{"body":"## Scope\n- Replace the fixed readiness deadline with a 30s quiet-phase budget reset by new BRAIN_PHASE stamps and a 300s aggregate cap; read the real persistent stderr sink.\n- Attribute Linux census survivors from NEXTEST_TEST_NAME; absent/unreadable identity stays explicitly unknown.\n- Run a real daemon-to-brain attribution positive control before the Linux battery, with bounded shutdown and retained diagnostics on unproven cleanup.\n- Await child exec before reading its attribution in the selftest.\n\n## Verification\n- Windows activity_link_push_e2e: 3/3 passed (11.94s).\n- Linux kitsubito activity_link_push_e2e: 3/3 passed (16.03s).\n- Linux census selftest and real broker/brain attribution control: passed; control cleanup confirmed both PIDs gone.\n- Focused rustfmt check and shell syntax passed.\n- traceable-reqs: 874 complete, 0 findings.\n\nLinux proof used isolated .worktrees/hertz-w2-proof at checkpoint 2c9a06c3 plus the exact three-file corrections committed in afa5af35; no shared checkout was patched. Full golden battery was not run. RCA-274 heavy reclassification is excluded. No public behavior/docs or changelog change: test-rig-only lane."}