Upload RESEARCH.md with huggingface_hub

RESEARCH.md

@@ -1,45 +1,45 @@
-# 🧬 ImmunoOrg: Intelligence & Architecture Research
-
-## Overview
-ImmunoOrg is a high-fidelity simulation of an autonomous, self-healing enterprise. It demonstrates "Winning-Tier" AI research patterns by integrating multi-agent coordination, retrieval-augmented generation (RAG), and explainable AI (XAI) within a competitive co-evolutionary framework.
-
-## 1. Adversarial Co-Evolution (Self-Play)
-The core of ImmunoOrg's intelligence is its **Co-Evolutionary Cycle**. Unlike static security simulations, ImmunoOrg employs a dynamic feedback loop between the Defender and the Adversary.
-
-### The Cycle:
-1. **Defender Adaptation**: The agent learns to identify and contain threats, improving its "Belief Map" and organizational efficiency.
-2. **Improvement Metric**: The `SelfImprovementEngine` tracks the rate of reward increase and time-to-containment across generations.
-3. **Adversary Evolution**: The `AttackEngine` dynamically increases adversary complexity (stealth, lateral movement capability, and vector variety) based on the defender's improvement rate.
-4. **New Equilibrium**: This creates a "Red Queen" effect, where the agent must continuously innovate its defense strategies to keep pace with an evolving threat.
-
-## 2. RAG-Powered CVE Intelligence
-To bridge the gap between simulated environments and real-world security, ImmunoOrg implements a **Retrieval-Augmented Generation (RAG)** system.
-
-- **Knowledge Base**: The `CVEKnowledgeBase` simulates a vector database of real-world CVEs (e.g., SQL Injection, Rootkits, Supply Chain attacks).
-- **Semantic Retrieval**: When a threat is detected, the agent retrieves technical details and recommended mitigations from the knowledge base.
-- **Observation Injection**: This intelligence is injected directly into the agent's observation stream, allowing the model to reason about *why* a specific mitigation (e.g., "parameterized queries") is the correct choice for a specific CVE.
-
-## 3. XAI: The Reasoning Heatmap
-To solve the "black box" problem of LLM agents, ImmunoOrg implements a **Reasoning Trace** architecture for full interpretability.
-
-### Trace Components:
-- **Decision Trigger**: The exact observation or alert that sparked the action (e.g., "RAG alert for CVE-2023-1234").
-- **Observation Snippet**: The raw data evidence used for the decision.
-- **Rationale**: The agent's internal chain-of-thought justification.
-
-### The Heatmap:
-These traces are aggregated into a **Reasoning Heatmap**, allowing human judges to audit the agent's logic in real-time. This transforms a simple action log into a transparent map of the agent's cognitive process.
-
-## 4. HITL: The Judge's Console
-The **Judge's Console** introduces Human-in-the-Loop (HITL) dynamics, simulating a corporate board of directors.
-
-- **Directive Injection**: Judges can inject high-level board directives (e.g., *"Prioritize Uptime over Isolation"*) mid-simulation.
-- **Constraint Adaptation**: The agent must dynamically shift its reward function and action priorities to align with these new constraints, demonstrating high-level cognitive flexibility and alignment.
-
-## Summary of Intelligence Tiers
-| Feature | Research Pattern | Value Proposition |
-| :--- | :--- | :--- |
-| **Co-Evolution** | Competitive Self-Play | Continuous improvement of agent robustness |
-| **RAG-CVE** | External Knowledge Retrieval | Grounding simulation in real-world security |
-| **Reasoning Heatmap** | Explainable AI (XAI) | Full transparency of decision-making |
-| **Judge's Console** | Human-in-the-Loop (HITL) | Dynamic alignment with executive goals |
+# 🧬 ImmunoOrg: Intelligence & Architecture Research
+
+## Overview
+ImmunoOrg is a high-fidelity simulation of an autonomous, self-healing enterprise. It demonstrates "Winning-Tier" AI research patterns by integrating multi-agent coordination, retrieval-augmented generation (RAG), and explainable AI (XAI) within a competitive co-evolutionary framework.
+
+## 1. Adversarial Co-Evolution (Self-Play)
+The core of ImmunoOrg's intelligence is its **Co-Evolutionary Cycle**. Unlike static security simulations, ImmunoOrg employs a dynamic feedback loop between the Defender and the Adversary.
+
+### The Cycle:
+1. **Defender Adaptation**: The agent learns to identify and contain threats, improving its "Belief Map" and organizational efficiency.
+2. **Improvement Metric**: The `SelfImprovementEngine` tracks the rate of reward increase and time-to-containment across generations.
+3. **Adversary Evolution**: The `AttackEngine` dynamically increases adversary complexity (stealth, lateral movement capability, and vector variety) based on the defender's improvement rate.
+4. **New Equilibrium**: This creates a "Red Queen" effect, where the agent must continuously innovate its defense strategies to keep pace with an evolving threat.
+
+## 2. RAG-Powered CVE Intelligence
+To bridge the gap between simulated environments and real-world security, ImmunoOrg implements a **Retrieval-Augmented Generation (RAG)** system.
+
+- **Knowledge Base**: The `CVEKnowledgeBase` simulates a vector database of real-world CVEs (e.g., SQL Injection, Rootkits, Supply Chain attacks).
+- **Semantic Retrieval**: When a threat is detected, the agent retrieves technical details and recommended mitigations from the knowledge base.
+- **Observation Injection**: This intelligence is injected directly into the agent's observation stream, allowing the model to reason about *why* a specific mitigation (e.g., "parameterized queries") is the correct choice for a specific CVE.
+
+## 3. XAI: The Reasoning Heatmap
+To solve the "black box" problem of LLM agents, ImmunoOrg implements a **Reasoning Trace** architecture for full interpretability.
+
+### Trace Components:
+- **Decision Trigger**: The exact observation or alert that sparked the action (e.g., "RAG alert for CVE-2023-1234").
+- **Observation Snippet**: The raw data evidence used for the decision.
+- **Rationale**: The agent's internal chain-of-thought justification.
+
+### The Heatmap:
+These traces are aggregated into a **Reasoning Heatmap**, allowing human judges to audit the agent's logic in real-time. This transforms a simple action log into a transparent map of the agent's cognitive process.
+
+## 4. HITL: The Judge's Console
+The **Judge's Console** introduces Human-in-the-Loop (HITL) dynamics, simulating a corporate board of directors.
+
+- **Directive Injection**: Judges can inject high-level board directives (e.g., *"Prioritize Uptime over Isolation"*) mid-simulation.
+- **Constraint Adaptation**: The agent must dynamically shift its reward function and action priorities to align with these new constraints, demonstrating high-level cognitive flexibility and alignment.
+
+## Summary of Intelligence Tiers
+| Feature | Research Pattern | Value Proposition |
+| :--- | :--- | :--- |
+| **Co-Evolution** | Competitive Self-Play | Continuous improvement of agent robustness |
+| **RAG-CVE** | External Knowledge Retrieval | Grounding simulation in real-world security |
+| **Reasoning Heatmap** | Explainable AI (XAI) | Full transparency of decision-making |
+| **Judge's Console** | Human-in-the-Loop (HITL) | Dynamic alignment with executive goals |