Where every number comes from.

Twenty-five-year time series (2000-2024) backing the layer's time slider. Values are interpolated from a network of weather stations onto a uniform 1km grid: estimates, not direct observations, and less certain where stations are sparse (upland/remote areas). NoData over sea and outside the UK land mask. Processing: extracted the `tas` variable per year, reprojected EPSG:27700 -> EPSG:3857 (bilinear), pre-colored with a fixed sequential ramp (2-14 degC, same scale across all years so they're visually comparable) via gdaldem color-relief, tiled to a raster PMTiles pyramid (zooms 2-7) per the approach confirmed in docs/DECISIONS.md #1. One PMTiles file per year; see pipeline/layers/haduk_annual_temp.py. UK-wide yearly series (added 2026-10-06, docs/DECISIONS_LOG.md #151): the Met Office's UK areal series of annual mean temperature from 1884 (https://www.metoffice.gov.uk/pub/data/weather/uk/climate/datasets/Tmean/date/UK.txt; file header "Areal values from HadUK-Grid 1km gridded climate data from land surface network. Source: Met Office National Climate Information Centre"), drawn as stripes coloured by difference from the 1961-1990 mean. The text file states no licence itself; the HadUK-Grid datasets it derives from are published under the Open Government Licence v3.0 (HadUK-Grid datasets page, read 2026-10-06), so it is treated the same: FLAGGED for the owner to confirm in the licence audit.

Twenty-five-year time series (2000-2024), same structure as the temperature layer. Variable is `rainfall` (standard_name lwe_thickness_of_precipitation_amount), annual sum in mm. Values across the years ranged ~317-6721mm (heavily right-skewed by mountainous areas: Lake District, Scottish Highlands, Snowdonia), matching CLAUDE.md's "hundreds to thousands of mm" plausibility rule. Same caveats as the temperature layer (interpolated from station data, no data over sea). Processing: gdaldem color-relief with a fixed sequential Blues ramp (500-3000mm, ColorBrewer, colour-blind safe) so years are visually comparable; see pipeline/layers/haduk_annual_rainfall.py. UK-wide yearly series (added 2026-10-06, docs/DECISIONS_LOG.md #151): the Met Office's UK areal series of annual total precipitation from 1836 (https://www.metoffice.gov.uk/pub/data/weather/uk/climate/datasets/Rainfall/date/UK.txt), same source, header, licence reasoning and flag as the temperature series in haduk-grid-tas-ann; stripes show each year's difference from the 1961-1990 mean, as a share of that mean.

Variable `sun`: hours of bright sunshine, summed over the year ("units: hour", cell_methods "time: sum"). Meaning and units were read from the NetCDF variable attributes of the 2024 file, not assumed. Same pipeline as the temperature layer (pipeline/layers/haduk_extra.py): fixed colour domain across years, bilinear reprojection to EPSG:3857, pre-coloured raster PMTiles (zoom 2-7), one file per year.

Variable `groundfrost`: number of days with ground (grass-level) minimum temperature below the threshold (variable attribute: "Days of ground frost", label_units "days"). Meaning and units were read from the NetCDF variable attributes of the 2024 file, not assumed. Same pipeline as the temperature layer (pipeline/layers/haduk_extra.py): fixed colour domain across years, bilinear reprojection to EPSG:3857, pre-coloured raster PMTiles (zoom 2-7), one file per year.

Variable `snowLying`: number of days with snow lying (variable attribute: "Days of snow lying", label_units "days"). Meaning and units were read from the NetCDF variable attributes of the 2024 file, not assumed. Same pipeline as the temperature layer (pipeline/layers/haduk_extra.py): fixed colour domain across years, bilinear reprojection to EPSG:3857, pre-coloured raster PMTiles (zoom 2-7), one file per year.

Variable `sfcWind`: mean wind speed at 10 m, m s-1. Meaning and units were read from the NetCDF variable attributes of the 2024 file, not assumed. Same pipeline as the temperature layer (pipeline/layers/haduk_extra.py): fixed colour domain across years, bilinear reprojection to EPSG:3857, pre-coloured raster PMTiles (zoom 2-7), one file per year.

Downloaded directly and anonymously (no OS Data Hub account or API key needed for this OpenData product). Verified by fetching the download URL directly (200 OK). GeoPackage contains two layers: `watercourse_link` (193,040 line features, the one rendered) and `hydro_node` (197,734 points, network topology only, not rendered). `form` values: inlandRiver, tidalRiver, canal, lake. Processing: reprojected to EPSG:4326, tiled with tippecanoe zoom 6-10 (dropped to 6-10 from an initial 6-12 build to stay well under Cloudflare Pages' 25MB single-file limit (see docs/DECISIONS.md #4 and #8), without needing R2 set up yet). This is dense real-world line data, so (unlike the Milestone 1 test fixtures) tippecanoe's default scale-dependent thinning is used deliberately, not disabled.

Licence and attribution copied from the dataset's own page. 464 sites: Excellent 297, Good 95, Sufficient 25, Poor 32, Not classified 13, Closed 2. Only slow-changing attributes are tiled (name, classification, water company, heavy-rain flag, year designated); the daily short-term pollution risk forecast is deliberately excluded because a static tile would show stale warnings. Re-run pipeline/layers/ea_bathing_waters.py after each bathing season (classifications update in the autumn).

90 designated bathing waters, each with its current water-quality classification (Excellent/Good/ Sufficient/Poor/Unclassified: 47/26/13/3/1). Filtered server-side to current_yn='Y' -- unlike England's storm overflow returns, every row in this service is already the latest classification, not a multi-year history to deduplicate. A companion to the England/Wales bathing water layer (ea-bathing-waters), not a duplicate: SEPA classifies on its own schedule, so counts and dates are not directly comparable across the border.

33 designated bathing water areas, first designated between 1988 and 2025. This is the designated AREA of each site, not this year's water-quality classification -- unlike SEPA's Scotland layer and the England/ Wales layer, this service has no classification field. A separate "Bathing Water Monitoring Points" item appears to carry sample results, but its licence text read ambiguously ("Open Government Licence with conditions; datasets availability pending") on inspection 2026-09-29, so it is not used until that is confirmed directly. The service's own id field is `OBJECTID_1`, not `OBJECTID` (the same recurring quirk already met in Scotland offshore wind, #63, and NI ASSI, #81); fetched directly in one request (33 features) rather than through the shared paged ArcGIS helper.

The credit "(c) Scottish Water" is the item's own attribution; the data.gov.uk record's OGL v3.0 statement is the licence (its credit line there reads "(c) Crown copyright. All content available under the Open Government Licence v3.0"). FLAGGED for the owner: Scottish Water's own Published Overflow Data page returned 403 to our fetch tool, so its wording was not read directly; the licence rests on the data.gov.uk record and the item text. The item warns the data "should not be interpreted as real-time conditions or confirmation of pollution events" and may include false activations; the layer's caveats repeat this. Only the static site details (id, name, receiving water, type, local authority, position) are tiled; each site's status is fetched in the browser every hour, matched by asset id (unique across all 2,017 sites, asserted at build). Built by pipeline/layers/sw_overflows.py (docs/DECISIONS.md #75).

Licence read from the dataset record's own metadata ("Open Government Licence"); attribution copied from its "Attribution statement". Built by pipeline/layers/ea_storm_overflows.py (docs/DECISIONS.md #66). 14,302 of the 14,303 outlets in the 2025 return are drawn (one has a malformed grid reference); each earlier year is linked to the 2025 outlet by permit reference, else site name plus grid reference, and is left empty rather than guessed when there is no unique match within 500 m (82% of 2021 spills, 83% of 2022, 89% of 2023 and 92% of 2024 are shown). Independent check: for every year the spill count and total duration summed over ALL rows equal the Environment Agency's own summary workbook (2025: 292,119 spills, 1,878,240 hours). 2020 is not used (its files have no locations). The August 2026 revision note says Long-Term Average and EDM operability figures for the revised sites may not fully align with the revised spill figures. Download host changed since docs/DECISIONS.md #48 (the older agrimetrics host no longer resolves); the files come from environment.data.gov.uk.

3,928 water bodies (3,772 river, 133 canal, 23 surface water transfer), MultiLineString. Ecological class: Bad 119, Poor 710, Moderate 2,220, Good 518, High 4, Not assessed 357 (only ~13% of all bodies, ~15% of assessed ones, are High or Good, consistent with the Environment Agency's published picture of English rivers). overall_water_body_class and the chemical classes are 'Not assessed' / 'Does not require assessment' for every row in this release, so they are not used; the per-element classes (fish, invertebrates, plants and algae, phosphate, dissolved oxygen, ammonia) are kept for click-inspect. Reprojected to EPSG:4326, simplified 15 m, tiled at zoom 5-12 (11.3MB). Ecological 'class' may be a 'potential' for heavily modified water bodies (per the dataset's layer descriptions).

Regional climate PROJECTIONS, not observations -- shown as change (anomaly) relative to the Met Office's own standard UKCP18 baseline, 1981-2000 (verified live; NOT the 1961-1990 originally guessed in this file's L2 entry). This specific 12km product covers only the RCP8.5 (high emissions) scenario -- no scenario selector needed, but always labelled as RCP8.5, not "the future". Computed as an ensemble mean across all 16 available model realisations (verified count via the CEDA archive directory listing; some third-party summaries say "12 members", which is wrong for this product). Six time slices, 20-year windows every decade from 2020s to 2070s (2011-2030 through 2061-2080). Ensemble-mean anomalies increase monotonically across them (+0.98 to +3.11 degC mean, entirely positive throughout) -- consistent with published RCP8.5 warming ranges for the UK. Baseline ensemble mean itself (9.46 degC) closely matches HadUK-Grid's real observed values for the same broad period, a useful cross-check between the two datasets. Processing (rewritten 2026-09-20, docs/DECISIONS.md #43): the native 82x112 12km ensemble-mean anomaly is interpolated (bilinear) onto HadUK-Grid's own 1km grid using LAND VALUES ONLY, each 12km cell weighted by the share of it that is HadUK-Grid land (so mostly-sea cells, whose values are sea values, barely influence the coast), lightly smoothed (Gaussian, sigma 6km, about half a model cell, to remove the interpolation's diamond pattern) and clipped to HadUK-Grid's real 1km land mask, so the outline is exactly the coastline of the observed-climate layers. It is weighted averages throughout, so every value stays within the range of the model's land cells and the UK land mean is unchanged (2070s +3.23 degC); no detail finer than the model's 12km is added. (This RCM has real, non-NaN values across its whole domain rectangle, sea, Ireland and France included, unlike HadUK-Grid, which is pre-masked to UK land.) Coloured with a diverging colour ramp centred exactly on zero change (ColorBrewer RdYlBu; the earlier RdBu made every 2070s value read as one flat dark red). Earlier versions used the 12km cells themselves (coastline of 12km squares; see #11 and #14). pipeline/layers/ukcp18_temp_projection.py.

Attribution wording: the UK-AIR page states 'All content is available under the Open Government Licence v3.0' and asks users to 'acknowledge Defra as the source' but gives no fixed attribution sentence, so the displayed wording combines the two. CSV -> float GeoTIFF (1 km, north-up) -> pre-coloured tiles, same pipeline tail as HadUK-Grid (common/raster_tiles.py). 2010-2019 files have 281,802 cells (~27k marked MISSING, coastal squares whose centre is offshore), 2020 has 412,483 cells (157,578 MISSING) and 2021-2024 have 254,905 cells with none missing, so coverage at the coast differs between year groups. The UK-AIR page warns that methodology varies annually and to 'be mindful of these year to year changes when comparing model outputs between years'. Background (not roadside) concentrations. Fixed colour domain across years. Values checked against plausibility: 2024 NO2 0.5-35.5 ug/m3 (London ~30), PM2.5 2.8-11.3 ug/m3.

Attribution wording: the UK-AIR page states 'All content is available under the Open Government Licence v3.0' and asks users to 'acknowledge Defra as the source' but gives no fixed attribution sentence, so the displayed wording combines the two. CSV -> float GeoTIFF (1 km, north-up) -> pre-coloured tiles, same pipeline tail as HadUK-Grid (common/raster_tiles.py). 2010-2019 files have 281,802 cells (~27k marked MISSING, coastal squares whose centre is offshore), 2020 has 412,483 cells (157,578 MISSING) and 2021-2024 have 254,905 cells with none missing, so coverage at the coast differs between year groups. The UK-AIR page warns that methodology varies annually and to 'be mindful of these year to year changes when comparing model outputs between years'. Background (not roadside) concentrations. Fixed colour domain across years. Values checked against plausibility: 2024 NO2 0.5-35.5 ug/m3 (London ~30), PM2.5 2.8-11.3 ug/m3.

Attribution wording: the UK-AIR page states 'All content is available under the Open Government Licence v3.0' and asks users to 'acknowledge Defra as the source' but gives no fixed attribution sentence, so the displayed wording combines the two. CSV -> float GeoTIFF (1 km, north-up) -> pre-coloured tiles, same pipeline tail as HadUK-Grid (common/raster_tiles.py). 2010-2019 files have 281,802 cells (~27k marked MISSING, coastal squares whose centre is offshore), 2020 has 412,483 cells (157,578 MISSING) and 2021-2024 have 254,905 cells with none missing, so coverage at the coast differs between year groups. The UK-AIR page warns that methodology varies annually and to 'be mindful of these year to year changes when comparing model outputs between years'. Background (not roadside) concentrations. Fixed colour domain across years. Values checked against plausibility: 2024 NO2 0.5-35.5 ug/m3 (London ~30), PM10 4.4-18.1 ug/m3.

A real, measured counterpart to the modelled background grids (defra-pcm-no2/pm25/pm10): these are actual monitor readings, not a model. NO2 and PM10 report roadside, urban background and rural background annual means; PM2.5 reports roadside and urban background; ozone reports urban background and rural background as the annual average of the daily maximum 8-hour mean (a different statistic from a simple annual mean, stated on the layer). Matched to AURN site metadata by station name: 187 of 188 stations with a 2025 reading matched (one, "Warrington", appears to have been renamed to "Warrington St. Peter's Park" in the metadata and was left out rather than guessed). A "data capture" percentage from the same tables is shown per reading; DEFRA's own guidance treats an annual mean based on less than about 75% data capture as less certain, stated in the caveats. Spot-checked: London Marylebone Road (a well-known roadside site) reads 30.6 ug/m3 NO2 in 2025, sensibly above the modelled background estimate at the same point (25.1 ug/m3 for 2024) since the model deliberately excludes the local traffic contribution a roadside monitor picks up.

By far the largest source file used so far: 813,627 polygon features, 5.86GB uncompressed GeoPackage (969MB zipped). Download is a two-step process verified live: GET a small JSON endpoint for a short-lived signed Azure Blob URL, then GET the ~970MB file from that URL. No account or API key needed for either step. Two zone values (flood_zone: FZ2 540,282 features, FZ3 273,345 features); flood_source is river (626,860), sea (91,912), "river and sea" (86,742), plus small undefined/unknown categories; origin records the modelling method (modelled, recorded, direct rainfall model, local evidence, or combinations). Per the dataset's own description (verified live, not assumed): "shows the extent of land at present day risk of flooding from rivers and the sea, ignoring the benefits of defences", i.e. these zones do not account for flood defences. Processing: reprojected EPSG:27700 -> EPSG:4326 with a 20m simplification tolerance applied during the same step (a browsing-scale national layer doesn't need engineering-precision boundary vertices, and skipping this made an already enormous dataset intractable to tile in reasonable time), then tiled with tippecanoe's default scale-dependent thinning (zoom 6-11), same reasoning as the rivers layer. See pipeline/layers/ea_flood_zones.py.

Zone definitions from the record's abstract: Zone 3 = rivers >=1% (1 in 100) or sea >=0.5% (1 in 200) annual probability; Zone 2 = rivers 0.1-1% or sea 0.1-0.5%; BOTH INCLUDE A CLIMATE CHANGE ALLOWANCE, unlike the England zones (present day). The record does not state whether flood defences are ignored, so the layer does not say either way. 335,144 polygons streamed from the WFS (1,000/page), simplified with ogr2ogr to ~5 m, tiled at zoom 6-10 (8.1MB, gitignored like the England layer; rebuild with pipeline/layers/nrw_flood_zones.py). mm_id holds the model names per polygon and is not shown.

Licence text and attribution copied from the ArcGIS item's own licenseInfo/accessInformation fields (live, 2026-09-19): "no public access constraints ... subject to the Open Government Licence". 4,128 polygons. Processing: server-side generalisation (maxAllowableOffset 0.0002 degrees, ~20 m) at download, tiled with tippecanoe at zoom 5-11 (4.0MB). Fields kept: NAME, REF_CODE, MEASURE, HYPERLINK. MEASURE's unit is not documented in the service metadata, so it is not displayed.

Sources of the wording (both read live 2026-09-19): the Scottish spatial data catalogue record says "Contains SNH information licensed under the Open Government Licence v3.0" and requires credit to NatureScot; the ArcGIS item says "(c) SNH, Contains Ordnance Survey data (c) Crown copyright and database right (2023)". The attribution string combines the two. 15,877 polygon features (multi-part sites appear as several features; TYPE is BIOLOGICAL, GEOLOGICAL or MIXED). Server-side generalisation (~20 m), tiled at zoom 5-11 (2.3MB). SITE_HA is assumed to be hectares from its name and values but the service does not document the unit, so it is shown as a raw number labelled Site ha.

Attribution copied verbatim from the DataMapWales metadata page on 2026-09-19. 1,088 polygon features (all unique ids). Tiled with tippecanoe at zoom 5-11 with --simplification=8 (0.84MB); no separate simplification step. Fields kept: sssi_name, sssi_code, type, cartesian_area_ha (hectares per the field name).

Licence read from the item's licenseInfo ("ASSI is licensed under the Open Government Licence" with the v3.0 link); the credit is the item's accessInformation, "DAERA NIEA dataset". The item is owned by a DAERA account behind DAERA's open data hub, and OpenDataNI lists the same dataset as OGL. 394 ASSIs (260 polygons, 134 multipolygons), all with a unique reference and all inside Northern Ireland. Habitat, species and earth science interests are kept where the source records one (its "No ASSI feature present in this field category" placeholder is dropped). Built by pipeline/layers/daera_assi.py (docs/DECISIONS.md #81).

Licence and credit read from the item's own text (owner gis.data@gov.scot): "Available under the terms of the Open Government Licence. The following attribution statement must be used ... (c) Scottish Government, contains Ordnance Survey data (c) Crown copyright and database right [insert year]"; its accessInformation gives the year, 2010. Two features: Cairngorms National Park (452,810 ha) and Loch Lomond and The Trossachs National Park (186,340 ha). Wales's three National Parks (NRW DataMapWales) were first left out for lack of a stated licence; the metadata page does state OGL, so they are now the separate layer nrw-national-parks (docs/DECISIONS.md #86). Built by pipeline/layers/sg_national_parks.py (docs/DECISIONS.md #82).

Licence and attribution copied from the ArcGIS item's own licenseInfo/accessInformation fields (live, 2026-09-19). 10 polygons. Server-side generalisation (~20 m) at download. Zoom 4-10. Fields kept: NAME, CODE, DESIG_DATE, STATUS.

Licence ("ogl3") and attribution text read from the dataset record at planning.data.gov.uk/dataset/area-of-outstanding-natural-beauty.json. 34 areas, as expected for England; their union covers 14.8% of England (Natural England says about 15%). The platform marks boundary quality as "some" rather than "authoritative". Fields kept: name, designation date, website.

Licence ("ogl3") and attribution read from the dataset record. 32 defined Heritage Coasts. Phase is beta. Fields kept: name (tidied from capitals), date defined.

Licence ("ogl3") and attribution read from the dataset record. 44,373 polygons (32,626 ASNW, 11,683 PAWS, 64 with a status code the platform does not define, shown as Other); together 3,659 km2, 2.8% of England. Natural England’s own ArcGIS service listed 53,638 polygons on 2026-09-19 (docs/DECISIONS.md #17), so this copy may be a subset or split differently: said in the layer caveats. Government guidance says woods under 2 ha may not appear on the inventory. Tiles 13.7 MB, zoom 8-12, gitignored (rebuild with pipeline/layers/pd_designations.py).

Licence (Open Government Licence v3.0) and "Contains public sector information licensed under the Open Government Licence v3.0" read from the data.gov.uk record; the download’s bundled metadata says "Available under the Open Government Licence" and gives the credit "(c) SNH, Contains Ordnance Survey data (c) Crown copyright and database right (2020)". The attribution combines the two, as for ns-sssi. The source stores each reserve as several polygons (all rows STATUS Current); they are dissolved by name into 49 reserves (asserted; SITE_HA agrees across the parts of each reserve). Dissolved area 1,731 km2 against 1,860 km2 summed from SITE_HA, the gap being reserves that overlap each other. Kind of interest is BIOLOGICAL, GEOLOGICAL or MIXED as in the source (689 of 748 rows MIXED).

The bundled metadata says "Available under the Open Government Licence" and requires "Copyright Scottish Natural Heritage Contains Ordnance Survey data (c) Crown copyright and database right (year)"; the data.gov.uk record instead names the OS Open Data licence and "Copyright NatureScot" (both permit commercial reuse; OS OpenData already accepted for OS Terrain 50 and Greenspace). The year is not given, so 2023 (the file date) is used: FLAGGED for the owner. 28,512 polygons in five antiquity classes (1a 6,305; 1b 708; 2a 8,263; 2b 10,687; 3 2,549), grouped for colour into ancient (semi-natural origin), long-established (plantation origin) and other (on the Roy map only). Total 3,528 km2. Tiles 12.7 MB, zoom 6-12, gitignored.

Licence (OGL v3.0) and credit read from the data.gov.uk record, whose statement ends "database right [insert year]": the year is not stated, so 2026 (the retrieval year) is used: FLAGGED for the owner, like the Scottish National Parks layer. 40 areas including offshore extensions, 13,811 km2 in total. The metadata says the boundaries were digitised against 1:50,000 OS raster data in 1998. Only the name is kept.

Licence ("Available under the Open Government Licence", version 3 link) and the credit "(c) NatureScot, Contains Ordnance Survey data (c) Crown copyright and database right (2024)" read from the download's bundled metadata; the same NatureScot download pattern as ns-nnr. The data.gov.uk record (read 2026-09-25, updated 17 January 2025) instead says "Available under an OS Open Data licence" and gives no credit wording, the same difference as ns-ancient-woodland; both licences permit commercial reuse, and the wording differs at source: FLAGGED for the owner to confirm. 131 polygons are 101 reserves (unique PA_CODE and name), all STATUS Current, dissolved per reserve; total SITE_HA 13,291 ha as summed over the pieces. The source LINK column (old council web addresses) is not shown.

Licence read from the download's bundled metadata ("Available under the Open Government Licence", version 3 link); unlike ns-nnr/ns-lnr/ns-ancient-woodland, this metadata states no bespoke Crown-copyright or Ordnance Survey credit sentence anywhere, so the standard OGL fallback wording is used, naming NatureScot (the metadata's own "Responsible organisation: Scottish Natural Heritage (SNH)") as publisher rather than inventing a credit the source does not give. 38 polygon rows are 36 sites (all STATUS Current); two sites (e.g. Mousa to Boddam) are split into parts and dissolved here by name, asserting the parts agree on lead body and site area. These are Nature Conservation MPAs designated under the Marine (Scotland) Act 2010 and the Marine and Coastal Access Act 2009 -- a different, complementary designation from the SAC/SPA/Ramsar sites JNCC already supplies (#44); 25 of 36 are led by NatureScot (inshore, within 12 nautical miles) and 13 by JNCC (offshore).

3 polygons (Pembrokeshire Coast 615 km2, Bannau Brycheiniog 1,350 km2, Eryri 2,140 km2, measured in British National Grid and matching the source area_ha field). docs/DECISIONS.md #82 recorded this layer as having no stated licence; the licence is a separate field on the metadata page that the first check did not read, so that decision is corrected in #86. Licence read from the layer's own metadata page (metadata_detail) on 2026-09-25, the same field and wording as nrw-sssi; attribution copied verbatim from the record.

6 polygons for five areas: the Wye Valley is supplied as separate England and Wales parts; the England part is dropped here because the England layer draws it. NRW says AONBs were rebranded National Landscapes in 2024, both names in use, AONB the legal name. Names are tidied from capitals and given their Welsh forms by hand (Llyn to Llŷn, Ynys Mon to Ynys Môn). Licence read from the layer's own metadata page (metadata_detail) on 2026-09-25, the same field and wording as nrw-sssi; attribution copied verbatim from the record.

76 polygons, all unique names. The source Notes field is free text with typos and declaration counts, so it is not shown; the declared date and area are kept. Licence read from the layer's own metadata page (metadata_detail) on 2026-09-25, the same field and wording as nrw-sssi; attribution copied verbatim from the record.

97 polygons with the declaring local authority and year. Licence read from the layer's own metadata page (metadata_detail) on 2026-09-25, the same field and wording as nrw-sssi; attribution copied verbatim from the record.

48,839 polygons; 2 flagged Deleted by the source are dropped, leaving 48,837 (26,675 ancient semi-natural, 11,452 restored, 8,943 plantation, 1,767 unknown category); together 947 km2, 4.6% of Wales. Tiles 11.5 MB, zoom 7-12, gitignored (rebuild with pipeline/layers/nrw_designations.py). Licence read from the layer's own metadata page (metadata_detail) on 2026-09-25, the same field and wording as nrw-sssi; attribution copied verbatim from the record.

14 stretches of coast defined as Heritage Coast in 1973-1984, mostly around Pembrokeshire, Llyn, Anglesey, Gower and the Glamorgan coast, a companion to England's Heritage Coast layer (na-heritage-coast, #84), not a duplicate -- Wales's own has its own record and credit. The source's 'Defined' field reads "Laterally" for 11 rows, "Completely" for 2 and "Llaterally" for 1 (Dinas Head) -- an evident typo, folded into "Laterally" rather than shown as a mysterious third category, since only two real definitions exist (a coastal strip only, or the whole parish/community). Licence and attribution read from the layer's own metadata page (metadata_detail) on 2026-09-28, the same field and wording as the other five NRW layers in this family.

25 segments of the official Wales Coast Path (opened 2012), the first continuous, signed coastal path covering an entire country. 16 segments marked "Main" (1,391 km) and 9 "Alternative" (100 km) -- an alternative route around a section of the main path, not a second, separate path. Licence and attribution read from the layer's own metadata page (metadata_detail) on 2026-09-29, the same field and wording as the other six NRW layers in this family. A line layer (a walking route), not a polygon like the rest of this family -- built with the same shared write() helper but its own registry entry rather than pdLayer(), which assumes fill geometry. CORRECTION 2026-10-22 (docs/DECISIONS_LOG.md #249): the source's `length` field is in miles (864.3 miles main, 61.9 miles alternative), which was first shown as km; measured against the geometry in EPSG:27700 the main route is 1,389 km, matching 864.3 miles (1,391 km), and the path is published as 870 miles. Lengths are now converted to km.

Licence read from the Open Data NI record and from the shapefile's bundled metadata ("(c)NIEA, 2019 ... AONB is licensed under the Open Government Licence"). The record states no separate credit wording, so the credit is the metadata's "(c)NIEA, 2019" plus the standard OGL statement: FLAGGED for the owner. 8 areas, as many as Northern Ireland has; CALC_AREA is hectares (asserted against the measured area of every polygon within 3%). Legislation codes (NCALO, Amenity Lands Act) are spelled out.

Licence read from the Open Data NI record and from the shapefile's bundled metadata ("(c)NIEA, 2019 ... NNR and NR is licensed under the Open Government Licence"). No separate credit wording is stated, so the credit combines the metadata's "(c)NIEA, 2019" with the standard OGL statement: FLAGGED for the owner. 50 polygons in one file called NNR_and_NR; TYPE is a habitat description and nothing says which are National Nature Reserves, so the layer is titled "Nature reserves".

Licence and attribution copied from the ArcGIS item's own licenseInfo/accessInformation fields (live, 2026-09-19). 225 polygons. Server-side generalisation (~20 m) at download. Zoom 5-11. Fields kept: NAME, REF_CODE.

Licence and attribution copied from the JNCC Resource Hub record (live, 2026-09-20). 576 polygon features (67.9 MB zipped shapefile). Fields kept: SITENAME, SITECODE, COUNTRY (letter codes E/S/W, read as England/Scotland/Wales; the record does not document them). Simplified to ~20 m. Zoom 4-11. Replaces the England-only Natural England SAC layer. Northern Ireland SACs are a separate JNCC resource, not used.

Licence and attribution copied from the JNCC Resource Hub record (live, 2026-09-20). 270 polygon features (43.9 MB zipped shapefile). Fields kept: SITENAME, SITECODE, Country (letter codes; 'O' is read as offshore, which the record does not document). Simplified to ~20 m. Zoom 4-11. Replaces the England-only Natural England SPA layer. Northern Ireland SPAs are a separate JNCC resource, not used.

Licence and attribution copied from the JNCC Resource Hub record (live, 2026-09-20). 152 polygon features (22.3 MB zipped shapefile). JNCC states the boundaries have not been updated since 2021 and points to the country agencies for definitive boundaries; the layer caveat says so. Fields kept: NAME, CODE, Country. Simplified to ~20 m. Zoom 4-11. Replaces the England-only Natural England Ramsar layer.

Licence and attribution copied from the ArcGIS item's own licenseInfo/accessInformation fields (live, 2026-09-19). 1728 polygon features (multi-part sites appear as several features). Server-side generalisation (~20 m) at download. Zoom 6-11. Fields kept: NAME, REF_CODE.

Licence and attribution copied from the ArcGIS item's own licenseInfo/accessInformation fields (live, 2026-09-19). 106 polygon features (multi-part sites appear as several features). Server-side generalisation (~20 m) at download. Zoom 4-10. Attribution year is 2023 per this item. Fields kept: MCZ_NAME, MCZ_CODE, STATUS.

Licence text and attribution copied from the ArcGIS item's own licenseInfo/accessInformation (live, 2026-09-19). 818,225 polygons (re-downloaded 2026-10-06; the first build on 2026-09-19 had 799,637). The service's layer number is now 1, not 0. Processing: server-side generalisation (0.0004 degrees, ~40 m), streamed as newline-delimited GeoJSON, tiled at zoom 9-11 (47.6MB; lower zooms stalled on very dense tiles). Derived property habitat_group groups the 60+ original MainHabs values (first habitat when several are listed) into 10 plain-English classes for colouring; the original label is kept as main_habitat. Fields kept: MainHabs, AreaHa.

The item's own licenseInfo (read 2026-10-25): "There are no public access constraints to this data. Use of this data is subject to the Open Government Licence". Its accessInformation carries "[Year]" twice; both are filled with 2026, the year the item was last updated, by the same convention as the other Natural England layers: FLAGGED for the owner with the others. The service's 31 Class values are grouped into seven for the legend, keeping the original in `source_class` for the click panel: the named habitats and "Other Priority Habitats" become "Existing priority habitat"; SSSI stays its own class; "Habitat Restoration-Creation" becomes "Being restored or created" and "Restorable Habitat" "Could be restored"; the three network zones become "Nearby: suitable for re-creating this habitat" (Enhancement Zone 1), "Nearby: other habitat or better management" (Enhancement Zone 2) and "Could link networks up" (Expansion Zone). The wording of each follows the item's own description. The build fails if a Class has no group. Eighteen habitat networks are combined here and Natural England settles the overlaps between them: asked about sixteen points (one per class, six well-known places, and four where the generalised copy has two polygons), the service returned exactly one class each time. The 40 m generalisation asked of the server does make neighbouring patches overlap slightly along their edges: 20% of 1,488 sampled points in five 0.1 degree test windows fall inside two or more polygons of this file, where the service gives one. The features are therefore written in precedence order (zones first, existing habitat last) and tiled with --preserve-input-order, so a shared edge shows the most specific class. The zones are buffers around existing habitat, 500 m generalised (1 km for blanket bog), narrowed by soils and in places hydrology, altitude and distance from the coast: not a survey, and not a designation. Zooms 9 to 11, as for the Priority Habitats Inventory. Gitignored, rebuilt with layers/na_habitat_networks.py.

Chosen over UKCEH's Land Cover Map after verifying UKCEH's licence live: bespoke/negotiated, not free for commercial use, £300+VAT fee even for non-commercial vector data. ESA WorldCover is fully open with no such restriction (see docs/DECISIONS.md #13 for the full comparison, done with the owner's explicit sign-off on this tradeoff). Downloaded 16 public S3 tiles (3x3 degree WGS84 grid, no auth needed) covering the UK bounding box (~530MB raw). Class codes, names and the official RGB palette were all read directly from the downloaded GeoTIFF's own embedded metadata/colour table, not invented or taken from secondary documentation. Processing: mosaic -> reproject to EPSG:3857 with *mode* resampling (never bilinear/average -- this is categorical class-code data, not a continuous field) at 400m (coarser than the 200m first attempt, which produced a 64.5MB file over Cloudflare Pages' 25MB limit) -> masked to UK land using HadUK-Grid's own land mask, since ESA WorldCover classifies open sea as "permanent water bodies" the same as inland lakes, and the bounding-box crop alone left ~41% of the raster as ocean masquerading as UK data -> expanded to RGBA using ESA's own embedded colour table -> tiled with nearest-neighbour overviews (same reasoning as docs/DECISIONS.md #11: categorical data must never be blended, and small isolated land like the Isle of Man risks disappearing at low zoom under bilinear/average resampling). Final UK class distribution (masked, land-only): grassland 59.3%, cropland 18.7%, tree cover 15.7%, built-up 3.9%, water bodies 2.1%, wetland/bare/shrubland/moss <1% combined -- plausible against general knowledge of UK land cover. See pipeline/layers/esa_worldcover_landuse.py.

Attribution wording: carbonintensity.org.uk states 'Our API is licensed under the CC BY 4.0 license' but gives no fixed attribution sentence, so the displayed wording credits the publisher and licence; the region-boundary credit is the NESO Open Licence statement recorded in neso-dno-regions.

Same API family and licence as neso-carbon-intensity, a different endpoint (/generation instead of /regional): the current GB-wide electricity generation fuel mix as percentages (wind, gas, nuclear, solar, biomass, hydro, imports, coal, other), one national reading rather than 14 DNO regions, so it has no region-boundary credit to add (unlike neso-carbon-intensity, which credits the boundary source too) and gets its own manifest entry rather than sharing that one. Verified live 2026-09-29. Indicative forecast of electricity-generation carbon intensity in gCO2/kWh for the 14 DNO-based regions (ids 1-14; ids 15-18 are England/Scotland/Wales/GB aggregates and are ignored). Fetched directly by the browser every 30 minutes while the layer is active; no proxy or key. Verified 2026-09-19: response carries access-control-allow-origin: *.

Licence and attribution copied from the dataset page (live, 2026-09-20), which also says use of OS OpenData is subject to os.uk/opendata/licence and that it "contains some data created by MHCLG" that is being replaced with authoritative sources (all 181 records here are marked quality: authoritative). The page describes the data as a snapshot of the authorities' annual returns (submitted in March, published in autumn) that will not show later boundary changes. 181 designations, one per local planning authority area; fields kept: name, green-belt-core (the belt or area it belongs to), local-authority-district. Independent check: the union area of all polygons is 16,332 km2, about 12.5% of England's area (roughly 130,000 km2). Simplified to ~20 m; zoom 4-11; 1.5 MB.

Licence and attribution copied from the ArcGIS item's own licenseInfo field (live, 2026-09-22): "Use of this data is subject to the Open Government Licence. Any use of the data requires the following attribution: '© Forestry Commission copyright. Contains Ordnance Survey data © Crown copyright and database right 2024.'" 641,092 total polygons; CATEGORY splits them 610,440 Woodland / 30,652 Non woodland (open-area gaps inside woodland boundaries, e.g. rides, ponds, buildings); only Woodland is downloaded and shown. IFT_IOA (25 values) grouped into 6 plain-English types for colouring; the original value is kept for click-inspect. Server-side generalisation via maxAllowableOffset. Zoom 8-11 (dense polygons, like the Priority Habitats Inventory, #22): a lower zoom either stalled or produced an unusably large file on a first attempt. Gitignored and rebuilt with layers/nfi_woodland.py.

7 polygons, one per Forest District Forestry England itself manages (Central England, East England, North England, South England, West England, Westonbirt, Yorkshire) -- already dissolved to district level, not sub-compartment/parcel level (an earlier check of this family found only parcel-level data; this named "Legal Boundary" item is a newer, simpler release). The item's own `cost_centre` coded-value domain also lists Scottish and Welsh Forest District names (Forestry England, Forestry and Land Scotland and Natural Resources Wales share the same schema from their shared Forestry Commission history), but only the 7 English codes are populated in this feature service -- confirmed by fetching every feature's code, not assumed from the domain list. Different from the National Forest Inventory GB layer (nfi-woodland): this is who manages the land, not what grows on it. The item's licenseInfo states a specific sentence for map use ("Any maps produced using this data should contain the following Forestry Commission acknowledgement: ..."), used verbatim here rather than the item's separate, differently-worded accessInformation field.

Attribution copied from the download's own Doc/licence.txt. 165,978 sites across Great Britain, classed by the `function` field into ten types (play space, religious grounds, public park or garden, playing field, other sports facility, allotments, cemetery, tennis court, bowling green, golf course); no name field. The product records the TYPE of greenspace, not whether the public may enter it (a separate OS product covers access), so the layer says a cemetery or private sports club can look the same as a public park. Simplified (~5 m) and tiled from zoom 9 so the country-wide view is not overwhelmed; 27.5 MB, gitignored and rebuilt with layers/os_open_greenspace.py.

Licence and the required attribution statement copied from the API's reference page (live, 2026-09-20). The API sends access-control-allow-origin: *, so the browser calls it directly; the documentation asks for a single latest-readings call at most every 15 minutes, which the layer's refresh interval respects. 3,213 stage gauges listed; 1,762 kept (894 have no published typical range, 556 have several stage measures so which one the typical range refers to is ambiguous, 1 has a bad range). Independent datum check at build: of 1,705 gauges with a current reading, 99.3% lie within their own lowest-highest level on record. Each gauge's typical range (exceeded 5% of the time at the top, 95% at the bottom, per the API documentation), highest recent level (typically the last five years) and record levels are stored in ea-river-levels.stations.json beside the tiles (0.8 MB) by layers/ea_river_levels.py; the live reading is compared with them in the browser. The API is a Beta service.

The copyright.txt beside every download (read 2026-10-11) says: "(c) European Union, 1995-2026. The Commission's reuse policy is implemented by the Commission Decision of 12 December 2011. Any copyright and/or sui generis right on the dataset is licensed under the Creative Commons Attribution 4.0 International (CC BY 4.0) licence. Reuse is allowed provided appropriate credit is given and any changes are indicated." The citation is copied from section 2.5.5 of the GHSL Data Package 2023 PDF that ships inside every tile zip. Method (same PDF, section 2.5): residential population from census units (CIESIN GPWv4.11) is spread onto a 100 m grid guided by mapped built-up volume from Sentinel-2 and Landsat, rescaled to UN World Urbanization Prospects 2018 city totals and UN World Population Prospects 2022 country totals, then summed to 1 km; Eurostat LAU2 figures were used as a control inside the EU. 2025 and 2030 in the same release are projections and are not used here. Processing: the 1 km Mollweide tiles over a 32 W to 45 E, 27 N to 72 N window (40 tiles, 36 with data) are mosaicked, resampled to a 700 m Web Mercator grid with nearest-cell sampling, grouped into nine density classes (1 to 3, 3 to 10, 10 to 30, 30 to 100, 100 to 300, 300 to 1,000, 1,000 to 3,000, 3,000 to 10,000, more than 10,000 per km2; under 1 and sea transparent) and written as 4-bit palette PNG tiles, zoom 3-8, about 23.6 MB per year. Checked at build against the 2020 grid: Iceland 366,669 people (reference about 366,000), the island of Ireland with the Isle of Man 6,904,667 (about 6.9 million), Paris centre 24,059, Madrid centre 30,844, Trafalgar Square 3,915 per km2, Ben Nevis 1 per km2, open Sahara 0. The window is a rectangle: it includes the UK, the EU and EFTA, Turkey, Ukraine, Belarus, western Russia and a strip of North Africa. The highest single cells in the window are 204,000 per km2 near Bursa (Turkey), 152,000 near Cairo and 130,000 near 58.0 N 38.9 E (Russia). Such values are modelling effects (a unit's population concentrated into a few cells), not real street densities, which is one reason the map shows bands, not exact figures.

The copyright.txt beside the download (read 2026-10-11) says: "(c) European Union, 1995-2026. The Commission's reuse policy is implemented by the Commission Decision of 12 December 2011. Any copyright and/or sui generis right on the dataset is licensed under the Creative Commons Attribution 4.0 International (CC BY 4.0) licence. Reuse is allowed provided appropriate credit is given and any changes are indicated." The citation is copied from section 2.1.7 of the GHSL Data Package 2023 PDF that ships inside every tile zip. Definition (same PDF, section 2.1.1): the built-up surface is the gross surface bounded by the building wall perimeter, at the detail of a 1:10,000 topographic map ("building footprint"); the share of a cell it covers is the built-up fraction. Method (section 2.1): a 10 m built-up fraction is estimated from a Sentinel-2 composite (2018) and from Landsat for the earlier epochs, and interpolated to 5-year steps; the process can only reduce built-up surface going back in time. The PDF says its error estimates were still being prepared (published 2024). 2025 and 2030 in the same release are projections and are not used here. Processing: the 1 km Mollweide tiles over a 32 W to 45 E, 27 N to 72 N window (36 tiles with data) are mosaicked, resampled to a 700 m Web Mercator grid with nearest-cell sampling, grouped into six classes of the share of the cell that is built (0.1 to 1, 1 to 3, 3 to 10, 10 to 20, 20 to 35 and 35 per cent or more; under 0.1 per cent and sea transparent) and written as 8-bit palette PNG tiles, zoom 3-8, 13 to 15 MB per year. Checked at build against the 2020 grid: Trafalgar Square 34.4 per cent, Paris centre 37.9, Madrid centre 45.4, Ben Nevis 0.05, open Sahara 0, open Atlantic 0 (the densest cells in the window are about 60 per cent, because streets are not buildings); read back by class from the tiles for Paris, Berlin, the Cotswolds and New Cairo, where the 1990 and 2020 grids differ (0.2 per cent in 1990, 33 per cent in 2020). The window is a rectangle: it includes the UK, the EU and EFTA, Turkey, Ukraine, Belarus, western Russia and a strip of North Africa (New Cairo, almost empty in 1990, is among the largest 1990 to 2020 changes inside it). Gitignored (4 files of 13 to 15 MB) and rebuilt with layers/ghsl_built.py.

The copyright.txt beside the download (read 2026-10-11) is the same notice as for GHS-POP and GHS-BUILT-S: "(c) European Union, 1995-2026 ... licensed under the Creative Commons Attribution 4.0 International (CC BY 4.0) licence. Reuse is allowed provided appropriate credit is given and any changes are indicated." The citation is copied from section 2.6.7 of the GHSL Data Package 2023 PDF that ships inside every tile zip. Method (same PDF, section 2.6): the Degree of Urbanisation method of the European Commission and Eurostat (2021) applied to the GHS-POP and GHS-BUILT-S grids gives each 1 km cell one of eight types (codes 30 urban centre, 23 dense urban cluster, 22 semi-dense urban cluster, 21 suburban or peri-urban, 13 rural cluster, 12 low density rural, 11 very low density rural, 10 water): for example an urban centre is a cluster of cells with more than 1,500 people per km2 (or a high built-up share) and more than 50,000 people in all. 2025 and 2030 in the same release are projections and are not used. Processing: the 1 km tiles over a 32 W to 45 E, 27 N to 72 N window (36 tiles with data) are mosaicked, resampled to a 700 m Web Mercator grid with nearest-cell sampling, the seven land types drawn as classes (water and no data transparent) and written as 8-bit palette PNG tiles, zoom 3-8, about 13 MB per year. Checked at build against the 2020 grid: only the documented codes occur; Paris, Madrid and Trafalgar Square are urban centres, Ben Nevis and the open Sahara are very low density rural (the Sahara is land, so it is not water), the open Atlantic is no data. Urban-centre cells grow from 323,959 in 1990 to 415,262 in 2020 in the window; read back by class from the tiles for Paris, the Cotswolds, the Highlands and a cell north-east of Cairo that is very low density rural in 1990 and an urban centre in 2020. The window is a rectangle (it includes North Africa and Turkey). Gitignored (4 files of about 13 MB) and rebuilt with layers/ghsl_smod.py.

The copyright.txt beside the download (read 2026-10-12) is the same notice as for the other GHSL products: "(c) European Union, 1995-2026 ... licensed under the Creative Commons Attribution 4.0 International (CC BY 4.0) licence. Reuse is allowed provided appropriate credit is given and any changes are indicated." The citation is copied from section 2.3.4 of the GHSL Data Package 2023 PDF that ships inside every tile zip. Method (same PDF, section 2.3): built-up volume (BUVOL, cubic metres of built space above ground) = average net building height of 2018 (GHS-BUILT-H, estimated from the AW3D30 and SRTM30 elevation models and Sentinel-2 shadows, at 100 m) times the built-up surface of the epoch (GHS-BUILT-S), so the heights are one 2018 estimate used for every epoch. Only 2020 is built: the other epochs would only repeat the built-up layer. Encoding UInt32 in m3, no data 4,294,967,295. Processing: the 1 km Mollweide tiles over a 32 W to 45 E, 27 N to 72 N window (36 tiles with data) are mosaicked, resampled to a 700 m Web Mercator grid with nearest-cell sampling, grouped into seven classes (1,000 to 10,000, 10,000 to 50,000, 50,000 to 200,000, 200,000 to 1 million, 1 to 3 million, 3 to 6 million, more than 6 million m3; under 1,000 and sea transparent) and written as 8-bit palette PNG tiles, zoom 3-8, 17 MB (gitignored; rebuilt with layers/ghsl_volume.py). Cells in the window: 6.8 million in the first class down to 8,074 in the top class. Checked at build against the 2020 grid (ranges chosen from what the places are, before looking): Trafalgar Square 4.2 million m3, Paris centre 8.2 million, Madrid centre 6.1 million, Ben Nevis 1,195 m3, open Sahara 0, open Atlantic 0; read back by class from the tiles in the browser for central Paris and La Defense (more than 6 million) and the Cotswolds (50,000 to 200,000); the highest cell in the data is about 13.7 million m3. The window is a rectangle (it includes North Africa and Turkey).

The copyright.txt beside the download (read 2026-10-12) is the same notice as for the other GHSL products: "(c) European Union, 1995-2026 ... licensed under the Creative Commons Attribution 4.0 International (CC BY 4.0) licence. Reuse is allowed provided appropriate credit is given and any changes are indicated." The citation is copied from section 2.1.3 of the product report (GHS_AGE_R2025.pdf, inside the zip). Definition (same report, section 2.1): for each cell, the epoch in which 50 per cent of the built-up surface of 2020 has been exceeded for the first time in the GHS-BUILT-S R2023A stack (Landsat and Sentinel-2), an estimate of the dominant age of the "built stock"; if all buildings in a cell had the same footprint it would be the median construction epoch. The report itself says the age ignores building renewal and buildings that no longer exist, that the more a cell was rebuilt after 1975 the less useful it is as a proxy for construction epoch, and that ground first built after 2020 is not in this version. Classes (the file's csv): 0 not built-up, 1 before 1975, 2 1975-1980 ... 10 2015-2020; no data 255. Processing: the 1 km global file cut to a 32 W to 45 E, 27 N to 72 N window, resampled to a 700 m Web Mercator grid with nearest-cell sampling, the ten classes drawn in viridis (the file's own colours are a rainbow; not colour-blind safe), not-built-up and no data transparent, 8-bit palette PNG tiles, zoom 3-8, 19 MB (gitignored; rebuilt with layers/ghsl_age.py). Cells in the window: before 1975 7.23 million, 1975-1980 1.18 million, then 1.00, 0.77, 0.80, 0.50, 0.52, 0.32, 0.22 and 0.19 million for the later periods. Checked at build against the source grid: Notre-Dame, the Colosseum and Tallinn Old Town are before 1975; Almere (a city on reclaimed land, founded 1976) reads 1975-1980; Disneyland Paris (opened 1992) reads 1995-2000; the open Sahara is not built-up. Three places I expected to read as new, Potsdamer Platz in Berlin (rebuilt 1993-2000), Canary Wharf (1988-2004) and HafenCity in Hamburg (from 2003), all read "before 1975": the cell's ground was already built on or is mostly older streets, which is exactly the renewal limit the report states, so they were taken out of the checks and the layer's caveat names them. Read back by class from the tiles for Paris, Rome, the Almeria greenhouse plain, eastern Istanbul and the Nile Delta. The window is a rectangle (it includes North Africa and Turkey).

Licence read from the record's own licence field through the figshare API (CC BY 4.0, 2026-10-16); the same record lists no other terms. Chosen over the Earth Observation Group's own annual VIIRS files (CC BY 4.0 on their page, but the downloads need a login) and over NOAA's DMSP archive (its pages have moved behind the same login). Content, from the record: 1992 to 2013 are DMSP-OLS night-time lights calibrated between satellites (F10 for 1992-1994, F12 for 1995-1996, F14 for 1997-2003, F16 for 2004-2009 and F18 for 2010-2013); 2014 to 2024 are VIIRS data converted to the DMSP scale ("simVIIRS"); 30 arc-seconds; whole numbers 0 to 63. The record suggests using cells with a value above 7; here cells at 11 and below are transparent, because the faintest cells (8 to 11) flicker from year to year and at the 2013/2014 join (cells above 7 fall by a third between 2013 and 2014 and swing by 10 to 25 per cent between other neighbouring years), which looks like twinkling when the years play. This is a harmonised, modelled series, not one instrument, and the join does not match: measured 2026-10-16 country by country (cells at 12 or more as a share of the country's 1 km cells), Germany falls from 37% in 2013 to 16% in 2014, France from 27% to 16% and Poland from 26% to 15%, Germany's mean brightness halves (13.6 to 6.7), and within the DMSP years Poland reads 52% in 2010, 32% in 2011, 44% in 2012 and 26% in 2013; the layer's caveats and the story say so (docs/DECISIONS_LOG.md #236). Processing: each year's global grid is warped to a 700 m Web Mercator grid over a 32 W to 45 E, 27 N to 72 N window (nearest cell), grouped into four classes (12 to 19, 20 to 31, 32 to 47 and 48 to 63) and written as 8-bit palette PNG tiles, zoom 3 to 8, about 10 MB a year. Checked at build against the 2020 grid: Paris, London and Madrid centres 40 to 63; the open North Atlantic 0; Ben Nevis 12 or less; open Sahara 5 or less; and the number of lit cells in the window every year (printed, so a broken year shows as a jump). Gitignored (33 files) and rebuilt with layers/night_lights.py.

The download page (read 2026-10-16) says: "This work is licensed under a Creative Commons Attribution 4.0 International License ... You are free to copy and redistribute the material in any medium or format, and to transform and build upon the material for any purpose, even commercially", with the display credit "Source: Hansen/UMD/Google/USGS/NASA" and the citation copied here. Layers used: treecover2000 (canopy closure in 2000 for vegetation taller than 5 m, 0 to 100 per cent) and lossyear (0 for no loss, or 1 to 24 for a stand-replacement disturbance detected in 2001 to 2024). Tree cover is taken as 30 per cent canopy or more in 2000 (the usual Global Forest Watch threshold). For each year the layer shows the share of each 16 x 16 block of cells (0.004 degrees, about 450 m north to south) still covered: the 2000 tree cover less the loss up to and including that year, warped to a 700 m Web Mercator grid (average), grouped into four classes (10 to 29, 30 to 54, 55 to 79 and 80 to 100 per cent; under 10 per cent and sea transparent) and written as 8-bit palette PNG tiles, zoom 3 to 8. The source's gain layer (2000 to 2012 only, no year) is not used, so regrowth and replanting are not added back; the page itself warns that sensors and methods changed over the series and that the 2001 to 2012 and 2011-onward parts were not processed alike. Checked at build against the 2000 cover: Black Forest 40 to 100 per cent, central Sahara 0 to 2, central London 0 to 20, Iceland's interior 0 to 2; the 10 x 10 degree tile over southern Sweden, Denmark and northern Poland loses about 21 per cent of its 2000 tree cover by 2024; and blocks in northern Euboea (Greece) fall from 91 to 98 per cent in 2020 to 0 to 3 per cent in 2021, the August 2021 fire. Gitignored (25 files, about 26 MB each) and rebuilt with layers/forest_loss.py.

Changed 2026-10-29 (docs/DECISIONS_LOG.md #255): the sea is now cut away from both layers with Natural Earth 1:10m land polygons. Measured first: 67% of everything the occurrence layer drew was sea, estuary or coastal water, and the Explorer maps coastal waters and only parts of some seas, so that sea was an uneven picture; the rebuild kept 39% of the occurrence cells and 90% of the change cells. Of the water on land, 63% lies outside any mapped lake or reservoir and 49% is there in under a third of observations, which is what no other layer here holds. The download page (read 2026-10-16) says: "All data here is produced under the Copernicus Programme and is provided free of charge, without restriction of use", asks for the credit "Source: EC JRC/Google" when the data are used as a layer in a published map, and for the Pekel and others (2016) citation copied here; the licence link on the page points to the Copernicus Regulation (the licence_url field is the Copernicus copyright and licences page). Version 1.5 (the page's "new dataset 1984-2024") is used because the page says its Occurrence layer was recomputed against the original baseline, to remove errors that the added years had caused in versions 1.2 to 1.4. Two layers are made: Occurrence (the per cent of valid observations in which a 30 m cell was water, averaged over each 16 x 16 block and then onto a 700 m Web Mercator grid, grouped into four classes: 2 to 9, 10 to 29, 30 to 59 and 60 to 100 per cent) and Transitions (the commonest of four changes in each block: new permanent, new seasonal, lost permanent and lost seasonal water; codes 2, 5, 3 and 6 in the Data Users Guide). Only one picture of the whole period is shown, because the yearly files are published for download only for 2015 to 2021 (from different releases), and the full yearly series exists only in Google Earth Engine, which this project does not use. The page warns that the 2022 to 2024 data use Landsat Collection 2 with a co-registration offset of up to one 30 m cell against the earlier years. Checked at build: Lake Geneva and Lake Constance read 90 to 100 per cent, central Madrid and Ben Nevis 0 to 10. Gitignored (2 files of about 12 MB each) and rebuilt with layers/surface_water.py.

The HydroLAKES page (read 2026-10-16) says: "The HydroLAKES database is licensed under a Creative Commons Attribution (CC-BY) 4.0 International License", and so does section 1 of the Technical Documentation v1.0 ("free for scientific, educational, and other uses"), which also gives the citation copied here; no credit wording is given, so the credit line is ours, built from that citation. The other HydroSHEDS products (HydroRIVERS, HydroBASINS and others) are under the separate HydroSHEDS licence agreement, which is NOT used here (docs/DECISIONS_LOG.md #235). 101,974 of the 1,427,688 lakes fall in the window (48 of them 500 km2 or more). Fields used: Lake_type (1 lake, 2 reservoir, 3 regulated lake; the documentation warns that "Lake" also includes all unidentified smaller human-made reservoirs), Lake_area, Vol_total (reported for lakes of 500 km2 and over and for reservoirs in GRanD, otherwise estimated by a statistical model: Vol_src), Depth_avg (volume divided by area), Elevation, Country and Lake_name (given only for lakes of 500 km2 or more, large reservoirs and lakes named in GLWD). Polygons are cut to the window with ogr2ogr; the Caspian Sea, which HydroLAKES counts as a lake, is cut at 45 E. Tiles at zoom 3 to 10; lakes of 500 km2 or more from zoom 3, 50 km2 from 5, 5 km2 from 7, 1 km2 from 8, the rest from 9. Checked at build against known lakes by true coordinate: Ladoga 17,444 km2, Onega 9,962, Vanern 5,486, Geneva 572, Constance 522, Balaton 577, Loch Ness 52. Gitignored (about 31 MB) and rebuilt with layers/europe_lakes.py.

The terms page (read 2026-10-16) says: "All versions of Natural Earth raster + vector map data found on this website are in the public domain ... invites you to use them for personal, educational, and commercial purposes. No permission is needed to use Natural Earth. Crediting the authors is unnecessary", and suggests the credit "Made with Natural Earth". Two themes are joined: the global rivers_lake_centerlines (the great rivers, scale ranks 1 to 9, 235 lines in the window, of which 50 are lake centrelines and 2 canals) and the Europe supplement (1,325 lines in all, ranks 10 to 12; 976 in the window, of which 332 are lake centrelines and 44 intermittent rivers). Both are selections made by cartographers for a 1:10 million map, not a full network; 829 river, canal and intermittent-river lines remain after the lake centrelines (lines drawn through lakes to join a river across them) are left out. The full European river network (HydroRIVERS) falls under the HydroSHEDS licence agreement, which allows distribution only to end users bound by a licence at least as protective as WWF's, with indemnification and record-keeping clauses: not usable on a public static site, so NOT used (docs/DECISIONS_LOG.md #235). Checked at build against known rivers by true coordinate (Danube at Budapest, Rhine at Cologne, Seine at Paris, Thames at London, Volga at Volgograd, Loire at Nantes: each within 0.03 degrees). About 1.7 MB, tiled at zoom 3 to 9 and rebuilt with layers/europe_rivers.py.

Same terms as natural-earth-rivers above (read 2026-10-16, re-checked 2026-11-06): public domain, no credit required. 384 ports in the window; not filtered to "major", because the dataset's own scalerank (1 to 8, used by Natural Earth for label placement on a small-scale map) does not track real cargo traffic closely enough to use as a filter: Rotterdam and Hamburg, Europe's two busiest ports by tonnage, are scalerank 5; Algeciras and Piraeus are scalerank 8, the dataset's lowest rank; Felixstowe, the UK's busiest container port, is not in the dataset at all. Checked at build against Rotterdam, Hamburg and Antwerp by true coordinate. About 0.2 MB, tiled at zoom 2 to 11 and rebuilt with layers/europe_ports.py.

Same terms as natural-earth-rivers above (read 2026-10-16, re-checked 2026-11-06): public domain, no credit required. 190 airports in the window; kept the 124 whose "type" field contains "major" (the publisher's own category, unlike the ports dataset's unreliable scalerank, see natural-earth-ports above), checked against Heathrow, Charles de Gaulle and Schiphol by true coordinate and IATA code. About 0.1 MB, tiled at zoom 2 to 11 and rebuilt with layers/europe_airports.py.

The licence and credit were re-read on 2026-10-12 from a downloaded tile's own metadata (gdalinfo -mdd all): "license=CC-BY 4.0 - https://creativecommons.org/licenses/by/4.0/" and "copyright=ESA WorldCover project 2021 / Contains modified Copernicus Sentinel data (2021) processed by ESA WorldCover consortium" (the same product as the UK land use entry above, see docs/DECISIONS.md #13). Class codes and ESA's official colours were read from the file's colour table (10 tree cover, 20 shrubland, 30 grassland, 40 cropland, 50 built-up, 60 bare or sparse vegetation, 70 snow and ice, 80 permanent water bodies, 90 herbaceous wetland, 95 mangroves, 100 moss and lichen). The tiles are 36,000 pixels square with overviews down to 562 pixels, so GDAL read only the overview levels needed for a 153 m grid (16 parallel strips, about 10 minutes). The first build was 500 m (zoom 3-8, 43 MB), where the Thames in London vanished; it was rebuilt at 153 m in Web Mercator metres, about 95 m on the ground in London, after the owner could not see the river (docs/DECISIONS_LOG.md #211) and took the most common class per cell (mode, never an average of class codes). Sea: WorldCover calls open sea "permanent water", so the sea is cut away with the Natural Earth 1:10m land polygons (public domain), leaving lakes and rivers inside the land as water; the coastline is therefore only as exact as that generalisation. One stray mangrove cell (there are none in Europe) was dropped and counted at build; the build stops if more than 49 appear. Processing: 8-bit palette PNG tiles, zoom 3-10, 298 MB (gitignored; rebuilt with layers/esa_worldcover_europe.py). Cells in the window (153 m cells, Mercator, not equal area): tree cover 500.3 million, grassland 279.6 million, cropland 227.9 million, bare 213.5 million, water 53.5 million, snow and ice 37.7 million, moss and lichen 30.9 million, wetland 21.8 million, built-up 16.8 million, shrubland 8.1 million. Checked at build against known places (Lake Geneva and Lake Constance water, the Beauce cropland, Białowieża forest tree cover, the Aletsch region bare or ice, the open Atlantic no data, the Sahara bare ground, Scottish moorland grass, the Thames at Westminster Bridge water) and read back by class from the tiles in the browser for Paris, the Beauce, Białowieża, Mont Blanc, Lake Geneva, the Danube delta wetland, Rannoch and the Sahara. Built-up land looks much rarer than it is because a 153 m cell is only built-up where buildings are the commonest class: said on the layer. A first test point (a spot east of Freiburg in the Black Forest) read as grassland, not trees, because the Black Forest is a mosaic of forest and meadow: it was replaced by Białowieża, a real forest. The window is a rectangle (it includes North Africa and Turkey).

Read 2026-10-11 from the EEA catalogue record and the metadata file that ships beside the data: "License CC-BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Copyright holder: Directorate-General for Environment (DG ENV). This data is provided for general information purposes only. Only the data possessed by the competent authorities of the Member States is authentic. Therefore, no rights or legal claims can be derived from the data displayed on this site." The record gives no citation sentence, so the citation above is built from its title, publication date and DOI. It names no country-by-country conditions on the spatial data: the older note in docs/EUROPE.md (Austria, Estonia, Hungary, Ireland, Romania and Turkey; credit lines for Estonia and Finland), which came from search results and not from the EEA, is not in this release's record, and the file holds sites from all 27 Member States. The record's one country note concerns species data: sensitive species information was removed for Austria, Belgium, Cyprus, Estonia, Finland, France, Germany, Greece, Ireland, Italy, Latvia, Lithuania, Luxembourg, Malta, Poland, Portugal, Slovakia, Spain and Sweden (it does not change the outlines shown here). The older UK note in the 2018 record does not apply: the UK is not in this release. 27,173 sites: 3,374 Special Protection Areas (type A, Birds Directive), 21,539 Sites of Community Importance or Special Areas of Conservation (type B, Habitats Directive) and 2,260 designated as both (type C). Processing: boundaries simplified to 30 m in EPSG:3035 (a shapefile of 1.14 GB), reprojected, and tiled with tippecanoe for zoom 3-11, dropping the smallest sites first at the zoomed-out levels (84 MB, gitignored, needs R2 at deploy; rebuilt with layers/protected_sites_europe.py). Checked against known sites by point-in-polygon: Doñana (Spain), the Camargue (France), Puszcza Białowieska (Poland), Niedersächsisches Wattenmeer (Germany) and the Waddenzee (Netherlands); site names stay in their national languages. Sites overlap (a bird area and a habitat area can cover the same ground), so the areas of the polygons must not be added up: their sum is about 1.67 million km2, more than the network's real extent.

Read 2026-10-11 from the EEA catalogue record and the metadata file beside the data: "License CC-BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Copyright holder: Council of Europe (CoE)." The record gives no citation sentence, so the citation above is built from its title, publication year and DOI. The Emerald Network is the Council of Europe's network of Areas of Special Conservation Interest under the Bern Convention (Recommendation No. 16 of 1989 and Resolution No. 3 of 1996), the counterpart of Natura 2000 for countries outside the EU; the EU contributes its Natura 2000 sites to it, so the EU is not in this file. 2,578 sites in 16 countries: United Kingdom 943, Norway 705, Ukraine 537, Georgia 66, Moldova 61, Serbia 61, Switzerland 37, North Macedonia 33, Montenegro 32, Bosnia and Herzegovina 29, Albania 25, Armenia 23, Azerbaijan 17, Iceland 5, Andorra 2, Liechtenstein 2. The UK's sites are the same places as its Special Areas of Conservation and Special Protection Areas (all four nations, 73 in Northern Ireland), so they overlap the JNCC layers. The file has no site-type field (the column is empty), so the layer has one class. Processing: boundaries simplified to 30 m in EPSG:3035, reprojected and tiled with tippecanoe for zoom 3-11, dropping the smallest sites first at the zoomed-out levels (16 MB, gitignored; rebuilt with layers/protected_sites_europe.py). Checked against known sites by point-in-polygon: Jotunheimen (Norway), Vatnajokull (Iceland), the Danube Biosphere Reserve (Ukraine), Tara (Serbia) and Kolkheti (Georgia). Three Ukrainian points failed in the raw shapefile as well as in the simplified data: Hoverla, the highest peak, lies about 280 m outside the Carpathian Biosphere Reserve polygon, the Chornobyl plant is in no listed polygon, and a point in the delta's water between islands is in none: the submitted boundaries are approximate, which the layer's caveat says. The older note in docs/EUROPE.md lists only the EEA and Natura 2000 for protected sites; this release's record names no per-country conditions.

Read from the licence PDF (3 pages, 2026-10-11): Article 4 grants reproduction, distribution, communication to the public and adaptation; Article 5 says free of charge; Article 6 requires the source notice, the "produced using ..." notice where the data were modified, the no-liability sentence in any legal notice, and that the user not suggest the EU, Airbus or ESA endorse the activity (all four are shown: the notice and sentence in the credit, the endorsement line in the caveats). The citation sentence is the one the Amazon open-data page asks for. It is a surface model (buildings and tree canopy included, not bare earth). Processing: each tile is a Cloud Optimized GeoTIFF with overviews, so the 30 m heights were read from the overviews and area-averaged into a 500 m grid (2,068 tiles over a 32 W to 45 E, 27 N to 72 N window, fetched in 16 parallel strips), then drawn in eleven bands (below sea level, 0 to 100, 100 to 200, 200 to 400, 400 to 700, 700 to 1,000, 1,000 to 1,500, 1,500 to 2,000, 2,000 to 2,500, 2,500 to 3,500, more than 3,500 m) as 4-bit palette PNG tiles, zoom 3-8. The coastal tiles store the sea as exactly 0: a cell of exactly 0.0 is blanked as sea (an average of land samples is never exactly 0.0). Checked at build against known heights in the 500 m grid (Mont Blanc 4,664 m, Ben Nevis 1,207, Zugspitze 2,799, Etna 3,272, Teide 3,590, Hvannadalshnukur 1,991, Galdhopiggen 2,340; Zuidplaspolder -7 m; four open-sea points 0 or no data; window maximum 5,557 m near Elbrus, minimum -427 m at the Dead Sea). A few countries' tiles (Armenia, Azerbaijan) are not public and are gaps. Gitignored (13 MB: first built at 35 MB, then rebuilt without the zoom-9 tiles the map never asks for, 2026-10-11) and rebuilt with layers/copernicus_dem.py. The same 500 m grid also feeds the layer copernicus-hillshade (shaded relief, added 2026-10-11): the light is computed in the equal-area projection (EPSG:3035), where cells are true 500 m squares, with gdaldem hillshade (light from several directions at once, vertical scale stretched eight times from 2026-10-18, four times before, to make the grey easier to see), then reprojected, cut into 32 levels (black for shadow, white for light, more opaque the stronger; flat ground and sea transparent) and written as WebP tiles with real transparency, zoom 3-8 (38 MB, gitignored, rebuilt with layers/copernicus_hillshade.py). It is a picture of relief, not a measurement.

The download's own licence.txt says "Your use of this data is subject to terms at www.ordnancesurvey.co.uk/opendata/licence. (c) Crown copyright and database right 2026."; the acknowledgement uses OS's standard "Contains OS data" form (the OS OpenData licence URL redirects to OGL v3.0, checked in #33). 50 m grid of height above Ordnance Datum, sea stored as exactly 0. Elevation layer: land cells only (0 treated as no data), averaged to 250 m, coloured with a fixed 0-900 m ramp, 9.8 MB of 8-bit PNG tiles, zoom 4-9. Checked against known heights at build: Ben Nevis 1345.1 m, Snowdon 1080.7 m, Scafell Pike 973.7 m, Holme Fen -3.5 m, Trafalgar Square 11.1 m. One tile (NR33, open sea) is stored as integers and skipped by the mosaic; the build asserts it is all zero. 65% of the grid is sea (0). Gitignored (large) and rebuilt with layers/os_terrain50.py.

The download's own licence.txt says "Your use of this data is subject to terms at www.ordnancesurvey.co.uk/opendata/licence. (c) Crown copyright and database right 2026."; the acknowledgement uses OS's standard "Contains OS data" form (the OS OpenData licence URL redirects to OGL v3.0, checked in #33). 50 m grid of height above Ordnance Datum, sea stored as exactly 0. Hillshade layer: computed from the full 50 m grid (multidirectional light, vertical exaggeration 3), averaged to 250 m, drawn as transparent black/white shading with flat ground fully transparent, 18.6 MB of 8-bit PNG tiles, zoom 4-9. Checked against known heights at build: Ben Nevis 1345.1 m, Snowdon 1080.7 m, Scafell Pike 973.7 m, Holme Fen -3.5 m, Trafalgar Square 11.1 m. One tile (NR33, open sea) is stored as integers and skipped by the mosaic; the build asserts it is all zero. 65% of the grid is sea (0). Gitignored (large) and rebuilt with layers/os_terrain50.py.

The BGS Open Government Licence page requires the acknowledgement "Contains British Geological Survey materials (c) UKRI [year]"; the year is not stated on the dataset page, so it is the year both releases were created (2022): flagged for the owner to confirm. 242,326 GB squares and 14,720 NI squares, each classed 1-6 by the highest radon potential inside the 1 km square (classes and official colours from the BGS User Guide, Table 3; the guide also credits the National Grid to Ordnance Survey). Indicative, not definitive: BGS points to the 1:50k Radon Potential dataset for that. Fields kept: class (CLASS_MAX / Class_max). Zoom 4-10, 10.7 MB, gitignored and rebuilt with layers/bgs_radon.py. Spot checks against the raw file: Redruth, Princetown (Dartmoor), Buxton and Bath class 6; Central London and Cambridge class 1.

The BGS Open Government Licence page requires the acknowledgement "Contains British Geological Survey materials (c) UKRI [year]"; the year is not stated on the dataset page, so it is the year the file was created (2024): flagged for the owner to confirm. Generalised from 1:50,000-scale mapping for the 2007 fifth edition bedrock map. 11,244 bedrock polygons. Coloured by geological age (from MAX_PERIOD, falling back to MAX_ERA/MAX_EON for Precambrian rocks), not rock type (86 distinct lithology descriptions, too many to colour usefully); colours are the official International Commission on Stratigraphy / Commission for the Geological Map of the World (CGMW) chronostratigraphic chart colours (stratigraphy.org/chart, CC BY 4.0 for the chart itself; our use of the BGS geology data remains under BGS's OGL v3.0), read from stratigraphy.org/ICSchart/CGMW_ICS_colour_codes.xlsx, not guessed. Independent check: six well-known locations queried directly against the raw file by true coordinate all match published geology (Dartmoor granite: Carboniferous-Permian; Lewisian Gneiss, Lochinver: Archaean; Lake District, Helvellyn: Ordovician volcanics; Yorkshire Dales, Malham: Carboniferous limestone; the Weald: Cretaceous; London: Palaeogene). Isle of Man is covered (Ordovician, matching the real Manx Group); the Republic of Ireland and Channel Islands are not. The GeoPackage also holds superficial deposits, dykes and fault lines, none built yet (see docs/DECISIONS.md #61). Gitignored and rebuilt with layers/bgs_bedrock_geology.py.

Same licence, credit and download as bgs-bedrock-geology (one GeoPackage holds both). 10,651 superficial (drift) polygons, classed by LEX_D into 14 deposit types (Till, Alluvium, Peat, Glacial Sand and Gravel, River Terrace Deposits, Raised Marine Deposits, Blown Sand, Clay-with-Flints, Sand and Gravel of Uncertain Age and Origin, Landslip, Brickearth, Lacustrine Deposits, Crag Group, plus a "Drift Geology Not Mapped" internal placeholder category). Colours are BGS's own, read from the GeoPackage's embedded QGIS style (layer_styles table, styleQML for 625k_V5_SUPERFICIAL_Geology), not chosen for this project. Independent check: four locations queried directly against the raw file by true coordinate match published geology (Holme Fen, Cambridgeshire: Peat; Ashridge, Chilterns: Clay-with-Flints; Huntly, Aberdeenshire: Till; the Stirling / Forth Valley carse: Raised Marine Deposits). Where bedrock is at or near the surface, this layer shows nothing, which is expected, not missing data. Rebuilt with layers/bgs_superficial_geology.py.

Same licence, credit and download as bgs-bedrock-geology (one GeoPackage holds bedrock, superficial deposits, dykes and faults). 3,263 dyke polygons (mostly unnamed igneous intrusions), coloured by geological age using the same ICS/CGMW colours and age rules as bedrock geology (Palaeogene 1,215; Silurian 1,005 [late Silurian to early Devonian, coloured by the oldest age]; Carboniferous 477; Palaeoproterozoic 239; Neoproterozoic 141; Ordovician 103; Archaean 32 [Lewisian Complex]; Devonian 32; Cambrian 19). Lightly simplified for web display. Built by pipeline/layers/bgs_dykes_faults.py (docs/DECISIONS.md #69).

Same licence, credit and download as bgs-bedrock-geology (one GeoPackage holds bedrock, superficial deposits, dykes and faults). 2,741 fault lines at rockhead: 2,603 faults and 138 thrust faults (barbs on the hanging-wall side). The source has no fault names (FLTNAME fields are empty for every feature) and no movement information, so the layer shows only the kind of fault. Lightly simplified for web display. Built by pipeline/layers/bgs_dykes_faults.py (docs/DECISIONS.md #69).

The BGS Open Government Licence page requires the acknowledgement "Contains British Geological Survey materials (c) UKRI [year]"; the year is not stated on the dataset page, so it is the year the record was last updated (2023): flagged for the owner to confirm. 4228 hexagons (2,026 Low / 1,510 Moderate / 692 Significant); fields kept: CLASS (1-3 read as Low/Moderate/Significant, matching the file's own Legend field) and the file's Advisory note. The file's own notice: "Not suitable for local or site specific analysis, data is generalised for national overview uses only". One of six hazards in one GeoPackage, each built as its own tile file (0.6 MB, zoom 3-9) by layers/bgs_geosure.py. The one-line hazard meanings in the layer text are plain-English summaries, not BGS wording.

The BGS Open Government Licence page requires the acknowledgement "Contains British Geological Survey materials (c) UKRI [year]"; the year is not stated on the dataset page, so it is the year the record was last updated (2023): flagged for the owner to confirm. 4233 hexagons (2,896 Low / 945 Moderate / 392 Significant); fields kept: CLASS (1-3 read as Low/Moderate/Significant, matching the file's own Legend field) and the file's Advisory note. The file's own notice: "Not suitable for local or site specific analysis, data is generalised for national overview uses only". One of six hazards in one GeoPackage, each built as its own tile file (0.6 MB, zoom 3-9) by layers/bgs_geosure.py. The one-line hazard meanings in the layer text are plain-English summaries, not BGS wording.

The BGS Open Government Licence page requires the acknowledgement "Contains British Geological Survey materials (c) UKRI [year]"; the year is not stated on the dataset page, so it is the year the record was last updated (2023): flagged for the owner to confirm. 4295 hexagons (3,837 Low / 176 Moderate / 282 Significant); fields kept: CLASS (1-3 read as Low/Moderate/Significant, matching the file's own Legend field) and the file's Advisory note. The file's own notice: "Not suitable for local or site specific analysis, data is generalised for national overview uses only". One of six hazards in one GeoPackage, each built as its own tile file (0.6 MB, zoom 3-9) by layers/bgs_geosure.py. The one-line hazard meanings in the layer text are plain-English summaries, not BGS wording.

The BGS Open Government Licence page requires the acknowledgement "Contains British Geological Survey materials (c) UKRI [year]"; the year is not stated on the dataset page, so it is the year the record was last updated (2023): flagged for the owner to confirm. 4233 hexagons (3,874 Low / 189 Moderate / 170 Significant); fields kept: CLASS (1-3 read as Low/Moderate/Significant, matching the file's own Legend field) and the file's Advisory note. The file's own notice: "Not suitable for local or site specific analysis, data is generalised for national overview uses only". One of six hazards in one GeoPackage, each built as its own tile file (0.6 MB, zoom 3-9) by layers/bgs_geosure.py. The one-line hazard meanings in the layer text are plain-English summaries, not BGS wording.

The BGS Open Government Licence page requires the acknowledgement "Contains British Geological Survey materials (c) UKRI [year]"; the year is not stated on the dataset page, so it is the year the record was last updated (2023): flagged for the owner to confirm. 4233 hexagons (3,704 Low / 1 Moderate / 528 Significant); fields kept: CLASS (1-3 read as Low/Moderate/Significant, matching the file's own Legend field) and the file's Advisory note. The file's own notice: "Not suitable for local or site specific analysis, data is generalised for national overview uses only". One of six hazards in one GeoPackage, each built as its own tile file (0.6 MB, zoom 3-9) by layers/bgs_geosure.py. The one-line hazard meanings in the layer text are plain-English summaries, not BGS wording.

The BGS Open Government Licence page requires the acknowledgement "Contains British Geological Survey materials (c) UKRI [year]"; the year is not stated on the dataset page, so it is the year the record was last updated (2023): flagged for the owner to confirm. 4233 hexagons (4,190 Low / 22 Moderate / 21 Significant); fields kept: CLASS (1-3 read as Low/Moderate/Significant, matching the file's own Legend field) and the file's Advisory note. The file's own notice: "Not suitable for local or site specific analysis, data is generalised for national overview uses only". One of six hazards in one GeoPackage, each built as its own tile file (0.6 MB, zoom 3-9) by layers/bgs_geosure.py. The one-line hazard meanings in the layer text are plain-English summaries, not BGS wording.

Licence and exact attribution wording copied from the ArcGIS item's own licenseInfo field (live, 2026-09-22) -- a different organisation and a plain OGL licence, unlike The Crown Estate's own England/Wales/NI offshore wind item, whose bespoke licence bans commercial gain and remains unused (see docs/DECISIONS.md #60). 58 sites: wind farm lease areas, OFTO transmission corridors, and substation/met-mast sites, across five project phases (Pre Planning, In Planning, Consented, Under Construction, Operational). Checked: Robin Rigg East's polygon (a well-known operational Solway Firth wind farm) is at the real, correct location by direct coordinate query. The service's unique-id field is OBJECTID_1, not the usual OBJECTID, so this was fetched directly rather than through common/arcgis.py's download_geojson helper (which assumes the latter and would need generalising for one dataset). Committed (0.31 MB, well under the project's gitignore threshold for large tiles) and rebuilt with layers/ces_offshore_wind_scotland.py.

Attribution wording: the gov.uk page states 'All content is available under the Open Government Licence v3.0, except where otherwise stated' and gives no fixed attribution sentence. 14,657 rows. Kept statuses Operational (3,132), Under Construction (457), Awaiting Construction (5,948) and Application Submitted (1,539); dropped refused, withdrawn, abandoned, expired, decommissioned, superseded ('Revised') and unsuccessful-appeal records, plus 36 rows with no coordinates: 11,040 points. Technologies grouped into 8 colour classes (the original technology label is kept; geothermal, 7 records, got its own colour on 2026-09-20). Fields kept: site, technology, capacity (MWelec), status, operator, country. Operator names are company names from the public register; no personal data is used. Re-run pipeline/layers/repd_renewables.py with a newer edition's CSV URL when DESNZ publishes one (they are dated, not stable URLs).

Attribution wording: the gov.uk chapter page states 'All content is available under the Open Government Licence v3.0, except where otherwise stated' and gives no fixed attribution sentence. Sheet '5.11 Full list', 1,364 stations, all with X/Y coordinates. Note 1 in the workbook: the list covers major power producer stations (other stations are in DESNZ's totals but not listed) and geographical information is published as supplied by the generators and not verified by DESNZ. Per-fuel capacity totals from the list (gas 33,676 MW, nuclear 5,883, onshore wind 12,953, offshore wind 16,885, solar 6,454, hydro 1,467, pumped storage 2,900, oil 1,044, biomass 3,316 + waste 522) are checked by the pipeline against DESNZ's own summary sheet '5.11 Summary' and it fails if they differ. The 'Changed and converted stations' sheet lists the closure of the last coal station (Ratcliffe-on-Soar, end of September 2024). Excel is read with GDAL's XLSX driver, so no extra Python package is needed. See pipeline/layers/dukes_power_stations.py and docs/DECISIONS.md #36.

The README inside the download (read 2026-10-16) says: "Copyright 2018-2021 World Resources Institute and Data Contributors" and "License: Creative Commons Attribution 4.0 International -- CC BY 4.0"; WRI's data portal record gives the same licence (cc-by) and the citation copied here (its "<date>" filled with the day it was accessed). The database holds grid-scale (1 MW and over) plants, one row per plant ("generally defined as a single transmission grid connection point"), "about 72% of the world's capacity" by its own estimate, built from many sources (national registers, Global Energy Observatory, wiki-solar and others); version 1.3.0 is the latest and was not updated afterwards. 10,889 of its 34,936 plants fall inside the 32 W to 45 E, 27 N to 72 N window. Checked at build: every fuel has a group, every capacity is above zero, Gravelines and Cattenom (France) are nuclear, large and where they should be; the per-fuel totals in the window are printed. Known gap, measured here: wind and solar are far from complete (Germany holds about 5,000 MW of wind and 6,800 MW of solar in the database), so the layer is for locating plants and comparing within a fuel, not for adding up. The year a plant started is present for about a quarter of the rows. Processing: plants in the window written as points with country, fuel group (the database's 15 fuels grouped into ten), the fuel as the database names it, capacity, year, owner and reporting source, and tiled with tippecanoe at zoom 2 to 12 (about 8 MB).

The site's Creative Commons page (read 2026-11-11): "Ember content is released under a Creative Commons Attribution Licence (CC-BY-4.0)... free to share and adapt... as long as you credit us" -- unlike Eurostat's own energy tables, which bar commercial reuse for any country outside the EU/EFTA/candidate list (docs/DECISIONS_LOG.md #265's "obvious filter would have lied" is the same family of check, done here for Ember instead; the actual public ZIP downloads anonymously, no account or licence-acceptance form). Two figures used: NET-P (net import, negative, or net export, positive, as a percentage of a country's own demand) and RES-E (variable renewable -- solar and wind -- share of generation), both 2024 monthly averages from country_monthly_chart_2024.csv, turned into one annual average per country here; and REF_NTC.csv's 2024 Net Transfer Capacity (MW) between every pair of neighbouring countries, both directions, read straight (not averaged). REF_NTC.csv holds 2024 and the 2030/2040 "Reference" scenario in one file, distinguished only by its own Year column: only 2024, the real, not-projected year, is read. Three "hub" codes (Baltic Hub, Celtic Hub, North Hub) are notional offshore-wind groupings, not real countries, and are left out of the border lines. Checked against known facts: France a net exporter, Italy and the UK net importers, all confirmed by the built 2024 figures. Rebuilt with layers/europe_electricity.py.

Attribution wording: the road traffic statistics site footer states 'All content is available under the Open Government Licence v3.0, except where otherwise stated' (c) Crown copyright, and data.gov.uk lists 'GB Road Traffic Counts' (DfT) under the UK Open Government Licence; no fixed attribution sentence is given. FLAGGED: the road geometry (Major Road Database) is probably derived from Ordnance Survey data, but no OS acknowledgement is stated on the DfT pages we could read (the metadata PDF could not be read as text); owner to confirm before launch. 17,941 junction-to-junction segments (1,120 motorway, 16,821 A road) each joined by count point id to its AADF for every year 2000-2025 (all motor vehicles); all joined. Checks in the build: busiest M25 segment in 2024 is 210,436 vehicles a day (expected roughly 180,000-240,000), and on the same segments traffic in 2020 was 76.7% of 2019 (DfT reports roughly a fifth lower). 24% of the 2025 major-road figures are direct counts, the rest DfT estimates. The 11 MB tile file is gitignored; rebuild with pipeline/layers/dft_road_traffic.py.

Every bus stop, rail station, metro/tram stop, ferry terminal and taxi rank in Great Britain: 435,641 rows in the full file, 388,066 marked Status=active (the rest inactive or pending, dropped). Of those, 31,410 (mostly on-street bus stops, checked by StopType) carry no coordinate at all in either column -- a real gap in what NaPTAN records for those rows, not a parsing bug -- leaving 356,656 mapped stops (344,508 bus, 6,184 rail, 3,942 metro/tram, 1,102 ferry, 796 taxi, 124 air). Rail/metro/ferry/air/taxi nodes are given a low tippecanoe per-feature minzoom (5) so they are visible at a country-wide view; ordinary bus stops only appear from city zoom onward (minzoom 11) and are still thinned by density at very dense zooms, the same "hierarchy, not random thinning" approach already used for the DfT road traffic layer (#40). 28.6MB of tiles, gitignored, rebuilt with pipeline/layers/naptan.py.

2,589 stations in the ORR table; joined to NaPTAN's rail/metro points (naptan.py) by name, normalising "Rail Station"/"Underground Station" suffixes, "&" vs "and", and footnote markers, since the two sources spell many station names differently. 2,570 of 2,589 (99.3%) matched this way; the rest are left off rather than guessed at. The dataset's own page previously read as stating no licence (checked 2026-09-20) -- re-checked 2026-09-29 and ORR's site-wide terms page states OGL v3.0 clearly, though the dataset page itself is a JavaScript application our tools cannot render directly, so this rests on the site-wide statement rather than page-specific wording. Top-ranked stations (London Liverpool Street, London Waterloo, London Paddington) match public knowledge, an independent sanity check the build does not itself assert but was checked by hand before shipping.

Freight handled at the UK's major ports, both directions, in thousand tonnes, for every year 2000 to 2025, carried as properties on one point per port so the year slider restyles the layer without new tiles (like the road traffic layer). The total is DfT's own "All cargo" figure; the split into liquid bulk, dry bulk, roll-on roll-off, containers and other general cargo is the sum of DfT's detailed cargo codes (11-19, 21-29, 51-69, 31-34 and 91-99), checked to add back to the total within 2% for every port and year. London's 2025 total of 56.6 million tonnes matches the DfT map table (PORT0304). Weight, not value. Ports that never reached a million tonnes in any year are left off (about 40 small ones). DfT notes kept in the layer's caveats: Fleetwood (ferries ended 2010) and Stranraer (closed 2011, moved to Loch Ryan) end; London and Hull revisions; 2020 affected by COVID-19 and the end of the UK's EU membership; containers before 2005 not directly comparable; from 2025 wood pellets and wind-farm products are counted separately. 53 positions: 2 from Natural Earth's ports (Southampton, Liverpool), 51 the well-known position of the port's main terminal (marked MANUAL in pipeline/layers/dft_ports.py); the coordinates are not DfT's.

37,209 line segments of the National Cycle Network, National routes, Regional routes and short Link routes. The publisher renamed from Sustrans to Walk Wheel Cycle Trust; the item's own credit text uses the new name. A live-refreshed source (updated weekly by the publisher), not a static annual snapshot. The service's unique-id field is `FID`, not the usual `OBJECTID` (the same recurring quirk already met in Scotland offshore wind, #63, NI ASSI, #81, and NI bathing water areas) -- paged directly with a small bespoke loop rather than generalising the shared ArcGIS helper. 14.1MB of tiles, gitignored, rebuilt with pipeline/layers/sustrans_ncn.py.

101,525 personal-injury road collisions reported in Great Britain in 2025; 101,523 carry a coordinate (2 do not and are dropped). Severity is coded 1/2/3 in the source (Fatal/Serious/Slight), decoded here. A different question from the existing road traffic layer (dft-road-traffic, #40), which shows how much traffic a road carries, not where collisions happen. Thinned by density at low zoom, the same approach already used for the listed buildings and bathing-water point layers, since a year of collisions is dense and fairly evenly spread rather than following NaPTAN's stop hierarchy. 16.4MB of tiles, gitignored, rebuilt with pipeline/layers/stats19_collisions.py.

Total goods exports and imports (all 123 commodity categories summed via the workbook's own "T Total" row, not summed by hand, so sub-categories are never double-counted with their parent category) for the UK's top 90 trading partners by 2025 combined trade (99% of the value of UK goods trade), every year 1997-2025 carried as a property so the year slider restyles the layer without loading new tiles. Aggregate rows (Whole world, EU totals) are excluded -- only real countries are shown. 26MB of tiles; small enough to commit rather than gitignore. The first version carried the top 40; it was widened to 90 on 2026-10-08 because 40 left out big partners such as Turkey, Indonesia, Nigeria and Algeria. Jersey and Guernsey are in the ONS figures but not in HMRC's port data, so they have lines but no mode or goods detail; Serbia is code XS in HMRC's tables. How goods crossed the border (added 2026-10-06, docs/DECISIONS_LOG.md #150): HMRC's Overseas Trade in Goods Statistics (OTS, Open Government Licence v3.0, "(c) Crown copyright" per the dataset record) record the port or airport of every declaration. Grouped by id into air (airports and airbases, ids 419-482), the Channel Tunnel and rail freight terminals, sea (every named seaport) and other (port not collected, estimates, postal, free zones, the Irish land border). Value and net mass summed by flow (exports = EU + non-EU exports, imports likewise), partner and year 2000-2025. It is the port, not the vehicle: a lorry on a ferry is sea. EU trade has no port on record before 2021 (exports) and 2022 (imports), so an icon is drawn only where at least 70% of the value has a recorded port. HMRC's trade-basis totals differ from ONS's balance-of-payments totals used for the line widths, so only shares are shown from this source. Lorry icon (added 2026-10-08, docs/DECISIONS_LOG.md #163): HMRC names the port, not the vehicle, so seaports are split by what the DfT says they mostly handle (layers/hmrc_trade_modes.py, FERRY_SHARE): a port where roll-on roll-off is at least 35% of its tonnage in 2025 (Dover, Immingham and Grimsby, Harwich, Holyhead, Heysham, Portsmouth, Fishguard, Poole, Belfast) is a ferry port, shown with a lorry, and every other seaport a ship. That is a stand-in for lorries on ferries, not a count of lorries, and it is judged by tonnes, which understates roll-on roll-off at mixed ports such as Hull. Lorry and train journeys on the click panel are drawn London, Dover (or the Channel Tunnel), Calais, then along hand-placed legs between about 90 European cities (pipeline/layers/land_lanes.py, docs/DECISIONS_LOG.md #164) to the country's capital or main entry city, with the Little Belt, Great Belt and Oresund bridges for Denmark and Sweden (Ireland: Holyhead to Dublin). They are drawing aids, straight lines between cities, not motorways or tracks. A lorry or train icon is shown only for a country such a route reaches; for any other (Morocco, Colombia, Malta...) goods that left by a ferry port or the tunnel go on by ship, so they count as sea. How "Play this route" travels (added 2026-10-07, docs/DECISIONS_LOG.md #159): the ship, plane and train routes on the click panel are drawing aids, not records. Ship routes follow about 150 hand-placed open-water legs (pipeline/layers/sea_lanes.py), each checked against Natural Earth's 1:10m land polygons so none cuts across land (the Suez Canal, and the Dardanelles, are the only exceptions, said in the code and tests), from Felixstowe, Southampton or Liverpool to the partner country's main container port. Flights run by great circle from London Heathrow to that country's main airport. Port and airport positions come from Natural Earth's 1:10m ports and airports (public domain, naturalearthdata.com, downloaded 2026-10-07) and were checked by name; five ports (Aarhus, Genoa, Busan, Port Klang, Constanta) and two airports (Larnaca, Bucharest) that Natural Earth does not list use the cities' well-known coordinates. A country with no coast is shown with the sea leg ending at its usual port in a neighbour (Rotterdam or Bremerhaven), and the goods go on overland. Journey times are the real length of each drawn route at a typical 40 km/h for a container ship and 850 km/h for a cargo jet. Real voyages vary with the carrier, weather and canal closures. Prices (added 2026-10-08, docs/DECISIONS_LOG.md #172): the card's Prices switch restates every year in 2025 pounds with the ONS consumer price index, series D7BT "CPI INDEX 00: ALL ITEMS 2015=100" (dataset MM23, Open Government Licence v3.0, annual averages, release of 15 September 2026, read live from https://www.ons.gov.uk/economy/inflationandpriceindices/timeseries/d7bt/mm23/data on 2026-10-08; pipeline/layers/ons_cpi.py writes web/public/layers/uk-cpi.json). The factor for a year is the 2025 index divided by that year's (1997: 1.97; 2020: 1.27). It is general consumer inflation, not the price of traded goods, said on the layer. One product at a time (added 2026-10-08, docs/DECISIONS_LOG.md #172): the card's Product picker redraws the map for one HS chapter (two-digit commodity code) from the same HMRC OData records as the goods lists, but for every chapter, not just the top five (pipeline/layers/hmrc_trade_goods.py writes web/public/layers/trade-products/<chapter>.json: exports and imports per partner and year, 2000-2025, GBP million, HMRC's own trade-basis values, which differ a little from ONS's totals). Figures are drawn at a scale of their own, so the biggest partner for a product stands out. A data fix made on the same day: HMRC lists the United States under two ids (902 "USA" to 2008, 400 "United States" from 2009), and the first version read only 902, so the USA had no port, mode or goods records after 2008; both ids are now read (layers/hmrc_trade_modes.py country_ids, test_trade_modes.py). Services (added 2026-10-08, docs/DECISIONS_LOG.md #172): ONS "UK trade in services: service type by partner country, non-seasonally adjusted" (Q1 2026 edition, released 24 July 2026, Open Government Licence v3.0; landing page https://www.ons.gov.uk/businessindustryandtrade/internationaltrade/datasets/uktradeinservicesservicetypebypartnercountrynonseasonallyadjusted, file https://www.ons.gov.uk/file?uri=/businessindustryandtrade/internationaltrade/datasets/uktradeinservicesservicetypebypartnercountrynonseasonallyadjusted/current/servicetypebycountry2026q1.xlsx, read 2026-10-08). Its "Sheet 1. Time Series" gives exports and imports in GBP million, current prices, for the 13 top-level service types and each country, annually 2016-2025. Each service type is a "product" of the map (web/public/layers/trade-products/S<n>.json, pipeline/layers/ons_services.py). Only the 61 of the 90 goods partners that ONS lists as a country of their own are drawn; ONS groups the rest (the Gulf states, Kenya, Nigeria...) into "other" rows that cannot sit on one country. Cells ONS withholds as confidential ("C") are left out. "United States" is ONS's "United States inc Puerto Rico". Services are not added to the goods totals.

Visits (trips) and spending by country of residence (inbound) and by main country visited (outbound), all purposes, from tables 9 and 7; how each country's visits were routed from the port tables 12 and 10 (airports summed to "air", sea ports, Channel Tunnel). Both workbooks carry the 4 September 2026 correction: ONS fixed an error in how destination airports were assigned for air passengers. Estimates are official statistics in development (2024 is a composite year of the old and new survey methods; fewer interviews in April to June); Great Britain only because Northern Ireland has been supplied by NISRA since April 2024. Group rows ("Other Asia", "Rest of Europe", "EU") are not drawn. "Other China" is read as China apart from Hong Kong, which the survey lists separately (an interpretation: the workbook's definitions sheet does not define it). Values the survey suppresses (fewer than 3 respondents) count as unallocated in a route's shares. Route icons show the port used, not the vehicle. The layer was built from the two workbooks' own totals, which the pipeline carries unchanged: for example France 3,520,000 visits to Great Britain and 9,059,000 visits by Great Britain residents. How "Play this route" travels (added 2026-10-07, docs/DECISIONS_LOG.md #159): the ship, plane and train routes on the click panel are drawing aids, not records. Ship routes follow about 150 hand-placed open-water legs (pipeline/layers/sea_lanes.py), each checked against Natural Earth's 1:10m land polygons so none cuts across land (the Suez Canal, and the Dardanelles, are the only exceptions, said in the code and tests), from Felixstowe, Southampton or Liverpool to the partner country's main container port. Flights run by great circle from London Heathrow to that country's main airport. Port and airport positions come from Natural Earth's 1:10m ports and airports (public domain, naturalearthdata.com, downloaded 2026-10-07) and were checked by name; five ports (Aarhus, Genoa, Busan, Port Klang, Constanta) and two airports (Larnaca, Bucharest) that Natural Earth does not list use the cities' well-known coordinates. A country with no coast is shown with the sea leg ending at its usual port in a neighbour (Rotterdam or Bremerhaven), and the goods go on overland. Journey times are the real length of each drawn route at a typical 40 km/h for a container ship and 850 km/h for a cargo jet. Real voyages vary with the carrier, weather and canal closures. Spending and ranges (added 2026-10-08, docs/DECISIONS_LOG.md #172): the spending measures use the same tables 9 and 7 (GBP million, 2024). The 95% ranges for visits are ONS's own, from table 21 (visits to Great Britain by country of residence) and table 16 of the other workbook (visits abroad by main country visited); ONS builds them from the survey alone, so they can differ a little from the published estimates (its note 3), and publishes no ranges for what Britons spend abroad ("[x]"). The ranges are written into the click panel and the partner table; nothing is estimated here.

The gov.uk page states "All content is available under the Open Government Licence v3.0, except where otherwise stated"; the attribution sentence is the standard OGL one, as the page gives none of its own. The boundary statements are the two from https://www.ons.gov.uk/methodology/geography/licences, the OS year taken as the year the boundary item was created (2024), as for the council maps: FLAGGED for the owner with the others. The map is coloured by the decile of the overall index (1 = the most deprived tenth of England's LSOAs); the click panel adds the overall rank and the deciles of the seven domains (income, employment, education, health, crime, barriers to housing and services, living environment). Codes in the CSV and the boundaries are the same 33,755 (the pipeline fails otherwise), and each decile holds 3,375 or 3,376 areas. Zooms 6 to 8 carry the deciles and council name only, with stronger simplification, to keep tiles near 500 kB; every field from zoom 9. Deciles rank areas within England in 2025: they say which areas are more deprived than others, not by how much, and not that everyone in an area is deprived. Wales, Scotland and Northern Ireland have their own indices on their own scales, which are not comparable with this one. Rebuilt with layers/iod_england.py.

StatsWales links to the Welsh Government copyright statement (https://www.gov.wales/copyright-statement), which allows re-use "under the terms of the Open Government Licence" (logos excepted) and gives no wording of its own, so the standard OGL sentence is used. The table holds a rank, decile, quintile, quartile and group for the overall index and eight domains for each of the 1,917 LSOAs, plus 22 council rows without figures (used for council names). The map uses the published deciles (1 = the most deprived tenth of Wales); the click panel adds the overall rank and each domain's decile. The domain weights are Wales's own (income 22%, employment 20%, health 15%, education 14%, access to services 10%, housing 9%, community safety 5%, physical environment 5%, from the StatsWales summary), so a Welsh decile is not comparable with an English one. Codes match the ONS LSOA 2021 boundaries exactly (the pipeline fails otherwise); rank 1 is Rhyl West 2. Rebuilt with layers/wimd_wales.py.

The collection page states "All content is available under the Open Government Licence v3.0, except for graphic assets and where otherwise stated" and that "SIMD 2020v2 ranks should now be used"; no later collection exists (2024, 2025 and 2026 addresses return 404). The Data Zone Boundaries 2011 record requires "Copyright Scottish Government, contains Ordnance Survey data © Crown copyright and database right (insert year)"; the year is the year the file was last updated (2021, the record's publication date and the shapefile's own date), the convention in docs/LICENCE_AUDIT.md: FLAGGED with the others. The map uses the published overall decile; domain deciles are worked out from the domain ranks as ceil(rank x 10 / 6,976), the rule that reproduces the published overall decile for all 6,976 data zones (checked on every run; tied domain ranks are halves). Data zones held 778 people on average in 2017 (the look-up's population column). Rank 1 is Greenock Town Centre and East Central - 02. Rebuilt with layers/simd_scotland.py.

NISRA's Crown copyright page allows re-use under the Open Government Licence "not including logos and Land & Property Services mapping data" and gives the attribution example "Source: NISRA : Website: www.nisra.gov.uk", used here. Because of that exclusion the boundaries were checked on their own: OpenDataNI's record "NISRA Open Data Boundaries - Super Output Areas 2001", which links this same NISRA zip, states "UK Open Government Licence (OGL)" version 3.0 (read 2026-10-23 through its API). NISRA publishes ranks only; deciles are worked out as ceil(rank x 10 / 890), 89 areas each. The workbook's codes include whole-ward (W) and Belfast city-centre (C) areas as well as ward parts (S); all 890 match the boundaries. Rank 1 is East (Derry City and Strabane). Rebuilt with layers/nimdm_ni.py.

The footer of the ONS dataset page for TS045 (read 2026-10-24) says "All content is available under the Open Government Licence v3.0, except where otherwise stated"; the credit sentences are the standard OGL one, the ONS statement and the OS statement for the boundaries (the OS year taken as the year the boundary item was created, 2024, as for the deprivation layers: FLAGGED for the owner with the others). Seven shares are worked out from published counts: households with no cars or vans / all households (TS045), households that rent privately and households that rent from a council or housing association, each / all households (TS054, the "Private rented" and "Social rented" groups), people who mainly work at or from home / all residents aged 16 and over in employment the week before the census (TS061), households with no central heating of any kind / all households (TS046), households whose occupancy rating for bedrooms is -1 or -2 or less, added together / all households (TS052), and residents answering "Bad health" or "Very bad health", added together / all usual residents (TS037). The 35,672 LSOA codes in each table and in the boundaries are the same set (the pipeline fails otherwise), and the neighbourhoods add up to the England and Wales figure in each table's country file (23.31% no car, 20.27% renting privately, 17.06% renting from a council or housing association, 31.22% working from home, 1.48% with no central heating, 4.25% overcrowded, 5.25% in bad or very bad health). The ONS swaps some records and adds small random changes to counts (cell key perturbation), so one neighbourhood's share can be a little off. Census day, 21 March 2021, was in the third national lockdown with many people on furlough, so the work-from-home share is far above a normal year; said on the layer. Health gets worse with age and the health share is not age-adjusted; said on the layer. Scotland's census was held in 2022 and is combined with this one on the same seven Great Britain-wide layers (census-gb), not its own (owner request 2026-11-06, docs/DECISIONS_LOG.md #262); Northern Ireland publishes these topics only by 2014 electoral ward and its boundary licence is stated two ways, so it is not used (docs/OWNER_NEEDS.md). Zooms 6 to 8 carry the seven shares and the council name only, with stronger simplification; every field from zoom 9. About 44 MB, gitignored, and rebuilt with layers/census_ew.py; census_gb.py combines it with Scotland's into the layers the web app actually reads.

The footer of scotlandscensus.gov.uk (read 2026-10-24) says "All content is available under the Open Government Licence v3.0, except for graphic assets and where otherwise stated"; the credit sentences are the standard OGL one, the source line, and the Scottish Government's own line for the boundaries, which is the map service's copyright text (the year, 2024, is therefore not guessed). The boundary record at spatialdata.gov.scot gives the statement with "(insert year)"; the service's text fills it. Seven shares are worked out from published counts: households with no cars or vans / all occupied households (UV405), households in "Private rented: Total" and households in "Social Rented: Council (LA) or Housing Association/ Registered Social Landlord", each / all occupied households (UV404), people who work mainly at or from home / all people aged 16 and over in employment the week before (UV701), households with "No central heating" / all occupied households (UV407), households at "Occupancy rating of bedrooms: -1 or less" / all occupied households (UV415, a single line where England and Wales separate -1 from -2 or less), and people answering "Bad" or "Very bad", added together / all people (UV302); a "-" in a table is a zero, and some row labels carry a leading tab. The 7,392 codes in each table, the boundaries and the lookup are the same set (the pipeline fails otherwise); the zone populations add up to 5,439,565, within 0.1% of NRS's published Census day estimate of 5,436,600 (the counts are perturbed); the boundary file's household counts agree with the car table's total (2,509,248). Scotland-wide shares: 26.37% no car, 12.88% privately rented, 31.59% working from home. Scotland's census was held a year after England and Wales's, with its own questions, tables and areas, but combined with England and Wales's into one Great Britain-wide layer each (census-gb), on the same classes and colours (owner request 2026-11-06, docs/DECISIONS_LOG.md #262, reversing the earlier choice to keep the two nations as separate layers). About 15 MB, gitignored, rebuilt with layers/census_scotland.py; census_gb.py combines it with England and Wales's into the layers the web app actually reads.

The item's licenseInfo reads "Use of this data is subject to the Open Government Licence"; its accessInformation gives "© Natural England copyright. Contains Ordnance Survey data © Crown copyright and database right [year]", the year taken as the item's last update (2023), the convention in docs/LICENCE_AUDIT.md: FLAGGED with the others. 1,826 polygons: Grade 1 to Grade 5, Non Agricultural, Urban and Exclusion (the service's own layer is titled "ALC Grades (Provisional) © ADAS & Defra"). Downloaded at full detail: with the service's own generalisation, and before the fix for holes described in docs/DECISIONS_LOG.md #250, grades overlapped (a point near Ely read grade 2 and 3; the service itself says grade 3). After the fix, 2,000 random points across the Midlands and the South each fall in at most one grade. Checked by point: Dartmoor and Helvellyn grade 5, Fen farmland near March grade 1, central London Urban. Rebuilt with layers/na_alc.py.

The catalogue record states "License CC-BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Copyright holder: European Environment Agency (EEA)" and "no limitations to public access"; it gives no citation sentence, so ours is built from the title, year and record address, and the credit names the changes made, as CC BY asks. The record says the maps "are assessment tools for use at the European scale. They may show results which differ from assessments made at national scale and should not be used to check formal compliance with regulated standards" (repeated in a caveat). Annual mean in µg/m3 on a 1 km grid; classes with an edge at the WHO 2021 guideline (10 µg/m3). Checked by point when built: Milan 31.6, Paris 26.8, Scottish Highlands 0.9, Engadin 2.0. No data for Ukraine, Belarus, Russia, Moldova, the Caucasus or North Africa. Validated maps exist for 2010 to 2024; 2010 and 2024 are used, as a pair (docs/DECISIONS_LOG.md #252), see the 2010 entries. Rebuilt with layers/eea_air_europe.py.

The catalogue record states "License CC-BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Copyright holder: European Environment Agency (EEA)" and "no limitations to public access"; it gives no citation sentence, so ours is built from the title, year and record address, and the credit names the changes made, as CC BY asks. The record says the maps "are assessment tools for use at the European scale. They may show results which differ from assessments made at national scale and should not be used to check formal compliance with regulated standards" (repeated in a caveat). Annual mean in µg/m3 on a 1 km grid; classes with an edge at the WHO 2021 guideline (5 µg/m3). Checked by point when built: Krakow 21.5, Milan 19.8, Scottish Highlands 3.7, Engadin 1.8; cells marked -99 (no estimate: Turkey) left transparent. No data for Ukraine, Belarus, Russia, Moldova, the Caucasus or North Africa. Validated maps exist for 2010 to 2024; 2010 and 2024 are used, as a pair (docs/DECISIONS_LOG.md #252), see the 2010 entries. Rebuilt with layers/eea_air_europe.py.

The catalogue record (read 2026-10-24, with the one for 2024) states "License CC-BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Copyright holder: European Environment Agency (EEA)" and "no limitations to public access"; it gives no citation sentence, so ours is built from the title, the revision year (2022) and the record address. The lineage text is the same as the 2024 record's apart from the method: the 2010 record cites "the methodology and the corresponding maps ... in the 2021 Eionet report 'European air quality maps for 2019 ...' (ETC/ATNI Report 1/2021)" and the 2024 record cites ETC HE report 2025/5 on the 2023 maps. Both list the same inputs (reported monitoring stations, EMEP stations, altitude, ECMWF meteorology, EMEP model results, satellite data) and the same regression-interpolation-merging method; neither says the two methods are identical. The 2010 map is therefore shown beside the 2024 one and as their difference, with that said on the layers. Same classes and colours as the 2024 layer. Checked by point when built: Milan, Paris, Scottish Highlands and Engadin (ranges in the script). No data for Turkey, Ukraine, Belarus, Russia, Moldova, the Caucasus or North Africa (the 2010 grids have none there). Rebuilt with layers/eea_air_europe.py.

The catalogue record (read 2026-10-24, with the one for 2024) states "License CC-BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Copyright holder: European Environment Agency (EEA)" and "no limitations to public access"; it gives no citation sentence, so ours is built from the title, the revision year (2022) and the record address. The lineage text is the same as the 2024 record's apart from the method: the 2010 record cites "the methodology and the corresponding maps ... in the 2021 Eionet report 'European air quality maps for 2019 ...' (ETC/ATNI Report 1/2021)" and the 2024 record cites ETC HE report 2025/5 on the 2023 maps. Both list the same inputs (reported monitoring stations, EMEP stations, altitude, ECMWF meteorology, EMEP model results, satellite data) and the same regression-interpolation-merging method; neither says the two methods are identical. The 2010 map is therefore shown beside the 2024 one and as their difference, with that said on the layers. Same classes and colours as the 2024 layer. Checked by point when built: Krakow, Milan, Scottish Highlands and Engadin (ranges in the script). No data for Turkey, Ukraine, Belarus, Russia, Moldova, the Caucasus or North Africa (the 2010 grids have none there). Rebuilt with layers/eea_air_europe.py.

The item's licenseInfo reads "Use of this data is subject to the Open Government Licence"; the credit is the item's accessInformation, copied as written. Its description defines peaty soil as soil with 20% or more organic content and says the layer maps where the England Peat Map model gives a probability above 50%; the model (NERR149, Natural England, 2025) was trained on survey data with satellite imagery, lidar, geology and historic land use as predictors. 249,743 polygons, one class, 11,047 km2 measured in British National Grid. The depth and probability rasters are a tiles-only image service that does not allow tile export, so they are not used. Downloaded through the fix for ArcGIS holes (docs/DECISIONS_LOG.md #250). Rebuilt with layers/na_peat.py.

The item's licenseInfo (read 2026-10-24) says "There are no public access constraints to this data. Use of this data is subject to the Open Government Licence"; its credit reads "(c) Natural England copyright. Contains Ordnance Survey data (c) Crown copyright and database right [year]", the year filled with 2023, the year the item was last updated (the convention in docs/LICENCE_AUDIT.md: FLAGGED for the owner with the others). 29,858 line features. The item describes rivers and streams that "exhibit a high degree of naturalness", scored on four components of habitat integrity (hydrological, physical, physico-chemical, biological) with headwaters judged from land cover; the other route into the map is a waterbody with high ecological and morphological status where monitoring was too thin to score, and sections added by local experts. The script sorts each line into one of those three routes from the service's own description text and fails on a description it does not know: 16,269 lines high naturalness, 288 high status with too little monitoring, 13,301 added locally; 2,961 lines are in waterbodies designated Heavily Modified. Lengths quoted in the habitats story are the service's own sums of Shape__Length (7,237 km in all, 3,967 km high naturalness). Tiled at zoom 6 to 12 (3 MB) and rebuilt with layers/na_priority_rivers.py.

Built specifically as the honest, verifiable companion to the "From empire to Commonwealth" story: this shows TODAY's Commonwealth membership, a voluntary association, not the historical empire. Several former British territories are not members today (the United States, Ireland, Myanmar...) and five members (Mozambique, Rwanda, Gabon, Togo, Cameroon) were never British colonies at all -- both facts are stated on the layer and used as the story's own honest closing point, rather than implying the shaded countries were all once part of the empire. No open, commercial-use-permitted dataset of historical colonial boundaries was found (CShapes 2.0, the standard academic dataset, is CC BY-NC-SA -- non-commercial only, ruled out under the same guardrail as iNaturalist and the Crown Estate's England offshore wind data), so no attempt is made to map the empire's own historical extent.

Combined open country + registered common land + Section 16 dedicated land, with MoD byelaw land, Section 28 exclusions, racecourses/aerodromes removed (per the item description). 42,059 features in the service; 42,045 had geometry and were kept. Derived property access_type from the OC/RCL/S16 flags (open country wins, then common land, then dedicated). Server-side generalisation ~30 m; tiles 4.7MB at zoom 5-11. NOTE: Natural England's 'Access Network Mapping' dataset in the same catalogue is licensed for NON-COMMERCIAL use only and was deliberately not used.

Licence is CC BY 4.0 per the item's licenseInfo; the credit line is the item's own accessInformation text plus the licence name. The item description states the data 'has been captured at approximately 1:50,000 scale and as such is not intended to show detailed ownership boundaries' and may include excepted land (buildings, gardens, cropped fields), so the layer never claims to show ownership. 1,173 Always Open + 519 Limited Access polygons, generalised ~20 m, 0.67MB.

415 polygon features, generalised ~20 m, 0.43MB at zoom 6-12. Licence and attribution copied from the ArcGIS item.

The 14 National Trails in England (Pennine Way, Offa's Dyke Path, South Downs Way, The Ridgeway, North Downs Way, Yorkshire Wolds Way, Peddars Way and Norfolk Coast Path, Thames Path, Hadrian's Wall Path, Cotswold Way, Pennine Bridleway, Cleveland Way, South West Coast Path, Coast to Coast). The item's own accessInformation gives the standard Natural England credit with a "[year]" placeholder, the same pattern already used for Priority Habitats, GeoSure and the radon atlas -- filled with the retrieval year (2026) since no other year is stated on the item. Northern Ireland has no National Trails; Scotland's Great Trails and Wales's own long-distance paths are not published as one clean combined open dataset (checked 2026-09-29: Scotland's routes are scattered across many separate council/NatureScot layers with inconsistent licensing) -- Wales's own coast path IS shown separately (nrw-coast-path), a genuine open single-dataset match, unlike Scotland's fragmented sources.

Same source item, licence and credit as he-nhle (its accessInformation gives the credit; its licenseInfo points to the Historic England Open Data Hub Terms and Conditions, whose page returned 403 to our fetch tool, flagged there for a manual look before any monetisation). 379,685 listed building points (Grade II 348,219; II* 22,125; I 9,341), one per record, from layer 0 of the service (MultiPoint geometry with a single point, converted to Point). Tiled at zoom 9 to 14 with density thinning at low zoom (72 MB, gitignored, rebuild with pipeline/layers/listed_buildings.py). Built by pipeline/layers/listed_buildings.py (docs/DECISIONS.md #71).

Licence and credit read from the download's own bundled Listed_Buildings.txt (OGL v3.0; the credit ends 'Crown copyright and database right [year]', the year filled as 2026 to match the other Historic Environment Scotland layers; and 'This data is indicative only', repeated in the layer caveats). 67,474 points (Category B 34,484; C 26,505; A 6,485) representing 46,747 distinct designation references, since some listings cover several buildings or parts. Tiled at zoom 7 to 14 with density thinning at low zoom (15 MB, gitignored). Built by pipeline/layers/listed_buildings.py (docs/DECISIONS.md #71).

Same licence statement and attribution form as cadw-designations (read from each layer's own DataMapWales metadata; DATE filled as the download date). 30,116 records: 30,113 Designated (Grade II 27,476; II* 2,142; I 494; one without a grade) and 3 under Interim Protection, which are left out. Tiled at zoom 7 to 14 with density thinning at low zoom (7 MB, gitignored). Built by pipeline/layers/listed_buildings.py (docs/DECISIONS.md #71).

Same licence and credit as hed-ni-designations (DfC Historic Environment Digital Datasets page: Open Government Licence). The Historic Buildings Record holds 15,862 records including buildings that are not listed; only the 9,198 with a listed grade are drawn (A 224; B+ 679; B1 3,434; B2 3,843; B 1,018). The 6,664 left out are Record Only (6,171), D1 Record Only (485) and Delisted (8); five records have an unusable position (all Record Only), so no listed building is lost. Tiled at zoom 7 to 14 (3.5 MB, gitignored). Built by pipeline/layers/listed_buildings.py (docs/DECISIONS.md #71).

The item's accessInformation reads '© Crown Copyright 2026. Contains Ordnance Survey data © Crown copyright and database right 2026. Released under OGL.' and its licenseInfo points to the Historic England Open Data Hub Terms and Conditions; data.gov.uk lists the sibling registers as OGL v3.0. The Terms page itself returned 403 to our fetch tool, so its exact wording was NOT read directly, worth a manual look before any monetisation. Counts: scheduled monuments 20,001 (8.1MB), parks and gardens 1,721, battlefields 47, World Heritage Sites 28. Listed buildings (379,685 points) are not yet included. Server-side generalisation ~5 m.

Licence and attribution verified independently for each of the four datasets, from each zip's own bundled .txt licence statement (not assumed from one to cover all four): "The [dataset name] data is licensed under UK Open Government Licence... Any reproduction of this data must be attributed with the following: Contains Historic Environment Scotland and Ordnance Survey data (c) Historic Environment Scotland - Scottish Charity No. SC045925 (c) Crown copyright and database right [year]." The [year] is not stated in the licence text itself; 2026 (the retrieval year) is used, flagged for the owner to confirm, same open question as other BGS/HES-style "[year]" placeholders in this project. Counts: 8,074 Scheduled Monuments, 371 Gardens and Designed Landscapes, 40 Battlefields, 10 World Heritage Site records (Scotland's 6 UNESCO sites, some split into separate Boundary and Buffer Zone polygons -- e.g. New Lanark, Antonine Wall, Heart of Neolithic Orkney each have both; Flow Country, the newest UK World Heritage Site (inscribed July 2024), is included). No grading scheme (no Grade I/II*/II equivalent for Gardens and Designed Landscapes, unlike England's Registered Parks and Gardens), so each layer is a single flat colour rather than graded like its English counterpart. Colours echo the same designation-type families used for the English National Heritage List layers (he-nhle) so the same kind of site reads the same regardless of nation. Rebuilt with layers/hes_designations.py.

Licence and attribution read from each layer's own DataMapWales metadata (GeoNode API): "There are no use restrictions on this data", re-use under the OGL provided the source and Cadw's copyright are acknowledged, with the attribution statement "Designated Historic Asset GIS Data, The Welsh Historic Environment Service (Cadw), DATE [the date that you received the data from Cadw], licensed under the Open Government Licence <URL>"; DATE is filled in as the download date, 2026-09-25. Cadw also asks that the data is used accurately and not in a misleading context, which the layer caveats say. Counts: 4,239 scheduled monuments, 385 registered parks and gardens (Grade I, II*, II), 58 registered historic landscapes (Outstanding or Special). Boundaries lightly simplified for web display. NOT built: Wales' World Heritage Sites (the layer's own metadata states no licence, only "see layer group metadata"), and Cadw's Listed Buildings and Protected Wrecks (same family and licence statement, left for a later pass). Built by pipeline/layers/cadw_designations.py (docs/DECISIONS.md #67).

Licence and attribution read from the data.gov.uk records (OpenDataNI is the publisher) and the DfC Historic Environment Digital Datasets page ("Open Government Licence"); the wording is the records' own statement. DfC states "Scheduled Zones are provided as an indicative dataset only", repeated in the layer caveats. Counts: 2,093 scheduled zones and 252 historic parks, gardens and demesnes (grades A* 23, A 133, B 95, one not recorded). The HED ArcGIS web service (services2.arcgis.com) is deliberately not used: its description limits it to HED's public viewer and its item states no licence. NOT built: Historic Buildings Record, Sites and Monuments Record, Industrial and Defence Heritage, Battlesites (not among DfC's published downloads, so its licence is unverified). Built by pipeline/layers/hed_ni_designations.py (docs/DECISIONS.md #68).

Licence ("ogl3") and attribution read from the dataset's own record at planning.data.gov.uk/dataset/conservation-area.json. The record's attribution ends "obtained on [date]"; the date is filled with the download date, and its markdown link is written as plain text. The record calls the dataset "a work in progress, and ... currently incomplete" that "contains a number of duplicate conservation areas we are working to reconcile". Each area has a quality flag: 5,393 "authoritative" (from the local planning authority) and 4,266 "some" (gathered from Historic England and data.gov.uk), coloured differently on the map. 119 exact duplicates (same name and position) were removed and 1 record without a polygon skipped, leaving 9,539 areas (5,323 authoritative, 4,216 other). All centroids fall inside England. Built by pipeline/layers/conservation_areas.py (docs/DECISIONS.md #74). 8 MB of tiles, gitignored.

Licence and credit read from the download's own bundled Conservation_Areas.txt (OGL v3.0; the credit ends 'Crown copyright and database right [year]', filled as 2026 to match the other Historic Environment Scotland layers; and 'This data is indicative only', repeated in the layer caveats). 670 conservation areas (661 polygons, 9 multipolygons), boundaries from the local authorities, names and authorities tidied from capitals. The designation year is 0 where not recorded and is then left out. Built by pipeline/layers/conservation_areas.py (docs/DECISIONS.md #74).

Licence ("ogl3") and attribution read from the dataset record; its "obtained on [date]" is filled with the download date. 5,490 entries, all flagged authoritative, all with a polygon. The Historic England Terms page still returned 403 to our fetch tool (see he-scheduled-monuments); flagged for a manual read before any monetisation.

Attribution wording: the product's own Doc/licence.txt lists the three OS acknowledgements ('Contains OS data (c) Crown Copyright and database rights 2026', 'Contains Royal Mail data ...', 'Contains National Statistics data ...') and points to the OS OpenData licence, whose URL redirects to OGL v3.0 (checked 2026-09-20); the fourth sentence credits the weather figures. Two layers: 141 hand-picked real place names (55 originally, plus 21 more added 2026-09-22 and 65 on 2026-09-26, each matched by exact name + county + place type -- the script fails if a match is missing or ambiguous), and ten 2024 weather extremes found by scanning the raw HadUK-Grid files. Weather values come from the nearest 1 km land square within 5 km. The Olympic-pool figure is rainfall x a 105 m x 68 m pitch / 2,500,000 litres. See pipeline/layers/curiosities.py and docs/DECISIONS.md #31 and #65.

Wikidata's licensing page (Wikidata:Licensing, read 2026-10-16) says its structured data "is released into the public domain under Creative Commons Zero" (CC0 1.0); an earlier read of the same page, on 2026-10-09, gave the wording "All data in Wikidata has a CC0 license". The list is hand-picked and light-hearted, not a dataset of funny names: each place was searched on Wikidata by name and country, and the item that is that place (a village, town, city, island or municipality) was chosen by hand; its position is that item's coordinate statement. The note on each place (what the name means in its own language, or one fact about the place) is written for this site in English, French, Spanish and German and is not taken from Wikidata. Wikidata can record a place as a village or as its municipality, so a point may sit on a village centre rather than on the whole municipality. The same places are in the homepage's "Surprise me" list (web/public/curiosities/surprise.json).

Attribution wording: the three acknowledgements are the ones in the product's own Doc/licence.txt (see the os-open-names entry). 29,463 towns, villages and hamlets (local types Town, City, Village, Hamlet; suburban areas excluded) are grouped by rule of thumb on the shape of their name: Old Norse (-by, -thorpe, -thwaite, -toft, -beck, -garth, -holme), Latin/Roman (-chester, -caster and variants), Brittonic (Llan-, Aber-, Caer- anywhere; Tre-, Pol-, Pen-, Pont-... only in Wales, Cornwall, Devon), Gaelic (Inver-, Kil-, Bal-, Auch-, Kin-, Dun-, Glen-... only in Scotland), saints (St ...) and Old English (-ton, -ham, -ley, -ford, -field, -ing, -worth...). 13,700 match a rule and are shown. The rules are a heuristic, not a scholarly etymology of each name; the build checks 13 well-known names (Whitby, Scunthorpe, Grimsby, Manchester, Winchester, Lancaster, Aberystwyth, Llandudno, Invergordon, Kilmarnock, Reading, Birmingham, St Ives) land in the expected group. See pipeline/layers/place_name_origins.py.

The two required boundary statements come from https://www.ons.gov.uk/methodology/geography/licences ("Source: Office for National Statistics licensed under the Open Government Licence v.3.0" and "Contains OS data © Crown copyright and database right [year]"); the [year] is not stated by the boundary item, so it is taken as the year the ArcGIS item was created (2025): FLAGGED for the owner to confirm before launch. The provider line names the data owner exactly as the ONS indicator page does. Area codes were matched to the boundary vintage they use (fails if over 2% are unmatched). Councils with no published value are shown grey. Shared tile file ons-councils-dec2024.pmtiles. Refreshed 2026-10-29 to the 2021-25 period (ONS published it on 06 October 2026, replacing 2020-24): the newer period names the 2023 unitary authorities, so the layer moved from the May 2022 boundary vintage to December 2024, and the OS credit year with it. ONS states on the indicator page that Cumberland, North Yorkshire, Somerset and Westmorland and Furness are based on 2025 data only and that North and West Northamptonshire are based on 2022 to 2025; the layer says so.

The two required boundary statements come from https://www.ons.gov.uk/methodology/geography/licences ("Source: Office for National Statistics licensed under the Open Government Licence v.3.0" and "Contains OS data © Crown copyright and database right [year]"); the [year] is not stated by the boundary item, so it is taken as the year the ArcGIS item was created (2022): FLAGGED for the owner to confirm before launch. The provider line names the data owner exactly as the ONS indicator page does. Area codes were matched to the boundary vintage they use (fails if over 2% are unmatched). Councils with no published value are shown grey. Shared tile file ons-councils-may2022.pmtiles.

The two required boundary statements come from https://www.ons.gov.uk/methodology/geography/licences ("Source: Office for National Statistics licensed under the Open Government Licence v.3.0" and "Contains OS data © Crown copyright and database right [year]"); the [year] is not stated by the boundary item, so it is taken as the year the ArcGIS item was created (2025): FLAGGED for the owner to confirm before launch. The provider line names the data owner exactly as the ONS indicator page does. Area codes were matched to the boundary vintage they use (fails if over 2% are unmatched). Councils with no published value are shown grey. Shared tile file ons-councils-dec2024.pmtiles.

The two required boundary statements come from https://www.ons.gov.uk/methodology/geography/licences ("Source: Office for National Statistics licensed under the Open Government Licence v.3.0" and "Contains OS data © Crown copyright and database right [year]"); the [year] is not stated by the boundary item, so it is taken as the year the ArcGIS item was created (2026): FLAGGED for the owner to confirm before launch. The provider line names the data owner exactly as the ONS indicator page does. Area codes were matched to the boundary vintage they use (fails if over 2% are unmatched). Councils with no published value are shown grey. Shared tile file ons-councils-may2026.pmtiles.

The two required boundary statements come from https://www.ons.gov.uk/methodology/geography/licences ("Source: Office for National Statistics licensed under the Open Government Licence v.3.0" and "Contains OS data © Crown copyright and database right [year]"); the [year] is not stated by the boundary item, so it is taken as the year the ArcGIS item was created (2025): FLAGGED for the owner to confirm before launch. The provider line names the data owner exactly as the ONS indicator page does. Area codes were matched to the boundary vintage they use (fails if over 2% are unmatched). Councils with no published value are shown grey. Shared tile file ons-councils-dec2024.pmtiles. Updated 2026-10-06: ONS published March 2026 on 24 September 2026 (found by the live audit), so the series now runs 2010 to 2026. The page says 2026 rates use the 2025 population estimates, the latest available, so the newest year is less firm.

The two required boundary statements come from https://www.ons.gov.uk/methodology/geography/licences ("Source: Office for National Statistics licensed under the Open Government Licence v.3.0" and "Contains OS data © Crown copyright and database right [year]"); the [year] is not stated by the boundary item, so it is taken as the year the ArcGIS item was created (2022): FLAGGED for the owner to confirm before launch. The provider line names the data owner exactly as the ONS indicator page does. Area codes were matched to the boundary vintage they use (fails if over 2% are unmatched). Councils with no published value are shown grey. Shared tile file ons-councils-may2022.pmtiles.

The two required boundary statements come from https://www.ons.gov.uk/methodology/geography/licences ("Source: Office for National Statistics licensed under the Open Government Licence v.3.0" and "Contains OS data © Crown copyright and database right [year]"); the [year] is not stated by the boundary item, so it is taken as the year the ArcGIS item was created (2025): FLAGGED for the owner to confirm before launch. The provider line names the data owner exactly as the ONS indicator page does. Area codes were matched to the boundary vintage they use (fails if over 2% are unmatched). Councils with no published value are shown grey. Shared tile file ons-councils-dec2024.pmtiles.

The two required boundary statements come from https://www.ons.gov.uk/methodology/geography/licences ("Source: Office for National Statistics licensed under the Open Government Licence v.3.0" and "Contains OS data © Crown copyright and database right [year]"); the [year] is not stated by the boundary item, so it is taken as the year the ArcGIS item was created (2026): FLAGGED for the owner to confirm before launch. The provider line names the data owner exactly as the ONS indicator page does. Area codes were matched to the boundary vintage they use (fails if over 2% are unmatched). Councils with no published value are shown grey. Shared tile file ons-councils-may2026.pmtiles.

Licence read from the release page's own "Open Government Licence v3.0" footer; the release says the data are provided by charge point operators under the Public Charge Point Regulations 2023 guidance and collated by Zapmap for DfT (the same source the ONS per-100,000 layer names). Sheet EVCI0102a is the count of public chargers by local authority for 1 January, 1 April and 1 July 2026; 1 July is shown. Independent check at build: the 361 local authority counts sum to 121,171, exactly the United Kingdom figure in the same release's EVCI0101a (country and region) table and DfT's headline total. Counts range from 16 to 2,893 (per class: 53, 115, 97, 50, 28, 18 councils). The same figures per 100,000 residents are the separate ons-ev-chargers layer. Extracted by pipeline/layers/dft_ev_chargers.py into the CSV shape the council build reads, then tiled by layers/ons_councils.py (docs/DECISIONS.md #76). The council boundary credit year follows the boundary vintage matched (May 2026).

The two required boundary statements come from https://www.ons.gov.uk/methodology/geography/licences; the [year] is not stated by the boundary item, so it is taken as the year the ArcGIS item was created (2025): FLAGGED for the owner to confirm before launch. The indicator page says the station analysis was done with Ordnance Survey and excludes some NaPTAN stations (e.g. heritage lines). Scotland and Northern Ireland are not in the data and are shown grey. Shared tile file ons-councils-dec2024.pmtiles.

The two required boundary statements come from https://www.ons.gov.uk/methodology/geography/licences ("Source: Office for National Statistics licensed under the Open Government Licence v.3.0" and "Contains OS data © Crown copyright and database right [year]"); the [year] is not stated by the boundary item, so it is taken as the year the ArcGIS item was created (2026): FLAGGED for the owner to confirm before launch. The provider line names the data owner exactly as the ONS indicator page does. Area codes were matched to the boundary vintage they use (fails if over 2% are unmatched). Councils with no published value are shown grey. Shared tile file ons-councils-may2026.pmtiles.

The two required boundary statements come from https://www.ons.gov.uk/methodology/geography/licences ("Source: Office for National Statistics licensed under the Open Government Licence v.3.0" and "Contains OS data © Crown copyright and database right [year]"); the [year] is not stated by the boundary item, so it is taken as the year the ArcGIS item was created (2026): FLAGGED for the owner to confirm before launch. The provider line names the data owner exactly as the ONS indicator page does. Area codes were matched to the boundary vintage they use (fails if over 2% are unmatched). Councils with no published value are shown grey. Shared tile file ons-councils-may2026.pmtiles.

The two required boundary statements come from https://www.ons.gov.uk/methodology/geography/licences ("Source: Office for National Statistics licensed under the Open Government Licence v.3.0" and "Contains OS data © Crown copyright and database right [year]"); the [year] is not stated by the boundary item, so it is taken as the year the ArcGIS item was created (2026): FLAGGED for the owner to confirm before launch. The provider line names the data owner exactly as the ONS indicator page does. Area codes were matched to the boundary vintage they use (fails if over 2% are unmatched). Councils with no published value are shown grey. Shared tile file ons-councils-may2026.pmtiles.

The two required boundary statements come from https://www.ons.gov.uk/methodology/geography/licences ("Source: Office for National Statistics licensed under the Open Government Licence v.3.0" and "Contains OS data © Crown copyright and database right [year]"); the [year] is not stated by the boundary item, so it is taken as the year the ArcGIS item was created (2025): FLAGGED for the owner to confirm before launch. The ONS indicator page labels this "official statistics in development" and gives the data source as the Office for National Statistics (published 05 December 2025); the layer is described in the page's own words as an estimated VALUE (£ millions, 2024 prices), not a physical measurement. Range in 2024 £0.1 million (Shetland Islands) to £47.7 million (Birmingham). Area codes were matched to the boundary vintage they use (fails if over 2% are unmatched). Councils with no published value are shown grey. Year slider starts at 2010 to keep the shared tile file ons-councils-dec2024.pmtiles small (earlier years exist in the ONS series). Shared tile file ons-councils-dec2024.pmtiles.

The two required boundary statements come from https://www.ons.gov.uk/methodology/geography/licences ("Source: Office for National Statistics licensed under the Open Government Licence v.3.0" and "Contains OS data © Crown copyright and database right [year]"); the [year] is not stated by the boundary item, so it is taken as the year the ArcGIS item was created (2025): FLAGGED for the owner to confirm before launch. The ONS indicator page labels this "official statistics in development" and gives the data source as the Office for National Statistics (published 05 December 2025); the layer is described in the page's own words as an estimated VALUE (£ millions, 2024 prices), not a physical measurement. Net values from -£258.9 million (Highland) to +£119.1 million (Argyll and Bute) in 2023; negative means net emitted (Fenland -£165.0 million in 2023: drained peat). Area codes were matched to the boundary vintage they use (fails if over 2% are unmatched). Councils with no published value are shown grey. Year slider starts at 2010 to keep the shared tile file ons-councils-dec2024.pmtiles small (earlier years exist in the ONS series). Shared tile file ons-councils-dec2024.pmtiles.

The two required boundary statements come from https://www.ons.gov.uk/methodology/geography/licences ("Source: Office for National Statistics licensed under the Open Government Licence v.3.0" and "Contains OS data © Crown copyright and database right [year]"); the [year] is not stated by the boundary item, so it is taken as the year the ArcGIS item was created (2025): FLAGGED for the owner to confirm before launch. The ONS indicator page labels this "official statistics in development" and gives the data source as the Office for National Statistics (published 05 December 2025); the layer is described in the page's own words as an estimated VALUE (£ millions, 2024 prices), not a physical measurement. Range in 2024 from about £0 (West Devon) to £18.3 million (Westminster); values can jump between years. Area codes were matched to the boundary vintage they use (fails if over 2% are unmatched). Councils with no published value are shown grey. Year slider starts at 2010 to keep the shared tile file ons-councils-dec2024.pmtiles small (earlier years exist in the ONS series). Shared tile file ons-councils-dec2024.pmtiles.

The two required boundary statements come from https://www.ons.gov.uk/methodology/geography/licences ("Source: Office for National Statistics licensed under the Open Government Licence v.3.0" and "Contains OS data © Crown copyright and database right [year]"); the [year] is not stated by the boundary item, so it is taken as the year the ArcGIS item was created (2026): FLAGGED for the owner to confirm before launch. ONS labels this an "official statistic in development", data source DESNZ, Low Income Low Energy Efficiency (LILEE) metric. England only (296 of 361 UK council areas). Only one year (2024) published so far, no time slider. Checked against the raw CSV: Hartlepool 8.7%, Kensington and Chelsea 10.7%, Blackpool 13.9%. Shared tile file ons-councils-may2026.pmtiles.

The two required boundary statements come from https://www.ons.gov.uk/methodology/geography/licences ("Source: Office for National Statistics licensed under the Open Government Licence v.3.0" and "Contains OS data © Crown copyright and database right [year]"); the [year] is not stated by the boundary item, so it is taken as the year the ArcGIS item was created (2026): FLAGGED for the owner to confirm before launch. Data source named on the ONS page: Ofcom. Coverage (can order the service), not take-up or actual speed. All 361 UK council areas have a value. Checked against the raw CSV: Hartlepool 96.9%, Kensington and Chelsea 93.3%, Na h-Eileanan Siar (Outer Hebrides, the lowest in the UK) 12.0%. Shared tile file ons-councils-may2026.pmtiles. Also listed on the Energy map (alsoIn).

The two required boundary statements come from https://www.ons.gov.uk/methodology/geography/licences ("Source: Office for National Statistics licensed under the Open Government Licence v.3.0" and "Contains OS data © Crown copyright and database right [year]"); the [year] is not stated by the boundary item, so it is taken as the year the ArcGIS item was created (2025, matching the median-age/population-density entries on the same boundary vintage): FLAGGED for the owner to confirm before launch. Jobs per resident aged 16-64, Great Britain only (Northern Ireland not covered by this indicator). A single year shown (2024); the ONS series runs 2011-2024 but a slider was not added given the extreme outlier (City of London ~57) would dominate every year's colour scale equally. Checked against the raw CSV: Manchester 1.21, Westminster 5.51, City of London 56.92 (the UK maximum). Shared tile file ons-councils-dec2024.pmtiles.

Median gross weekly pay for all employee jobs (full-time and part-time) by where people live, from the Annual Survey of Hours and Earnings (ONS and NISRA), April 2025; the ONS series runs 2008 to 2025 but only the latest year is shown. 359 council areas have a value (England, Scotland, Wales and Northern Ireland). Checked against the raw CSV: Hartlepool 533.9, minimum 515, median 628.8, maximum 955.2 pounds a week. The ONS page says the 2025 estimates are provisional for Great Britain, that users should consider confidence intervals, and that the Northern Ireland country figure comes from Nomis. The boundary credit's [year] is taken as the year the boundary ArcGIS item was created, as for the other council layers: FLAGGED for the owner. Shared tile file for its boundary vintage.

Share of children aged 0 to 15 in households below 60% of median UK income after housing costs (DWP). Definition, source, provisional status and the survey of about 20,000 households were read from the ONS indicator page. 363 areas have a value. Checked against the raw CSV: East Dunbartonshire 7.79 (minimum), Tower Hamlets 50.34 (maximum), Hartlepool 31.95, Manchester 42.27. Two council codes (E08000016 Barnsley and E08000019 Sheffield, recoded in 2024) have no polygon in the matched boundary vintage and are left grey, within the pipeline's 2% tolerance. The boundary credit's [year] is taken as the year the boundary ArcGIS item was created (2026 for the May 2026 boundaries this indicator matches), as for the other council layers: FLAGGED for the owner. Shared tile file for its boundary vintage.

Residence-based housing affordability ratio: median house price divided by median gross annual full-time earnings of residents (ASHE), England and Wales. The source says house prices are not mix-adjusted and that five years of earnings is used as a broad affordability threshold. 316 areas have a value. Checked against the raw CSV: Hyndburn 3.86 (minimum), Kensington and Chelsea 21.87 (maximum), Hartlepool 4.86, Fenland 6.80. The boundary credit's [year] is taken as the year the boundary ArcGIS item was created (2026 for the May 2026 boundaries this indicator matches), as for the other council layers: FLAGGED for the owner. Shared tile file for its boundary vintage.

Adults aged 18 and over with a body mass index of 30 or more, from Sport England's Active Lives Adult Survey (self-reported height and weight, adjusted), published by OHID, with 95% confidence intervals. 296 areas have a value. Checked against the raw CSV: Camden 10.40 (minimum), Sunderland 42.60 (maximum), Hartlepool 35.76 (32.32 to 39.23), Manchester 26.96. The boundary credit's [year] is taken as the year the boundary ArcGIS item was created, as for the other council layers: FLAGGED for the owner. Shared tile file for its boundary vintage.

The two required boundary statements come from https://www.ons.gov.uk/methodology/geography/licences ("Source: Office for National Statistics licensed under the Open Government Licence v.3.0" and "Contains OS data © Crown copyright and database right [year]"); the [year] is not stated by the boundary item, so it is taken as the year the ArcGIS item was created (2022, matching the smokers/happiness entries on the same boundary vintage): FLAGGED for the owner to confirm before launch. Supermarkets per 10,000 people, a single count as at August 2023 (the oldest source date of any layer on the map). 6 area codes (post-2023 unitary authorities in Cumbria and North Yorkshire) have no boundary on this vintage and are left off the map, matching the pipeline's own printed note; within its 2% mismatch tolerance. Checked against the raw CSV: Fenland 2.54, Cambridge 3.30, City of London 37.28 (the UK maximum). Shared tile file ons-councils-may2022.pmtiles.

The two required boundary statements come from https://www.ons.gov.uk/methodology/geography/licences ("Source: Office for National Statistics licensed under the Open Government Licence v.3.0" and "Contains OS data © Crown copyright and database right [year]"); the [year] is not stated by the boundary item, so it is taken as the year the ArcGIS item was created (2022, matching the supermarkets/smokers entries on the same boundary vintage): FLAGGED for the owner to confirm before launch. England only. Sports facilities per 10,000 people, a single count as at 5 October 2023. Shares the same 6 area codes (post-2023 Cumbria and North Yorkshire unitary authorities) with no boundary on this vintage as the supermarkets layer, left off the map, within the pipeline's 2% mismatch tolerance. Checked against the raw CSV: Fenland 15.62, Manchester 10.29, City of London 111.85 (the UK maximum). Shared tile file ons-councils-may2022.pmtiles.

The two required boundary statements come from https://www.ons.gov.uk/methodology/geography/licences ("Source: Office for National Statistics licensed under the Open Government Licence v.3.0" and "Contains OS data © Crown copyright and database right [year]"); the [year] is not stated by the boundary item, so it is taken as the year the ArcGIS item was created (2026): FLAGGED for the owner to confirm before launch. Survey estimate with a 95% confidence range (has_ci), data source DESNZ successor department DCMS. England only (295 of 361 UK council areas). Checked against the raw CSV: Fenland 27% (22-32%), Cambridge 41% (36-46%), City of London the UK minimum at 14%, Richmond upon Thames the UK maximum at 48%. Shared tile file ons-councils-may2026.pmtiles.

Total installed renewable electricity capacity (every technology summed) by local authority, UK-wide (England, Scotland, Wales and Northern Ireland) -- wider coverage than DUKES power stations (major stations only) or REPD (individual projects, not summed per council). Only the 2025 sheet is used: the workbook's own 2014-2025 time series uses different local authority boundaries in earlier years (410 councils in 2014 vs 365 in 2025), the same "older years use older council codes" problem already found for household electricity/gas (#41), so a full slider would show misleading grey gaps rather than genuine boundary reorganisation. Written as a CSV in the shape layers/ons_councils.py reads (the pattern already used for the DfT EV charger count layer, #76) so the shared council pipeline matches and tiles it.

Boundaries are described by NESO as approximate and possibly slightly outdated. Used as the polygon geometry for the carbon-intensity layer: the API's 14 regions follow DNO areas, matched by area name to the API region ids (mapping in pipeline/layers/neso_dno_regions.py). Reprojected to EPSG:4326 and simplified to 100 m; 0.22MB tiles.

Attribution statements copied from ONS's licences page on 2026-09-19 (year placeholders filled with 2026, the release year). That page does not give Gridlink wording even though the ONSPD README lists Gridlink IP, flagged for the owner to check before any monetisation. Processing: 2,729,090 rows -> kept 1,748,230 live GB postcodes; dropped NI (63,293), terminated (905,581) and rows with no grid reference (11,986). Output is one JSON shard per outcode (2,863 files, ~58MB uncompressed; gitignored, rebuilt by pipeline/layers/onspd_postcodes.py) holding lat/long (5 dp) and indices into name tables. Local authority names come from ONS's LAD names file (April 2025 vintage; codes in the 2026 data all matched except catch-all codes), region names from ONS's region file, with the country as a fallback outside England.

The readme beside the downloads (read 2026-10-11) says "This work is licensed under a Creative Commons Attribution 4.0 License" and that the data is provided "as is". It names no credit wording, so the credit above is ours: the creator, a link and the licence, as CC BY asks. Used for the "search a place" box on the map, not as a layer: 242 countries and 166,658 places (City, Town or Village by size and administrative seat) written to web/public/places/places-world.json (6.1 MB, about 2 MB compressed) by pipeline/layers/world_places.py (docs/DECISIONS.md #199), biggest first. Each place carries its region and country (GeoNames' ASCII region names). Great Britain is left out of this file because the search already has every named place there from OS Open Names; Northern Ireland, the Channel Islands and the Isle of Man are kept. Other spellings of the name are added for about 60 big places by hand (Munich and Muenchen for Munchen, Peking, Bombay...), not taken from GeoNames' own alternate-names column, which is mostly transliterations and airport codes. The data is as GeoNames has it: populations, spellings and positions can be out of date or wrong, and a place smaller than 1,000 people is not in it (type a latitude and longitude instead).

The three acknowledgements are the ones in the product's own Doc/licence.txt (see the os-open-names entry). Used for the "search a place" box on the map, not as a layer: 43,219 populated places (City 71, Town 1,352, Suburban Area 10,861, Village 15,150, Other Settlement 2,895, Hamlet 12,890) written to web/public/places/places.json (1.5 MB, about 0.5 MB compressed) as name, type, county or unitary authority (else district or region) and WGS84 position rounded to 4 decimals (about 10 m), by pipeline/layers/place_index.py (docs/DECISIONS.md #72). Great Britain only.

Wikidata states: "All data in Wikidata has a CC0 license" (Wikidata:Licensing, read 2026-10-09). Used for the small 3D pictures of famous places and for the search box (web/public/landmark-symbols-europe.geojson and web/public/places/places-europe.json), not as a layer. Which places: every item of the country that is an instance of (or of a subclass of) a castle, chateau, fortification, palace, cathedral, basilica, church, abbey, mountain, lake, island, waterfall, cave, cliff, lighthouse, bridge, viaduct, tower, monument, museum, megalith, archaeological site, garden, theme park, stadium, railway station, theatre or beach, and is described on at least 10 to 25 language editions of Wikipedia (the bar differs by kind). The count of language editions ("sitelinks") is the measure of how famous a place is: it measures fame, not quality or importance, and it favours places with a large international audience. Wikidata's classes are uneven (a town can be filed as a castle, a range as a mountain), so the build drops a short written list of such records (DROP_NAME in pipeline/tools/build_europe_landmark_symbols.py) and writes down by hand, with the reason, the famous places the classes miss (the Atomium, the Moulin Rouge, the Dune of Pilat...). Items with no "country" statement are missed. The Republic of Ireland only: Northern Ireland's places are in the UK set. Sites of atrocity (camps, massacre sites, slave forts) and places whose destruction or rebuilding would make a toy picture misleading are dropped by name. Raw results (data/raw/wikidata-landmarks/) are not committed. The pictures are AI-generated in Canva (the same terms question as docs/DECISIONS_LOG.md #153, #162, #179, #182).