We compare each sale with our estimate from the day before it sold. Pick a group to see how close we came.
Grade 8, every grader · 1 to 4 sales in the 90 days before · evidence: limited
Only 0 sales in this group. We show figures for groups of 50 sales or more.
Sales before Oct 18, 2026 are checked against the estimate as it stood that day, rebuilt from earlier sales only. From Oct 18, 2026, we use the estimate we showed.
Groups by era, grader and grade with 200 or more sales since January 2025. Tap one to see it.
Each pair is one sale we counted and our estimate for that card, grade and printing as it stood the day before the sale, moved into the sale's venue terms (Buy It Now and auction-house prices carry the venue gap measured for the era). The miss is |ln(estimate / price)|, shown as exp(x) - 1: the Stage 0 symmetric error, so a sale at half or at double the estimate counts the same. "Half landed within" is the median miss; "80% within" is its 80th percentile. Groups with fewer than 50 pairs show no figures.
Sales before Oct 18, 2026: the causal replay (Stage 0), on the frozen evaluation set of every sale we counted from January 1, 2025 on the replayed cards. Each estimate is rebuilt from the sales and sale checks available before the sale's day, with no human review decisions; grades with fewer than 5 sales in 90 days use the modeled estimate as of that week. Sales from Oct 18, 2026: the estimate published the night before (version 1.2 where it was published, else the modeled estimate). The backfilled daily history before October 17, 2026 is not used: it was computed later with later sale checks.
Two copies of the same card and grade sold within 7 days of each other typically differ by about 7.5 percent, so no estimate can be expected to land much closer than that to a single sale. Sales in 90 days counts the sales we counted of that grade in the 90 days before the sale. Evidence is the word shown with the estimate at the time; from similar cards means the modeled estimate.