I want to thank the reviewers and the editor for making the review process a constructive, pleasant, and exciting experience! I am grateful! n/n

Another “unintended” conclusion from this study (thank you Reviewer #3) is that the DL blocking reconstruction lends support to tropical Pacific gradient reconstructions that are based on very sparse networks. 8/n

The deep learning reconstruction also shows an increased influence of ENSO post-1600 coinciding with a weaker tropical Pacific zonal temperature gradient: so much to unpack here! 7/n

A weaker gradient is consistent with hemispherically reduced blocking as CMIP models project, but if GHG forcing eventually leads to an increased gradient, we may also see increased atmospheric blocking and subsequent extremes. 6/n

This underscores how important it is for climate models to simulate trends in the tropical Pacific surface temperature gradient accurately. 5/n

The results provide paleoclimatic evidence supporting modeling studies that show that changes in the pattern of mean surface temperature in the tropical Pacific are an important factor in explaining patterns of change in blocking frequencies. 4/n

This new way to infer synoptic signals from low temporal resolution paleoclimate records is a flexible temporal downscaling method that can be used to derive variables with complex/unknown relationships to proxies that are not typically targeted in paleo reconstructions. 3/n

The study shows that even though the DL model ingests paleodata assimilation products and has no information on the location of paleoclimate proxies, it is implicitly constrained by them. 2/n