If you have built anything in UK proptech, you have almost certainly stood on Price Paid Data. HM Land Registry publishes every residential sale in England and Wales lodged for registration since January 1995, updated monthly, under the Open Government Licence v3.0.
It is free for commercial reuse. You can pull it as a bulk download, run the Standard Reports application to build a report for any area down to postcode sector, or query the linked-data endpoint at landregistry.data.gov.uk.
As a demand-side record it is genuinely strong, and a large share of the UK industry runs on it. The supply side is the half that sits somewhere else, and it is the half worth talking about.
What Price Paid Data covers, and where its scope ends
The official guidance is precise about scope. Price Paid Data is residential, and it excludes commercial property and land, sales not lodged with HM Land Registry, sales not for value, right-to-buy sales at a discount, transfers on divorce, gifts, compulsory purchase orders, court orders, leases of seven years or less, and prices below £100 or above £99,999,999.
Read that list as an engineer. A site sold as land falls outside the dataset by definition, and so does the company-to-company deal that usually moves one.
The second register is Commercial and Corporate Ownership Data, a monthly file of registered titles in England and Wales held by UK-incorporated companies and corporate bodies, each linked to its registered proprietor. It is an ownership index, so what it records is the holder of each title, with the consideration paid sitting outside its scope.
The third leg is planning. England has 311 local planning authorities, of which 73 currently feed planning.data.gov.uk, the national platform standardising planning data, with eight datasets being compiled from multiple local sources to give national coverage.
Historically each authority ran its own system and its own schema, so querying across authorities meant scraping. The direction of travel is obvious enough, and the coverage figure is the one to watch.
So three questions about any British development site, what the land cost, what it is permitted to become, and who else wanted it, are questions a UK model has to estimate.
Singapore publishes the supply side
Now the contrast case. Singapore sells most development land itself, through a programme called Government Land Sales, or GLS, and publishes the tender outcome in full, including every bid, every bidder, the site area, the plot ratio, and the maximum gross floor area fixed before the sale.
Work it on one parcel. The Bedok Rise site, in the east of the island, went to tender under the second-half 2025 programme, closed on 27 November 2025 with 10 bids, and was awarded on 2 December 2025.
Ten bids made it the most contested GLS residential tender of 2025. Bellis Residential, a vehicle of Allgreen Properties, which is the Singapore real-estate arm of the Kuok Group, took it at S$464.8 million.
When we read a tender result, we divide the winning bid by the maximum permitted floor area before looking at anything else, because land is the one input to a future launch price that is fully public. S$464.8 million over 32,470 sqm of permitted floor area is about $1,330 per square foot per plot ratio, which is the local unit for a normalised land cost.
The scheme going up on it is Bedok Rise Residences, around 380 homes on the official yield estimate, on a 99-year lease running from the award. The site is 20,293.6 sqm, or about 218,507 sq ft, at a plot ratio of 1.6, zoned fully residential.
The runner-up is the row with no British equivalent
Here is the line a UK team cannot buy at any price. Hoi Hup Realty bid S$462.8 million, or $1,324 psf ppr, about 0.4 per cent behind the winner.
That is a second independent valuation of the same asset, on the same day, by a different firm running its own cost model. It is published alongside the rest of the field.
Count what that hands you. You get a point estimate, a competing estimate, the spread between the two, and the number of firms that thought the asset was worth bidding for at all.
The demand side, and the sample-size trap
Singapore’s answer to Price Paid Data is the caveat record. A transaction is lodged and then published, so the file holds prices actually paid rather than prices asked.
The trap is the row count. When we compare two medians, we check how many caveats sit behind each one before quoting either, because three caveats and 411 caveats are not the same evidence.
Take Bagnall Haus, a freehold project of 113 units in the same district. Its median is $2,494 per square foot across 91 caveats covering the whole launch, from January 2025 to April 2026, and it has to be quoted that way because the trailing 12 months hold only six rows.
A 12-month median computed on those six rows would then be printed with the same confidence as one computed on 411. Sceneca Residence shows the same thing.
Its recent record holds three caveats at a median of $1,778, against a launch-era median of around $2,080. For scale, Pinery Residences sits at a median of $2,537 on 411 caveats, and Vela Bay at $2,861 on 385.
The file also settles an argument that usually gets conducted on instinct. Across District 16 resales over the 24 months to August 2026, freehold and leasehold prices sat within two per cent of each other, and freehold Bagnall Haus at $2,494 sits slightly under leasehold Pinery Residences at $2,537.
Those per-square-foot figures come from URA caveats read on 15 September 2026.
Both files trail, and most dashboards hide it
This is the paragraph a UK technical reader should recognise from their own stack. Price Paid Data is captured at registration, typically two weeks to two months after the sale, which is why the most recent two months of it are incomplete.
Singapore’s caveats lag by weeks for the same structural reason. The record is a byproduct of a legal filing, and filings arrive on their own schedule.
Both registers trail. In a moving market that means the rows you have fewest of are the rows that matter most, and a median that prints without its lag and its row count beside it is lying by omission.
When we hand a buyer a computed figure, we put the read date and the sample count on the same line as the number, because a figure with nothing attached to it travels much further than it deserves to.
Where both records stop
The boundary is identical in both countries, and it sits at the front door of the building. Land registers describe parcels and completed transactions, not the individual home a buyer is choosing between.
The query stops there.
Unit mix, layouts, and the area of each apartment are developer documents, issued at preview as a brochure and a set of drawings. A Bedok Rise Residences floor plan is a PDF that a person reads, and the last mile of any valuation model is exactly that, in Britain as much as in Singapore.
A median price per square foot is an average across five things at once, which are stack, floor, facing, layout efficiency, and launch phase. The open files carry the price and the area without any of the five.
That is not an argument for scraping harder. It is an argument for drawing the line on the diagram and labelling it honestly.
What transfers back to a UK model
Three habits carry most of the value here, and none of them depend on which register you happen to have.
First, compute only what the files support, and estimate the rest out loud, in a field a reviewer can see. A land cost per buildable square foot is a floor and a shape, not a price, because construction, finance, marketing, and margin all stack on top of it.
Second, print the row count and the read date beside every median you publish. A thin sample and a deep one look identical once they reach a chart.
Third, mark the exact point where the open record ends and a private document begins. On this Singapore parcel that point is the front door, and in England and Wales it sits earlier, somewhere before the site changes hands.
So the useful question for a UK team is not whether Britain ought to publish tender bids. It is how much of your current model rests on an estimate that another market simply looks up, and what you would build differently if that one file arrived tomorrow.