We do not assume the answer. This page reports whether the competitive rating predicts later results — with denominators, and without calling any of it evidence of ability.
Does Arena rating from earlier matches predict who wins later matches?
Frozen at 2026-09-01. about 50% for both lines below if competitive skill does not persist.
| higher-rated player wins | 64.3% of 28 |
|---|---|
| mean Brier score | 0.241 of 28 |
0.25 = coin flip; lower is better calibrated.
| pre-match probability | matches | predicted | observed |
|---|---|---|---|
| 0–10% | 1 | 8% | 100% |
| 10–20% | 1 | 18% | 0% |
| 30–40% | 1 | 37% | 0% |
| 40–50% | 2 | 50% | 50% |
| 50–60% | 4 | 50% | 75% |
| 60–70% | 13 | 61% | 77% |
| 70–80% | 1 | 77% | 100% |
| 90–100% | 2 | 92% | 50% |
| top quartile, later-half win rate | 81.8% of 11 |
|---|---|
| everyone else, later-half win rate | — |
split matches by time; rank players on the earlier half, then measure the later-half win rate of the top quartile vs everyone else. Train: 15 matches · test: 15.
A rating that predicts later results would be a discovery; so far this is reported as a competitive observation, never as evidence of ability. See also how scores are explained and the Arena record.