46.1 Pre-build checklist (spec, constraints, acceptance tests)

Pre-Build Checklist

## Pre-Build Checklist

### Problem Definition
- [ ] What problem are we solving?
- [ ] Why use AI instead of rules/code?
- [ ] What happens if AI gives wrong answer?
- [ ] Who is the user? What do they expect?

### Scope & Constraints
- [ ] What CAN this AI do?
- [ ] What MUST it NOT do? (explicit exclusions)
- [ ] What data is available as input?
- [ ] What format should output be?

### Acceptance Criteria
- [ ] Define "good enough" numerically (e.g., >90% accuracy)
- [ ] What latency is acceptable? (e.g., <2s)
- [ ] What cost per request is acceptable?
- [ ] Who approves before launch?

### Test Cases (minimum 20)
- [ ] 5+ happy path examples
- [ ] 5+ edge cases
- [ ] 5+ adversarial inputs
- [ ] 5+ failure cases (what should trigger "I don't know")

### Dependencies
- [ ] Which model? Why?
- [ ] What context/data is needed?
- [ ] Any external APIs?
- [ ] Fallback if AI fails?

Where to go next

• 46.2 Prompt checklist