Tight band lands within 3% of the real sale price.
Blind out-of-sample: trained on 985,346 Dubai Land Department sales spanning 23 years of registered transactions, then tested against 18,327 sales from 18 Apr to 12 Jun2026 it had never seen. Predicted against the price each unit actually sold for. For comparison, DLD's own free Smart Valuation reports a 5–8% margin of error.
3%
Tight band
with same-building comps
6.15%
All apartments
median error
91%
Conformal coverage
90% target hit honestly
Two tiers of accuracy: Tight band on units where we find recent same-building sales (lands within 3%); standard band elsewhere (within 6.15% median). For ultra-luxury units above AED 25M with no comparable sales, the engine refuses to value and points to a RERA-licensed valuer instead. Honest abstention is a feature, not a gap.
Predicted vs. what it actually sold for
160 blind sales, each dot is one
The closer a dot sits to the diagonal, the closer the algorithm was. It never saw these prices.
How far off, exactly
Distribution of the miss across the 160 sales shown.
0 to 5%
82 · 51%
5 to 10%
29 · 18%
10 to 15%
20 · 13%
15 to 20%
10 · 6%
over 20%
19 · 12%
Real calls, side by side
DAMAC LAGOONS - VALENCIA
1BR · 740 ft² · Off-Plan
valued AED 1.24M, sold AED 1.24M
0.0% off
THE POLO RESIDENCE
1BR · 806 ft² · Ready
valued AED 1.23M, sold AED 1.19M
3.3% off
Beach Walk Residences 4 by Imtiaz
1BR · 886 ft² · Off-Plan
valued AED 2.39M, sold AED 2.58M
7.4% off
THE FIFTH TOWER
2BR · 1,117 ft² · Off-Plan
valued AED 1.64M, sold AED 1.21M
35.5% off
Bought, then resold higher, on the public record
Real Dubai buildings where registered Land Department sales climbed over 2026. Bought and Now are the actual median resale prices on record. Folio says is the algorithm valuing the same unit today, same engine, no inside knowledge.
Dubai Creek Harbour
Harbour Gate
2 bed · ~1,141 ft² · Ready
Worth now
AED 2.65M
8.2%
Bought AED 2.45M in Feb 2026, resold AED 2.65M by Apr 2026
Folio's algorithm values it at AED 2.62M today, in line with the latest resales.
Out of sample means no leakage: the algorithm was built first, then scored on sales it had never seen. A handful of registered transfers (bulk deals, related party sales, mislabeled sizes) sit well outside the market and pull individual errors high. The median absorbs them. This measures the algorithm, not a promise on any single unit.
