Where’s My MEP Pro
How the vote predictions work
84%
accuracy on 5,728 held-out votes, each predicted from the rest of the record, then checked against how the MEP actually voted.
Every prediction is a transparent rule, not a black box. For a vote an MEP missed, we blend three signals from the public roll-call record:
- Group majority (50%), how the MEP’s own political group actually voted on that exact vote.
- Own topic history (30%), how the MEP has voted on the same subject in the past.
- Recent lean (20%), their overall direction across recent cast votes.
A vote with no signal is skipped, never guessed. Every prediction shows its confidence and the exact reasons behind it.
Is the confidence honest?
When we say 90% confident, we should be right about 90% of the time. Measured on the same held-out set:
| Stated confidence | Actual accuracy | Sample |
|---|---|---|
| 50–60% | 43% | 175 |
| 60–70% | 65% | 810 |
| 70–80% | 79% | 533 |
| 80–90% | 93% | 492 |
| 90–95% | 90% | 3,670 |
Recomputed daily · last run 11 Jul 2026