Which inputs changed before labels matured?
Decision-time feature distributions as of 2024-10-01. This comparison uses application and feature events, with no outcome file or model fitting.
Reference decisions: 2023-07-01 to 2023-12-31. Current window: 2024-07-01 to 2024-12-31, restricted to decisions on or before the cutoff. Every feature must have been available by its own decision date.
Input by input
Distances range from 0 (identical empirical distributions) to 1. KS distance is the largest absolute difference between the two empirical cumulative distributions. Total variation compares the two segment proportions. Means retain source units; principal_cents is in cents.
Swipe or scroll the tables horizontally to see every column.
| Input | Distance | Measure | Reference mean | Current mean |
|---|---|---|---|---|
| utilization | 0.096 | KS distance | 0.51151645 | 0.46970877 |
| debt_to_income | 0.057 | KS distance | 0.45630308 | 0.44372536 |
| prior_delinquencies | 0.050 | KS distance | 0.52916667 | 0.45625 |
| principal_cents | 0.055 | KS distance | 1602671.16666667 | 1675230.225 |
| segment | 0.019 | Total variation | — | — |
By current origination month
Every month uses the same full reference population. Months with fewer than 20 applications have no distance estimate.
| Vintage | Applications | Utilization | Debt / income | Prior delinquencies | Principal | Segment |
|---|---|---|---|---|---|---|
| 2024-07 | 80 | 0.154 | 0.067 | 0.113 | 0.123 | 0.100 |
| 2024-08 | 80 | 0.177 | 0.113 | 0.037 | 0.081 | 0.000 |
| 2024-09 | 80 | 0.131 | 0.179 | 0.100 | 0.100 | 0.025 |
| 2024-10 | 80 | 0.067 | 0.135 | 0.025 | 0.106 | 0.000 |
These are marginal distribution distances, not statistical significance, model deterioration, default rates or an automatic decision rule. They do not detect every joint-distribution change. Missing decision-time features fail validation rather than being filled with zero; missing-rate trends are a separate, unimplemented check.