Methodology & Engine

The engine is fed from the official, government-grade record — not listings or estimates. The 8-dimensional state vector splits into four Chaos dimensions (instability) and four Fidelity dimensions (structural strength), each from a named source with measured coverage:

Freshness: transactions register with the DLD continuously; we treat the most recent 2–3 months as provisional and revise them upward as late registrations arrive.

Each node carries the eight dimensions above on a 0–100 scale. They collapse into a single order parameter, Ψ, a transparent weighted blend: Ψ = 0.5·(1 − mean Chaos/100) + 0.5·(mean Fidelity/100) — high fidelity and low chaos push Ψ up. Ψ maps to an exposure band:

The score is a transparent weighted blend, not a black-box model — you can reproduce any node's Ψ from its eight inputs. Cross-dimension propagation and a backtested band calibration are on the roadmap (see Transparency below).

• Sales only (mortgages & gifts excluded), AED per m², dated 2000-onward.
• p5/p95 winsorised within each node — clips distressed, bulk and data-error deals so the median tracks price, not the odd transaction.
• Quarterly median, 3-quarter smoothed; contiguity-gated so we never draw a line across a multi-year gap.
• Areas precomputed (instant), buildings & projects computed live; thin nodes fall back to their parent community, clearly labelled.
• Headline change & CAGR are anchored on multi-quarter averages at both ends, not a single noisy quarter.

For any building, project or unit-type we rank the most-similar peers in the same master community by a weighted distance: price/m² (primary), median worth (separates villas from apartments), and the Ψ state. Peers that are stale (old last-sale) or thin(< 10 sales) are demoted, so a 2010-dated outlier can't masquerade as a current comparable. Unit-type comps are matched to the same room type.

Gross rental yield = median annual Ejari rent per m² ÷ median DLD sale price per m², resolved at the project (else area) level and applied per node. It differentiates by unit type — smaller units and offices typically show higher gross yields.

We publish our method and flag what's provisional — the opposite of an opaque "97%-accurate" black box. Where a number is an estimate, we say so.