Tie background results to their rendered source, borrow cached mattes for the approximation notice, and cover provenance and matte access with regressions. Co-authored-by: Codex <noreply@openai.com>