Key concepts
Ten ideas carry the rest of this documentation, and they mean the same thing in every edition of the BuiltForProd Baseline: AWS, Azure and GCP. Each gets a paragraph here and a full page in the concepts section or the glossary.
How the pieces nest
Landing zone
A landing zone is the multi-account foundation every application runs on: the top of the cloud's hierarchy and the accounts, subscriptions or projects under it, the network that connects them, the security and compliance services that watch them, identity and access for people and pipelines, and the state and delivery machinery that manages all of it. Each edition of the Baseline is a landing zone held in one repository. Application teams do not work in it; they build on what it publishes. Read landing zones.
Account, subscription or project
Each cloud's isolation unit is the hard boundary for access, billing and blast radius, and every edition creates 15 of them: core holds shared, single-purpose units (root, security, audit, identity, network, dns, artifacts, auto, corp, public, backup) and plat holds one workload unit per stage (sandbox, dev, staging, prod). On GCP the backup project sits in a top-level folder of its own.
- AWS Enterprise Baseline. 15 AWS accounts in two organizational units, the management account plus 14 members; each receives
account-baseline(default EBS encryption, the S3 public-access block, AWS Config, the deployer role and the read-only reader role). - Azure Enterprise Baseline. 15 subscriptions under three management groups; each receives
subscription-baseline(resource-provider registration, Defender for Cloud plans, the Activity Log export, Network Watcher, a locked baseline resource group). - GCP Enterprise Baseline. 15 projects in three folders; each receives
project-baseline(APIs, a de-privileged default service account, the project's deployer, log retention, OS Login, Essential Contacts).
The architecture overview draws all three.
Unit
A unit is one deployable module instance with one state file: a VPC or VNet, a firewall policy, a detector. Its definition lives once in units/<unit>/terragrunt.hcl and says which module to use, which dependencies it reads and which inputs it passes. Per-instance data arrives as values. The same definition serves every account, subscription, project or stage that lists it. Read Terragrunt units and stacks.
Stack
A stack file, terragrunt.stack.hcl, sits in a region or global folder and lists which units that folder deploys, with their values. Repeated sets are templates under stacks/: the units every workload unit has are plat-account (AWS, seven units), plat-subscription (Azure, eight) and plat-project (GCP, four), and every edition's blueprints have webapp-stage and etl-stage. Terragrunt generates the units into a git-ignored .terragrunt-stack/ tree; nothing under environments/ is hand-written unit code. Read Terragrunt units and stacks.
Blueprint
A blueprint is an opinionated, production-ready reference architecture for a real workload, delivered as its own repositories and deployed into the workload stages. Each edition has three: Web App, Data and ETL, and Secrets.
- AWS. The Web App Blueprint (EKS, ArgoCD, DocumentDB, ElastiCache, CloudFront), the Data and ETL Blueprint (S3 data lake, Glue, Lake Formation, Lambda) and the Secrets Blueprint (SOPS and KMS to SSM Parameter Store).
- Azure. The Web App Blueprint (AKS, ArgoCD, Front Door, Cosmos DB for MongoDB vCore, Azure Managed Redis), the Data and ETL Blueprint (Data Lake Storage, Azure Databricks with Unity Catalog, a Container Apps job trigger) and the Secrets Blueprint (SOPS with Key Vault keys, synced to Key Vault).
- GCP. The Web App Blueprint (GKE, ArgoCD, Cloud Armor, a Cloud Run front end behind Cloud CDN, Firestore, Memorystore for Valkey), the Data and ETL Blueprint (Cloud Storage lake, Dataplex, BigQuery, Dataproc Serverless, an Eventarc-driven Cloud Run trigger) and the Secrets Blueprint (SOPS with Cloud KMS, synced to Secret Manager).
A blueprint never creates what the landing zone owns; it reads network IDs, DNS zones, the registry and identities from the landing-zone contract. Read the blueprints.
Entitlement
An entitlement is the product your organization holds, and it decides which documentation sets you can open. The Overview set is public. The AWS, Azure and GCP Baseline sets and each blueprint set show their overview to everyone and the rest to organizations entitled to that product. Entitlements are stored on your organization by BuiltForProd staff; you see them after signing in. Read how to use these docs.
Environment
In the Baseline, an environment is a region slug: usw2 for us-west-2 on AWS, eus2 for eastus2 on Azure, usw1 for us-west1 on GCP, and glob in every edition for the global folders that hold organization-wide units. A stage is the workload unit's role: sbx, dev, stg, prd. Resource names combine namespace, environment and stage: acme-usw2-prd-vpc, acme-eus2-prd-aks, acme-usw1-prd-gke. Read environments and promotion.
Promotion
Promotion moves a change from dev through staging to prod without rebuilding it. Infrastructure merges to main and applies: the blueprint infrastructure repositories run a dev, staging and prod matrix whose prod job waits for the prod GitHub Environment, and the Baseline applies in one job under the prod Environment. Application images are built once, tagged immutably as main-<short sha>, given a vX.Y.Z tag at release, and never rebuilt for a later stage. Rollback is a Git revert or a promotion of the previous tag. Read environments and promotion.
GitOps
GitOps means the desired state of a system lives in Git and a controller reconciles the running system to it. In the Web App Blueprint of every edition, ArgoCD reads the Helm chart and per-stage values from the GitOps repository. No pipeline pushes to that repository directly: the application repository opens a pull request per rollout, dev and staging merge themselves, prod is merged by a person and synced manually. Read GitOps.
Placeholder values
The repositories are templates. Every value that must be set per deployment carries a TODO: comment and every optional choice an @optional: comment, so git grep lists both. This site follows the same idea: acme is the namespace, ACME the group prefix, company.com the primary domain, and the home regions are us-west-2, eastus2 and us-west1. The AWS accounts use 111111111111 through 151515151515; the Azure tenant is aaaaaaaa-aaaa-aaaa-aaaa-aaaaaaaaaaaa with one repeating GUID per subscription, such as 11111111-1111-1111-1111-111111111111; the GCP organization is 161616161616 and the billing account AAAAAA-BBBBBB-CCCCCC. GCP project IDs are built from the namespace, such as acme-plat-dev. Signed-in readers whose organization profile is filled in see their own values in place of the samples. Read how to use these docs.
Where these ideas come from
The vocabulary is the one the repositories use. The contributor rules define unit, unit definition, stack file, template, values and generated tree exactly as above, and the checks in every repository enforce them: a stack file beside every region.hcl, no hand-written unit under environments/, every definition referenced, every dependency mocked so a new environment can be planned end to end before anything exists. The infrastructure as code page explains why that discipline matters.