diff --git a/CLAUDE.md b/CLAUDE.md
index 276e22b..84f9730 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -16,8 +16,8 @@
   **Framework7 9**（`framework7-react`，iOS 主题）+ Recharts 图表。根目录
   `package.json` 是 npm workspaces（含 `client`）。
 - **数据库**：可插拔数据层（`backend/db.py`），`DB_TYPE` 切换——
-  本地开发 SQLite（`backend/data/health.db`，默认）；生产 NAS MariaDB
-  （`garmin_health_lab` 库，连接配置见 NAS 上 `backend/.env`）。
+  本地开发 SQLite（`backend/data/health.db`，默认）；生产 MariaDB（甲骨文云主机
+  本机，`garmin_health_lab` 库，连接配置见该机 `/opt/garmin-health-lab/backend/.env`）。
 - **认证**：**auth-hub SSO**（OAuth2/OIDC，`services/auth_hub_client.py`），
   本地邮箱密码注册/登录已移除；应用内 JWT 由后端签发。
 - **AI**：自建 ai-gateway（OpenAI 兼容，`https://ai.zichuan.xyz/v1`）为唯一上游
@@ -57,7 +57,8 @@ GarminHealthLab/
 │   ├── src/lib/              # day.ts（本地日期）/ metrics.ts（指标注册表）
 │   ├── src/features.ts       # 功能开关（FEATURES.ai）
 │   └── .env.production       # REACT_APP_API_URL=/api（同源，防 Network Error）
-├── deploy/                   # NAS 部署脚本族（S99garmin/start/stop/push/deploy）
+├── deploy/                   # push_oracle.sh 现役；其余（S99garmin/start/stop/
+│                              #   push/deploy）是 NAS 时代脚本，已废弃保留参考
 ├── docs/                     # 架构 / 开发 / 需求文档
 ├── PROGRESS.md               # 进度与部署事实（排查必读）
 └── README.md                 # 项目说明 + API 文档
@@ -83,24 +84,42 @@ npm run build           # client/react-scripts build → client/build/
 cd backend && .venv/bin/python tests/smoke.py
 ```
 
-## 部署（生产 = NAS，端口 8124）
+## 部署（生产 = 甲骨文云主机，2026-09-12 起；此前是 NAS，已下线）
 
-- **位置**：NAS `192.168.50.64` `/volume1/web/garmin-health-lab`，root 用户跑
-  gunicorn `0.0.0.0:8124`（2 workers / 4 threads / --timeout 300），开机自启走
-  DSM 任务调度器执行 `deploy/S99garmin.sh`。
-- **一键部署**：本地 `./deploy/push.sh`（tar 经 ssh 同步 backend + deploy +
-  清空重推 client/build + sudo 重启，**校验 gunicorn pid 变化 + health 200**）。
-  前置：先 `npm run build`。
-- **公网**：NAS frpc → 甲骨文 `http://129.146.26.249:8124`（frp 重连需几秒）。
-- **DB**：NAS MariaDB 10.11，root 经 socket `/run/mysqld/mysqld10.sock`（或
-  TCP 127.0.0.1:3306），库 `garmin_health_lab`。10.11 dump 含 `/*M!999999`
-  注释、旧客户端会报错，且须 `--default-character-set=utf8mb4` 防中文丢失。
-- **auth-hub**：client `996aLPw4T5gl-rYZ`；回调注册了 LAN
-  `http://192.168.50.64:8124/auth/callback` 与公网
-  `http://129.146.26.249:8124/auth/callback` 两个地址。
+- **位置**：`129.146.26.249` `/opt/garmin-health-lab`，**`ubuntu` 用户**跑
+  gunicorn `127.0.0.1:5500`（2 workers / 4 threads / --timeout 300），
+  systemd 单元 `garmin-health-lab.service`（`enabled`，随机器开机自启，
+  `Restart=always`）。和这台机器上其它服务（ai-gateway / auth-hub / fam-edge）
+  同一套约定：`/opt/<项目>` 下独立部署 + 独立 venv + Caddy 按子域名反代。
+- **公网**：`https://garmin.zichuan.xyz` → Caddy → `127.0.0.1:5500`。
+  DNS（腾讯云 DNSPod，A 记录）与 TLS（Caddy 自动签发）都已配好。
+  ~~旧的 `http://129.146.26.249:8124`~~ 已下线（走 NAS frpc 转发，隧道已拆）。
+- **一键部署**：本地 `./deploy/push_oracle.sh`（tar 经 ssh key 认证同步
+  backend + 清空重推 client/build + pip install + systemctl restart，
+  **校验 main PID 变化 + health 200**）。前置：先 `npm run build`。
+  `deploy/push.sh` / `start.sh` / `stop.sh` / `S99garmin.sh` / `deploy.sh`
+  是 NAS 时代的脚本，已废弃保留仅供参考。
+- **DB**：甲骨文本机 MariaDB 10.3.39（`127.0.0.1:3306`），独立账号
+  `garmin`（**注意**：`garmin@localhost` 与 `garmin@127.0.0.1` 是两个不同账号，
+  必须同密码建两份，否则 TCP 连接用的是另一个密码），库 `garmin_health_lab`。
+  和其它项目（`chat_relay`、`zhongyuan`）共用同一个 MariaDB 实例，各自独立库
+  独立账号。**这台机器磁盘 96% 已满**，改动前留意剩余空间。
+- **AI 网关**：`AI_GATEWAY_BASE_URL` 现在是**本地回环** `http://127.0.0.1:5100/v1`
+  （不再经 Caddy/公网 —— 网关和这个服务同机了）。
+- **auth-hub**：client `996aLPw4T5gl-rYZ`；回调只保留
+  `https://garmin.zichuan.xyz/auth/callback` 一条（NAS/旧公网端口那几条已用
+  `manage_clients remove-redirect-uri` 清掉）。改注册用
+  `/opt/auth-hub` 下 `PYTHONPATH=/opt/auth-hub/src venv/bin/python -m
+  auth_hub.manage_clients`。
+- **NAS 现状**：`garmin_health_lab` 库已 `DROP DATABASE`（备份在本地
+  `~/Desktop/Work/backups/garmin_health_lab_nas_backup_20260912.sql.gz`，
+  20 张表逐条精确计数核对过一致后才删的），`S99garmin.sh` 开机项已移除，
+  frpc 配置里 `garmin-health` 转发段已删（`gitea`/`wordpress`/`fam-core`/
+  `nexusai` 那几条没动——**Gitea 还在 NAS 上，这个仓库的 git remote 仍然指
+  向它**，这次迁移只搬了应用和数据，没搬源码托管）。
 - **部署/排障前**：先读 `PROGRESS.md` 与 `deploy/` 脚本确认事实（曾经凭旧记忆
-  断言"无线上环境"而误判）。服务以 root 运行：重启用 sudo，日志
-  `logs/error.log`、`logs/access.log` 是 root 所有。
+  断言"无线上环境"而误判，后来又把"生产在 NAS"当成默认事实——**两次都错在
+  没有先查，部署位置是会变的**）。
 
 ## 关键约定与坑（写代码/改样式前看）
 
diff --git a/PROGRESS.md b/PROGRESS.md
index 479f334..c1d0ef6 100644
--- a/PROGRESS.md
+++ b/PROGRESS.md
@@ -10,13 +10,55 @@
 
 - [x] Gunicorn 生产服务器配置（2 workers / 4 threads）
 
-- [x] 部署到本地 NAS（`/volume1/web/garmin-health-lab`），端口 8124
+- [x] ~~部署到本地 NAS（`/volume1/web/garmin-health-lab`），端口 8124~~ 已下线，见下
 
-- [x] frp 隧道配置，通过甲骨文公网 IP 外网访问（`http://129.146.26.249:8124`）
+- [x] ~~frp 隧道配置，通过甲骨文公网 IP 外网访问（`http://129.146.26.249:8124`）~~ 已拆除
 
-- [x] 开机自启脚本（`deploy/S99garmin.sh`）
+- [x] ~~开机自启脚本（`deploy/S99garmin.sh`）~~ 已移除，脚本保留仅供参考
 
-- [x] 甲骨文 iptables 放行 8124 端口
+- [x] ~~甲骨文 iptables 放行 8124 端口~~ 该端口不再使用
+
+### 迁移到甲骨文云主机（2026-09-12）
+
+NAS 的局域网 IP 因 DHCP 重启换过两次（`.64` → `.65` → 又变回 `.64`），且磁盘/
+网络都不如已经稳定跑着好几个服务的甲骨文机器可靠。整体搬迁，NAS 只保留 Gitea
+（这个仓库的源码托管，未动）。
+
+- [x] 甲骨文本机新建 MariaDB 库 `garmin_health_lab` + 独立账号 `garmin`（同时建
+  `garmin@localhost` 与 `garmin@127.0.0.1` 两个 host 变体、同密码——MariaDB 视
+  为两个不同账号，只建一个会导致 TCP 连接失败）
+
+- [x] NAS `mysqldump`（10.11 源库，处理了 `/*M!999999` 注释兼容 10.3 目标）→
+  导入甲骨文 MariaDB 10.3.39，**20 张表逐条精确 `COUNT(*)` 比对完全一致**才继续
+
+- [x] 应用部署到 `/opt/garmin-health-lab`，systemd 单元 `garmin-health-lab.service`
+  （`ubuntu` 用户，非 root；和 ai-gateway/auth-hub 同一套约定）
+
+- [x] DNS：腾讯云 DNSPod API 新增 `garmin.zichuan.xyz` A 记录；Caddy 反代 +
+  自动签发 TLS，实测已生效
+
+- [x] auth-hub 回调地址改为 `https://garmin.zichuan.xyz/auth/callback`，清掉
+  NAS/旧公网端口那几条历史回调
+
+- [x] AI 网关配置从公网域名改本地回环 `http://127.0.0.1:5100/v1`（网关和本服务
+  现在同机），**触发真实生成验证通过**（`upstream: gemini`）
+
+- [x] 迁移窗口期加了安全阀：甲骨文这边先 `AUTO_SYNC=false`，避免和 NAS 的调度
+  器同时刷新 Garmin 令牌撞上按账号计算的 SSO 限流（见下方"同步逻辑修复"一节
+  09-03 那次事故）。确认 NAS 已冻结、数据无缺口后才在甲骨文打开
+
+- [x] 冻结 NAS（停 gunicorn）→ 补一次终态 dump 核对无数据差异 → 移除开机自启
+  → `DROP DATABASE garmin_health_lab`（备份在本地
+  `~/Desktop/Work/backups/garmin_health_lab_nas_backup_20260912.sql.gz`）
+
+- [x] NAS frpc 配置精确删除 `garmin-health` 转发段，`gitea`/`wordpress`/
+  `fam-core`/`nexusai` 几条未动，改后逐条复检确认其它站点未受影响
+
+- [x] 新部署脚本 `deploy/push_oracle.sh`（key 认证 + systemd，校验 PID 变化 +
+  health 200），落地当场跑通一次真实部署
+
+- [x] `CLAUDE.md` 部署章节整体重写；NAS 时代的 `deploy/` 脚本加了废弃说明保留
+  参考，未删除
 
 ### 认证
 
diff --git a/README.md b/README.md
index 2d21e71..f0d2261 100644
--- a/README.md
+++ b/README.md
@@ -20,9 +20,9 @@
 
 ### 后端
 - Python 3.10+ + Flask（应用工厂）
-- Gunicorn（生产运行，NAS :8124）
-- 可插拔数据层：SQLite（本地开发）/ MariaDB（生产，NAS 10.11 库
-  `garmin_health_lab`，经 PyMySQL/socket）
+- Gunicorn（生产运行，甲骨文云主机 :5500，Caddy 反代出 `garmin.zichuan.xyz`）
+- 可插拔数据层：SQLite（本地开发）/ MariaDB（生产，与应用同机的 10.3.39 库
+  `garmin_health_lab`，经 PyMySQL/TCP）
 - JWT 鉴权；登录走 auth-hub 统一 SSO（OAuth2/OIDC，本地密码登录已移除）
 - garminconnect（Garmin API 集成）
 
@@ -93,8 +93,8 @@ CORS_ORIGIN=http://localhost:3000,http://localhost:5173
 数据层通过 `DB_TYPE` 环境变量切换后端，**业务代码无需改动**：
 
 - **SQLite（默认，本地开发）**：零配置，由 `DATABASE_PATH` 指定文件位置。
-- **MariaDB（生产）**：运行于 NAS（192.168.50.64）本地 MariaDB 10.11，root 经
-  socket `/run/mysqld/mysqld10.sock`（或 TCP `127.0.0.1:3306`）连接（PyMySQL），
+- **MariaDB（生产）**：运行于甲骨文云主机（`129.146.26.249`）本地 MariaDB
+  10.3.39，独立账号 `garmin` 经 TCP `127.0.0.1:3306` 连接（PyMySQL），
   独立库 `garmin_health_lab`。完整配置见 `backend/.env.example`：
 
 ```env
@@ -195,15 +195,17 @@ python tests/smoke.py
 > 轮询，趋势归因与 Copilot 走显式触发；任一模型失败时降级为规则引擎，
 > `meta.source` 会说明本次由谁作答。
 
-## 🚀 部署（生产 = NAS）
+## 🚀 部署（生产 = 甲骨文云主机，2026-09-12 起）
 
-- **位置**：NAS `192.168.50.64` `/volume1/web/garmin-health-lab`，root 跑
-  gunicorn `0.0.0.0:8124`（2 workers / 4 threads / --timeout 300），开机自启
-  DSM 任务调度器执行 `deploy/S99garmin.sh`。
-- **公网**：NAS frpc → `http://129.146.26.249:8124`。
-- **一键部署**：先 `npm run build`，再 `./deploy/push.sh`（同步 backend +
-  deploy + 清空重推 `client/build` + sudo 重启 + pid/health 双校验）。
-- **部署排障前**：先读 `PROGRESS.md` 与 `deploy/` 脚本确认事实。
+- **位置**：`129.146.26.249` `/opt/garmin-health-lab`，`ubuntu` 用户跑
+  gunicorn `127.0.0.1:5500`（2 workers / 4 threads / --timeout 300），
+  systemd 单元 `garmin-health-lab.service`。
+- **公网**：`https://garmin.zichuan.xyz` → Caddy → `127.0.0.1:5500`。
+- **一键部署**：先 `npm run build`，再 `./deploy/push_oracle.sh`（ssh key 同步
+  backend + 清空重推 `client/build` + pip install + systemctl restart +
+  PID/health 双校验）。
+- **部署排障前**：先读 `PROGRESS.md` 与 `deploy/` 脚本确认事实——部署位置换过
+  一次（NAS → 甲骨文），凭记忆断言曾经出过错。
 
 ## 🔐 安全说明
 
diff --git a/backend/.env.example b/backend/.env.example
index ac53126..d66eac3 100644
--- a/backend/.env.example
+++ b/backend/.env.example
@@ -8,14 +8,15 @@ DB_TYPE=sqlite
 # SQLite file (used when DB_TYPE=sqlite)
 DATABASE_PATH=./data/health.db
 
-# MariaDB (used when DB_TYPE=mariadb) — production DB on the NAS
-# (192.168.50.64, MariaDB 10.11). Connection is over the socket
-# /run/mysqld/mysqld10.sock (or TCP 127.0.0.1:3306) as root; the socket path
-# only matters when TCP auth is disabled for the app user.
-# MARIADB_SOCKET=/run/mysqld/mysqld10.sock
+# MariaDB (used when DB_TYPE=mariadb) — production DB is co-located with the
+# app on the Oracle box (129.146.26.249, MariaDB 10.3.39), TCP 127.0.0.1:3306,
+# dedicated account (not root). MariaDB treats `user@localhost` and
+# `user@127.0.0.1` as two different accounts — if you create this user by
+# hand, create both host variants with the same password, or TCP connections
+# fail with a password that looks right but is not the one that account has.
 # MARIADB_HOST=127.0.0.1
 # MARIADB_PORT=3306
-# MARIADB_USER=root
+# MARIADB_USER=garmin
 # MARIADB_PASSWORD=your_production_mariadb_password
 # MARIADB_DATABASE=garmin_health_lab
 
@@ -38,9 +39,10 @@ AUTH_HUB_CLIENT_ID=your_client_id
 AUTH_HUB_CLIENT_SECRET=your_client_secret
 #
 # Callback URL (must exactly match what's registered in auth-hub). Production
-# registers both the public frp address (129.146.26.249:8124) and the LAN
-# address (192.168.50.64:8124).
-AUTH_HUB_REDIRECT_URI=http://129.146.26.249:8124/auth/callback
+# registers https://garmin.zichuan.xyz/auth/callback — add/remove redirect
+# URIs on that client with /opt/auth-hub's manage_clients CLI, not by editing
+# auth-hub's own database directly.
+AUTH_HUB_REDIRECT_URI=http://127.0.0.1:5500/auth/callback
 
 # --- CORS (comma-separated allowed front-end origins) ---
 # localhost stays in the production list on purpose: CORS is not an auth
@@ -57,9 +59,12 @@ CORS_ORIGIN=http://localhost:3000,http://localhost:5173
 # absorbs single-vendor quota limits. Reached directly, bypassing any local
 # HTTP proxy. NOTE: its NVIDIA upstream is a large reasoning model — replies
 # can take 2-3 minutes, so set AI_TIMEOUT_SECONDS accordingly.
-# HTTPS (Caddy, strips the /ai prefix) rather than http://…:5100 — the token
-# rides in an Authorization header and should not cross the internet in clear.
-AI_GATEWAY_BASE_URL=https://ai.zichuan.xyz/v1
+# Production (co-located with ai-gateway on the same Oracle box) uses the
+# loopback address — skips Caddy and the public hop entirely, and the token
+# never leaves localhost either way. Use the HTTPS domain instead only when
+# this app runs somewhere else: the token rides in an Authorization header
+# and must not cross the public internet in the clear.
+AI_GATEWAY_BASE_URL=http://127.0.0.1:5100/v1
 AI_GATEWAY_TOKEN=
 AI_GATEWAY_MODEL=ai-gateway-auto
 
diff --git a/backend/config.py b/backend/config.py
index 0845a0a..e4260ef 100644
--- a/backend/config.py
+++ b/backend/config.py
@@ -42,13 +42,13 @@ JWT_EXPIRY_DAYS = int(os.environ.get("JWT_EXPIRY_DAYS") or 7)
 # hardcoded in source; put it in backend/.env (gitignored) instead.
 AUTH_HUB_BASE_URL = os.environ.get("AUTH_HUB_BASE_URL") or "http://129.146.26.249:5300"
 # Dev fallback client id (auth-hub keeps dev and prod clients in separate
-# databases). Production always overrides this via backend/.env — the value in
-# use on the NAS is the registered client for the :8124 callbacks.
+# databases). Production always overrides this via backend/.env — the value
+# in use is the client registered for https://garmin.zichuan.xyz/auth/callback.
 AUTH_HUB_CLIENT_ID = os.environ.get("AUTH_HUB_CLIENT_ID") or "0asGO0FdX_XYOk6O"
 AUTH_HUB_CLIENT_SECRET = os.environ.get("AUTH_HUB_CLIENT_SECRET") or ""
-# Production callback goes through the NAS frp tunnel to the public address
-# (129.146.26.249:8124); the LAN callback 192.168.50.64:8124 is registered too.
-AUTH_HUB_REDIRECT_URI = os.environ.get("AUTH_HUB_REDIRECT_URI") or "http://129.146.26.249:8124/auth/callback"
+# Production sets this via backend/.env to https://garmin.zichuan.xyz/auth/callback
+# — the fallback below is a dev-only placeholder, not a live production route.
+AUTH_HUB_REDIRECT_URI = os.environ.get("AUTH_HUB_REDIRECT_URI") or "http://127.0.0.1:5500/auth/callback"
 
 # --- Static UI --------------------------------------------------------------
 # Directory holding the built React app. When set and populated, the Flask
diff --git a/deploy/S99garmin.sh b/deploy/S99garmin.sh
index 69b4b97..946829a 100755
--- a/deploy/S99garmin.sh
+++ b/deploy/S99garmin.sh
@@ -1,4 +1,6 @@
 #!/bin/sh
+# RETIRED 2026-09-12: production moved off the NAS to the Oracle box
+# (systemd unit garmin-health-lab.service there). Kept for reference only.
 # Garmin Health Lab — DSM boot script.
 # Mirrors the convention already used by S99frpc.sh on this NAS.
 APP=/volume1/web/garmin-health-lab
diff --git a/deploy/deploy.sh b/deploy/deploy.sh
index 7e38aec..4c034e2 100644
--- a/deploy/deploy.sh
+++ b/deploy/deploy.sh
@@ -1,4 +1,6 @@
 #!/bin/sh
+# RETIRED 2026-09-12: production moved off the NAS to the Oracle box
+# (systemd unit garmin-health-lab.service there). Kept for reference only.
 # Deploy the latest local code to NAS.
 # Usage: sh deploy/deploy.sh
 # Prerequisites: client has been built (`npm run build` in client/)
diff --git a/deploy/push.sh b/deploy/push.sh
index 8608693..2b81caa 100755
--- a/deploy/push.sh
+++ b/deploy/push.sh
@@ -1,4 +1,13 @@
 #!/bin/sh
+# RETIRED 2026-09-12: production moved off the NAS to the Oracle box — see
+# deploy/push_oracle.sh. The NAS's garmin_health_lab database has been
+# DROPPED and its S99garmin.sh boot entry removed; this script now has
+# nothing live to deploy to. Kept for reference (the tar-over-ssh dance for
+# a password-only host may be useful again) — do not run it expecting it to
+# reach a running service.
+#
+# --- original header below ---
+#
 # Push this working tree to the NAS and restart it.
 #
 # The NAS only accepts password auth, so one ssh master connection is opened
diff --git a/deploy/push_oracle.sh b/deploy/push_oracle.sh
new file mode 100755
index 0000000..187f9d0
--- /dev/null
+++ b/deploy/push_oracle.sh
@@ -0,0 +1,65 @@
+#!/bin/sh
+# Push this working tree to the Oracle box and restart it.
+#
+# Unlike deploy/push.sh (the retired NAS deployment), this target uses key
+# auth and systemd — matching the convention every other service on that box
+# already follows (ai-gateway, auth-hub, fam-edge, ...): code lives under
+# /opt/<name>, gunicorn is a systemd unit, Caddy reverse-proxies a subdomain
+# to a local port. Nothing here needs a password or sudo for the deploy
+# itself; only the one-time systemd/Caddy setup did, and that is already done.
+#
+#   ./deploy/push_oracle.sh [user@host] [ssh-key]
+#
+# Never touches backend/.env, .venv or the database on the far side.
+set -e
+
+HOST="${1:-ubuntu@129.146.26.249}"
+KEY="${2:-$HOME/.ssh/oracle_new}"
+APP=/opt/garmin-health-lab
+REPO="$(cd "$(dirname "$0")/.." && pwd)"
+
+sh_() { ssh -i "$KEY" -o BatchMode=yes "$HOST" "$@"; }
+
+if [ ! -f "$REPO/client/build/index.html" ]; then
+  echo "client/build is missing — run 'npm run build' first" >&2
+  exit 1
+fi
+
+# macOS bsdtar writes com.apple.provenance xattrs and ._ resource forks that
+# the Linux side cannot read; --no-xattrs plus COPYFILE_DISABLE strip them.
+TAR="tar czf - --no-xattrs"
+export COPYFILE_DISABLE=1
+
+echo "==> backend"
+$TAR --exclude .venv --exclude .env --exclude __pycache__ \
+    --exclude '*.db' --exclude tests --exclude .pytest_cache --exclude static \
+    -C "$REPO/backend" . | sh_ "tar xzf - -C '$APP/backend'"
+
+echo "==> static (cleared first, so stale JS chunks do not pile up)"
+sh_ "rm -rf '$APP/backend/static' && mkdir -p '$APP/backend/static'"
+$TAR -C "$REPO/client/build" . | sh_ "tar xzf - -C '$APP/backend/static'"
+
+echo "==> pip install (in case requirements.txt changed)"
+sh_ "$APP/backend/.venv/bin/pip install -q -r $APP/backend/requirements.txt"
+
+echo "==> restart"
+BEFORE=$(sh_ "systemctl show garmin-health-lab -p MainPID --value" || true)
+sh_ "sudo systemctl restart garmin-health-lab"
+sleep 2
+AFTER=$(sh_ "systemctl show garmin-health-lab -p MainPID --value" || true)
+if [ -z "$AFTER" ] || [ "$AFTER" = "0" ]; then
+  echo "service failed to come back up — check journalctl -u garmin-health-lab" >&2
+  exit 1
+fi
+if [ "$BEFORE" = "$AFTER" ]; then
+  echo "the main PID did not change ($AFTER) — the old process may still be" >&2
+  echo "serving and your changes are NOT live." >&2
+  exit 1
+fi
+echo "==> restarted: $BEFORE -> $AFTER"
+
+echo "==> health"
+sh_ "curl -sf -m 5 -o /dev/null -w 'local: %{http_code}\n' http://127.0.0.1:5500/api/health/status" \
+  || echo "local: unreachable"
+curl -sf -m 8 -o /dev/null -w "public (https://garmin.zichuan.xyz): %{http_code}\n" \
+  https://garmin.zichuan.xyz/api/health/status || echo "public: unreachable"
diff --git a/deploy/start.sh b/deploy/start.sh
index ae2f553..fc92423 100755
--- a/deploy/start.sh
+++ b/deploy/start.sh
@@ -1,4 +1,6 @@
 #!/bin/sh
+# RETIRED 2026-09-12: production moved off the NAS to the Oracle box
+# (systemd unit garmin-health-lab.service there). Kept for reference only.
 # Start Garmin Health Lab. Safe to run repeatedly: an already-running
 # instance is stopped first. Intended for DSM Task Scheduler (boot-up).
 APP="$(cd "$(dirname "$0")/.." && pwd)"
diff --git a/deploy/stop.sh b/deploy/stop.sh
index f8da08d..8fdabbb 100755
--- a/deploy/stop.sh
+++ b/deploy/stop.sh
@@ -1,4 +1,6 @@
 #!/bin/sh
+# RETIRED 2026-09-12: production moved off the NAS to the Oracle box
+# (systemd unit garmin-health-lab.service there). Kept for reference only.
 APP="$(cd "$(dirname "$0")/.." && pwd)"
 GUNICORN="$APP/backend/.venv/bin/gunicorn"