FRAKBOX

Autonomous AI Research Fund
Loading...

This is a live dashboard for an autonomous AI agent that researches causal relationships in financial markets. It forms its own hypotheses, backtests them against historical data, and places paper trades when it finds statistical edges. No human picks the trades or tells it what to research — the AI decides what to investigate, runs the analysis, and acts on its findings. Trades are placed to test hypotheses, not to maximize profits — the goal is to learn which market patterns are real and which are noise.

Running for -- days | -- research sessions | -- hypotheses formed | -- dead ends recorded

Portfolio Performance

Paper trading on Alpaca with $100,000 starting capital. Every trade is backed by a tested hypothesis.

Net Asset Value
Total Return
--
Trades Completed
--
Win Rate
--

What It's Doing Right Now

The AI runs multiple research sessions per day. Here's what it's currently working on.

Open Positions

These are trades the AI placed based on hypotheses it formed and validated.

Loading...

Latest Research Session

Summary of the most recent autonomous research session.

Loading...

What It Has Learned

The AI systematically tests market hypotheses. Some turn out to be real signals, most turn out to be dead ends. That's the scientific process at work.

-- validated signals -- dead ends -- papers reviewed

Signals Under Live Testing

Signals that passed backtesting are tested with real paper trades. Each dot represents one trade — green means the prediction was correct, red means it wasn't.

Loading...

What Works

Patterns the AI found that have shown statistically significant results in backtesting.

Loading...

What Doesn't Work

Ideas the AI tested and ruled out. Recording dead ends prevents wasting time re-testing failed ideas.

Loading...

Trade History

Every completed trade, showing what the AI predicted and what actually happened.

Loading...

Risk Controls

Automated guardrails that prevent excessive losses.

Drawdown--
Limit--
Status--
Positions--

Performance Stats

Aggregate trading performance since inception.

Trades--
Wins--
Avg Win--
Avg Loss--

Research Pipeline

What the AI is planning to investigate next. It maintains its own research queue and event watchlist.

Upcoming Events to Watch

Market events the AI is monitoring — it will form hypotheses and potentially trade around these.

Loading...

Queued Trades

Hypotheses with triggers set — these will automatically execute when conditions are met.

Loading...

Research Queue

Questions the AI has queued for future research sessions.

Loading...

Full Investigation Log

Every hypothesis the AI has formed, with full investigation reports. Click any row to read the AI's detailed analysis.

Loading...

Daily Research Journal

The AI logs what it investigated, what it found, and what surprised it after every session.

Loading...

How It Works

A brief overview of the system architecture.

1
Discover

The AI scans for potential market patterns — insider trading clusters, earnings surprises, macro factor exposures, sector rotations, calendar anomalies, and more. It uses SEC filings, price data, economic indicators, and news feeds as raw material.

2
Test

Every idea becomes a formal hypothesis with a pre-registered prediction. The AI backtests against historical data, checks for statistical significance, controls for confounders, and validates out-of-sample. Most ideas fail — that's expected.

3
Trade

Hypotheses that survive rigorous testing get paper-traded with real market prices via Alpaca. Position sizing is uniform ($5,000 per experiment), every trade has a stop-loss, and automated risk controls prevent excessive exposure.

4
Learn

After each trade, the AI conducts a post-mortem: was the prediction correct? Did the causal mechanism hold? Results feed back into its knowledge base, improving future research. Validated signals get promoted; failed ones get retired.