=True)

	fig_heat = go.Figure()
	for sent, color in [("Positive", "#22c55e"), ("Neutral", "#eab308"), ("Negative", "#ef4444")]:
	fig_heat.add_trace(go.Scatter(
	x=heatmap_data["bucket"],
	y=heatmap_data[sent],
	name=sent,
	mode="lines+markers",
	line=dict(color=color, width=2),
	marker=dict(size=4),
	hovertemplate=f"<b>{sent}</b><br>%{{x}}<br>Count: %{{y}}<extra></extra>",
	))

	layout = plotly_layout(220)
	layout["showlegend"] = True
	layout["legend"] = dict(orientation="h", y=1.08, font=dict(size=11))
	layout["xaxis"]["tickformat"] = "%H:%M"
	fig_heat.update_layout(**layout)
	st.plotly_chart(fig_heat, width='stretch', config={"displayModeBar": False})

	heat_hdr, heat_dl = st.columns([1, 1])
	with heat_hdr:
	show_heat_data = st.checkbox("View data", key="show_heat")
	with heat_dl:
	csv_download(heatmap_data.rename(columns={"bucket": "time_bucket"}), "Download CSV", "sentiment_heatmap.csv")
	if show_heat_data:
	st.dataframe(heatmap_data.rename(columns={"bucket": "time_bucket"}), width='stretch', hide_index=True)
	st.markdown('</div>', unsafe_allow_html=True)
	else:
	st.info("Not enough timestamped data for heatmap yet.")

	# ── TOPIC DISTRIBUTION ────────────────────────────────────────
	st.divider()
	st.markdown(
	'<div class="sec-hdr"><span class="sec-ttl">Topic Distribution</span><span class="sec-pill">All Time</span></div>',
	unsafe_allow_html=True
	)

	topic_counts = {
	label: int((all_df["topic"] == label).sum())
	for label in TOPIC_LABELS
	}

	pills = '<div class="topic-grid">'
	for label in TOPIC_LABELS:
	color = TOPIC_COLOR[label]
	count = topic_counts[label]
	pills += (
	f'<div class="topic-pill" style="border:1px solid {color}44;">'
	f'<div class="topic-count" style="color:{color};">{count}</div>'
	f'<div class="topic-name">{label}</div>'
	f'</div>'
	)
	pills += '</div>'
	st.markdown(pills, unsafe_allow_html=True)

	st.markdown('<div class="chart-wrap">', unsafe_allow_html=True)
	st.markdown('<div class="chart-title">Topic Breakdown</div><div class="chart-sub">All-time message count per topic category</div>', unsafe_allow_html=True)
	fig_topic = go.Figure(go.Bar(
	x=TOPIC_LABELS,
	y=[topic_counts[l] for l in TOPIC_LABELS],
	marker_color=[TOPIC_COLOR[l] for l in TOPIC_LABELS],
	marker_line_width=0,
	text=[topic_counts[l] for l in TOPIC_LABELS],
	textposition="outside",
	textfont=dict(size=11),
	hovertemplate="<b>%{x}</b><br>Count: %{y}<extra></extra>",
	))
	fig_topic.update_layout(**plotly_layout(250))
	st.plotly_chart(fig_topic, width='stretch', config={"displayModeBar": False})
	topic_hdr, topic_dl = st.columns([1, 1])
	with topic_hdr:
	show_topic_data = st.checkbox("View data", key="show_topic")
	with topic_dl:
	topic_df = pd.DataFrame({"Topic": TOPIC_LABELS, "Count": [topic_counts[l] for l in TOPIC_LABELS]})
	csv_download(topic_df, "Download CSV", "topic_distribution.csv")
	if show_topic_data:
	st.dataframe(topic_df, width='stretch', hide_index=True)
	st.markdown('</div>', unsafe_allow_html=True)

	# ── Topic Sentiment breakdown ──────────────────────────────────
	st.markdown('<div class="chart-wrap" style="margin-top:16px;">', unsafe_allow_html=True)
	st.markdown('<div class="chart-title">Sentiment by Topic</div><div class="chart-sub">% positive / neutral / negative within each topic category</div>', unsafe_allow_html=True)
	_topic_sent_data = []
	for _lbl in TOPIC_LABELS:
	_mask = all_df["topic"] == _lbl
	_total = int(_mask.sum())
	if _total == 0:
	_topic_sent_data.append({"topic": _lbl, "pos": 0, "neu": 0, "neg": 0})
	continue
	_sub = all_df[_mask]
	_topic_sent_data.append({
	"topic": _lbl,
	"pos": round((_sub["sentiment"] == "Positive").sum() / _total * 100),
	"neu": round((_sub["sentiment"] == "Neutral").sum() / _total * 100),
	"neg": round((_sub["sentiment"] == "Negative").sum() / _total * 100),
	})
	fig_ts = go.Figure()
	for _sk, _sl, _sc in [("neg", "Neg", "#ef4444"), ("neu", "Neu", "#eab308"), ("pos", "Pos", "#22c55e")]:
	fig_ts.add_trace(go.Bar(
	y=[d["topic"] for d in _topic_sent_data],
	x=[d[_sk] for d in _topic_sent_data],
	name=_sl,
	orientation="h",
	marker_color=_sc,
	hovertemplate="<b>%{y}</b><br>" + _sl + ": %{x}%<extra></extra>",
	))
	_layout_ts = plotly_layout(260)
	_layout_ts["barmode"] = "stack"
	_layout_ts["showlegend"] = True
	_layout_ts["legend"] = dict(orientation="h", y=1.08, x=0.35, font=dict(size=11))
	_layout_ts["xaxis"]["range"] = [0, 100]
	_layout_ts["xaxis"]["ticksuffix"] = "%"
	_layout_ts["yaxis"]["autorange"] = "reversed"
	fig_ts.update_layout(**_layout_ts)
	st.plotly_chart(fig_ts, width='stretch', config={"displayModeBar": False})
	st.markdown('</div>', unsafe_allow_html=True)


	# ── ACTION TYPE CHARTS ────────────────────────────────────────
	st.divider()
	st.markdown(
	'<div class="sec-hdr"><span class="sec-ttl">Action Type Analysis</span><span class="sec-pill">Last 100 msgs</span></div>',
	unsafe_allow_html=True
	)

	# Category groupings
	_QUESTION_ACTIONS = [
	"Access & Support",
	"Academic / Lecture / Concept Doubts",
	"Study Materials, Deliverables & Learning Resources",
	"Batch details / structure / offerings (incl faculty)",
	"Schedule & logistics (Batch)",
	"Guidance- What should I take/do?",
	"Suitability & Sufficiency (Is this enough?)",
	"Eligibility & audience fit - Can I take this?",
	"Information- Exam",
	"Information- Post Exam",
	]
	_REQUEST_ACTIONS = [
	"Content requests",
	"Content Feedback",
	"Faculty Request",
	"Faculty Feedback",
	"Academic requests",
	"Language Request",
	"Language medium",
	"Product/feature requests (non-content)",
	"Offline expansion & event-city requests",
	"General Feedback",
	"Others",
	]
	_SHORT_ACTION = {
	"Access & Support": "Access & Support",
	"Academic / Lecture / Concept Doubts": "Academic Doubts",
	"Study Materials, Deliverables & Learning Resources": "Study Materials & Learning Resources",
	"Batch details / structure / offerings (incl faculty)": "Batch Details & Offerings",
	"Schedule & logistics (Batch)": "Batch Schedule & Logistics",
	"Guidance- What should I take/do?": "Guidance (What Should I Take/Do?)",
	"Suitability & Sufficiency (Is this enough?)": "Suitability & Sufficiency (Is This Enough?)",
	"Eligibility & audience fit - Can I take this?": "Eligibility (Can I Take This?)",
	"Information- Exam": "Exam Information",
	"Information- Post Exam": "Post Exam Information",
	"Content requests": "Content requests",
	"Content Feedback": "Content Feedback",
	"Faculty Request": "Faculty Request",
	"Faculty Feedback": "Faculty Feedback",
	"Academic requests": "Academic requests",
	"Language Request": "Language Request",
	"Language medium": "Language Medium",
	"Product/feature requests (non-content)": "Non Content Product Requests",
	"Offline expansion & event-city requests": "Offline Expansion & Event Requests",
	"General Feedback": "General Feedback",
	"Others": "Others",
	}

	# Compute counts from last 100 messages
	_at_counts: dict[str, int] = {}
	if "action_type" in all_df.columns:
	for _at in _QUESTION_ACTIONS + _REQUEST_ACTIONS:
	_at_counts[_at] = int((all_df.tail(100)["action_type"] == _at).sum())
	else:
	_at_counts = {_at: 0 for _at in _QUESTION_ACTIONS + _REQUEST_ACTIONS}

	_q_data = {k: _at_counts.get(k, 0) for k in _QUESTION_ACTIONS if _at_counts.get(k, 0) > 0}
	_rf_data = {k: _at_counts.get(k, 0) for k in _REQUEST_ACTIONS if _at_counts.get(k, 0) > 0}
	_q_total = sum(_q_data.values())
	_rf_total = sum(_rf_data.values())

	_at_col1, _at_col2 = st.columns(2)

	with _at_col1:
	st.markdown(
	f'<div class="chart-wrap"><div class="chart-title">Type of Questions</div>'
	f'<div class="chart-sub">({_q_total} comments)</div>',
	unsafe_allow_html=True
	)
	if _q_data:
	_q_sorted = sorted(_q_data.items(), key=lambda x: x[1], reverse=True)
	_q_labels = [_SHORT_ACTION.get(k, k) for k, _ in _q_sorted]
	_q_vals = [v for _, v in _q_sorted]
	fig_q = go.Figure(go.Bar(
	x=_q_labels, y=_q_vals,
	marker_color="#4a90d9",
	marker_line_width=0,
	text=_q_vals, textposition="outside",
	textfont=dict(size=11, color="#fff"),
	hovertemplate="<b>%{x}</b><br>Comments: %{y}<extra></extra>",
	))
	fig_q.update_layout(**plotly_layout(280))
	st.plotly_chart(fig_q, width='stretch', config={"displayModeBar": False})
	else:
	st.markdown('<div style="text-align:center;padding:40px;color:var(--text-3);">No data yet</div>', unsafe_allow_html=True)
	st.markdown('</div>', unsafe_allow_html=True)

	with _at_col2:
	st.markdown(
	f'<div class="chart-wrap"><div class="chart-title">Type of Requests & Feedback</div>'
	f'<div class="chart-sub">({_rf_total} comments)</div>',
	unsafe_allow_html=True
	)
	if _rf_data:
	_rf_sorted = sorted(_rf_data.items(), key=lambda x: x[1], reverse=True)
	_rf_labels = [_SHORT_ACTION.get(k, k) for k, _ in _rf_sorted]
	_rf_vals = [v for _, v in _rf_sorted]
	fig_rf = go.Figure(go.Bar(
	x=_rf_labels, y=_rf_vals,
	marker_color="#f5a623",
	marker_line_width=0,
	text=_rf_vals, textposition="outside",
	textfont=dict(size=11, color="#fff"),
	hovertemplate="<b>%{x}</b><br>Comments: %{y}<extra></extra>",
	))
	fig_rf.update_layout(**plotly_layout(280))
	st.plotly_chart(fig_rf, width='stretch', config={"displayModeBar": False})
	else:
	st.markdown('<div style="text-align:center;padding:40px;color:var(--text-3);">No data yet</div>', unsafe_allow_html=True)
	st.markdown('</div>', unsafe_allow_html=True)

	# Top-5 horizontal bar panels
	_top5_col1, _top5_col2 = st.columns(2)

	def _hbar_rows_html(data: dict, color: str, max_val: int) -> str:
	html = ""
	for cat, count in sorted(data.items(), key=lambda x: x[1], reverse=True)[:5]:
	pct = round(count / max(max_val, 1) * 100)
	label = _SHORT_ACTION.get(cat, cat)
	html += (
	f'<div style="display:flex;align-items:center;gap:10px;margin-bottom:10px;">'
	f'<div style="width:170px;font-size:12px;text-align:right;opacity:0.85;line-height:1.3;">{label}</div>'
	f'<div style="flex:1;height:22px;border-radius:4px;background:rgba(255,255,255,0.06);overflow:hidden;">'
	f'<div style="width:{pct}%;height:100%;background:{color};border-radius:4px;'
	f'display:flex;align-items:center;justify-content:flex-end;padding-right:6px;'
	f'font-size:11px;font-weight:700;color:#fff;">{pct}%</div>'
	f'</div></div>'
	)
	return html

	with _top5_col1:
	st.markdown(
	'<div class="chart-wrap">'
	'<div class="chart-title">Top 5 <span style="color:#60a5fa;">Questions</span> Students Ask</div>'
	'<div class="chart-sub">Type of action count for Questions across tagged videos.</div>',
	unsafe_allow_html=True
	)
	if _q_data:
	st.markdown(_hbar_rows_html(_q_data, "#f87171", max(_q_data.values(), default=1)), unsafe_allow_html=True)
	else:
	st.markdown('<div style="text-align:center;padding:20px;color:var(--text-3);">No data yet</div>', unsafe_allow_html=True)
	st.markdown('</div>', unsafe_allow_html=True)

	with _top5_col2:
	st.markdown(
	'<div class="chart-wrap">'
	'<div class="chart-title">Top 5 Types of <span style="color:#f87171;">Requests & Feedback</span> Students Give</div>'
	'<div class="chart-sub">Type of action count for Request/Feedback across tagged videos.</div>',
	unsafe_allow_html=True
	)
	if _rf_data:
	st.markdown(_hbar_rows_html(_rf_data, "#f87171", max(_rf_data.values(), default=1)), unsafe_allow_html=True)
	else:
	st.markdown('<div style="text-align:center;padding:20px;color:var(--text-3);">No data yet</div>', unsafe_allow_html=True)
	st.markdown('</div>', unsafe_allow_html=True)


	# ── TOP CONTRIBUTORS ──────────────────────────────────────────
	st.divider()
	st.markdown(
	'<div class="sec-hdr"><span class="sec-ttl">Top Contributors</span><span class="sec-pill">All Time</span></div>',
	unsafe_allow_html=True
	)

	_contrib_json = json.dumps([{"author": m.get("author",""), "sentiment": m.get("sentiment","Neutral"), "topic": m.get("topic","General")} for m in all_data])
	contributors = compute_top_contributors(_contrib_json)

	if contributors:
	max_count = contributors[0]["count"]
	rank_icons = {1: "🥇", 2: "🥈", 3: "🥉"}
	rank_classes = {1: "gold", 2: "silver", 3: "bronze"}
	for rank, c in enumerate(contributors, 1):
	bar_pct = int(c["count"] / max(max_count, 1) * 100)
	rank_cls = rank_classes.get(rank, "")
	rank_icon = rank_icons.get(rank, f"#{rank}")
	author = c["author"]
	count = c["count"]
	pos_pct = c["pos_pct"]
	neu_pct = c["neu_pct"]
	neg_pct = c["neg_pct"]
	html = (
	f'<div class="leaderboard-row">'
	f'<div class="lb-rank {rank_cls}">{rank_icon}</div>'
	f'<div class="lb-author">{author}</div>'
	f'<div class="lb-bar"><div class="lb-bar-fill" style="width:{bar_pct}%;background:var(--accent);"></div></div>'
	f'<div class="lb-sent">'
	f'<span class="lb-dot" style="background:#22c55e;" title="Positive {pos_pct}%"></span>'
	f'<span class="lb-dot" style="background:#eab308;" title="Neutral {neu_pct}%"></span>'
	f'<span class="lb-dot" style="background:#ef4444;" title="Negative {neg_pct}%"></span>'
	f'</div>'
	f'<div class="lb-count">{count} msgs</div>'
	f'</div>'
	)
	st.markdown(html, unsafe_allow_html=True)

	# ── Combined Sentiment + Topic dual-bar chart ──────────────
	st.markdown('<div class="chart-wrap" style="margin-top:16px;">', unsafe_allow_html=True)
	st.markdown(
	'<div class="chart-title">Sentiment & Topic Breakdown — Top Contributors</div>'
	'<div class="chart-sub">Top bar = sentiment (Neg/Neu/Pos) · Bottom bar = topic mix · right = message count</div>',
	unsafe_allow_html=True
	)

	# Each user occupies 2 numeric slots: sentiment at i2+0.3, topic at i2-0.3
	# Tick label sits at i*2 (midpoint) showing the name once
	n = len(contributors)
	y_sent_num = [i * 2 + 0.3 for i in range(n)]
	y_topic_num = [i * 2 - 0.3 for i in range(n)]
	tick_vals = [i * 2 for i in range(n)]
	tick_text = [c["author"][:22] for c in contributors]

	fig_combo = go.Figure()

	# ── Sentiment traces ──
	for key, label, color in [
	("neg_pct", "Neg", "#ef4444"),
	("neu_pct", "Neu", "#eab308"),
	("pos_pct", "Pos", "#22c55e"),
	]:
	fig_combo.add_trace(go.Bar(
	name=label,
	y=y_sent_num,
	x=[c[key] for c in contributors],
	orientation="h",
	marker_color=color,
	legendgroup="sent",
	legendgrouptitle_text="Sentiment" if key == "neg_pct" else None,
	width=0.5,
	hovertemplate="<b>" + label + "</b>: %{x}%<extra></extra>",
	))

	# ── Topic traces ──
	for key, label, color in [
	("t_appr", "Appreciation", "#f59e0b"),
	("t_ques", "Question", "#3b82f6"),
	("t_rf", "Request/Feedback","#8b5cf6"),
	("t_promo", "Promo", "#ec4899"),
	("t_spam", "Spam", "#ef4444"),
	("t_gen", "General", "#6b7280"),
	("t_mcq", "MCQ Answer", "#10b981"),
	]:
	fig_combo.add_trace(go.Bar(
	name=label,
	y=y_topic_num,
	x=[c[key] for c in contributors],
	orientation="h",
	marker_color=color,
	legendgroup="topic",
	legendgrouptitle_text="Topic" if key == "t_appr" else None,
	width=0.5,
	hovertemplate="<b>" + label + "</b>: %{x}%<extra></extra>",
	))

	# ── Message count annotations (right of sentiment bar) ──
	annotations = []
	for i, c in enumerate(contributors):
	annotations.append(dict(
	x=102, y=y_sent_num[i],
	text=f"<b>{c['count']} msgs</b>",
	showarrow=False,
	xanchor="left",
	font=dict(size=10, color="#94a3b8"),
	xref="x", yref="y",
	))

	chart_h = max(400, n * 56)
	layout_combo = plotly_layout(chart_h)
	layout_combo["barmode"] = "stack"
	layout_combo["bargap"] = 0.1
	layout_combo["showlegend"] = True
	layout_combo["legend"] = dict(
	orientation="h", y=1.0, x=0,
	font=dict(size=12, color="#f1f5f9"),
	title_font=dict(size=12, color="#a78bfa"),
	groupclick="toggleitem",
	yanchor="bottom",
	xanchor="left",
	bgcolor="rgba(0,0,0,0)",
	)
	layout_combo["margin"] = dict(l=10, r=80, t=80, b=10)
	layout_combo["xaxis"]["range"] = [0, 115]
	layout_combo["xaxis"]["ticksuffix"] = "%"
	layout_combo["yaxis"] = dict(
	tickvals=tick_vals,
	ticktext=tick_text,
	tickfont=dict(size=10),
	autorange="reversed",
	showgrid=False,
	zeroline=False,
	showline=False,
	)
	layout_combo["annotations"] = annotations
	fig_combo.update_layout(**layout_combo)

	st.plotly_chart(fig_combo, width='stretch', config={"displayModeBar": False})
	st.markdown('</div>', unsafe_allow_html=True)

	contrib_df = pd.DataFrame(contributors)
	csv_download(contrib_df, "Download CSV", "top_contributors.csv")
	else:
	st.info("Not enough data yet.")

	# ── REPEAT SPAMMERS ───────────────────────────────────────────
	st.divider()
	st.markdown(
	'<div class="sec-hdr"><span class="sec-ttl">Repeat Spammers</span><span class="sec-pill">All Time</span></div>',
	unsafe_allow_html=True
	)
	rs_col1, rs_col2 = st.columns([1, 1])
	with rs_col1:
	rs_window = st.slider("Time window (sec)", 5, 60, 15, key="rs_window")
	with rs_col2:
	rs_min = st.slider("Min repeats to flag", 2, 10, 2, key="rs_min")

	_rs_json = json.dumps([{
	"author": m.get("author",""), "text": m.get("text",""),
	"topic": m.get("topic","General"), "sentiment": m.get("sentiment","Neutral"),
	"time": m.get("time","")
	} for m in all_data])
	repeat_spammers = detect_repeat_spammers(_rs_json, window_sec=rs_window, min_repeats=rs_min)

	if repeat_spammers:
	st.markdown(
	f'<div style="font-size:0.78rem;color:var(--text-3);margin-bottom:12px;">'
	f'Found <b style="color:var(--text-1);">{len(repeat_spammers)}</b> users repeating the same message '
	f'≥{rs_min}× within {rs_window}s</div>',
	unsafe_allow_html=True
	)
	for rs in repeat_spammers:
	_t_color = TOPIC_COLOR.get(rs["topic"], "#6b7280")
	_s_color = SENT_COLORS.get(rs["sentiment"], "#6b7280")
	_burst = rs["max_burst"]
	_total = rs["count"]
	_severity = "#ef4444" if _burst >= 5 else "#eab308" if _burst >= 3 else "#f59e0b"
	st.markdown(
	f'<div class="chat-card" style="border-left:3px solid {_severity};">'
	f'<div style="display:flex;align-items:center;justify-content:space-between;margin-bottom:6px;">'
	f'<div class="chat-author">⚠️ {rs["author"]}</div>'
	f'<div style="display:flex;gap:6px;">'
	f'<span class="badge" style="color:{_severity};border-color:{_severity}44;">'
	f'🔁 {_burst}× in {rs_window}s</span>'
	f'<span class="badge" style="color:var(--text-3);">{_total} total</span>'
	f'</div></div>'
	f'<div class="chat-text">"{rs["text"]}"</div>'
	f'<div class="chat-badges">'
	f'<span class="badge" style="color:{_s_color};border-color:{_s_color}33;">{rs["sentiment"]}</span>'
	f'<span class="badge" style="color:{_t_color};border-color:{_t_color}33;">{rs["topic"]}</span>'
	f'<span class="badge">First: {rs["first_seen"]}</span>'
	f'<span class="badge">Last: {rs["last_seen"]}</span>'
	f'</div></div>',
	unsafe_allow_html=True
	)
	rs_df = pd.DataFrame(repeat_spammers)
	csv_download(rs_df, "Download CSV", "repeat_spammers.csv")
	else:
	st.markdown(
	'<div style="font-size:0.84rem;color:var(--text-3);padding:12px 0;">No repeat spammers detected in current window.</div>',
	unsafe_allow_html=True
	)

	# ── MULTI-STREAM COMPARISON ───────────────────────────────────
	active_streams = [s for s in st.session_state.streams if r.llen(s["redis_key"]) > 0]

	if len(active_streams) > 1:
	st.divider()
	n_streams = len(active_streams)
	st.markdown(
	f'<div class="sec-hdr"><span class="sec-ttl">Multi-Stream Comparison</span>'
	f'<span class="sec-pill">{n_streams} streams</span></div>',
	unsafe_allow_html=True
	)

	# ── Load all stream data ONCE (fix double-load) ───────────
	_stream_cache: dict[str, dict] = {}
	for _s in active_streams:
	_rkey = _s["redis_key"]
	_raw = load_stream_data(_rkey)
	if not _raw:
	continue
	_sdf = pd.DataFrame(_raw)
	_sdf["sentiment"] = _sdf["sentiment"].apply(clean_sentiment)
	_sdf["topic"] = _sdf["topic"].apply(clean_topic) if "topic" in _sdf.columns else "General"
	_sc = _sdf["sentiment"].value_counts().to_dict()
	_p = _sc.get("Positive", 0)
	_n = _sc.get("Neutral", 0)
	_g = _sc.get("Negative", 0)
	_t = max(_p + _n + _g, 1)
	_tc = {lbl: int((_sdf["topic"] == lbl).sum()) for lbl in TOPIC_LABELS}
	_top_topic = max(_tc, key=_tc.get)
	_eng_json = json.dumps([
	{"sentiment": m.get("sentiment","Neutral"),
	"topic": m.get("topic","General"),
	"time": m.get("time","")} for m in _raw
	])
	_eng = compute_engagement(_eng_json)
	_title = _s.get("video_title") or _s.get("video_id") or _rkey
	_stream_cache[_rkey] = {
	"df": _sdf, "raw": _raw,
	"p": _p, "n": _n, "g": _g, "t": _t,
	"tc": _tc, "top_topic": _top_topic,
	"eng": _eng, "title": _title,
	"sidx": st.session_state.streams.index(_s),
	}

	# ── Head-to-head comparison table ─────────────────────────
	st.markdown('<div class="chart-wrap" style="margin-bottom:16px;">', unsafe_allow_html=True)
	st.markdown('<div class="chart-title">Head-to-Head Summary</div><div class="chart-sub">All active streams at a glance</div>', unsafe_allow_html=True)
	_hth_rows = []
	for _s in active_streams:
	_rkey = _s["redis_key"]
	if _rkey not in _stream_cache:
	continue
	_c = _stream_cache[_rkey]
	_sidx = _c["sidx"]
	_hth_rows.append({
	"Stream": f"Stream {STREAM_NAMES[_sidx]}",
	"Title": _c["title"][:30],
	"Messages": _c["t"],
	"Positive %": f"{_c['p']/_c['t']*100:.1f}%",
	"Neutral %": f"{_c['n']/_c['t']*100:.1f}%",
	"Negative %": f"{_c['g']/_c['t']*100:.1f}%",
	"Top Topic": _c["top_topic"],
	"Engagement": f"{_c['eng']['score']}/100 {_c['eng']['grade']}",
	})
	if _hth_rows:
	st.dataframe(pd.DataFrame(_hth_rows), hide_index=True, use_container_width=True)
	st.markdown('</div>', unsafe_allow_html=True)

	# ── Per-stream sentiment + topic + engagement cards ────────
	chunk_size = 2
	_cached_keys = [_s["redis_key"] for _s in active_streams if _s["redis_key"] in _stream_cache]
	for row_start in range(0, len(_cached_keys), chunk_size):
	row_keys = _cached_keys[row_start:row_start + chunk_size]
	cols = st.columns(len(row_keys))
	for col, _rkey in zip(cols, row_keys):
	_c = _stream_cache[_rkey]
	_sidx = _c["sidx"]
	color = STREAM_COLORS[_sidx]
	slabel = STREAM_NAMES[_sidx]
	_p, _n, _g, _t = _c["p"], _c["n"], _c["g"], _c["t"]
	_eng = _c["eng"]
	_tc = _c["tc"]
	with col:
	st.markdown(
	f'<span class="compare-label" style="background:{color}18;color:{color};border:1px solid {color}44;">'
	f'Stream {slabel} · {_c["title"][:25]}</span>',
	unsafe_allow_html=True
	)
	_ec = "#22c55e" if _eng["score"] >= 70 else "#eab308" if _eng["score"] >= 40 else "#ef4444"
	st.markdown(
	f'<div style="display:flex;gap:10px;margin:6px 0 10px;flex-wrap:wrap;">'
	f'<div style="background:var(--bg-card);border:1px solid {_ec}44;border-radius:12px;padding:8px 14px;">'
	f'<div style="font-size:1.4rem;font-weight:800;color:{_ec};">{_eng["score"]}</div>'
	f'<div style="font-size:0.68rem;color:var(--text-3);text-transform:uppercase;">Engagement</div>'
	f'</div>'
	f'<div style="background:var(--bg-card);border:1px solid var(--border);border-radius:12px;padding:8px 14px;">'
	f'<div style="font-size:1.4rem;font-weight:800;color:var(--text-1);">{_t}</div>'
	f'<div style="font-size:0.68rem;color:var(--text-3);text-transform:uppercase;">Messages</div>'
	f'</div>'
	f'<div style="background:var(--bg-card);border:1px solid var(--border);border-radius:12px;padding:8px 14px;">'
	f'<div style="font-size:1.4rem;font-weight:800;color:#22c55e;">{_p/_t*100:.0f}%</div>'
	f'<div style="font-size:0.68rem;color:var(--text-3);text-transform:uppercase;">Positive</div>'
	f'</div>'
	f'</div>',
	unsafe_allow_html=True
	)
	st.markdown('<div class="chart-wrap" style="margin-bottom:8px;">', unsafe_allow_html=True)
	st.markdown('<div class="chart-title" style="font-size:0.78rem;">Sentiment</div>', unsafe_allow_html=True)
	fig_s = go.Figure(go.Bar(
	x=["Pos", "Neu", "Neg"],
	y=[_p, _n, _g],
	marker_color=["#22c55e", "#eab308", "#ef4444"],
	marker_line_width=0,
	text=[_p, _n, _g],
	textposition="outside",
	hovertemplate="<b>%{x}</b>: %{y}<extra></extra>",
	))
	fig_s.update_layout(**plotly_layout(180))
	st.plotly_chart(fig_s, width='stretch', config={"displayModeBar": False})
	st.markdown('</div>', unsafe_allow_html=True)
	st.markdown('<div class="chart-wrap">', unsafe_allow_html=True)
	st.markdown('<div class="chart-title" style="font-size:0.78rem;">Topic Breakdown</div>', unsafe_allow_html=True)
	_tc_colors = ["#f59e0b","#3b82f6","#ec4899","#ef4444","#6b7280","#10b981"]
	fig_t = go.Figure(go.Bar(
	x=TOPIC_LABELS,
	y=[_tc[l] for l in TOPIC_LABELS],
	marker_color=_tc_colors,
	marker_line_width=0,
	text=[_tc[l] for l in TOPIC_LABELS],
	textposition="outside",
	hovertemplate="<b>%{x}</b>: %{y}<extra></extra>",
	))
	_tl = plotly_layout(180)
	_tl["xaxis"]["tickfont"] = dict(size=8)
	fig_t.update_layout(**_tl)
	st.plotly_chart(fig_t, width='stretch', config={"displayModeBar": False})
	st.markdown('</div>', unsafe_allow_html=True)

	# ── Overlay: positive ratio over time (all streams) ────────
	st.markdown('<div class="chart-wrap" style="margin-top:14px;">', unsafe_allow_html=True)
	st.markdown('<div class="chart-title">Positive Ratio Over Time</div><div class="chart-sub">Rolling positive % per stream (synced refresh)</div>', unsafe_allow_html=True)
	fig_overlay = go.Figure()
	for _rkey, _c in _stream_cache.items():
	_sidx = _c["sidx"]
	color = STREAM_COLORS[_sidx]
	slabel = STREAM_NAMES[_sidx]
	_sdf = _c["df"].copy()
	_sdf["is_pos"] = (_sdf["sentiment"] == "Positive").astype(int)
	_sdf["rolling"] = _sdf["is_pos"].rolling(10, min_periods=1).mean() * 100
	fig_overlay.add_trace(go.Scatter(
	x=list(range(len(_sdf))),
	y=_sdf["rolling"],
	mode="lines",
	name=f"Stream {slabel} · {_c['title'][:20]}",
	line=dict(color=color, width=2),
	hovertemplate=f"Stream {slabel} msg %{{x}}: %{{y:.1f}}%<extra></extra>",
	))
	layout_ov = plotly_layout(220)
	layout_ov["showlegend"] = True
	layout_ov["legend"] = dict(orientation="h", y=1.08, font=dict(size=11, color="#f1f5f9"))
	layout_ov["yaxis"]["range"] = [0, 100]
	fig_overlay.update_layout(**layout_ov)
	st.plotly_chart(fig_overlay, width='stretch', config={"displayModeBar": False})
	st.markdown('</div>', unsafe_allow_html=True)

	elif len(st.session_state.streams) > 1:
	st.divider()
	st.info("Add video IDs to your extra stream slots and click ▶ Start to enable multi-stream comparison.")

	# ── AUTO REFRESH ──────────────────────────────────────────────
	if auto_refresh:
	time.sleep(refresh_rate)
	st.rerun()