diff --git a/docs/impl/plans/CAN-BITRATE.md b/docs/impl/plans/CAN-BITRATE.md
index 9c35a17..3df9b32 100644
--- a/docs/impl/plans/CAN-BITRATE.md
+++ b/docs/impl/plans/CAN-BITRATE.md
@@ -16,10 +16,11 @@ The OBD port is **not** gatewayed: broadcast traffic is visible passively.
 
 Persisted (boot and hot-plug):
 
-- `/etc/udev/rules.d/71-can-up.rules`: on add of the PEAK (USB 0c72:000c, matched like `70-can-names.rules`), wants
-  `can-up@can_pcan.service`. It first matched `NAME=="can_pcan"`, which never matches on udev 249, so hot-plug left the
-  interface down and the 2026-10-13 run recorded nothing. Fixed that day; check `can_pcan` is UP after the next replug.
-- `/etc/systemd/system/can-up@.service`: oneshot; reads `/etc/default/can-<iface>`, sets the bitrate and options, brings the interface up, and stops with the device
+- `can-up@can_pcan` is **enabled with `WantedBy=sys-subsystem-net-devices-%i.device`**: systemd starts it whenever
+  `can_pcan` appears, at boot or on hot-plug. It replaced `/etc/udev/rules.d/71-can-up.rules` (2026-10-13), whose
+  `SYSTEMD_WANTS` never reached systemd, by `NAME==` or by USB id: the interface stayed down on hot-plug and at boot,
+  so the recorder never opened a session and refused marks ("no session").
+- `/etc/systemd/system/can-up@.service`: oneshot, with that `[Install]` section; reads `/etc/default/can-<iface>`, sets the bitrate and options, brings the interface up, and stops with the device
 - `/etc/default/can-can_pcan`: `BITRATE=500000`, `CAN_OPTS="restart-ms 100"`
 
 **Listen-only was on during the sniffing phase**, so the adapter never ACKed on the car's bus. Dropped 2026-10-12 for `obd-poller`