app.py

# Smart Office Attendance — YOLO (Real ML) + Seeded Inputs (no DB) + InsightFace (CPU) + Daily CSV
# Views: Live / Employees / Reports
# POC polish + Readonly/Kiosk URLs + Light Blue Theme + Borders
# CHANGES in this version:
# - (NEW) Per-employee live timers on video overlays: Working and Idle counters (mm:ss)
# - (NEW) Employees view shows the same live Working/Idle seconds
# - (Patch I–K) Real-time streaming pace: higher default FPS, up to 60 FPS slider, process every frame when possible, time-synced display (no fixed sleeps).
# - (Patch M) Real-time catch-up reader (clip plays in wall time; drop backlog if behind)
# - (Patch O) Downscale only for inference (rescale boxes for draw)
# - (Patch P) Run InsightFace every ~5 processed frames (reuse embeddings in between)
# - (Patch Q) Update progress bar every ~10 frames

import os, json, math, tempfile, time
from datetime import datetime, date
from pathlib import Path
from dataclasses import dataclass

import cv2
import numpy as np
import pandas as pd
import plotly.express as px
from PIL import Image
import streamlit as st
from ultralytics import YOLO
from collections import deque  # temporal votes

# ---------- InsightFace (CPU) ----------
from insightface.app import FaceAnalysis
from sklearn.metrics.pairwise import cosine_similarity

# -------------------- Page / Theme --------------------
st.set_page_config(
    page_title="Smart Office Attendance — YOLO + FaceID (CPU) + CSV",
    page_icon="📡",
    layout="wide",
    menu_items={"Get help": None, "Report a Bug": None, "About": None},
)

# URL flags
qp = st.query_params
KIOSK = str(qp.get("kiosk", ["0"])[0]).lower() in ("1", "true", "yes")
READ_ONLY = str(qp.get("readonly", ["0"])[0]).lower() in ("1", "true", "yes")

if KIOSK:
    st.markdown("""
    <style>
      [data-testid="stToolbar"] { display:none !important; }
      footer, [data-testid="stStatusWidget"] { visibility:hidden !important; }
      [data-testid="stDecoration"] { display:none !important; }
    </style>
    """, unsafe_allow_html=True)

CSS = """
:root{
  /* Light Blue Theme */
  --bg:#eef6ff;           /* app background */
  --card:#ffffff;         /* card background */
  --bd:#b7cffb;           /* borders (stronger) */
  --muted:#5b6b8c;        /* secondary text */
  --txt:#0b1a33;          /* primary text */
  --accent:#2563eb;       /* links / accents */
}
html,body,[data-testid="stAppViewContainer"]{background:var(--bg);}
.block-container {padding-top: 0.8rem; padding-bottom: 3rem; max-width: 1240px;}
.navbar {
  position: sticky; top: 0; z-index: 999; backdrop-filter: blur(8px);
  background: linear-gradient(180deg, rgba(255,255,255,0.92), rgba(255,255,255,0.86));
  border-bottom: 2px solid var(--bd);
  box-shadow: 0 6px 16px rgba(30,64,175,0.08);
}
.navbar-inner {display:flex; align-items:center; justify-content:space-between; padding: 12px 0;}
.navbrand {display:flex; gap:.7rem; align-items:center}
.brandbox {width:32px; height:32px; border-radius:10px; background:#e3edff; color:#1e40af;
  display:grid; place-items:center; font-weight:700}

.card {
  border:2px solid var(--bd); border-radius:14px; padding:14px; background: var(--card);
  box-shadow: 0 6px 16px rgba(30,64,175,0.06);
}
.card:hover{border-color:#9db9fb}
.badge {
  display:inline-flex; align-items:center; gap:.4rem; border:1px solid #cfe0ff;
  border-radius:10px; padding:2px 8px; font-size:.75rem; color:#1e3a8a; background:#f3f7ff;
}
.pill {display:inline-flex; align-items:center; gap:.35rem; border-radius:999px; padding:4px 10px; font-size:.75rem;}
.pill-success {background:#e8fff6; color:#065f46; border:1px solid #b7f3df}
.pill-warn   {background:#fff4e6; color:#92400e; border:1px solid #ffd3a6}
.pill-info   {background:#eef2ff; color:#1e3a8a; border:1px solid #cfd8ff}
.pill-danger {background:#ffeaea; color:#7f1d1d; border:1px solid #ffc9c9}

.grid {display:grid; gap:14px}
@media(min-width: 768px){ .grid.cols-2 {grid-template-columns: repeat(2, 1fr);} }
@media(min-width: 1200px){ .grid.cols-3 {grid-template-columns: repeat(3, 1fr);} }

.video {
  height: 220px; border-radius:12px; border:2px dashed #b7cffb;
  display:grid; place-items:center; color:var(--muted); background:#fbfdff;
}

.state {text-align:center; color:var(--muted); padding:20px}
.footerpad {height: 40px}
[data-testid="stMetricValue"]{color:var(--txt);}
"""
st.markdown(f"<style>{CSS}</style>", unsafe_allow_html=True)

# -------------------- Demo mode flag (kept) --------------------
POC_DEMO = True
SHOW_DEV_CONTROLS = not POC_DEMO and not READ_ONLY

# -------------------- Seed loading (inputs only) --------------------
def load_seed():
    p = Path("seed.json")
    if p.exists():
        with open(p, "r", encoding="utf-8") as f:
            return json.load(f)
    return {
        "cameras": {
            "cam-1":{"name":"Lobby Cam","zone":"Lobby","online":True,"latencyMs":120,"fps":15},
            "cam-2":{"name":"West Corridor","zone":"Corridor","online":True,"latencyMs":95,"fps":22},
            "cam-3":{"name":"Open Desk A","zone":"Desk","online":True,"latencyMs":135,"fps":18},
        },
        "employees":{
            "e1":{"name":"Aparna Rao","title":"Project Manager","deskId":"D-14","avatarUrl":"https://placehold.co/64","status":"At Desk","team":"Ops"},
            "e2":{"name":"Karthik S","title":"ML Engineer","deskId":"D-27","avatarUrl":"https://placehold.co/64","status":"At Desk","team":"Eng"},
            "e3":{"name":"Nisha M","title":"Ops Lead","deskId":"C-05","avatarUrl":"https://placehold.co/64","status":"Away","team":"Ops"}
        },
        "detections": {},
        "sessions": {},
        "alerts": {},
        "appSettings": {"default":{"onPhoneSec":60,"idleMin":10,"blurFaces":False,"retentionDays":30}},
        "trainingImages": {}
    }

SEED = load_seed()
EMP = SEED.get("employees", {})
CAM = SEED.get("cameras", {})
SESS = SEED.get("sessions", {})
ALR = SEED.get("alerts", {})
APPS = SEED.get("appSettings", {}).get("default", {"onPhoneSec":60,"idleMin":10,"blurFaces":False,"retentionDays":30})

# -------------------- Session State --------------------
if "events" not in st.session_state:
    st.session_state.events = pd.DataFrame(columns=["id","ts","camera","employee","activity","zone","confidence","run_id"])
if "current_run_id" not in st.session_state:
    st.session_state.current_run_id = None

st.session_state.setdefault("selected_site_floor", None)
st.session_state.setdefault("selected_camera", None)
st.session_state.setdefault("privacy_blur", bool(APPS.get("blurFaces", False)))

for _i in range(3):
    st.session_state.setdefault(f"last_video_bytes_{_i}", None)
    st.session_state.setdefault(f"last_video_suffix_{_i}", ".mp4")

st.session_state.setdefault("run_meta", {})
st.session_state.setdefault("frame_cache", {})
st.session_state.setdefault("frame_paths", {})
st.session_state.setdefault("last_uploaded_names", [])

st.session_state.setdefault("emp_summary", {})
st.session_state.setdefault("face_demo_map", {})
st.session_state.setdefault("face_demo_used", set())

st.session_state.setdefault("face_db", {})
st.session_state.setdefault("face_ready", False)

st.session_state.setdefault("persist_count", 0)
st.session_state.setdefault("persisted_csv", "")
st.session_state.setdefault("daily_summary_path", "")

# live counters store
st.session_state.setdefault("emp_counters", {})

# temporal votes + previous boxes (per name)
st.session_state.setdefault("act_votes", {})   # {name: deque(maxlen=5) of "phone"/"not"}
st.session_state.setdefault("prev_boxes", {})  # {name: last person box}

# On Phone time in nanoseconds (per name)
st.session_state.setdefault("on_phone_start_ns", {})  # {name: ns or None}
st.session_state.setdefault("on_phone_accum_ns", {})  # {name: total ns}

# -------------------- ML (YOLO + face utils) --------------------
@dataclass
class DetBox:
    cls: str
    conf: float
    box: tuple

@st.cache_resource(show_spinner=False)
def load_model():
    return YOLO("yolov8n.pt")

@st.cache_resource(show_spinner=False)
def load_face_cascade():
    return cv2.CascadeClassifier(cv2.data.haarcascades + "haarcascade_frontalface_default.xml")

@st.cache_resource(show_spinner=False)
def load_face_analysis():
    app = FaceAnalysis(name="buffalo_l")
    app.prepare(ctx_id=-1, det_size=(640, 640))
    return app

def run_yolo_on_frame(model: YOLO, frame_bgr, conf_thres=0.25):
    res = model.predict(frame_bgr, verbose=False, conf=conf_thres)[0]
    names = res.names
    out = []
    if res.boxes is None: return out
    for b in res.boxes:
        c = int(b.cls.item())
        conf = float(b.conf.item()) if b.conf is not None else 0.0
        xyxy = tuple(map(int, b.xyxy[0].tolist()))
        out.append(DetBox(names[c], conf, xyxy))
    return out

def iou(a, b):
    xA, yA = max(a[0], b[0]), max(a[1], b[1])
    xB, yB = min(a[2], b[2]), min(a[3], b[3])
    inter = max(0, xB-xA) * max(0, yB-yA)
    if inter == 0: return 0.0
    areaA = (a[2]-a[0])*(a[3]-a[1]); areaB = (b[2]-b[0])*(b[3]-b[1])
    return inter / float(areaA + areaB - inter + 1e-6)

def phone_near_head(person_box, phone_boxes, face_boxes):
    (px1, py1, px2, py2) = person_box
    head_h = int(py1 + 0.4 * (py2 - py1))
    head_box = (px1, py1, px2, head_h)
    exp_faces = []
    for f in face_boxes:
        if not isinstance(f, (list, tuple)) or len(f) != 4:
            continue
        try:
            fx1, fy1, fx2, fy2 = map(int, f)
        except Exception:
            continue
        w = fx2 - fx1
        h = fy2 - fy1
        ex = (max(px1, fx1 - w//6), max(py1, fy1 - h//6),
              min(px2, fx2 + w//6), min(py2, fy2 + h//6))
        exp_faces.append(ex)
    for ph in phone_boxes:
        if not hasattr(ph, "box") or not isinstance(ph.box, (list, tuple)) or len(ph.box) != 4:
            continue
        try:
            if any(isinstance(f, (list, tuple)) and len(f) == 4 and iou(ph.box, f) > 0.02 for f in exp_faces):
                return True
            if iou(ph.box, head_box) > 0.05:
                return True
        except Exception:
            continue
    return False

def blur_faces_if_needed(frame_bgr, face_cascade, enable):
    if not enable: return frame_bgr
    gray = cv2.cvtColor(frame_bgr, cv2.COLOR_BGR2GRAY)
    faces = face_cascade.detectMultiScale(gray, scaleFactor=1.2, minNeighbors=5, minSize=(32,32))
    out = frame_bgr.copy()
    for (x, y, w, h) in faces:
        roi = out[y:y+h, x:x+w]
        roi = cv2.GaussianBlur(roi, (31,31), 15)
        out[y:y+h, x:x+w] = roi
    return out

def phone_like_rect_near_face(frame_bgr, gray, person_box, face_boxes):
    (px1, py1, px2, py2) = person_box
    head_h = int(py1 + 0.45 * (py2 - py1))
    head_box = (max(0,px1), max(0,py1), max(0,min(px2, frame_bgr.shape[1]-1)), max(0,min(head_h, frame_bgr.shape[0]-1)))
    hx1, hy1, hx2, hy2 = head_box
    if hx2 <= hx1 or hy2 <= hy1: return False
    roi_gray = gray[hy1:hy2, hx1:hx2]
    if roi_gray.size == 0: return False
    edges = cv2.Canny(roi_gray, 80, 160)
    contours, _ = cv2.findContours(edges, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
    mean_int = float(np.mean(roi_gray)) if roi_gray.size else 255
    for cnt in contours:
        x,y,w,h = cv2.boundingRect(cnt)
        area = w*h
        if area < 120 or area > (roi_gray.shape[0]*roi_gray.shape[1]*0.25):
            continue
        ar = w / float(h+1e-6)
        if 0.4 < ar < 3.5:
            patch = roi_gray[y:y+h, x:x+w]
            if patch.size == 0: continue
            patch_mean = float(np.mean(patch))
            if patch_mean + 15 < mean_int:
                return True
    return False

def estimate_activity(person_box, phones, prev_box=None, motion_thresh=8.0):
    if prev_box is not None:
        cx0 = (prev_box[0]+prev_box[2])/2; cy0 = (prev_box[1]+prev_box[3])/2
        cx1 = (person_box[0]+person_box[2])/2; cy1 = (person_box[1]+person_box[3])/2
        if math.hypot(cx1-cx0) < motion_thresh: return "Idle"
        else: return "Working"
    return "Working"

# -------------------- InsightFace helpers --------------------
def get_face_pairs_insight(rgb_frame):
    app = load_face_analysis()
    bgr = cv2.cvtColor(rgb_frame, cv2.COLOR_RGB2BGR)
    faces = app.get(bgr)
    pairs = []
    for f in faces:
        x1, y1, x2, y2 = map(int, f.bbox)
        emb = f.embedding
        if emb is not None and emb.size > 0:
            pairs.append(((x1, y1, x2, y2), emb.astype(np.float32)))
    return pairs

def build_face_db_from_videos_insight(videos, max_people=6, frames_to_scan=200, every_n=5):
    names = [v.get("name","Unknown") for v in EMP.values()] or [f"Emp{i+1}" for i in range(max_people)]
    db = {n: [] for n in names}
    collected = []
    for fv in videos:
        with tempfile.NamedTemporaryFile(delete=False, suffix=os.path.splitext(fv.name)[1]) as t:
            data = fv.read()
            t.write(data); path = t.name
        cap = cv2.VideoCapture(path)
        count = 0
        while cap.isOpened() and count < frames_to_scan:
            ret, frame = cap.read()
            if not ret: break
            frame_no = int(cap.get(cv2.CAP_PROP_POS_FRAMES))
            if frame_no % every_n != 0:
                continue
            rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
            pairs = get_face_pairs_insight(rgb)
            for (_, emb) in pairs:
                collected.append(emb)
                if len(collected) >= max_people:
                    break
            if len(collected) >= max_people:
                break
            count += 1
        cap.release()
        os.remove(path)
        if len(collected) >= max_people:
            break
    i = 0
    for emb in collected:
        db[names[i % len(names)]].append(emb)
        i += 1
    db = {k:v for k,v in db.items() if v}
    return db

def recognize_name_from_embedding(emb, db, cosine_thresh=0.55, l2_tolerance=1.2):
    if not db: return None
    best_name, best_cos = None, -1.0
    for name, enc_list in db.items():
        arr = np.stack(enc_list, axis=0)
        sim = float(np.mean(cosine_similarity([emb], arr)))
        if sim > best_cos:
            best_cos, best_name = sim, name
    if best_name is not None and best_cos >= cosine_thresh:
        all_encs = np.stack(db[best_name], axis=0)
        dists = np.linalg.norm(all_encs - emb, axis=1)
        if float(np.mean(dists)) <= l2_tolerance:
            return best_name
    return None

def seed_name_order():
    return [v.get("name","Unknown") for v in EMP.values()][:3]

def demo_map_name(raw_label: str) -> str:
    if not raw_label or (not raw_label.startswith("Unknown") and raw_label in seed_name_order()):
        return raw_label
    if raw_label in st.session_state.face_demo_map:
        return st.session_state.face_demo_map[raw_label]
    for seed_nm in seed_name_order():
        if seed_nm not in st.session_state.face_demo_map.values():
            st.session_state.face_demo_map[raw_label] = seed_nm
            st.session_state.face_demo_used.add(seed_nm)
            return seed_nm
    return raw_label

# -------------------- Helpers --------------------
def latency_badge(latency_ms: int, fps: int) -> str:
    tone = "#10b981" if latency_ms < 120 else ("#f59e0b" if latency_ms < 200 else "#ef4444")
    return f"<span class='badge' style='border-color:{tone};color:{tone}'>{latency_ms}ms | {fps} FPS</span>"

def kpis_from_events(df: pd.DataFrame):
    if df is None or df.empty:
        return [("Present %","—"),("Avg Working hrs","—"),("Avg Idle","—"),("Corridor time","—"),("Alerts", str(len(ALR)))]
    working_min = int((df["activity"]=="Working").sum() * 0.5)
    idle_min = int((df["activity"]=="Idle").sum() * 0.5)
    corr_min = int((df["zone"]=="Corridor").sum() * 0.5)
    present_pct = f"{min(100, 40 + len(df)//2)}%"
    avg_work_hrs = round(6.5 + (working_min/120), 1)
    return [
        ("Present %", present_pct),
        ("Avg Working hrs", str(avg_work_hrs)),
        ("Avg Idle", f"{idle_min}m"),
        ("Corridor time", f"{corr_min}m"),
        ("Alerts", str(len(ALR))),
    ]

def save_run_summary_csv(run_id: str, site_floor: str, camera_name: str) -> str:
    df = st.session_state.events.copy()
    if df.empty: return ""
    df = df[df["run_id"] == run_id].copy()
    if df.empty: return ""
    df["ts"] = pd.to_datetime(df["ts"], errors="coerce", utc=True)
    counts = df.groupby(["employee","activity"]).size().unstack(fill_value=0)
    for col in ["Working","On Phone","Idle","Away"]:
        if col not in counts.columns: counts[col] = 0
    counts["total"] = counts["Working"] + counts["On Phone"] + counts["Idle"] + counts["Away"]
    pct = counts[["Working","On Phone","Idle","Away"]].div(counts["total"].replace(0,1), axis=0) * 100.0
    pct = pct.round(2)
    out = counts.join(pct.add_suffix(" %")).reset_index()
    out.insert(0, "camera", camera_name)
    out.insert(0, "site_floor", site_floor)
    out.insert(0, "run_id", run_id)
    out.insert(0, "date", date.today().isoformat())
    out_path = f"/tmp/run_summary_{run_id}.csv"
    out.to_csv(out_path, index=False)
    return out_path

def cache_event_frame(event_id: str, vis_bgr: np.ndarray):
    try:
        img = Image.fromarray(cv2.cvtColor(vis_bgr, cv2.COLOR_BGR2RGB))
        st.session_state.frame_cache[event_id] = img
        thumb = img.copy(); thumb.thumbnail((96, 96))
        thumb_path = f"/tmp/frame_{event_id}.png"
        thumb.save(thumb_path, format="PNG")
        st.session_state.frame_paths[event_id] = thumb_path
        if len(st.session_state.frame_cache) > 500:
            drop = list(st.session_state.frame_cache.keys())[:-400]
            for k in drop:
                st.session_state.frame_cache.pop(k, None)
                st.session_state.frame_paths.pop(k, None)
    except Exception:
        pass

def write_daily_emp_summary_csv():
    if not st.session_state.emp_summary:
        return ""
    today = date.today().strftime("%Y%m%d")
    path = f"/tmp/summary_{today}.csv"
    rows = []
    for emp, c in st.session_state.emp_summary.items():
        w = int(c.get("Working", 0)); ph = int(c.get("On Phone", 0)); idle = int(c.get("Idle", 0)); away = int(c.get("Away", 0))
        total = max(1, w + ph + idle + away)
        rows.append({
            "Employee": emp,
            "Working": w, "On Phone": ph, "Idle": idle, "Away": away,
            "Working %": round(w/total*100, 2), "On Phone %": round(ph/total*100, 2),
            "Idle %": round(idle/total*100, 2), "Away %": round(away/total*100, 2)
        })
    df = pd.DataFrame(rows)
    df.to_csv(path, index=False)
    st.session_state.daily_summary_path = path
    return path

def recent_frames_for_employee(emp_name: str, n: int = 3):
    if st.session_state.events is None or st.session_state.events.empty:
        return []
    df = st.session_state.events.copy()
    try:
        df["ts"] = pd.to_datetime(df["ts"], errors="coerce", utc=True)
    except Exception:
        pass
    df = df[df["employee"] == emp_name].sort_values("ts", ascending=False)
    frames = []
    for _, r in df.iterrows():
        eid = r["id"]
        img = st.session_state.frame_cache.get(eid)
        if img is not None:
            frames.append((eid, img))
        if len(frames) >= n:
            break
    return frames

def fmt_mmss(secs: float) -> str:
    secs = max(0, int(round(secs)))
    m, s = divmod(secs, 60)
    return f"{m:02d}:{s:02d}"

def fmt_hhmmss_ns(ns_total: int) -> str:
    secs, ns = divmod(int(max(0, ns_total)), 1_000_000_000)
    mins, secs = divmod(secs, 60)
    hrs, mins = divmod(mins, 60)
    return f"{int(hrs):02}:{int(mins):02}:{int(secs):02}.{int(ns):09d}"

def center_speed(cur_box, prev_box):
    if prev_box is None or cur_box is None:
        return 0.0
    cx0 = (prev_box[0]+prev_box[2]) * 0.5; cy0 = (prev_box[1]+prev_box[3]) * 0.5
    cx1 = (cur_box[0]+cur_box[2]) * 0.5; cy1 = (cur_box[1]+cur_box[3]) * 0.5
    return math.hypot(cx1 - cx0, cy1 - cy0)

# Patch K helper: clock-paced UI display
def _should_display(last_t, target_dt):
    """Return True if it's time to push next frame to UI."""
    return (time.perf_counter() - last_t) >= target_dt

# -------------------- Navbar --------------------
st.markdown("""
<div class='navbar'>
  <div class='navbar-inner'>
    <div class='navbrand'>
      <div class='brandbox'>SO</div>
      <div>
        <div style='font-size:12px;color:#60749b;text-transform:uppercase;letter-spacing:.08em'>Smart Office Attendance</div>
        <div style='font-weight:600'>Live • Employees • Reports</div>
      </div>
    </div>
    <div style='display:flex; gap:.5rem; align-items:center'>
      <span class='badge'>Sync: <code>External CRM (mock)</code></span>
    </div>
  </div>
</div>
""", unsafe_allow_html=True)

# -------------------- Global line (Settings removed) / Privacy / Clock --------------------
top = st.columns([1,1,1,1,1,1])
with top[-2]:
    st.session_state.privacy_blur = st.toggle(
        "Privacy: blur faces",
        value=st.session_state.get("privacy_blur", bool(APPS.get("blurFaces", False))),
        help="Reason: Compliance / PII policy",
        disabled=READ_ONLY
    )
with top[-1]:
    st.write(":alarm_clock:", datetime.now().strftime("%Y-%m-%d %H:%M:%S"))

# -------------------- Tabs --------------------
tab_live, tab_emp, tab_reports = st.tabs(["🟢 Live", "👥 Employees", "📊 Reports"])

# ======================================================
# LIVE
# ======================================================
with tab_live:
    h1, h2, h3 = st.columns([1.2, 1.0, 1.0])
    with h1:
        st.session_state.selected_site_floor = st.selectbox(
            "Site / Floor (required before upload)",
            ["HQ • L2","HQ • L3","Annex • L1"], index=0, key="site_selector", disabled=READ_ONLY
        )
    with h2:
        cam_names = [v["name"] for v in CAM.values()] or ["Lobby Cam"]
        st.session_state.selected_camera = st.selectbox(
            "Camera (required before upload)", cam_names, index=0, key="cam_selector", disabled=READ_ONLY
        )
    with h3:
        st.toggle("Maximize stream", value=st.session_state.get("stream_full", False),
                  key="stream_full", help="Show one live stream full-width", disabled=READ_ONLY)

    online = [c for c in CAM.values() if c.get("online")]
    cols = st.columns(3)
    stream_canvases = [None, None, None]
    replay_btns = [False, False, False]

    if online and not st.session_state.get("stream_full", False):
        for i in range(3):
            cam = online[i] if i < len(online) else {"name": f"Camera {i+1}", "zone": "—", "latencyMs": 0, "fps": 0}
            with cols[i]:
                st.markdown("<div class='card'>", unsafe_allow_html=True)
                st.markdown(f"**{cam['name']}** — {cam.get('zone','—')}")
                st.markdown(latency_badge(cam.get("latencyMs",0), cam.get("fps",0)), unsafe_allow_html=True)
                stream_canvases[i] = st.empty()
                replay_btns[i] = st.button("Replay", key=f"replay_tile_{i}", disabled=READ_ONLY)
                st.markdown("</div>", unsafe_allow_html=True)
    else:
        st.markdown("<div class='card'>", unsafe_allow_html=True)
        st.caption("Live stream (maximized)")
        stream_canvases[0] = st.empty()
        replay_btns[0] = st.button("Replay", key="replay_tile_0_max", disabled=READ_ONLY)
        st.markdown("</div>", unsafe_allow_html=True)

    for i in range(3):
        if stream_canvases[i] is None:
            stream_canvases[i] = st.empty()

    # --- Advanced defaults (Patch I): push baseline up + disable skip by default
    conf_thres, sample_fps, max_secs, motion_px = 0.25, 24, 30, 8
    stream_live, skip_near_identical, diff_thresh, event_cooldown_sec = True, False, 3.5, 1.0

    if not READ_ONLY:
        with st.expander("Advanced (ML & System)", expanded=False):
            c1, c2, c3, c4 = st.columns(4)
            with c1: conf_thres = st.slider("Confidence", 0.1, 0.6, 0.25, 0.05)
            # slider up to 60 FPS (Patch I)
            with c2: sample_fps = st.slider("Sample FPS", 1, 60, 24)
            with c3: max_secs   = st.slider("Max seconds", 5, 120, 30)
            with c4: motion_px  = st.slider("Idle motion px", 2, 20, 8)

            dd1, dd2, dd3 = st.columns(3)
            with dd1: stream_live = st.checkbox("Live preview (stream)", value=True)
            with dd2: skip_near_identical = st.checkbox("Skip near-identical frames", value=False)
            with dd3: diff_thresh = st.slider("Frame diff threshold", 1.0, 20.0, 3.5, 0.5)

            cd1, cd2 = st.columns(2)
            with cd1: event_cooldown_sec = st.slider("Event cooldown (sec)", 0.0, 5.0, 1.0, 0.5)
            with cd2: st.caption("Replay per placeholder (buttons above)")
    else:
        st.caption("Read-only mode active — uploads, replay, clear & advanced controls are disabled.")

    def ensure_face_db(videos):
        if st.session_state.face_ready or READ_ONLY:
            return
        if not videos:
            return
        mem_files = []
        for fv in videos:
            data = fv.getvalue() if hasattr(fv, "getvalue") else fv.read()
            mem = tempfile.NamedTemporaryFile(delete=False, suffix=os.path.splitext(fv.name)[1])
            mem.write(data); mem.flush(); mem.close()
            class _MemWrap:
                def __init__(self, p, n): self._p=p; self.name=n
                def read(self):
                    with open(self._p, "rb") as f: return f.read()
            mem_files.append(_MemWrap(mem.name, fv.name))
        db = build_face_db_from_videos_insight(mem_files, max_people=max(3, len(EMP)))
        for mf in mem_files:
            try: os.remove(mf._p)
            except: pass
        st.session_state.face_db = db
        st.session_state.face_ready = True
        if db:
            st.success(f"Face DB ready for {len(db)} person(s): {', '.join(list(db.keys())[:6])}")
        else:
            st.warning("No faces found to build DB. Events will be tagged as Unknown.")

    def _ensure_counter(name: str):
        if name not in st.session_state.emp_counters:
            st.session_state.emp_counters[name] = {
                "state": None,
                "working_seconds": 0.0,
                "idle_seconds": 0.0
            }

    def _draw_timers(vis, x1, y1, name, act):
        c = st.session_state.emp_counters.get(name, {})
        w = fmt_mmss(c.get("working_seconds", 0.0))
        i = fmt_mmss(c.get("idle_seconds", 0.0))
        text = f"W {w} | I {i}"
        box_w = min(260, max(140, len(text)*7 + 16))
        y_top = max(0, y1 - 42)
        # line 1: Working/Idle
        cv2.rectangle(vis, (x1, y_top), (x1 + box_w, y_top + 18), (230, 240, 255), -1)
        cv2.putText(vis, text, (x1 + 6, y_top + 13), cv2.FONT_HERSHEY_SIMPLEX, 0.42, (30, 40, 80), 1, cv2.LINE_AA)
        # line 2: On Phone (ns)
        start_map = st.session_state.on_phone_start_ns
        accum_map = st.session_state.on_phone_accum_ns
        total_ns = int(accum_map.get(name, 0))
        if start_map.get(name) is not None:
            total_ns += (time.time_ns() - int(start_map[name]))
        pdisp = fmt_hhmmss_ns(total_ns)
        p_text = f"P {pdisp}"
        y2 = y_top + 20
        cv2.rectangle(vis, (x1, y2), (x1 + max(box_w, min(300, len(p_text)*7 + 16)), y2 + 18), (230, 240, 255), -1)
        cv2.putText(vis, p_text, (x1 + 6, y2 + 13), cv2.FONT_HERSHEY_SIMPLEX, 0.42, (30, 40, 80), 1, cv2.LINE_AA)

    def run_in_tile(tile_index: int, payload_bytes: bytes, suffix: str, tag_site_floor: str, tag_camera: str):
        st.session_state.current_run_id = datetime.utcnow().strftime("%Y%m%d-%H%M%S")
        run_id = f"{st.session_state.current_run_id}-T{tile_index+1}"
        st.session_state.run_meta[run_id] = {"site_floor": tag_site_floor, "camera": tag_camera}

        model = load_model()
        face_cascade = load_face_cascade()

        with tempfile.NamedTemporaryFile(delete=False, suffix=suffix) as t:
            t.write(payload_bytes)
            tmp_path = t.name

        cap = cv2.VideoCapture(tmp_path)
        src_fps = cap.get(cv2.CAP_PROP_FPS) or 25.0

        # Patch J: process every frame when target >= source fps
        if sample_fps >= (src_fps or 25.0):
            step = 1
        else:
            step = int(max(1, round((src_fps or 25.0) / sample_fps)))

        total = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
        max_frames = int(min(total, src_fps * max_secs))

        # per-frame time quantum (sec) we add to current activity
        frame_dt = float(step / (src_fps if src_fps > 0 else 25.0))

        prev_person_boxes, frame_no = {}, 0
        prev_gray = None
        last_tag_time = {}

        prog = st.progress(0.0)

        # Patch K: pacing clock
        last_display_t = time.perf_counter()
        target_dt = 1.0 / float(src_fps if src_fps and src_fps > 0 else 25.0)

        # Patch M: wall-clock anchor (real-time catch-up)
        wall_t0 = time.perf_counter()

        while cap.isOpened() and frame_no < max_frames:
            # ---- Patch M: real-time catch-up reader ----
            elapsed = time.perf_counter() - wall_t0
            target_idx = int(elapsed * (src_fps if src_fps and src_fps > 0 else 25.0))
            cur_idx = int(cap.get(cv2.CAP_PROP_POS_FRAMES))
            if target_idx > cur_idx + 1:
                cap.set(cv2.CAP_PROP_POS_FRAMES, target_idx)

            ret, frame = cap.read()
            if not ret:
                break
            pos = int(cap.get(cv2.CAP_PROP_POS_FRAMES))
            if pos % step != 0:
                frame_no += 1
                # still push raw frame at real-time pace if streaming
                if st.session_state.get("stream_live", True) and stream_live and stream_canvases[tile_index] is not None:
                    if _should_display(last_display_t, target_dt):
                        stream_canvases[tile_index].image(cv2.cvtColor(frame, cv2.COLOR_BGR2RGB), channels="RGB", use_column_width=True)
                        last_display_t = time.perf_counter()
                continue

            # ---- Patch O: downscale only for inference; rescale boxes for draw
            H0, W0 = frame.shape[:2]
            infer_scale = 1.0
            if W0 > 1280:
                infer_scale = 1280.0 / W0
                infer_frame = cv2.resize(frame, (int(W0*infer_scale), int(H0*infer_scale)), interpolation=cv2.INTER_AREA)
            else:
                infer_frame = frame

            def _inv(b):
                if infer_scale == 1.0: return b
                x1,y1,x2,y2 = b
                inv = 1.0 / infer_scale
                return (int(x1*inv), int(y1*inv), int(x2*inv), int(y2*inv))

            do_process = True
            gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
            if prev_gray is not None and skip_near_identical:
                mad = float(np.mean(cv2.absdiff(gray, prev_gray)))
                if mad < diff_thresh:
                    do_process = False
            prev_gray = gray

            if not do_process:
                # Even if we skip analytics, still display at real-time pace
                if stream_live and stream_canvases[tile_index] is not None and _should_display(last_display_t, target_dt):
                    stream_canvases[tile_index].image(cv2.cvtColor(frame, cv2.COLOR_BGR2RGB), channels="RGB", use_column_width=True)
                    last_display_t = time.perf_counter()
                frame_no += 1
                continue

            # YOLO on infer_frame, rescale boxes back
            dets = run_yolo_on_frame(model, infer_frame, conf_thres)
            persons = [DetBox(d.cls, d.conf, _inv(d.box)) for d in dets if d.cls == "person"]
            phones  = [DetBox(d.cls, d.conf, _inv(d.box)) for d in dets if d.cls in ("cell phone","mobile phone","phone")]

            vis = blur_faces_if_needed(frame, face_cascade, st.session_state.privacy_blur)

            # ---- Patch P: run InsightFace every ~5 processed frames; reuse otherwise
            if 'last_face_pairs' not in st.session_state: st.session_state.last_face_pairs = []
            if 'last_face_update_pos' not in st.session_state: st.session_state.last_face_update_pos = -999
            if (pos - st.session_state.last_face_update_pos) >= 5:
                rgb_infer = cv2.cvtColor(infer_frame, cv2.COLOR_BGR2RGB)
                face_pairs = get_face_pairs_insight(rgb_infer)
                st.session_state.last_face_pairs = face_pairs
                st.session_state.last_face_update_pos = pos
            else:
                face_pairs = st.session_state.last_face_pairs
            face_boxes_only = [_inv(b) for (b, _) in face_pairs]

            # (legacy var kept; harmless)
            rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)

            for idx, p in enumerate(persons):
                # motion / base activity
                if idx in prev_person_boxes:
                    base_act = "Idle" if estimate_activity(p.box, phones, prev_person_boxes[idx]) == "Idle" else "Working"
                else:
                    base_act = "Working"

                # identify person
                name = None
                for (fbox, emb) in face_pairs:
                    fx1, fy1, fx2, fy2 = _inv(fbox)
                    cx = (fx1+fx2)/2; cy = (fy1+fy2)/2
                    if (p.box[0] <= cx <= p.box[2]) and (p.box[1] <= cy <= p.box[3]):
                        name = recognize_name_from_embedding(emb, st.session_state.face_db)
                        if name: break
                if not name:
                    name = f"Unknown #{idx+1}"
                name = demo_map_name(name)

                # stricter + temporal phone logic
                prev_named_box = st.session_state.prev_boxes.get(name)
                spd = center_speed(p.box, prev_named_box)  # pixels per processed frame
                st.session_state.prev_boxes[name] = p.box

                near_head_phone = phone_near_head(p.box, phones, face_boxes_only)
                rect_like_phone = phone_like_rect_near_face(frame, gray, p.box, face_boxes_only)
                phone_raw = (near_head_phone and rect_like_phone)

                if spd > 10.0:
                    phone_raw = False

                dq = st.session_state.act_votes.get(name)
                if dq is None:
                    dq = deque(maxlen=5)
                    st.session_state.act_votes[name] = dq
                dq.append("phone" if phone_raw else "not")
                phone_votes = dq.count("phone")
                on_phone_stable = phone_votes >= 3  # need 3/5

                act = "On Phone" if on_phone_stable else base_act

                _ensure_counter(name)
                c = st.session_state.emp_counters[name]
                prev_state = c.get("state")

                # Advance timers based on current act
                if act == "Working":
                    c["working_seconds"] = c.get("working_seconds", 0.0) + frame_dt
                elif act == "Idle":
                    c["idle_seconds"] = c.get("idle_seconds", 0.0) + frame_dt

                # On Phone time ns start/stop per name
                start_map = st.session_state.on_phone_start_ns
                accum_map = st.session_state.on_phone_accum_ns
                now_ns = time.time_ns()
                if act == "On Phone":
                    if start_map.get(name) is None:
                        start_map[name] = now_ns
                else:
                    if start_map.get(name) is not None:
                        accum_map[name] = int(accum_map.get(name, 0)) + (now_ns - int(start_map[name]))
                        start_map[name] = None

                if prev_state != act:
                    c["state"] = act

                conf = int(p.conf * 100)
                (x1,y1,x2,y2) = p.box
                color = (16,185,129) if act=="Working" else (245,158,11) if act=="On Phone" \
                        else (113,113,122) if act=="Idle" else (239,68,68)
                cv2.rectangle(vis,(x1,y1),(x2,y2),color,2)
                label = f"{name} ({act} {conf}%)"
                cv2.rectangle(vis,(x1,y1-20),(x1+min(280, x2-x1), y1), color, -1)
                cv2.putText(vis,label,(x1+5,y1-6),cv2.FONT_HERSHEY_SIMPLEX,0.45,(0,0,0),1,cv2.LINE_AA)

                _draw_timers(vis, x1, y1-22, name, act)

                cam_name = tag_camera or f"Tile {tile_index+1}"
                zone = "Desk"
                if CAM:
                    seed_list = list(CAM.values())
                    zone = seed_list[min(tile_index, len(seed_list)-1)].get("zone","Desk")

                now_ts = datetime.utcnow()
                event_tag = f"{cam_name}|{act}|{zone}|{name}|T{tile_index+1}"
                can_emit = True
                cooldown = 1.0
                try:
                    cooldown = float(event_cooldown_sec)
                except Exception:
                    pass
                last_t = last_tag_time.get(event_tag)
                if last_t is not None and (now_ts - last_t).total_seconds() < cooldown:
                    can_emit = False

                if can_emit:
                    evt_id = f"evt-{len(st.session_state.events)+1}"
                    row = {
                        "id": evt_id,
                        "ts": now_ts,
                        "camera": cam_name,
                        "employee": name,
                        "activity": act,
                        "zone": zone,
                        "confidence": conf,
                        "run_id": run_id
                    }
                    st.session_state.events = pd.concat(
                        [st.session_state.events, pd.DataFrame([row])],
                        ignore_index=True
                    )
                    last_tag_time[event_tag] = now_ts
                    cache_event_frame(evt_id, vis)

                    if name not in st.session_state.emp_summary:
                        st.session_state.emp_summary[name] = {"Working":0,"On Phone":0,"Idle":0,"Away":0,"Corridor":0}
                    st.session_state.emp_summary[name][act] = st.session_state.emp_summary[name].get(act,0) + 1
                    if zone in ("Corridor","Lobby","Common"):
                        st.session_state.emp_summary[name]["Corridor"] = st.session_state.emp_summary[name].get("Corridor",0) + 1

                prev_person_boxes[idx] = p.box

            for ph in phones:
                (x1,y1,x2,y2) = ph.box
                cv2.rectangle(vis,(x1,y1),(x2,y2),(58,131,255),2)
                cv2.putText(vis,"phone",(x1,y1-6),cv2.FONT_HERSHEY_SIMPLEX,0.45,(58,131,255),1,cv2.LINE_AA)

            # Patch K: clock-paced UI push (no fixed sleep)
            if st.session_state.get("stream_live", True) and stream_live and stream_canvases[tile_index] is not None:
                if _should_display(last_display_t, target_dt):
                    stream_canvases[tile_index].image(cv2.cvtColor(vis, cv2.COLOR_BGR2RGB), channels="RGB", use_column_width=True)
                    last_display_t = time.perf_counter()

            frame_no += 1
            # Patch Q: throttle progress updates
            if frame_no % 10 == 0:
                prog.progress(min(1.0, frame_no/max_frames))

        cap.release(); os.remove(tmp_path); prog.progress(1.0)
        st.success(f"Tile {tile_index+1}: processed {frame_no} frames")

        out_csv = save_run_summary_csv(run_id, tag_site_floor, tag_camera)
        if out_csv and not READ_ONLY:
            with open(out_csv, "rb") as f:
                st.download_button(
                    label=f"Download run summary CSV ({Path(out_csv).name})",
                    data=f.read(),
                    file_name=Path(out_csv).name,
                    mime="text/csv",
                    key=f"dl_{run_id}"
                )

        write_daily_emp_summary_csv()

    # ---- Upload (multi) ----
    tagging_ready = bool(st.session_state.selected_site_floor) and bool(st.session_state.selected_camera)
    ups = st.file_uploader(
        "Upload one or more demo videos (MP4/AVI/MOV/MKV)",
        type=["mp4","avi","mov","mkv"],
        accept_multiple_files=True,
        key="uploader_multi",
        disabled=(not tagging_ready) or READ_ONLY
    )
    if not tagging_ready and not READ_ONLY:
        st.info("Pick **Site / Floor** and **Camera** first to enable uploads.")
    if READ_ONLY:
        st.info("Read-only URL — uploads, replay and clear are disabled. Use the normal URL to process videos.")

    names = [u.name for u in ups] if ups else []

    if ups and st.session_state.last_uploaded_names != names and not READ_ONLY:
        st.session_state.last_uploaded_names = names
        ensure_face_db(ups)
        for i in range(min(3, len(ups))):
            payload = ups[i].read()
            st.session_state[f"last_video_bytes_{i}"] = payload
            st.session_state[f"last_video_suffix_{i}"] = os.path.splitext(ups[i].name)[1]
            run_in_tile(
                i, payload, st.session_state[f"last_video_suffix_{i}"],
                st.session_state.selected_site_floor, st.session_state.selected_camera
            )

    for i in range(3):
        if replay_btns[i] and not READ_ONLY:
            key_b = f"last_video_bytes_{i}"
            key_s = f"last_video_suffix_{i}"
            if st.session_state.get(key_b) is None:
                st.warning(f"Tile {i+1}: no previous video to replay.")
            else:
                run_in_tile(
                    i, st.session_state[key_b], st.session_state.get(key_s, ".mp4"),
                    st.session_state.selected_site_floor or "—",
                    st.session_state.selected_camera or f"Tile {i+1}"
                )

    # ---- Detection Events ----
    st.markdown("<div class='card'>", unsafe_allow_html=True)
    st.subheader("Detection Events")
    f1, f2, f3 = st.columns(3)
    with f1: st.selectbox("Employee", ["All"] + [v.get("name") for v in EMP.values()], key="events_filter_employee")
    with f2: st.selectbox("Activity", ["All","Working","On Phone","Idle","Away"], key="events_filter_activity")
    with f3: st.selectbox("Zone", ["All","Desk","Corridor","Lobby","Common"], key="events_filter_zone")

    only_this_run = st.checkbox("Show only detections from this run", value=True, key="events_only_this_run")
    df_show = st.session_state.events.copy()
    if not df_show.empty:
        df_show["ts"] = pd.to_datetime(df_show["ts"], errors="coerce", utc=True)
        if only_this_run and st.session_state.current_run_id:
            df_show = df_show[df_show["run_id"].str.startswith(st.session_state.current_run_id, na=False)]
        df_show = df_show.sort_values("ts", ascending=False)
        sub = df_show.tail(20).reset_index(drop=True)

        for _, row in sub.iterrows():
            colA, colB, colC, colD = st.columns([1.6, 1.3, 0.8, 0.8])
            with colA:
                st.markdown(f"**{row['activity']}** — {row['employee']}")
                st.caption(f"{row['camera']} | {row['zone']}")
            with colB:
                ts_str = row["ts"]
                try: ts_str = row["ts"].strftime("%Y-%m-%d %H:%M:%S %Z")
                except: pass
                st.caption(ts_str)
            with colC:
                st.write(f"{int(row.get('confidence',0))}%")
            with colD:
                st.write(row["id"])
            st.divider()
    else:
        st.info("No detections yet — upload a video to generate events.")
    st.markdown("</div>", unsafe_allow_html=True)

    # ---- Per-Employee Distribution (current run) ----
    st.markdown("<div class='card'>", unsafe_allow_html=True)
    st.subheader("Per-Employee Distribution (current run)")
    if not st.session_state.events.empty and st.session_state.current_run_id:
        run_mask = st.session_state.events["run_id"].str.startswith(st.session_state.current_run_id, na=False)
        cur = st.session_state.events[run_mask].copy()
        if not cur.empty:
            counts = cur.groupby(["employee","activity"]).size().unstack(fill_value=0)
            for col in ["Working","On Phone","Idle","Away"]:
                if col not in counts.columns: counts[col] = 0
            counts["total"] = counts.sum(axis=1)
            dist = counts[["Working","On Phone","Idle","Away"]].div(counts["total"].replace(0,1), axis=0)*100.0
            dist = dist.round(2)
            dist_df = counts.join(dist.add_suffix(" %")).reset_index().rename(columns={"employee":"Employee"})
            st.dataframe(dist_df, use_container_width=True, hide_index=True)
            emp_opts = list(dist_df["Employee"])
            if emp_opts:
                sel_emp = st.selectbox("Employee breakdown", emp_opts, key="emp_breakdown")
                rowb = dist_df[dist_df["Employee"]==sel_emp].iloc[0]
                chart_df = pd.DataFrame({
                    "Activity":["Working","On Phone","Idle","Away"],
                    "Percent":[rowb["Working %"],rowb["On Phone %"],rowb["Idle %"],rowb["Away %"]]
                })
                figb = px.bar(chart_df, x="Activity", y="Percent", range_y=[0,100])
                figb.update_layout(height=260, margin=dict(l=0,r=0,t=10,b=0),
                                   paper_bgcolor="rgba(0,0,0,0)", plot_bgcolor="rgba(0,0,0,0)")
                st.plotly_chart(figb, use_container_width=True, config={"displayModeBar": False})
        else:
            st.caption("No events in this run yet.")
    else:
        st.caption("Run a video to see per-employee distribution.")
    st.markdown("</div>", unsafe_allow_html=True)

    c1, c2, c3 = st.columns(3)
    with c1:
        if st.button("Clear detections (all)", disabled=READ_ONLY):
            st.session_state.events = st.session_state.events.iloc[0:0]
            st.session_state.frame_cache = {}
            st.session_state.frame_paths = {}
            st.session_state.emp_summary = {}
            st.session_state.emp_counters = {}
            st.session_state.face_demo_map = {}
            st.session_state.face_demo_used = set()
            st.session_state.daily_summary_path = ""
            st.session_state.on_phone_start_ns = {}
            st.session_state.on_phone_accum_ns = {}
            st.session_state.act_votes = {}
            st.session_state.prev_boxes = {}
            st.success("Cleared all detections.")
    if SHOW_DEV_CONTROLS:
        with c2:
            if st.button("Simulate stream error"):
                st.error("Stream lost, reconnecting…")
        with c3:
            if st.button("Simulate Superbse token expiry"):
                st.toast("Superbse auth expired.", icon="❌")

# ======================================================
# EMPLOYEES
# ======================================================
with tab_emp:
    left, right = st.columns([1,2])
    with left:
        st.markdown("<div class='card'>", unsafe_allow_html=True)
        q = st.text_input("Search employees", disabled=False)
        df_emp = pd.DataFrame([
            {"id": k, "name": v.get("name"), "title": v.get("title"), "desk": v.get("deskId"),
             "status": v.get("status"), "avatar": v.get("avatarUrl")}
            for k, v in EMP.items()
        ])
        if q:
            df_emp = df_emp[df_emp["name"].str.contains(q, case=False)]
        for _, row in df_emp.iterrows():
            c = st.container()
            cols = c.columns([0.2, 1, 0.6])
            with cols[0]: st.image(row["avatar"], width=40)
            with cols[1]: st.write(f"**{row['name']}**\n\n{row['title']} • {row['desk']}")
            with cols[2]:
                summ = st.session_state.emp_summary.get(row["name"], {})
                status = "At Desk"
                if summ.get("Away",0) > 0: status = "Away"
                badge = "pill-success" if status=="At Desk" else "pill-danger"
                st.markdown(f"<span class='pill {badge}'>{status}</span>", unsafe_allow_html=True)
        st.markdown("</div>", unsafe_allow_html=True)

    with right:
        st.markdown("<div class='card'>", unsafe_allow_html=True)
        if not df_emp.empty:
            active = df_emp.iloc[0]
            b1,b2,b3 = st.columns(3)
            with b1:
                st.button("Sync to Superbse", disabled=True if READ_ONLY else False)
            with b2:
                st.button("Open Employee in Superbse", disabled=True)
            with b3:
                st.button("Export Day as PDF", disabled=True if READ_ONLY else False)

            st.subheader(f"{active['name']} — {active['title']}")
            st.caption(f"Desk {active['desk']}")

            st.markdown("**Face Gallery**")
            frames = recent_frames_for_employee(active["name"], n=3)
            g1, g2, g3 = st.columns(3)
            cols = [g1,g2,g3]
            for i in range(3):
                with cols[i]:
                    if i < len(frames):
                        eid, img = frames[i]
                        st.image(img, use_column_width=True, caption=f"Event {eid}")
                    else:
                        st.image("https://placehold.co/160x120?text=Face", use_column_width=True)
                    st.button("Approve", key=f"approve:{active['name']}:{i}", disabled=True)
                    st.button("Reject",  key=f"reject:{active['name']}:{i}", disabled=True)

            st.markdown("**Today summary (live timers)**")
            c1,c2,c3,c4 = st.columns(4)
            t = st.session_state.emp_summary.get(active["name"], {})
            counters = st.session_state.emp_counters.get(active["name"], {"working_seconds":0.0,"idle_seconds":0.0})
            c1.metric("In / Out", "—")
            c2.metric("Working (sec)", fmt_mmss(counters.get("working_seconds",0.0)))
            c3.metric("Idle (sec)", fmt_mmss(counters.get("idle_seconds",0.0)))
            c4.metric("Corridor (frames)", int(t.get("Corridor",0)))
        else:
            st.caption("No employees in seed.json")
        st.markdown("</div>", unsafe_allow_html=True)

# ======================================================
# REPORTS
# ======================================================
with tab_reports:
    kpi_base = st.session_state.events
    if st.session_state.current_run_id:
        kpi_base = kpi_base[kpi_base.get("run_id").str.startswith(st.session_state.current_run_id, na=False)]
    KPI_LIST = kpis_from_events(kpi_base)

    kcols = st.columns(5)
    for i, (label, val) in enumerate(KPI_LIST):
        with kcols[i % 5]:
            st.markdown("<div class='card'>", unsafe_allow_html=True)
            st.write(label); st.subheader(val)
            st.markdown("</div>", unsafe_allow_html=True)

    dept_df = pd.DataFrame([
        {"dept":"Eng","Working":6.8,"Idle":0.9,"OnPhone":0.5,"Away":0.3},
        {"dept":"Ops","Working":7.2,"Idle":0.6,"OnPhone":0.4,"Away":0.2},
        {"dept":"HR","Working":6.3,"Idle":1.1,"OnPhone":0.3,"Away":0.4},
    ])
    c1,c2 = st.columns(2)
    with c1:
        st.markdown("<div class='card'>", unsafe_allow_html=True)
        st.write("By Department (Stacked)")
        dfm = dept_df.melt(id_vars=["dept"], var_name="state", value_name="hours")
        fig = px.bar(dfm, x="dept", y="hours", color="state", barmode="stack")
        fig.update_layout(height=320, margin=dict(l=0,r=0,t=0,b=0),
                          paper_bgcolor="rgba(0,0,0,0)", plot_bgcolor="rgba(0,0,0,0)")
        st.plotly_chart(fig, use_container_width=True, config={"displayModeBar": False})
        st.markdown("</div>", unsafe_allow_html=True)
    with c2:
        st.markdown("<div class='card'>", unsafe_allow_html=True)
        st.write("Working Hours (30 days) — Trend")
        trend = pd.DataFrame({"day": list(range(1,31)), "hours": [6.5 + np.sin(i/5)*0.7 + np.random.rand()*0.3 for i in range(30)]})
        fig2 = px.line(trend, x="day", y="hours")
        fig2.update_layout(height=320, margin=dict(l=0,r=0,t=0,b=0),
                           paper_bgcolor="rgba(0,0,0,0)", plot_bgcolor="rgba(0,0,0,0)")
        st.plotly_chart(fig2, use_container_width=True, config={"displayModeBar": False})
        st.markdown("</div>", unsafe_allow_html=True)

    st.markdown("<div class='card'>", unsafe_allow_html=True)
    st.subheader("Report Builder")
    rb1, rb2, rb3, rb4, rb5 = st.columns([1,1,1,1,2])
    with rb1: st.date_input("Date range start", key="reports_date_start")
    with rb2: st.date_input("Date range end", key="reports_date_end")
    with rb3: st.selectbox("Team", ["All","Eng","Ops","HR"], key="reports_team")
    with rb4: st.selectbox("Zone", ["All","Desk","Corridor","Lobby","Common"], key="reports_zone")
    with rb5: st.slider("Activity threshold (Idle min)", 0, 120, 30, key="reports_idle_thresh")

    rep = None
    if st.session_state.get("daily_summary_path") and os.path.exists(st.session_state.daily_summary_path):
        rep = pd.read_csv(st.session_state.daily_summary_path)
        if "Employee" in rep.columns and "Working" in rep.columns:
            rep = rep.copy()
            if "Corridor" not in rep.columns:
                rep["Corridor"] = 0
            rep["Alerts"] = 0
    else:
        rep = pd.DataFrame([
            {"Employee":"Aparna Rao","In":"09:12","Out":"18:07","Working":382,"Idle":46,"OnPhone":18,"Away":10,"Corridor":21,"Alerts":0},
            {"Employee":"Karthik S","In":"09:15","Out":"18:05","Working":368,"Idle":43,"OnPhone":22,"Away":12,"Corridor":19,"Alerts":1},
            {"Employee":"Nisha M","In":"09:20","Out":"18:10","Working":372,"Idle":40,"OnPhone":24,"Away":11,"Corridor":17,"Alerts":2},
        ])
    st.dataframe(rep, use_container_width=True, hide_index=True)

    d1, d2, d3 = st.columns([1,1,2])
    with d1:
        st.download_button("Export CSV", rep.to_csv(index=False).encode("utf-8"),
                           file_name="report.csv", mime="text/csv")
    with d2:
        if st.button("Export PDF", disabled=True if READ_ONLY else False):
            st.toast("PDF export queued (demo)", icon="📄")
    with d3:
        st.toggle("Schedule to Superbse", value=False, help="Creates/refreshes daily summary (mock)", disabled=True)

# -------------------- Daily detections CSV (append-only existing) --------------------
def persist_events_daily():
    df = st.session_state.events
    if df is None or df.empty:
        return
    new_count = len(df)
    if new_count <= st.session_state.persist_count:
        return
    today_str = date.today().strftime("%Y%m%d")
    file_path = f"/tmp/detections_{today_str}.csv"
    chunk = df.iloc[st.session_state.persist_count:].copy()
    chunk["ts"] = pd.to_datetime(chunk["ts"], errors="coerce", utc=True).astype(str)
    if os.path.exists(file_path):
        chunk.to_csv(file_path, mode="a", index=False, header=False)
    else:
        chunk.to_csv(file_path, index=False)
    st.session_state.persist_count = new_count
    st.session_state.persisted_csv = file_path

persist_events_daily()
if st.session_state.get("persisted_csv"):
    st.caption(f"📁 Appending to `{Path(st.session_state.persisted_csv).name}` in /tmp (auto every batch)")
st.markdown("<div class='footerpad'></div>", unsafe_allow_html=True)