diff --git a/docs/BOM.md b/docs/BOM.md
index bebc8a6..c2202f9 100644
--- a/docs/BOM.md
+++ b/docs/BOM.md
@@ -164,8 +164,10 @@ Header facts (carrier board spec SP-11324-001 v1.3, Table 3-3 and 5-3; raw copy
   plus a realign event, so a bad alignment can be corrected afterwards. Align on level ground: a slope becomes part of the mount.
 - I²C mode (`FF AA 62`) exposes the raw MPU6050 only (no angles) and needs 4.7 kΩ pull-ups. Not used; v24-0429 lists SDA/SCL as not connected.
 - **Mounting matters more than the sensor.** An IMU needs a rigid path to the body; a box on carpet, Velcro or isolators smooths the motion
-  and drifts. Under consideration: the IMU on **USB** (WT61 behind a 3.3 V USB-UART adapter, or a GNSS+IMU module such as the NEO-M8U over
-  USB-C), mounted apart from the Orin's box on the **roll cage**, clamped close to a foot or a joint rather than mid-span.
+  and drifts. **Decided (2026-10-12): the IMU lives outside the enclosure on one USB-A port**
+  (WT61 behind a 3.3 V USB-UART adapter now; a NEO-M8U over USB-C as the GNSS+IMU upgrade), mounted on the **roll cage**, clamped close
+  to a foot or a joint rather than mid-span. The header wiring above stays valid but is no longer the plan. A fault in the sensor now
+  lands on a USB port (0.5 A limit) rather than the header's 3.3 V rail.
 - **No reverse-polarity protection is documented.** A 4-way Dupont housing put on the wrong way round swaps VCC and GND. Check VCC and GND
   with a meter before first power-up.