Opportunity matrices

One giant ranking conceals too much. This lab keeps five views because they answer different questions. The current values are published in the generated matrices.md; python dogfood.py check fails if that artifact drifts from the JSON registry.

1. Course × repository coverage

This view finds both transfer opportunities and blind spots. A technique may fit several repositories, but that does not mean every repository should adopt it.

2. Opportunity score

The score adds pain, fit, expected value, evidence and reversibility, then subtracts cost and risk. Every dimension is 0–5. It orders investigation only.

Two safeguards are essential:

• the score cannot change status or authority;
• a high score with weak evidence remains discovery, not incubation.

3. Promotion state

The state machine is deliberately explicit:

discovered → experimenting → incubating → integrating → adopted
       └──────────────→ rejected                         → retired

Promoted states require evidence artifacts. adopted additionally requires a confirmed verdict. Rejected candidates remain in the registry so another agent does not repeat the same failed idea.

4. Results and evidence

This view answers: what was actually measured, where is the artifact, and when should the conclusion be revisited? A running agent, comment, label or plausible narrative is not evidence.

5. Course-method quality

The method itself is dogfood. The matrix tracks executable examples, journal evidence, locale parity, adoption feedback and agentic efficiency.

Exercise: add a candidate without overstating it

Add an opportunity with a real observed pain, a simpler alternative and a falsifier. Keep it discovered until a baseline exists. Run python dogfood.py write and python -m unittest -v.