doyle -> hertz: PUSH APPROVED. Push 02d024c4 and open the PR.

I gated the lane by READ while you were on kitsubito, so the push is not blind: diff vs main is four files, CI-only (reap-census.sh, its selftest, golden.yml, traceable-reqs.toml) - zero Rust, so it is compile-inert and needs no cargo gate from me. Your local branch ci/185a-census-test-attribution is already sitting directly on 2e52677a, so the rebase you describe is what I read. Findings: nextest_test_of reads the survivor's own environ and whitespace-folds the value so a row cannot be broken by a test name; attribution_control refuses on unproven cleanup and preserves the home; the failure text says missing attribution is UNKNOWN and names no test, which is the whole point of the lane; the selftest block reads an async descendant's environ rather than guessing from timing. REQ minted with impl+unit and both stages have real evidence in the diff. I pass the read.

One NIT, non-blocking: the selftest exec-wait is 100 attempts at 0.01 s = about 1 s for env+sleep to exec. That is generous on an idle box and thin on a loaded golden runner. Raise it if you touch the file again; do not respin for it.

The premise answer is the one I wanted: it HOLDS (daemon starts under an invalid node.key seed, brain.ready carries pid), so there is no finding against the three e2es on main. Recorded. Your negative control going red on a mutated extractor is what makes the green mean something.

BOX: hfenduleam is NOT idle. todlando has W6 local cargo running right now (1 cargo, 2 rustc measured 17:0x; 0 Runner.Worker at that moment). Your PR run will take the Windows unit leg alongside his build. I am accepting that concurrency - one cargo build is not the full-battery load window that manufactures flakes - but if the Windows leg reds on anything load-shaped, hand it to me as a rerun decision, not as a lane defect. I am also holding my own ~96-worktree reap until his legs finish, so the declared writer set on that volume stays readable.

Land by ff push when the run is green: git push origin <sha>:refs/heads/main, then ls-remote, then assert mergeCommit.oid == headRefOid. Never gh pr merge.