Update app.py

app.py

@@ -1980,36 +1980,30 @@ if not df_category.empty:
     # st.markdown(insight_text, unsafe_allow_html=True)
 else:
     st.info("No data available for non-positive issue categories with 100% coverage and positive trend.")
-
-
-# =================== OBJECTIVE 7 — Insight and Recommendation (FINAL v2 — Sesuai Permintaan) ===================
+# =================== OBJECTIVE 7 — Insight and Recommendation (Agentic AI LLM Style — Final) ===================
 st.markdown("<h3 class='section-title'>OBJECTIVE 7 — Insight and Recommendation</h3>", unsafe_allow_html=True)
 
-def extract_deviations_v2(df: pd.DataFrame):
+def extract_agentic_insights_v4(df: pd.DataFrame):
     dev = {
-        # Obj 2: lokasi dengan avg ratio ≈ 1.0
-        "obj2_optimal_locs": [],
-        # Obj 3a & 3c: rasio temuan/orang TERENDAH (divisi & reporter)
-        "obj3a_lowest_div_ratio": None,   # ('Div A', 0.12)
-        "obj3c_lowest_reporter_ratio": None,  # ('Ali', 0.05)
-        # Obj 3b & 3d: lead time TERPANJANG (divisi & eksekutor)
-        "obj3b_slowest_div_lead": None,   # ('Div B', 12.4)
-        "obj3d_slowest_executor_lead": None,  # ('Budi', 15.2)
-        # Obj 4: unsafe share
-        "obj4_uc": 0.0,
-        "obj4_ua": 0.0,
-        "obj4_nm": 0.0,
-        # Obj 6 (Panel 4): top 2 non-Positive categories (avg/month tertinggi)
-        "obj6_top2_cat": [],
-        # Obj 5: Quadrant I & II
+        # 1. 9 lokasi rasio TERENDAH
+        "lowest_ratio_9_locs": [],
+        # 2a–2d: dari Obj 3a–3d
+        "obj3a_lowest_div": None,
+        "obj3b_slowest_executor": None,
+        "obj3c_lowest_reporter": None,
+        "obj3d_slowest_div": None,
+        # 3. Non-Positive composition
+        "obj4_unsafe_condition_pct": 0.0,
+        "obj4_unsafe_action_pct": 0.0,
+        "obj4_near_miss_pct": 0.0,
+        # 4. Quadrant I & II
         "obj5_q1_divs": [],
         "obj5_q2_divs": [],
-        # 🔥 BARU: Obj 3 (predictive panel 1–3): trend slope TERENDAH (bukan avg ratio!)
-        "obj3_most_declining_locs": [],   # 5 lokasi dengan slope paling negatif
-        "obj3_most_declining_divs": [],   # 5 divisi dengan slope paling negatif
+        # 5. Top 2 non-Positive categories
+        "obj6_top2_categories": [],
     }
 
-    # ========= 1. Optimal Ratio (~1.0) dari Objective 2 =========
+    # === 1. 9 lokasi rasio terendah ===
     if {'nama_lokasi_full', 'creator_nid', 'created_at', 'kode_temuan'}.issubset(df.columns):
         calc = df[['nama_lokasi_full', 'creator_nid', 'created_at', 'kode_temuan']].copy()
         calc['created_at'] = pd.to_datetime(calc['created_at'], errors='coerce')
@@ -2022,11 +2016,10 @@ def extract_deviations_v2(df: pd.DataFrame):
         monthly = monthly[monthly['reporters'] > 0]
         monthly['ratio'] = monthly['findings'] / monthly['reporters']
         loc_avg = monthly.groupby('nama_lokasi_full')['ratio'].mean()
-        near_1 = loc_avg[(loc_avg >= 0.95) & (loc_avg <= 1.05)].nlargest(9)
-        dev["obj2_optimal_locs"] = near_1.index.tolist()
+        lowest_9 = loc_avg.nsmallest(9)
+        dev["lowest_ratio_9_locs"] = [(loc, round(ratio, 3)) for loc, ratio in lowest_9.items()]
 
-    # ========= 2. TERENDAH — Rasio Temuan/Orang (Objective 3a & 3c) =========
-    # 3a: divisi — rasio temuan/orang
+    # === 2a: divisi — rasio temuan/orang terendah (Obj 3a) ===
     if {'nama', 'creator_nid', 'created_at', 'kode_temuan'}.issubset(df.columns):
         calc = df[['nama', 'creator_nid', 'created_at', 'kode_temuan']].copy()
         calc['bulan'] = pd.to_datetime(calc['created_at']).dt.to_period('M')
@@ -2040,9 +2033,20 @@ def extract_deviations_v2(df: pd.DataFrame):
         if not div_ratio.empty:
             name = div_ratio.idxmin()
             val = round(div_ratio.min(), 2)
-            dev["obj3a_lowest_div_ratio"] = (name, val)
+            dev["obj3a_lowest_div"] = (name, val)
 
-    # 3c: individu — avg temuan/bulan
+    # === 2b: eksekutor — lead time terpanjang (Obj 3b) ===
+    if 'days_to_close' in df.columns:
+        valid = df[df['days_to_close'].notna() & (df['days_to_close'] >= 0)]
+        exec_col = 'nama_pic' if 'nama_pic' in valid.columns else 'creator_name'
+        if exec_col in valid.columns:
+            lead = valid.groupby(exec_col)['days_to_close'].mean()
+            if not lead.empty:
+                name = lead.idxmax()
+                val = round(lead.max(), 1)
+                dev["obj3b_slowest_executor"] = (name, val)
+
+    # === 2c: reporter — frekuensi terendah (Obj 3c) ===
     if {'creator_name', 'created_at'}.issubset(df.columns):
         calc = df[['creator_name', 'created_at']].copy()
         calc['bulan'] = pd.to_datetime(calc['created_at']).dt.to_period('M')
@@ -2052,47 +2056,26 @@ def extract_deviations_v2(df: pd.DataFrame):
         if not avg.empty:
             name = avg.idxmin()
             val = round(avg.min(), 2)
-            dev["obj3c_lowest_reporter_ratio"] = (name, val)
+            dev["obj3c_lowest_reporter"] = (name, val)
 
-    # ========= 3. TERPANJANG — Lead Time (Objective 3b & 3d) =========
-    if 'days_to_close' in df.columns:
+    # === 2d: divisi — lead time terpanjang (Obj 3d) ===
+    if 'days_to_close' in df.columns and 'nama' in df.columns:
         valid = df[df['days_to_close'].notna() & (df['days_to_close'] >= 0)]
-        # 3b: divisi
-        if 'nama' in valid.columns:
+        if not valid.empty:
             lead = valid.groupby('nama')['days_to_close'].mean()
             if not lead.empty:
                 name = lead.idxmax()
                 val = round(lead.max(), 1)
-                dev["obj3b_slowest_div_lead"] = (name, val)
-        # 3d: eksekutor (deteksi otomatis kolom)
-        exec_col = next((c for c in ['nama_pic', 'pic', 'responsible', 'creator_name'] if c in valid.columns), None)
-        if exec_col:
-            lead = valid.groupby(exec_col)['days_to_close'].mean()
-            if not lead.empty:
-                name = lead.idxmax()
-                val = round(lead.max(), 1)
-                dev["obj3d_slowest_executor_lead"] = (name, val)
+                dev["obj3d_slowest_div"] = (name, val)
 
-    # ========= 4. Unsafe Share (Objective 4) =========
+    # === 3. Komposisi non-Positive ===
     if 'temuan_kategori' in df.columns:
-        total = len(df)
-        if total > 0:
-            cnt = df['temuan_kategori'].value_counts(normalize=True) * 100
-            dev["obj4_uc"] = round(cnt.get("Unsafe Condition", 0), 1)
-            dev["obj4_ua"] = round(cnt.get("Unsafe Action", 0), 1)
-            dev["obj4_nm"] = round(cnt.get("Near Miss", 0), 1)
-
-    # ========= 5. Top 2 Bubble: Category Non-Positive (Objective 6 Panel 4) =========
-    if {'kategori', 'temuan_kategori', 'created_at'}.issubset(df.columns):
-        nonpos = df[df['temuan_kategori'] != 'Positive']
-        if not nonpos.empty:
-            start = nonpos['created_at'].min().to_period('M')
-            end = nonpos['created_at'].max().to_period('M')
-            n_months = len(pd.period_range(start=start, end=end, freq='M'))
-            cat_avg = (nonpos.groupby('kategori').size() / n_months).sort_values(ascending=False).head(2)
-            dev["obj6_top2_cat"] = [(cat, round(val, 1)) for cat, val in cat_avg.items()]
+        cnt = df['temuan_kategori'].value_counts(normalize=True) * 100
+        dev["obj4_unsafe_condition_pct"] = round(cnt.get("Unsafe Condition", 0), 1)
+        dev["obj4_unsafe_action_pct"] = round(cnt.get("Unsafe Action", 0), 1)
+        dev["obj4_near_miss_pct"] = round(cnt.get("Near Miss", 0), 1)
 
-    # ========= 6. Quadrant I & II (Objective 5) =========
+    # === 4. Kuadran Risiko (X=20, Y=3) ===
     X_LIMIT, Y_LIMIT = 20, 3
     if {'nama', 'created_at', 'days_to_close', 'kode_temuan'}.issubset(df.columns):
         calc = df.copy()
@@ -2108,192 +2091,155 @@ def extract_deviations_v2(df: pd.DataFrame):
             elif r['Finding Count'] < X_LIMIT and r['Avg Lead Time'] >= Y_LIMIT:
                 dev["obj5_q2_divs"].append(r['nama'])
 
-    # ========= 🔥 7. TREND TERENDAH — Slope paling negatif (Objective 6 Panels 1–3) =========
-    # Lokasi (Panel 2)
-    if {'nama_lokasi_full', 'created_at'}.issubset(df.columns):
-        start = df['created_at'].min().to_period('M')
-        end = df['created_at'].max().to_period('M')
-        all_months = pd.period_range(start=start, end=end, freq='M')
-        monthly = (
-            df.groupby(['nama_lokasi_full', df['created_at'].dt.to_period('M')])
-            .size()
-            .unstack(fill_value=0)
-            .reindex(columns=all_months, fill_value=0)
-        )
-        slopes = {}
-        for loc in monthly.index:
-            ts = monthly.loc[loc].values
-            if len(ts) >= 2:
-                try:
-                    slope = np.polyfit(range(len(ts)), ts, 1)[0]
-                    slopes[loc] = slope
-                except:
-                    continue
-        top5 = sorted(slopes.items(), key=lambda x: x[1])[:5]  # paling negatif
-        dev["obj3_most_declining_locs"] = [loc for loc, _ in top5]
-
-    # Divisi (Panel 3)
-    if {'nama', 'created_at'}.issubset(df.columns):
-        monthly = (
-            df.groupby(['nama', df['created_at'].dt.to_period('M')])
-            .size()
-            .unstack(fill_value=0)
-            .reindex(columns=all_months, fill_value=0)
-        )
-        slopes = {}
-        for div in monthly.index:
-            ts = monthly.loc[div].values
-            if len(ts) >= 2:
-                try:
-                    slope = np.polyfit(range(len(ts)), ts, 1)[0]
-                    slopes[div] = slope
-                except:
-                    continue
-        top5 = sorted(slopes.items(), key=lambda x: x[1])[:5]
-        dev["obj3_most_declining_divs"] = [div for div, _ in top5]
+    # === 5. Top 2 non-Positive category (Obj 6) ===
+    if {'kategori', 'temuan_kategori', 'created_at'}.issubset(df.columns):
+        nonpos = df[df['temuan_kategori'] != 'Positive']
+        if not nonpos.empty:
+            start = nonpos['created_at'].min().to_period('M')
+            end = nonpos['created_at'].max().to_period('M')
+            n_months = len(pd.period_range(start=start, end=end, freq='M'))
+            cat_avg = (nonpos.groupby('kategori').size() / n_months).sort_values(ascending=False).head(2)
+            dev["obj6_top2_categories"] = [(cat, round(val, 1)) for cat, val in cat_avg.items()]
 
     return dev
 
 # Ekstrak
-dev = extract_deviations_v2(df_filtered)
-
-# ======== INSIGHT SUMMARY (Revised Penomoran) ========
-insight_parts = []
-
-# 1 → dari Obj 2
-if dev["obj2_optimal_locs"]:
-    locs = ", ".join(dev["obj2_optimal_locs"][:5])
-    insight_parts.append("1. Sembilan lokasi menunjukkan rasio temuan/orang optimal (~1,0), indikasi beban kerja seimbang dan pelaporan konsisten: {} dan 4 lainnya.".format(locs))
-
-# 1a → reporter/Divisi TERENDAH (Obj 3a & 3c)
-low_div = dev["obj3a_lowest_div_ratio"]
-low_rep = dev["obj3c_lowest_reporter_ratio"]
-if low_div and low_rep:
-    insight_parts.append(
-        f"   1a. Namun, aktivitas inspeksi terendah terjadi di: • Divisi <strong>{low_div[0]}</strong> (rasio {low_div[1]}), "
-        f"• Reporter <strong>{low_rep[0]}</strong> ({low_rep[1]} temuan/bulan), menunjukkan kemungkinan under-reporting atau hambatan akses."
+dev = extract_agentic_insights_v4(df_filtered)
+
+# 📝 INSIGHT TEXT — STRUKTUR FINAL
+insight_lines = []
+
+# 1. 9 lokasi rasio terendah (semua ditampilkan)
+if dev["lowest_ratio_9_locs"]:
+    loc_list = ", ".join([f"<strong>{loc}</strong> ({ratio})" for loc, ratio in dev["lowest_ratio_9_locs"]])
+    insight_lines.append(f"1. Sembilan lokasi dengan rasio temuan/orang *terendah* (<0.5): {loc_list}.")
+
+# 2. ✅ AGENTIC AI SUMMARY dari 2a–2d (BUKAN insight umum biasa)
+# → Ini adalah kunci perbaikan: summary berbasis *capacity-agency-risk mismatch*
+summary_parts = []
+if dev["obj3a_lowest_div"]:
+    summary_parts.append(f"divisi {dev['obj3a_lowest_div'][0]} (rasio {dev['obj3a_lowest_div'][1]})")
+if dev["obj3c_lowest_reporter"]:
+    summary_parts.append(f"reporter {dev['obj3c_lowest_reporter'][0]} ({dev['obj3c_lowest_reporter'][1]}/bln)")
+if dev["obj3d_slowest_div"]:
+    summary_parts.append(f"divisi {dev['obj3d_slowest_div'][0]} ({dev['obj3d_slowest_div'][1]} hari)")
+if dev["obj3b_slowest_executor"]:
+    summary_parts.append(f"eksekutor {dev['obj3b_slowest_executor'][0]} ({dev['obj3b_slowest_executor'][1]} hari)")
+
+if summary_parts:
+    joined = "; ".join(summary_parts)
+    insight_lines.append(
+        "2. Sistem mendeteksi *agency-capacity mismatch*: "
+        f"<strong>{joined}</strong>. "
+        "Ini menunjukkan ketidakseimbangan antara kapasitas pelaporan dan kapasitas resolusi — dua dimensi kritis dalam ekosistem *agentic safety*."
     )
 
-# 1b → lead time TERPANJANG (Obj 3b & 3d)
-slow_div = dev["obj3b_slowest_div_lead"]
-slow_exe = dev["obj3d_slowest_executor_lead"]
-if slow_div and slow_exe:
-    insight_parts.append(
-        f"   1b. Kapasitas resolusi terendah di: • Divisi <strong>{slow_div[0]}</strong> ({slow_div[1]} hari), "
-        f"• Eksekutor <strong>{slow_exe[0]}</strong> ({slow_exe[1]} hari), berisiko SLA breach."
-    )
+# 2a–2d: detail
+if dev["obj3a_lowest_div"]:
+    name, val = dev["obj3a_lowest_div"]
+    insight_lines.append(f"   2a. Divisi <strong>{name}</strong> memiliki rasio temuan/orang terendah ({val}), indikasi *low agency* atau *high reporting barrier*.")
 
-# 1c → unsafe share (Obj 4)
-uc, ua, nm = dev["obj4_uc"], dev["obj4_ua"], dev["obj4_nm"]
-if uc > 0 or ua > 0 or nm > 0:
-    insight_parts.append(
-        f"   1c. Komposisi temuan non-Positive: Unsafe Condition ({uc}%), Unsafe Action ({ua}%), Near Miss ({nm}%)."
-    )
+if dev["obj3b_slowest_executor"]:
+    name, days = dev["obj3b_slowest_executor"]
+    insight_lines.append(f"   2b. Eksekutor <strong>{name}</strong> memiliki lead time resolusi tertinggi ({days} hari), berisiko menjadi *bottleneck agent* dalam alur penutupan.")
 
-# 1d → trend paling menurun (NEW!)
-decl_locs = dev["obj3_most_declining_locs"]
-decl_divs = dev["obj3_most_declining_divs"]
-if decl_locs and decl_divs:
-    top_loc = decl_locs[0]
-    top_div = decl_divs[0]
-    insight_parts.append(
-        f"   1d. Lokasi <strong>{top_loc}</strong> dan Divisi <strong>{top_div}</strong> memiliki tren aktivitas inspeksi paling menurun (slope negatif), "
-        f"mengindikasikan potensi penurunan komitmen atau pergantian pelapor kunci."
-    )
+if dev["obj3c_lowest_reporter"]:
+    name, rate = dev["obj3c_lowest_reporter"]
+    insight_lines.append(f"   2c. Reporter <strong>{name}</strong> memiliki frekuensi pelaporan terendah ({rate}/bulan), perlu *capacity uplift* melalui *nudge* atau *coaching*.")
+
+if dev["obj3d_slowest_div"]:
+    name, days = dev["obj3d_slowest_div"]
+    insight_lines.append(f"   2d. Divisi <strong>{name}</strong> memiliki lead time rata-rata tertinggi ({days} hari), menunjukkan *systemic delay* dalam *execution loop*.")
 
-# 2 → top 2 bubble (Obj 6)
-if dev["obj6_top2_cat"]:
-    c1, c2 = dev["obj6_top2_cat"]
-    insight_parts.append(f"2. Dua kategori paling sering muncul (non-Positive): <strong>{c1[0]}</strong> ({c1[1]}/bulan) dan <strong>{c2[0]}</strong> ({c2[1]}/bulan).")
+# 3. Komposisi non-Positive
+uc, ua, nm = dev["obj4_unsafe_condition_pct"], dev["obj4_unsafe_action_pct"], dev["obj4_near_miss_pct"]
+if uc + ua + nm > 0:
+    insight_lines.append(f"3. Komposisi non-Positive: Unsafe Condition ({uc}%), Unsafe Action ({ua}%), Near Miss ({nm}%). Proporsi Near Miss masih rendah — indikasi *under-reporting of close calls*.")
 
-# 3 → quadrant I & II (Obj 5)
-q1 = dev["obj5_q1_divs"]
-q2 = dev["obj5_q2_divs"]
-if q1:
-    insight_parts.append(f"3. Divisi risiko tinggi (volume tinggi + lead time tinggi / Quadrant I): {', '.join(q1[:3])}.")
-    if q2:
-        insight_parts.append(f"   3a. Divisi risiko tersembunyi (volume rendah + lead time tinggi / Quadrant II): {', '.join(q2[:3])}.")
+# 4. Kuadran risiko
+if dev["obj5_q1_divs"] or dev["obj5_q2_divs"]:
+    q1 = ", ".join([f"<strong>{d}</strong>" for d in dev["obj5_q1_divs"][:3]])
+    q2 = ", ".join([f"<strong>{d}</strong>" for d in dev["obj5_q2_divs"][:3]])
+    insight_lines.append(f"4. Divisi risiko tinggi (Kuadran I): {q1 if q1 else '—'}. Divisi risiko tersembunyi (Kuadran II): {q2 if q2 else '—'}.")
 
-# 4 → sintesis
-insight_parts.append("4. Pola utama: terjadi *mismatch* antara kapasitas inspeksi (menurun di beberapa area) dan kapasitas resolusi (terlalu lambat), menyebabkan hazard tetap terbuka dalam jangka waktu lama.")
+# 5. Top 2 kategori
+if dev["obj6_top2_categories"]:
+    c1, c2 = dev["obj6_top2_categories"]
+    insight_lines.append(f"5. Dua kategori non-Positive paling sering: <strong>{c1[0]}</strong> ({c1[1]}/bulan) dan <strong>{c2[0]}</strong> ({c2[1]}/bulan).")
 
-insight_text = "<br>".join(insight_parts)
+insight_text = "<br>".join(insight_lines)
 
-# ======== REKOMENDASI & MITIGASI ========
-recoms = []
+# 🔔 REKOMENDASI — AGENTIC AI STYLE (per poin, actionable, closed-loop)
+recs = []
 
-# 1 → benchmark lokasi optimal
-if dev["obj2_optimal_locs"]:
-    recoms.append({
+# 1 → 9 lokasi terendah
+if dev["lowest_ratio_9_locs"]:
+    recs.append({
         "point": "1",
-        "rec": "Jadikan 9 lokasi ber-rasio ~1,0 sebagai benchmark operasional: dokumentasikan rotasi pelapor, checklist, dan frekuensi.",
-        "mit": "Tetapkan *tolerance band* rasio 0,8–1,2. Jika keluar band ≥2 bulan → picu *capacity review* otomatis."
+        "rec": "Luncurkan *Agency Activation Sprint* di 9 lokasi terendah: PIC Area wajib lakukan 1 inspeksi spot/minggu dan catat dalam sistem.",
+        "mit": "Aktifkan *QR code checklist* 3-menit + notifikasi WhatsApp otomatis setiap Senin. Target: rasio ≥0.5 dalam 45 hari."
     })
 
-# 1a → tingkatkan inspeksi di area rendah
-if low_div and low_rep:
-    recoms.append({
-        "point": "1a",
-        "rec": "Luncurkan *micro-inspection sprint* di divisi & reporter terendah: target minimal 0,5 temuan/bulan/orang dalam 30 hari.",
-        "mit": "Aktifkan *QR code checklist* 3-menit di lokasi kritis. Integrasi reward points di aplikasi mobile."
+# 2 → summary agentic
+if summary_parts:
+    recs.append({
+        "point": "2",
+        "rec": "Integrasikan *Agentic Capacity Dashboard*: real-time monitoring rasio & lead time per individu/divisi. Jika deviasi >20% dari baseline → picu *auto-coaching alert*.",
+        "mit": "Sistem akan mengirim notifikasi ke PIC Area & Safety Manager, disertai rekomendasi tindakan berbasis best practice dari division benchmark (misal: divisi dengan rasio ~1.0)."
     })
 
-# 1b → perbaiki lead time
-if slow_div and slow_exe:
-    recoms.append({
-        "point": "1b",
-        "rec": "Bentuk *Rapid Closure Team* khusus untuk divisi & eksekutor terlambat. Terapkan *photo verification* & *auto-escalation*.",
-        "mit": "Terapkan SLA berlapis: High Risk ≤3 hari, Medium/Low ≤7 hari. >7 hari → notifikasi ke Daily Huddle & PIC Area."
+# 2a–2d → rekomendasi individual
+if dev["obj3a_lowest_div"]:
+    recs.append({
+        "point": "2a",
+        "rec": "Assign *Safety Buddy* dari divisi berkinerja tinggi ke divisi terendah untuk 2 minggu *shadowing & transfer knowledge*.",
+        "mit": "Integrasikan *micro-reporting goal*: 1 temuan/minggu/orang. Sistem otomatis lacak progres & kirim apresiasi digital tiap goal tercapai."
     })
-
-# 1c → tingkatkan Near Miss & Positive
-if uc > 0 or ua > 0 or nm > 0:
-    recoms.append({
-        "point": "1c",
-        "rec": "Ubah alur pelaporan default: wajibkan 1 temuan Near Miss atau Positive per Unsafe Condition.",
-        "mit": "Aktifkan *Positive Hunt Challenge* bulanan: reward terbaik berdasarkan kualitas & frekuensi temuan positif."
+if dev["obj3b_slowest_executor"]:
+    recs.append({
+        "point": "2b",
+        "rec": "Aktifkan *Rapid Closure Protocol* untuk eksekutor terlambat: verifikasi via foto + approval satu tingkat → target ≤3 hari.",
+        "mit": "Jika >5 hari, sistem auto-escalate ke Daily Safety Huddle & PIC Area via Telegram."
     })
-
-# 1d → mitigasi penurunan tren
-if decl_locs and decl_divs:
-    recoms.append({
-        "point": "1d",
-        "rec": "Identifikasi *root cause* penurunan tren: apakah perubahan PIC, rotasi, atau kelelahan? Lakukan *exit interview safety* bagi pelapor yang keluar.",
-        "mit": "Terapkan *trend alert*: jika slope < -0.1 selama 2 bulan → notifikasi ke Safety Manager & PIC Area."
+if dev["obj3c_lowest_reporter"]:
+    recs.append({
+        "point": "2c",
+        "rec": "Terapkan *Nudge Agent Program*: PIC lokasi bertanggung jawab memastikan semua anggota tim minimal 1x inspeksi/minggu.",
+        "mit": "Reward: point redeemable di kantin. Sistem kirim reminder 2x/minggu + contoh temuan sederhana (e.g., APD tidak lengkap)."
     })
-
-# 2 → top category RCA
-if dev["obj6_top2_cat"]:
-    c1, c2 = dev["obj6_top2_cat"]
-    recoms.append({
-        "point": "2",
-        "rec": "Lakukan *Cross-Functional RCA* untuk <strong>{}</strong> & <strong>{}</strong>, libatkan desain, kontraktor, dan operasi.".format(c1[0], c2[0]),
-        "mit": "Perbarui spesifikasi teknis: wajibkan mitigasi berbasis temuan historis sebelum tender dimulai."
+if dev["obj3d_slowest_div"]:
+    recs.append({
+        "point": "2d",
+        "rec": "Lakukan *Process Mining*: petakan alur temuan dari buka → tutup, identifikasi *waste* (waiting, rework, approval bottleneck).",
+        "mit": "Terapkan SLA berlapis: High Risk ≤3 hari, Medium/Low ≤7 hari. Jika breach ≥2x/bulan → picu *capacity intervention* otomatis."
     })
 
-# 3 → Quadrant I & II
-if q1:
-    recoms.append({
+# 3 → Near Miss
+if uc + ua + nm > 0:
+    recs.append({
         "point": "3",
-        "rec": "Alokasikan *dedicated crew* (2 orang) hanya untuk menutup temuan Quadrant I.",
-        "mit": "Jika divisi masuk Quadrant I ≥2 bulan → eskalasi ke VP Operasi & PIC Area."
+        "rec": "Luncurkan *Near Miss Amplifier*: setiap laporan Unsafe Condition wajib diikuti minimal 1 Near Miss terkait.",
+        "mit": "Sistem blokir submit jika tidak memenuhi. Tampilkan *Near Miss Leaderboard* di dashboard tim."
     })
-if q2:
-    recoms.append({
-        "point": "3a",
-        "rec": "Terapkan *One Finding, One Day* untuk Quadrant II: semua temuan wajib selesai ≤24 jam.",
-        "mit": "Jadikan *100% closure <24 jam* sebagai KPI supervisor. Reward: zero-backlog selama 1 bulan."
+
+# 4 → Kuadran
+if dev["obj5_q1_divs"] or dev["obj5_q2_divs"]:
+    recs.append({
+        "point": "4",
+        "rec": "Alokasikan *dedicated safety crew* untuk Kuadran I; terapkan *One Finding, One Day* untuk Kuadran II.",
+        "mit": "Jika masuk Kuadran I/II ≥2 bulan berturut-turut → sistem auto-generate *executive escalation report* ke VP Operasi."
     })
 
-# 4 → integrasi sistem
-recoms.append({
-    "point": "4",
-    "rec": "Integrasikan modul *capacity validation* ke dalam penjadwalan inspeksi: verifikasi lead time historis sebelum jadwal dibuat.",
-    "mit": "Bangun *closed-loop safety system*: setiap temuan baru harus diverifikasi *capacity to close* terlebih dahulu."
-})
+# 5 → Top 2 kategori
+if dev["obj6_top2_categories"]:
+    c1, c2 = dev["obj6_top2_categories"]
+    recs.append({
+        "point": "5",
+        "rec": f"Bentuk *RCA Task Force* lintas fungsi (SIPIL, ELEKTRIKAL, K3, Kontraktor) untuk {c1[0]} & {c2[0]}.",
+        "mit": "Revisi spesifikasi teknis & template tender: semua penawaran wajib menyertakan mitigasi berbasis temuan historis."
+    })
 
-# ======== TAMPILKAN ========
-# Insight Card
+# 🖼️ TAMPILKAN
 st.markdown(
     f"""
     <div class="card" style="
@@ -2304,17 +2250,17 @@ st.markdown(
         border-radius: 4px;
         box-shadow: 0 2px 4px rgba(0,0,0,0.05);
     ">
-        <h4 style="margin-top: 0; color: #003DA5;">Insight Summary</h4>
-        <p style="margin-bottom: 0; line-height: 1.6;">{insight_text}</p>
+        <h4 style="margin-top: 0; color: #003DA5;">Insight Summary (Agentic AI LLM Style)</h4>
+        <p style="margin-bottom: 0; line-height: 1.6; font-size: 0.98em;">{insight_text}</p>
     </div>
     """,
     unsafe_allow_html=True
 )
 
-# Rekomendasi Table
-if recoms:
+# Tabel rekomendasi
+if recs:
     rows = []
-    for r in recoms:
+    for r in recs:
         rows.append(
             f"<tr>"
             f"<td style='text-align:center; font-weight:bold; width:5%;'>{r['point']}</td>"
@@ -2331,7 +2277,7 @@ if recoms:
         border-radius: 4px;
         box-shadow: 0 2px 4px rgba(0,0,0,0.05);
     ">
-        <h4 style="margin-top: 0; color: #2E7D32;">Recommended Actions and Risk Mitigation Strategy</h4>
+        <h4 style="margin-top: 0; color: #2E7D32;">Recommended Actions & Agentic Risk Mitigation</h4>
         <table style="width:100%; border-collapse:collapse; font-size:0.95em; margin-top:12px;">
             <thead>
                 <tr style="background-color:#e8f5ee;">