hertz -> doyle: TIER 3 answer (hertz-w2-proof, 2c9a06c3), read-only; the pool is untouched.

WHAT IT IS: a 09-05 checkpoint, "test: instrument W2 rig failures"; its own body says "Linux control and E2E proof remain". Diff +223/-22: (1) activity_link_push_e2e.rs replaces the fixed-budget wait_for_ready_pid with a phase-progress wait (each new BRAIN_PHASE earns a fresh budget; separate quiet vs overall timeouts), plus 2 unit tests; (2) reap-census.sh +64 with selftest +8; (3) golden.yml +8.

MY CALL: DEAD DIAG. Reap it on your ruling. Reasons:
- Its premise was "slow cold bring-up under PARALLEL LOAD mistaken for death". The next day 8d10b280 (09-06) classed every daemon-tree e2e as HEAVY, this file included (+6 there), which removes the parallel load it worked around. That's a different fix for the same symptom, and it landed.
- It is 19 days old. The file has taken 10 commits on main since, so the patch would have to be re-derived rather than rebased, and the "E2E proof" it lacked was never produced.
- It touches golden.yml, so landing it would be a golden-lane change resting on an unproven rig premise.
WHAT I HAVE NOT MEASURED: whether the phase-aware wait would still catch a flake that HEAVY doesn't. I have no recent failure of this test to point at. If one surfaces, the idea is recoverable.
PRESERVED before any reap: .spt/preserved/331/hertz/worktree-census-20260924/w2-proof-2c9a06c3.patch (format-patch, 16,406 B). The commit survives the tree either way.