Kristian Fagerlie demonstrates two experimental features in Sikt Intelligence: replaying a historical news snapshotBacktesting applies a strategy to historical data to estimate how it would have behaved under past market conditions. and asking an AI system to map possible future scenarios. He describes timestamped saved pages intended to restrict retrieval to material available before a chosen date. That retrieval boundary is a design claim in this demonstration, not an independent audit of training-data leakage or hindsight biasLook-ahead bias occurs when a historical evaluation uses information that would not have been available at the original decision time..
Kristian Fagerlie first revisits a reported diesel-policy development, then tests an inflation-release scenario. He shows the model assigning low probabilityConfidence calibration measures whether an AI system's stated certainty matches how often its predictions or assumptions are actually correct. to some outcomes and missing part of an inflation result. A later exercise supplies the actual inflation figures as a hypothetical input before asking about market direction, which is different from predicting both the release and its consequences in advance.
Kristian Fagerlie argues that scenario enumeration can reveal paths worth monitoring even when the model assigns them little probability. He also acknowledges that one case is not proof and that a selected scenario did not match every observed condition. The examples support investigating a testing method, not claims of reliable trading returns or validated economic predictions.
Watch on YouTube
