The Proof Room · Galaxy Calibration

Trust is an architecture, not a tagline.

Every credibility receipt the platform publishes, gathered in one place. No fabricated picks, no invented stats, no silent edits. Each number stays gated until the settled sample is large enough to back it honestly. And once it settles, it stays in the record, win or loss.

The calibration we measure ourselves on is the calibration of that market-implied probability on our settled two-way moneyline picks: the average implied probability across books, normalised to remove the vig, fixed at publish time and committed to the pick's proof receipt, never recomputed. Confidence is a ranking score and is not part of that measurement; on spreads and totals it does not include the market's probability of the outcome, and on moneylines it is anchored on the market's win probability and labeled as such.

The interactive chart below groups settled picks by confidence score, and its “expected” line is that score read as a percentage. It is a separate view of the same record, not the market-implied measurement described above, and it is not a forecast. We have measured how far off it runs: across 2,385 settled picks, the rows we scored 80 and above won 51.9% of the time, not the 86.6% the score reads as. It also does not climb in a straight line - our 75-79 rows have won more often than our 90-plus rows. Read it as a picture of how our selection score has behaved, not as a rate we are claiming.