[woksin]

Changed

• The native runner documentation now describes what happens when a Microsoft.Testing.Platform session is stopped. Running specification classes finish their cleanup once, queued classes do not start, and the platform does not deliver results after a hard cancellation. (part of Graduate the native runner from experimental #77)

Fixed

• The experimental native runner now reports a cancelled Establish, Because or Destroy the same way whether the specification uses generated or reflection-based lifecycle calls. Both raise SpecificationCancelledException with the phase and the original exception. (part of Graduate the native runner from experimental #77)

[woksin]

Reviewer notes

• Closes the evidence gaps an independent review found on Graduate the native runner from experimental #77: a lifecycle cancellation mismatch between generated and reflection code (reproduced sync and async before the fix), no shared lifecycle matrix across adapters, and no real platform-level Stop coverage.
• CI gains two steps in build.yml. Twelve shared setup, act, teardown and cancellation cases run through real xUnit v2/v3, NUnit 4/5, TUnit and runner consumers; verify.mjs asserts 148 terminal results, including the documented framework teardown-reporting differences. stop.mjs drives a real MTP server-session $/cancelRequest with two concurrent classes. The build workflow's path filter now includes Integration/Lifecycle/**.
• The runner stays experimental. Graduate the native runner from experimental #77 stays open for graduation after Fundamentals NativeAOT (AddTypeDiscovery throws PlatformNotSupportedException under NativeAOT Fundamentals#1217).
• Local: Release build 0 warnings/0 errors; nine non-integration suites, 1,009 specs passed; actionlint ok. The full trimmed/AOT CI matrix was not rerun locally.
• Review: full round 1 passed (Opus and Sol), no blocking findings. I applied the path-filter advisory; the other advisory (verify.mjs counts only some phases for teardown-once) is left as is.