Your current implementation for the Jensen-Shannon Distance (JSD) is leveraging scipy.spatial.distance.jensenshannon, which defaults to computing the JSD using logarithms with base e. Under this definition, the JSD is bound between 0 and sqrt(ln(2))=0.83255... (see, for example, the JSD Wikipedia article)

I believe you want to compute JSD using base 2 logarithms, so that it is bound between 0 and 1 as you state in your data drift detection blog post.

Here is code which reproduces this issue by computing the JSD between two extremely drifted distributions:

from evidently.calculations.stattests import jensenshannon_stat_test
import numpy as np
import pandas as pd

x = pd.Series(np.random.normal(100, 10, 100_000))
y = pd.Series(np.random.normal(10_000, 10, 100_000))

print(jensenshannon_stat_test(x, y, feature_type='num',threshold=0.1))
>>> StatTestResult(drift_score=0.8325546111576977, drifted=True, actual_threshold=0.1)

Note that the drift_score is close to the current maximum theoretical value of sqrt(ln(2))=0.83255..., instead of the desired value of 1.