A recent query of Hive and Tornilo showed a significant discrepancy in data on banners displayed on mobile. Tornilo numbers from beacon/impression with the banner_shown status for the stewnoms_mobile banner are around 30% higher than the actual calls to Special:BannerLoader, as queried from Hive. This is odd, because that status should only be set after the banner has been fetched and injected. Though the data from beacon/impression is sampled at 1%, it shouldn't be consistently so much higher than actual banner fetches.
We should investigate the cause. It could be due to some issue in the queries used, or in the analytics pipelines. An actual problem with client-side CN code seems to be a more remote possibility, though it shouldn't be discounted, either.