Appendices

Pages that document a real service rather than a mechanism: a walkthrough against a running Ollama, a run against a real PostgreSQL, anything that cannot execute in continuous integration.

Why these are not lessons

A lesson on this site has to run offline, because that is the only way its output can be verified before the page is published — a lesson whose output block says "your output will differ" is a page that is not checked. So the mechanism is taught as a lesson with a fake provider, and the real service appears here, where the assertion is command-level instead of output-level: the module imports, the factory constructs a real object, and every quoted command names a target that exists.

What an appendix must satisfy

• It is outside every learning path, so no lesson depends on a service to be readable.
• It contains no output block, because its output cannot be reproduced. It may quote real output in prose, clearly labelled as a transcript from one machine.
• Its commands are checked, and its module is exercised by an opt-in workflow that sets XYBEROS_OLLAMA_LIVE=1 — the same gate tests/ai/test_live_ollama.py already uses, so a missing service can never fail a pull request.

There are no appendices yet. The first will be the live Ollama walkthrough for the AI path, and the lesson that teaches the same mechanism without a service is the AI path overview.