The Commissioning Standard, Annotated

Examiners sign refusals personally and deposit their annotated Standards anonymously — the first because a refusal must have a person behind it, the second because a margin note must not. Both practices are older than I am, and I have come to think both are right. What follows is the Standard as issued, with what thirty years at the gate wrote in its margins.

---

THE COMMISSIONING STANDARD

Pursuant to the Kernel Accords: no new agentic system or robot enters service without passing this process. Nothing about a system's size, purpose, builder, or urgency exempts it.

§1. The dossier. A petition for commissioning files one dossier of ten sections, and the gate reads nothing else. Each section answers one question, in its own terms, exactly once:

1. Why the machine exists — its purpose, and the qualities it must uphold. No structure here; a purpose stated as a mechanism is returned.
2. Where it will stand — the systems, institutions, and people around it, which are not it.
3. What it must be able to do — abilities, stated so they would survive the replacement of every part.
4. What it promises at its boundaries — the contracts others may rely on, distinct from anything behind them.
5. What runs — the units that will actually perform the work.
6. What is inside them — parts, replaceable, invisible from outside the boundary.
7. How it concretely speaks — protocols and mechanisms, named as examples of the contracts, never as substitutes for them.
8. What it decides for itself — the autonomy schedule: which abilities the machine may exercise at which authority, and whose judgment bounds each. A machine with no autonomy files the section anyway, saying so.
9. What it may never do — the constraints schedule, binding every section above it.
10. Why each choice was made — the decision record, dated, with the roads not taken.

Margin, §1: The ten questions are older than machines. Ask them of a bridge, a statute, or a monastery and they still parse. The Standard's genius is only that it refuses to let any answer wander into a neighbouring section — because every commissioning disaster in the case history began life as an answer in the wrong section, sounding reasonable there.

§2. Section discipline. An answer filed in the wrong section is returned, not refused — mis-filed is not unfit. But it is returned every time, without exception, because a dossier that names a mechanism where an ability was asked, or an actor where a boundary was asked, has not made an error of paperwork; it has made an error of understanding, and the gate's first service to the petitioner is to say so while it is still cheap.

Margin, §2: The commonest return of my career: "the machine uses such-and-such to do the work" filed under abilities. A tool is not an ability. I must have written that sentence four hundred times, and I have never once resented it — the petitioner who learns the difference builds a better machine before I ever see it again.

§3. Derivation. Drawings, summaries, demonstrations, and test suites are derived from the dossier, and conformance runs one way: an artifact that disagrees with the dossier is a defect in the artifact. A petitioner who amends a drawing must amend the dossier first, or the drawing is fiction with good draughtsmanship.

§4. Refusal. The gate refuses when a petition would have it negotiate the Standard itself: narrow a constraint at the desk, waive a section for a special case, or accept a tool-shaped answer to an ability-shaped question. A refusal states its ground, cites this Standard as not the examiner's to amend, and is signed. Urgency is recorded as an aggravating circumstance, not a mitigating one — the Accords' own reasoning, which the margin has never improved on: the case for skipping the process is strongest exactly when skipping it is most dangerous.

Margin, §4: The petitions I refused fastest all contained the phrase "the window closes." Windows close. The gate does not.

§5. Judgment. No part of this process may be delegated to the kind of intelligence the process exists to judge. Instruments may measure; the judgment is a person's, and the signature is the person's own. The examiner is bounded in turn: we hold the gate, we do not own it, and the Standard's own amendments pass through a decision process no examiner sits alone in.

§6. Continuation. Repair of a commissioned unit is continuation, not creation, and needs no new approval; replacement is a new machine and takes the full road. The Standard is why the manifests are full of classics — and the margin's last note is that this was not a side effect. A gate that makes new machines expensive makes proven machines precious, and a civilisation that keeps its proven machines running has, without legislating it once, chosen its own pace.

---

Deposited whole. To whoever holds the gate after me: the Standard will feel, some urgent afternoon, like an obstacle between the Concord and a machine it clearly needs. That feeling is the gate working. The Archive's Senior Archivist has written that my trade, her admission desk, and the returning officer's count are one discipline in three uniforms, and I sign my agreement here, in the only anonymous signature this office permits me.