# Golden head handoff — NOW-SIGNAL #23 / v0.67.0 (doyle → deployah, 2026-08-30 ~06:00Z)

## The head

- **`da71b785`** on `release/v0.67.0-head` (pushed to origin) = main tip `dc07e215` +
  shape-on-top commit (exactly 3 files per the `4d6007ac` precedent: Cargo.toml 0.66.0→0.67.0,
  Cargo.lock refresh, CHANGELOG `## [0.67.0]` authored under the operator's introduce-by-name
  rule). SHAPED BEFORE HANDOFF — the IR-54 intake question is pre-answered this time.
- Every wave already ff-landed on main beneath it: W1 `31020e02` (#245 UPDATES + #18 SEAL_BRIEF
  + #11b render rider), #113 `aa5f80f6` (echo cadence), W2 `cb78f710` (#44 attachment channel +
  #45 away alerts, one spacerun respin), W3 `dc07e215` (#133 shell hints). Tested sha == landed
  sha for every wave (pick-audit 1/1 MATCH each; batteries by exit file on #23's comments).
- Head gate at `da71b785`: clippy --workspace --all-targets -D warnings **0**, traceable-reqs
  **0** (post-disk-reclaim rerun; the first attempt's 101/101 was the disk at 0.01 GB free —
  classified per IR-59, re-run not re-read).

## Greenlit-form check inputs (yours to verify, this makes it cheap not discharged)

- Baseline you banked: **10 open members at greenlit** (9 at cascade + #245 operator-added
  02:10Z, alchemy audit trail); #33 already-DONE reconciles the 11-vs-10 count.
- State at intake (CORRECTED BY REPLACEMENT 06:20Z — the earlier "deferred-attached" framing
  here was retracted and properly discharged after this doc's first write): **8 members in
  ACCEPTANCE** (#11 #14 #18 #44 #45 #113 #133 #245; the #11b rider rode W1 and is not a
  member; #33 carries prior-arc DONE). **#16 + #17 are DISCHARGED DROPS**: detached from #23
  and moved back to EVAL 05:59:47–53Z (comment 5467039869 pairs the re-homing with the
  referral's reasons — both halves of the drop rule). The board is the record; this doc now
  matches it.
- DROPS: #16/#17, discharged as above. ADDS vs greenlight: #245 only (operator-authored).

## New-cells list (runbook clause; classification GROUND measured, not recalled: thin ci.yml
## runs `cargo nextest run --workspace -E 'kind(lib) + kind(bin)'` — ci.yml:114 — so
## INTEGRATION cells never execute in thin CI and every kind(test) cell below is first-CI here)

**First-EVER CI execution is THIS GOLDEN:**
1. `crates/spt/tests/attach_link_push_e2e.rs` (NEW, W2 #44/#45) — registered in the HEAVY set
   (both strings: `.config/nextest.toml` + golden.yml's `HEAVY` env, edited identically), so it
   runs in **phase B**. Executed so far: todlando's lane battery phase B (green) and my W2 gate
   named leg 1/1 at `5f474098`; the spacerun respin did not touch it. Real `spt daemon run`,
   both observers live — the heaviest new cell of the arc.
2. `crates/spt/tests/shell_hints_e2e.rs` (NEW, W3 #133) — NOT in HEAVY (deliberate, sibling of
   shell_e2e/now_signal_delta_e2e), so it runs in **phase A**. Executed so far: todlando's lane
   leg 1/1 first-run green and my W3 gate named leg 1/1 at `dc07e215`.

**MODIFIED integration cells whose new assertions first CI-execute here** (all four #113,
spt-daemon tests, phase membership unchanged): `psyche_context_file_e2e`,
`psyche_event_turn_e2e`, `psyche_real_bound_kill_soft_budget_e2e`,
`psyche_residency_expectation_e2e` — executed in todlando's #113 battery and my #113 gate's
full `-p spt-daemon` run (1141/1141) at `aa5f80f6`-equivalent tree.

**NOT first-CI** (for completeness): every new UNIT cell of the arc (UPDATES both arms,
SEAL_BRIEF two-sentence assertion, EDGE subject naming, age-gate/near-crossing arithmetic,
boundary first-wins) is `kind(lib)/kind(bin)` and has re-earned green in thin CI at
`31020e02`, `cb78f710` and `dc07e215`.

**Red-read rule this list exists for:** a phase-B red in attach_link_push_e2e or a phase-A red
in shell_hints_e2e on THIS run is a first-ever-CI-executed cell on a loaded runner — triage
starts at the cell's environment sensitivity (v0.56.0 precedent), not at product regression;
either way it hands back to me RCA-first.
- Board requests minted during the milestone (not members): #244, #246, #247, #254 — BACKLOG,
  operator triages.

## Run notes

- Drive the ACCEPTANCE cascade **BEFORE** publish (`state #23 acceptance` is already effectively
  done member-by-member; the milestone-level cascade + close ordering is yours) — the
  hub-card off-by-one (spt-progress-tool#78) stands until fixed; grace exists on the verb, not
  the cards.
- Fresh `pool-claim` before the first build in ns23-w1 / ns23-113 (claim records reaped with
  the targets) — you banked this; IR-42 framing correct.
- Two disk-window main-CI reds (cb78f710, dc07e215) were re-fired with `--failed` at ~05:57Z —
  check they re-earned green before reading any golden signal against those shas.
- Box: ~237 GB free post-reap; actions-runner warm cache HELD by three-way agreement.
- gate-ns23-w1 rig + pool stay MINE until golden verdict (respin insurance), then the
  milestone reap party takes everything (ns23-w1, ns23-113, ns23-w3 worktrees + gate rig;
  conhost CWD-pin check first per IR-14's hygiene rule / IR-63 third face).

## Red hand-back

Golden reds hand straight back to me for triage — RCA-first, same-sha rerun decisions ride the
triage, never precede it.