Together with @cocov we found that the HitPattern class confuses A and C side. The official nomenclature defines even numbers as C side and odd numbers as A side https://cds.cern.ch/record/2683878/files/LHCb-PUB-2019-008.pdf but here in the HitPattern algorithm was the other way round. This was spotted in the monitoring (already last year, but we didn't go so deep and thus didn't find this). This simple fix should address the issue. cc @gligorov who pinged me.
Should go together with Detector!547 (merged) and Rec!3877 (merged)