Skip to content

Reproduce a campaign

Use Windows with Hyper-V administration, Visual Studio C++17 tools, Windows SDK, Git, CMake and Python. Building user-mode benchmarks does not require changing the drivers or installing the WDK in the guest.

python -m venv .venv
.venv\Scripts\python -m pip install -r requirements.txt
python scripts\prepare-sources.py
python scripts\prepare-vjoy-sdk.py
python scripts\fetch_artifacts.py
cmake -S . -B build -A x64
cmake --build build --config Release

Inspect reports/repository-audit.json and reports/artifacts.json: they pin source commits and official release downloads. The SDK and driver licenses remain separate from the MIT benchmark code. Create a new dedicated VM with automation/provision-vm.ps1, supplying an already stopped lab parent VM/disk and a local encrypted PowerShell credential file. Do not use a personal VM or install input filters on your workstation.

. .\automation\Lab.ps1
$lab = Get-BenchmarkLab
.\automation\initialize-guest.ps1
Checkpoint-VM -VM (Get-VM -Id $lab.vm_id) -SnapshotName Clean-Benchmark-Baseline
.\automation\install-titancsa.ps1
.\automation\run-virtual-device-tests.ps1 -Repetitions 5 -Samples 1000
.\automation\run-gamepad-tests.ps1 -Backend vdevice -Repetitions 3
.\automation\reset-vm.ps1
.\automation\install-competitors.ps1
.\automation\run-gamepad-tests.ps1 -Backend vigem -Repetitions 3
.venv\Scripts\python scripts\report.py
.\automation\analyze-results.ps1
.venv\Scripts\python -m mkdocs build --strict

Keep the one-hour demo allowance intact. Do not edit trial constants, system clocks, signatures or activation code. Campaigns must fit within the permitted evaluation session. Retain failures, crash records and raw data; start new run IDs rather than overwriting a series.

The first pilot lacked an orchestration completion record and was invalidated. A subsequent file-handoff race produced an empty JSON read; the handoff was changed to atomic rename. Those failed records remain in the raw dataset.

run-all-benchmarks.ps1 covers the initial vendor-HID/XUSB comparison, not the entire matrix. Other families have dedicated runners (run-input-tests.ps1, run-controller-filter-tests.ps1, run-joystick-tests.ps1, run-profile-validation.ps1, run-stability-tests.ps1 and trace-scenario.ps1). Keep controller teardown experiments isolated while WDF-001 is unresolved; no repeated-crash qualification or complete protocol suite is claimed.

The self-hosted workflow is prepared, not executed. A dedicated runner must have Hyper-V rights, MSVC/CMake and an external private lab configuration path in TITANCSA_BENCHMARK_LAB_CONFIG. Keep the configuration and encrypted guest credential outside the Actions checkout so a checkout cleanup cannot remove them. No runner was registered on the user's workstation.