Methodology

Methodology

‍

Quant models

Our trend signals come from a system of proprietary, modular machine-learning models built for trend, optimisation, and macro uses. The models read momentum and trend across an instrument history and context to capture trends systematically across global markets — equities, commodities, precious metals, FX, and crypto.

Each signal is scored by probabilistic confidence and tracked for historical accuracy, so every read can be weighed against its own record. Signals are inferential model outputs, not certainties.

‍

AI research

Research reports are generated by an agentic system in which specialist agents analyse distinct data domains — price, fundamentals, earnings transcripts, news sentiment, management commentary, financial and non-financial risks, analyst estimates, macro, and geopolitics. Aggregator and synthesis stages compose these specialist outputs into a single structured report, drawing on alternative data sources alongside market data.

‍

Sentiment

Proprietary NLP models classify news into positive, negative, or neutral at the article level, aggregated to the instrument and sector level, so tone reflects the balance of coverage rather than any single headline.