Calibration curve
Each row is a confidence decile. The bar shows the resolved hit rate among theses we closed in that bucket.
While grading is confidence-derived, this curve reflects how the desk’s confidence resolved rather than a
calibration against independent outcomes, so it cannot yet be read as a true reliability diagram.
Biggest misses
hormuz_transit_drop_7d ·
closed at 22%Closed at 22 percent confidence. Thesis statement was not supported by the evidence over the horizon.