Futures Flow methodology and limitations
Futures Flow combines dated CFTC positioning data, statistical analysis and AI-generated commentary. The underlying observations describe trader exposure; analytical patterns and narrative are interpretations. Readers should verify the observation date, report format and source numbers before using any output in a decision.
Reviewed
Where does the data come from?
The U.S. Commodity Futures Trading Commission publishes COT reports. Futures Flow stores observations and processes Legacy, TFF and Disaggregated formats. Weekly reports normally describe Tuesday positions and are released on Friday. Release delays, missing observations and different market naming conventions can affect availability. The CFTC does not endorse Futures Flow.
How are positions compared?
Net positioning equals long contracts minus short contracts. Normalized indices and z-scores provide context within a selected historical window. The public market pages show Legacy categories and identify the number of stored observations used for their empirical percentile rank. Percentile conventions differ across tools; compare the definition and sample rather than the label alone.
How do the market maps work?
PCA and UMAP reduce the dimensions of positioning features. Clustering groups observations with similar inputs. Normalization windows, model versions and coverage affect the result. Visual proximity is not a measure of expected returns, and a cluster identifier does not permanently correspond to a named economic regime.
How are weekly narratives produced?
The report pipeline assembles stored positioning and analytical context for language-model generation. The code includes structured checks and repair logic, but checks have limited coverage. Retrieved historical summaries can be older than the current positioning observation. The report observation date and publication timestamp are displayed independently in the archive.
What should readers verify?
Check the actual observation date, category, futures-only or combined coverage, units and lookback. Compare any quoted numbers with the source table. Distinguish model output from direct CFTC observations. AI-generated text may make numerical, contextual or reasoning errors even when its output passes automated checks.
What does this analysis not establish?
The product does not establish causation, validate a profitable trading strategy or provide personalized investment advice. COT data is delayed and aggregated. A crowded position can persist, correlations can change, and historical analogues may fail. Readers remain responsible for independent verification and risk decisions.