Skip to main content

The BuiltForProd Standard

The BuiltForProd Standard defines what production-ready means: a system is built for production when it is Secure, Reliable, Observable, Scalable, Automated, Recoverable, Compliant and Operable. Each property is a design constraint, not a later upgrade, and each pillar page on this site names the units, services and workflows in each Baseline edition and its blueprints that deliver it.

The eight pillars​

PillarA BuiltForProd system...Page
Secureprotects identities, data, workloads and supply chainsSecure
Reliablecontinues operating through expected failuresReliable
Observablemakes its state and behavior visible to operatorsObservable
Scalablegrows predictably without architectural failureScalable
Automatedmakes routine provisioning, deployment, validation and operations repeatableAutomated
Recoverablehas tested recovery mechanisms for failures and disastersRecoverable
Compliantcan demonstrate controls and evidence consistentlyCompliant
Operablecan be understood, operated, troubleshot and changed safely by humansOperable

The BuiltForProd test​

The test is one question, "Would you trust this in production?", asked eight ways:

DimensionQuestion
SecurityIs it secure by default?
ReliabilityWhat happens when something fails?
ObservabilityCan operators see what is happening?
ScalabilityWhat happens when demand grows 10x?
AutomationHow much depends on manual intervention?
RecoveryCan the system actually be restored?
ComplianceCan controls and evidence be demonstrated?
OperationsCan another engineer operate it safely?

A demo passes none of these. A system that "works" answers only the first half of each question. The pillar pages answer all eight for each edition, the AWS, Azure and GCP Enterprise Baseline, and the blueprints that deploy on it, with the concrete mechanism and how you verify it in your own environment.

How the pillars map to the platform​

Each pillar is delivered by several parts of the platform, and most parts serve more than one pillar. The table shows the principal mechanisms per edition; the pillar pages hold the full lists.

PillarAWS Enterprise BaselineAzure Enterprise BaselineGCP Enterprise Baseline
SecureThree service control policies, IAM Identity Center, GitHub OIDC, KMS, SOPS, GuardDuty, Security Hub, Firewall Manager WAF, isolation domainsAzure Policy guardrails at mg-acme, Entra groups with Conditional Access and PIM, federated identities, Key Vault keys, Defender for Cloud, Front Door WAF, Azure Firewall19 organization policy constraints, IAM deny policies, Google groups with Privileged Access Manager, Workload Identity Federation, Cloud KMS, Security Command Center, Cloud Armor
ReliableThree Availability Zones, NAT per AZ, multi-AZ DocumentDB and Redis in staging and prod, ArgoCD HA, autoscalers and disruption budgetsFirewall and runners across three zones, zone-redundant storage, zone-redundant Cosmos DB and Redis in staging and prod, ArgoCD HA, autoscalers and disruption budgetsRegional GKE across three zones in staging and prod, a Valkey replica with automatic failover, regional Firestore, ArgoCD HA, autoscalers and budgets
ObservableOrganization CloudTrail, VPC Flow Logs, the CloudWatch OAM sink, seven CIS alarms, Fluent Bit, X-RayActivity Log and Entra logs in the audit workspace, VNet flow logs, 10 activity-log and 6 KQL alerts, Defender alerts, Container Insights, Application InsightsTwo aggregated organization sinks, flow logs on every workload subnet, a metrics scope over every project, 13 log-based detections, Cloud Trace
ScalableA /8 plan for 16 regions and 8 stages per region, IPAM, Karpenter and Cluster Autoscaler, Spot runners that scale to zeroA /8 plan, AKS pods outside the VNet, Node Auto Provisioning under a vCPU ceiling, Container Apps jobs that scale from zero, Databricks job clustersA /8 plan with eight stage blocks per region, node auto-provisioning under CPU and memory ceilings, bounded Dataproc Serverless batches
AutomatedPlan on pull request, apply on merge, drift detection, guard scripts, immutable images promoted by pull request, secrets synced on mergePlan and apply through a federated identity on Container Apps runners, drift detection, guard scripts, locked registry tags, secrets synced to Key VaultPlan and apply through Workload Identity Federation, drift detection, guard scripts, immutable Artifact Registry tags, secrets synced to Secret Manager
RecoverableVersioned and locked state, DocumentDB backups and deletion protection in prod, S3 versioning, Git revert, promotion of a previous tagVersioned state with 90-day soft delete, delete locks on the prod data stores, immutable audit storage, Git revert, promotion of a previous tagVersioned state with soft delete, Firestore point-in-time recovery and daily backups in prod, deletion protection, Git revert, previous tag
CompliantThe SOC 2 baseline conformance pack, 28 further packs, Security Hub standards, an audit-protection SCP, auditor permission sets that cannot read dataA custom SOC 2 baseline initiative in Audit, built-in framework initiatives in DoNotEnforce, DenyAction on the audit trail, write-once audit storageCIS Google Cloud Foundations detections, Security Command Center, an IAM deny policy on the audit trail, retention-locked archives, auditor deny policies
OperableWorkbooks and runbooks, CODEOWNERS by blast radius, twelve named roles, TODO: and @optional: tags, one priced switch file per concernCODEOWNERS by blast radius, twelve named roles as Entra groups, TODO: and @optional: tags, network.hcl and security.hcl with the price beside each switchCODEOWNERS by blast radius, twelve named roles as Google groups, TODO: and @optional: tags, network.hcl, security.hcl and identity.hcl

How to read the pillar pages​

Every pillar page has the same three parts: what the pillar means, how each edition delivers it, and how you verify it in your own environment without changing anything. The architecture overview gives the map the pillar pages assume, and the security posture summary condenses the Secure and Compliant pillars for a reviewer. The public overviews of the AWS, Azure and GCP Enterprise Baselines introduce each edition.