Anomality is ambiguous when scores are tied

[rohori]

Currently, anomalies are defined as entities satisfying both

• score > threshold, and
• rank(score) <= n.

This poses ambiguity when there are ties. Current implementation does not consider this and actually depends on sorting algorithms rather than using DataFrame.rank method directly.

It can be left undefined without any realistic problem, but can be fixed by rank(method). This is a theoretically breaking change.