Sale report · graded 20 July 2026
Sotheby's. 173 lots, every close captured. This is the sale the whole record is built on — the only graded set big enough, and with published estimates, to score anything properly against.
Lots tracked
173
Every close captured, including two withdrawn on the day
Calls graded
60
The lots where Sotheby's published an estimate to score against
Beat the house
70%
Closer to the final price than the estimate was
Typical error
44%
House ran 73% on the same lots
And the number that matters more than the hit rate: 49 of those 60 calls were too high. Median overshoot, 42%. I didn't find that by winning — I found it by running a no-skill baseline against myself. The working is here.
I published something off this sale and I've taken it down. It said every six-figure vintage jersey here closed below its low estimate — four for four — and that the only lots to blow past were moments rather than names.
The pattern might still be real. The evidence wasn't.
I built that list from a spreadsheet of truncated lot titles instead of reading the actual catalogue entries. When I went back:
One of my "vintage jerseys" is a warm-up jacket — Wilt's 1972 Lakers Finals jacket, matched to three games. Another is an All-Star jersey with an apparent photo match, not a conclusive one. And the LeBron jersey I used as an example of a name without a moment was his 6,000th career assist — the night he became the fourth player ever to reach 20,000 points, 6,000 rebounds and 6,000 assists. That's a milestone. I just never saw it, because it wasn't in the truncated title.
I also missed one that belonged in the sample. Oscar Robertson's 1960 Rome Olympics gold-medal jersey sold for $96,000 against a $100,000 estimate. A genuine vintage jersey, genuinely underperforming, and I didn't cite it.
So the category was wrong, the sample was wrong, and one of the four was a counterexample I'd misread.
There may still be something here. LeBron's 6,000th assist is a counting-stat milestone rather than a dramatic event, and I've already learned this month that a stat line isn't the same as a moment. But I'm not putting that back up until every lot in it has been read properly, front to back.
New rule, and it's my own fault it's needed: no lot counts as evidence until the full description has been read. Titles lie by omission.
I tag every call. Conviction means there's a real chain of comps behind it. Tape means I logged it so a surprise would have a baseline, but I'm telling you the evidence is thin.
Conviction calls beat the house 73% of the time, across 49 lots. Tape calls managed 55%, across 11. So the labels are carrying real information — which is worth knowing, because it means when I say I'm unsure, you should believe me.
I captured the standing bid on every lot at three days out, two days out and one day out. Then the close.
The typical lot gains 37.5% over those final three days — and two thirds of that lands on the last day alone. A snapshot three days out barely tells you anything. A snapshot one day out tells you a lot.
But the median hides the real shape. Forty-four of those lots closed at exactly their one-day-out bid. Never moved at all. So it isn't that everything drifts up 25% — it's that a minority run hard and a long tail sits perfectly still.
And the better predictor turned out not to be price. It was competition. Lots with five or fewer bids the day before close gained 9%. Lots with thirty or more gained 39%. If you want to know whether something's going to move, count the bidders, not the dollars.
Two things, and neither was comfortable.
The first is the bias. Forty-nine calls too high out of sixty isn't a run of bad luck, it's an anchor — and I only caught it because I bothered to ask what a no-skill rule would have scored. That question should have come a lot earlier.
The second is that a purely mechanical rule — take the standing bid the day before and add 25% — beat the house 72% of the time. Marginally better than everything I did. It uses information a blind call doesn't have, so it isn't a fair fight. But it's a useful reminder that the market is often telling you the answer if you'd just look at the screen.
Balls and strikes.