Real server coverage parity probe
Run with PowerShell7 on Windows against a built/staged server. The script launches
its own hidden server on a free loopback port, creates synthetic data under a fresh
timestamped .local/rest-coverage-* directory, captures25 requests/responses and
asserts diagnostic parity, invalid-write rejection, source preservation and restart.
It stops only its own process, including on failure. It never attaches to an
existing server and does not use user credentials. The token it creates is fixed
synthetic probe data, not suitable for a deployed server.
From the repository root, replace the executable path with the artifact under test:
$env:PATH = "$env:SystemRoot\System32;$env:SystemRoot;$env:SystemRoot\System32\Wbem"
& harness/fixtures/rest-coverage/run-coverage-probe.ps1 -ServerExe 'C:\path\to\staged\arcadia-server.exe'
For gnullvm builds, stage the required runtime DLL beside the executable. Clean PATH is deliberate: success should not depend on a developer compiler directory. The first launch and restart use the same isolated store; restart is a forced process stop after an acknowledged saved revision. This tests local persistence, not graceful-drain handling. Output directories and evidence are retained.
coverage-probe.json has incomplete, repaired and invalid source bundles with
explicit synthetic UUIDs, plus five exact expected warning tuples. The repair
deliberately leaves an independent sequence_gap warning. The final reintroduction
of the incomplete bundle is an explicit replacement, not service Undo. No new API
or production code is required by the probe.
Observed Windows execution and limitations are recorded in
harness/release-rest-coverage-verification-20260906.md. This launch wrapper is
Windows-specific; the fixture and HTTP oracles can be reused on other platforms.