Agent-utilities scale claim register

This is the AU-side, source-anchored register for scaling, W3 deployment assets, and capacity statements. It deliberately does not copy epistemic-graph claims: the source anchors below are files in this repository, and a claim is only as strong as the AU source or evidence named here.

Status vocabulary

The status is an evidence level, not a promise about an operator's cluster:

• DESIGNED — an AU plan or workload model exists; no implementation or run is implied.
• IMPLEMENTED — the AU source path or committed reference asset exists. This does not mean it has been deployed.
• UNIT-PROVEN — a focused automated fixture exercises the AU path. The fixture still needs to pass in the current release before it is a release claim.
• LAB-PROVEN — a bounded external run passed against a named release and environment; no such run is implied by a unit fixture.
• LIVE — an operator observed the exact deployed artifact, configuration, and health evidence at a recorded time.
• 1M-CERTIFIED — the signed production certification campaign passed the exact one-million-resident workload contract and SLOs.

reference and staged describe committed deployment inputs, not additional status levels. A reference/staged manifest is not deployable until its release compatibility gate, exact digest rendering, and asset checks pass. A deployable render is not LIVE until an operator records deployment evidence. No AU claim below is currently LAB-PROVEN, LIVE, or 1M-CERTIFIED.

Machine-checked claims

The checker in scripts/check_scale_claims.py requires each source anchor and source fragment to remain present. It also rejects duplicate IDs, path escape, unsupported statuses, and ungrounded LIVE/1M-CERTIFIED rows. Run it from the repository root before changing a scale claim.

ID	Status	AU source anchor	Required source fragment	Evidence scope
AU-SCALE-001	IMPLEMENTED	agent_utilities/orchestration/fleet_actuation.py#class KubernetesActuator	class KubernetesActuator:	optional kubectl reference actuator; source capability, not live cluster proof
AU-SCALE-002	IMPLEMENTED	agent_utilities/orchestration/fleet_actuation.py#selection in ("k8s", "kubernetes")	selection in ("k8s", "kubernetes")	config selection falls back to dry-run when kubectl is unavailable
AU-SCALE-003	IMPLEMENTED	agent_utilities/core/config.py#fleet_actuator: str	fleet_actuator: str = Field(default="dryrun"	default is inert; mutation remains policy-gated
AU-SCALE-004	IMPLEMENTED	agent_utilities/orchestration/fleet_reconciler.py#def parse_scaling_spec	Required: ``max`` (ceiling), ``signal`` and ``target`` (>0).	explicit floor/ceiling/target contract; invalid blocks are ignored
AU-SCALE-005	UNIT-PROVEN	agent_utilities/orchestration/fleet_autoscaler.py#def compute_desired_replicas	desired = math.ceil(eff * per_replica / spec.target)	focused math path; root must run the AU unit suite
AU-SCALE-006	IMPLEMENTED	agent_utilities/orchestration/scaling_signals.py#_PROMQL_TEMPLATES	container_label_com_docker_swarm_service_name	built-in CPU query is Swarm-shaped; K8s needs a custom/injected metric mapping
AU-SCALE-007	IMPLEMENTED	docs/scaling/capacity_model.py#RESIDENTS_PER_ENGINE_SHARD	RESIDENTS_PER_ENGINE_SHARD = 50_000	AU planning constant, not a measured multi-host capacity
AU-SCALE-008	DESIGNED	docs/scaling/capacity_model.md#1,000,000 residents	MODELED — the documented reference case	one-million case is a workload/capacity model until certification evidence exists
AU-SCALE-009	UNIT-PROVEN	tests/scale/test_capacity_model.py#def test_one_million_matches_documented_numbers	p.kafka_partitions == 8	arithmetic/doc consistency only; not a 1M runtime certification
AU-SCALE-010	IMPLEMENTED	deploy/k8s/production-cell/autoscaling.yaml#kind: HorizontalPodAutoscaler	kind: HorizontalPodAutoscaler	committed HPA reference/staged asset; not evidence of a live HPA
AU-SCALE-011	IMPLEMENTED	deploy/README.md#check-graphos-compatibility	kubectl apply -k RENDERED_DIRECTORY	exact release gate + rendered directory are required before deployment
AU-SCALE-012	UNIT-PROVEN	tests/unit/test_fleet_actuation.py#def test_scale_service_issues_real_kubectl_scale_call	deployment/graph-os-dispatch	mocked argv contract only; no cluster was contacted by the fixture