Infrastructure code proved wrong without a cloud account, with what that cannot show stated on the same screenful.
What it is. Two Terraform-compatible binaries, three modules, a Helm chart and an Ansible playbook, every one applied to moto and none to an account. 100 tests run with nothing installed; nine more need an emulator, a Docker runtime or helm, and each is in its own CI job so the boundary of the test rig is drawn rather than blurred.
The measurements.
- Terraform 1.16.0 against OpenTofu 1.12.6: 13 differing leaves in the JSON plan, 0 of them about what would be created. Exemptions are anchored to the path each was measured at.
- Both binaries apply twice and the second run must find nothing to do.
- 3 of the 4 policy documents the modules write are linted offline. The fourth interpolates ARNs that do not exist at plan time, so it has no body to lint, and it is named rather than skipped.
- Five state lock facts re-derived against a real PostgreSQL in CI, transcripts recording their own commands.
- A message on an SQS queue reaches an SNS topic through a Lambda the emulator invoked itself.
What it does not show. Four things, on the first screenful and generated from src/quiltz/boundary.py rather than typed beside it: IAM conditions are ignored by the emulator, bucket configuration propagation is not reproduced, requests cost nothing, and quotas are infinite.
Five claims moved when they were finally measured, and the README lists them rather than quietly correcting them.