# #331 golden-head intake — candidate 5507c487 (doyle, 2026-09-25) — NOT HANDED OFF

Candidate: gate/331-v073 @5507c487 = main 456c8554 + "release: shape v0.73.0" (4 files: Cargo.toml, Cargo.lock 14 pairs, CHANGELOG.md, docs-site/src/changelog.md).
Shape leg: `git show 5507c487:Cargo.toml` -> version = "0.73.0"; CHANGELOG -> ## [0.73.0] - 2026-09-25. Carries the W9 restart-once line.
Compile gate (isolated .worktrees/golden-331, fresh pool): cargo build --workspace --all-targets exit 0, 7m09s. traceable-reqs check exit 0. xtask gen: no drift.
MISFIRE: I pushed golden/331-v073 to share the sha. That fired golden.yml run 36104664271 before intake. Cancelled ~2 min in; ref moved to gate/331-v073. Memory: gater-never-pushes-golden-ref.

## Seam-consumer rule (runbook 152-166): spt/tests + spt-daemon/tests full, both OSes
Lists: doyle/intake-5507c487/handoff-lists.md (subagent; 2 MUST-RUN-FULL crates, 4 golden-only legs n1-gate/twohost/notify/oneliner last run at 73f19d51).
Windows (hfenduleam, my leg, env CONTAMINATED: OWL_SESSION_ID/SPT_ENDPOINT_ID/SPT_HOST_PID... inherited from the agent shell):
- Phase A (ci-windows profile) exit 0, 206/206 (2 leaky), 06:49:42-06:59:13Z
- Phase B exit 100, 238/242, 4 FAIL, 06:59-07:22:54Z
- Scrubbed rerun of the 4 (env -u all OWL_/SPT_): attach_wedge PASS, job_escape PASS, refresh_freeze PASS, webserve_attachment FAIL.
  - attach_wedge: env mechanism PROVEN (DAEMON_STOP_REFUSED: $OWL_SESSION_ID is set).
  - job_escape (reachable=false) and refresh_freeze (sustained ack false): each green ONCE when scrubbed; NO mechanism read. Treat as unexplained, not cleared. Golden re-executes them in a clean runner env.
Linux (kitsubito, hertz, 0 OWL_/SPT_ vars measured): Phase A 199/199; Phase B 231/234. Record: hertz/gate331-linux/REPORT.md
- webserve_attachment_e2e: W9 PRODUCT race (autostart readiness = seed ping; port exists only after the brain announces). Both OSes. -> todlando fix/305-docs-port-pending, RULED 07:27Z (pending flag + {gen,0} announce + CLI-only 15 s bounded retry -> DOCS_LISTENER_PENDING; amend REQ-WEB-BOUND-PORT-GENERATION-SCOPED).
- source_ladder_e2e: RIG (path_without_gh drops /usr/bin, and tar goes with it) -> hertz.
- auto_apply_e2e TIMEOUT 240 s: RIG budget vs nextest terminate-after -> hertz.
First-ever Linux cells: see REPORT; 7 of 9 green.
Subagent finding: the W10 gate record claims PR run closed busy_window_axis_e2e's Linux gap; that run was units-only. It is now closed for real by hertz's leg (PASS 0.69 s).

## Next
Land todlando's fix + hertz's rig PR (ff) -> RE-SHAPE v0.73.0 fresh on the new main tip (not-an-ancestor of 5507c487) -> compile gate -> re-run the changed binaries both OSes (webserve_attachment, serving_in_brain, source_ladder, auto_apply) -> reap golden-331 pool (73.6 GB allocated, 0 escaped) -> floor (128GB; C: 55.9 GB at 06:48Z pre-reap; runner _work target 70.0 GB) -> hand to deployah.

## Progress 08:20Z
- #282 (hertz rig fixes) gated on Windows at 0c467bfc: source_ladder + auto_apply PASS 104.8 s, no leak row, 0 temp spt.exe survivors, xtask check OK. PR CI 36110117837: Windows `Unit tests` success, then `DISK floor end` RED (15.8 GB vs 32 GiB) = RESOURCE red caused by my golden-331 pool. Ruled by doyle; comment 5829221926. LANDED by ff push, main = 0c467bfc, PR auto-MERGED at the same sha.
- #283 rebased onto 0c467bfc -> d3c90499. All 8 lane files blob-identical to 37c94d24 (checked by doyle).
- REAP golden-331 pool 08:17:22-08:20:23Z: outbound = real dir (no LinkType); inbound: no links in the earlier census. apparent 92,395,969,040 B; allocated 92,482,071,552 B; escaped 0 (0 inodes). Free C: 81,758,519,296 -> 178,401,632,256 (+96,643,560,704 = allocated +4.16 GB). Box census during the window: Runner.Worker 60596 -> 30780, CI cargo + rustc live, so the delta is UNINTERPRETABLE (a whole-volume write) rather than a finding. Fresh pool rebuild needed for the final head.

## Respin r2 (09:10Z) — candidate gate/331-v073-r2 @8d402d6b
- #283 gated on Windows at d3c90499: 33/33 (serving_in_brain 3/3 incl the new autostart arm, webserve_attachment PASS), clippy --workspace 0, treqs 0. Linux (todlando) 4/4. PR CI 36112215689 all green. LANDED by ff, main = d3c90499, PR MERGED at the same sha.
- Pools: todlando w9-pending closed both boxes (record todlando-w9-pending-retire-ir150-record.txt); his 145 MB hfenduleam gap overlapped MY xtask gen build.
- Fresh shape 8d402d6b (parent d3c90499): Cargo 0.73.0, CHANGELOG ## [0.73.0] - 2026-09-25, same text as 5507c487; NOT an ancestor of 5507c487 (checked).
- #283 changed the DocsStatus/announce seam after the 5507c487 full run, so BOTH crates run in full again on both OSes (Windows: doyle, env scrubbed; Linux: hertz).