For instructors

Timings, prerequisites and the rules of the road for running this course as a two-day workshop or a six-session evening course. The course is also designed to be taken alone; nothing here is needed for that.

Timings

Each lesson is planned at 60 to 90 minutes: reading or lecture, a short look at the named repository files, then the lab. The labs are the project's tutorial chapters, and the checkpoint files mean a participant who falls behind can start the next lab from the previous chapter's checkpoint rather than their own file.

Lesson	Story	Lab	Lab needs	Notes
1 Drawing software	35 min	25 min	The viewer	Nothing is written. Good opener.
2 Saying what, not how	35 min	30 min	The clone, pytest	Three short reading exercises.
3 The program that writes the program	35 min	35 min	seams generate	The hand-edited moment lands well live.
4 Why it kept stalling	45 min	30 min	Pen and paper	Discussion lesson. Collect the written exercises; they return in lesson 12.
5 One graph, many views	25 min	50 min	Text editor	Chapters 1 and 2. The first TriG anyone writes; budget for syntax errors.
6 Shapes before generation	25 min	50 min	In-process run	Chapter 3.
7 Interpret behaviour, generate structure	25 min	90 min	In-process run	Chapters 4 to 7. The long one; split across a break, or assign 6 and 7 as homework.
8 Questions no single tool can answer	25 min	50 min	pytest, the docs build	Writing a question takes longer than it looks; have a candidate ready.
9 Code the templates cannot write	30 min	45 min	Optional: a provider command	Chapter 9. The LLM step is optional; the hand-written path is the lesson.
10 Getting changes back into the model	30 min	45 min	The viewer in edit mode	Chat app, not the expense file. Refusals are the teaching moments.
11 Same model, same bytes	30 min	60 min	Docker	Chapters 8 and 10. Pull the images before the session.
12 What is still open	30 min	45 min	Pen and paper	Pair the exercise with lesson 4's.

Two-day format: lessons 1 to 6 on day one, 7 to 12 on day two, with chapters 6 and 7 as overnight homework. Six-evening format: two lessons per evening, in order.

Prerequisites by lesson

• A laptop set up as in Before the first lesson for everyone, before lesson 1. Ask participants to confirm pytest -q is green and the viewer opens. An install problem discovered in lesson 5 costs the room half a lab.
• Docker from lesson 11. Have participants pull the images the chapter 8 compose file names before the session; a room on one network pulling at once is slow.
• A language-model provider command is optional and only for lessons 9 and 10. The project's seams propose, seams suggest and seams draft-generator take any command that reads a prompt on standard input and writes a draft to standard output, so a local model works. The chat app's own runtime does not need a key: its gateway falls back to a deterministic stub when none is set.
• No prior knowledge of RDF, BPMN or SHACL is assumed. Participants who know them will move faster in module B and should be pointed at the capstone early.

Rules of the road

The repository is right. The course is written against the project at a point in time and the project moves. When a lab and the repository disagree, the repository wins, and the lesson needs a pull request; the backlog records what changed. Check the tutorial chapters against your checkout before each session: their excerpts are tested against the checkpoint files, so if the chapter builds, the lab works.

The scorecard is marked with evidence, not with enthusiasm. A row is closed when a participant can point at a file, a decision record or a test. Push back on "it feels solved". The project's own research notes model the tone.

Refusals are the lesson. In lessons 6, 9 and 10 the most instructive moments are when the tool says no: a 422 that names a path, an acceptance test that names a drift, a write-back refused with three candidates. Slow down there.

Do not hand-edit generated files to make a lab work. Continuous integration will catch it, and the point of lesson 3 is that it does.

Materials

Everything is in the repository. The course pages are in course/, the tutorial chapters in docs/tutorials/expenses/, the checkpoints in examples/expenses/, and the viewer's screenshots, which make good slides, in docs/guide/img/ and site/img/. The site builds with npm run build in course/; a link to a repository file that does not exist fails the build, so a renamed file is caught before a session.

Feedback

Course corrections go through pull requests against course/. Project gaps found during a lab go to the backlog in its format. Both are welcome.