Protofin Macro Intelligence PROTOTYPE

AUTONOMOUS BLACK-LITTERMAN ENGINE

Your key is used only from your browser, straight to the provider you name. It is never sent to this site. Works with OpenAI, Groq, OpenRouter, Together, or a local Ollama (http://localhost:11434/v1, with OLLAMA_ORIGINS set). URL options: ?provider=byok&model=gpt-4o-mini&baseUrl=..., key via #key=... (fragment, not sent to any server).

Pipeline Flow archify render · each step lights as its values arrive

full archify page
  1. Run the pipeline to see each step's values here.

1. Data Ingestion

2. Quant SVAR Shock

3. Agent Debate (Chain of Thought)
i
Source: the model provider you choose above (temperature 0).
Calculation: Multi-agent prompt engineering forces an adversarial simulation. The Economist argues growth, the Risk Manager argues downside, and the CIO forces a quantitative synthesis.

4. The "Views" (Q Vector) & Grounding

5. Black-Litterman Portfolio
i
Source: Python NumPy Math Engine.
Calculation: Historical covariance (Σ) from 1y daily returns. 60/40 equilibrium prior (Π). BL posterior:
E[R] = [(τΣ)⁻¹ + PᵀΩ⁻¹P]⁻¹ [(τΣ)⁻¹Π + PᵀΩ⁻¹Q]Then a long-only, fully invested mean-variance optimum inside a 40–80% SPY tilt band, a 2pp no-trade band against current holdings, and a 10pp per-run move cap. Unchanged evidence returns the stored decision.