- added errors to output
- replaced metric calculations by binomial wilson interval
- averaging only once over all events instead of each event
- added GNN output mapped to each hit