Energy Desk — local energy-news analysis with Basal and MetaTrader 5 integration

I am developing a private Energy Desk: an analytical environment for the energy market, connected to MetaTrader 5. It combines broker data, price and volume analysis, volume profiles, CVD, context across multiple timeframes, and public information about energy, macroeconomics and geopolitics. The goal is to support daily market analysis and organise the information needed for decisions.

Interpreting news is one challenge. Headlines about a supply disruption, a planned production change or a denial of a report can contain similar words while conveying very different information. This is where I use basal-1.0-1.5B locally.

The model analyses each headline in seven categories: supply disruptions, restoration of supply, sanctions, production, inventories, monetary policy and geopolitical conflicts. In each category, it distinguishes an event described as having occurred, a proposal or intention, a denial, uncertainty and insufficient information. This separates the topic from the way an event is presented.

Results are automatically archived with their source, the time the information became available, model version and analysis configuration. I also compare them with text rules, watching disagreements, classification confidence and latency. Basal’s answer distributions form 29 input features for an experimental “Decision” module that combines the news context with market data.

The Desk works with a signal EA in MT5 that presents context and scenarios directly on the chart. The trading decision remains with the user. Basal’s probabilities describe text classification; their relationship with later price movements needs separate evaluation.

The integration runs locally on CPU. In my review of the latest configuration, I collected 102 unique headlines, or 714 category answers. Median processing time for one headline with all seven questions was around 81 seconds. This is a useful reference for further optimisation and evaluating the setup on an ordinary laptop.

The project has a working integration and is undergoing validation. Next, I will compare models using Basal’s features with versions without those features, on later data. I am particularly interested in combining local text understanding with measurable quantitative analysis, while preserving sources, chronology and the ability to verify what the model actually contributes.