[IanFrelinger]

The post-merge project-inventory guard for #901 found the new release-evidence evaluator missing from docs/ProjectTiers.md. The protocol gate also exposed deployment-profile tests changing a process-wide environment variable while MCP bridge tests ran in parallel. An earlier macOS readiness failure exposed an independent race while Docker command tests read console buffers that asynchronous logging was still writing.

Register the evaluator and run the existing inventory guard on PRs that change projects, the inventory, or the guard itself. Serialize the deployment-profile test collection, restore the prior profile, and remove its now-stale convention exception. Take console snapshots under the same locks as console writers and reuse the snapshot. Production validation and every existing test assertion are unchanged.

Local validation: the entire MCP-server assembly passed 78/78 with zero skips in three isolated .NET 8 runs; all three focused CLI regressions passed with zero skips on .NET 10. These runs used isolated patches over 330f993 whose changed source files match producer 98b1ff33c653a368d520ae62740cedebe6e1f256. The final CLI build/test ran offline after restoring the existing PublicApiGenerator dependency, without version changes. All 48 repository gates passed on 98b1ff33; project inventory covers 108/108 tracked projects and all 62 workflow triggers parse and match the updated inventory.

Two post-commit mutation controls on 98b1ff33 each failed the intended convention assertion (1 failed / 5 executed), restored exact source with empty porcelain, and passed 5/5 on the restored tree, with zero skips. They remove the nonparallel flag and collection membership respectively. The stock mutation driver used real offline Docker runs, without a runner override.

Hosted CI on 98b1ff33: all five required checks and all applicable ancillary checks passed on the first attempt, including certification (3,989/3,989, zero skips), full MCP/A2A protocols, the newly PR-triggered inventory guard, UAT, coverage, and Linux/macOS/Windows readiness. Platform run. The summary records partial lanes_ran=3 lanes_skipped=1: native, container and CLI-image groups passed; production-image builds are intentionally skipped on PRs. This is merge evidence, not a release or compliance sign-off.

Original failures: project inventory, protocol gate, and macOS capture. The network-disconnected mechanism-only capture probe reproduced four matching buffer exceptions in 700 attempted raw reads, with zero exceptions in 1,100 synchronized reads; it did not execute the original macOS/base suite.

This follow-up does not change package versions, licensing boundaries, deployment configuration, or pilot/marketing/autonomy holds. Use a merge commit to preserve producer 98b1ff33c653a368d520ae62740cedebe6e1f256 for the retained local negative controls; required and applicable checks must pass before merging.