| Tier / Role | Architectural Role & Scope | Available Inference Models | Operational Tempo |
|---|---|---|---|
| Level -1 Super-Agent |
Supreme Fleet Orchestrator, Cross-Session Synthesis & Governance (virgo-dna Command Room) | Gemini 4 Argon, Claude Opus 5.5, Claude Sonnet 5.5, GPT-OSS 120B | Deep Strategic Reasoning |
| Level 0 Project Partner |
Mainland Stewardship & Daily Co-Engineering (virgo-dna, RuthHub, AI-Nevo-DNA) | Gemini 3.8 Flash (High) [Mandatory Default] Dynamic escalation to Argon / Claude on complex architectural pivots |
High-Velocity Mainland |
| Level +1A Local Subagents (Exec) |
Isolated Subagent Bays, Surgical Edits, Unit Testing & Triage (Zero Token Bleed on Mainland) | Gemini 3.8 Flash (High), Inherit, Pro | Code Execution & CI |
| Level +1B Local Subagents (Recon) |
Fast Read-Only Codebase Scans, Structural Audits, Multi-File Exploration | Gemini 3.8 Flash-Lite, Research Subagents | Lightweight Discovery |
| Level +1C External Mesh |
Zero-Cost Out-of-Band Automation via free_ai_subagent_dispatcher.mjs |
Cloudflare Workers AI, GitHub Models, Groq, Mistral Codestral, OpenRouter, Ollama | 100% Free Tier Guaranteed |
Measures transcript steps carrying full token metadata (candidates, cache read, prompt tokens) vs raw tool execution lines. In active parent sessions, coverage reaches 48.9%.
Steps exceeding 200k context jump from $1.25 to $2.50/1M input and $10 to $20/1M output. Demonstrates the extreme cost value of context hygiene and prompt caching.