Reevaluate tie-breaking logic for peak signal selection in 3D reflections #58
Description
In #55, I introduced a deterministic tie-breaking rule for selecting the peak signal in 3D reflections. This was part of a broader bugfix to resolve non-deterministic output resulting from incorrect assumptions I had made when adding 3d connect component analysis to the codebase.
The current tie-breaking approach selects the signal with the lowest z, then y, then x coordinate when multiple signals have the same maximum intensity. This mirrors DIALS behaviour and ensures consistency across runs, but - as brought up in discussion - this is not necessarily the "correct" method to use.
Some non-exhaustive alternatives include:
- Nearest peak to centroid distance
- Average peak to centroid distance
- Furthest peak to centroid distance
We should revisit whether the current z/y/x-based tie-breaking is justified - both here and in DIALS - or whether another method would yield more meaningful or robust results.