Predictions published in public before the sale closes. Graded after, hits and misses both. Built on a verified dataset of every lot in every sale — not the ones that flatter the record.
The state of play
An auction house publishes an estimate. The lot sells. Everyone moves on. Nobody goes back to check whether the estimate was any good — so nothing corrects, and the numbers drift.
1,534
Lots verified
Two houses. Every lot in thirteen Sotheby's sales, plus three years of Heritage results
63%
Sold under estimate
219 lots — every one where the house published a number to be measured against
91
Calls graded
Every one published before the sale closed. None edited afterwards.
15%
What guessing scores
A no-skill rule on the same lots. That's the number the research has to beat.
Findings
Ten game-issued jerseys, one game, one sale, all matched to the same locker-room image, all bench players. The only variable was which team won. The distributions don't even overlap — the cheapest jersey from the winning side beat the most expensive from the losing one.
The same player's jersey, from the same event, sold twice this year. In the premier sale: $177,800. In the weekly clearing sale: $12,160. And the cheap one carried a 39-point triple-double.
One pair of Michael Jordan's 1998 Finals shoes, split by half. The photo-matched half sold for $2.2 million. The half where the match failed sold for $102,000. Same game, same feet, same night. Across 450 Heritage lots the effect holds at 5.28×, in three separate years.
A shoe bought direct from the athlete at $70,400 resold ten weeks later for $48,800. Nothing about it changed. What changed is that the event was over.
Two lots in one sale, described word for word identically. One sold for $512. The other for $2,560. Some of what looks like a bad call is just which minute a lot closed in.
Open now
Closes 3 August
Six calls. The bet is a photo-matched Pascal Siakam over Kevin Durant — evidence over name.
Read the calls →
Closes 8 August
Eighteen calls. One is already lost with ten days to run, and I caused it.
Read the calls →
Closes 21–23 August
Twenty-six blind calls made before bidding opened. Ten say the lot doesn't sell at all — including the Hand of God ball.
Read the calls →
How this works
Every call goes up before the sale closes and gets archived the same day at the Internet Archive, so nobody has to take my word for the timing. Nothing is edited afterwards. When a finding turns out to be built on bad data, it gets withdrawn in public with the reasoning attached — because a record that only contains wins isn't a record.