GPU TPC: When running cluster rejection based on interpolation, also reject during update with current cluster position

[davidrohr]

No description provided.

[github-actions]

REQUEST FOR PRODUCTION RELEASES:
To request your PR to be included in production software, please add the corresponding labels called "async-" to your PR. Add the labels directly (if you have the permissions) or add a comment of the form (note that labels are separated by a ",")

+async-label <label1>, <label2>, !<label3> ...

This will add <label1> and <label2> and removes <label3>.

The following labels are available
async-2023-pbpb-apass4
async-2023-pp-apass4
async-2024-pp-apass1
async-2022-pp-apass7
async-2024-pp-cpass0
async-2024-PbPb-apass1
async-2024-ppRef-apass1
async-2024-PbPb-apass2
async-2023-PbPb-apass5

[miranov25]

Hello David and all,

As discussed during today’s TPC meeting, I’d like to briefly summarize the motivation and context behind this fix needed for the apass5 but influencing the past data.

---

TPC QC mattermost message summary: 🐞 Bug in Tracking Update Logic

A bug was identified in the track reconstruction logic: the χ² cut was being applied at a different point than where the actual track update takes place. This inconsistency meant that even a single outlier cluster could cause the entire track update to fail — resulting in track loss and an invalid χ² assignment.

This issue was particularly problematic in regions where the distortion map and its inverse differ slightly due to approximations, making the tracking unreasonably sensitive to local imperfections.

David has already implemented a fix for this behavior in [PR #14188](#14188), discussed in the 14:00 TPC meeting.

---

✅ Updated Return Logic

The fix introduces a more granular return code from the update step:

• retVal = 0: Update successful
• retVal = 1: Update failed → abort track
• retVal >= 2: Cluster rejected (but track continues)
  • retVal = 2: Cluster outside allowed search window
  • retVal = 3: Edge cluster rejected due to rejectEdgeClustersInTrackFit
  • retVal = 4: Rejected during interpolation
  • retVal = 5: Rejected due to large χ² during update

---

This change makes the tracking more robust against local issues and should significantly reduce unnecessary track loss. We still need to monitor and identify regions where the map and inverse deviate too much — but this update is an essential step toward more stable reconstruction.

Best regards,
Marian I. Ivanov

[miranov25]

Hello @davidrohr and @shahor02,

Let’s proceed with the test.

I expect that a significant fraction of the issues observed in apass5 (and even in older passes) will be resolved. I've previously noticed non-physical, large TPC χ² values not only in the Pb–Pb apass5 data but also in earlier productions. In the specific case of apass5, it seems the inconsistency between the distortion map and its inverse is more pronounced than usual.

If the fix proves effective, we should consider restarting the production with corrected settings.

Best regards,
Marian

[alibuild]

Error while checking build/O2/fullCI_slc9 for 6940a98 at 2025-04-16 16:48:

## sw/BUILD/O2-latest/log
/sw/SOURCES/O2/14188-slc9_x86-64/0/GPU/GPUTracking/Merger/GPUTPCGMPropagator.h:56:31: error: missing ',' between enumerators
Error: /sw/slc9_x86-64/ROOT/v6-32-06-alice7-3/bin/rootcling: compilation failure (/sw/BUILD/f6e68dd875af9a43559c7e007d2ff5d62d2c0a83/O2/GPU/GPUTracking/G__O2GPUTrackingc07ff93271_dictUmbrella.h)
ninja: build stopped: subcommand failed.

Full log here.