Every rate on EdgeUp is printed with the fraction it came from, and the games in that fraction are the ones the player started, read out of the play-by-play. Not every game he appeared in. This page is the measurements behind that choice, including the test that failed and the finding that was retracted.
Andrés Giménez, TOR. Over 0.5 hits, 2026 regular season. Read from the database at Sep 6, 12:31 AM ET, so it moves as he plays. A push is void at a book, so it leaves both denominators rather than counting as a miss.
The numerator is the same 72 games either way: every game over the line was one he started. The entire gap is the denominator moving, which is what makes this the teaching case rather than the widest one.
The middle fraction is ours too. No competitor prints one, so we computed theirs the way they compute it, to show what their percentage is standing on. This player is chosen, not sampled; the population figures are directly below. Check both numbers on his page.
measured 2026-08-31 and 2026-09-01. On most markets the correction is small. On roughly one in four it is large enough to change what you think you are looking at, and you cannot tell which is which unless the fraction is on the screen. That is the entire reason it is on the screen.
We tested our own history against the closing market and got r = +0.0102 (t = 0.76) across 5,538 settled props. That is noise. A site that had found an edge would not print that number; we print it because it is what we measured.
We also withdrew a +7.87% ROI we thought we had found in NBA minutes. 220 near-DNP games out of 4,140 carried the whole result, and a book voids those. Above 20 minutes the same bet loses. The finding is retracted in writing rather than quietly deleted.
So there is no edge score, no confidence rating and no projected line anywhere on this site. Every figure is either something that already happened, with its denominator, or a price a book is offering right now.
To be exact about what that test showed: it says our numbers carry nothing the closing line does not already contain. It is not a claim that no one can price these markets better than a book, which is a different question and one we have not tested. If we ever do compare prices across books, it will be a comparison of prices and labelled as one, not a projection.