docs(wiki): trainer as a compose profile — one running service is expected, first-run pull/login, why the collector TAG bump mattered, park-2 needs no bump until a pin

Claude-Session: https://claude.ai/code/session_01FWncR69HgGPuei1dLrW3cU

wiki/log.md

@@ -3144,6 +3144,13 @@ run; the Quadro FX 3800 is unusable (cc 1.3), the HD P530 irrelevant, the Xeon E
 compose seam drops the GPU reservation; cloud GPU rejected (crops stay on premises). Linked from
 [[opencv-anpr-service]], [[vision-review-outbox]], index. User: "No build just yet."
 
+## [2026-09-07] ingest | Trainer deployed as a profile; operating notes
+User pushed `stage-f7a262a`, bumped the `wash-collector` TAG, redeployed, and asked why only one
+service runs on art-docker-station. Expected: the trainer is a compose profile, never started or
+pulled by a deploy; run by hand, exits. Recorded on [[bodytype-classifier-training]] §Operating
+notes (why the TAG bump still mattered, park-2 needs no bump until a pin, the registry login for
+the first pull, the order of commands) and [[fleet-deployment-komodo]].
+
 ## [2026-09-07] build | Phase B trainer + the classifier stage on the booth
 User: "Shall we go and build the trainer for the Xeon?" Built `apps/trainer` (`parking-trainer`:
 `inspect` / `train` / `evaluate` / `publish`; reads the collector volume read-only, time split,