,

Methodology & model assumptions

Geology & terrain

Each region carries a simplified but real stratigraphic stack (Coal Measures and Magnesian Limestone in the North East; gritstone and Coal Measures in the Pennine regions; chalk, boulder clay and fen in the East; the Weald sequence in the South East; granite and Devonian slate in the South West). Strata are projected along a regional dip; relief comes from differential erosion of those units. The geology is three-dimensional: a folded, faulted, sometimes domed structural surface carries a stratigraphic column of formation thicknesses, and the outcrop anywhere is the formation found where today's land surface intersects that column. Valleys therefore cut down to older rocks (inliers), resistant caps survive on hills, and outcrop boundaries make V-patterns across valleys, exactly as on a real geological sheet. Relief and outcrop are co-evolved through differential-erosion passes, so scarps and vales emerge from the resistance feedback rather than being stamped on. Massif fronts are lobed with spur ridges; in the three northern regions, glacial dales with U-shaped cross-profiles are carved radially from the upland and the drainage then finds them. Coastlines are erosion-controlled: bays bite into soft strata and resistant rock stands out as headlands, which is also why an all-sea "island" sheet comes out as an irregular massif rather than a square. Drainage is derived by priority-flood depression filling and D8 flow accumulation over an orographic rainfall field; requested rivers carry an implied off-map catchment so they always run.

Settlement & names

The main town is sited the way English towns actually were: at the lowest bridging point of the major river, at a sheltered harbour, or on a dry spring-line site, with an origin story (Roman fort, minster, burh, Norman plantation) that constrains its name. Place names are assembled from Old English, Old Norse, Brittonic and Norman elements with regionally correct frequencies (−by/−thorpe in the Danelaw, −combe and Tre−/Pen− in the South West), and suffixes conditioned on the site (−ford at crossings, −mouth at estuaries). River names use Brittonic roots, as most real English river names do.

Growth

The town is grown era by era, and not as concentric rings: each settlement carries persistent Hoyt-style sector preferences, so growth runs out in wedges; Victorian industry goes to flat land by water and rail on the north-eastern, downwind side (the prevailing wind is south-westerly, which is why English west ends are the nice ends); terraces pack around the works and villas take the upwind hills. Interwar semis ribbon hard along the arterial roads; a green belt is designated around the main town in the 1950s, so post-war estates leapfrog it; late-century cul-de-sacs pull toward the motorway junction; the 2000s add centre flats and dockside conversion. Roads are routed over a cost surface with field-scale noise, so they wander the way enclosure-era lanes do, and every settlement is knitted into a lane web; railways are routed at ruling gradients before the motorway exists, and a share of branches close in the 1960s and persist as green corridors. Zoom in on any built-up area for era-conditioned street grain: tangled medieval cores, gridded Victorian terraces, curling estates.

Streets and people

Streets are the growth medium, not decoration: each era first extends a settlement's street graph with era-specific geometry (organic medieval lanes, bye-law grids snapped to the main-road bearing, curving C20 suburbs), and only land served by a street becomes developable, so the drawn network is the structure the town actually grew on (99% of the housing stock ends street-served). The population is a household-level microsimulation: roughly one household per 2.3 residents is synthesized into the stock, with a socio-economic segment drawn from the housing it occupies and an ethnicity drawn from regional census-style priors. Where groups end up living is then NOT painted: a few rounds of residential choice with a chain-migration preference (the mechanism behind real English ethnic geography) produce emergent clustering, calibrated so a Pennine textile borough develops inner wards around 40-55% single-community, as the census shows, while a rural county stays dispersed. Every working household is matched to actual sector-resolved jobs by segment preference and distance decay, and the commuting matrix the traffic model loads is the aggregate of those real matches. Ward statistics are aggregations of this simulated population.

Wards

Ward boundaries are grown by a barrier-weighted geodesic flood from urban seeds and parish centres: rivers, railway lines and main roads are expensive to cross, so boundaries snap onto them, as real English ward boundaries do. Urban wards target ~5,000 residents; each ward also carries an idiosyncratic small-area term, because adjacent real wards differ more than any smooth model predicts.

Economy

Every job is allocated to one of ten sectors from its land use. Contiguous industrial and commercial clusters become named firms with period- and region-appropriate names and founding dates (worsted mills and rolling mills where the geology says coal and wool; science parks by the late-century campus), with peak-employment figures for the old staples. The borough also gets its institutional roster: cathedral or minster where earned, universities, the NHS trust, a football club at a plausible league tier, a newspaper, market day, reservoir, sewage works, all consistent with what the growth model actually built.

The labour market closes as an accounting identity: resident workers split into internal and outbound commuters, local jobs into internal and inbound, with the two margins sharing the internal term, so residents equal internal plus outbound and jobs equal internal plus inbound exactly. Workplace capacity is a hard constraint in the match; a worker who finds no cell with headroom commutes out. The inbound flow is sized to the residual jobs (about a quarter, as in real English boroughs), and both inbound and outbound movements are routed through the sheet-edge gateways in the traffic assignment. Primary-school provision responds to the modelled roll: sites are added in the most child-heavy districts until planned capacity covers the roll with a 17% surplus, matching England’s roughly 10–15% spare places plus bulge headroom; persistent local overflow triggers additional provision, and children assign to the nearest school with places.

Demography

Ward profiles are derived from the housing stock the growth model actually built, seeded with 2021-census-calibrated regional priors. Minority communities concentrate in inner Victorian terraces near former industry (South Asian communities in the Pennine textile belt, Indian communities in Midlands manufacturing towns), student belts and centre flats; outer estates and villages run whiter than the borough average. Deprivation, tenure, unemployment, car ownership and degree share co-vary with the stock; life expectancy applies the real English IMD gradient (~8–9 years for men across deciles) to regional baselines.

Traffic & environment

The origin-destination matrix is aggregated from the actual household-to-job matches of the microsimulation, with car availability by segment and ward. Assignment is by congested shortest paths with the BPR volume-delay function, iterated; a v/c above 1 on an inner link should be read as queueing demand. Rail loading comes from station catchments against a typical peak seat supply. Crime is generated per ward from the emerged composition (unemployment, affluence, night-time economy, students), scaled to the England average of roughly 85 police-recorded offences per 1,000 residents per year. NO₂ is background + urban increment + a traffic kernel (the dominant term in English monitoring data) with industrial plumes trending north-east; noise is an Lden-style proxy from roads, rail and industry; biodiversity scores habitats in a UK style, where estuary, marsh, old woodland and reclaimed colliery spoil score high, and improved arable scores low.

External traffic

The network is not closed to the sheet: every classified road reaching the edge is a gateway. Roughly a quarter of the borough's jobs are filled by in-commuters entering through those gateways (capped at each gateway's approach capacity), and strategic through-movements run between motorway and trunk gateways, so the motorway and radial A-roads carry traffic that has nothing to do with the town, as they do in reality. Each ward records where its own workers actually end up, shown in its profile.

Biodiversity

Habitat value is set by land cover and use, then modified by landscape configuration: riparian buffers along watercourses, hedgerow field margins in the farmed matrix, woodland interior versus edge, and a connectivity term so isolated green in a built-up matrix scores below the same habitat embedded in a green network. The best-connected patches are promoted to named designations at County Wildlife Site, Local Nature Reserve or SSSI tier by area.

Public transport & cycling

Bus provision is modelled at ward level as buses per hour, following density and position on the radial classified corridors; high provision suppresses car commuting in the mode model. Cycle-to-work share follows the census pattern: ~3% nationally, rising with flat terrain and student population.

Calibration & validation

Large-city ethnic composition is additionally anchored to published Census 2021 metropolitan-borough figures, solved in person space so that census-true differences in household size and age structure between groups do not bias the realized shares; verified mean absolute error against the regional anchors is under 2 percentage points.

The raw authority table, the deterministic train/holdout split, the 19-to-10 group concordance and the fitting script ship alongside this file in the calibration bundle, so the headline figures below can be re-derived independently.

Most model constants are hand-tuned so that batches of generated counties roughly match published England aggregates (ONS workplace employment, GVA by sector, census travel-to-work, police.uk ward crime rates, the ONS life-expectancy gradient). That is plausibility tuning, not statistical calibration, and every indicator on the sheet should be read as a modelled index for a fictional place.

One family of outputs is statistically fitted: district ethnic composition. The region-and-class priors are fitted to ONS Census 2021 ethnic-group profiles (TS021, 19+1 classification, aggregated here to ten groups) for 23 English local authorities, split deterministically by authority code into 14 training and 9 holdout authorities. Against the holdout, generated borough compositions show a mean absolute error of 1.6 percentage points per group, and real values fall within the 3-seed ensemble envelope for 69% of group-cells. Caveats, stated plainly: one model defect was diagnosed by inspecting holdout errors, so the holdout is not perfectly clean and these are best read as development-set scores; the reference sample is 23 of 331 authorities, with no London boroughs and only one mill town, so high-diversity cities are under-generated (synthetic ~73% White British where a Nottingham-class city is nearer 57%); and the fit covers this one variable family, not the full multivariate programme. Cells with no region-class example use nationally pooled priors.

Typical measured outputs across a generated batch, for orientation: mean home-work separation ~5.8 km straight-line; commuting car share ~55% (census: ~60% outside London); urban male life-expectancy span ~8 years (ONS decile gradient: 7-10); ward crime rates ~50-160 per 1,000 (police.uk: ~40-200); peak single-community ward shares of 35-45% in textile regions, as in Bradford or Blackburn.

Limitations

A specialist would recognise simplifications at every stage. The advertised price-to-composition coupling is a single post-hoc affordability adjustment applied after households are settled; it does not move households or feed back into demographics, income, tenure or voting. Hydrology is D8 flow on a coarse grid, recomputed in full after estuary widening converts land to tidal water so the network drains to the final coastline. Traffic runs a short method-of-successive-averages loop, not a solved equilibrium; on hard networks it can fail to converge, and the diagnostics say so rather than pretending otherwise. Pollution is a kernel proxy, not dispersion modelling. The street network is a mid-scale growth skeleton, routable by the traffic model but not parcel-faithful; there is no block subdivision or cul-de-sac plan. Households are a settled cross-section with no lifecycle: no births, deaths or moves through time. The claim throughout is not precision but that the couplings (geology to industry to housing to travel to health) run in the right direction with roughly the right magnitudes.

References

ONS, Census 2021, Ethnic group (TS021), local authority tables via the Ethnicity Facts and Figures service, retrieved 18 July 2026 (calibration and holdout data).
Barnes, Lehman and Mulla (2014), Priority-Flood: an optimal depression-filling and watershed-labeling algorithm, Computers & Geosciences 62 (drainage preprocessing).
O'Callaghan and Mark (1984), The extraction of drainage networks from digital elevation data, CVGIP 28 (D8 flow routing).
Wardrop (1952), Some theoretical aspects of road traffic research, Proc. ICE (equilibrium concept); Bureau of Public Roads (1964) volume-delay function; assignment by the method of successive averages.
Department for Transport, National Travel Survey 2024 (trip-rate and mode-share magnitudes).
MHCLG, English Indices of Deprivation (concept for the deprivation index; the synthetic index is not the official IMD).
Zipf (1949) rank-size regularity for the settlement hierarchy.
Historic England, Historic Landscape Characterisation programme (field-system and enclosure patterning).
Choupani and Mamdoohi (2016), Population synthesis using iterative proportional fitting, Transportation Research Procedia, doi:10.1016/j.trpro.2016.11.078 (population-synthesis background).
Boeing (2017), OSMnx: new methods for street networks, CEUS 65, doi:10.1016/j.compenvurbsys.2017.05.004 (network-metric concepts).
ONS, Census 2021 built-up-area characteristics, ons.gov.uk (settlement-size context).
Ethnicity Facts and Figures service: ethnicity-facts-figures.service.gov.uk/uk-population-by-ethnicity (source CSVs).
Priority-Flood: doi:10.1016/j.cageo.2013.04.024. National Travel Survey 2024: gov.uk/government/statistics/national-travel-survey-2024. English Indices of Deprivation: gov.uk/government/collections/english-indices-of-deprivation. Historic Landscape Characterisation: historicengland.org.uk/research/methods/characterisation.