What we did
RefCal inspected real Sentinel-2 results on its common delivery grid. One check sampled delivered results from two real swaths to see whether each carried a full confidence vector across all 13 bands. A second check followed the spatial state for one real 696-cell window through processing and into the delivered cells.
What we saw
All six sampled cells returned a full 13-band confidence vector. Spatial state and its contribution to confidence persisted through the real processing path. A subsequent run completed the full 696-cell window and stamped every delivered cell with the active state generation for that run.
What it means
RefCal has carried band-specific confidence and spatial state into real common-grid outputs end to end. That earns an observed-in-real-data stage. It does not yet show that the propagated covariance is scientifically faithful or that the resulting uncertainty has passed independent validation.