U principle §6: bind mounts only, never named volumes

Promote 'don't use named volumes' from a mysql/redis-specific
explanation to a hard rule that applies to anything needing
persistence (ssh host keys, uploads, queue state, etc). Lead
with the 'docker compose down -v wipes named volumes' rationale
since that's the keystroke this rule is really protecting against.
Add a counter-pattern example.

PRINCIPLES/dockerization-and-deployment.md

@@ -393,30 +393,64 @@ that's why each one starts with the app's slug.
 
 ---
 
-## 6. Persistent data: `./docker-data/` in the repo, gitignored
+## 6. Persistent data: `./docker-data/` bind mounts, never named volumes
 
-Bind-mount mysql and redis storage to a `docker-data/` folder right next
+**Rule: any service that needs to keep state between container restarts
-to the source:
+gets a bind mount under `./docker-data/<service>/`. Never a named docker
+volume.** This applies to mysql, redis, ssh host keys, app uploads,
+queue state, every "this dir needs to survive" case — same shape, same
+location, no exceptions.
 
 ```
 docker-data/
   mysql/        # mysql:8.0 datadir
   redis/        # redis dump.rdb
+  …             # whatever else needs persistence
 ```
 
-The folder is gitignored (`docker-data/` in `.gitignore`). It survives
+The folder is gitignored (`docker-data/` in `.gitignore`). The deploy
-`docker compose down`. The deploy script `mkdir -p`s it on first run so
+script `mkdir -p`s it on first run so fresh boxes Just Work.
-fresh boxes Just Work.
 
-### Why not named docker volumes?
+### Why bind mounts, not named volumes
+
+The headline reason: **`docker compose down -v` wipes named volumes**.
+That command is in too many people's muscle memory ("nuke the stack and
+start clean") for the production datastore to be one accidental keystroke
+away from gone. Bind mounts under the repo are immune — `down -v`
+doesn't touch them.
+
+The rest of the rationale:
 
 - Trivially backed up — `tar -caf data.tar.xz docker-data/` from the
   repo root is the entire prod state.
 - Survives container/image churn including accidental
-  `docker volume prune -af`.
+  `docker volume prune -af` and `docker system prune --volumes`.
 - Discoverable — anyone with the repo can see where the data lives.
 - The repo path identifies which app owns the data when you have ten
   apps on one host.
+- Restoring on a new host is `git clone && rsync docker-data/` — no
+  per-volume `docker volume create && docker run --rm -v ... tar` dance.
+
+### Counter-pattern (do not do this)
+
+```yaml
+# WRONG — named volume; `docker compose down -v` deletes the data.
+services:
+  mysql:
+    volumes:
+      - mysql-data:/var/lib/mysql
+
+volumes:
+  mysql-data:
+```
+
+```yaml
+# RIGHT — bind mount; survives `down -v`.
+services:
+  mysql:
+    volumes:
+      - ./docker-data/mysql:/var/lib/mysql
+```
 
 ---