## What's Missing

Tests run against stubbed hardware (`OMARCHY_DMI_ID_PATH`, `OMARCHY_PCI_DEVICES_PATH`). This works for CI but means no test ever runs against actual GPU, Wi-Fi, NVMe, or EC hardware. Contributors are told to "test on real hardware before PR," but there's no harness to do that.

## The Gap

- `test/shell.d/hw-msi-test.sh` passes with stubs. It does not verify that `coolercontrold` is actually running, that `r8125` is actually loaded, or that `nvidia-powerd` is actually active.
- A contributor with an MSI Titan can run the stubbed tests, but there's no automated way to say "on this real hardware, these 5 services are running, these 3 modules are loaded, these 2 configs are correct."
- CI can't catch what stubs don't model. Example: PR #11549 enables `coolercontrold`, but only real hardware can confirm the daemon actually starts and fans respond.

## Proposal

Add a `./test/hardware` runner that:

1. Detects real hardware (no stubs)
2. Runs the same test suite against actual devices
3. Reports pass/fail for: loaded modules, active services, config files present, devices visible
4. Requires no special setup — runs on the contributor's machine

CI continues to run stubbed tests. Real hardware tests are opt-in: contributor runs `./test/hardware` before submitting.

## Effect

PR that enables `coolercontrold` on MSI → contributor runs `./test/hardware` on their Titan → test confirms `coolercontrold` is running and responding → CI runs stubbed tests → merge with confidence.

Bugs like "service enabled but daemon fails to start on real hardware" get caught by the contributor before PR, not by users after release.