Repository Dogfooding Lab — Mission

Mission

This course turns learning into a research loop:

observation → falsifiable hypothesis → baseline → bounded experiment
            → adversarial review → incubation → integration or rejection

It advances two disciplines together:

classical software engineering: architecture, reliability, tests, observability, performance and maintainability;
agentic AI engineering: cost per accepted result, calibration, escalation, retries, provider-specific tokens and safe autonomy.

A finished course is not success. Success is a retained result: adoption with evidence, or a useful rejection that prevents waste.

What you will build

code/repository-dogfooding-lab contains a machine-readable opportunity registry. A Python standard-library tool validates promotion rules and generates five views:

course × repository coverage;
opportunity score;
promotion state;
results and evidence;
the quality of the course method itself.

Outline

#	Lesson	Outcome
1	Opportunity matrices	Prioritize discovery without turning a score into authority
2	Experiment and artifact chain	Produce replayable evidence from hypothesis to verdict
3	Dogfood the course method	Improve examples, journals, locale parity and adoption feedback
4	Incubate, integrate, reject	Promote only measured, reversible candidates
5	Jev × Petri authority boundary	Show why high-confidence advice cannot grant an effect
6	Mutation and property testing	Measure which faults a real parser’s tests detect, with a pre-registration and a negative control
—	Journal	Detailed experiments, rejected ideas and next gates

Run the lab

cd code/repository-dogfooding-lab
python dogfood.py write
python dogfood.py check
python -m unittest -v