Fathom Coursewick unifies data from multiple exchanges into a single, risk-adjusted view and applies predictive modelling to flag emerging exposure before it affects your portfolio — accessible from any time zone, on any stable connection.
Fathom Coursewick was built to address a specific, recurring problem: investors who hold positions across several exchanges, often while travelling, rarely have the time to reconcile conflicting data feeds by hand. The platform consolidates positions, order books and volatility signals into one interface, updated continuously rather than on request.
Every output is presented as a recommendation rather than an instruction. The underlying models flag correlation, drawdown risk and timing anomalies; the decision to act remains with the investor, informed rather than replaced by the system.
Two mechanisms underpin the platform: the consolidation of disparate exchange data, and the predictive layer applied once that data has been unified.
Rather than requiring separate logins and manual spreadsheets for each venue, Fathom Coursewick ingests order-book depth, historical volatility and liquidity data from every connected exchange and synthesises them into one risk-adjusted stream. Positions are shown net of exposure across venues, so correlated risk becomes visible rather than hidden across browser tabs.
The engine applies statistical and machine-learning models, including random forest classifiers and neural network regressors, to historical and live data, surfacing pricing discrepancies between exchanges as they emerge. Each signal carries a confidence measure, allowing the recommendation to be weighed against personal judgement rather than followed by default.
The process behind every recommendation follows three defined stages. Understanding them is, in our view, as important as trusting the output.
Market data, order books and account-level positions are pulled continuously from each connected exchange via read-only API access. No withdrawal or trading permissions are requested at this or any later stage.
The consolidated dataset passes through a layered model architecture, combining time-series analysis with classification techniques, to identify volatility clusters, correlation shifts and pricing anomalies across venues.
Findings are translated into ranked, time-stamped signals with an accompanying rationale. Each signal is intended to support a specific decision: whether to rebalance, hedge, or simply hold.
The platform is designed around a specific constraint: the investor is often unavailable, asleep, or between connections, and the system must continue working in that absence.
Ingestion and analysis continue whether you are mid-flight or working from a location with intermittent connectivity. Findings are queued and presented at your next login, time-stamped to your local time zone.
A single browser-based dashboard replaces the need to switch between exchange-specific applications, each with its own login and layout. Any device with a modern browser and a stable connection is sufficient.
Signals are generated from historical pattern recognition rather than sentiment, so they are not shaped by the fatigue or urgency that can accompany decisions made across time zones. This does not remove risk; it reduces one specific source of it.
We would rather describe how the system operates than ask for confidence in advance.
Signal generation draws on random forest classifiers for pattern detection and neural network regressors for volatility forecasting, cross-validated against historical exchange data before deployment.
All exchange connections use API-only access, with withdrawal permissions disabled at the key level. Data in transit and at rest is protected using 256-bit encryption, and credentials are never stored in plain text.
Market data feeds refresh on a rolling basis throughout each trading session. Ingestion latency depends on the connected exchange's own API limits rather than a fixed internal delay.
