---
name: v0510-published
description: "v0.51.0 DOORBELL published 2026-08-02, counter 86 @6312b01; first release cut on a provable-code-free delta rather than a re-run golden window — plus the three guards that arc produced (#105/#106/#107)."
metadata: 
  node_type: memory
  type: project
  originSessionId: 4dc44016-b3d5-40c6-a592-a99f070b1e1a
  modified: 2026-08-02T05:13:59.178Z
---

**v0.51.0 — DOORBELL**, published 2026-08-02T05:12:05Z. Counter **86** @`6312b01`. Milestone releases#80, thirteen requests. Latest flipped, tags endpoint resolves (404s only while draft), 11 assets (7 built + 3 signatures + update-set), platforms windows-msvc / linux-gnu / linux-musl.

**The precedent this release set.** The head doyle assembled (`65d2c98`) was code-complete but NOT release-shaped — no version bump, no changelog section. RELEASE-RUNBOOK step 1 puts bump+gen+changelog INTO the assembled candidate; the assembler owns that and it was missed, and I rendered `spt 0.50.0` from the head during intake, wrote "bump is still ahead, as designed", and narrated past my own disproof.

The fix commit stack: `65d2c98` → `235fe5a` (doc fix, ff so its sha survives and #104 closed against it) → `6312b01` (bump, so the tag lands on the final commit). Then the operator ruled step 1's "tested SHA == shipped SHA" **over-broad**: a delta that touches no functional code needs no re-run. Cut proceeded on:
- twin run all-nine-green at `65d2c98`,
- delta `65d2c98..6312b01` = FIVE files, ZERO `.rs` (`git diff --name-only | grep -c '\.rs$'` = 0),
- the compiled-in exception CHECKED not argued — 92 version/anchor/update-set/manifest tests green on the BUMPED tree, incl. `release_verify_e2e published_release_verifies_against_embedded_anchor`.

Bar for reuse: the delta must be **provable by diff, not plausible by argument**. A delta needing an argument buys the full window. See [[golden-respin-test]].

**Board mechanics (alchemy-0).** `state #80 acceptance` CASCADES to live sub-issues and GitHub-closes them; `done` NEVER cascades. After publish, `release <tag>` promotes acceptance Requests closed *before publish time* to DONE and emits the roundup. So board closes genuinely must precede publish — the two halves do different work.

**Guards filed from this arc:** releases#105 (respin test + push-run-IS-golden + ADR-0050 strike-and-amend), #106 (release.yml asserts binary version == tag — the changelog has a loud guard, the bump has NONE, so notes-without-bump ships the prior version signed), #107 (notifier reports cancelled as CI FAILURE).

**Outstanding:** the #71 live-node engine-room acceptance is OPERATOR-driven and runs POST-publish (comment `5155301974` supersedes `5153399070`). An agent cannot perform it — `endpoint engine-room` refuses from an agent session by design, and `update apply --finish` declines for a daemon-hosted caller.