Q1 approval-rate push
Loosening the manual-referral queue and adding two underwriters should raise the approval rate. CR-09 is carried because the way to raise approvals is to approve worse applications, and that is what the guard is for.
Recorded 2025-12-18, before the window opened on 2026-01-01. A hypothesis written afterwards is a description, and the difference matters when the result is read.
The goal was met AND a guard moved the wrong way inside the same window. R-G-17 reports both at equal weight: this is not a success, it is the pattern the guard was invented to catch.
Campaign window 2026-01-01 → 2026-04-30 against a baseline of the same length immediately before it, 2025-09-03 → 2025-12-31. The baseline is not chosen — it is derived from the window, so it cannot be picked after the result is known.
| METRIC | BASELINE MEAN | WINDOW MEAN | CHANGE | GOAL |
|---|---|---|---|---|
| Provision expense vs budget CR-05 · higher is better | 51.6% 3 pts | 53.6% 4 pts | +2.01pp improved | 53.0% |
| Gross NPL before write-off + write-off volume CR-09 · lower is better · guard | 28.5 3 pts | 30 4 pts | +1.53 moved the wrong way | — |
- 1 other campaign ran over part of this window: Operations throughput, Nov–Jan.
- CR-05 was already moving in this direction BEFORE the campaign began — the baseline itself improved on the period before it. Part of what you are reading is a trend the campaign did not start.
- A movement that coincides with a campaign is not a movement caused by it. This platform reports the coincidence and does not attribute it — attribution needs a control group, and there is not one.
The goal of 53 is the campaign’s, not the registry’s. CR-05’s target is still whatever the registry says it is, its threshold is unchanged, and closing this campaign wrote nothing to either. R-G-16.