The ledger
Our identity is radical transparency. Every prediction is timestamped, locked at kickoff, and scored properly after full-time — wins and losses alike. The misses stay visible, permanently. See our methodology for exactly how these numbers are computed.
And you don’t have to take our word for it — every scored call is sealed into a public SHA-256 hash chain, each row locking the one before it, so any change to a past result breaks the chain in a way anyone can check. Tamper-evident, not just promised.
Running record
0.57
Mean Brier score
0 best, 2 worst — lower is better. Losses included.
0.90
Mean log loss
Punishes confident misses — lower is better.
Across 18 scored predictions — wins and losses alike. The outcome we leaned towards came in 5 of 18 (28%); every miss is counted in full.
Sample size matters. Small samples are noisy; these numbers only mean something over dozens-to-hundreds of scored predictions. We show the count alongside every metric so the record is honest about its own limits.
Calibration — checking our work
Calibration asks a simple question: when we say 30%, does it happen about 30% of the time? Each band groups every home, draw and away probability we assigned, so 18 matches give 54 data points. Well-calibrated overall — when we say 45%, it lands about 46% of the time.
| Confidence band | Data points | We predicted | It happened |
|---|---|---|---|
| 0–10% | 3 | 0% | 0% |
| 10–20% | 13 | 10% | 8% |
| 20–30% | 0 | — | — |
| 30–40% | 6 | 33% | 33% |
| 40–50% | 26 | 45% | 46% |
| 50–60% | 6 | 50% | 50% |
| 60–70% | 0 | — | — |
| 70–80% | 0 | — | — |
| 80–90% | 0 | — | — |
| 90–100% | 0 | — | — |
Every scored call
Newest first. Lower Brier is better (0 to 2). Tap any match for its full breakdown, including log loss.
Probabilities from a third-party model; context, not a guarantee.
Analysis and probabilities only — not betting advice. Outcomes are uncertain; we do not guarantee results and we do not claim to beat the market.