Intermittent Missing Hits in Vertex/Tracker During Reconstruction Process

I have encountered an intermittent issue where a hit is occasionally missing in the vertex or tracker during the reconstruction process. This seems to occur randomly and is not specific to any detector model, theta, or momentum. This problem does not correspond to a specific event but affects the entire reconstruction process. Interestingly, relaunching the reconstruction with the same conditions resolves the issue.

I add some examples:
1-
First reconstruction for 80 and 89 deg (third set of points) and r position of the hits in the vertex:

Second reconstruction with the same input resulted in:

2-
With the CLD_o3_v01 geometry, the first reconstruction showed these pT and d0 resolution:

While a simple relaunch altered the results, indicating a possible missing hit in the tracker as the difference in d0 is quite small. Unfortunately, I no longer have access to the files for further analysis

This issue has been observed multiple times, although it appears to be random.
Do you any idea what could be the problem?

No source file, test, event, or reproducible input is identified in the report. Start by obtaining a minimal reconstruction case and comparing repeated runs with identical conditions; the work is done when the intermittent missing hit is reproduced, its cause is identified, and a regression check demonstrates consistent results.

Written by the indexing model from the issue text.