⎯⎯⎯  ON THE PUBLIC RECORD  ⎯⎯⎯

How the verdict
gets its number.

We don't predict — we cite. Every figure on Verdict ties back to a registered deed you can name out loud. Here is exactly how each number is made.

⎯⎯ SOURCE
Dubai Land Department
Official deed registry
⎯⎯ DEEDS IN SCOPE
—
settled, registered
⎯⎯ AREAS INDEXED
—
distinct districts
⎯⎯ AS OF
—
refreshed daily · 06:00 GST
I.

The promise

Verdict is a registry-driven valuation tool, not a market estimator. Every claim on the site has a chain of evidence that ends in a settled, government-registered transaction. Here's what that means in practice:

  • We cite, we don't predict. No machine-learning forecast, no LLM «opinions» — only weighted aggregations of filed deeds.
  • No portal asks. We never read Bayut, Property Finder, or any listing site. Asking prices are user-provided; everything else is registry.
  • Auditable. Each comparable on a verdict can be checked against the official deed (transaction id and date are exposed).
  • Explicit about gaps. Where evidence is thin, we say so — sample size, recency, and confidence are part of every output.
II.

Source of truth

Data originates from the Dubai Land Department's public deed feed. We mirror it nightly into a Postgres warehouse and aggregate from there.

Refresh cadence

Daily at 06:00 GST. Numbers visible across the site reflect the most recent successful refresh — see the «As of» date in the source strip above.

Schema disclosure

Every metric on the site reads from one Postgres table, transactions. The columns we touch:

event_dateday of registration · used for windows, recency, freshness
area_name_endistrict · used for aggregation, search, comparables
project_name_enproject · prioritised in comparable matching
building_name_enspecific building · highest-priority comparable signal
rooms_enunit type (e.g. 1 B/R, Studio) · used for unit filter
reg_type_enExisting Properties or Off-Plan Properties
trans_group_enSales, Mortgages, Gifts · we filter to Sales
actual_worthtotal transaction price in AED
procedure_areaunit area in square metres
meter_sale_priceAED per square metre · primary unit-economics field
III.

Coverage & filters

Default scope across the site:

  • Sales only · trans_group_en = 'Sales'. Mortgages and gift transfers are excluded from price aggregations because they don't reflect arms-length pricing.
  • Both resale and off-plan · reg_type_en IN ('Existing Properties', 'Off-Plan Properties'). The off-plan share is shown explicitly per area so you can weigh structural mix.
  • Valid prices and sizes · meter_sale_price > 0 and procedure_area > 0. Records with missing or zero values are dropped silently.

Time windows used on the site

  • Verdict comparables · last 365 days. Older deeds get exponentially down-weighted (~18-month half-life).
  • Liquidity · last 90 days.
  • Trends league · last 12 calendar months for the trace; +1 month back to compute YoY.
  • Momentum / volume change · last 3 months vs the prior 3 months.
IV.

Glossary of metrics

Every metric on the site is defined here exactly once, with formula and where it appears.

Recorded fair · midpoint

⎯⎯ shown on the verdict report — /, Section II
weighted median AED/sqm of selected comparables

Weights combine recency (≈18-month half-life), size similarity (closer-to-subject deeds matter more), unit-type match, and source weight (resale vs off-plan).

Fair band (low — high)

⎯⎯ verdict report and Compare unified scale
midpoint × (1 − dispersion) … midpoint × (1 + dispersion)

Dispersion is observed in the comparable pool, bounded between 6% (very tight market) and 18% (sparse evidence).

Premium / discount %

⎯⎯ verdict delta chip and Compare decision banner
(ask − fair_mid) ÷ fair_mid × 100

Confidence score · label

⎯⎯ verdict report and Compare unit cards
composite of sample size, dispersion, and best-tier evidence

Score is reported 0–100; label is High / Medium / Low. High requires same-project evidence at sample size ≥ 8 with low dispersion.

Liquidity · 90d

⎯⎯ verdict and Compare unit cards
COUNT(*) WHERE event_date >= now − 90d

A measure of how often deeds settle in this area or project — proxy for market depth.

Momentum · 3 mo

⎯⎯ shown on /trends league and row detail
avg(last 3 months avg AED/sqm) ÷ avg(prior 3 months) − 1

Captures direction this quarter — distinguishes accelerating markets from cooling ones, where YoY alone would smooth over both.

Volume change · 3 mo

⎯⎯ trends league and detail
sum(deeds last 3 months) ÷ sum(deeds prior 3 months) − 1

Liquidity direction — rising volume suggests growing market depth, falling volume suggests buyer pullback.

YoY %

⎯⎯ trends league
(current month avg) ÷ (same month 12 months ago) − 1

Dispersion %

⎯⎯ trends row detail
(12-mo high − 12-mo low) ÷ current × 100

A high number means the area's price moved a lot over the year; low means it stayed flat.

Off-plan share

⎯⎯ verdict, Compare, trends
off-plan deeds ÷ total deeds × 100

Structural read: high share means the area is dominated by new-build sales; low share means most activity is resale.

Project share

⎯⎯ verdict report (header above evidence)
comparables in same project ÷ total comparables × 100 (computed client-side)

If a verdict's evidence is mostly in the same project, the answer is project-specific. Lower share means we widened to area-level evidence.

Walk-away

⎯⎯ verdict and Compare unit cards
fair_mid × 1.03

A 3% headroom over the registry's midpoint — a soft ceiling for negotiations.

Fees

⎯⎯ verdict report Fees block, Compare
Transfer 4% · Agency 2.1% (buy) or 2% (sell) · Other AED 5,000

DLD transfer fee plus agency commission plus a fixed-cost stub for NOC, trustee, etc. Marked 'Estimated' until observed against your specific transaction.

All-in cost

⎯⎯ verdict and Compare
ask + total fees

Percentile placement

⎯⎯ verdict report under delta
% of comparables priced at or below the asking price per sqm

«P78» means the asking price is at the 78th percentile — higher than 78% of comparable settled deeds, per sqm.

Verdict pill thresholds

Used on the home verdict report and Compare unit cards.

Premium / discount %
Pill
Reading
< −5%
PURSUE
Materially below the registry — actionable upside
−5% to +5%
WATCH
Within fair-band noise — neither cheap nor expensive
> +5%
PASS
Materially above the registry — overpriced relative to filed deeds

Momentum chip tiers

Used on /trends league.

3-mo momentum
Chip
≥ +5%
↑↑ strong up
+1% to +5%
↑ up
±1%
→ flat
−1% to −5%
↓ down
≤ −5%
↓↓ strong down
V.

Fair-price algorithm

When you file a verdict, this is what runs server-side:

  1. Filter base pool. Only Sales transactions in the last 365 days for the subject area, with valid price and size.
  2. Tier the comparables. Priority order: same building × same unit type > same project × same unit > same project, any unit > area × same unit > area-only fallback. The highest-priority tier with sufficient sample wins.
  3. Score each comparable. Weight = recency_weight × size_similarity × unit_match × kind_weight. Recency uses an 18-month half-life: a deed from 9 months ago carries roughly 70% the weight of a deed from this month.
  4. Compute the midpoint. Weighted median of AED/sqm across the pool — robust to outliers (a single penthouse won't pull the band).
  5. Compute the band. Observed dispersion (CV of weighted prices), bounded to 6%–18% spread above and below the midpoint.
  6. Score confidence. Larger sample, tighter dispersion, and higher-tier evidence → higher confidence. Reported 0–100 with Low / Medium / High label.
  7. Surface evidence. Top 5 comparables by similarity weight are returned with date, project, size, price, and similarity score so you can audit the answer.

Minimum sample. A verdict requires at least 6 comparable deeds after filtering. If the registry slice is too thin, the API falls back to a static comparables fixture and explicitly downgrades confidence to Low.

VI.

What we don't do

  • We don't scrape listing portals. No Bayut, Property Finder, or Dubizzle data is used. Asking prices are typed by the user; everything else is registry.
  • We don't use AI predictions. No LLM, no neural network, no «AI-powered estimates». Every figure is a deterministic aggregation of filed deeds.
  • We don't forecast the future. Trends show what already settled. Twelve-month traces are descriptive, not predictive.
  • We don't mix mortgages or gifts into pricing. Only arms-length sales (trans_group_en = 'Sales') feed the price aggregations.
  • We don't expose listing data. If a unit hasn't settled and registered, it doesn't exist in our numbers. No «sold but not filed» phantom inventory.
VII.

Limitations & honesty

Things we know we don't do well, in plain text:

  • No geographic radius yet. Project and building coordinates aren't in the source schema, so proximity comparables fall back to same-project then same-area evidence rather than a literal 1-km circle.
  • Off-plan assignments and pre-handover flips can be mixed inside the broader off-plan procedures bucket — the source schema doesn't expose them as a distinct type.
  • USD conversion is approximate. We use a hardcoded reference rate of 1 AED ≈ 0.272 USD, updated manually. USD figures shown across the site are for orientation only.
  • «UNKNOWN» building names appear in the registry feed. Where building is unknown, we display the project name to keep evidence rows readable.
  • Sparse-area fallback. If an area has fewer than 6 valid comparables in 365 days, the verdict falls back to a static reference fixture and the confidence label drops to Low.
  • Sqft is computed. The registry stores square metres. We display sqft as sqm × 10.7639 for comfort; any rounding error originates here.
  • Refresh failures. If a nightly refresh fails, the «As of» date in the source strip will lag — that's your signal that you're looking at older data, not a silent bug.