[JasonStys]

Approved C# nonzero returns were reported as partial coverage because Coverlet 8.0.1 collapses the target exit into its own failure code. Version 0.4.3 runs a fixed isolated Python bridge that waits for the real instrumented .NET child and exclusively records its exact return, source, case and stdin binding. The worker independently requires collector success, a valid fresh observation and real coverage XML; unexpected exits and genuine collector errors/timeouts remain failures.

The current compiler lane now requires measured C# cases and separate target/collector evidence. It exercises exits 0/1/2/7/255, malformed input, an incorrect expected return, an exception and real collector argument rejection, post-target report-write failure and timeout. The immutable 0.4.2 baseline and published package test preserve the historical limitation through an explicitly source-bound policy.

Focused tests cover actual inert child-process behavior and strict request/receipt/measurement failures; only hosted Actions executes .NET/Coverlet. Native package tests require the exact bundled bridge bytes. Source/application/worker versions are coherently 0.4.3 with unchanged dependencies. The owner's installed app, data and Docker are not modified, and existing 0.4.2 release assets are not overwritten.

Validation: both hosted source lanes passed 70 realistic scenarios; current C# exits 0/1/2/7/255/2 were measured, and all three genuine collector faults were blocked. Full hosted suite: 410 tests and 28 subtests passed, with 85.01% controller/training coverage against the unchanged 80% floor. Browser, isolated worker/fuzz, private-history replay, native packages on three operating systems and CodeQL passed. Focused local suite: 238 passed and two explicit Windows platform skips. No new release or local upgrade is claimed.

Detailed design and dated run/source evidence: docs/csharp-coverage.md and docs/reports/csharp-coverage-2026-11-10.md.