A comprehensive assessment of the MOEX stock forecasting project has been conducted. The system is built on FastAPI with an asynchronous architecture, caching, and support for 50 issuers. It is based on an ensemble of models (Random Forest + ElasticNet), supplemented with dynamic weights and market filters (ABI/ADR/Hughes), which is rare in private projects.

The forecast accuracy is recorded at MAPE -4.8%, which is an acceptable result but requires further calibration to -3%. The architecture and philosophy of the project were rated 10/10 — the "fish in the flow" and "sideline" approach were recognized as unique and architecturally sound.

The overall project score is 7.5/10. The system is considered "smarter than 90% of what is found among private traders." The immediate plans include implementing dynamic weights, improving accuracy to -3% MAPE, expanding backtesting to 3–5 years, and automating reporting for clients.

Editorial Comment

While others measure in pips, we measure in systems. 7.5/10 is not just a number — it's a statement: the project is smarter than most private traders. But, as always, the devil is in the details: -4.8% MAPE is not ideal, but a reason to work. However, we never promised magic. We promised a system, and it exists. Now we just need to finish it. The editorial opinion may not coincide with the position of the authors and other experts.