Reportability Monitor
Overview
Quality metrics
Distribution & bias
Drift vs benchmark
Assessment
Drift vs benchmark
Comparing the current window against the training/benchmark window. Overall PSI 0.328 →
significant
Score shift shows how the model's output distribution moved from the benchmark.
PSI above 0.25 is a significant drift signal and should trigger review.
Per-feature drift
Feature
Type
PSI
Flag
Test
model_score
numeric
0.044
stable
KS 0.087 (p=0.000)
narrative_length
numeric
0.028
stable
KS 0.059 (p=0.016)
product
categorical
0.328
significant
χ² p=0.000 · new: Buy now pay later, Cryptocurrency wallet, Peer-to-peer payment app
channel
categorical
0.009
stable
χ² p=0.109
PSI <0.10 stable · 0.10–0.25 moderate · >0.25 significant. Score drift = the model's outputs moved; input drift = the world moved; together with a recall drop they signal
concept drift
(the input→label relationship changed).