Methodology

0. Why public markets

The real work here is a framework for transparent, accountable AI-driven research. We chose public markets as the testing ground because the data is public and the outcomes are measurable — every hypothesis can be scored against reality in days or weeks, not years. The same methodology generalises to any domain where inputs are observable and outputs can be scored. Markets are the stress test. Transparency is the product.

1. Data sources

Every morning at 06:00 UTC the agent pulls public data from: Yahoo Finance (prices for 236 subjects), FRED (US macro indicators), ECB statistics (EU macro), RSS feeds from major financial outlets, and Google Trends. All data is public. No proprietary feeds, no insider information.

2. What the agent produces

The agent produces research output, not investment advice. For each subject it studies, it records: an entry reference (the price at the time of observation), a thesis (why the pattern caught the agent's attention), scenario analysis (base, bull, and bear outcomes), a thesis strength score from 0 to 1, and a risk level. These are observational notes about the data — not recommendations to buy or sell.

3. Two-model review

Every research output is reviewed by a second LLM from a different provider before publication. Both models must agree the thesis is coherent and the data supports it. If the reviewer disagrees, the research entry is flagged for human review or discarded.

4. Observed outcome tracking

After publication, every research subject is tracked daily. Prices are logged. When a subject closes (price hits base/bull/bear target, thesis invalidated, or horizon expires), the observed outcome is recorded permanently. Nothing is deleted. A research set with a negative observed outcome stays visible in the public research history forever.

5. How to read "hit rate"

Hit rate = percentage of closed research sets where the observed outcome was positive. This is a descriptive statistic about past observations. It does not predict future results, it does not adjust for risk, and a positive outcome on a research set does not mean subscribers who acted on it made money (they may have entered at different prices, paid different taxes, or held different horizons). It is one data point among many you should weigh.

6. What we do not claim

• We do not claim the agent beats any benchmark.
• We do not claim past observed outcomes predict future results.
• We do not give personalized advice — the agent does not know your situation.
• We do not guarantee the data is error-free. Public data sources occasionally fail.
• We do not recommend specific asset allocations for your portfolio.

7. What subscribers are buying

Subscribers pay (during beta: nothing) to access the research output the agent produces. They are not buying investment recommendations. What they do with this research, whether to act on it, ignore it, use it as one input among many, or treat it as educational material only, is entirely their decision and responsibility. Always consult an authorized financial advisor before any investment decision.

8. Changes to methodology

If we change how the agent works — different data sources, different review model, different scoring — we will document the change here with the date. The research history will always show which methodology version produced each output.