Trial environments

A trial environment is a running instance of the built application, built from the same snapshot as the deliverable, deployed near you, and destroyed when you are finished with it.

The one rule

The trial environment is the artifact. Not a demo of it, not an approximation of it, not a separately-configured variant with relaxed limits.

This is worth stating plainly because the alternative is common and quietly corrosive. A demo environment with fake data, no rate limiting, and a different configuration tells you about the demo environment. It tells you nothing about the thing you are about to ship, and the gap between the two is exactly where production surprises live.

Here, the only differences between the trial environment and the promoted build are:

  • The data is seeded, not real.
  • The lifetime is short, and it is torn down automatically.
  • The hostname is temporary, and credentials are scoped to the tenant.

Everything else — the code, the configuration shape, the migrations, the dependencies — is the same.

Lifecycle

queued ──▶ provisioning ──▶ ready ──▶ active ──▶ tearing-down ──▶ destroyed
                │                          │
                └────── failed ◀───────────┘

Provisioning creates the runtime, applies migrations, and loads the seed data. The environment gets its own hostname, its own credentials, and its own outbound allowlist.

Ready means the health check passed and the seed data is loaded. The environment URL and the credentials to reach it are issued at this point.

Active is the ordinary state. Usage is metered here.

Tearing-down destroys the runtime and deletes the storage the environment owned. It is not reversible, and it is not meant to be.

Idle reclamation runs automatically. An environment that has not been touched for the idle window is torn down and can be re-provisioned on demand from the same snapshot — which is fast, because the snapshot and the build are immutable. Re-provisioning gives a fresh environment with fresh seeded data, which is useful when you want to retry a flow from a clean state.

Seed data

Seed data is derived from your specification. It contains the entities your scenarios reference, in the states your scenarios mention — the settled order, the order outside the refund window, the customer without a subscription, the subscription at the boundary.

That matters for two reasons.

It makes the environment immediately usable. You do not have to invent fixtures before you can try anything.

It makes the specification testable by a human. If the seed data does not contain a case your scenario implies, that is a gap in your specification, and you find out before you find it in production.

By default seed data is synthetic. Connecting a real dataset is an explicit opt-in per project, and it is logged as such, because the environment then holds real data and the usual obligations apply.

Regions

Trial environments are deployed in the region closest to you: us-east-1, eu-west-1, or ap-southeast-1, among others.

The control plane runs in a single region, which is a deliberate simplification — but it does not constrain where the trial environment lives. The artifact runs where the user is, because the trial is the thing being evaluated, and an evaluation performed over a 200-millisecond round trip is a worse evaluation.

The platform chooses the region from your account setting, and you can override it per project. Where a build depends on a regional service — a payment provider with a domestic requirement, for instance — the project can pin the region, and the platform will tell you when a build has run in a region that does not match.

Isolation

Each environment is single-tenant and disposable:

  • Its own credentials, scoped to that environment.
  • A read-only root filesystem.
  • An outbound allowlist — only the dependency registries and mock services the build declared.
  • No access to other tenants’ storage, queues, or networks.
  • No persistent volumes that outlive the environment.

Secrets are injected into the running process and never written to logs. The build log records which secret names were present, never their values.

Access

You get a URL and scoped credentials when the environment is ready. Access is authenticated, and the credentials are valid only for that environment and only for its lifetime.

For sharing a trial with someone outside your tenant, the platform issues a time-limited, read-only observation link rather than sharing credentials. The link can be revoked, and it expires on its own.

Cost and limits

Environment time is metered from ready to teardown, by region. The idle window is generous on the free tier and configurable on paid tiers.

Region matters for cost, not just for latency: the same environment hour costs different amounts in different regions. The pricing on the pricing page reflects this, and the build report states which region a build ran in.

When to promote

Promotion is your action, not the platform’s. You take the source, the build recipe, and the verification suite, and deploy them into your own infrastructure through a pipeline you already have.

The platform does not hold your deployment, does not become a runtime dependency of your production system, and does not get a vote on when you ship. The trial environment exists to answer the question "is this the right thing?" — and then it goes away.

Next