diff --git a/.dockerignore b/.dockerignore
new file mode 100644
index 0000000..eec608d
--- /dev/null
+++ b/.dockerignore
@@ -0,0 +1,9 @@
+# Artifak runtime & env lokal — jangan masuk build context
+venv/
+.git/
+.env
+voters.db
+.oc/
+__pycache__/
+*.pyc
+*.log
diff --git a/.env.example b/.env.example
index aab6e8d..83fcda0 100644
--- a/.env.example
+++ b/.env.example
@@ -42,3 +42,17 @@ DATABISNIS_API=https://api.databisnis.id
 
 # Dashboard: TTL cache (detik) untuk saldo liquid — gagal fetch juga di-cooldown
 DASH_LIQUID_TTL=60
+
+# — Stack swarm (docker-compose.yml) —
+# Interpolasi compose; override default di bawah lewat .env bila perlu.
+# TZ dipakai juga sebagai zona waktu jadwal skan (`datetime.now()` lokal).
+TZ=Asia/Jakarta
+SCAN_HOURS=0,8,16
+# Skan sekali saat container start (tabel voters langsung ada; dashboard ⊥ 500)
+SCAN_RUN_ON_START=1
+
+# Kredensial mariadb stack — wajib konsisten antar-service (mariadb/web/scan)
+MARIADB_ROOT_PASSWORD=root
+MARIADB_DATABASE=databisnisid
+MARIADB_USER=databisnis
+MARIADB_PASSWORD=databisnis
diff --git a/AGENTS.md b/AGENTS.md
index 76d64e9..0e4c28d 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -7,11 +7,12 @@ Two tools scan the Vexanium blockchain voters table for accounts whose only vote
 - Node reference script: `node get_voters.js` (no deps, Node 18+).
 - Python tool (production): `./venv/bin/python get_voters.py`. venv is Python 3.12, deps `requests` + `flask` + `python-dotenv` + `gunicorn` + `pymysql` (mysql backend only) (`requirements.txt`). Install with `./venv/bin/pip install -r requirements.txt`.
 - Storage backend: `db.py` abstracts it. Default `sqlite` (`VEX_DB_PATH`, stdlib `sqlite3`, WAL). Optional `mysql` (`VEX_DB_BACKEND=mysql` + `VEX_DB_HOST/PORT/USER/PASS/NAME`, PyMySQL). Oracle tests stay on sqlite; `test_mariadb.py` is opt-in (skips unless `VEX_DB_BACKEND=mysql`). Query SQL is written once with `%s` placeholders (translated to `?` for sqlite); `db.query` always returns a list.
-- Test MariaDB/MySQL via docker: `docker compose up -d` (mariadb:11 container `databisnisid-mariadb`, localhost-only `127.0.0.1:3306`, db/user/pass `databisnisid`/`databisnis`/`databisnis`, named volume, healthcheck). Stop/remove with `docker compose down`; wipe data with `docker compose down -v`. Verify with `docker compose exec mariadb mariadb -u databisnis -pdatabisnis databisnisid -e 'SELECT 1'`. Smoke against the container:
+- Test MariaDB/MySQL via docker: `docker compose -f docker-compose.dev.yml up -d` (mariadb:11 container `databisnisid-mariadb`, localhost-only `127.0.0.1:3306`, db/user/pass `databisnisid`/`databisnis`/`databisnis`, named volume, healthcheck). Stop/remove with `docker compose -f docker-compose.dev.yml down`; wipe data with `docker compose -f docker-compose.dev.yml down -v`. Verify with `docker compose -f docker-compose.dev.yml exec mariadb mariadb -u databisnis -pdatabisnis databisnisid -e 'SELECT 1'`. Smoke against the container:
   `VEX_DB_BACKEND=mysql VEX_DB_HOST=127.0.0.1 VEX_DB_PORT=3306 VEX_DB_USER=databisnis VEX_DB_PASS=databisnis VEX_DB_NAME=databisnisid ./venv/bin/python test_mariadb.py`
+- Docker Swarm (production stack): images are registry-pushed `git.proit.id/proitlab/databisnisid-web` + `databisnisid-scan` — build & push them first (`docker build -t git.proit.id/proitlab/databisnisid-web . && docker push git.proit.id/proitlab/databisnisid-web`, same for the scan image), then from a swarm manager run `docker stack deploy -c docker-compose.yml databisnisid`. Stack = mariadb (internal) + web (gunicorn dashboard, published `:5000`) + scan (`scan_loop.py`, runs `get_voters.py` at `SCAN_HOURS` default `0,8,16`, local timezone `TZ` default `Asia/Jakarta`, plus one scan at container start via `SCAN_RUN_ON_START`). `mariadb` is pinned by `placement.constraints: node.hostname == server5.saltis.id` because its data lives in the host bind mount `/data/db/mariadb/databisnisid/data` on that node; `web`/`scan` can run on any node and reach it over the overlay network `appnet`. Inspect: `docker stack services databisnisid`, `docker service logs databisnisid_scan`, `docker stack rm databisnisid`. `docker stack deploy` ignores `build:` (images must already be in the registry) and ignores `env_file` (env is inlined with `${VAR}` interpolation from `.env`). The local dev box has no swarm anymore (torn down) — if you re-init one there, the mariadb constraint leaves that task **Pending** since no node is named `server5.saltis.id`.
 - Web dashboard (read-only, reads the store via `db.py`): production `./venv/bin/gunicorn -c gunicorn.conf.py dashboard:app` → http://127.0.0.1:5000/ (run from repo dir). Easier: `./run.sh` (same command, works from any cwd, `$@` passed through). `gunicorn.conf.py` imports `config` → `.env` honored; `DASH_WORKERS` (default 2) controls workers, `DASH_HOST`/`DASH_PORT` the bind. Dev server (single-process) still works via `./venv/bin/python dashboard.py`. Paged 50/page, sorted `staked DESC`, live owner search (`/api/search`). Data freshness comes from the daily scan run — the dashboard never scans.
 - Config: all tunables load from env / `.env` via `config.py` (python-dotenv): `VEX_TARGET_BP`, `VEX_API_NODE`, `VEX_DB_PATH`, `VEX_DB_BACKEND`, `VEX_DB_HOST`, `VEX_DB_PORT`, `VEX_DB_USER`, `VEX_DB_PASS`, `VEX_DB_NAME`, `VEX_MIN_STAKED_VEX`, `DASH_PAGE_SIZE`, `DASH_HOST`, `DASH_PORT`, `DASH_WORKERS`, `VEX_STALE_DAYS`, `DATABISNIS_API`, `DASH_LIQUID_TTL`. Copy `.env.example` → `.env` to override; `.env` is gitignored. Chain constants (`vexcore`/scope/table) stay hardcoded.
-- Syntax check: `node --check get_voters.js`, `./venv/bin/python -m py_compile config.py get_voters.py dashboard.py db.py gunicorn.conf.py test_get_voters.py test_dashboard.py test_mariadb.py`.
+- Syntax check: `node --check get_voters.js`, `./venv/bin/python -m py_compile config.py get_voters.py dashboard.py db.py gunicorn.conf.py scan_loop.py test_get_voters.py test_dashboard.py test_mariadb.py`.
 - Tests (the verification oracles): `./venv/bin/python test_get_voters.py` and `./venv/bin/python test_dashboard.py` must both exit 0. They mock the network / use a temp DB — the live node is too flaky/slow for a full-scan test. Run after touching the relevant file.
 
 ## Key facts
@@ -36,7 +37,9 @@ Two tools scan the Vexanium blockchain voters table for accounts whose only vote
 - `dashboard.py` + `templates/index.html` + `static/style.css` + `static/app.js` — Flask web dashboard; reads the store via `db.py`, styled per `DESIGN.md`; `app.js` = debounced live owner search (fetch `/api/search`), degrades to the server-side `?q=` GET form if JS is off.
 - `gunicorn.conf.py` — gunicorn production config (bind/workers from `config`, sync worker). `run.sh` — launcher: `./run.sh` = `./venv/bin/gunicorn -c gunicorn.conf.py dashboard:app` from any cwd.
 - `config.py` — loads env/`.env` (python-dotenv) → `TARGET_BP`, `API_NODE`, `DB_PATH`, `DB_BACKEND`, `DB_HOST`, `DB_PORT`, `DB_USER`, `DB_PASS`, `DB_NAME`, `MIN_STAKED_VEX`, `PAGE_SIZE`, `DASH_HOST`, `DASH_PORT`, `DASH_WORKERS`, `VEX_STALE_DAYS`, `DATABISNIS_API`, `DASH_LIQUID_TTL`; shared by get_voters & dashboard.
-- `docker-compose.yml` — mariadb:11 test container (`databisnisid-mariadb`, localhost-only 3306) for the mysql backend.
+- `scan_loop.py` — scheduler dalam container stack: menunggu batas `SCAN_HOURS` (lokal via `TZ`), panggil `get_voters.main()`; skan-awal `SCAN_RUN_ON_START` + tunggu DB siap; loop tak pernah keluar.
+- `Dockerfile` — image web `databisnisid-web` (gunicorn dashboard; `DASH_HOST=0.0.0.0` di stack agar ingress menjangkaunya). `Dockerfile.scan` — image `databisnisid-scan` (scan_loop; sertakan `tzdata`). Di stack produksi keduanya di-push ke registry `git.proit.id/proitlab/databisnisid-{web,scan}` (⊥ `build:` di compose). `.dockerignore` — venv/.env/artifak tak masuk build context.
+- `docker-compose.yml` — STACK SWARM PRODUKSI (mariadb internal + web `:5000` + scan); image dari registry `git.proit.id/proitlab/databisnisid-*`; mariadb bind mount `/data/db/mariadb/databisnisid/data` + placement `node.hostname == server5.saltis.id`; network overlay `appnet`. `docker-compose.dev.yml` — mariadb uji lokal (`127.0.0.1:3306`, volume `mariadb_data`).
 - `DESIGN.md` — Bugatti austere style guide; the dashboard's CSS maps its tokens (canvas #000000, hairline #262626, weight 400 everywhere, fonts Saira Condensed / EB Garamond / JetBrains Mono).
 - `voters.db` — SQLite output (daily snapshot, gitignored in spirit).
 
diff --git a/Dockerfile b/Dockerfile
new file mode 100644
index 0000000..55a5fd2
--- /dev/null
+++ b/Dockerfile
@@ -0,0 +1,13 @@
+FROM python:3.12-slim
+ENV PYTHONUNBUFFERED=1 PIP_NO_CACHE_DIR=1
+WORKDIR /app
+
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+
+COPY config.py db.py dashboard.py gunicorn.conf.py ./
+COPY templates ./templates
+COPY static ./static
+
+EXPOSE 5000
+CMD ["gunicorn", "-c", "gunicorn.conf.py", "dashboard:app"]
diff --git a/Dockerfile.scan b/Dockerfile.scan
new file mode 100644
index 0000000..22dcd9e
--- /dev/null
+++ b/Dockerfile.scan
@@ -0,0 +1,15 @@
+FROM python:3.12-slim
+ENV PYTHONUNBUFFERED=1 PIP_NO_CACHE_DIR=1
+
+# tzdata → TZ env (jadwal skan lokal, default Asia/Jakarta) bekerja di container
+RUN apt-get update && apt-get install -y --no-install-recommends tzdata \
+ && rm -rf /var/lib/apt/lists/*
+
+WORKDIR /app
+
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+
+COPY config.py db.py get_voters.py scan_loop.py ./
+
+CMD ["python", "scan_loop.py"]
diff --git a/SPEC.md b/SPEC.md
index 62af059..8f9ab89 100644
--- a/SPEC.md
+++ b/SPEC.md
@@ -8,7 +8,7 @@ Re-runnable scheduled. Python 3.12, venv `./venv`.
 - venv `./venv` exists (py3.12, only pip) — install deps there, ⊥ global
 - lib: `requests`, `flask`, `python-dotenv`, `gunicorn` ! pip-installed; `pymysql` utk backend mysql (opt-in); others stdlib
 - config via env/`.env` (`config.py`, python-dotenv); default = konstanta lama
-- db: `db.py` abstraksi backend — `sqlite` (default, stdlib `sqlite3` → `DB_PATH`, WAL → pembaca tak terblokir) atau `mysql` (MariaDB/MySQL via PyMySQL, `VEX_DB_*`); container uji via `docker-compose.yml`
+- db: `db.py` abstraksi backend — `sqlite` (default, stdlib `sqlite3` → `DB_PATH`, WAL → pembaca tak terblokir) atau `mysql` (MariaDB/MySQL via PyMySQL, `VEX_DB_*`); container uji via `docker-compose.dev.yml`
 - api: `API_NODE` (default `https://v2.vexascan.com:2096`, public, flaky → retry)
 - saldo liquid akun BP: API databisnis (`DATABISNIS_API`, default `https://api.databisnis.id`) — diambil dashboard, cache TTL `DASH_LIQUID_TTL` (default 60s), cooldown kegagalan
 - contract=scope=`vexcore` (≠ `vexio`, EOS convention ⊥) — ⊥ env-able
@@ -20,6 +20,7 @@ Re-runnable scheduled. Python 3.12, venv `./venv`.
 - voter basi (revote > `VEX_STALE_DAYS`) atau `last_vote` tak diketahui → dibuang di `normalize` (⊥ disimpan); hanya pemilih segar yang masuk DB
 - output/comments: Indonesian
 - run harian → ganti snapshot (DROP+INSERT tiap skan; ⊥ history/append)
+- produksi stack: Docker Swarm multi-node (node DB `server5.saltis.id`) — image dari registry `git.proit.id/proitlab/databisnisid-web` & `-scan` (⊥ `build:` di stack), deploy `docker stack deploy -c docker-compose.yml databisnisid`; mariadb bind mount `/data/db/mariadb/databisnisid/data` pd node `server5.saltis.id` (placement constraint `node.hostname == server5.saltis.id`), web/scan di node mana pun via overlay `appnet`; scan berjalan pd jam `SCAN_HOURS` (default 0,8,16, waktu lokal `TZ` default Asia/Jakarta) + skan-awal `SCAN_RUN_ON_START`
 
 ## §I — Interfaces
 api: POST `https://v2.vexascan.com:2096/v1/chain/get_table_rows`
@@ -30,7 +31,7 @@ cmd: `./venv/bin/python get_voters.py` → stdout summary (id-ID)
 web: GET `/` (Flask, disajikan gunicorn di produksi) → HTML spec-list, paged 50/halaman, `ORDER BY staked DESC, owner ASC`; ⊥ mutation (read-only)
 web: GET `/` + `?q=<substring>` → filter owner (server-side, no-JS fallback); pager bawa `q`
 web: GET `/api/search?q=` → JSON `{query,count,cap,results:[{owner,staked,weight,rank,last_vote}]}`, rank global, cap 500
-env: `VEX_TARGET_BP`, `VEX_API_NODE`, `VEX_DB_PATH`, `VEX_DB_BACKEND`, `VEX_DB_HOST`, `VEX_DB_PORT`, `VEX_DB_USER`, `VEX_DB_PASS`, `VEX_DB_NAME`, `VEX_MIN_STAKED_VEX`, `DASH_PAGE_SIZE`, `DASH_HOST`, `DASH_PORT`, `DASH_WORKERS`, `VEX_STALE_DAYS`, `DATABISNIS_API`, `DASH_LIQUID_TTL` — via `config.py` (`.env`)
+env: `VEX_TARGET_BP`, `VEX_API_NODE`, `VEX_DB_PATH`, `VEX_DB_BACKEND`, `VEX_DB_HOST`, `VEX_DB_PORT`, `VEX_DB_USER`, `VEX_DB_PASS`, `VEX_DB_NAME`, `VEX_MIN_STAKED_VEX`, `DASH_PAGE_SIZE`, `DASH_HOST`, `DASH_PORT`, `DASH_WORKERS`, `VEX_STALE_DAYS`, `DATABISNIS_API`, `DASH_LIQUID_TTL`, `SCAN_HOURS`, `SCAN_RUN_ON_START`, `TZ` — via `config.py` (`.env`)
 
 ## §V — Invariants
 V1: ∀ fetch → retry ≥3 on timeout/http err (like JS fetchWithRetry)
@@ -49,6 +50,7 @@ V13: `last_vote` = `2000-01-01 + log2(last_vote_weight / (staked×10000)) × 365
 V15: web list = kolom RANK/AKUN/STAKE (VEX)/VOTE TERAKHIR — ⊥ BOBOT SUARA, ⊥ penanda/stat/filter basi; stats = 3 sel (TOTAL PEMILIH/TOTAL VEX/STAKE TERTINGGI); `/api/search` hasil tetap `{owner,staked,weight,rank,last_vote}` (weight disimpan & di-API, ⊥ dirender)
 V16: saldo liquid akun `TARGET_BP` dari `GET {DATABISNIS_API}/v2/state/get_account` (parse `account.core_liquid_balance`) → sel stat SALDO LIQUID; cache in-memory per worker TTL `DASH_LIQUID_TTL`; gagal fetch → nilai lama (atau `—` bila belum pernah sukses) + `at` ikut diset (cooldown, ⊥ pukulan berulang); read-only, ⊥ tulis DB
 V17: penyimpanan lewat `db.py` (backend `VEX_DB_BACKEND` = `sqlite` default | `mysql`); query berbagi sintaks pakai placeholder `%s` (diterjemahkan `?` utk sqlite); LIKE escape pakai `ESCAPE '!'` — backslash memutus literal string MySQL (B2); `db.query` → list (fetchall sqlite=list / PyMySQL=tuple diseragamkan, B3); replace snapshot mysql = CREATE IF NOT EXISTS + DELETE + INSERT satu transaksi (MVCC → pembaca dapat snapshot konsisten saat ganti harian, tak kena torn read); oracle utama (test_get_voters/test_dashboard) tetap sqlite hermetik; `test_mariadb.py` opt-in (skip exit 0 bila `VEX_DB_BACKEND != mysql`)
+V18: stack produksi = Docker Swarm (`docker-compose.yml`, `docker stack deploy -c docker-compose.yml databisnisid`): mariadb bind mount `/data/db/mariadb/databisnisid/data` pd node `server5.saltis.id` + placement constraint `node.hostname == server5.saltis.id` (data node-lokal → mariadb harus selalu di node itu; di swarm yg ⊥ punya node itu task-nya Pending, bukan bug); web (gunicorn, publik `:5000`) + scan (`scan_loop.py`) di node mana pun, terhubung via overlay `appnet`; image dari registry `git.proit.id/proitlab/databisnisid-web` & `-scan` (⊥ `build:`); scan berjalan pd jam `SCAN_HOURS` (default 0,8,16, lokal `TZ` default Asia/Jakarta via `datetime.now()`) + skan-awal `SCAN_RUN_ON_START` (setelah DB siap) → tabel voters langsung ada (dashboard ⊥ 500); loop tak pernah keluar; `stack deploy` ⊥ `env_file` (env inline via `${VAR}` interpolasi `.env`); image web set `DASH_HOST=0.0.0.0` agar ingress menjangkau; healthcheck web = cek socket TCP (⊥ ketergantungan isi DB)
 
 ## §T — Tasks
 id|status|task|cites
@@ -69,6 +71,7 @@ T14|x|stale-vote: last_vote di scan (rasio bobot/stake); basi/unverifiable dibua
 T15|x|hapus bobot suara + UI basi di web (kolom/stats/penanda/filter `?stale=1`); oracle diperbarui|V15,I.web
 T16|x|saldo liquid akun di dashboard: API databisnis + cache TTL + cooldown; sel SALDO LIQUID|V16,I.web,I.env
 T17|x|backend shared db.py (sqlite|mysql, `%s`→`?`, ESCAPE '!'), config VEX_DB_*, refactor fetcher+dashboard, test_mariadb.py opt-in, runbook docker-compose|V17,I.db,I.env
+T18|x|stack swarm: Dockerfile web + Dockerfile.scan, scan_loop.py (SCAN_HOURS 0,8,16 lokal, skan-awal), docker-compose.yml stack (mariadb+web+scan), docker-compose.dev.yml utk uji lokal|V18,I.web,I.env
 
 ## §B — Bug log
 id|date|cause|fix
diff --git a/docker-compose.dev.yml b/docker-compose.dev.yml
new file mode 100644
index 0000000..80b34e4
--- /dev/null
+++ b/docker-compose.dev.yml
@@ -0,0 +1,22 @@
+services:
+  mariadb:
+    image: mariadb:11
+    container_name: databisnisid-mariadb
+    restart: unless-stopped
+    environment:
+      MARIADB_ROOT_PASSWORD: ${MARIADB_ROOT_PASSWORD:-root}
+      MARIADB_DATABASE: ${MARIADB_DATABASE:-databisnisid}
+      MARIADB_USER: ${MARIADB_USER:-databisnis}
+      MARIADB_PASSWORD: ${MARIADB_PASSWORD:-databisnis}
+    ports:
+      - "127.0.0.1:3306:3306"
+    volumes:
+      - mariadb_data:/var/lib/mysql
+    healthcheck:
+      test: ["CMD", "healthcheck.sh", "--connect", "--innodb_initialized"]
+      interval: 5s
+      timeout: 5s
+      retries: 12
+
+volumes:
+  mariadb_data:
diff --git a/docker-compose.yml b/docker-compose.yml
index 80b34e4..3a2ebe7 100644
--- a/docker-compose.yml
+++ b/docker-compose.yml
@@ -1,22 +1,93 @@
+# Stack Docker Swarm — deploy via `docker stack deploy -c docker-compose.yml databisnisid`
+#
+# 3 service: mariadb (storage, internal), web (dashboard gunicorn, publik :5000),
+# scan (skan get_voters.py pada SCAN_HOURS, waktu lokal TZ).
+#
+# Catatan swarm:
+# - `build:` hanya untuk `docker compose build`; `docker stack deploy` memakai `image:`
+#   (harus dibangun lebih dulu; multi-node perlu registry).
+# - `env_file` ⊥ didukung stack → environment inline dgn interpolasi `.env`.
+# - Tanpa `container_name`/`restart:` — pakai `deploy.restart_policy`.
+
 services:
   mariadb:
     image: mariadb:11
-    container_name: databisnisid-mariadb
-    restart: unless-stopped
     environment:
       MARIADB_ROOT_PASSWORD: ${MARIADB_ROOT_PASSWORD:-root}
       MARIADB_DATABASE: ${MARIADB_DATABASE:-databisnisid}
       MARIADB_USER: ${MARIADB_USER:-databisnis}
       MARIADB_PASSWORD: ${MARIADB_PASSWORD:-databisnis}
-    ports:
-      - "127.0.0.1:3306:3306"
     volumes:
-      - mariadb_data:/var/lib/mysql
+      - /data/db/mariadb/databisnisid/data:/var/lib/mysql
+    networks:
+      - appnet
     healthcheck:
       test: ["CMD", "healthcheck.sh", "--connect", "--innodb_initialized"]
       interval: 5s
       timeout: 5s
       retries: 12
+    deploy:
+      replicas: 1
+      placement:
+        constraints:
+          - node.hostname == server5.saltis.id
+      restart_policy:
+        condition: any
 
+  web:
+    image: git.proit.id/proitlab/databisnisid-web:latest
+    environment:
+      VEX_DB_BACKEND: mysql
+      VEX_DB_HOST: mariadb
+      VEX_DB_PORT: "3306"
+      VEX_DB_USER: ${MARIADB_USER:-databisnis}
+      VEX_DB_PASS: ${MARIADB_PASSWORD:-databisnis}
+      VEX_DB_NAME: ${MARIADB_DATABASE:-databisnisid}
+      DASH_HOST: "0.0.0.0"          # wajib di container → gunicorn bind terbuka utk ingress
+      DASH_PORT: "5000"
+      DASH_WORKERS: "2"
+      DATABISNIS_API: ${DATABISNIS_API:-https://api.databisnis.id}
+      DASH_LIQUID_TTL: "60"
+    ports:
+      - published: 5000
+        target: 5000
+    networks:
+      - appnet
+    healthcheck:
+      test: ["CMD", "python", "-c", "import socket,sys;sys.exit(0 if socket.create_connection(('127.0.0.1',5000),5) else 1)"]
+      interval: 30s
+      timeout: 5s
+      retries: 3
+      start_period: 20s
+    deploy:
+      replicas: 1
+      restart_policy:
+        condition: any
+
+  scan:
+    image: git.proit.id/proitlab/databisnisid-scan:latest
+    command: ["python", "scan_loop.py"]
+    environment:
+      VEX_DB_BACKEND: mysql
+      VEX_DB_HOST: mariadb
+      VEX_DB_PORT: "3306"
+      VEX_DB_USER: ${MARIADB_USER:-databisnis}
+      VEX_DB_PASS: ${MARIADB_PASSWORD:-databisnis}
+      VEX_DB_NAME: ${MARIADB_DATABASE:-databisnisid}
+      TZ: ${TZ:-Asia/Jakarta}
+      SCAN_HOURS: ${SCAN_HOURS:-0,8,16}
+      SCAN_RUN_ON_START: ${SCAN_RUN_ON_START:-1}
+    networks:
+      - appnet
+    deploy:
+      replicas: 1
+      restart_policy:
+        condition: any
+
+networks:
+  appnet:
+    driver: overlay
+
+# volume khusus stack — jangan pakai mariadb_data (milik docker-compose.dev.yml)
 volumes:
-  mariadb_data:
+  db_data:
diff --git a/scan_loop.py b/scan_loop.py
new file mode 100644
index 0000000..734ef3a
--- /dev/null
+++ b/scan_loop.py
@@ -0,0 +1,83 @@
+"""Jadwal skan berkala: jalankan get_voters.py pada jam SCAN_HOURS (waktu lokal).
+
+Menunggu hingga batas jam berikutnya dalam daftar `SCAN_HOURS` (default 0,8,16)
+lalu memanggil `get_voters.main()`. Loop tak pernah keluar — service dijaga
+hidup oleh restart_policy swarm. Gagal skan dicatat lalu lanjut ke batas
+berikutnya (tidak membunuh loop). `TZ` env menentukan jam lokal via
+`datetime.now()` (libc); image Dockerfile.scan sudah menyertakan tzdata.
+
+`SCAN_RUN_ON_START` (default 1): skan sekali saat container start setelah
+basis data siap — penting agar tabel voters langsung ada (dashboard ⊥ 500
+sebelum batas jam pertama).
+"""
+
+import os
+import time
+from datetime import datetime, timedelta
+
+import db
+import get_voters
+
+
+def parse_hours(raw):
+    """Daftar jam integer dari string CSV, diurutkan naik."""
+    hours = sorted({int(h) for h in raw.split(',') if h.strip()})
+    if not hours or not all(0 <= h <= 23 for h in hours):
+        raise ValueError(f'SCAN_HOURS tak valid: {raw!r}')
+    return hours
+
+
+def next_boundary(now, hours):
+    """datetime berikutnya pada jam-jam `hours` yang lebih besar dari `now`."""
+    for h in hours:
+        cand = now.replace(hour=h, minute=0, second=0, microsecond=0)
+        if cand > now:
+            return cand
+    # semua jam hari ini sudah lewat → besok jam pertama
+    nxt = now + timedelta(days=1)
+    return nxt.replace(hour=hours[0], minute=0, second=0, microsecond=0)
+
+
+def _wait_db(timeout=120):
+    """Tunggu basis data siap (mariadb baru boot) sebelum skan awal."""
+    start = time.monotonic()
+    while time.monotonic() - start < timeout:
+        try:
+            db.queryone('SELECT 1')
+            return
+        except Exception:
+            time.sleep(3)
+    raise RuntimeError('basis data tak siap setelah menunggu')
+
+
+def _run_scan(label):
+    print(f'[{label}] memulai skan...', flush=True)
+    try:
+        rc = get_voters.main()
+    except Exception as exc:
+        print(f'[Error] Skan gagal: {exc}', flush=True)
+        return
+    if rc != 0:
+        print(f'[Peringatan] get_voters keluar dengan kode {rc}', flush=True)
+
+
+def main():
+    hours = parse_hours(os.getenv('SCAN_HOURS', '0,8,16'))
+    run_on_start = os.getenv('SCAN_RUN_ON_START', '1').lower() not in ('0', 'false', 'no')
+    print(f'Jadwal skan: jam {", ".join(str(h) for h in hours)} (waktu lokal); '
+          f'skan-awal: {"ya" if run_on_start else "tidak"}')
+    if run_on_start:
+        _wait_db()
+        _run_scan('skan awal')
+    while True:
+        now = datetime.now()
+        target = next_boundary(now, hours)
+        wait = (target - now).total_seconds()
+        print(f'{now.isoformat(timespec="seconds")} tidur {wait / 3600:.2f} jam '
+              f'sampai {target.isoformat(timespec="seconds")}', flush=True)
+        time.sleep(max(wait, 0))
+        _run_scan('skan terjadwal')
+
+
+if __name__ == '__main__':
+    main()