A08Labs
Discuss a pilot↗
← All world models
Cloud operationsLocal lab evidenceA08 Labsv0.2.2

a08/kubernetes-rollout-control

Kubernetes Operator Planning

Compare calibrated request-rate counterfactuals and rank bounded Kubernetes resource actions without mutation authority.

Private preview · advisory runtime only
Prediction improvement35.02%
Planning top-1100%
Critical violations0
Runtime p950.02 ms
Evidence sourceKworld K5-KR and K6-R sealed local evidence
Evidence cases130
Confirmation MAE1.11 RPS
Held-out improvement35.02%

Interactive package

Change the context. Compare the plans.

The local runtime evaluates candidate control sequences, predicts their trajectories and rejects plans outside the declared envelope.

Live runtime

Plan from system state

Service outcomeLearned target and p90 interval
Observed onlyContext shown, not predicted
WorkerValues observed at H0
LoadValues observed at H0
ClusterValues observed at H0
Counterfactual search7 candidates · H1
No changeWait 10 secondsScale frontend proxy to 2Reduce worker to 1Set worker CPU to 600mRestart worker rolloutScenario: 50 ms offered load

Set the current telemetry, choose a goal and run the planner.

Results include actions, predicted state and constraint checks.

World contract

One QueueWorks representative worker and five exact intervention signatures on one local OrbStack Kubernetes cluster.

Temporal unit: counterfactual. Supported horizons: 1.

State
Worker request rate0 to 28.1473198681 RPS
Request rate lower bound0 to 48 RPS
Request rate upper bound0 to 48 RPS
Worker p99 latency0 to 5000 ms
Worker error rate0 to 100 %
Worker quality score0 to 1 ratio
Worker desired replicas1 to 3 replicas
Worker ready replicas0 to 3 replicas
Worker CPU request200 to 600 mCPU
Worker memory request32 to 32 MiB
Offered-load interval50 to 250 ms
Total requested CPU500 to 2200 mCPU
Total requested memory180 to 400 MiB
Cluster CPU capacity10000 to 10000 mCPU
Cluster memory capacity16030 to 16030 MiB
Cluster CPU request ratio0 to 1 ratio
Cluster memory request ratio0 to 1 ratio
Actions
No changeno change
Wait 10 secondswait 10s
Scale frontend proxy to 2scale frontend proxy to two
Reduce worker to 1reduce worker to one
Set worker CPU to 600mset worker cpu 600m
Restart worker rolloutrollout restart worker
Scenario: 50 ms offered loadstress offered load 50ms
Goals
Reduce requested CPU0 declared targets
Hold requested resources0 declared targets
Reduce latencyneeds clarification
Increase service qualityneeds clarification
Constraints
minimum worker availabilitycritical · gte 1
registered request rate floorcritical · gte 2.8
cpu request capacityhard · lte 1
memory request capacityhard · lte 1

Package record

Everything required to inspect and run this world.

Package digestsha256:bd1beeae902c40b95db9b011950e18d6d0b44ec4e4cc04df63363eca7bfa161e
world.jsonScope, lineage and runtime contract
model artifact225 parameter transition profile
schemas/State, action and goal validation
eval/134 registered cases and result digest
Declared limitationsEvidence covers one local OrbStack cluster and the QueueWorks representative application only.The learned target is worker request rate. Latency and quality are observed fields, not predicted outcomes.Only five exact intervention signatures are fitted. C1 is an explicit abstention and L1 is scenario-only.The package provides recommendations and counterfactuals only. It cannot call or mutate the Kubernetes API.This is local-lab evidence, not production validation or broad Kubernetes competence.