Read Machine Learning Insights and Factor Attribution
Updated 31 October 2026 · 2 min read · 10 Contents
Plain-language meaning
Read Machine Learning Insights and Factor Attribution is a source-backed research workflow for one exact exchange listing or a clearly stated market scope. It describes available evidence and never creates a recommendation or trade instruction.
The ML Insights tab on Stock Details (Max plan) presents 20-session benchmark outperformance probabilities, a Why the model says this block with feature tailwinds and headwinds, the volatility-expansion and three-year compounding model heads, macro market regimes, and walk-forward validation diagnostics. The same probabilities are available as Max fields in the screener.
1. Confirm exchange and instrument. 2. Select the intended period or scope. 3. Read the as-of time, unit and native currency. 4. Inspect available rows. 5. Open supporting evidence. 6. Treat missing inputs as unavailable. 7. Save or compare only like-for-like evidence.
Illustrative only: a stock can show a 74% outperformance probability with 200-session trend alignment as the largest tailwind and a high RSI as a headwind; contributions are in the model's own scale (log-odds for the logistic model, probability points for the forest challenger), not percentages or returns, and when the feature history is still maturing it displays an honest building state The example is not a live quote, forecast or expected result.
Higher, lower, positive and negative values retain the definition of the displayed field and period. A genuine zero is evidence; unavailable is missing evidence. No single field determines whether an instrument is suitable.
A genuine source-reported or calculated zero is displayed as 0. Unavailable (—) means verified evidence is absent, unsupported, stale or not computable; it is never silently changed to zero.
Coverage, reporting periods, stale observations, corporate actions, methodology changes and small samples can limit comparison. Never combine unlike currencies, scales, sessions or instruments without an explicit compatible basis.
ML models are trained over walk-forward folds per exchange; listings without sufficient matured price history display an honest building status without simulated numbers.
Continue with Data sources and freshness, Exchange and universe scope, and the linked A-to-Z concepts. Verify the current primary source before relying on time-sensitive evidence.
This guide is descriptive education, not investment advice, a prediction, a recommendation, or an instruction to buy, sell, rebalance or place an order.
Educational information only; not investment advice or an order service.
