How close is SetScout’s fair value to the real sale price?

Every set page in SetScout shows a fair value: our estimate of what a new, sealed copy sells for. Each month we check those fair values against the price the set sold for later, and publish the result whatever it shows.

Typical fair value error, 1 month aheadProvisional · scoring until Oct 30, 2026

9.6%

across 1,888 sets valued in July 2026 and sold in August 2026

Half of our fair values were within 9.6% of what the set sold for a month later.

52%

within ±10% of the sale price

80%

within ±20%

32%

of sales inside our likely range (aim: 50%)n = 1,879

How does that compare?

On the same sets, the 6-month sales average came closer, with 8.5% typical error. It does well here because only frequently traded sets can be scored, and that is where an average of past sales works best.1,882 sets

Our fair values ran high: 67% of sales landed below our likely range, 1% above it. Fair value also weighs current asking prices, which sit above what sets actually sell for. The next model version gives actual sales more weight to correct this; its results will be published next to these.1,879 sets

Last month’s sale price: not enough data yet (14 sets).

Fair value also gives an estimate for sets with too few sales for any average. This check cannot score those sets.

See the fair value for any setCollector plan

Results by horizon

How far ahead the check looks. Lower error is better. Longer horizons need time to pass before they can be scored.

Horizon	Typical error	6-month sales avg	Within ±10%	Sets
1 monthProvisional	9.6%	8.5%n = 1,882	52%	1,888
3 months	Pending · first results after Nov 15, 2026
6 months	Pending · first results after Feb 15, 2027
12 months	Pending · first results after Aug 15, 2027
3-year return forecast	Pending · first results after Aug 15, 2029

Results by month valued1 month

One row per month of valuations, so you can see whether accuracy changes over time.

Valued in	Typical error	6-month sales avg	Scored	Status
July 2026 Rules designed after seeing this month	9.6%	8.5%n = 1,882	1,888 of 11,346	Scoring until Oct 30, 2026
August 2026	Fewer than 30 scored so far: not shown		0 of 11,548	Scoring until Nov 29, 2026
September 2026	Fewer than 30 scored so far: not shown		0 of 11,571	Scoring until Dec 30, 2026
October 2026	Fewer than 30 scored so far: not shown		0 of 11,488	Scoring until Jan 29, 2027

Only 100% of the July 2026 valuations could be scored: a set needs at least 3 resales in the target month. Scored sets are the more frequently traded ones.

How we measure

What we check. The fair value shown on each set page estimates what a new, sealed copy sells for today. We compare it with the median resale price for new copies shipped to Germany, one, three, six and twelve months later. Sale prices come from the largest online marketplace for new and used LEGO, recorded monthly.

Fixed rules. Scoring rules are version trackrecord-v1. We take the first valuation of each set each month and never recompute a published valuation. Changing a rule creates a new version, shown separately.

What counts as a result. A month needs at least 3 sales. Sets without enough sales stay unscored and count against the scoring rate above.

Limits. Prices are resale prices before fees and shipping. We designed these rules in October 2026 after seeing the July 2026 results, so July 2026 is marked. We have no results before July 2026: we did not store the history a backtest would need.

Model versions. These results count valuations from 1 model version: valuation-2026-07-v1. A new model version never changes a valuation already published; the details below show each version’s results.

Uncertainty and model versions

95% intervals from 1,000 bootstrap resamples of 93 themes. Whole themes are resampled, because sets in one theme tend to move together.

Measure	Value	95% interval	Sets
Fair value: typical error	9.6%	8.9% to 10.8%	1,888
6-month sales average: typical error	8.5%	7.9% to 9.4%	1,882
6-month sales average: how much smaller our error was	-12.8%	-24.4% to -1.2%	1,882

By model version

Model version	Typical error	Sets
valuation-2026-07-v1	9.6%	1,888