Things learned the hard way while building and racing with RaceVisor. None of this is in a manual. Each note says what went wrong, what it looked like, what fixed it, and what the rig was, so you can tell whether it applies to you. Newest first.
The fixes here were worked out live on the author's rig with the logs open, by an AI assistant working alongside the author. Several of them are the same things a computer technician would try first; they are written up because nobody thinks of them until they have lost a few sessions to the problem.
Check your own rig. Rig Doctor, a free one-file program, does all of this and more: it reads your machine and your last VR session and names the bottleneck with its fix, and it watches live while you drive. The lighter Check_Rig.ps1 script reads your machine and says which of these notes apply: where the headset sits on USB, who shares its hub, whether Windows may power that port down, which OpenXR runtime is active, whether the wheel software is up. It reads only, changes nothing, and nothing leaves your PC. It is printed in full on the launchers page so you can read it before you run it. Your hardware stays yours; share the result if you want help, or do not.
Link dropped every few minutes, and the face sensor watcher was doing it#
Rig: Meta Quest 3 over Link cable, Windows 11, the Quest Face Sensor Off watcher from the launcher scripts installed.
What it looked like. "PC disconnected", the headset froze, Link came back a few seconds later. Over and over, on track and in the menus, with or without a charger on the cable. The Meta service log said WinUsb_WritePipe ... failed: (31) A device attached to the system is not functioning, then device error, with the cable still attached.
The cause. Versions of Quest_FaceSensorWatch.ps1 downloaded before 24 October 2026 re-sent the sensor command to the headset every 5 minutes. That command goes over adb on the same USB cable that Link streams on. In one 90-minute session, 10 of 11 Link drops landed in the same second as one of those re-sends: the watcher's log said error: closed and Link ended at that moment. Re-sends that did not collide went through with no drop. Starting the old watcher while Link was up did it too.
The fix.
- Download the launcher scripts again (the watcher in them is dated 24 October 2026 or later).
- Replace your
Quest_FaceSensorWatch.ps1with the new one. - Run
Quest_FaceSensorWatch_Install.batagain.
The new watcher turns the sensor off once each time the headset appears (boot, wake, re-plug) and then leaves the headset alone. It recognises the headset from the device list, without opening a command session on it.
The rule that came out of it. Do not send adb commands to the headset (adb shell, logcat, dumpsys, battery readers) while Link is running. Do them with Link stopped.
What it was not. Not the charger and not the cable. Those were tested first, with the watcher switched off for the tests, which is why the tests looked clean and pointed the wrong way.
The face sensor came back on by itself, and the drops came back with it#
Rig: Meta Quest 3 over Link, Windows 11, RaceVisor launcher scripts.
What it looked like. The sensor had been turned off for the session, the drops had stopped, and then days later they were back: 1:43, 1:33, 1:55 into three sessions. The launcher had run, the sensor command was in it, nothing had changed on the PC.
The cause. The prox_close broadcast lasts until the headset restarts. Every headset reboot, including the one you do to get Link back after a drop, turns the sensor on again, and a one-second lift of the headset then ends the Link session. The launcher only sends the command once, at start; a reboot in the middle of the day undoes it silently.
The fix: a watcher, not a one-off. The Quest_FaceSensorWatch_Install.bat in the launcher scripts installs a logon task, Quest Face Sensor Off, that watches for the headset every 10 seconds and turns the sensor off each time it appears, with no window. (Before 24 October 2026 it also re-sent the command every 5 minutes, and that caused Link drops of its own - see the note above.) The recovery scripts send the command too, so a headset reboot from the recovery path never leaves the sensor on. If you want the sensor back, remove the task (-Remove) or run the launcher with -FaceSensorOn.
What it was not. Not the cable, not the hub (that had already been fixed), not a PC crash.
Also seen the same day. WinUsb_Initialize error 8 and write error 31 in the Meta service log on a drop: the PC lost the USB device. That one is the hub power note below, not the sensor.
The dash sat way above my head at every start#
Rig: Meta Quest 3 over Link, RaceVisor VR groups, OpenKneeboard.
What it looked like. The widgets were centred left-right and front-back, but a foot or two above eye level, every time RaceVisor came up, and a Recenter fixed them every time. Saving a position did not help, because there is none: the widgets sit relative to where the head was when the origin was taken.
The cause. The origin was taken at the layer's very first frame, which is wherever the headset was at that instant: on the desk, on a hook, lifted onto the forehead. Centred but high is exactly a headset standing on a desk in front of the seat.
The fix. The layer now waits until the headset is worn and the head has been still for a second and a half before it takes the origin, and draws nothing before that. On top of it, RaceVisor recentres the kneeboard and the widgets itself the first time the car is on track after a session load (Settings → Recenter the kneeboard and widgets when the car gets on track). If anything is ever off, the Kneeboard widget's Recenter, or its hotkey, moves everything to your head.
After a Link drop the headset comes back black: get back in the race fast#
Rig: Meta Quest 3 over Link, RTX 4090, SteamVR, iRacing.
What it looked like. Mid-session the headset dropped and the car went straight on. Link reconnected and showed the SteamVR lobby, recentering worked, but the scene stayed black. Restarting the headset did not bring it back. The session was lost.
What the logs said. The video stream was perfect to the last moment: no glitches, normal network times, no graphics driver error, nothing from Windows. Then the headset's tracking data stopped for a quarter of a second ("Prediction interval too high: 0.26 s" in Meta's Client log) and the PC side ended the session. After that every reconnect failed with OVR error -3000 (ovrError_Initialize) in SteamVR's vrserver log: the Meta service on the PC was stuck, so SteamVR only ever got a black screen.
The cause. The drop started in the headset. The black screen afterwards is the PC: the Meta service does not recover by itself, and restarting the headset does not touch it.
The fix: back in within a minute, no PC restart.
- Close SteamVR. Close the sim if it is frozen; your race session keeps running on the server.
- Restart the Meta service: right-click Start, choose Terminal (Admin), and type
Restart-Service OVRService -Force
- Start Link in the headset, then SteamVR.
- Start the sim and rejoin the same session from the race page while it is still running. You go back out from the pits and keep the laps already done, so you can still salvage points.
- Before the next session, turn the face sensor off (see the note below). A one-second lift of the headset is enough to start this whole chain.
What it was not. Not the PC (no crash, no driver reset), not the network, not the video bitrate.
Quest Link drops mid-race: "PC disconnected", screen freezes, back to the Meta lobby#
Rig: Meta Quest 3 over Link cable, ASUS ROG Strix X670E-E, Ryzen 9 7900X, RTX 4090, Logitech G29, Logitech Lightspeed keyboard and receivers. Custom build.
What it looked like. Four time trials in a row, on a straight each time with no head movement, the headset dropped to the Meta lobby and the sim froze. One monitor flashed and locked. Cables were tight; shaking the cable while not driving never dropped it. Power cycling everything did not help.
What the logs said. Windows' Kernel-PnP device log recorded the headset as surprise removed, reported as missing on the bus a minute before the Oculus service logged HMD not attached and gave up. The headset vanished from USB; it did not go to sleep.
The cause. The rear USB port the Quest was in belongs to the motherboard's chipset USB controller, which sits behind two PCIe switches, and on that board those ports share one hub chip with the wheel, the keyboard and six wireless receivers. Windows was also allowed to power that hub and controller down when it judged them idle. Under racing load, with the wheel's force feedback and the headset's video on the same hub, the hub reset and the heaviest device on it, the headset, fell off.
The fix.
- Move the Quest cable to a rear port wired straight to the CPU. On this board those are the ports the keyboard and a TV tuner stick happened to be in. The Quest took the tuner's port and the tuner took the Quest's. Nothing else shares the CPU ports.
- Tell Windows it may not power down the chipset USB controller or the hub under it (Device Manager, the controller's Power Management tab, untick "Allow the computer to turn off this device"). Also check USB selective suspend is off in the power plan.
How to tell which ports are which on your board. In PowerShell, list the headset's parents: each hop up ends at a controller. "USB 3.10 ... 1.20" on an AMD board is the CPU's own controller; "USB 3.20 ... 1.10" behind "PCI Express Downstream Switch Port" is the chipset. Any device on a CPU-direct port tells you where those ports are on the back panel.
Also, separately: the headset's face sensor. Lift the headset for one second and the Quest sleeps and ends the Link session, which on the PC looks exactly like a disconnect. Turn the sensor off for the session (adb shell am broadcast -a com.oculus.vrpowermanager.prox_close; a headset restart turns it back on). The launcher scripts do this by default; add -FaceSensorOn to the launcher line to keep your sensor working. If your phone or another Android device is also connected to adb, a plain adb shell refuses ("more than one device") and the sensor stays on: aim at the headset with adb -s <serial>. The launchers now find the Quest by name.
What it was not. Not the cable (shake-tested), not a PC crash (every reboot in the log was a clean user restart), not sleep (the sensor was already off), and not, as far as the logs can say, the graphics driver, although a driver revert lined up with the first drops in time.
Does a disconnect in an iRacing time trial count against you? No. Time trials have no incident points and no safety-rating or iRating effect. A disconnect ends that attempt; you keep any clean lap already completed.
Turned the Quest resolution up and it started to stutter#
Rig: Quest 3 over Link, RTX 4090.
What it looked like. Render resolution to 1.3× and bitrate to 500 Mbps with H.265 for a sharper image. The GPU sat at 75 percent. It stuttered everywhere.
What the logs said. Meta's service log records, once a minute, how long the headset takes to decode each frame. It was 22.5 ms on every percentile, against an 11.1 ms budget at 90 Hz, and the PC encoder had backed up thousands of frames waiting for it.
The cause. The Quest's own video decoder, not the PC. It cannot decode H.265 at 500 Mbps in time. No graphics setting on the PC changes that.
The fix.
- Open the Oculus Debug Tool (Meta Horizon folder, Support, oculus-diagnostics).
- Set Encode Bitrate (Mbps) to 250 with H.265. Use 200 for a long race.
- Close the tool, end the Link session in the headset, start it again.
- Drive a lap and check: decode should sit under the frame budget. Here it dropped from 22 ms to 9 ms, reprojection to zero, glitches to none, and the 1.3× resolution stayed.
Automobilista 2 minimised and would not come back#
What it looked like. Starting RaceVisor after the sim was already full screen ejected AMS2 from its display mode; it sat minimised on the taskbar and would not restore.
The fix.
- Set the sim to windowed or borderless, never exclusive fullscreen.
- Start RaceVisor before the sim, so the overlay is already there when the sim goes full screen.
- Or use the launcher scripts, which start everything in that order.
The wheel used default rotation and force feedback#
What it looked like. G29 felt wrong: default rotation, weak force feedback, no button profile. Settings snapped back the moment G HUB was started.
The cause. The wheel's profile is applied by the G HUB app, not the driver. If G HUB is not running when the sim binds the wheel, the sim gets the defaults.
The fix.
- Start G HUB first, before the sim and before RaceVisor.
- Wait a few seconds for it to apply the profile.
- Then start everything else. The launcher scripts do exactly this.
OBS captured a blurry, upscaled postage stamp#
What it looked like. The OpenXR Mirror source in OBS was a soft, 8x-upscaled mess.
The cause. The mirror source latches the eye-texture size when it initialises. Start OBS while nothing is rendering in VR and it grabs a placeholder a few hundred pixels across.
The fix.
- Start the sim and get in the car, so it is rendering in the headset.
- Only then start OBS. If OBS is already open, remove and re-add the mirror source.
- Or use the launcher scripts, which wait for the sim's driving process before starting OBS.
The kneeboard was blank in the headset#
What it looked like. OpenKneeboard showed nothing, while the overlay was clearly running on the monitor.
The cause. OpenKneeboard's window-capture tab is pinned to the exact path of the program it captures. Launching a different copy of the overlay, even an identical one from another folder, runs fine and captures nothing.
The fix.
- In OpenKneeboard, open the window-capture tab's settings and note the executable path it points at.
- Launch that exact executable, every time. A copy in another folder captures nothing.
- The launcher scripts find it by a fixed search order for this reason.
The launcher window flashed red and closed instantly#
What it looked like. Double-clicking a .bat gave a window full of red parse errors for a fraction of a second, then nothing launched.
The cause. A PowerShell script saved as UTF-8 without a byte-order mark, containing one curly quote or em dash. Windows PowerShell 5.1 reads such a file as ANSI, the mangled bytes contain a quote character, and every string after it breaks.
The fix.
- Open the script in Notepad and replace any curly quotes or long dashes with plain ones.
- Save it as UTF-8 with BOM, or keep it ASCII only. Every script on this site is plain ASCII.
- Run the .bat again.
Nothing launched: "powershell is not recognized"#
What it looked like. The same instant-close, with "'powershell' is not recognized".
The cause. PATH is not reliable from a double-click or from some shells.
The fix.
- Download the .bat files from this site again: they call PowerShell by its full path, PowerShell 7 first, then Windows PowerShell.
- If you wrote your own, replace
powershellwithC:\Windows\System32\WindowsPowerShell\v1.0\powershell.exe.