Two-Layer Validation

How is data validated before publication?

Intratec applies a two-layer validation approach to all data prior to publication: an automated processing stage followed by expert human review. The automated stage prepares and tests the data, while market experts — analysts with sector-specific knowledge — act as the final quality gate. Both layers run every month before any figure is released, and the same process applies uniformly across every Intratec solution rather than varying dataset by dataset. Validation does not stop at publication: every forecast and every preliminary figure is later confronted with the final official value once it arrives, so deviations between what the models estimated and what was officially reported can be traced, investigated, and fed back into recalibrating or replacing the underlying model. Together, the two checks — before release and after — keep both the immediate figure and the underlying model itself accountable to real outcomes.

What does automated cross-referencing check?

The automated stage cross-references multiple official sources and applies mathematical models to identify inconsistencies between them. Cross-referencing means comparing the same data point across independent official records to confirm it agrees. Automation improves quality by reducing errors introduced through manual handling and by removing bias from market participants, since the system applies the same consistent, validated rules every month.

What is a modeled figure cross-checked against?

The review framework compares each modeled price against seven independent reference points: the same commodity as reported by independent sources; the same commodity assessed at different geographic locations; exporter and importer trade reports for the same commodity; more actively traded neighboring specifications, such as a closely related grade or delivery term; adjacent time frames, covering prior months and seasonal patterns; feedstocks and derivatives, tracing upstream and downstream price relationships; and different transport types, trade volumes, and shipping routes. The purpose is to catch inconsistencies arising from missing data, mathematical errors, technical issues, or data anomalies. Where the source data does not reflect typical market behavior after these comparisons, the model output is published instead of the data-derived result, so that a single anomalous data point does not distort the published series.

What does expert human review add?

After the automated stage, market experts review the model outputs before publication and serve as the final quality gate — the last check a figure passes before release. This human review covers a representative sample of each month's output: any value flagged as anomalous by the automated stage is always reviewed, along with a sample of the remainder. It adds contextual judgment that automated rules alone cannot provide, keeping processing consistent while ensuring outputs remain sensible in real market conditions.

How is the reviewed sample chosen?

Selection is guided rather than random — a risk-based review. Values that the automated stage flags as anomalous or inconsistent are always routed to expert review, so human attention concentrates where the probability of an issue is highest. A sample of the remaining, unflagged values is also reviewed each month, providing an independent check that the automated rules themselves are not silently missing problems. The combination directs expert judgment where it matters most while keeping the whole pipeline under human oversight.