Risk Driven Development & Agent Driven Evaluation
Two co-equal disciplines. One decides what to prove first; the other decides what correct means.
Risk Driven Development
Section titled “Risk Driven Development”Taste every candy before making the recipe. The riskiest question — do these candies actually melt together the way I think they do — gets proven on a real, disposable candybox first. Reality is the only ground truth.
— tenet 5
Documentation and code are hypotheses. When being wrong would invalidate the plan, the assumption gets proven on a live disposable bed before the work is designed around it. The archetypal unknown is composition at latest versions: does this actually build, together, today?
The instrument is a spike — small, time-boxed, thrown away once it has answered the question. Only the knowledge is kept, never the batch. A spike finds out how; it never decides whether, and it never shrinks the work.
When a bed contradicts a document, the document is wrong and gets fixed the same day. That rule is why this site is generated: a page projected from the source cannot disagree with it.
Agent Driven Evaluation
Section titled “Agent Driven Evaluation”Write down what “good” means, and have an agent taste it. The plan’s deterministic
check:steps verify the measurable; for the subtle “is it actually right?” anagent-check:step has an agent taste the live batch and judge.— tenet 6
Every candy ships a plan: — a runnable acceptance spec, baked into the image and executable
against a live deployment. You can read it on every candy’s page in the
candy reference.
Steps carry an explicit intent:
| Intent | Meaning |
|---|---|
run: |
changes state |
check: |
an idempotent probe — deterministic, and the mandatory minimum |
agent-run: |
an agent-performed action |
agent-check: |
an agent grades the live deployment; an unparseable or timed-out grader fails the step |
include: |
compose another entity’s plan |
The important property is that the spec is not a document about the candy — it is the test, and it runs.
Together
Section titled “Together”RDD proves the risky assumptions a behavior rests on. ADE pins down what correct behavior is and grades the running system against it. The first keeps you from building on a guess; the second keeps “it works” from being an opinion.
See also
Section titled “See also”- The check verb reference — every probe verb and the bed model.
- Candyboxing — why running these beds is cheap enough to do constantly.