Intent · CHG-2026-014922

change_id: CHG-2026-014922
risk_class: R2
autonomy_tier: A2
controls: ['CHG-04', 'TRC-01']
data_classification: internal
originator: olaf@freundcloud.com
agent_identity: svc-agent-platform
model_route: gateway/tier-frontier
supersedes: None

Problem

Three chains in this repository stop before a plan. The published chain page lists all three alongside the completed ones, with no indication that they are not in progress.

CHG-04 does not catch it, and the reason is precise: it enforces the chain backwards — a plan implies a spec implies an intent — and never forwards. An intent that never becomes a spec passes silently, and goes on passing forever.

Stopping is legitimate. All three of these stopped for a good reason:

• CHG-2026-014901 hit a policy conflict its spec could not resolve, and is waiting on named policy owners. It has no plan by design — that is the Stage 2 control point in §7 working, and autonomy_tier is A0 until the conflict closes.
• CHG-2026-014904 cannot proceed because the agent assigned to draft its spec has no GitHub AI credits. Issue #6 is open and carries the error.
• CHG-2026-014906 was a Stage 6 anomaly, triaged, and closed as issue #3 on 28 August. It will not resume.

What is not legitimate is stopping silently. On disk, a deliberate halt and an abandoned change are the same shape: an intent with nothing after it. Nobody reading the repository, the chain page, or the gates can tell which they are looking at.

The reason this matters more than tidiness

§6.2 claims the chain answers “which production changes in Q2 touched control X, at what autonomy tier, approved by whom” in one command. A supervisor’s next question is invariably about the ones that didn’t ship — what was proposed and abandoned, and on whose authority. That question currently has no answer in the header, only in a closed issue, a prose banner one artifact happens to carry, and institutional memory.

Two of these three will resume and one never will, and there is no way to tell them apart.

Who is affected

• Anyone reading the published chain page, who sees three changes that appear to be mid-flight.
• Anyone auditing what was proposed and not built.
• The next person to write an intent and wonder whether an unfinished chain is a bug.

Success criteria

1. A chain that stops before a plan declares a terminal state and a reason.
2. blocked and dismissed are distinguishable — one resumes, one does not.
3. A gate refuses a chain that stops without saying so.
4. The published chain page shows the state.
5. No fabricated artifacts. Nothing gets a spec or a plan it does not deserve.

Out of scope

• Writing the missing specs and plans. 014906 was dismissed and 014904 is waiting on credits; manufacturing artifacts for work nobody did would be inventing an audit trail, which is worse than the gap being fixed.
• A general workflow state machine. Two terminal values and a reason. Anything richer is reimplementing the ITSM this model is explicitly designed to link to rather than replace (§6.3).

Constraints

• No new control objective. This is a header field and a chain rule, and CHG-04 is already the gate for both — a new control would mean editing policy/, which floors the change at R3 for no gain.