Methodology
We don't predict the future.
We expose non-obvious links.
At the core of the system is Robert Axelrod's theory on the evolution of cooperation and the principle of cascading chains. We build a graph of events, agents and assets, and compute the probability of transition from the first trigger to tertiary effects.
24/7 source ingestion
50+ RSS feeds, satellite data, prediction markets, exchange tickers. Parsing and normalisation into events.
Relationship graph
TigerGraph stores 50 countries, 200 companies, tickers, sectors, commodities and Person/Org entities. Every PRECURSOR_TO edge carries a probability weight.
AI agents on Gemma 3
5 parallel agents (Analyzer, Skeptic, ScenarioWriter, Notifier, Calibrator) debate the chain and produce 3 scenarios: A — main, B — alternative, C — black swan.
Accuracy calibration
Every forecast window is closed by fact or timeout. Brier Score, ROC-AUC and calibration curves are published openly.
Event taxonomy
Extended classification for 2nd–3rd order cascades: from macro and central banks to logistics and regulation.
Event relationship graph
Interactive TigerGraph view — soonTigerGraph links countries, companies, commodities, and institutions. Each PRECURSOR_TO edge carries a probability weight so we can score 2nd- and 3rd-order chains, not just headlines.
Methodology and open stats are live; graph visualization ships in Q3 2026.
Open accuracy statistics
Full calibration history →Calibration is accumulating
Verified outcomes are still insufficient; the system will update automatically after forecast windows close.
Rolling accuracy from closed forecast windows; recomputed by the calibration service. Not investment advice.
Why «Axelrod»
Robert Axelrod showed that in repeated games a simple tit-for-tat strategy beats complex ones. In markets the analogue is to track recurring chains rather than individual news. If a chain has triggered before — its weight in the graph grows. If it has failed — it decreases.
What we measure and publish
- Brier Score — mean squared error between predicted probability and reality. Scale 0–1, lower is better. A coin flip yields 0.25.
- ROC-AUC — area under the ROC curve. 1.0 is perfect separation, 0.5 is random.
- Calibration — for predictions with 70% probability the realised frequency must be ~70%.
Where we are wrong
Systemically — on black swans not represented in the graph, and in regimes where correlations break (fiscal crises, new regulations, war). That's why we never give financial advice, only show scenarios and their probabilities — the decision is yours.
Editorial policy
Axelrod's Oracle Editorial policy
Public blog posts, case studies, and Crystal AI answers follow shared transparency and responsibility rules.
- We publish analysis, not investment advice. Scenario probabilities are estimates, not guarantees.
- Event sources are shown on scenario cards; AI content (blog, FAQ, Crystal) is generated and spot-checked editorially.
- Errors and stale data are corrected when found; calibration stats are published openly at /stats.
- Conflict of interest: the team does not give personal trading signals through Oracle public channels.
- Feedback: support@axelrodsoracle.ru — for factual corrections and methodology questions.
Editorial materials do not replace professional legal or investment advice.