Missing VELO Raw Banks using hlt2_PbPb_2025_bgi_prescaled sequence in HLT2CALIB stream

We've found that running the current HLT2 version for PbPb (https://gitlab.cern.ch/lhcb/Moore/-/tree/v57r15-patches) with the hlt2_PbPb_2025_bgi_prescaled sequence (https://gitlab.cern.ch/lhcb/Moore/-/blob/v57r15-patches/Hlt/Hlt2Conf/python/Hlt2Conf/settings/hlt2_PbPb_2025_bgi_prescaled.py) and later we apply a BGI RB filter, some events miss raw banks and DecReports the HLT2CALIB stream.

The error found is:

VeloClusterTrackingSIMD_9d6e67b1    WARNING Suppressing message: 'Missing velo raw banks (check raw bank type)'
Hlt1DecReportsDecoder_e139c7d0        ERROR Hlt1DecReportsDecoder_e139c7d0 :  No HltDecReports RawBank with sourceID 1 found
Hlt1DecReportsDecoder_e139c7d0        ERROR Maximum number of errors ( 'ErrorMax':1) reached.
HLTControlFlowMgr                     FATAL Event failed in Node LumiPVs_nobeamline/LumiPVs_nobeamline_dd7d238c
HLTControlFlowMgr                      INFO Application return code set to 3
HLTControlFlowMgr                     FATAL Event 26886 on slot 0 failed!
HLTControlFlowMgr                     FATAL *** Too many consecutive failures 1, stopping now ***

When the sequence is run without prescaling (so prescale=1.0) and we apply the same filter, this issue has not been found. Indeed, we even tried to modify the prescaling factor and the issue appears, but if the factor is set to 1, everything works fine.

For the lumi program we should be fine, since the HLT2CALIB with unprescaled BGI lines is what matters. Nevertheless, this behaviour is not understood, and everything indicates that the presaling is introducing some undefined effect in the hlt2calib_lines.

We tried to add the setting detectors=["VP"] in the hlt2calib stream as suggested here !5493 (comment 10448093) but it has not resolved the issue.

In order to reproduce the issue:

Run HLT2 on HLT1 mdfs from run 333906

/localdisk/stack_mcolonna/stack/Moore/build.x86_64_v3-el9-gcc13+cuda12_4-opt+g/run gaudirun.py /home/tmombach/tags.py /localdisk/stack_mcolonna/stack/Moore/Hlt/Hlt2Conf/options/hlt2_PbPb_2025_bgi_prescaled.py | tee hlt2.log

Apply the BGI RB filter on the HLT2CALIB stream:

/localdisk/stack_mcolonna/stack/Moore/run gaudirun.py setOpt.py pvrecoBGI.py | tee test_prescale_2025data_HLT2CALIBmdf_prescale099.log

Python files for the filters:

pvrecoBGI.pysetOpt.py @mfontana @tmombach @efranzos