Accuracy
How accurate is the Quirkies model?
Every score comes from the newest chronological hold-out; the target sale is excluded from training, comparables, and its causal floor proxy.
Median accuracy (100−APE)
93.5%
Mean accuracy (100−APE)
90.6%
Within ±5%
39%
Within ±10%
70%
Within ±20%
90%
Bias
-2.6%
Accuracy by market tier
Floor, middle-market, and grail sales are scored separately so aggregate accuracy cannot hide a weak segment
| Tier | Sales | Median accuracy | Within ±10% | Bias |
|---|---|---|---|---|
| floor | 1132 | 93.6% | 71% | -2.3% |
| middle | 18 | 66.7% | 11% | -21.0% |
| grail | 1 | 31.5% | 0% | -68.5% |
Weekly median accuracy
Chronological hold-out accuracy by sale week
Error distribution
Signed hold-out error; positive means over-predicted
Scored hold-out sales
60 out-of-sample predictions frozen at evaluation time
| Item | Tier / evidence | Date | Sale | Model | Error | Accuracy |
|---|---|---|---|---|---|---|
| #3966 | floor · high | Aug 3, 2026 | 0.900Ξ | 1.033Ξ | +14.8% | 85.2% |
| #4846 | floor · high | Aug 1, 2026 | 1.080Ξ | 1.058Ξ | -2.0% | 98.0% |
| #1016 | floor · high | Aug 1, 2026 | 1.490Ξ | 1.264Ξ | -15.2% | 84.8% |
| #804 | floor · high | Aug 1, 2026 | 1.066Ξ | 0.978Ξ | -8.3% | 91.7% |
| #4044 | floor · high | Jul 31, 2026 | 1.080Ξ | 0.997Ξ | -7.7% | 92.3% |
| #1204 | floor · high | Jul 31, 2026 | 1.075Ξ | 0.969Ξ | -9.9% | 90.1% |
| #1016 | floor · high | Jul 31, 2026 | 1.230Ξ | 1.307Ξ | +6.3% | 93.7% |
| #4932 | floor · high | Jul 30, 2026 | 0.927Ξ | 1.033Ξ | +11.4% | 88.6% |
| #3965 | floor · high | Jul 30, 2026 | 0.927Ξ | 1.029Ξ | +11.0% | 89.0% |
| #804 | floor · high | Jul 30, 2026 | 0.950Ξ | 1.032Ξ | +8.6% | 91.4% |
| #3740 | floor · high | Jul 29, 2026 | 1.082Ξ | 1.033Ξ | -4.5% | 95.5% |
| #2730 | floor · high | Jul 29, 2026 | 1.090Ξ | 0.996Ξ | -8.6% | 91.4% |
| #230 | floor · high | Jul 28, 2026 | 1.079Ξ | 0.952Ξ | -11.8% | 88.2% |
| #1204 | floor · medium | Jul 27, 2026 | 0.950Ξ | 1.015Ξ | +6.8% | 93.2% |
| #230 | floor · high | Jul 27, 2026 | 0.951Ξ | 1.046Ξ | +10.0% | 90.0% |
| #1081 | floor · high | Jul 27, 2026 | 1.089Ξ | 0.977Ξ | -10.3% | 89.7% |
| #1081 | floor · high | Jul 27, 2026 | 0.940Ξ | 1.127Ξ | +19.9% | 80.1% |
| #3345 | floor · high | Jul 26, 2026 | 1.100Ξ | 0.976Ξ | -11.3% | 88.7% |
| #4242 | floor · medium | Jul 26, 2026 | 1.100Ξ | 1.066Ξ | -3.1% | 96.9% |
| #1615 | floor · medium | Jul 26, 2026 | 1.100Ξ | 1.000Ξ | -9.1% | 90.9% |
| #153 | floor · medium | Jul 26, 2026 | 1.200Ξ | 1.111Ξ | -7.4% | 92.6% |
| #2119 | floor · high | Jul 26, 2026 | 1.088Ξ | 0.965Ξ | -11.3% | 88.7% |
| #2999 | floor · high | Jul 26, 2026 | 1.100Ξ | 1.024Ξ | -6.9% | 93.1% |
| #2921 | floor · high | Jul 26, 2026 | 1.030Ξ | 0.954Ξ | -7.4% | 92.6% |
| #1125 | floor · high | Jul 26, 2026 | 1.050Ξ | 0.997Ξ | -5.1% | 95.0% |
| #4555 | floor · high | Jul 26, 2026 | 1.039Ξ | 1.014Ξ | -2.4% | 97.6% |
| #1219 | floor · high | Jul 26, 2026 | 1.040Ξ | 0.949Ξ | -8.8% | 91.3% |
| #2831 | floor · medium | Jul 26, 2026 | 1.080Ξ | 1.056Ξ | -2.2% | 97.8% |
| #4339 | floor · medium | Jul 26, 2026 | 1.030Ξ | 0.956Ξ | -7.2% | 92.8% |
| #4678 | floor · high | Jul 26, 2026 | 1.074Ξ | 0.976Ξ | -9.1% | 90.9% |
| #1719 | floor · high | Jul 26, 2026 | 1.035Ξ | 0.953Ξ | -7.9% | 92.1% |
| #4031 | floor · high | Jul 26, 2026 | 1.040Ξ | 0.980Ξ | -5.8% | 94.2% |
| #4253 | floor · high | Jul 26, 2026 | 1.030Ξ | 0.948Ξ | -8.0% | 92.0% |
| #2458 | floor · high | Jul 25, 2026 | 1.028Ξ | 0.965Ξ | -6.1% | 93.9% |
| #2458 | floor · high | Jul 24, 2026 | 0.906Ξ | 1.054Ξ | +16.3% | 83.7% |
| #897 | floor · high | Jul 23, 2026 | 1.080Ξ | 1.043Ξ | -3.4% | 96.6% |
| #4366 | floor · high | Jul 22, 2026 | 1.030Ξ | 1.052Ξ | +2.1% | 97.9% |
| #1045 | floor · high | Jul 22, 2026 | 1.026Ξ | 0.952Ξ | -7.2% | 92.8% |
| #2921 | floor · high | Jul 22, 2026 | 0.915Ξ | 1.003Ξ | +9.6% | 90.4% |
| #771 | floor · high | Jul 21, 2026 | 1.020Ξ | 0.981Ξ | -3.8% | 96.2% |
Model quirkies-1785736763-2026-08-03.2 · evaluated through 8/3/2026. Method details on the methodology page.