How We Use AI
Human ideas, researched with AI.
The ideas, editorial direction and messaging come from Yuriy Matso. AI supports the research behind them, along with data analysis and development. That applies to this site and The Filings Report letter on Substack.
Where the direction comes from
Yuriy Matso chooses the questions to investigate, the point of view and what the publication wants to communicate. AI helps pursue that inquiry. It is not the source of the publication’s editorial agenda or the person accountable for it.
The same distinction applies to the Substack letter: the ideas and messaging are Yuriy’s; AI assists the supporting research.
What AI helps with
- Research: finding relevant records, comparing disclosures, organizing evidence and exploring alternative explanations.
- Data analysis: writing queries and calculations, investigating unusual results and helping check whether a claim is supported by the available data.
- Engineering: building extraction tools, tables, charts and site features, and helping diagnose errors and test changes.
- Presentation: helping organize and edit explanations under Yuriy’s direction, while preserving his intended message.
Where the published figures come from
The financial tables, fund positions, transaction values and price charts use saved source data and calculations implemented in code. A language model’s answer is not used as a substitute for a reported value.
For example, a derived fourth quarter follows a stated arithmetic rule, and an insider transaction value is reported shares multiplied by the reported price. Missing statement values are not filled by asking AI to estimate them.
AI can help write that code, and code can still contain mistakes. The documented rules, reconciliation checks and published datasets make the calculations open to inspection; they do not make the work infallible.
Research assistance still needs evidence
AI can misread a filing, confuse a reporting period with a filing date, overlook a footnote or supply a plausible explanation that the record does not support. A fluent answer is not evidence.
Our standard is to trace factual claims to the relevant record, distinguish interpretation from a reported fact and keep the dates and limitations attached. Readers should be able to follow a number back to its dataset or filing, regardless of which tools assisted the research.
Responsibility stays with the publication
AI is a research and production tool, not an author who can take responsibility for a claim. Yuriy remains responsible for the publication’s editorial choices and messaging.
If a figure, explanation or source link looks wrong, send the page, company or fund and relevant period to info@thetrading.tools. Source records and reproducible examples are especially useful for investigating a correction.