DOE Framework

The DOE (Directive-Orchestration-Execution) framework is a 3-layer architecture designed to maximize AI reliability by separating probabilistic decision-making from deterministic execution.

🏗️ The 3-Layer Architecture

1. Layer 1: Directive (What to do)

   • Located in directives/
   • Standard Operating Procedures (SOPs) written in Markdown.
   • Defines goals, inputs, tools, and edge cases in natural language.

2. Layer 2: Orchestration (Decision Making)

   • The AI agent (e.g., Gemini, Claude) acts as the intelligent router.
   • Reads directives, calls execution tools, handles errors, and updates instructions.

3. Layer 3: Execution (Doing the work)

   • Located in execution/
   • Deterministic Python scripts that handle API calls, data processing, and interactions.
   • Ensures consistent, testable, and reliable results.

📂 Project Structure

├── directives/      # SOPs and instruction sets
├── execution/       # Deterministic Python scripts (the tools)
├── .tmp/            # Intermediate processing files (git-ignored)
├── .env             # Environment variables and API keys (git-ignored)
├── GEMINI.md        # AI Agent instructions (mirrored across environments)
└── README.md        # Project documentation

🛠️ Key Principles

• Check for tools first: Always look in execution/ before creating a new script.
• Self-Annealing: When something breaks, fix the script, update the tool, and then update the directive to include the new learning.
• Cloud Deliverables: Local files in .tmp/ are temporary. Final outputs should reside in cloud services (e.g., Google Sheets, Slides).

---

Built with reliability and scale in mind.