Sure! My AR app's 3D interaction logic is regression-tested today only by device-tier XCUITests against the live camera (i.e. me holding the device while it adds a model, checks it can select it with a tap, and transform it as expected). In practice that caps how much AR coverage I can maintain, since every regression pass needs a human and a device. I keep a library of recorded session .movs that reproduce specific scenarios; Xcode replays them fine under the Run action, but the scheme's replay setting (DeviceSensorReplayFileReference) doesn't apply to the Test action (and isn't valid inside one) so XCUITest always gets the live camera. Replay in tests would also unlock checks I can't do at all today: deterministically verifying that a saved AR scene relocalizes correctly on reload, which right now only works if I'm physically standing in the room where the scene was saved. What I'd want, in order of preference: a supported way to attach a replay file per test plan or per test, so AR XCUITests can run unattended (locally or in CI); failing that, ARSession replay in the simulator. I'll file Feedback with this use case, and I'll try the ARFrame-serialization harness you suggested for the unit tier in the meantime. That seems to cover the math/logic layer nicely, though not the render/hit-test path the XCUITests exercise, so the replay ask above still stands. Thanks!

Thanks! Since the Ultra Wide is already streaming during world tracking: Is there any supported way today to get Ultra Wide pixels while an ARSession runs? (My understanding is a concurrent AVCaptureSession gets interrupted once ARKit owns the cameras.) If not, would read-only access to those frames — e.g. a secondary pixel buffer on ARFrame with its own intrinsics/extrinsics, no configurability, in the same spirit as scene depth — be a reasonable Feedback request? Or is there a deeper reason (calibration, exposure control, bandwidth) that feed can't be surfaced for rendering?

I'm also having this issue