Guides
Cross-cutting, turn-by-turn guides for teams running the PARR pipeline. Each one starts from a user story, gives the exact commands, and states what you should see at each checkpoint so you can tell a working step from a step that merely did not error.
Per-service reference lives with each product: PFactory, AIFactory, TFactory, CFactory.
Start here
| Guide | You want to |
|---|---|
| Onboarding | Get from nothing to a first verified change |
| Working as a team | Set up roles, approval gates and shared conventions |
| Boards and work items | Drive work from GitHub, GitLab or Azure DevOps issues |
Scenario guides
| Guide | Scenario |
|---|---|
| Cloud development | Build services that target cloud infrastructure |
| Legacy rewrites | Change code nobody remembers writing |
| Enterprise scale | Many repositories, many contributors, audit obligations |
| Testing scenarios | Decide what “verified” means and enforce it |
What these guides promise
They document what is shipped today, not what is designed. Where a capability is planned but not built, it is named as planned with a link to the RFC, rather than described in the present tense.
That distinction is deliberate. A guide that describes an endpoint which does not exist costs more than no guide at all: the reader configures against it, sees nothing work, and cannot tell whether they made a mistake or the feature did. Two examples from this codebase’s own history are documented in Testing scenarios because the lesson generalises.
Every command below has been run against a live fleet. Where output is quoted, it is real output.
Conventions used
Checkpoints. Each numbered step ends with what you should observe. If you see something else, stop there rather than continuing — a later step failing for an earlier reason is the hardest kind of problem to unpick.
Options tables. Every flag or environment variable is given with its options, its default, and what happens when it is unset. “Unset” is a real configuration, and usually the one you are in when something is not working.
Nothing is assumed green. Where a guide tells you to check a gate, it also tells you how to confirm the gate actually ran, because a check that is skipped and a check that passed look identical in most user interfaces.