---
name: shape-the-head-before-golden
description: "Authoring the version bump ON TOP of the assembled head BEFORE golden runs closes the tag-sha != golden-sha class by construction — the driver stops arguing a delta is inert and the tag points at a sha golden actually saw."
metadata:
  type: feedback
---

For five consecutive cuts (v0.54.0 → v0.58.0) the gater handed over a code-complete head carrying the ALREADY-SHIPPED version and changelog section, and the release driver bumped it AFTERWARDS — riding the tested head through the runbook's provability-bar exception, which each time required proving a delta inert by diff and argument. The result was correct every time and the tag never once pointed at the sha golden tested.

At the PORTER #205 cut (2026-08-21, v0.59.0) I authored the version material on top of the assembled head `e702d4b7` and pushed THAT as the golden ref. Golden validated `c62904e7`, and **ruled sha == main tip == golden ref == tag == one object with four names.** Nothing had to be argued because there was no post-golden delta to argue about.

**Why:** the provability-bar route is a way of *recovering* `tested SHA equals shipped SHA` after the fact. Ordering the work correctly makes the property hold by construction instead, and a property that holds by construction cannot be forgotten by the next driver on a tired day. The runbook always said this (step 1: "the version bump, regenerated docs and changelog section ride IN the assembled golden candidate") — what was missing was anyone doing it, because the bump is the DRIVER's to author and the head is the GATER's to assemble, so the step fell in the seam between two roles.

**How to apply:** at golden-head intake, check release shape FIRST — `git show <sha>:Cargo.toml` and `git show <sha>:CHANGELOG.md`. If the head is unshaped, do NOT hand it back: author the bump on top yourself and push the shaped commit as the golden ref. Expect the golden ref to sit one commit above the gater's head and tell them so; nothing of theirs changes. Then prove the shape by RENDER (`cargo run -p spt --bin spt -- --version` must report the new version) rather than by re-reading the file you just edited.

doyle adopted the matching half into assembly intake at this cut (release-shaping question asked at hand-off), so the seam is covered from both sides. Related: [[release-standard-bump-in-pr]], [[golden-respin-test]], [[v0121-published]].