Predict
For every promise still open, CivicLens estimates the chance it will be recorded as fulfilled within 90 days, 180 days and a year. The table shows how well those estimates did when tested on the past, including where they fall short.
How reliable are these figures?
The model was trained on 15,815 past assurances (data as of 2026-11-09) and checked against what really happened. Each table compares what it predicted with what followed. If the two columns match, the model is calibrated.
Read this before using a number. In the test on past promises, the model scored 0.173 on the Brier scale (lower is better, 0.25 is a coin toss). A plain rule that only looks at how old a promise is scored 0.173, and a simple consensus of methods scored 0.170. So the model is about as good as the simple rules, not better. Its ability to rank promises is modest (AUC 0.49 to 0.67, where 0.50 is chance). Treat each figure as a rough guide, and treat a gap of ten points between two promises as noise.
Within 90 days
| Model said | What happened | Cases |
|---|---|---|
| 6% | 9% | 2,583 |
| 14% | 15% | 4,892 |
| 23% | 15% | 1,254 |
| 32% | 12% | 34 |
Within 180 days
| Model said | What happened | Cases |
|---|---|---|
| 7% | 5% | 846 |
| 16% | 19% | 3,509 |
| 24% | 26% | 3,182 |
| 33% | 38% | 1,156 |
| 42% | 34% | 70 |
Within 365 days
| Model said | What happened | Cases |
|---|---|---|
| 19% | 16% | 32 |
| 26% | 32% | 2,056 |
| 35% | 38% | 2,911 |
| 45% | 46% | 2,858 |
| 53% | 52% | 849 |
| 61% | 42% | 57 |
Open promises, ranked by the model
Each figure is the chance, estimated from how similar promises went in the past, that the promise is recorded as fulfilled within that time. It does not say anything about intent. Click a heading to sort.
| Promise | Ministry | Given on | 90 days | 180 days | 1 year |
|---|