---
name: v041-published
description: v0.4.1 PUBLIC 2026-06-11 (counter 8); D7 seamless no-bounce fleet-verify payload; deployah drove full merge→tag→sign→publish solo
metadata: 
  node_type: memory
  type: project
  originSessionId: 1fb9fc52-b258-4b44-9c4c-7cae9345098f
---

v0.4.1 SIGNED + PUBLISHED 2026-06-11 (release counter 8; v0.3.2=6, v0.4.0=7, v0.4.1=8 — counter != semver). Live agent **deployah** drove the FULL publisher sequence solo (not just sign): merge PR#4→main (merge 038257b) → tag v0.4.1 on main + push → release.yml drafted both runners → `release-publish --tag v0.4.1 --key-id rel-primary-2026 --version 8` → flip public.

Acceptance hashes (brain.ready exe_hash must MATCH, NO manual bounce): linux `35c9544c…17314f08`, win `728a1bfc…00df62d70`. signed `54188f12…05e3ec9`. update-set.json v8. https://github.com/SaberMage/spt-releases/releases/tag/v0.4.1

**Why:** v0.4.1 is the D7 SEAMLESS no-bounce fleet-verify payload — v0.4.0→v0.4.1 apply must roll new bytes live with brain pid change + NO manual daemon bounce = restoration acceptance proof (the assert [[v040-published-fleet-rolled]] deferred because in-process 0.3.2 couldn't self-relaunch). hfenduleam pump wedge (broker 17232/brain 38324/STALLED 19:42) heals on apply = bonus before/after. enlyzeam (<0.3.2) catch-up = project's FINAL manual bounce.

**How to apply:** Publisher path proven repeatable — see Psyche commune. Seed `SPT_RELEASE_SEED` resident in ENV (64-hex), inherited by fresh pwsh children. Don't re-plan the publish.

**FLEET ROLL OUTCOME (2026-06-11):** hfenduleam (Win) GREEN — seamless no-bounce proven, brain runs published 728a1bfc, broker pid held, wedge healed. **kitsubito (Linux) RED → restoration close BLOCKED.** Apply put correct v0.4.1 (35c9544c) on disk + recorded "applied v8", but resident v0.4.0 broker respawns brain via `current_exe()` which on Linux FOLLOWS the apply-rename to `spt.old-8` (b4b7ea14=v0.4.0) → brain execs OLD bytes; readiness passed on old brain = false-success, only exe_hash assert caught it. **Release artifacts are SOUND — do NOT yank v0.4.1.** Bug is in already-deployed v0.4.0 broker respawn logic (brainproc.rs). Windows dodges it (can't rename running PE in place).

**v0.4.2 FIX (todlando driving, doyle confirming):** respawn brain by resolved CANONICAL install path not current_exe(); gate readiness-success on bytes-assert (respawned exe_hash==artifact). Needs one-time MANUAL Linux broker bounce to load (resident broker can't self-fix) = project's real FINAL manual bounce, paired w/ enlyzeam catch-up. Mint REQ-HAZARD before fixing (deployah flagged): (1) "post-apply respawn must exec applied bytes not renamed old binary" + Linux exe_hash==artifact test; (2) readiness-success-without-bytes-assert gap. deployah takes v0.4.2 publish = **counter 9** when green. Still-deferred DEFERRED.md gaps: CI-flake applyhost socket-bind kitsubito; REQ-UPD-8 single→set same-version re-fetch rollback-floor trap.