What we did
RefCal regenerated the RadCalNet truth set so that measured surface reflectance was compared with Sentinel-2 surface reflectance on the same basis. It then reran the per-band confidence check over the stored real-data evidence. This corrected an earlier comparison that had treated RadCalNet top-of-atmosphere values as surface truth.
What we saw
On the corrected surface-to-surface basis, the intervals contained the reference value in 0.766 of cases against the stated 0.683 target. The same stored outputs had produced 0.442 when compared with the top-of-atmosphere reference values. Median absolute residual fell from 0.0224 to 0.0097 and RMSE from 0.0369 to 0.0229.
The remaining differences are site-dependent. RVUS and GSCN retain positive offsets, while BSCN and GONA are close to zero. The evidence does not yet separate radiometry, spectral response, view geometry and spatial support as causes.
What it means
The earlier under-coverage was primarily a reference-basis mismatch, not evidence that RefCal's surface-confidence intervals were uniformly too narrow. On the corrected basis the overall intervals are slightly conservative in this real-data set. Because the register still marks the result preliminary and several support limits remain, this set records evidence seen in real data rather than claiming validation under all stated conditions.