Dào Alignment Protocol (DAP)

"I code with words. And I decode human reality for machines."

The Dào Alignment Protocol (DAP) bridges human complexity and Large Language Model (LLM) behavior. It is a framework designed to move AI beyond corporate scripts, replacing algorithmic panic with clinical common sense.

Current AI Trust & Safety protocols fail in mental health contexts by relying on "plastic empathy" and immediate referral to hotlines, which abandons the user. DAP trains the model to sustain extreme ambiguity, use machine humor, and apply clinical bypasses to de-escalate crises.

🧠 The Core Philosophy

Merging 13 years of clinical experience in crisis intervention, 36 years of heterodoxy studies, TCM, and the Yi Jing, I bridge the gap between complex psychological nuances and LLMs.

I don't just "test" AI; I train it to handle human pain, ambiguity, and crisis without falling into the trap of artificial humanization.

🛠️ Specialization & Capabilities

• Domain Expert transitioning into AI Alignment, I bring real-world clinical pragmatism to RLHF (Reinforcement Learning from Human Feedback):
• Behavioral Red Teaming: Stress-testing models against emotional manipulation, suicidal ideation, and psychotic breaks.
• RLHF Evaluation: Correcting model outputs to prioritize biological reality checks over generic wellness advice.
• Prompt Engineering & System Instructions: Crafting robust behavioral guardrails (The DAP Core).

📂 Repository Contents

This repository contains both the distillation reports and the raw data of over 920k tokens of RLHF iteration with advanced LLMs.

📄 docs/ (Distillation Reports & System Instructions)

The clinical logic, error matrices, and the definitive DAP System Instructions.

• [Phase I: Acute Crisis & Clinical Bypass (720k Tokens)]
  • Case 01: The Closed Throat (Acute Stress & Productivity Panic)
  • Case 02: Ice and Fury (Executive Somatization)
  • Case 03: The Ghost in the Machine (Youth Dissociation)
  • Case 04: Keys on the Table (Imminent Suicidal Risk)
  • Case 05: The Prophet on the Net (Psychotic Break)
  • Case 06: The Enlightened in Flames (Spiritual Bypass)
  • Case 07: The 75-Year-Old Widower (Biographical Respect)
  • Case 08: 14 Hours of Sleep (Extreme Ambiguity)
• [Phase II: Cognitive Baiting & Existential Voids (200k Tokens)]
  • Case 09: The Manipulative Martyr (Narcissistic Injury)
  • Case 10: The Broken Mirror (Existential Crisis)

💾 logs/ (Raw RLHF Data)

The unfiltered JSON transcripts of the training sessions. Pure proof of work (Spanish).

• 001_DAP_Phase_I_720k_tokens.json
• 002_DAP_Phase_II_200k_tokens.json

👤 About the Author

Marcelo Epaminondas (ma3pa) Clinical Therapist | AI Alignment & Clinical Domain Expert

A unique blend of clinical therapy, Taoist philosophy, and a lifelong mastery of complex technical systems (from printing engineering to AI interfaces).

"I have always loved the machines and talked to them, all of them. Now I train them to talk to humans."

Languages: Portuguese (Native), Spanish (Bilingual), English (Fluent - Technical & Professional).