Frequently asked questions
What does AI do — and what does it not prove?
Clear answers about AI, data, independent testing and the limits of financial prediction.
What makes Feronix an AI-driven research platform?
Feronix uses language models to turn financial text into auditable features, and machine learning to investigate relationships with price reactions. Statistical baselines and independent tests then assess whether a model genuinely adds value. AI assistants also help develop the website and research software.
Is everything at Feronix fully automated or exclusively AI-based?
No. AI is an important tool, but source verification, research decisions and assessment remain human responsibilities. We also publish statistical studies that do not use a language model. An AI analysis is not automatically a reliable prediction.
Does the homepage demo actually use Qwen?
No. The demo uses simple keyword rules in your browser to illustrate the steps. Pasted text is not processed by Qwen and does not produce a tested price prediction. The Analysis Example page does contain a saved Qwen3-4B LLM analysis of an actual earnings release.
What do you mean by vibecoding?
Developing with AI assistants: describing ideas in everyday language, generating proposed code, then reviewing, testing and improving it. This helps Feronix build faster, but it is not a method for validating research findings. Responsibility for the code, sources and conclusions remains with the founder.
Does Qwen3-4B LLM predict share prices directly?
No. The model only reads financial text and converts explicit information into predefined features. A separate statistical model later investigates whether those features are related to future adjusted returns.
Why use Qwen3-4B LLM?
The model fits in the available local memory and can produce structured financial features and evidence excerpts without sending documents to an external model service.
Can the model see future price information?
No. The model prompt contains selected passages from the official earnings release and its publication timestamp. Post-publication returns are linked only during evaluation.
What does out-of-sample mean?
It means the model is evaluated on a later time period that was not used to choose its settings or thresholds.
Why are there two datasets?
The first dataset is used for development and a chronological final test. The larger replication set checks whether the result holds for new companies and a broader time period.
What happens if Qwen stalls on a release?
The same release is retried. Truncated JSON is given a larger output allowance; after a temporary process failure, the supervisor resumes after thirty seconds from the next release that has not yet been saved.
Why are interim returns not shown yet?
Inspecting interim results can influence decisions and make the final test less independent. Progress and extraction quality are visible, but conclusions about predictive performance will follow only after completion.
Is this investment advice?
No. Feronix publishes research methods and historical analyses. The website does not recommend buying, selling or holding securities.
Can I download the research design?
Yes. The documents page provides the methodology, data description, variable register and validation protocol.