The approval is the checkpoint, and what a declaration is #10

Merged
jschoubben merged 3 commits from design/approval-is-the-checkpoint into main 2026-08-26 18:37:43 +00:00
2 changed files with 79 additions and 1 deletions
Showing only changes of commit 23232e019a - Show all commits
+1 -1
View File
@@ -41,7 +41,7 @@ incident behind it is not written down, and the fix is to write it down, not to
| **Never bypass the pipeline** | No manual database edit, no manual restart as a workaround. Fix the cause and deploy. A workaround that works is a workaround that is never removed, and the next person cannot tell the node from its declaration. |
| **Never create a symlink** | A hand-made link caused production data loss through container volume resolution, and the judgement needed to make a safe exception is exactly the judgement unavailable at the moment it matters. **The mesh creates none at all** ([ADR 0018](../02-DECISIONS/0018-the-mesh-creates-no-symlinks.md), which supersedes [ADR 0011](../02-DECISIONS/0011-the-installer-owns-linking.md)). The links the installer still reconciles are a migration, not a permission. |
| **Never push directly to the main branch** | Branch, push, review, merge. Every merge is a human checkpoint, without exception — **including in this repository**. A documentation repository is not a lower tier of care; a decision record lands the same way a service does. |
| **One change per pull request, and never merge your own** | Unrelated improvements bundled together cannot be reviewed or reverted separately. Self-merging removes the checkpoint that is the entire point. |
| **One change per pull request, and never merge unapproved work** | Unrelated improvements bundled together cannot be reviewed or reverted separately. And the checkpoint is **a person deciding, not a person clicking** — work may be merged by whoever wrote it once a human has explicitly approved *that merge*, and never on a standing permission, an instruction to do the work, silence, or the author's own judgement that it is ready. [ADR 0042](../02-DECISIONS/0042-approval-is-the-checkpoint.md) |
| **Never open a pull request unprompted** | A permissions list saying it is allowed is not a request. |
| **A failed step fails the job** | A sequence that continues past a failure does the next thing in the wrong place. Gate each step on the last. [ADR 0008](../02-DECISIONS/0008-a-failed-step-fails-the-job.md), and §5. |
@@ -0,0 +1,78 @@
---
status: accepted
date: 2026-08-26
deciders: jochen
reconstructed: false
---
# 42. The approval is the checkpoint, not the second pair of hands
## Context
[§2](../00-META/how-we-build.md) reads *"one change per pull request, and never merge your
own"*, and gives the reason: *"self-merging removes the checkpoint that is the entire point."*
The rule was written for people. Applied literally to an agent it produces a contradiction
that surfaced immediately: an agent asked to merge cannot merge, because it authored what it
is being asked to merge. Every merge this repository has seen was therefore either performed
by the thing that wrote it, or not performed at all.
Stated by the operator: *"you are allowed to self-merge, after I've given you explicit
approval."*
## Considered options
1. **Read it literally.** An agent never merges; the operator merges by hand every time.
Rejected: it makes the operator a mechanical step rather than a reviewer, and a step
performed dozens of times without reading is not a checkpoint — it is the *shape* of one,
which is worse, because the record then claims a review that did not happen.
2. **Drop the rule for agents.** Rejected: the rule is right, and the failure it prevents —
work merged with nobody having looked — is not one an agent is less prone to.
3. **Name what the checkpoint actually is.** Chosen.
## Decision
**The checkpoint is a person deciding, not a person clicking.**
An agent may merge its own work **when a human has explicitly approved that merge**. The
approval is the review; the merge is bookkeeping that follows it.
Without an explicit approval, nothing changes: the agent does not merge, and
[§2](../00-META/how-we-build.md)'s *never open a pull request unprompted* continues to mean
that a permissions list is not a request.
**What "explicit" excludes**, because this is the half that can rot:
- A standing permission granted once and cited forever.
- An instruction to do the work, read as approval to merge it.
- Silence.
- The agent's own judgement that the work is ready.
## Consequences
- **The record must show which it was.** A merge performed on approval and a merge performed
unilaterally look identical afterwards, and the difference is the whole rule. Until the forge
can record an approval, the merge commit says so — which is weaker than a recorded review and
is stated here rather than left to be discovered.
- **Branch protection would make this structural instead of textual.** Required approvals turn
the rule into something the forge enforces. It cannot be set through the API on the forge
version in use, so this remains a rule held by intention — the condition
[§5](../00-META/how-we-build.md) exists to name.
- **The one-change-per-pull-request half is untouched**, and was broken twice on 2026-08-26.
Both breaches are recorded in the pull requests themselves rather than in a conversation.
- **This narrows a rule rather than relaxing one.** The obligation moves from *who performs the
merge* to *whether a person decided*, which is a higher bar in the case the original wording
actually permits: a reviewer who merges someone else's work without reading it satisfies the
old rule and fails this one.
## Sync
Run and verified 2026-08-26. §2 in the enforced page now reads *never merge unapproved work*,
confirmed by reading the rule back out of the live page rather than by trusting the publish —
the previous sync reported success and changed nothing.
## References
- [`how-we-build.md`](../00-META/how-we-build.md) §2 — the rule, now carrying this.
- [ADR 0008](0008-a-failed-step-fails-the-job.md) — the standard a checkpoint is held to: a
step that reports success without doing anything is the fault, not the shortcut.