Research
Findings built from a verified dataset of 5,769 auction lots across eight houses. Every claim carries a number, a sample size, and a way to be proven wrong.
The flagship. 5,769 lots across eight sale streams, sport and entertainment. The market asks every lot two questions — was it there, and did it matter — and pays in multiples for the answers. Includes five findings we published and had to withdraw.
I published that the market pays for the moment, then bet against my own rule eighteen times at Goldin and lost. The thesis held. Three forces — the moment, the floor, the name — and the errors that separate them.
Every sale is captured from the auction house's own published results page — the full population, not the lots I happened to call. Lot by lot, into a dataset that currently runs to 5,769 entries across eight houses — Sotheby’s, Heritage, Goldin, Grey Flannel, SCP, Julien’s, Heritage Entertainment and NFL Auction.
That sounds obvious. It isn't what I was doing three days ago, and it cost me a published finding. That retraction is here.
If you want the scoring rules and the baseline a no-skill rule achieves, that's the methodology page.