—
Pick an address to fill in size, age and energy rating from its certificate, or type the details. Blank fields are fine; the estimate is just less precise.
How far each kind of information moved the value from a typical Doncaster sale, worked out from the model's own trees (SHAP). Hover a bar for the detail.
Re-runs the model with the changes and compares the value gained with a 2026 guide cost (sources in Method).
Nearest sales of the same type in the last 3 years, moved to today's prices.
| Address | Today | Away |
|---|
Darker dots sold for more in today's money; the ring is this property.
Straight-line distances. Flood risk is the Environment Agency's rivers-and-sea rating for the postcode; check the address on the long-term flood risk service. None of these improved the estimate when tested, as nearby prices already reflect them, so they're here for your judgement.
Schools: DfE and Ofsted (31 March 2026). Stations: NaPTAN. Shops, parks, GPs and roads © OpenStreetMap contributors.
Land Registry sales of the address you picked, moved to today's prices. The last one is filled in for the model.
| Sold | Price | In today's money | Note |
|---|
A rough cross-check: where the band sits among similar homes in the neighbourhood, against where the estimate sits among recent sales. Bands date from 1991 values.
This home as an investment
—
—
No open data gives the rent of one home, so this links rent to value: pricier homes and areas rent for more, but by less than their prices.
—
Starts from the property on the first tab: the condition-adjusted estimate (or the asking price, if given) and the rent estimate above (or the rent from a pasted listing). Every input can be changed, and the figures above follow.
Before capital gains tax.
Monthly cash flow after tax if things go wrong, one at a time and all together.
| Scenario | Monthly cash flow | Change |
|---|
Monthly cash flow in year 1 at every rate from 0% to —. Drag the slider, or across the chart, to try a rate: the whole deal follows.
The 10-year IRR after tax, or the first year's monthly cash flow after tax, with each input lowered and raised by a set amount, one at a time, everything else as entered. The longest bars are the figures most worth checking.
Defaults are Doncaster's own last-10-year growth: rents from ONS, prices from this model's index. Sold at year 10 after selling costs, before capital gains tax.
| Year | Rent | Cash flow after tax | Property value | Loan | Equity |
|---|
—
Buys at your purchase price from the deal below, or, if you haven't given one, at the model's value for the condition the works put right; sells at the value after them, both from the repeat-sales calibration. Works are Yorkshire guide prices plus a contingency (sources in Method). A kitchen and a rewire grow only with the square root of floor area, each bathroom and the boiler are fixed amounts, and decorating, flooring, plastering and radiators go by floor area; kitchens and bathrooms are cheaper in cheaper homes unless you pick a finish. Finance is a bridging loan on part of the price: interest on the loan and an arrangement fee; the rest of the price, the works and the buying costs are your cash. Set the loan to 0% for a cash purchase.
As of October 2026. Check the source before relying on any of these.
Doncaster's 39 neighbourhoods (ONS middle-layer areas, named by the House of Commons Library). Try "Bessacarr", "Thorne", "DN4 5" or a full postcode.
—
Neighbourhood
—
—
| Measure | Here | Doncaster |
|---|
Pick a period with the slider or a quick button (or drag across Doncaster's line at the bottom). The map shows each neighbourhood's growth over it, from its repeat-sales index: blue grew more than Doncaster, red less. Switch to each year's change to see the years prices rose and fell. The bars race the neighbourhoods on growth over the period. Click a neighbourhood to see its figures.
—
—
—
Blue is better for a landlord, red is worse, grey is about the Doncaster average. Click a neighbourhood to see its figures. The ring is the property on the first tab; squares are railway stations.
Boundaries: ONS LSOA 2021 (super generalised), joined into neighbourhoods (MSOA 2021); contains OS data © Crown copyright and database right 2026. Neighbourhood names: House of Commons Library, Open Parliament Licence.
—
—
—
ONS average monthly rent in Doncaster.
ONS average rent against the Local Housing Allowance cap for Doncaster, 2026-27. Tenants on benefits get help up to the cap.
| Bedrooms | Average rent | LHA cap | Gap |
|---|
Quality-adjusted (the same mix of types and neighbourhoods each month), January 2015 = 100.
Postcode sectors split into thirds by their 2015-2019 median price, January 2015 = 100.
3-month rolling median of sale prices.
Standard-price sales registered with HM Land Registry; the latest month is still filling in.
Possession claims a landlord takes to court, per 1,000 privately rented homes, over the previous 12 months. Standard claims include rent-arrears cases; accelerated claims were the paper route for no-fault Section 21 notices, which ended on 1 May 2026.
Ministry of Justice mortgage and landlord possession statistics (county court claims, by the tenant's address), with Census 2021 privately rented households. Claims cover private landlords; accelerated claims and bailiff evictions can include a few social landlord cases. These figures exist for the whole borough only, not for neighbourhoods.
Tested on 5,995 sales from January 2025 to August 2026 that the model never saw; trained on 2015-2023 and tuned on 2024.
Median error on the same 2025-26 sales, each method using only sales at least two months old (by then most sales are published, though new builds and some others take months longer). Index methods need a previous sale, so they cover 72% of homes. The hedonic regression is the fair baseline: a linear model on this model's own features, with a fixed effect for each postcode sector, trained and tested the same way.
Each year valued by a model trained on sales up to two years before, with a price index rebuilt as it would have been known at the time.
The published test lets a sale take its size, rooms and build details from a certificate lodged up to 90 days after it, or from the home's first certificate if none came before. Here only certificates lodged by the sale date count. The 2025 deprivation scores and plot sizes from today's title plans have no dated versions, so the third row drops them altogether. That is a sensitivity test, not a measure of their timing: it also loses what they legitimately add. Each row is the average of three runs with different random seeds; the change is paired with the published rules seed by seed.
| Check | Sales | Median error | Change, range over seeds | Within 10% |
|---|
The headline is for homes whose address, certificate and last sale are known. Typed in by hand with fewer details, estimates are looser; the estimate panel shows the typical error for each home.
| What is entered | Median error | Price inside the 80% range |
|---|
2025-26 test sales with their other details blanked out; calibrated on 2024 (src/model_v8_typed.py).
A second small model predicts how far off each estimate is likely to be, from things like how many sales there are nearby and whether the last sale is known. Trained on 2020-23 errors, scaled on 2024, checked here on 2025-26. Homes typed in by hand are scaled separately.
The checks property-tax assessors use (IAAO ratio studies), on the ratio of estimate to sale price.
Median error on the same test sales as each layer is added.
1,500 random test sales; on the diagonal is a perfect estimate.
Test sales, by group. Land Registry registers some sales months late, new builds most of all: they are 4% of the 2025 test sales and none from February 2026, and August 2026 has 82 sales against a usual 300 to 400. New builds are the easiest homes to value, so the lag doesn't flatter the 2026 part of the test: January to August 2026 scores much the same as the same months of 2025. It will be scored again once Land Registry catches up.
| Group | Sales | Median error | Within 10% |
|---|
Share of the model's total gain, top 12 features.
Each change fitted 3 times with different random seeds; a change smaller than its seed-to-seed spread is noise. Negative means a lower (better) median error.
| Change | Median error | Seed spread | Decision |
|---|
Median gap between sale price and estimate on 2025-26 test sales, sectors with 40+ sales. Positive means homes sold for more than the estimate. The local correction shifts each estimate by half the model's median miss on the 40 nearest recent sales.
| Sector | Sales | Before | After correction |
|---|
Homes resold within 3 to 18 months are mostly bought, refurbished and resold. Against a model without last-sale information, they bought 7.6% below the estimate and sold 7.7% above it. Homes resold after 5 years or more also sold about 1.4% above it, at both ends, for reasons that aren't a quick refurbishment, so that is taken off. Each home's sales are kept in the same fold, so the model never learns a home's price from its own other sale. With a last sale entered the full model already knows part of it: a purchase in need of renovation then sits 4.7% below (on the 546 purchases that had an earlier sale, a median 10 years before), and a quick resale only 0.6% above, so those figures apply when the last sale is known (for "good" and "recently refurbished" only when it was within 18 months; works since an older sale aren't visible to the model, and keep the first column).
| Condition | Adjustment | Middle half, low | Middle half, high | With a last sale |
|---|
Every step only uses information that existed before the sale being valued.
Private to you. "Estimate now" re-runs today's model on the saved details. Open one to load it back into the value and buy-to-let tabs.
| Property | Saved | Asking | Estimate then | Estimate now | Asking vs now | Rent | Gross yield | Cash flow a month | Notes |
|---|
Every version of the dashboard, newest first.
This public copy leaves out house-level address data: the address lookup, sales history and saved list.