# Lab method: E1/E2/E3, M-V/M-P, PVE templates and snapshots

> How Windows.Homes produces evidence: official sources, PVE templates, GPU passthrough measurements, raw data and the limits of each level.

**Tested on**

- Build: 26300.9711 / 26100.33650
- Edition: Lab
- Architecture: x64
- Evidence: E2, M-V
- Verified: 2026-10-27

## Evidence levels

| Level | Meaning | What it can support |
|---|---|---|
| E1 | Official documentation or specification, not executed here | “Microsoft documents this behavior” |
| E2 | Reproduced in a fixed PVE template | “In our 26H2 VM, this happened” |
| E3 | Reproduced on a physical PC or a second host | “On this hardware, this happened” |
| M-V | Measured in a GPU-passthrough VM, raw data kept | Same-environment deltas, not absolute FPS |
| M-P | Measured on bare metal, raw data kept | Performance claims with hardware context |

> **Caution: The VM is not the physical PC**
>
> A passthrough VM is excellent for reproducing UI, policy, driver install and service behavior. It is not a substitute for a bare-metal frame-time result. The site labels every measured row with its mode.

## Template baseline

- PVE host with an isolated bridge; Windows lab traffic never bridges to the production LAN.
- Windows 11 26H2: q35, OVMF, vTPM 2.0, VirtIO SCSI single, VirtIO NIC, QEMU Guest Agent.
- VirtIO drivers pinned to **0.1.271**; the template records the ISO and driver version.
- Server 2025: Desktop and Core templates, Evaluation edition, `expiresAt` recorded.
- Every experiment starts from a snapshot; the snapshot name goes into the test record.

## Procedure

- [ ] **Clone the template** (PVE host)

  Do not mutate the template. Clone, rename, and record the template version and clone id.

  Expected: A linked clone boots with the expected build and no pending driver warnings.

- [ ] **Snapshot before the change** (Virtual machine)

  Snapshot name: `<page>-<change>-before`. If the change touches storage or the boot path, take a full clone instead.

  Expected: A named snapshot exists and can be reverted.

- [ ] **Run the measurement** (Virtual machine)

  Discard the first cold run. Keep the median and 1% low; note the room temperature and power plan.

  Expected: Five passes of the same scene, raw CSV saved, build and driver recorded.

- [ ] **Archive the evidence** (Terminal (admin))

  The page can only claim `E2`/`M-V` if the archive exists. A screenshot without a recipe is not evidence.

  Expected: Recipe, VM config, raw data and screenshot are stored under the page id.

## Anti-cheat and passthrough

Some games refuse to run in a VM. This is a per-title policy, not a bug in the lab. The test record marks those titles “not testable in VM” and, if a physical machine is available, repeats them as `M-P`.

## What the lab will not do

- Bypass TPM/Secure Boot to install on unsupported hardware and call it supported.
- Use “rearm” scripts to extend an evaluation.
- Publish a measurement without the raw data and the hardware list.
- Hide a failed run.

## Sources

- [Proxmox Wiki — Windows 11 guest best practices](https://pve.proxmox.com/wiki/Windows_11_guest_best_practices)
- [Proxmox Wiki — Windows VirtIO drivers](https://pve.proxmox.com/wiki/Windows_VirtIO_Drivers)

---

Source: https://windows.homes/en/lab/method/