Hyper-V lab isolation
TitanCSA-Benchmark-Lab uses a dedicated differencing VHDX, 2 vCPU and 3 GiB fixed RAM. Dynamic Memory is disabled. Secure Boot is enabled. A stopped existing lab image supplies the immutable parent; the parent's VM is not started or changed. The owned lab's identity is recorded under ignored .local/lab.json and checked before restore, copy or installation.
The initial inherited image contained historical Filter packages. Those were discovered through PnP inspection and removed only in the dedicated child VM. The original checkpoint is retained as Inherited-Image-Before-Input-Cleanup; it is not the corrected clean baseline. After removing the historical packages and uninstalling the measured drivers, Windows was restarted and Clean-Benchmark-Baseline was created. reports/clean-baseline-v2.txt records the checked driver inventory. Older pilot data retain their original provenance and are not claimed to be pristine-image measurements.
Driver-specific *-Cleaned-Ready checkpoints isolate later configurations. Restores apply only to the recorded benchmark VM. Export results before restoring: guest results intentionally disappear with the checkpoint. No checkpoint or clock manipulation is used to extend a demonstration allowance; each evaluation stays within the published session terms.
PowerShell Direct controls the guest without embedding a password in scripts or publications. Compilers run on the host, producing static-CRT executables copied into the guest. Drivers and workloads run only in the guest. WPR is already installed there. Installing a second compiler and WDK in the measured VM is unnecessary for these prebuilt signed-driver scenarios and would change background services and storage load.
Hyper-V does not supply universal native USB passthrough. No physical USB controller is assigned here; RDP redirection is not treated as direct USB hardware. Results are explicitly virtual-device software-path measurements.