Learn by property
The method grades a project on seven properties, one letter each. Study them one at a time: for each, read what it means, see the checks that enforce it, and practise it on a project.
Five of the seven are the ones a non-engineer can care about, because they decide whether a system survives scrutiny. They spell SAVED: Self-describing, Auditable, Verifiable, Executable, Defended. The other two, Bounded and Composable, keep the five true under change. The rubric holds the full definitions.
| Property | The question it asks | Read the definition | See the checks | Practise it |
|---|---|---|---|---|
| Self-describing | Can a stranger, human or AI, understand the system from its own artifacts? | Rubric | Gates | Spec completeness · Your first hour · Templates |
| Auditable | Are the decisions on the record: what, why and what was ruled out? | Rubric | Gates | Repository discipline · Existing project |
| Verifiable | Is correctness computed, not claimed? | Rubric | Gates | New project |
| Executable | Is the spec bound to the running system, not to a description of an older one? | Rubric | Gates | Coherence between spec and code |
| Defended | Are the rules enforced, not suggested? | Rubric | Gates · Gate library | Run structural gates and remediate |
| Bounded | Does every unit of work have explicit scope and seams? | Rubric | Gates | Structural disciplines |
| Composable | Can a change stay local, without unexpected coupling? | Rubric | Gates | Structural disciplines · Migrate to a new stack |
Related ideas, each in one place
- The sentinel, the small map that bounds what the assistant loads: taught in the free course and set up in Your first hour.
- The ratchet, so every miss leaves something permanent: Quality gates and Refine the spec, triage first.
- Enforcement versus suggestion: The Andon Method, a proposed team method built on the same thesis.
- Grades and honesty: the rubric says what its letters do not claim.