[harder]

Adds a manually invoked acceptance kit for Windows, WSL and macOS, with no scheduled automation. The runner keeps results outside the checkout and reports failures, interrupted work and missing coverage explicitly.

The PR contains reusable scripts, agent prompts, fixture fixes, startup documentation and a sanitized aggregate acceptance summary. Personal reports, device evidence, machine dumps, logs and terminal captures have been removed from both the PR tree and its commit history. Future testing instructions explicitly keep those artifacts outside Git.

Fixes include CI-compatible sequential locked AOT publishes; rejection of result locations beneath ancestor .git markers; isolation of Git context in a bounded collision fixture; and Windows fixture-owner normalization that skips redundant writes, with file/directory/link regressions. Startup-input tests and documentation clarify the inventory-driven switch to Installed. No product source or dependency locks changed.

Validation of the retained code: WSL unit suite 2,698 passed, 30 explicitly skipped, zero failures; integration suite 11/11; Python runner tests 10/10; Release builds and scoped formatting passed. Focused native Windows fixture regressions passed 3/3 after failing with the original helper. Focused macOS collision/startup tests and settled-startup PTY checks passed. Windows-only bodies that return early on Linux do not establish Windows coverage. The Python smoke tests and diff checks are repeated after the evidence cleanup.

Remaining release follow-ups are explicit in the summary: positive Windows security-metadata-dependent proofs require an already-admitted SeSecurityPrivilege context; the native Windows terminal walkthrough is incomplete; and a full Windows suite pass after the fixture correction remains unverified. These are separate from merging the testing kit. Merge after required CI checks pass.