Dogfood the course method

Courses can teach good engineering while being produced by a weak process. This course evaluates its own production system.

Five quality dimensions

1. Executable examples: every claimed output comes from runnable code or is marked untested.
2. Journal evidence: hypothesis, baseline, result, verdict and artifact are distinct fields.
3. Locale parity: EN, FR and ES share the same files and links; automation checks structure, humans review meaning.
4. Adoption feedback: the journal follows candidates beyond publication into rejection, incubation or repository integration.
5. Agentic efficiency: measure cost per accepted result, not agent activity or tokens alone.

A better publication gate

A course page is ready when:

• official sources support time-sensitive facts;
• commands and outputs were reproduced on the stated environment;
• untested statements are explicit;
• the journal links each measured claim to code or raw evidence;
• translations mirror the source structure;
• repository recommendations remain hypotheses until tested there.

Automation can prove file parity, links, generated-matrix freshness and test results. It cannot prove that a translation is idiomatic or an architectural recommendation is wise; those remain review responsibilities.

Exercise: deliberately stale the matrix

Change one score in opportunities.json without regenerating matrices.md, then run python dogfood.py check. The command must fail. Restore the file or run python dogfood.py write and re-run the tests.