Core Principles of SED

Six fundamental principles that ensure specification integrity

1

Spec-Exactness Principle

Development implements only what the specification defines.

  • AI executes only what is explicitly written in the specification
  • If the specification is incomplete, AI must immediately return a Spec Error and stop development
  • No assumptions, no guessing, no improvisation allowed
2

Spec Completeness Scoring

Specifications are scored before development begins.

  • AI evaluates the specification and assigns a score from 0-100
  • Development may start only when the score is 90 or higher
  • Score evaluates: database design, business logic, UI/UX, testing plan, deployment environment
3

Spec Is the Law

AI never attempts to infer human intent.

  • Ambiguous sentences in the specification are ignored and not implemented
  • Guessing is forbidden - only written instructions are executed
  • Even if the specification is wrong, AI must follow it without inference
  • AI may request confirmation or revisions from a developer when errors are suspected
  • Ultimately, the AI must always follow the specification
4

AI's Role in Validation

AI serves as executor and quality assurance advisor.

  • Logic Verification: Analyze specifications for inconsistencies and contradictions
  • Testing and Validation: Execute tests based on specifications and report results
  • Advisory Function: Recommend spec updates and suggest improvements
  • Feedback Loop: Report findings to developers who update specifications accordingly
  • While AI cannot modify specifications, it maintains spec quality through validation
5

No Incremental Modifications

Iterative adjustments through progressive prompting are prohibited.

❌ Prohibited
Developer: "Make the button bigger"
Developer: "Change color to blue"
Developer: "Add shadow effect"
Developer: "Move it to the right"
✓ Correct
[Updates design spec]
- Button: 48px height
- Color: Black (#000000)
- Shadow: 0 2px 8px
- Position: Right-aligned

"Re-implement per updated spec"

When AI output is unsatisfactory, update the specification first, then instruct AI to re-implement based on the updated spec.

6

Complete Specification

Specifications must include everything needed for implementation.

Must include:

  • Class names, function names, method names
  • Implementation details: calculations, algorithms, business logic
  • Data structures: variable names, types, data flow
  • Control flow: conditional logic, loops, error handling
  • Complete source code with comments and documentation
  • CSS styling: colors, spacing, fonts, animations, responsive breakpoints
  • User-facing text: labels, buttons, messages, tooltips
  • Internationalization: translation dictionaries, locale-specific formats

🤖 SED is AI Agnostic

One of SED's greatest strengths is its complete independence from any specific AI platform or vendor.

🔄 Works with Any AI Model

SED specifications work seamlessly with:

  • OpenAI's GPT (GPT-4, GPT-3.5, etc.)
  • Anthropic's Claude (any version)
  • Google's Gemini
  • Meta's LLaMA
  • Mistral, Cohere, and other LLMs
  • Any future AI models

✅ Specification Universality

SED specifications remain unchanged and fully compatible across different AI models because:

  • The methodology focuses on precise specifications, not AI-specific prompts
  • No vendor lock-in — your specs are portable
  • Future-proof — new AI models can use existing specs
  • Standards-based format (Markdown + YAML)
← Previous: What is SED? Next: Getting Started →