deploy: drop ineffective mkdir, document state-dir bootstrap

2 files changed, 4 insertions, 3 deletions

diff --git a/README.md b/README.md
index a8e8ec1..2550450 100644
--- a/README.md
+++ b/README.md
@@ -53,3 +53,5 @@ runs.jsonl         server run log; doubles as the RSS render source
 - Training stays on the local PC; the server only scores (low RAM).
 - Deploy `serve` only after a `train` run on enough organic stars shows an
   honest precision@15 in report.json; tiny validation sets lie.
+- Server state dir: `sudo install -d -o <user> -g <user> ~/fluxrec-data`
+  **before the first `compose up`** (Docker otherwise creates it as root).
diff --git a/justfile b/justfile
index 7cfa3aa..4ba4997 100644
--- a/justfile
+++ b/justfile
@@ -52,10 +52,9 @@ retrain: check-auth
 # SERVER is an ssh config alias; the agent does auth. No hostnames in this
 # file. Run `ship` only after eyeballing report.json — retrain never deploys.
 
-# Ship a freshly trained model.json to the server (compose mounts
-# ~/fluxrec-data at the container's state dir)
+# Ship a freshly trained model.json to the server state dir (bind-mounted
+# at /data; created once on the VPS: see README "Deployment")
 deploy SERVER:
-    ssh {{SERVER}} 'mkdir -p ~/fluxrec-data'
     scp model.json {{SERVER}}:~/fluxrec-data/
 
 # Bounce the serve container so it picks up the new model (fluxrec runs