Project cards: sales history, launch-to-latest prices and unit mix (module project_detail)
Script: scripts/modules/project_detail.py. Data through 2026-11-05 (last complete month October 2026, last
complete quarter 2026-Q3). All inputs are Dubai Land Department (DLD) open data published on data.dubai under the
Dubai Open Data Licence:
| DLD dataset | Staging table / derived file | Used for |
|---|---|---|
| Real Estate Transactions | transactions |
residential sales per project (project_number) |
| Real Estate Transactions, Building and Property Project Records, Real Estate Projects register | enriched residential basis of the sales_market module (transaction-level, internal file) |
market (Ready / Off-plan), imputed villa / townhouse segment and bedrooms, estimated built-up area per sale |
| Real Estate Projects register, Registered Real Estate Units, Building and Property Project Records | public project table of the supply_stock module (supply_stock/projects.csv) |
project labels, status, register dates, planned units, withheld_brand_name |
| Registered Real Estate Units, Building and Property Project Records | units, buildings_raw |
unit mix per project (registered apartments, villas and stacked townhouses) |
Nothing here comes from a commercial data provider. The module runs after sales_market and supply_stock and
rebuilds every output from the read-only inputs.
1. Sales basis
A sale is counted for a project when the DLD transaction carries that project_number. The basis is the enriched
residential basis of the sales_market module (sales_market.md, section (b)): title-deed and Oqood sales of
apartments, villas and townhouses with a residential usage, plus villa sales recorded with another usage and houses
that DLD registers on their plot (type Land) inside villa projects (bulk plot registrations, which are not house
sales, are left out as in sales_market.md). Every sale passed the staging date and price checks
(q_date_ok, q_price_ok). Sales without a project number (about 17% of residential sales, mostly older buildings
outside the off-plan projects register) are not attributed to any project. The window ends with the latest complete
month; developer-level "Development" procedures are included, as in sales_market.
Difference from supply_stock/projects.csv. That table's sales_12m and median_aed_sqft_12m rest on another
basis, so the two public tables differ in both directions and must not be reconciled one to one:
supply_stockcounts every sale registered under the project number, including commercial units (shops, offices), hotel units and plots (bulk plot registrations too), and does not requireq_price_ok; it excludes "Development" procedures. This module counts residential sales only (the enriched basis above) and includes "Development" procedures. In the build of 9 November 2026 the 12-month counts are equal for about 1,890 projects, lower here for about 400 (commercial units in the project, e.g. a tower with shops or offices) and higher here for about 40 (Development procedures).supply_stocktakes its median per sq ft on the DLD unit area (plot area for houses registered on their plot) with the staging area check only; this module prices on the built-up area with the 100-50,000 AED per sq ft bounds, so it drops implausibly cheap registrations and, for villa projects, gives a built-up-area price that is far below the plot-area figure. The medians differ by more than 5% in roughly one project in ten.
The site shows this module's figures in the projects table and in the card, so the row and the card agree; the
downloadable supply_stock table keeps its own basis, documented in supply_stock.md.
- Market.
Ready= title-deed sales (and plot-registered houses in projects that the register shows as FINISHED by the sale date);Off-plan= Oqood registrations (and plot-registered houses in unfinished projects);All= both, plus the few plot-registered houses whose market could not be determined. - Bedrooms come from the DLD room label, imputed for houses from the DLD buildings register when missing
(
bedrooms_imputedofsales_market). Buckets:studio,b1,b2,b3,b4,b5plus,unknown;all= every sale. - Price per sq ft is the sale price divided by the built-up area: the DLD unit area for apartments and stacked
townhouses, and the estimated built-up area from the DLD buildings register for villas and plot-registered houses
(
bua_sqft_est; the headline "Built-up area (est.)" basis of the price index). A sale enters the per-sq-ft statistics only when its built-up area is between 100 and 100,000 sq ft and the result is between 100 and 50,000 AED per sq ft; the sale still counts inn_sales. - No outlier trimming. Project cells are small, so the medians and quartiles are taken over all measured sales of the cell (the price index trims by a log-IQR rule inside much larger community cells). The same untrimmed basis is used in every table of this module, so the summary, the quarterly table and the bedroom table agree.
- Median prices (
median_price_aed*) are published to the nearest 1,000 AED: in a small cell the median is the exact price of one transaction, and the project level does not restate a single deal to the dirham. Medians and quartiles per sq ft and sizes are whole numbers.
2. Quarters and windows
- Quarters are calendar quarters (
YYYY-Qn). - Latest complete quarter = the quarter of the latest complete month when that month is the last month of its
quarter (for example September 2026 → 2026-Q3); otherwise the quarter before. In this build the latest complete month
is October 2026, the first month of 2026-Q4, so the latest complete quarter is still 2026-Q3 and October 2026 sales
are not in
sales_quarterly.sales_quarterlyand the latest window end with this quarter; the summary's month-based figures (last_sale_month, the 12-month columns) run to the latest complete month. - Last 12 complete months (
12m): 2025-11-01 to 2026-10-31. - Other summary windows (
1m,3m,6m,24m): the latest complete month alone (2026-10) and the last 3, 6 and 24 complete months, also ending with the latest complete month: 2026-08 to 2026-10, 2026-05 to 2026-10 and 2024-11 to 2026-10. The site's projects table lets the reader choose among 1, 3, 6, 12 and 24 months. In a single month most projects have fewer than 5 sales, so the 1-month count is often empty and the 1-month median is published for a small number of projects only. - Launch window = the project's first two calendar quarters that have sales (any market), pooled.
- Latest window = the last four complete quarters ending with the latest complete quarter (2025-Q4 to 2026-Q3).
3. Project summary (summary)
One row per project of supply_stock/projects.csv (3,783 rows: the 3,474 register projects and the projects known
only from the units register or from transactions). Key project_number (empty for projects known only from the units
register; project_id is also carried, and the site joins on project_number when present, else project_id).
| Column | Definition |
|---|---|
project_number, project_id, project_name_en, master_project, area_name_en, status |
As in supply_stock/projects.csv (cleaned place labels, register status or NOT_IN_REGISTER). |
withheld_brand_name |
True when the project's name carries a developer brand (supply_stock.md, section 1). Every sales-derived column below is then empty; see section 7. |
first_sale_month, last_sale_month |
Month (YYYY-MM) of the project's earliest and latest residential sale, any market. |
first_offplan_sale_month |
Month of the earliest off-plan sale. |
n_sales_all_time, n_sales_offplan_all_time |
Residential sales since January 2008 (the start of the sales basis), all markets and off-plan only. |
n_sales_12m, n_sales_12m_offplan |
The same in the last 12 complete months. |
offplan_sales_per_unit |
n_sales_offplan_all_time divided by
homes_for_ratio. Two decimals. Empty when the denominator is missing or zero, when fewer than 5 off-plan sales exist, or when the ratio is above 2 (the plan then does not describe what was sold). Upper bound of the share sold, see section 4. |
homes_for_ratio |
Denominator of offplan_sales_per_unit, a register fact: the larger of the register plan (no_of_units + no_of_villas + no_of_lands; villa projects often register their homes as plots) and what is already registered for the project (reg_units + reg_villa_records of supply_stock/projects). Empty when both are zero or missing. |
launch_aed_sqft |
Median AED per sq ft over the launch window (first two quarters with sales); at least 5 measured sales, else empty. |
latest_aed_sqft |
Median AED per sq ft over the latest window (last four complete quarters); at least 5 measured sales, else empty. |
price_chg_since_launch_pct |
latest_aed_sqft / launch_aed_sqft − 1, in percent, one decimal; empty when either side is empty. Not quality-adjusted: the mix of homes sold at launch and recently differs, and for projects launched in 2008-2010 the "launch" quarters are dominated by late registration of earlier contracts. |
median_aed_sqft_12m, median_price_aed_12m |
Median AED per sq ft (at least 5 measured sales) and median price (at least 5 sales, nearest 1,000 AED) in the last 12 complete months. |
n_sales_1m, n_sales_3m, n_sales_6m, n_sales_24m |
Residential sales in the latest complete month and in the last 3, 6 and 24 complete months (section 2), all markets; counts of 1 to 4 are empty. |
median_aed_sqft_1m, median_aed_sqft_3m, median_aed_sqft_6m, median_aed_sqft_24m |
Median AED per sq ft in the same windows; at least 5 measured sales, else empty. Same price basis as the 12-month median (built-up area, no outlier trimming). |
handover_delay_months |
FINISHED projects: whole months from project_end_date to completion_date (negative when completed early; 0 when the register carries the same date in both fields, which is the usual case). Unfinished projects whose project_end_date lies before the latest complete month: months overdue as of that month (positive). Otherwise empty. A register fact, published for every project. |
Sale months are published only when the count behind them is published (at least 5 sales; 5 off-plan sales for
first_offplan_sale_month), because a date of one to four sales describes individual transactions.
4. The upper-bound caveat of offplan_sales_per_unit
DLD registers a developer's first sale of an off-plan unit and a later off-plan resale of the same unit with the same
procedures (Oqood registration), and the open transactions file carries no unit identifier and no seller type. A unit
sold by the developer and resold twice before handover therefore counts three off-plan sales. The ratio is thus an
upper bound of "share of the project sold": it can exceed 1 (538 of 1,951 published ratios do; the median is 0.86),
and it rises over time while a project is traded off-plan. There is no initial-versus-resale split anywhere in this
portal for the same reason (sales_market.md, section (e)). Ratios far above 1 also occur where the register's planned count does
not describe what was sold: re-planned projects and projects whose units register is incomplete. Plots are counted in
the denominator, so villa projects that register their homes as plots get a meaningful ratio; ratios above 2 are not
published.
5. Quarterly sales (sales_quarterly) and bedrooms in the last 12 months (sales_bedrooms_12m)
sales_quarterly: one row per project, calendar quarter, bedrooms bucket (all plus the seven buckets) and market
(All, Off-plan, Ready), full history since the project's first sale up to and including the latest complete
quarter (section 2): n_sales, median_price_aed (nearest 1,000 AED), median_aed_sqft, p25_aed_sqft,
p75_aed_sqft, median_sqft (median built-up area of the sold homes). Sales of the months after the latest complete
quarter (one or two months when the latest complete month is the first or second month of a quarter) are not in this
table, so no row describes a partial quarter under a full quarter's label; the module stops with an error if the last
quarter of the table is not the latest complete quarter. Rows with fewer than 5 sales are omitted, not emptied, so
the table has no suppressed rows; a per-sq-ft or size statistic resting on fewer than 5 measured sales is empty while
the count stays. 121,008 rows in the build of 9 November 2026 (26,672 of them all / All).
sales_bedrooms_12m: one row per project and bedrooms bucket (all plus buckets) over the last 12 complete months:
n_sales, median_price_aed, median_aed_sqft, median_sqft, offplan_share_pct (share of the sales that are
off-plan, one decimal). Rows with fewer than 5 sales are omitted. 5,117 rows.
The site file carries the all / All quarterly series (count, median AED per sq ft, median price) and the full
bedroom table; the CSV files carry everything.
6. Unit mix (unit_mix)
From the DLD units register and the villa records of the buildings dataset, not from sales, so it is published for
every project including brand-named ones. Units are classified exactly as in supply_stock.md, section 4 (the two
modules share the classification code): apartment = flats and units (kind apartment); villa_townhouse = villa
records and stacked townhouses (kinds villa and townhouse). Hotel, commercial, labour and unclassified units are
not in this table. One row per project_id, kind and bedrooms bucket: n_units (1-4 → empty), median_sqft,
p25_sqft, p75_sqft of the unit size (empty when fewer than 5 units have a plausible size; sizes as in
supply_stock: unit area 150-30,000 sq ft, villa built-up area 500-60,000 sq ft). project_number comes from the
project table and is empty for projects known only from the units register. Summing n_units over the two kinds
reproduces the res_* columns of supply_stock/projects.csv, except where a 1-4 cell of one kind is suppressed here
but absorbed into the project total there. 10,161 rows.
7. Privacy, suppression and the brand rule
- No company names: labels are the already cleaned place labels of
supply_stock/projects.csv; the other tables carry identifiers only. - No unit-level rows and no join of sales to units or rents; sales and the unit mix meet only at project level.
- Counts of 1-4 are empty; statistics from fewer than 5 records are empty; in the sales tables rows under 5 sales are
omitted. There is no secondary suppression (
00-publication-compliance.md, section 6). - Brand rule (
00-publication-compliance.md, section 4; owner decision of 7 October 2026, until data.dubai answers the open question): projects whose registered or published name carries a developer brand (withheld_brand_name = true, 823 projects) have no sales-derived value insummaryand no rows at all insales_quarterlyandsales_bedrooms_12m; their sales stay inside community and area totals of the other modules. Identifiers, labels, status,handover_delay_monthsand the unit mix (register facts) are shown. The publication scanner (scripts/common/compliance.py) checks every public table keyed byproject_numberagainst the flag in the project table and reports any sales-derived value of a withheld project; its self-test covers the check. - Exception inside brand-named master communities. The flag is inherited from
supply_stock/projects.csv, and the rule skips a leading official DLD place name that carries a brand (BRAND_PLACE_NAMESinscripts/common/compliance.py: DAMAC Hills, DAMAC Hills 2, DAMAC Lagoons, Sobha Hartland, Emaar South, and the brand-free Jumeirah Golf Estates, Dubai Hills Estate and Dubai Investment Park). A sub-community or phase of such a master community (DAMAC Hills 2 - <cluster>,DAMAC Lagoons - <cluster>,Sobha Hartland <phase>) is therefore published with its full sales history here, although its label starts with the brand word. These rows can be summed by that word and rebuild part of the developer's volume inside those master communities (in the build of 9 November 2026, about 135 such projects with about 5,600 residential sales in the last 12 months). This is the documented owner choice for official place names, pending the data.dubai answer; the list is limited to names that exist as DLD area, master-project or master-community names, and the scanner self-test fails when an entry does not. Project-series names that DLD uses only inside project names (a "<Brand> <Name> 1 .. 60" series) were removed from the list on 9 October 2026, which withheld 85 further projects.
8. Known limits
- The projects register is refreshed less often than the transactions; in this build its snapshot is of 2026-11-01,
the transactions snapshot of 2026-11-06 (sales through 2026-11-05). Projects launched after the register snapshot
are
NOT_IN_REGISTERwith no status, dates or planned units, so they have nooffplan_sales_per_unitand nohandover_delay_months, andACTIVEprojects completed after it may show an overdue delay. In this build no project with sales is newer than the register (its highest project number, 4,615, is above every number in the transactions); the 309NOT_IN_REGISTERrows are mostly older projects that are no longer in the register. The lag grows with every monthly run until the register is republished on data.dubai. project_end_dateis the developer's schedule filed with DLD and is re-filed when projects are re-planned; for most FINISHED projects the register carries the completion date in both fields, so the delay reads 0.- Project numbers group sales as DLD assigns them: a phase or a tower can be its own project, or several buildings can share one number.
- The estimated built-up area of houses is a register match or a project median (
sales_market.md); its errors pass into the per-sq-ft figures of villa projects. - Medians are not quality-adjusted; the change since launch mixes price movement with a changed mix of sold homes and the counting of resales.
- The launch window is the project's first two calendar quarters with sales, whatever their length: for a project
whose first sale falls in the months after the latest complete quarter, those months are its (so far partial) first
quarter, and
last_sale_monthcan lie after the last quarter ofsales_quarterly.
Outputs
data/out/project_detail/: summary, sales_quarterly, sales_bedrooms_12m, unit_mix (each as .parquet and
.csv). Site payload: site/data/project_detail.json (summary without the label columns, the all / All quarterly
series with the sales count and the median per sq ft only, because the site's quarterly chart draws just those two (the
quarterly median price stays in the CSV), the bedroom table and the unit mix; meta.windows gives the first and last
month of the 1-, 3-, 6-, 12- and 24-month windows; meta.notes records any trimming applied to stay within the size
budget of 2 MB per site file, none in this build, where the file is about 1.86 MB, so roughly 240 KB are left, about
seven quarters of growth). The first trim would drop median_aed_sqft_24m from the site payload, keeping it in the
CSV; the site then no longer offers the 24-month period, and scripts/qa/static_checks.py fails on it ("all five
periods"), so run it before publishing.