Found that at least one Tsnow is sending corrupt data messages, presumably the same issue we saw in the previous 2 projects on the soil wisard sensors. This is happening even though both the Tsnows and the soils are now running versions of the PIC code with double buffering (v2.38 and v1.08, respectively). For the moment, I've set limits on the Tsnow data in sensor_catalog to remove wildly incorrect values, though I guess if the really bad corrupt messages are happening, that more subtle corruption also can occur.
After this change, I restarted both statsprocs...