# GB railway stations through time, v1 Every railway station Great Britain has had, living and closed: **9,663 dated stations** (3,011 open today), each with an opening year, a closing year where it has gone, coordinates, and a per-date record of where that date came from, plus the population-weighted distance-to-nearest-station percentiles per year that the chart draws. ## Scope Every railway station **except trams**: National Rail, London Underground, the metros (Tyne and Wear, DLR, Glasgow Subway), airport transit, monorails and funiculars, heritage and miniature lines, and the closed stations of all of them. The `system` column says which (national-rail 8,513, heritage 542, london-underground 276, metro 127, industrial 80). **If you want heavy rail only, filter `system == 'national-rail'`.** Great Britain means the island and its offshore islands, **not** the Crown Dependencies: Northern Ireland and the Isle of Man are excluded. Open National Rail stations come to 2,618 rows and **2,612 distinct CRS codes**. Checked against DfT NaPTAN directly (`check_naptan_coverage.py`, joining NaPTAN's ATCO code to CRS through CORPUS, so no name matching is involved), **coverage of the live network is complete**: of 2,635 active NaPTAN rail stations, 2,603 are open here, 18 more are second codes for a station we already have open (high and low level platforms), and 7 are stations NaPTAN still lists as active that closed years ago (Wedgwood, Barlaston, Teesside Airport, IBM). 7 remain (Bristol Brabazon Rail Station, Frosterley Rail Station, Porthmadog Harbour Ffestiniog Railway Station, Southease - Piddinghoe Road, Southsea Hoverport, Stanhope Rail Station, Wolsingham Rail Station), and they are not a coverage gap either: Frosterley, Stanhope, Wolsingham and Porthmadog Harbour are heritage railways, Southsea Hoverport is not a railway station, Southease holds a second code, and Bristol Brabazon is a 2026 opening. These counts are recomputed at every publish (`check_naptan_coverage.py`). ## Where the dates come from Each date carries its own basis in `opened_src` / `closed_src` / `reopened_src`, and its own source URL in `opened_source` / `closed_source` / `reopened_source`. A `+` in a basis means "year from the first source, precision from the second". | basis | openings | closures | |---|---|---| | `distribution-est` | 0 | 81 | | `manual` | 287 | 230 | | `osm` | 6 | 3 | | `quick` | 8,308 | 5,620 | | `segment-est` | 265 | 128 | | `unknown` | 0 | 173 | | `wikidata` | 248 | 72 | | `wikidata+prose` | 4 | 0 | | `wikidata+wikipedia` | 467 | 20 | | `wikipedia` | 54 | 626 | | `wikipedia+category` | 10 | 5 | | `wikipedia+prose` | 8 | 8 | | `wikipedia-category` | 3 | 40 | | `wikipedia-list` | 0 | 20 | | `wikipedia-prose` | 3 | 7 | Of 6,652 closed stations (rows not open today; a further 381 open stations carry a past closure and a reopening), **6,270 (94.3%) carry a station-level closure record**. 93% of openings and 93% of closures carry a month or a day, not just a year. One source outranks every automatic tier, and two bases are estimates that should never be read as facts about an individual station: - **`quick`**: Michael Quick, *Railway Passenger Stations in Great Britain: a Chronology*, version 5.05 (2023); each date links to the page of the PDF it comes from. "op" is the first day fare-paying passengers were carried on a timetabled train, "clo" the first day services did not run, the same convention as this table. The Chronology is a source: its dates are cited, page by page, and neither the PDF nor its text is redistributed. Quick excludes preserved lines, so heritage stations never take dates from it. - **`segment-est`**, the station inherits the opening or closing year of the route-history segment it sits on. Right for the great cull years, wrong where a station outlived or predeceased its line. The `*_source` field names TWO separate doubts, because they are different questions: * **association**, is the station really on that segment? Distance decides, and it is bimodal: p50 36 m, p75 58 m, then p90 222 m. Beyond 100 m the string says "association uncertain" (72 openings, 38 closures). Moor Road inherits from an UNNAMED segment 157 m away. * **date quality**, is the segment's OWN date sourced, or itself an estimate? An inherited date resting on an estimate is an estimate of an estimate (53 openings, 21 closures). Only 11 openings carry both. Neither is refused: for those rows it is the best answer available, and refusing would drop real places out of the table. They say so instead. - **`distribution-est`**, no closure record anywhere; a year drawn deterministically from the closure distribution of stations opened within ±15 years. Right in aggregate, wrong per station. Capped at 1975, so it cannot invent a modern closure. **`unknown`** means the station is not open today, no source records a closure, and imputing one would have been implausible. The year is deliberately null rather than fabricated. ## How much of this is checked Dates are held by up to six sources: Quick's Chronology, the Wikipedia infobox event ladder, the article prose, the `opened in YYYY` / `closed in YYYY` categories, Wikipedia's closed-station lists, and Wikidata. They are separately held, not independent: Wikidata imports from Wikipedia, the four Wikipedia readings come from the same editors, and Wikipedia articles cite Quick and Butt for their dates. So corroboration below measures agreement between records, not independent confirmation; most dates are corroborated in that sense. Recomputed at every publish over every row with a Wikidata id; a source "holds" a year when it gives one for the row, and a date is corroborated when two or more sources give the year the table carries: | | openings | closures | |---|---|---| | corroborated: two or more sources give the year | **85.6%** | **85.9%** | | one source only holds a year | 2.4% | 4.8% | | sources held years but fewer than two give ours | 7.7% | 4.5% | | modelled, nothing to check against | 2.7% | 3.0% | | no source holds a year (hand entry, OSM, no Wikidata id) | 1.6% | 1.7% | Where two sources both hold a year, they agree with each other **88.2%** of the time on openings and **88.1%** on closures. Most of the closure gap is convention, not error: Wikipedia's `closed in` category tends to the FINAL closure while this file records closure to passengers. Separately, on 2026-09-01 a random sample of 25 rows was put beside each station's article and adjudicated by hand by the same researcher who built the table (so a check, not an independent audit): 23 clean, one arguable (a station opened under one name and renamed later, so the year depends which name you are dating) and one wrong, an article covering two stations whose bare infobox fields straddled the pair. Both are fixed and cited. ## Conventions - **A closure is a closure to PASSENGERS.** "Closed to passengers 1958, goods until 1964" is 1958. Wikidata's own P3999 often holds the goods year and is corrected against the article where it does. - **`public_passenger` is about the row's dated life, not the line's status today.** Pickering served the public network 1845-1965 and is `public_passenger=1` for those years even though its line is preserved now; its closure year ends its public life without help. A station is excluded (438 rows) only when it had no public life to record, a preservation-era station opened after its line's public closure, or an adjudicated goods depot, mail platform, private halt, tourist operation or never-opened site. `pp_basis` says which. - **`duplicate_of` marks a station present twice** (18 rows), two OSM objects for one place. **Filter on it whenever you count stations.** ## The percentiles file `station_distance_percentiles.csv`: for each year, the distance from where people live to the nearest station then open, population- weighted percentiles (p50/p75/p90/p95/p99), in miles and km. It counts **public passenger stations only**, which is what the external validation row counts. It also skips a station that is not open today and carries no closure year: its span is unknown, and treating a null closure as "never closed" would keep it in the modern count. Two deliberate conventions, stated plainly: - **Today's population, every year.** People are weighted by the 2021/2022 census output-area centroids (ONS + NRS, OGL v3) in 1850 as much as in 2026. The question answered is "how far would TODAY'S population have been from a station in year Y", not "how far were the people of year Y". **That bias has now been measured, and it is not a uniform offset.** A parallel workstream has built a year-native population surface and run both weightings over this same station table. Median distance in miles, the people of the year against today's people: | year | contemporary | today's (this file) | |---|---|---| | 1851 | 1.72 | 1.52 | | 1881 | 0.81 | 0.83 | | 1891 | 0.71 | 0.76 | | 1911 | 0.55 | 0.69 | | 1921 | 0.56 | 0.69 | | 1961 | 0.76 | 0.81 | These rows are read at publish time from the population workstream's `distance_percentiles.csv`, which is the series the site charts; the contemporary column there is Great Britain in every census year, and its error bands are in that file. They cross over around 1881. Before then the people of the time were FURTHER from their own network than today's population would be, because the early network served towns and modern settlement has since grown around it; after it, the reverse. **So this file flatters the Victorian years and penalises the interwar peak.** This file remains the present-day check: its final year is the external validation row below, and it is the only series here that can be compared with a NaPTAN-based measurement. - **Straight-line distance**, not travel distance. Latest year (2026): p50 1.0 mi, p90 4.03 mi. The external validation row is computed at every publish by `naptan_distance_check.py` from DfT NaPTAN's active rail stops and the same population centroids, no station of ours involved (`naptan_distance_check.json`, inputs pinned by hash, run 2026-09-13): median **1.05 mi**, p90 **4.05 mi** over 2,639 stops. NaPTAN's scope is not this table's: it carries heritage-line stations and not London Underground or the metros. Dropping the 10 stops this table marks heritage or not public passenger (the like-series variant) gives 1.06 / 4.08; the Underground and metro difference is not recoverable from NaPTAN. ## gb_stations.csv schema - `station_id`, Wikidata QID where known, else the OSM object id - `name`, `lat`, `lon` (EPSG:4326), `easting`, `northing` (EPSG:27700) - `position_src`, `osm` or `wikidata`, whichever placed the station; `wikipedia` for a station with no Wikidata coordinate, placed from its own article's `{coord}` (or a line article's); `village-centre` for the few stations that neither holds, named in a closed-station list and placed at the centre of their village (Wikidata's coordinate for the settlement). Those rows are the village's position, not the station's: fine for a distance-to- nearest-station surface at parish scale, wrong by up to a mile for anything finer. `between-stations` for a station Quick's Chronology places between two named neighbours we hold and nothing else locates: the midpoint of the two, so wrong by up to half their spacing. Worth having: both coordinate errors found by auditing this table came from Wikidata, copied out of the article's own `{coord}`. - `opened`, `closed`, `reopened`, years; `closed` empty = still open. **`closed` is closure to PASSENGERS**, which is what a question about public transport access needs. Most closed British stations shut in stages, so it is often not the day the site fell silent: Quainton Road lost its passengers on 4 March 1963 and its goods service on 4 July 1966. - `closed_final`, `closed_final_date`, `closed_final_kind`, `closed_final_src`, `closed_final_source`, the LAST closure the article records, when it is later than the passenger one. Populated for 1,710 of 6,652 closed stations; empty means no source records a later event, not that none happened. `closed_final_kind` says what ended, `goods` (797), `complete` (446), `workmen` (18, miners' and school trains outliving the service) or `other` (449, a closure whose traffic the text does not name). A temporary or wartime closure is never used. **Use `closed` for access questions and `closed_final` for "when did the railway stop coming here".** - `opened_date`, `closed_date`, `reopened_date`, the same events as precise as the SOURCE is: `1862-06-02`, `1928-07` or `1862`, never padded - `open_today`, 1/0, in service on any system in scope - `heritage`, 1 if the station sits on a heritage line **today**; see `public_passenger` for whether it had a public life - `public_passenger`, `pp_basis`, see Conventions - `duplicate_of`, the row this one repeats, empty if unique - `system`, `access`, `function` and their `*_basis` columns , `system` is evidenced and usable; `access` and `function` are mostly `default`, meaning "nothing said otherwise", and the basis column says so on every row. Do not read their totals as findings. - `lines_within_100m`, `nearest_line`, `nearest_line_m` - `kind`, station | halt - `opened_src` / `closed_src` / `reopened_src`, the basis table above - `opened_source` / `closed_source` / `reopened_source`, the URL - `crs`, `osm_id`, `wikidata`, joins to other datasets ## Known limitations - **Stations are missing, and not slightly.** Checked against Wikipedia's own closed-station lists (5,326 names): after matching every name, old name and partial name we hold, **220 stations those lists name are absent from this table**, and the lists cover only some counties, so the true figure is higher. There is no independent national count of every station that ever existed to set a ceiling on it, so this figure is a floor with no upper bound. It reconciles with no independent national count: closures of public stations in the Beeching years 1963-70 come to 1,783 here, against 2,363 stations listed for closure in the 1963 report (Beeching, *The Reshaping of British Railways*, 1963, p. 97), which is a proposal, not an outcome, so the two are not comparable. These were stations people could walk to, so their absence biases historical distances LONG, most in the 1850s-80s and 1930s where the missing cluster. They are missing because neither OSM nor Wikidata holds a position for them; most have no Wikipedia article either. The names are in `missing_from_lists.json` in the build directory. A further 95 of the listed names are in the table at village-centre precision and 23 at a coordinate taken from a line article (see `position_src`); the rest have no settlement to stand in for them, or a name too ambiguous to place. - Reopening dates are honoured only where a dated later opening exists; closure-then-silence on an open station is treated as a records artefact, matching the route-history convention. - 65 dates are contested between an article's infobox and its own category, and are not guessed, the table keeps the majority reading (`contested_dates.tsv`). - Four Tanfield Railway stations carry the 1725 wagonway date, which is wrong; no source gives their real opening. They are `public_passenger=0`. - `segment-est` inherits from the route-history dataset, where a RENAMED route carries the renaming year. Guarded against here, but it is why a handful of inherited openings look modern. Licence: ODbL (positions from OSM); dates CC0/CC BY-SA per source. Attribute "Mind the Gates railway atlas / © OpenStreetMap contributors".