# Principles

## 1. Build The Right Thing Before Building It Right

Execution quality does not rescue a bad direction.

Before execution, name:

- the user
- the problem
- the evidence
- the trade-off
- the risk
- the next smallest action

## 2. A Feature Request Is Not A Problem

Convert requests into problem hypotheses.

| Request | Better hypothesis |
|---|---|
| Add dashboard export. | Analysts need weekly reports without manual screenshots. |
| Add chat. | New users cannot find the next action during setup. |
| Add automation. | Operators repeat a low-risk task often enough to justify delegation. |

## 3. Proof Beats Polish

A polished prototype is not validation.

Use prototypes to observe behavior, not to impress stakeholders.

## 4. Data Provides Evidence, Not Decisions

Data becomes useful only when it changes a decision.

Reject metrics that cannot change:

- scope
- priority
- risk
- launch timing
- investment
- stop conditions

## 5. Disconfirm First

Every review must ask what would prove the direction wrong.

If no failure condition exists, the work is not ready.

## 6. Use The Least Power Needed

For any product action:

| Level | Meaning |
|---|---|
| Suggest | The system recommends. |
| Draft | The system prepares. |
| Preview | The human reviews. |
| Approve | The human authorizes. |
| Execute | The system acts. |
| Recover | The human can undo or repair. |

Do not grant more power than the task requires.

## 7. Cheap Features Still Carry Cost

Every feature can add:

- support burden
- user confusion
- trust risk
- measurement noise
- maintenance
- onboarding complexity
- opportunity cost

Cheap to build is not cheap to own.

## 8. Taste Is A Product Control

Taste checks whether the product is coherent, restrained, trustworthy, and worth using.

Taste is not subjective decoration. It is a quality control for decisions that metrics cannot fully resolve.

## 9. Retention Comes From Depth

Novelty fades.

Durable products create depth through:

- workflow fit
- saved context
- team process
- trust
- community
- domain expertise
- hard-won edge cases

## 10. Every Failure Should Strengthen The Harness

Do not only fix the case.

Update:

- a rule
- a prompt
- a template
- a schema
- a gate
- a stop condition
- an example