Deployment & Infrastructure View
build
intermediate
Environment Topology
Which environments exist, what each one proves, what data it holds and who may deploy to it.
flowchart LR
subgraph e1["Ephemeral / PR"]
direction TB
a1["scaled to 1<br/>mocked externals<br/>synthetic data<br/><i>proves: it builds and unit-passes</i><br/>auto-destroy 24h"]
end
subgraph e2["Dev"]
direction TB
a2["shared<br/>real internal deps<br/>synthetic data<br/><i>proves: integration wiring</i><br/>anyone deploys"]
end
subgraph e3["Staging"]
direction TB
a3["prod-shaped topology<br/>real deps · sandbox externals<br/>masked prod-like data<br/><i>proves: contracts + performance</i><br/>pipeline only"]
end
subgraph e4["Production"]
direction TB
a4["full scale · multi-AZ<br/>real data<br/><i>proves: nothing — it serves</i><br/>pipeline + gate"]
end
e1 --> e2 --> e3 --> e4
note["Same artifact promoted throughout.<br/>Config differs; the image digest does not."]What it is
The environment ladder with, for each rung, the three facts that actually matter: what it proves, what data it holds, and who may deploy to it. An environment that proves nothing is pure cost, and most estates have at least one.
When you produce it
When setting up delivery for a new platform, and during any cost programme — non-production is routinely thirty to fifty percent of cloud spend and the first place to look.
Who reads it
Engineers, who need to know where to test what. Security and compliance, who care which environments hold real data. Finance, who fund the copies.
What good looks like
- Every environment states what it proves, in one clause.
- Data classification per environment, and production data is not in any of the lower ones unless masked with a stated method.
- Deployment permissions per environment.
- The same artifact is promoted, not rebuilt — the diagram should assert this.
- Non-production is scaled down and scheduled off outside working hours, and the diagram shows it.
Common mistakes
- Six environments where three would do, each drifting from the others.
- Staging that is not production-shaped, so it cannot prove the thing it exists to prove.
- Real production data in development, which is a breach waiting for an auditor.
- Rebuilding per environment, so what was tested is not what ships.