Update app.py

app.py

@@ -2,11 +2,20 @@
 """
 IPLM 2025 — Final (Target Sampel 33.88% per Jenis) — TANPA Kinerja Relatif / Percentile
 
-✅ Ringkasan (Jenis + Keseluruhan) kini MEMUAT:
-- Rata2_sub_koleksi, Rata2_sub_sdm, Rata2_sub_pelayanan, Rata2_sub_pengelolaan
-- Rata2_dim_kepatuhan, Rata2_dim_kinerja
-
-✅ Tabel Ringkasan tsb menjadi RUJUKAN UTAMA LLM untuk narasi laporan Word.
+KONSEP INTI
+A) Level entitas:
+   Yeo-Johnson per indikator -> MinMax global (0–1) -> sub -> dim -> Indeks_Dasar_0_100
+B) Penyesuaian kecukupan sampel (target 33.88% per jenis):
+   faktor_penyesuaian_jenis = min(n_jenis / target_total_jenis, 1.0)
+   Indeks_Final_Agregat_0_100 = Indeks_Dasar_Agregat_0_100 * faktor_penyesuaian_jenis
+C) Keseluruhan (FIX avg3):
+   indeks keseluruhan = (sekolah + umum + khusus)/3, missing=0 tetap /3
+D) Dashboard KPI:
+   hanya 2 kartu: Indeks Final & Indeks Dasar (tanpa coverage KPI)
+E) Analisis LLM:
+   wajib merujuk TABEL Ringkasan (Jenis + Keseluruhan),
+   netral-deskriptif, wajib kutip angka, tanpa label normatif,
+   dan memuat tindakan teknis untuk menaikkan indeks (berbasis pola angka).
 """
 
 import os
@@ -14,6 +23,7 @@ import re
 import time
 import tempfile
 from pathlib import Path
+from typing import List, Optional
 
 import gradio as gr
 import numpy as np
@@ -42,7 +52,7 @@ except Exception:
 # 1) KONFIGURASI
 # ============================================================
 
-DATA_FILE   = os.getenv("DATA_FILE",   "DATA CLEAN GABUNGAN SANGGAH-TIDAK SANGGAH - ALL190226.xlsx")
+DATA_FILE   = os.getenv("DATA_FILE",   "DATA CLEAN GABUNGAN SANGGAH-TIDAK SANGGAH - ALLL.xlsx")
 POP_KAB     = os.getenv("POP_KAB",     "Data_populasi_Kab_kota_fixed.xlsx")
 POP_PROV    = os.getenv("POP_PROV",    "Data_populasi_propinsi.xlsx")
 POP_KHUSUS  = os.getenv("POP_KHUSUS",  "Data_populasi_perp_khusus.xlsx")
@@ -79,7 +89,7 @@ def _disp_text(x):
     t = str(x).strip().upper()
     return " ".join(t.split())
 
-def pick_col(df, candidates):
+def pick_col(df, candidates: List[str]) -> Optional[str]:
     if df is None or df.empty:
         return None
     for c in candidates:
@@ -101,6 +111,7 @@ def coerce_num(val):
     t = t.replace("\u00a0", " ").replace("Rp", "").replace("%", "")
     t = re.sub(r"[^0-9,.\-]", "", t)
 
+    # smart decimal
     if t.count(".") > 1 and t.count(",") == 1:
         t = t.replace(".", "").replace(",", ".")
     elif t.count(",") > 1 and t.count(".") == 1:
@@ -179,26 +190,16 @@ def safe_div(num, den):
     return float(num) / float(den)
 
 def faktor_penyesuaian_total(n_total: float, target_total: float) -> float:
+    """
+    faktor = min(n / target, 1.0)
+    - Jika target <= 0 -> default 1.0 (tidak menghukum)
+    """
     if target_total is None or pd.isna(target_total) or float(target_total) <= 0:
         return 1.0
     if n_total is None or pd.isna(n_total) or float(n_total) < 0:
         n_total = 0.0
     return float(min(float(n_total) / float(target_total), 1.0))
 
-def df_to_markdown_table(df: pd.DataFrame, max_rows: int = 50) -> str:
-    if df is None or df.empty:
-        return "(tabel kosong)"
-    d = df.copy()
-    if len(d) > max_rows:
-        d = d.head(max_rows)
-    for c in d.columns:
-        if pd.api.types.is_float_dtype(d[c]):
-            d[c] = d[c].map(lambda x: "" if pd.isna(x) else f"{float(x):.3f}")
-    try:
-        return d.to_markdown(index=False)
-    except Exception:
-        return d.to_string(index=False)
-
 
 # ============================================================
 # 3) INDIKATOR IPLM
@@ -272,11 +273,20 @@ def _mean_norm_cols(row, cols):
     return float(np.mean(vals)) if vals else 0.0
 
 def prepare_global(df_src: pd.DataFrame) -> pd.DataFrame:
+    """
+    Transform + normalisasi indikator pada level entitas:
+    - rename kolom indikator (alias)
+    - coerce numeric
+    - Yeo-Johnson per indikator (standardize=False)
+    - MinMax global 0-1
+    - hitung sub_*, dim_*, Indeks_Dasar_0_100
+    """
     if df_src is None or df_src.empty:
         return df_src
 
     df = df_src.copy()
 
+    # rename indikator
     rename_map = {}
     for col in df.columns:
         c = _canon(col)
@@ -294,6 +304,7 @@ def prepare_global(df_src: pd.DataFrame) -> pd.DataFrame:
     for c in available:
         df[c] = df[c].apply(coerce_num)
 
+    # YJ per indikator + MinMax global
     for c in available:
         x = pd.to_numeric(df[c], errors="coerce").astype(float).values
         mask = ~np.isnan(x)
@@ -325,9 +336,27 @@ def prepare_global(df_src: pd.DataFrame) -> pd.DataFrame:
 # 5) CACHE LOADER (NO UPLOAD)
 # ============================================================
 
-_CACHE = {"key": None, "df_all": None, "df_raw": None, "pop_kab": None, "pop_prov": None, "pop_khusus": None, "meta": None, "info": None}
+_CACHE = {
+    "key": None,
+    "df_all": None,
+    "df_raw": None,
+    "pop_kab": None,
+    "pop_prov": None,
+    "pop_khusus": None,
+    "meta": None,
+    "info": None
+}
 
 def _parse_pop_khusus(path_xlsx: str) -> pd.DataFrame:
+    """
+    POP_KHUSUS format campuran:
+    - Baris 'PROVINSI X' -> level PROV
+    - Baris berikutnya -> KAB/KOTA dibawah prov tsb
+    Output:
+      LEVEL: PROV / KAB
+      prov_key / kab_key
+      Pop_Total_Jenis
+    """
     df = pd.read_excel(path_xlsx)
     if df is None or df.empty:
         return pd.DataFrame()
@@ -349,6 +378,7 @@ def _parse_pop_khusus(path_xlsx: str) -> pd.DataFrame:
 
     rows = []
     current_prov = None
+
     for m, pval in zip(mix.tolist(), pop_series.tolist()):
         mm = _disp_text(m) or ""
         if mm == "":
@@ -357,10 +387,20 @@ def _parse_pop_khusus(path_xlsx: str) -> pd.DataFrame:
         if mm.startswith("PROVINSI "):
             prov_name = mm.replace("PROVINSI", "").strip()
             current_prov = prov_name
-            rows.append({"LEVEL": "PROV", "Provinsi_Label": f"PROVINSI {prov_name}", "Kab_Kota_Label": None, "Pop_Total_Jenis": pval})
+            rows.append({
+                "LEVEL": "PROV",
+                "Provinsi_Label": f"PROVINSI {prov_name}",
+                "Kab_Kota_Label": None,
+                "Pop_Total_Jenis": pval,
+            })
             continue
 
-        rows.append({"LEVEL": "KAB", "Provinsi_Label": f"PROVINSI {current_prov}" if current_prov else None, "Kab_Kota_Label": mm, "Pop_Total_Jenis": pval})
+        rows.append({
+            "LEVEL": "KAB",
+            "Provinsi_Label": f"PROVINSI {current_prov}" if current_prov else None,
+            "Kab_Kota_Label": mm,
+            "Pop_Total_Jenis": pval,
+        })
 
     pop = pd.DataFrame(rows)
     if pop.empty:
@@ -372,13 +412,25 @@ def _parse_pop_khusus(path_xlsx: str) -> pd.DataFrame:
     return pop
 
 def load_default_files(force=False):
-    key = (DATA_FILE, POP_KAB, POP_PROV, POP_KHUSUS, _mtime(DATA_FILE), _mtime(POP_KAB), _mtime(POP_PROV), _mtime(POP_KHUSUS))
+    """
+    Load:
+    - DM (DATA_FILE) multi-sheet -> concat
+    - POP_KAB, POP_PROV, POP_KHUSUS
+    - Standarisasi kolom wilayah & jenis
+    - Dedup baris DM
+    - prepare_global() (YJ+MinMax+Indeks_Dasar)
+    """
+    key = (
+        DATA_FILE, POP_KAB, POP_PROV, POP_KHUSUS,
+        _mtime(DATA_FILE), _mtime(POP_KAB), _mtime(POP_PROV), _mtime(POP_KHUSUS)
+    )
+
     if (not force) and _CACHE["key"] == key and _CACHE["df_all"] is not None:
         return _CACHE["df_all"], _CACHE["df_raw"], _CACHE["pop_kab"], _CACHE["pop_prov"], _CACHE["pop_khusus"], _CACHE["meta"], _CACHE["info"]
 
     for p, label in [(DATA_FILE, "DM"), (POP_KAB, "POP_KAB"), (POP_PROV, "POP_PROV"), (POP_KHUSUS, "POP_KHUSUS")]:
         if not Path(p).exists():
-            info = f"❌ File {label} tidak ditemukan: `{p}`"
+            info = f"File {label} tidak ditemukan: `{p}`"
             _CACHE.update({"key": key, "df_all": None, "df_raw": None, "pop_kab": None, "pop_prov": None, "pop_khusus": None, "meta": {}, "info": info})
             return None, None, None, None, None, {}, info
 
@@ -399,7 +451,7 @@ def load_default_files(force=False):
     if kew_col is None:   missing.append("Kewenangan")
     if jenis_col is None: missing.append("Jenis Perpustakaan")
     if missing:
-        info = f"❌ Kolom wajib tidak ditemukan di DM: {', '.join(missing)}"
+        info = f"Kolom wajib tidak ditemukan di DM: {', '.join(missing)}"
         _CACHE.update({"key": key, "df_all": None, "df_raw": None, "pop_kab": None, "pop_prov": None, "pop_khusus": None, "meta": {}, "info": info})
         return None, None, None, None, None, {}, info
 
@@ -416,6 +468,7 @@ def load_default_files(force=False):
     df_raw["prov_key"]  = df_raw["PROV_DISP"].apply(norm_prov_label)
     df_raw["kab_key"]   = df_raw["KAB_DISP"].apply(norm_kab_label)
 
+    # Dedup aman berdasarkan (prov,kab,kew,jenis,nama_perpus)
     if nama_col and nama_col in df_raw.columns:
         kcols = [prov_col, kab_col, kew_col, jenis_col, nama_col]
     else:
@@ -427,11 +480,12 @@ def load_default_files(force=False):
     df_raw = df_raw.drop_duplicates(subset=["_row_key"], keep="first").copy()
     after = len(df_raw)
 
+    # POP KAB
     pk = pd.read_excel(POP_KAB)
     c_kab  = pick_col(pk, ["KABUPATEN_KOTA","Kab/Kota","Kabupaten/Kota","KAB/KOTA","Kabupaten_Kota","kab_kota","kabupaten_kota"])
     c_prov = pick_col(pk, ["PROVINSI","Provinsi","provinsi"])
     if c_kab is None:
-        info = "❌ POP_KAB: wajib ada kolom Kab/Kota."
+        info = "POP_KAB: wajib ada kolom Kab/Kota."
         _CACHE.update({"key": key, "df_all": None, "df_raw": None, "pop_kab": None, "pop_prov": None, "pop_khusus": None, "meta": {}, "info": info})
         return None, None, None, None, None, {}, info
 
@@ -441,10 +495,11 @@ def load_default_files(force=False):
     pop_kab["kab_key"] = pop_kab["Kab_Kota_Label"].apply(norm_kab_label)
     pop_kab = pop_kab.groupby("kab_key", as_index=False).first()
 
+    # POP PROV
     pp = pd.read_excel(POP_PROV)
     c_pr = pick_col(pp, ["Provinsi","PROVINSI","provinsi","Propinsi","PROPINSI","propinsi"])
     if c_pr is None:
-        info = "❌ POP_PROV: wajib ada kolom Provinsi."
+        info = "POP_PROV: wajib ada kolom Provinsi."
         _CACHE.update({"key": key, "df_all": None, "df_raw": None, "pop_kab": None, "pop_prov": None, "pop_khusus": None, "meta": {}, "info": info})
         return None, None, None, None, None, {}, info
 
@@ -453,10 +508,11 @@ def load_default_files(force=False):
     pop_prov["prov_key"] = pop_prov["Provinsi_Label"].apply(norm_prov_label)
     pop_prov = pop_prov.groupby("prov_key", as_index=False).first()
 
+    # POP KHUSUS
     try:
         pop_khusus = _parse_pop_khusus(POP_KHUSUS)
     except Exception as e:
-        info = f"❌ POP_KHUSUS gagal dibaca: {repr(e)}"
+        info = f"POP_KHUSUS gagal dibaca: {repr(e)}"
         _CACHE.update({"key": key, "df_all": None, "df_raw": None, "pop_kab": None, "pop_prov": None, "pop_khusus": None, "meta": {}, "info": info})
         return None, None, None, None, None, {}, info
 
@@ -464,16 +520,25 @@ def load_default_files(force=False):
     meta = dict(prov_col=prov_col, kab_col=kab_col, kew_col=kew_col, jenis_col=jenis_col, nama_col=nama_col)
 
     info = (
-        f"✅ Mode NO UPLOAD (cache aktif)<br>"
-        f"✅ DM: <b>{fp.name}</b> | Baris: {before} → dedup: {after}<br>"
-        f"✅ POP_KAB: <b>{Path(POP_KAB).name}</b> (n={len(pop_kab)})<br>"
-        f"✅ POP_PROV: <b>{Path(POP_PROV).name}</b> (n={len(pop_prov)})<br>"
-        f"✅ POP_KHUSUS: <b>{Path(POP_KHUSUS).name}</b> (n={len(pop_khusus)}) — (PROV row ikut dihitung)<br>"
-        f"✅ TARGET sampel per jenis: <b>{TARGET_RATIO*100:.2f}%</b><br>"
-        f"🕒 mtime: DM={time.ctime(_mtime(DATA_FILE))} | Kab={time.ctime(_mtime(POP_KAB))} | Prov={time.ctime(_mtime(POP_PROV))} | Khusus={time.ctime(_mtime(POP_KHUSUS))}"
+        f"Mode NO UPLOAD (cache aktif)\n"
+        f"DM: {fp.name} | Baris: {before} -> dedup: {after}\n"
+        f"POP_KAB: {Path(POP_KAB).name} (n={len(pop_kab)})\n"
+        f"POP_PROV: {Path(POP_PROV).name} (n={len(pop_prov)})\n"
+        f"POP_KHUSUS: {Path(POP_KHUSUS).name} (n={len(pop_khusus)})\n"
+        f"TARGET sampel per jenis: {TARGET_RATIO*100:.2f}%\n"
+        f"mtime: DM={time.ctime(_mtime(DATA_FILE))} | Kab={time.ctime(_mtime(POP_KAB))} | Prov={time.ctime(_mtime(POP_PROV))} | Khusus={time.ctime(_mtime(POP_KHUSUS))}"
     )
 
-    _CACHE.update({"key": key, "df_all": df_all, "df_raw": df_raw, "pop_kab": pop_kab, "pop_prov": pop_prov, "pop_khusus": pop_khusus, "meta": meta, "info": info})
+    _CACHE.update({
+        "key": key,
+        "df_all": df_all,
+        "df_raw": df_raw,
+        "pop_kab": pop_kab,
+        "pop_prov": pop_prov,
+        "pop_khusus": pop_khusus,
+        "meta": meta,
+        "info": info
+    })
     return df_all, df_raw, pop_kab, pop_prov, pop_khusus, meta, info
 
 
@@ -481,7 +546,19 @@ def load_default_files(force=False):
 # 6) FAKTOR WILAYAH — PER JENIS (TARGET 33.88%)
 # ============================================================
 
-def build_faktor_wilayah_jenis(df_filtered: pd.DataFrame, pop_kab: pd.DataFrame, pop_prov: pd.DataFrame, pop_khusus: pd.DataFrame, kew_value: str):
+def build_faktor_wilayah_jenis(
+    df_filtered: pd.DataFrame,
+    pop_kab: pd.DataFrame,
+    pop_prov: pd.DataFrame,
+    pop_khusus: pd.DataFrame,
+    kew_value: str
+):
+    """
+    Output:
+    group_key + (Kab/Kota atau Provinsi) + Jenis
+    n_jenis, pop_total_jenis, target_total_33_88_jenis,
+    coverage_jenis_%, faktor_penyesuaian_jenis, gap_target33_88_jenis
+    """
     if df_filtered is None or df_filtered.empty:
         return pd.DataFrame()
 
@@ -493,6 +570,7 @@ def build_faktor_wilayah_jenis(df_filtered: pd.DataFrame, pop_kab: pd.DataFrame,
 
     jenis_list = ["sekolah", "umum", "khusus"]
 
+    # tentukan level berdasarkan kewenangan
     if "PROV" in kew_norm:
         key_col, label_col, label_name, mode = "prov_key", "PROV_DISP", "Provinsi", "PROV"
         base_pop = pop_prov.copy() if (pop_prov is not None and not pop_prov.empty) else pd.DataFrame()
@@ -507,7 +585,10 @@ def build_faktor_wilayah_jenis(df_filtered: pd.DataFrame, pop_kab: pd.DataFrame,
         base_pop = base_pop.set_index("kab_key") if (not base_pop.empty and "kab_key" in base_pop.columns) else pd.DataFrame().set_index(pd.Index([]))
 
     base_keys = df[[key_col, label_col]].drop_duplicates().rename(columns={key_col: "group_key", label_col: label_name})
-    full = base_keys.assign(_tmp=1).merge(pd.DataFrame({"Jenis": jenis_list, "_tmp": 1}), on="_tmp").drop(columns="_tmp")
+    full = base_keys.assign(_tmp=1).merge(
+        pd.DataFrame({"Jenis": jenis_list, "_tmp": 1}),
+        on="_tmp"
+    ).drop(columns="_tmp")
 
     cnt = (
         df.groupby([key_col, label_col, "_dataset"], dropna=False)
@@ -523,10 +604,12 @@ def build_faktor_wilayah_jenis(df_filtered: pd.DataFrame, pop_kab: pd.DataFrame,
     base_n["target_total_33_88_jenis"] = 0.0
     base_n["pop_total_jenis"] = 0.0
 
+    # SEKOLAH + UMUM dari POP_KAB/POP_PROV
     if not base_pop.empty:
         if mode == "KAB":
             pop_sekolah = pd.to_numeric(base_pop.get("jumlah_populasi_sekolah", 0), errors="coerce").fillna(0.0)
             pop_umum    = pd.to_numeric(base_pop.get("jumlah_populasi_umum", 0), errors="coerce").fillna(0.0)
+
             tgt_sekolah = pop_sekolah * float(TARGET_RATIO)
             tgt_umum    = pop_umum * float(TARGET_RATIO)
         else:
@@ -535,6 +618,7 @@ def build_faktor_wilayah_jenis(df_filtered: pd.DataFrame, pop_kab: pd.DataFrame,
             slb = pd.to_numeric(base_pop.get("slb", 0), errors="coerce").fillna(0.0)
             pop_sekolah = sma + smk + slb
             tgt_sekolah = pop_sekolah * float(TARGET_RATIO)
+
             pop_umum = pd.to_numeric(base_pop.get("perpus_umum_prop", 0), errors="coerce").fillna(0.0)
             tgt_umum = pop_umum * float(TARGET_RATIO)
 
@@ -546,6 +630,7 @@ def build_faktor_wilayah_jenis(df_filtered: pd.DataFrame, pop_kab: pd.DataFrame,
         base_n.loc[m, "pop_total_jenis"] = base_n.loc[m, "group_key"].map(pop_umum).fillna(0.0).values
         base_n.loc[m, "target_total_33_88_jenis"] = base_n.loc[m, "group_key"].map(tgt_umum).fillna(0.0).values
 
+    # KHUSUS dari POP_KHUSUS
     if pop_khusus is not None and not pop_khusus.empty:
         pk = pop_khusus.copy()
         pk["Pop_Total_Jenis"] = pd.to_numeric(pk.get("Pop_Total_Jenis", 0), errors="coerce").fillna(0.0)
@@ -609,6 +694,13 @@ def build_faktor_wilayah_jenis(df_filtered: pd.DataFrame, pop_kab: pd.DataFrame,
 # ============================================================
 
 def build_agg_wilayah_jenis(df_filtered: pd.DataFrame, faktor_wilayah_jenis: pd.DataFrame, kew_value: str):
+    """
+    Agregasi wilayah × jenis:
+      - Jumlah entitas
+      - rata-rata sub/dim
+      - Indeks_Dasar_Agregat_0_100 = mean(Indeks_Dasar_0_100)
+      - Indeks_Final_Agregat_0_100 = Indeks_Dasar_Agregat_0_100 * faktor_penyesuaian_jenis
+    """
     if df_filtered is None or df_filtered.empty:
         return pd.DataFrame()
 
@@ -627,7 +719,10 @@ def build_agg_wilayah_jenis(df_filtered: pd.DataFrame, faktor_wilayah_jenis: pd.
     jenis_list = ["sekolah", "umum", "khusus"]
 
     base_keys = df[[key_col, label_col]].drop_duplicates().rename(columns={key_col: "group_key", label_col: label_name})
-    full = base_keys.assign(_tmp=1).merge(pd.DataFrame({"Jenis": jenis_list, "_tmp": 1}), on="_tmp").drop(columns="_tmp")
+    full = base_keys.assign(_tmp=1).merge(
+        pd.DataFrame({"Jenis": jenis_list, "_tmp": 1}),
+        on="_tmp"
+    ).drop(columns="_tmp")
 
     agg_real = df.groupby([key_col, label_col, "_dataset"], dropna=False).agg(
         Jumlah=("Indeks_Dasar_0_100", "size"),
@@ -641,10 +736,12 @@ def build_agg_wilayah_jenis(df_filtered: pd.DataFrame, faktor_wilayah_jenis: pd.
     ).reset_index().rename(columns={key_col: "group_key", label_col: label_name, "_dataset": "Jenis"})
 
     agg_real["Jenis"] = agg_real["Jenis"].astype(str).str.lower().str.strip()
-    agg = full.merge(agg_real, on=["group_key", label_name, "Jenis"], how="left")
 
-    for c in ["Jumlah","Rata2_sub_koleksi","Rata2_sub_sdm","Rata2_sub_pelayanan","Rata2_sub_pengelolaan",
-              "Rata2_dim_kepatuhan","Rata2_dim_kinerja","Indeks_Dasar_Agregat_0_100"]:
+    agg = full.merge(agg_real, on=["group_key", label_name, "Jenis"], how="left")
+    for c in [
+        "Jumlah","Rata2_sub_koleksi","Rata2_sub_sdm","Rata2_sub_pelayanan","Rata2_sub_pengelolaan",
+        "Rata2_dim_kepatuhan","Rata2_dim_kinerja","Indeks_Dasar_Agregat_0_100"
+    ]:
         if c in agg.columns:
             agg[c] = pd.to_numeric(agg[c], errors="coerce").fillna(0.0)
 
@@ -660,6 +757,7 @@ def build_agg_wilayah_jenis(df_filtered: pd.DataFrame, faktor_wilayah_jenis: pd.
     else:
         fw = faktor_wilayah_jenis.copy()
         fw["Jenis"] = fw["Jenis"].astype(str).str.lower().str.strip()
+
         keep = ["group_key", label_name, "Jenis",
                 "faktor_penyesuaian_jenis", "target_total_33_88_jenis", "pop_total_jenis",
                 "coverage_jenis_%", "gap_target33_88_jenis", "n_jenis"]
@@ -700,6 +798,11 @@ def build_agg_wilayah_jenis(df_filtered: pd.DataFrame, faktor_wilayah_jenis: pd.
 # ============================================================
 
 def build_agg_wilayah_total_from_jenis(agg_jenis: pd.DataFrame, faktor_wilayah_jenis: pd.DataFrame, kew_value: str):
+    """
+    Keseluruhan dari agg_jenis, FIX avg3:
+      Indeks_Dasar_Agregat_0_100 (keseluruhan) = mean(dasar_3jenis) [missing=0, tetap /3]
+      Indeks_Final_Wilayah_0_100 (keseluruhan) = mean(final_3jenis) [missing=0, tetap /3]
+    """
     if agg_jenis is None or agg_jenis.empty:
         return pd.DataFrame()
 
@@ -711,15 +814,19 @@ def build_agg_wilayah_total_from_jenis(agg_jenis: pd.DataFrame, faktor_wilayah_j
     a["Jenis"] = a["Jenis"].astype(str).str.lower().str.strip()
 
     base_keys = a[["group_key", label_name]].drop_duplicates()
-    full = base_keys.assign(_tmp=1).merge(pd.DataFrame({"Jenis": jenis_list, "_tmp": 1}), on="_tmp").drop(columns="_tmp")
+    full = base_keys.assign(_tmp=1).merge(
+        pd.DataFrame({"Jenis": jenis_list, "_tmp": 1}),
+        on="_tmp"
+    ).drop(columns="_tmp")
 
-    cols_present = [c for c in [
+    cols_need = [
         "Jumlah",
         "Rata2_sub_koleksi","Rata2_sub_sdm","Rata2_sub_pelayanan","Rata2_sub_pengelolaan",
         "Rata2_dim_kepatuhan","Rata2_dim_kinerja",
         "Indeks_Dasar_Agregat_0_100",
         "Indeks_Final_Agregat_0_100",
-    ] if c in a.columns]
+    ]
+    cols_present = [c for c in cols_need if c in a.columns]
 
     full = full.merge(
         a[["group_key", label_name, "Jenis"] + cols_present],
@@ -742,9 +849,11 @@ def build_agg_wilayah_total_from_jenis(agg_jenis: pd.DataFrame, faktor_wilayah_j
         Indeks_Final_Wilayah_0_100=("Indeks_Final_Agregat_0_100", "mean"),
     )
 
+    # Tempel info Pop/Target/N per jenis + total (untuk verifikasi/ekspor)
     if faktor_wilayah_jenis is not None and not faktor_wilayah_jenis.empty:
         fw = faktor_wilayah_jenis.copy()
         fw["Jenis"] = fw["Jenis"].astype(str).str.lower().str.strip()
+
         piv = fw.pivot_table(
             index=["group_key", label_name],
             columns="Jenis",
@@ -784,7 +893,8 @@ def build_agg_wilayah_total_from_jenis(agg_jenis: pd.DataFrame, faktor_wilayah_j
 
         out["coverage_target33_88_all_%"] = np.where(
             pd.to_numeric(out["target_total_33_88_all"], errors="coerce").fillna(0).values > 0,
-            (pd.to_numeric(out["terkumpul_all"], errors="coerce").fillna(0).values / pd.to_numeric(out["target_total_33_88_all"], errors="coerce").fillna(0).values) * 100.0,
+            (pd.to_numeric(out["terkumpul_all"], errors="coerce").fillna(0).values /
+             pd.to_numeric(out["target_total_33_88_all"], errors="coerce").fillna(0).values) * 100.0,
             0.0
         )
         out["coverage_target33_88_all_%"] = pd.to_numeric(out["coverage_target33_88_all_%"], errors="coerce").fillna(0.0).round(2)
@@ -805,22 +915,25 @@ def build_agg_wilayah_total_from_jenis(agg_jenis: pd.DataFrame, faktor_wilayah_j
 
 
 # ============================================================
-# 9) RINGKASAN (PER JENIS) + KESELURUHAN (MEMUAT Rata2_sub/dim)
+# 9) SUMMARY (PER JENIS) + KESELURUHAN  + sub/dim
 # ============================================================
 
 def build_summary_per_jenis(agg_jenis: pd.DataFrame, agg_total: pd.DataFrame):
+    """
+    Ringkasan nasional (per jenis + keseluruhan) sebagai RUJUKAN UTAMA.
+    Memuat:
+      - Pop/Target33.88/Terkumpul/Coverage
+      - Indeks_Dasar_0_100, Indeks_Final_Disesuaikan_0_100, Penyesuaian_Poin
+      - Rata2 sub/dim: koleksi, sdm, pelayanan, pengelolaan, dim_kepatuhan, dim_kinerja
+
+    Catatan:
+      - Untuk 'keseluruhan': indeks dasar/final mengikuti aturan avg3 (missing=0 tetap /3),
+        sub/dim keseluruhan dihitung avg3 dari 3 jenis (missing=0 tetap /3).
+    """
     jenis_list = ["sekolah", "umum", "khusus"]
-    metric_cols = [
-        "Rata2_sub_koleksi",
-        "Rata2_sub_sdm",
-        "Rata2_sub_pelayanan",
-        "Rata2_sub_pengelolaan",
-        "Rata2_dim_kepatuhan",
-        "Rata2_dim_kinerja",
-    ]
 
     def _row_default(jenis):
-        base = {
+        return {
             "Jenis": jenis,
             "Jumlah_Wilayah": 0,
             "Total_Perpus": 0,
@@ -831,10 +944,13 @@ def build_summary_per_jenis(agg_jenis: pd.DataFrame, agg_total: pd.DataFrame):
             "Indeks_Dasar_0_100": 0.0,
             "Indeks_Final_Disesuaikan_0_100": 0.0,
             "Penyesuaian_Poin": 0.0,
+            "Rata2_sub_koleksi": 0.0,
+            "Rata2_sub_sdm": 0.0,
+            "Rata2_sub_pelayanan": 0.0,
+            "Rata2_sub_pengelolaan": 0.0,
+            "Rata2_dim_kepatuhan": 0.0,
+            "Rata2_dim_kinerja": 0.0,
         }
-        for c in metric_cols:
-            base[c] = 0.0
-        return base
 
     rows_by_jenis = {j: _row_default(j) for j in jenis_list}
 
@@ -848,8 +964,13 @@ def build_summary_per_jenis(agg_jenis: pd.DataFrame, agg_total: pd.DataFrame):
             "Indeks_Final_Agregat_0_100",
             "pop_total_jenis",
             "target_total_33_88_jenis",
-        ] + metric_cols
-
+            "Rata2_sub_koleksi",
+            "Rata2_sub_sdm",
+            "Rata2_sub_pelayanan",
+            "Rata2_sub_pengelolaan",
+            "Rata2_dim_kepatuhan",
+            "Rata2_dim_kinerja",
+        ]
         for c in num_cols:
             if c in a.columns:
                 a[c] = pd.to_numeric(a[c], errors="coerce").fillna(0.0)
@@ -860,15 +981,22 @@ def build_summary_per_jenis(agg_jenis: pd.DataFrame, agg_total: pd.DataFrame):
                 continue
 
             jumlah_wilayah = int(sub.shape[0])
-            terkumpul = int(pd.to_numeric(sub.get("Jumlah", 0), errors="coerce").fillna(0).sum())
-            pop_total = int(pd.to_numeric(sub.get("pop_total_jenis", 0), errors="coerce").fillna(0).sum())
-            target3388 = int(pd.to_numeric(sub.get("target_total_33_88_jenis", 0), errors="coerce").fillna(0).sum())
-
+            terkumpul = int(sub.get("Jumlah", 0).sum())
+            pop_total = int(sub.get("pop_total_jenis", 0).sum())
+            target3388 = int(sub.get("target_total_33_88_jenis", 0).sum())
             coverage = (terkumpul / target3388 * 100.0) if target3388 > 0 else 0.0
-            dasar = float(pd.to_numeric(sub.get("Indeks_Dasar_Agregat_0_100", 0), errors="coerce").fillna(0).mean())
-            final = float(pd.to_numeric(sub.get("Indeks_Final_Agregat_0_100", 0), errors="coerce").fillna(0).mean())
 
-            row = {
+            dasar = float(sub.get("Indeks_Dasar_Agregat_0_100", 0).mean())
+            final = float(sub.get("Indeks_Final_Agregat_0_100", 0).mean())
+
+            r_kol = float(sub.get("Rata2_sub_koleksi", 0).mean())
+            r_sdm = float(sub.get("Rata2_sub_sdm", 0).mean())
+            r_pel = float(sub.get("Rata2_sub_pelayanan", 0).mean())
+            r_png = float(sub.get("Rata2_sub_pengelolaan", 0).mean())
+            r_kep = float(sub.get("Rata2_dim_kepatuhan", 0).mean())
+            r_kin = float(sub.get("Rata2_dim_kinerja", 0).mean())
+
+            rows_by_jenis[jenis] = {
                 "Jenis": jenis,
                 "Jumlah_Wilayah": jumlah_wilayah,
                 "Total_Perpus": terkumpul,
@@ -879,23 +1007,42 @@ def build_summary_per_jenis(agg_jenis: pd.DataFrame, agg_total: pd.DataFrame):
                 "Indeks_Dasar_0_100": float(dasar),
                 "Indeks_Final_Disesuaikan_0_100": float(final),
                 "Penyesuaian_Poin": float(final - dasar),
+                "Rata2_sub_koleksi": r_kol,
+                "Rata2_sub_sdm": r_sdm,
+                "Rata2_sub_pelayanan": r_pel,
+                "Rata2_sub_pengelolaan": r_png,
+                "Rata2_dim_kepatuhan": r_kep,
+                "Rata2_dim_kinerja": r_kin,
             }
-            for c in metric_cols:
-                row[c] = float(pd.to_numeric(sub.get(c, 0), errors="coerce").fillna(0).mean())
-
-            rows_by_jenis[jenis] = row
 
     rows = [rows_by_jenis[j] for j in jenis_list]
 
-    def _avg3(colname: str) -> float:
-        return (
-            float(rows_by_jenis["sekolah"].get(colname, 0.0))
-            + float(rows_by_jenis["umum"].get(colname, 0.0))
-            + float(rows_by_jenis["khusus"].get(colname, 0.0))
-        ) / 3.0
-
-    dasar_all = _avg3("Indeks_Dasar_0_100")
-    final_all = _avg3("Indeks_Final_Disesuaikan_0_100")
+    dasar_all = (rows_by_jenis["sekolah"]["Indeks_Dasar_0_100"]
+                 + rows_by_jenis["umum"]["Indeks_Dasar_0_100"]
+                 + rows_by_jenis["khusus"]["Indeks_Dasar_0_100"]) / 3.0
+
+    final_all = (rows_by_jenis["sekolah"]["Indeks_Final_Disesuaikan_0_100"]
+                 + rows_by_jenis["umum"]["Indeks_Final_Disesuaikan_0_100"]
+                 + rows_by_jenis["khusus"]["Indeks_Final_Disesuaikan_0_100"]) / 3.0
+
+    r_kol_all = (rows_by_jenis["sekolah"]["Rata2_sub_koleksi"]
+                 + rows_by_jenis["umum"]["Rata2_sub_koleksi"]
+                 + rows_by_jenis["khusus"]["Rata2_sub_koleksi"]) / 3.0
+    r_sdm_all = (rows_by_jenis["sekolah"]["Rata2_sub_sdm"]
+                 + rows_by_jenis["umum"]["Rata2_sub_sdm"]
+                 + rows_by_jenis["khusus"]["Rata2_sub_sdm"]) / 3.0
+    r_pel_all = (rows_by_jenis["sekolah"]["Rata2_sub_pelayanan"]
+                 + rows_by_jenis["umum"]["Rata2_sub_pelayanan"]
+                 + rows_by_jenis["khusus"]["Rata2_sub_pelayanan"]) / 3.0
+    r_png_all = (rows_by_jenis["sekolah"]["Rata2_sub_pengelolaan"]
+                 + rows_by_jenis["umum"]["Rata2_sub_pengelolaan"]
+                 + rows_by_jenis["khusus"]["Rata2_sub_pengelolaan"]) / 3.0
+    r_kep_all = (rows_by_jenis["sekolah"]["Rata2_dim_kepatuhan"]
+                 + rows_by_jenis["umum"]["Rata2_dim_kepatuhan"]
+                 + rows_by_jenis["khusus"]["Rata2_dim_kepatuhan"]) / 3.0
+    r_kin_all = (rows_by_jenis["sekolah"]["Rata2_dim_kinerja"]
+                 + rows_by_jenis["umum"]["Rata2_dim_kinerja"]
+                 + rows_by_jenis["khusus"]["Rata2_dim_kinerja"]) / 3.0
 
     pop_all = int(rows_by_jenis["sekolah"]["Pop_Total_Jenis"]
                   + rows_by_jenis["umum"]["Pop_Total_Jenis"]
@@ -917,7 +1064,7 @@ def build_summary_per_jenis(agg_jenis: pd.DataFrame, agg_total: pd.DataFrame):
             rows_by_jenis["khusus"]["Jumlah_Wilayah"])
     )
 
-    row_all = {
+    rows.append({
         "Jenis": "keseluruhan",
         "Jumlah_Wilayah": jumlah_wilayah_all,
         "Total_Perpus": terkumpul_all,
@@ -928,24 +1075,27 @@ def build_summary_per_jenis(agg_jenis: pd.DataFrame, agg_total: pd.DataFrame):
         "Indeks_Dasar_0_100": float(dasar_all),
         "Indeks_Final_Disesuaikan_0_100": float(final_all),
         "Penyesuaian_Poin": float(final_all - dasar_all),
-    }
-    for c in metric_cols:
-        row_all[c] = float(_avg3(c))
+        "Rata2_sub_koleksi": float(r_kol_all),
+        "Rata2_sub_sdm": float(r_sdm_all),
+        "Rata2_sub_pelayanan": float(r_pel_all),
+        "Rata2_sub_pengelolaan": float(r_png_all),
+        "Rata2_dim_kepatuhan": float(r_kep_all),
+        "Rata2_dim_kinerja": float(r_kin_all),
+    })
 
-    rows.append(row_all)
     out = pd.DataFrame(rows)
 
     for c in ["Jumlah_Wilayah","Total_Perpus","Pop_Total_Jenis","Target33_88_Total_Jenis","Terkumpul_Jenis"]:
-        if c in out.columns:
-            out[c] = pd.to_numeric(out[c], errors="coerce").fillna(0).round(0).astype(int)
+        out[c] = pd.to_numeric(out[c], errors="coerce").fillna(0).round(0).astype(int)
 
-    for c in ["Coverage_Target33_88_Jenis_%","Indeks_Dasar_0_100","Indeks_Final_Disesuaikan_0_100","Penyesuaian_Poin"] + metric_cols:
-        if c in out.columns:
-            out[c] = pd.to_numeric(out[c], errors="coerce").fillna(0.0).round(3)
+    for c in ["Coverage_Target33_88_Jenis_%","Indeks_Dasar_0_100","Indeks_Final_Disesuaikan_0_100","Penyesuaian_Poin"]:
+        out[c] = pd.to_numeric(out[c], errors="coerce").fillna(0.0).round(2)
 
-    for c in ["Indeks_Dasar_0_100","Indeks_Final_Disesuaikan_0_100","Penyesuaian_Poin","Coverage_Target33_88_Jenis_%"]:
-        if c in out.columns:
-            out[c] = pd.to_numeric(out[c], errors="coerce").fillna(0.0).round(2)
+    for c in [
+        "Rata2_sub_koleksi","Rata2_sub_sdm","Rata2_sub_pelayanan","Rata2_sub_pengelolaan",
+        "Rata2_dim_kepatuhan","Rata2_dim_kinerja"
+    ]:
+        out[c] = pd.to_numeric(out[c], errors="coerce").fillna(0.0).round(3)
 
     return out
 
@@ -1036,11 +1186,17 @@ def build_verif_jenis(faktor_wilayah_jenis: pd.DataFrame, kew_value: str):
 
 
 # ============================================================
-# 12) BELL CURVE — Indeks Dasar per Entitas (per Jenis) + Hover Nama
+# 12) BELL CURVE — Indeks Dasar per Entitas (per Jenis) + Hover Nama Perpus
 # ============================================================
 
-def _make_bell_curve_entitas(dfp: pd.DataFrame, title: str, xcol: str = "Indeks_Dasar_0_100",
-                            label_col: str = "nm_perpustakaan", hover_cols=None, min_points: int = 2):
+def _make_bell_curve_entitas(
+    dfp: pd.DataFrame,
+    title: str,
+    xcol: str = "Indeks_Dasar_0_100",
+    label_col: str = "nm_perpustakaan",
+    hover_cols: Optional[List[str]] = None,
+    min_points: int = 2
+):
     fig = go.Figure()
     fig.update_layout(
         title=title,
@@ -1096,8 +1252,12 @@ def _make_bell_curve_entitas(dfp: pd.DataFrame, title: str, xcol: str = "Indeks_
 
     if len(x) < min_points:
         x_single = float(x[0])
-        fig.add_trace(go.Scatter(x=[x_single], y=[0], mode="markers", showlegend=False,
-                                 hovertext=[hover_text[0]] if hover_text else None, hoverinfo="text"))
+        fig.add_trace(go.Scatter(
+            x=[x_single], y=[0],
+            mode="markers", showlegend=False,
+            hovertext=[hover_text[0]] if hover_text else None,
+            hoverinfo="text"
+        ))
         fig.add_vline(x=x_single, line_width=1, line_dash="dash",
                       annotation_text=f"Nilai: {x_single:.1f}", annotation_position="top")
         fig.update_xaxes(range=[0, 100])
@@ -1114,8 +1274,12 @@ def _make_bell_curve_entitas(dfp: pd.DataFrame, title: str, xcol: str = "Indeks_
     pdf = (1.0 / (sigma * np.sqrt(2 * np.pi))) * np.exp(-0.5 * ((xs - mu) / sigma) ** 2)
 
     fig.add_trace(go.Scatter(x=xs, y=pdf, mode="lines", name="Kurva Normal (fit)"))
-    fig.add_trace(go.Scatter(x=x, y=np.zeros_like(x), mode="markers", showlegend=False,
-                             hovertext=hover_text if hover_text else None, hoverinfo="text"))
+    fig.add_trace(go.Scatter(
+        x=x, y=np.zeros_like(x),
+        mode="markers", showlegend=False,
+        hovertext=hover_text if hover_text else None,
+        hoverinfo="text"
+    ))
 
     q1, q2, q3 = np.percentile(x, [25, 50, 75])
     for xv, lab in [(q1, "Q1"), (q2, "Q2 (Median)"), (q3, "Q3"), (mu, "Mean")]:
@@ -1142,37 +1306,42 @@ def _safe_first(df, col, default=0.0, where=None):
     return float(pd.to_numeric(sub[col], errors="coerce").fillna(default).iloc[0])
 
 def compute_dashboard_kpis(summary_jenis: pd.DataFrame):
-    final_all = _safe_first(summary_jenis, "Indeks_Final_Disesuaikan_0_100", 0.0, where=summary_jenis["Jenis"].astype(str).str.lower().eq("keseluruhan"))
-    dasar_all = _safe_first(summary_jenis, "Indeks_Dasar_0_100", 0.0, where=summary_jenis["Jenis"].astype(str).str.lower().eq("keseluruhan"))
+    final_all = _safe_first(summary_jenis, "Indeks_Final_Disesuaikan_0_100", 0.0,
+                            where=summary_jenis["Jenis"].astype(str).str.lower().eq("keseluruhan"))
+    dasar_all = _safe_first(summary_jenis, "Indeks_Dasar_0_100", 0.0,
+                            where=summary_jenis["Jenis"].astype(str).str.lower().eq("keseluruhan"))
     return {"final_all": final_all, "dasar_all": dasar_all}
 
 def build_kpi_markdown(summary_jenis: pd.DataFrame) -> str:
     if summary_jenis is None or summary_jenis.empty:
         return ""
+
     k = compute_dashboard_kpis(summary_jenis)
 
     def fmt(x, nd=2):
         return "NA" if pd.isna(x) else f"{x:.{nd}f}"
 
+    target_pct = TARGET_RATIO * 100.0
+
     return f"""
 <div style="display:flex; gap:12px; flex-wrap:wrap;">
   <div style="border:1px solid #333; border-radius:10px; padding:10px 12px; min-width:260px;">
-    <div style="opacity:0.8;">Indeks IPLM FINAL (Disesuaikan 33.88%)</div>
+    <div style="opacity:0.8;">Indeks IPLM FINAL (Disesuaikan {target_pct:.2f}%)</div>
     <div style="font-size:26px; font-weight:700;">{fmt(k["final_all"],2)}</div>
-    <div style="opacity:0.7;">Skor absolut (untuk akuntabilitas)</div>
+    <div style="opacity:0.7;">final = dasar × faktor kecukupan sampel</div>
   </div>
 
   <div style="border:1px solid #333; border-radius:10px; padding:10px 12px; min-width:260px;">
     <div style="opacity:0.8;">Indeks Dasar (Tanpa Penyesuaian)</div>
     <div style="font-size:26px; font-weight:700;">{fmt(k["dasar_all"],2)}</div>
-    <div style="opacity:0.7;">Sebelum faktor kecukupan sampel</div>
+    <div style="opacity:0.7;">skor absolut sebelum faktor kecukupan data</div>
   </div>
 </div>
 """.strip()
 
 
 # ============================================================
-# 14) LLM + WORD (LLM RUJUKAN UTAMA = TABEL RINGKASAN)
+# 14) LLM + WORD (OPSIONAL)
 # ============================================================
 
 _HF_CLIENT = None
@@ -1191,95 +1360,126 @@ def get_llm_client():
         _HF_CLIENT = None
         return None
 
-def generate_llm_analysis(summary_jenis, wilayah, kew):
+def _summary_md_for_llm(summary_jenis: pd.DataFrame) -> str:
+    """
+    Buat tabel markdown yang stabil untuk LLM (hanya 4 baris wajib).
+    """
+    if summary_jenis is None or summary_jenis.empty:
+        return "(Tabel ringkasan kosong)"
+
+    s = summary_jenis.copy()
+    s["Jenis"] = s["Jenis"].astype(str).str.lower().str.strip()
+
+    order = ["sekolah", "umum", "khusus", "keseluruhan"]
+    s = s[s["Jenis"].isin(order)].copy()
+    if s.empty:
+        return "(Tabel ringkasan tidak memuat baris sekolah/umum/khusus/keseluruhan)"
+
+    s["_ord"] = s["Jenis"].map({k: i for i, k in enumerate(order)})
+    s = s.sort_values("_ord").drop(columns=["_ord"])
+
+    cols = [
+        "Jenis",
+        "Jumlah_Wilayah",
+        "Total_Perpus",
+        "Pop_Total_Jenis",
+        "Target33_88_Total_Jenis",
+        "Terkumpul_Jenis",
+        "Coverage_Target33_88_Jenis_%",
+        "Indeks_Dasar_0_100",
+        "Indeks_Final_Disesuaikan_0_100",
+        "Penyesuaian_Poin",
+        "Rata2_sub_koleksi",
+        "Rata2_sub_sdm",
+        "Rata2_sub_pelayanan",
+        "Rata2_sub_pengelolaan",
+        "Rata2_dim_kepatuhan",
+        "Rata2_dim_kinerja",
+    ]
+    cols = [c for c in cols if c in s.columns]
+    s = s[cols].copy()
+
+    def _fmt(v, col):
+        if pd.isna(v):
+            return ""
+        if col in ["Jumlah_Wilayah", "Total_Perpus", "Pop_Total_Jenis", "Target33_88_Total_Jenis", "Terkumpul_Jenis"]:
+            try:
+                return str(int(v))
+            except Exception:
+                return str(v)
+        if col in ["Coverage_Target33_88_Jenis_%", "Indeks_Dasar_0_100", "Indeks_Final_Disesuaikan_0_100", "Penyesuaian_Poin"]:
+            return f"{float(v):.2f}"
+        if col.startswith("Rata2_"):
+            return f"{float(v):.3f}"
+        return str(v)
+
+    s_disp = s.copy()
+    for c in s_disp.columns:
+        s_disp[c] = [_fmt(v, c) for v in s_disp[c].tolist()]
+
+    try:
+        return s_disp.to_markdown(index=False)
+    except Exception:
+        return s_disp.to_string(index=False)
+
+def generate_llm_analysis(summary_jenis, agg_total, verif_total, wilayah, kew):
     client = get_llm_client()
     if client is None or (not USE_LLM):
         return "Analisis otomatis (LLM) tidak digunakan / tidak tersedia."
 
-    ctx = f"Wilayah={wilayah} | Kewenangan={kew} | Target={TARGET_RATIO*100:.2f}%"
-    summary_md = df_to_markdown_table(summary_jenis, max_rows=10)
+    ctx = f"Wilayah={wilayah} | Kewenangan={kew} | Target_per_jenis={TARGET_RATIO*100:.2f}%"
+    summary_md = _summary_md_for_llm(summary_jenis)
 
     prompt = f"""
-
 Anda adalah analis kebijakan perpustakaan di Indonesia.
-
-Tulis analisis berbasis DATA yang sepenuhnya merujuk pada
-TABEL 'Ringkasan (Jenis + Keseluruhan)'.
-
-ATURAN:
-1. Semua angka harus diambil persis dari tabel.
-2. Dilarang membuat angka baru atau membulatkan ulang.
-3. Dilarang menggunakan percentile atau benchmarking eksternal.
-4. Hindari kata evaluatif seperti:
-   baik, buruk, optimal, kurang, memuaskan.
-5. Interpretasi hanya boleh berbasis relasi angka:
-   selisih absolut, rentang, perbandingan lebih besar/kecil,
-   perubahan numerik, kontribusi.
-
-WAJIB DISEBUT:
-- Baris: keseluruhan, sekolah, umum, khusus
-- Kolom:
-  Indeks_Dasar_0_100
-  Indeks_Final_Disesuaikan_0_100
-  Penyesuaian_Poin
-  Rata2_sub_koleksi
-  Rata2_sub_sdm
-  Rata2_sub_pelayanan
-  Rata2_sub_pengelolaan
-  Rata2_dim_kepatuhan
-  Rata2_dim_kinerja
-
-WAJIB DIJELASKAN:
-Indeks_Final_Disesuaikan_0_100 merupakan hasil
-Indeks_Dasar_0_100 dikalikan faktor kecukupan sampel
-berdasarkan target 33.88% per jenis.
-Perubahan antara dasar dan final adalah efek penyesuaian tersebut.
-
-FORMAT:
-Tepat 3 paragraf.
-
-Paragraf 1:
-Deskripsi numerik Indeks_Dasar_0_100 dan profil sub/dimensi
-untuk keseluruhan dan tiga jenis.
-Sebutkan selisih atau rentang angka bila relevan.
-
-Paragraf 2:
-Deskripsi Indeks_Final_Disesuaikan_0_100 dan Penyesuaian_Poin.
-Jelaskan perubahan absolutnya dan kaitkan dengan faktor kecukupan sampel.
-
-Paragraf 3:
-Tampilkan opsi teknis peningkatan indeks IPLM
-yang diturunkan langsung dari pola angka.
-Contoh:
-- Sub dengan nilai paling kecil → area ekspansi program.
-- Gap antar dimensi → arah harmonisasi kebijakan.
-- Jenis dengan penyesuaian negatif terbesar → fokus peningkatan coverage.
-Jangan gunakan bahasa normatif, tetap berbasis angka.
+Tulis analisis berbasis DATA, tanpa percentile/benchmarking.
+
+GAYA & ATURAN (WAJIB):
+- Netral dan deskriptif: dilarang memakai label normatif seperti “baik/buruk”, “tinggi/sedang/rendah”, “memuaskan/kurang”, “optimal/tidak optimal”.
+- Interpretasi hanya boleh memakai relasi angka: “lebih besar/kecil”, “selisih”, “gap”, “konsisten/tidak konsisten”, “dominan”, “terkonsentrasi”, “proporsi”, “kontribusi”, “perubahan absolut”.
+- DILARANG mengarang angka. Semua angka yang disebut harus muncul di TABEL.
+
+WAJIB (RUJUKAN UTAMA):
+- Jadikan TABEL 'Ringkasan (Jenis + Keseluruhan)' sebagai rujukan utama.
+- Kutip angka penting minimal untuk baris: sekolah, umum, khusus, keseluruhan.
+- Bahas metrik sub/dimensi:
+  Rata2_sub_koleksi, Rata2_sub_sdm, Rata2_sub_pelayanan, Rata2_sub_pengelolaan,
+  Rata2_dim_kepatuhan, Rata2_dim_kinerja.
+- Jelaskan makna penyesuaian {TARGET_RATIO*100:.2f}%:
+  tekankan bahwa Indeks_Final = Indeks_Dasar × faktor kecukupan sampel (target per jenis),
+  sehingga selisih final vs dasar merepresentasikan konsekuensi kecukupan data (cakupan),
+  bukan perubahan capaian layanan itu sendiri.
+
+FORMAT OUTPUT (WAJIB): tepat 3 paragraf, tanpa bullet list.
+(1) Gambaran Indeks Dasar + profil sub/dimensi (pakai angka; sebut relasi “lebih besar/kecil” dan gap antar-sub/dimensi).
+(2) Dampak penyesuaian {TARGET_RATIO*100:.2f}%: bandingkan Indeks_Final vs Indeks_Dasar per jenis dan keseluruhan
+    (gunakan angka Dasar, Final, dan/atau Penyesuaian_Poin dari tabel).
+(3) Tindakan teknis untuk menaikkan indeks (tanpa menghakimi): 3–4 kalimat tindakan berbasis pola angka,
+    misalnya sub yang lebih kecil -> prioritas penguatan; gap kepatuhan vs kinerja -> arah intervensi;
+    konsistensi antar-jenis -> standardisasi; serta langkah perbaikan kecukupan data agar selisih final vs dasar mengecil.
 
 Konteks:
 {ctx}
 
-TABEL RINGKASAN:
+TABEL RINGKASAN (RUJUKAN UTAMA):
 {summary_md}
-""".strip()
-
 """.strip()
 
     try:
         resp = client.chat_completion(
             model=LLM_MODEL_NAME,
             messages=[
-                {"role":"system","content":"Tulis analisis ringkas, transparan, dan berbasis angka tabel."},
+                {"role":"system","content":"Ikuti instruksi pengguna secara ketat. Jangan menilai kualitas; hanya relasi angka."},
                 {"role":"user","content":prompt}
             ],
-            max_tokens=700,
-            temperature=0.25,
+            max_tokens=650,
+            temperature=0.15,
             top_p=0.9,
         )
         text = resp.choices[0].message.content.strip()
         return text if text else "LLM mengembalikan respon kosong."
     except Exception as e:
-        return f"⚠️ Error LLM: {repr(e)}"
+        return f"Error LLM: {repr(e)}"
 
 def generate_word_report(wilayah, summary_jenis, analysis_text):
     if (not DOCX_AVAILABLE) or (Document is None):
@@ -1301,13 +1501,14 @@ def generate_word_report(wilayah, summary_jenis, analysis_text):
                 if pd.isna(v):
                     cells[i].text = ""
                 elif isinstance(v, (float, np.floating)):
-                    cells[i].text = f"{float(v):.3f}"
+                    # angka ringkasan sudah dibulatkan, tapi tulis rapi
+                    cells[i].text = f"{float(v)}"
                 elif isinstance(v, (int, np.integer)):
                     cells[i].text = str(int(v))
                 else:
                     cells[i].text = str(v)
 
-    doc.add_heading("Analisis (LLM)", level=2)
+    doc.add_heading("Analisis (LLM, opsional)", level=2)
     for p in (analysis_text or "").split("\n"):
         if p.strip():
             doc.add_paragraph(p.strip())
@@ -1321,7 +1522,7 @@ def generate_word_report(wilayah, summary_jenis, analysis_text):
 # 15) CORE RUN
 # ============================================================
 
-def _empty_outputs(msg="⚠️ Data belum siap."):
+def _empty_outputs(msg="Data belum siap."):
     empty = pd.DataFrame()
     empty_fig = go.Figure()
     return (
@@ -1335,7 +1536,7 @@ def _empty_outputs(msg="⚠️ Data belum siap."):
 def run_calc(prov_value, kab_value, kew_value, df_all, df_raw, pop_kab, pop_prov, pop_khusus, meta):
     try:
         if df_all is None or df_all.empty or df_raw is None or df_raw.empty:
-            return _empty_outputs("⚠️ Data belum ter-load. Pastikan file tersedia di repo/server.")
+            return _empty_outputs("Data belum ter-load. Pastikan file tersedia di repo/server.")
 
         df = df_all.copy()
         if prov_value and prov_value != "(Semua)":
@@ -1353,17 +1554,16 @@ def run_calc(prov_value, kab_value, kew_value, df_all, df_raw, pop_kab, pop_prov
         agg_jenis_full = build_agg_wilayah_jenis(df, faktor_wilayah_jenis, kew_norm)
         agg_total = build_agg_wilayah_total_from_jenis(agg_jenis_full, faktor_wilayah_jenis, kew_norm)
 
-        # ✅ Ringkasan sekarang sudah mengandung Rata2_sub* dan Rata2_dim*
         summary_jenis = build_summary_per_jenis(agg_jenis_full, agg_total)
         verif_total = build_verif_jenis(faktor_wilayah_jenis, kew_norm)
         detail_view = attach_final_to_detail(df, agg_total, meta, kew_norm)
 
-        # UI agg_jenis: tampil sampai indeks dasar + juga rata2 sub/dim
+        # agg_jenis view: tampilkan sampai indeks dasar agregat
         if agg_jenis_full is None or agg_jenis_full.empty:
             agg_jenis_view = agg_jenis_full
         else:
             kew_norm2 = str(kew_norm).upper()
-            label_name = "Kab/Kota" if ("KAB" in kew_norm2 or "KOTA" in kew_norm2) else ("Provinsi" if "PROV" in kew_norm2 else "Kab/Kota")
+            label_name = "Provinsi" if "PROV" in kew_norm2 else "Kab/Kota"
             cols_upto = [
                 "group_key",
                 label_name,
@@ -1376,6 +1576,7 @@ def run_calc(prov_value, kab_value, kew_value, df_all, df_raw, pop_kab, pop_prov
             cols_upto = [c for c in cols_upto if c in agg_jenis_full.columns]
             agg_jenis_view = agg_jenis_full[cols_upto].copy()
 
+        # RAW untuk download (hasil filter dari df_raw)
         raw = df_raw.copy()
         if prov_value and prov_value != "(Semua)":
             raw = raw[raw["PROV_DISP"] == prov_value]
@@ -1384,7 +1585,7 @@ def run_calc(prov_value, kab_value, kew_value, df_all, df_raw, pop_kab, pop_prov
         if kew_value and kew_value != "(Semua)":
             raw = raw[raw["KEW_NORM"] == kew_value]
 
-        # bell curve per jenis
+        # Bell curve per jenis (pakai detail_view)
         if detail_view is None or detail_view.empty:
             fig_umum = _make_bell_curve_entitas(pd.DataFrame(), "Bell Curve — Jenis: Umum")
             fig_sekolah = _make_bell_curve_entitas(pd.DataFrame(), "Bell Curve — Jenis: Sekolah")
@@ -1407,9 +1608,10 @@ def run_calc(prov_value, kab_value, kew_value, df_all, df_raw, pop_kab, pop_prov
             fig_umum = _fig("umum")
             fig_khusus = _fig("khusus")
 
+        # KPI 2 kartu
         kpi_md = build_kpi_markdown(summary_jenis)
 
-        # export
+        # Export
         tmpdir = tempfile.mkdtemp()
         prov_slug = (_canon(prov_value or "SEMUA").upper() or "SEMUA")
         kab_slug = (_canon(kab_value or "SEMUA").upper() or "SEMUA")
@@ -1428,15 +1630,13 @@ def run_calc(prov_value, kab_value, kew_value, df_all, df_raw, pop_kab, pop_prov
         verif_total.to_excel(p_verif, index=False)
 
         wilayah_txt = kab_value if (kab_value and kab_value != "(Semua)") else (prov_value if (prov_value and prov_value != "(Semua)") else "Nasional/All")
-
-        # ✅ LLM merujuk TABEL RINGKASAN (summary_jenis)
-        analysis_text = generate_llm_analysis(summary_jenis, wilayah_txt, kew_value or "(Semua)")
+        analysis_text = generate_llm_analysis(summary_jenis, agg_total, verif_total, wilayah_txt, kew_value or "(Semua)")
         word_path = generate_word_report(wilayah_txt, summary_jenis, analysis_text)
 
         msg = (
-            f"✅ Selesai (TARGET {TARGET_RATIO*100:.2f}%): raw={len(raw)} | entitas={len(detail_view)} | "
+            f"Selesai (TARGET {TARGET_RATIO*100:.2f}%): raw={len(raw)} | entitas={len(detail_view)} | "
             f"wilayah(keseluruhan)={len(agg_total)} | jenis(agregat)={len(agg_jenis_full)}"
-            + ("" if DOCX_AVAILABLE else "<br>⚠️ python-docx tidak tersedia → laporan Word dimatikan.")
+            + ("" if DOCX_AVAILABLE else "\npython-docx tidak tersedia -> laporan Word dimatikan.")
         )
 
         return (
@@ -1448,7 +1648,7 @@ def run_calc(prov_value, kab_value, kew_value, df_all, df_raw, pop_kab, pop_prov
         )
 
     except Exception as e:
-        return _empty_outputs(f"⚠️ Runtime error: {repr(e)}")
+        return _empty_outputs(f"Runtime error: {repr(e)}")
 
 
 # ============================================================
@@ -1493,34 +1693,29 @@ def on_prov_change(prov_value):
 
 
 with gr.Blocks() as demo:
-    gr.Markdown(f"""
-# IPLM 2025 — Final (Target Sampel **33.88%** per Jenis) — TANPA Kinerja Relatif / Percentile
-
-**Mode NO UPLOAD (cache aktif).** File dibaca dari repo/server:
-target_pct = TARGET_RATIO * 100
+    target_pct = TARGET_RATIO * 100.0
 
-with gr.Blocks() as demo:
     gr.Markdown(f"""
 # IPLM 2025 — Final (Target Sampel **{target_pct:.2f}%** per Jenis) — TANPA Kinerja Relatif / Percentile
 
-**Mode NO UPLOAD (cache aktif).** File dibaca dari repo/server:
-- `DATA_FILE`   = **{DATA_FILE}**
-- `POP_KAB`     = **{POP_KAB}**
-- `POP_PROV`    = **{POP_PROV}**
-- `POP_KHUSUS`  = **{POP_KHUSUS}**
+Mode NO UPLOAD (cache aktif). File dibaca dari repo/server:
+- DATA_FILE   = {DATA_FILE}
+- POP_KAB     = {POP_KAB}
+- POP_PROV    = {POP_PROV}
+- POP_KHUSUS  = {POP_KHUSUS}
 
-**TARGET RATIO (per jenis): {target_pct:.2f}%**
+TARGET RATIO (per jenis): {target_pct:.2f}%
 
-Dashboard KPI:
+Dashboard KPI menampilkan:
 - Indeks IPLM FINAL (disesuaikan {target_pct:.2f}%)
 - Indeks Dasar (tanpa penyesuaian)
 
 Ringkasan (Jenis + Keseluruhan) memuat:
-- Rata2_sub_koleksi, Rata2_sub_sdm, Rata2_sub_pelayanan, Rata2_sub_pengelolaan
-- Rata2_dim_kepatuhan, Rata2_dim_kinerja
+- Pop/Target/Terkumpul/Coverage + Penyesuaian
+- Rata2 sub/dim: koleksi, sdm, pelayanan, pengelolaan, dim_kepatuhan, dim_kinerja
 
-LLM untuk laporan Word wajib merujuk tabel ringkasan tersebut.
-""")
+Analisis otomatis (opsional) wajib merujuk tabel ringkasan tersebut.
+""".strip())
 
     state_df = gr.State(None)
     state_raw = gr.State(None)
@@ -1540,22 +1735,21 @@ LLM untuk laporan Word wajib merujuk tabel ringkasan tersebut.
 
     run_btn = gr.Button("Jalankan Perhitungan")
     msg_out = gr.Markdown()
-
     kpi_out = gr.Markdown()
 
-    gr.Markdown("## Ringkasan (Jenis + Keseluruhan) — Pop/Target33.88/Terkumpul/Coverage + Penyesuaian + Rata2 sub/dim")
+    gr.Markdown("## Ringkasan (Jenis + Keseluruhan) — Pop/Target/Terkumpul/Coverage + Penyesuaian + Rata2 sub/dim")
     out_summary = gr.DataFrame(interactive=False)
 
     gr.Markdown("## Agregat Wilayah (Keseluruhan) — FIX avg3")
     out_agg_total = gr.DataFrame(interactive=False)
 
-    gr.Markdown("## Agregat Wilayah × Jenis — (ditampilkan sampai Indeks_Dasar_Agregat_0_100)")
+    gr.Markdown("## Agregat Wilayah × Jenis — ditampilkan sampai Indeks_Dasar_Agregat_0_100")
     out_agg_jenis = gr.DataFrame(interactive=False)
 
     gr.Markdown("## Detail Entitas (Final menempel dari wilayah)")
     out_detail = gr.DataFrame(interactive=False)
 
-    gr.Markdown("## Kecukupan Sampel 33.88%")
+    gr.Markdown("## Kecukupan Sampel (Target per jenis)")
     out_verif = gr.DataFrame(interactive=False)
 
     gr.Markdown("## Bell Curve — Indeks Dasar per Entitas (per Jenis) + Nama Perpustakaan")
@@ -1568,7 +1762,7 @@ LLM untuk laporan Word wajib merujuk tabel ringkasan tersebut.
     gr.Markdown("### Perpustakaan Khusus")
     bell_khusus = gr.Plot(scale=1)
 
-    gr.Markdown("## Analisis Otomatis (LLM) — Rujukan Utama: Tabel Ringkasan")
+    gr.Markdown("## Analisis Otomatis (opsional)")
     analysis_out = gr.Markdown()
 
     with gr.Row():
@@ -1596,4 +1790,4 @@ LLM untuk laporan Word wajib merujuk tabel ringkasan tersebut.
         outputs=[state_df, state_raw, state_pop_kab, state_pop_prov, state_pop_khusus, state_meta, info_box, dd_prov, dd_kab, dd_kew]
     )
 
-demo.launch()
+demo.launch()