Changelog

What changed in the atlas, and when: errors found and fixed, routes added and the map changed, each dated and credited. Most fixes ship within a day of being reported.

An atlas like this is never finished and never flawless. This page keeps the fixes honest: what was wrong, when it was corrected, and who caught it. Spotted something? Email ceoflightmussy@gmail.com, open an issue on GitHub or mention it wherever the atlas is being discussed; it will be fixed and end up here.

8 November 2026

Seven routes join the map where a reader found towns with daily trains that no route served: the Cotswold Line from Hereford through Worcester and Oxford to London Paddington; Greater Anglia’s West Anglia Main Line from Ely through Cambridge to London Liverpool Street; Great Northern from London King’s Cross through Cambridge and Ely to King’s Lynn; Northern’s Calder Valley line from Blackpool through Preston, Hebden Bridge and Bradford to York; in Victoria, V/Line’s Warrnambool line through Geelong and its Bendigo line through Castlemaine; and in Queensland, the Gold Coast line from Varsity Lakes to Brisbane. Suggested by kangerluswag on Reddit.
SNCB’s Intercity from Ostend to Eupen is now one route, through Bruges, Ghent, the three Brussels stations, Leuven, Liège and Verviers, on to Eupen. Suggested by MofiPrano on Reddit.
Three routes in western Jutland join the map: GoCollective’s trains from Esbjerg through Ribe to Tønder and from Nørre Nebel through Varde to Esbjerg on the Nebelbanen, and Midtjyske Jernbaner’s from Skjern past Ringkøbing to Holstebro. Suggested by SapphicCelestialy on Reddit.
Two routes in North Rhine-Westphalia join the map: RE 57 from Dortmund up the Ruhr through Arnsberg and Meschede to Warburg, and RB 25, the Oberbergische Bahn, from Cologne through Gummersbach to Lüdenscheid. Suggested by InvestigatorBusy5856 on Reddit.
Nine more routes join the map. Amtrak’s Pere Marquette runs from Grand Rapids by Holland and St. Joseph to Chicago, and VIA’s Sarnia train by London, Stratford, Kitchener and Guelph to Toronto. In Poland, ŁKA’s trains run from Częstochowa over the line to Chorzew Siemkowice, open to passengers again since December 2024, to Łódź, Polregio’s from Siedlce by Czeremcha and Hajnówka to Białystok, and the Wrocław–Międzylesie railway from the Czech border at Lichkov by Kłodzko to Wrocław. The S41 runs from Verviers by Liège and Tongeren to Hasselt, and Arriva’s Vechtdal line from Zwolle to Emmen. In Japan, the limited express Fujikawa runs from Shizuoka up the Fuji River to Kofu, and the Shiosai from Tokyo by Chiba to Chōshi. Found by the atlas’s own audit sweep.
Two lines that had been left off the map because buses ran in place of their trains are on it now: a temporary bus does not make a railway any less a passenger railway. The TER from Épinal by Saint-Dié and over the Col de Saales to Strasbourg had been refused on the old news that buses served Arches to Saint-Dié; trains came back there in December 2021. Banova Jaruga to Daruvar in Croatia is being rebuilt, and its card says that buses replace the trains until 12 December 2026. Changed at the maker’s request.
The Kalambaka branch’s card now says that storm Daniel damaged the line in September 2023 and that buses replace its trains from Palaiofarsalos until it is repaired. Changed at the maker’s request.
850 routes now have a timetable that confirmed their own train, up from 830. Four in Australia are confirmed for the first time, each against its operator’s own published timetable: the Gold Coast line against TransLink’s, and V/Line’s Bairnsdale, Warrnambool and Bendigo lines against Victoria’s, each train running the drawn line within a few metres. Pere Marquette and the Sarnia train are confirmed against Amtrak’s and VIA’s own timetables. Found by the atlas’s own audit sweep.
All thirteen routes added earlier today list their stations. Four lists come from their own train’s OpenStreetMap relation and nine from the operator’s timetable, because those relations leave out stations the trains call at (Cambridge North and Cambridge South, Blackburn, Hebden Bridge and Leeds, Jerne, Ribe and Dyreby, Köln Hauptbahnhof, Pimpama, and on the Warrnambool line ten stations between Waurn Ponds and Footscray) or, for Bendigo, map its trains only in pieces. The same check found the Bairnsdale line missing Traralgon, the Sarnia train missing Brampton, Guelph and Kitchener, the Lichkov train missing Krosnowice Kłodzkie and Bardo Śląskie, and the Shiosai missing Chiba; all four are complete now. Found by the atlas’s own audit sweep.
66 station names on 24 routes no longer end in a “Station” that was the word and not the name, among them Goulburn, Junee and Moss Vale in New South Wales, Knokke and Oostende on the Belgian coast, the Wellington Cable Car’s Kelburn and Lambton Quay, and Amtrak’s Southern Pines. Union Station, Penn Station and a funicular’s Upper and Lower Station keep theirs. Found by the atlas’s own audit sweep.
On a first visit the map prepares two fewer drawing programs on the graphics card: the station names now draw with the same program as their discs, and the coastlines, borders, rivers and the grid with the same program as the route lines. On a laptop’s built-in graphics that is about a fifth of a second less of a frozen page while the map first appears, and nothing on the map changes colour or width. Changed at the maker’s request.

7 November 2026

The Filters button stands on the panel’s name line, a small blue button right of the night and day switch, and the filters open beside the panel, joined to it and as high as it: route types, trip length, how often and the two map layers side by side, every switch a button that stays lit while what it names is on the map. A laptop gets the same strip with the counts left to the headings; a narrower window shows the filters as one column beside the panel, or under it. The line that says which route types are switched off stands under the doors. Changed at the maker’s request.
On a phone the filters’ circle is light blue, the filters take at most half the screen with “Scroll down for more” at their foot, and Station names and the OpenStreetMap railways are buttons like the rest. Changed at the maker’s request.

1 November 2026

From any station to any station, on the map: the panel’s From and To fields take any two of the 14,457 stations the map’s lines call at, and the map draws the whole way, each line in its own colour, with every change and the distance, in a card that opens each line. It is marked as an alpha version while it is tried out, and the city to city page now leads to it. Changed at the maker’s request.
The route card is redrawn: the photograph, then the route’s type, the country’s flag and the operator with its logo, the name, the two ends with the journey time on a line between them, the distance and the two speeds, and the description in larger type. On a phone the card takes less of the screen and says when there is more to scroll. Changed at the maker’s request.
Every page is laid out as Movie Scene Map’s: a wide page of boxes with breadcrumbs, the photograph beside the facts, long lists that open a few at a time, and tables with a photograph on each row. The label over a page’s title is gone from every page but a route’s, whose type stays. Changed at the maker’s request.
Operator logos, from Wikimedia Commons with their licences, stand on the operators page, on route pages and on the cards: 272 of the 594 companies the map names have one there. Changed at the maker’s request.
Train routes by country holds every country in a box of its own on one page, its best known routes first. Changed at the maker’s request.
All 670 city guides now carry a photograph, on the city to city page and on each guide’s own page: the 416 guides that had none show a line of their journey. Changed at the maker’s request.
Flags are pictures everywhere on the pages, so a flag no longer draws as two letters on Windows. Changed at the maker’s request.

27 October 2026

Three of Amtrak’s trains that divide on the way now show the half that was missing: the Empire Builder’s Portland section, which leaves the Seattle cars at Spokane and follows the Columbia River from Pasco to Portland; the Lake Shore Limited’s Boston section, which parts from the New York train at Albany–Rensselaer and crosses the Berkshires through Pittsfield, Springfield and Worcester; and the Northeast Regional’s trains from Boston to Roanoke, which leave the Crescent’s line at Lynchburg. About 950 km of their lines were under no other route. Each card shares its train’s line as far as the train divides and runs on OpenStreetMap track from there, past every station its train calls at, and lies on the route shape Amtrak publishes with its timetable from end to end. Suggested by a reader.
830 routes now have a timetable that confirmed their own train, up from 827. Found by the atlas’s own audit sweep.
The Empire Builder’s Portland section lists its 40 stations and the Lake Shore Limited’s Boston section its 22, each from its own train’s OpenStreetMap relation. Found by the atlas’s own audit sweep.
In the United States a route no open source can list now takes its stations from Amtrak’s own timetable first, and only then from ÖBB Scotty: the Northeast Regional to Roanoke lists the 27 stations of train 171, the Floridian 45, the Maple Leaf from New York to Toronto 18, and the Auto Train its two, each from the train Amtrak’s timetable check matched, as their pages say. Changed at the maker’s request.
Choosing a route on the map no longer moves the map: a click on a line, or on a row of the list of lines under the pointer, opens the route’s card and leaves the view where it was, and on a phone the map no longer frames a chosen route again when the search box and buttons unfold. A route chosen in the search box, with Surprise me or from a link to the map still takes the map to it. Changed at the maker’s request.
On a computer the route card now stands still beside the point where the pointer met the line, as on Movie Scene Map, instead of following the pointer along it, and a card opened with a click stays there instead of moving to the right edge of the map. While the map is dragged the card steps aside, and it comes back beside the line under the pointer once the map stops. Changed at the maker’s request.
The links to the sister maps, at the foot of every page, in the More maps cards and on the page-not-found page, now tell search engines not to count them as recommendations (rel="nofollow"): one person owns all six sites, and a search engine counted 95 of every 100 links to this one from their footers. The pages themselves are unchanged, so their dates in the sitemap do not move. robots.txt also keeps crawlers off the map’s region links and off the MCP endpoint, which answers only to MCP clients and had nothing to index. Changed at the maker’s request.

25 October 2026

On a phone the map’s top bar is now one panel, as on Movie Scene Map: the name with four round buttons beside it, for the filters, the language, the night or day chart, and a fold that hides the search box and the three buttons under the panel or brings them back. The fold is remembered in the browser, and a route the map has framed is framed again when the buttons unfold over it. Changed at the maker’s request.
Seventeen routes join the map: Palermo to Catania across the middle of Sicily; RE 7 from Erfurt over the Thuringian Forest to Würzburg; the Glasgow South Western Line from Carlisle through Dumfries and Kilmarnock to Glasgow; the Franz-Josefs-Bahn from Vienna to Gmünd and České Velenice; the West of England line from London Waterloo through Salisbury to Exeter; the Stranraer line from Stranraer through Girvan and Ayr to Kilmarnock; Padua to Belluno at the foot of the Dolomites; RE 3 from Saarbrücken down the Nahe valley to Mainz and Frankfurt; Pau to Bedous up the Aspe valley; Bayonne to Saint-Jean-Pied-de-Port up the Nive valley; RB 55 from Munich past the Schliersee to Bayrischzell; the Yorkshire Coast Line from Hull to Scarborough; the Hastings line from London Charing Cross through Tunbridge Wells to Hastings; Limoges to Ussel over the Plateau de Millevaches; RE 5 from Cuxhaven along the Lower Elbe to Hamburg; Svendborgbanen from Svendborg to Odense; and Givet to Charleville-Mézières up the Meuse through the Ardennes. About 1,900 km of their lines were under no other route. Each is drawn from its own OpenStreetMap relation, lies on OpenStreetMap track from end to end, was checked against the train the planner runs on it, and is dated by the year its whole journey became possible. Found by the atlas’s own audit sweep.
Twelve routes were redrawn where their line left the railway their train takes, and each now follows it: the railjet from Vienna to Venice runs from Udine through Pordenone and Treviso, where it calls, instead of through Portogruaro; the TGV from Paris to Barcelona takes the high-speed bypass of Nîmes and Montpellier and calls at Nîmes Pont du Gard and Montpellier Sud de France; the TGV from Paris to Lille stays on the high-speed line instead of running through Arras; the Frecciarossa from Milan to Bari runs to Bologna on the main line through Parma, Reggio Emilia and Modena, where it calls; the railjet from Budapest to Berlin leaves from Budapest-Nyugati; SJ’s trains from Stockholm to Mora and to Sundsvall run through Arlanda airport, where they call; the X2000 from Stockholm to Gothenburg takes the Grödinge line to Södertälje Syd; the night train from Kyiv to Warsaw runs from Pilawa through Otwock; the AVE from Madrid to Oviedo goes through the Pajares base tunnels instead of over the old line across the pass; Bucharest to Sofia runs through Gorna Oryahovitsa station rather than 4 km beside it; and the Avant from Málaga to Sevilla runs into Córdoba, where it calls and turns back, instead of turning back 7 km short of the station. Eleven of them were confirmed against the planner’s timetable twice, the second time for 17 November. Found by the atlas’s own audit sweep.
827 routes now have a timetable that confirmed their own train, up from 799. Found by the atlas’s own audit sweep.
Twenty-one more routes list their stations: the seventeen new ones, sixteen from their own OpenStreetMap relations and RE 5 from another train’s, and the railjet from Vienna to Venice, the Frecciarossa from Milan to Bari, the night train from Kyiv to Warsaw and the X2000 from Stockholm to Sundsvall from the operator’s timetable, as their pages say. The TGVs from Paris to Barcelona and to Lille and the railjet from Budapest to Berlin now list the stations of the train their line follows, from the timetable; the TGV to Lille runs without a stop and no longer lists Arras. SJ’s train from Stockholm to Mora adds Arlanda, and the Avant from Málaga to Sevilla adds Córdoba. CrossCountry from Aberdeen to Plymouth now lists the 35 stations its 08:20 calls at, from the timetable: its OpenStreetMap relation named Dunbar, Morpeth, Tamworth and Worcestershire Parkway, where the train does not stop, and left out Stonehaven, Montrose and Arbroath, where it does. Found by the atlas’s own audit sweep.

24 October 2026

The Floridian, Amtrak’s train from Chicago to Miami through Washington, is now drawn on its own railway from end to end. A quarter of its line had been sketched rather than laid on track, and around Washington it strayed up to 13 km from the railway. It runs on OpenStreetMap track now, past all 46 stations it calls at, Tampa included, and lies within 122 m of the route shape Amtrak publishes with its timetable everywhere, so its page now says it was checked against Amtrak’s timetable. Found by the atlas’s own audit sweep.
The Texas Eagle from Chicago to San Antonio now crosses the Mississippi at St. Louis on the Merchants Bridge, as Amtrak’s trains do, instead of the MacArthur Bridge, and runs into San Antonio on the line through Kirby that Amtrak’s route shape follows, instead of the other of Union Pacific’s two lines from New Braunfels. Amtrak’s route shape now confirms the whole line; before, only its stations had been checked against the timetable. Found by the atlas’s own audit sweep.
The Silver Star card is retired, and its page now leads to the Floridian’s. The Silver Star last ran on 9 November 2024; since then the Floridian has run its route from Washington to Miami, Tampa included, and Amtrak’s timetable, which now runs to October 2027, has no Silver Star. Amtrak calls the merger temporary; if the Silver Star returns, so will its card. Found by the atlas’s own audit sweep.
Five more of Amtrak’s lines were redrawn on OpenStreetMap track and checked against the route shapes Amtrak publishes with its timetable: the Coast Starlight, sketched from Eugene to San Luis Obispo, which also ran on past Los Angeles Union Station and came back in a straight line; the California Zephyr near Winnemucca, drawn on the other of Union Pacific’s two paired tracks; the Sunset Limited between Beaumont and Houston, drawn on the other of two railways; the Maple Leaf, sketched from Niagara Falls to Toronto; and the Capitol Corridor between Roseville and Auburn. All five now lie on their own trains’ track. Found by the atlas’s own audit sweep.
799 routes now have a timetable that confirmed their own train, up from 793. Found by the atlas’s own audit sweep.
R10 Krakonoš from Prague to Trutnov now lists its 15 stations, from the timetable, as its page says. Found by the atlas’s own audit sweep.
World Beach Map and Detourmap join Movie Scene Map, Castle Map and Sunshine Atlas under More maps from the same creator, on the About page and at the end of the page of all routes, each with a picture of its map. Changed at the maker’s request.

19 October 2026

The route card has a new layout, the one Movie Scene Map’s place card has: the route’s type, its fame rank and the year it opened as chips on one line, ended by the close button, now a raised button like every other button on the map; Full page and Wikipedia beside the route’s name; the countries with their flags as pictures, which Windows used to show as two letters; and a taller photograph with its licence beside the photographer’s name. Changed at the maker’s request.
A new line on the route card names the timetable that confirmed the route’s own train, for the 793 routes that have one. On an iPhone, a card pulled past its top or its end now shows the map behind it instead of a band of the card’s colour. Changed at the maker’s request.
Sunshine Atlas joins Movie Scene Map and Castle Map under More maps from the same creator, on the About page and at the end of the page of all routes, with a picture of its map. Changed at the maker’s request.
The search box’s clear button and the map’s other close buttons are the same raised button. The map’s credits fold into their (i) five seconds after the map appears, on every screen. Changed at the maker’s request.
The welcome card no longer comes on the map. Changed at the maker’s request.
The strip at the top of every page is a plain, compact dark blue without the photograph, with the chip Partnership and a shorter pitch. Buy me a coffee has a line of its own under it: a coffee supports Mr. Flightmussy personally while he develops World Train Map further. Changed at the maker’s request.
Station names are on when the map opens, and their switch now sits under the filters. A press still turns them off, and the map remembers it. Changed at the maker’s request.
The About page no longer says that a person checks every fact. It says that AI helps write the code and gather and check the data, and that each route’s page says what its line was checked against. Changed at the maker’s request.
Three routes join the map: IC Łempicka from Lublin through Radom, Kielce, Częstochowa and Opole to Wrocław; IC Kasztelan from Kraków through Tarnów, Mielec, Stalowa Wola, Biłgoraj and Zamość to Hrubieszów; and the IC-04 from Poperinge through Ypres, Menen and Kortrijk to Ghent and Antwerp. About 690 km of their lines were under no other route. Each is drawn from its own OpenStreetMap relation, checked against the train the planner runs on it, and dated by the year its last section opened: 1911, 1976 and 1970. Found by the atlas’s own audit sweep.
The Antwerp to Ghent card said 1847, the year its line reached Ghent. Its trains reached Antwerp’s own stations only on 1 January 1970, through the Kennedy tunnel under the Scheldt, so the card now says 1970, by the rule every card follows: the year the whole journey became possible. Found by the atlas’s own audit sweep.
Five more routes list their stations: the three new ones from their own OpenStreetMap relations, and the Frecciargento from Rome to Lecce and the EC Transalpin from Zürich to Graz from the operator’s timetable, as their pages say. IC Brda and Poznań to Gdynia now spell their stations in Polish, Łowicz Główny rather than Lowicz Glowny, from their own relations; the stations are the same. Found by the atlas’s own audit sweep.
Hull Trains from London to Hull is now confirmed against a live timetable. It was one of 138 routes in Europe asked again because they had no recent answer from the planner; most of the others are heritage and mountain lines the planner does not carry, and they stay as they were. Found by the atlas’s own audit sweep.

12 October 2026

Every page now opens with a strip for a partner, from edge to edge: the photograph of Bergensbanen under a dark tint, the words Your brand here, and a sentence saying what it is for. World Train Map is free and independent, and we are looking for one partner to support it, with full exclusivity: their brand alone, on every page and on the map. A line under it says the advertising is never invasive and that each partnership is tailored with care to both the readers’ experience and the partner’s wishes. On the photograph, Learn more leads to Flightmussy Studio, and beside it, at the same size, the Buy me a coffee sign and its cat are there for anyone who would like to support the creator; a press anywhere else on the strip opens the studio too. The strip stands outside each page’s own content, so search engines read the pages as before, and the photograph names its author. It is not on the privacy notice. Movie Scene Map’s pages carry the same strip. Changed at the maker’s request.
The cat with the Buy me a coffee sign no longer flies across the foot of the map. It stands on the strip at the top of every page instead, and the foot of every page still links the same Buy Me a Coffee page. The entries of 2 and 6 October describe the banner that is now gone, and the privacy notice says where the sign is now. Changed at the maker’s request.
The welcome card on the map is smaller and comes on every visit to the map, once, a minute in, for ten seconds, with a blue line along its foot that runs down with them. On a computer it stands in the map’s top right corner, level with the panel, and moves left of a route’s card when one is pinned; on a phone it lies across the foot of the screen, and while a route, the panel or the search is open there it steps aside, coming back for the rest of its seconds when that is closed. Nothing done with the map takes it away before its time: the X and the Escape key close it at once, and a press on the card keeps it until it is closed. Its photograph no longer moves as it arrives. Until today it came only when you arrived from another site or switched language, ninety seconds in, and gave way as soon as you used the map; the entries of 3 and 10 October describe that card. Movie Scene Map’s card changed the same way. Changed at the maker’s request.
The About page and the page of all routes now end their account of the map with More maps from the same creator: Movie Scene Map and Castle Map as cards, each with a picture of its map, and a sentence saying that Mr. Flightmussy, who makes this map, makes and runs both at Flightmussy Studio. On the About page they take the place of a line of links. The About page’s description no longer says no ads, and its studio card names the studio’s three maps. Changed at the maker’s request.
The sister site that was called Destinations Map is Sunshine Atlas again, at sunshineatlas.com, and the foot of every page and the Contact page now link it by that name. Changed at the maker’s request.
Dashes used as punctuation are gone from the site’s own writing. The route hooks and stories, their German translations, the page titles, the About, Contact and privacy pages, the data stories and this log used a long dash where a comma, a colon, a full stop or a pair of brackets says the same thing more plainly; each is now the punctuation it stood for, in all three editions, and the wording is otherwise unchanged. Route page titles separate the route’s name from the rest with a colon, and the one route whose name carried a dash is now West Highland Line to Oban. What stays is what a dash is for: a range such as Paris–Lyon or four to five trains a day written as 4–5, the names of stations and lines that are written with one, and the mark for a figure nobody publishes in the table of every route. Changed at the maker’s request.
The About page is shorter, plainer and more concrete. It opens with one real route as its example, Bergensbanen from Oslo to Bergen, with the length, fastest journey, operator and opening year the map itself holds for it, and it now says how much line the map holds as well as how many routes. The three rules for what goes on the map are a list, and the account of how the drawings were checked keeps its facts in fewer words. The history of how the map grew and the comparison with OpenRailwayMap are gone, in all three editions. Changed at the maker’s request.
On a phone or a small tablet the top of every page takes less room, which matters because it stays in view while you scroll. The name, the language flags and the light and dark switch now share the first row, and the links sit under them on one row, or two on a narrow phone. Before, the same header took up to four rows. Changed at the maker’s request.

10 October 2026

The welcome card no longer opens with the map. When you arrive at the map from somewhere else, or switch it to another language, the card now comes ninety seconds later. The seconds are seconds of the map on your screen: while it sits in a tab you are not looking at, the clock stands still. If you have a route open when the time comes, or are typing in the search, or have the panel open on a phone, the card waits until you have closed it, so it never lands on what you are reading. The entry of 3 October said the map opens with the card; this replaces that. Its blue Continue to the map button is gone: a bold X on a solid round button in the corner of the card closes it. The line under its heading, The first thing visitors to the map see, is gone too, since it no longer is. Movie Scene Map’s card changed the same way on the same day. Changed at the maker’s request.
What stays as it was. The card still comes only on those arrivals, never on a reload, a step back, a link from the site’s own pages or a link to a route, and nothing is stored about it. Once it is there it stays for at least five seconds and then gives way as soon as you use the map; what you did with the map before it came does not count. The X and the Escape key take it away at once. The coffee cat still waits while the card is on the map, and at ninety seconds the card now comes before the cat’s two minutes. Changed at the maker’s request.
What it does not do. The ninety seconds belong to one visit of the map: leave it for one of the site’s pages before they are up and come back by a link, and the card does not come, because a link from the site’s own pages never opens it. Changed at the maker’s request.

7 October 2026

Eighty-eight more routes are now drawn from OpenStreetMap’s own route relations, node for node: 1,381 of the 1,516, and 1,094 of them from end to end. Twenty-three routes that a relation drew for only part of their length are now covered further, 20,805 km of line in all. The relations were found in new ways that do not depend on the language a relation is named in: the trains whose stops lie along the route, anywhere on the planet; the route’s own Wikidata item; the railway lines Wikidata says its stations are on; and the lines whose track runs under it. Many are named in their own country’s script, such as Железопътна линия 16 for Septemvri to Dobrinishte. Every new drawing moved its line by less than a kilometre and passed the same checks as before. Where it would have left a station the train calls at, as at Kilkenny, Huesca and Valencia, or where the timetable no longer confirmed it, the old line was kept. Each route’s page names the relations it is drawn from. Changed at the maker’s request.
Google Analytics is gone from the site. Since 8 September every page had loaded it from Google; now no page does, nothing about a visit is sent to Google, and the only counting left is this site’s own, which is anonymous and stores nothing on your device. The cookies Google Analytics set are read by nobody now, and the map deletes them the next time it opens. The privacy notice says so, and the site’s security policy no longer lets a page reach Google’s addresses at all. Changed at the maker’s request.
On the About page, Featured in and the thanks to the two forums whose members have corrected the map the most now come at the end of the page, after the account of how the map is made and checked, instead of at the top. Changed at the maker’s request.
The About page now says who keeps the map, the way Castle Map’s does: it is made by one person, under the pseudonym Mr. Flightmussy, with a card quoting his own words from X beside a photograph of his desk, and under it Flightmussy Studio, a professional studio for interactive maps in Oslo, which also makes Movie Scene Map, Castle Map and Destinations Map, with the studio’s logo, postal address and site. Neither it nor the Contact page calls the map a hobby project any more. Changed at the maker’s request.
Seventy-five more routes list their stations, 1,153 of the 1,516, with 16,190 stations in all. Seventy-one of the new lists come from the operator’s timetable, for routes that no open source could list: the calls of the train the atlas’s own timetable check had identified as the route’s, read from ÖBB Scotty on one date, checked against the route’s line the way every other list is, and named by OpenStreetMap’s station where it has one. The route’s page says so, names the train and the date, and says that other runs of the train may skip some of the stations. Such a list is one train’s calling pattern on one day: the nonstop TGV from Paris to Bordeaux lists its two ends, the Vereina car shuttle its two. A list is still refused where a call lies off the route’s line, as the InterRegio from Bucharest to Iași via Tecuci does, where the train the planner found is another operator’s, such as RegioJet’s for the ČD Pendolino, or where it is a day train for a night train’s route. The other four new lists come from OpenStreetMap, where a station of a Wikipedia list that nothing had placed is now placed by the planet’s station of that name along the line, and a list that ran out short of an end is completed with the station there. Some existing lists changed with the day’s redrawing of lines from OpenStreetMap relations and with OpenStreetMap’s own edits: a few halts that are gone from the map are gone from the lists, and the Valencia to Alicante line lists eight more of its stations. Junctions, signal stations and track mis-tagged as a stop no longer get into a list, and English names lose their "Railway Station". Changed at the maker’s request.
On a phone, the search box now stands on the map itself, above the Browse routes, Surprise me and About buttons, the way Castle Map’s does, so a route can be found without opening the panel. Its results open over the buttons, and choosing one pins the route. On a phone the panel no longer carries a search box of its own, nor a Filters heading: the box above the buttons is the search, and the sliders pill in the top bar is what opens the filters, so the panel opens straight on Station names and the filters, the way Castle Map’s does. On a computer the panel keeps its search box and its Filters fold. Three things Castle Map had already put right came along the same day: the list of results is opaque, so the buttons’ labels no longer read through its rows; it is cut to the part of the screen the keyboard leaves and scrolls, instead of standing on under the keyboard out of reach; and a tap anywhere on the map, or Escape, closes it, where before only a tap outside the box did and a tap on the map itself did not count. Changed at the maker’s request.
The map no longer opens on Antarctica. On a computer the first screen is centred a little further north, at 27°N instead of 20°N, so the Arctic lies under the top edge and the Antarctic coast under the bottom one while every route from Narvik to Patagonia stays on the screen. On a phone, where the whole world at its own width was a strip across the middle of the screen with Antarctica the largest thing on it, the map now opens with the North Pole at the top of the screen and 57°S at the bottom, Europe in the middle, the way Castle Map has opened since the same day; the whole world is still one pinch away. Returning to the world view goes to the same picture. Changed at the maker’s request.
The map now says, on its first screen and on the About page, what sets it apart: it maps every passenger rail service that runs today, each route checked before it goes on, and nothing else. A map drawn from every railway line in OpenStreetMap shows many lines that carry no passenger train any more; here there are no freight-only lines and no closed railways, and a line under construction waits until there is a train to ride. The line under the map’s name, the About page’s opening and the page descriptions say so in all three editions. Changed at the maker’s request.
The panel beside the map no longer ends with the More maps line and the Follow on X line. The sister maps are still linked at the foot of every page. On a phone, the About button now sits in the first row beside Browse routes and Surprise me, instead of on a row of its own under them. Changed at the maker’s request.
The library that draws the map, MapLibre GL JS, moved from version 5.24 to 6.13. Nothing visible changes: the opening views on a computer and on a phone are the same picture to the pixel as before, a zoomed view of the Alps differs only along the seams of the relief tiles, panning, zooming, hovering and pinning a route run as smoothly as before, and on an iPhone, where the map runs on under the status bar and the toolbar, the bounds of the world are still judged on the visible part of the screen, now by the library itself rather than by the site’s own copy of its rule. One thing a reader may notice: the map now needs WebGL 2, which every current browser has had for years; a browser without it is told that the map cannot be drawn, as before, instead of being left with a blank space. The upgrade closes a security advisory about the attribution text that the site’s security policy had already blocked, and the dependencies of the build and of the MCP server were brought up to date with their own advisories at the same time. Changed at the maker’s request.

6 October 2026

The cat with the Buy me a coffee sign no longer flies in as the map opens and stays at its foot. It now comes two minutes after the map has opened, is there for thirty seconds, and leaves the way it came: it turns round and pushes its sign back out to the left. Ten minutes after it has left it comes again, for thirty seconds again, and so on for as long as the map stays open. The entries of 2 October (it flies in when the map opens) and 3 October (it flies in once the welcome card has gone) are replaced by this clock, and the cat still never comes while the welcome card is on the map. The cat on Movie Scene Map keeps the same clock. Changed at the maker’s request.
The minutes are minutes of the map on your screen. While the map sits in a tab you are not looking at, the clock stands still, so the cat does not come and go unseen. It is not taken from under you either: if you are moving your pointer over the sign when the thirty seconds are up, the cat waits until the pointer has moved off or lain still for a few seconds, and if you followed the sign to Buy Me a Coffee, it is still there with its cup when you come back and leaves a few seconds later. Changed at the maker’s request.
What stays as it was: a press on the cat sends it away at once, and a cat sent away with a press does not come back until the map is loaded again. A gust still moves it, once in the middle of each visit. With reduced motion set in your browser the sign appears and disappears without flying. Nothing is stored about any of this: the clock is a timer in the page, and it starts again with every load. What it does not do: on a phone an open route card still covers the sign, so a visit of the cat can pass behind it unseen. Whoever wants the page at another moment finds the same link at the foot of every page. Changed at the maker’s request.

5 October 2026

Ninety-six more routes list their stations, 1,078 of the 1,516, with 15,279 stations in all. The new lists come from OpenStreetMap in three new ways. A route that is a line takes the stations that lie on that line’s own track in OpenStreetMap, the stop positions and stations on its rails and the station buildings its track runs through, so a high-speed line now has its stations too, without the older line that often runs beside it: Beijing to Shanghai lists its 23. A route that is a train takes the stops of a train mapped in OpenStreetMap that runs it end to end, found anywhere in the world rather than only near the two ends as before, such as the IC 56 from Norddeich Mole to Leipzig or the İzmir Mavi Treni. And a funicular lists its two stations, one at each end, where before a list needed three. Changed at the maker’s request.
Every new list was read before it went up, and a list is still left off rather than guessed. A train’s list is refused when the mapped train is plainly not the one on the card, such as a local train for an Intercity, or when it runs past a city’s main station without a stop, which says some of its stops were never mapped. Where OpenStreetMap does not record which stations a train calls at, as for most long-distance and luxury trains, the route has no list yet. Changed at the maker’s request.
Stations look different on the map. A station where two or more lines meet, stopping or ending, is a white disc with a dark ring and its name in bold, and the disc grows a little with the number of lines. The end of a line that no other line reaches is a disc in the line’s own colour, so the line visibly ends in it, and it now carries its name from a country’s zoom; at a city’s zoom an end is named after its station, Paris Gare du Nord rather than Paris. The stations of a route you point at or pin are white beads ringed in its colour, a little wider than the line so they stand out on it, and the stations where other lines meet it are the white discs. The same beads and discs are drawn when Station names is on. Every marker sits on its line, and a name tries the side away from the line first. On a route’s own page the list of stations draws them the same way, and a +N after a station’s name says how many other routes on this map stop or end there. Changed at the maker’s request.

3 October 2026

When you arrive at the map it now opens with a card. It says that Flightmussy Studio is seeking an exclusive partner for World Train Map, and it is the thing on offer: the card a partner would have, with a photograph, a name, a line of their own and a message. No partner is in it yet, so the heading reads Your brand here and the photograph shows Bergensbanen in Norway, as an example. The picture is most of the card, and the words under it are short. It has two buttons and nothing else to press: Continue to the map, in blue, takes the card away at once, and so does the Escape key; the other opens the studio’s site in a new tab. Movie Scene Map opens with the same card. Changed at the maker’s request.
The card opens when you arrive at the map from somewhere else, a search engine, a link or the address bar, and when you switch the map to another language. It does not open when you reload, when you step back, when you come to the map from one of the site’s own pages, or when you follow a link to a route, which opens on that route instead. Nothing is stored about it: the card reads only where you came from. It stays for at least five seconds once it has appeared. After that it also leaves by itself as soon as you press, drag or zoom the map, choose a route, or open the panel on a phone; in its first five seconds the map can be used under it, and the card gives way when the five seconds are up. While the card is on the map the coffee banner waits, and the cat flies in once the card has gone. Changed at the maker’s request.
The photograph is a train from Oslo to Bergen mirrored in a lake on the Hardangervidda, between Haugastøl and Finse, by Kabelleger / David Gubler, from Wikimedia Commons under CC BY-SA 4.0, and credited on the card with a link to its page. It is served from this site, so the card asks nothing of Wikimedia. What the card is and is not: the site offering its own first screen to a partner, in its own words. The site still loads nothing from an advertising network and sells nothing about its readers, and when a partner takes the card, the card will say whose it is. Changed at the maker’s request.

2 October 2026

At the foot of the map there is now a purple button that says Buy me a coffee, and a ginger cat tows it there the way a small plane tows a banner. The cat is Mr. Flightmussy, the same one that flies on Movie Scene Map. It flies in from the left when the map opens, a gust moves it now and then, it rolls when you press the button, and when you come back from the page the button opened it is holding a cup. At rest nothing moves, and with reduced motion set in your browser nothing moves at all. Press the cat instead and it turns round and pushes its sign back out the way it came, so the map is left clear; nothing is remembered, and the cat is back the next time the map opens. Changed at the maker’s request.
The button opens the maker’s page on Buy Me a Coffee in a new tab, in German on the German edition and in English on the others, because those are the languages that site has; its words are the same in every edition, because they are the name of the service. Nothing on this site loads from Buy Me a Coffee: the button is an ordinary link, so that company learns of a visit only when someone follows it. The Donate button to PayPal at the end of the About page is gone, the foot of every page links the Buy Me a Coffee page instead, and the privacy notice says so. PayPal is no longer linked anywhere on the site. Changed at the maker’s request.

28 September 2026

The two links at the foot of a pinned route card, Full route guide and Wikipedia, are buttons now, raised like the panel’s: the blue one opens the route’s own page, the grey one its Wikipedia article, and both stay in view at the foot of the card while a longer card scrolls past them. On a computer, a card taller than the window now scrolls inside it instead of running off the bottom, which a phone held sideways could do. Movie Scene Map’s cards took the same step on 26 September. Changed at the maker’s request.

27 September 2026

Station lists now reach 982 of the map’s 1,516 routes, 13,973 stations in all, up from 695 routes and 9,190 stations. A route’s own OpenStreetMap relation is still where its list comes from first, and 727 are read that way. Where a route has none, the list comes from another train’s relation that runs the whole route and carries the route’s name, or, for a line, from any train that runs it end to end (32 routes); then from Wikidata, where each station names the train or the line it belongs to (102); for a line, from the stations OpenStreetMap maps along its track (101), though never for a high-speed line, which runs beside an older line for long stretches and picked up that line’s stations; and last, from the route’s own Wikipedia article (20): a section listing the train’s stops, or a table marking which of its trains stop where, or, for a line, its route diagram or station list, each station found through its Wikidata item, and only where every station the list names can be placed on our line, so a list with one missing station is left out rather than shown incomplete. From a table only the trains that run the whole route count: the Hida’s article also marks the stops of trains that turn back at Takayama and of trains from Osaka that join the line at Gifu, and the Hida from Nagoya to Toyama lists the ten stations where the trains that run the whole way stop. That last source brought the Kamui, the Sōya, the Kairi, the Hida, the Shinano, the Yakumo, the Kuroshio, the Okhotsk, the SL Banetsu Monogatari and the SL Paleo Express, the Island Express to Kanniyakumari, the V-Train through the Baekdudaegan gorges and the Beijing–Harbin, Nanning–Kunming and Zhengzhou–Xi’an high-speed lines, among others. A train’s relation named differently from the route may also count, when the journey planner shows the route’s own train calling at exactly its stations; of the eight routes asked, none did: the relations of the other trains between Hamburg and Berlin stop at Büchen and the ICE does not, the Lötschberger’s relation leaves out Münsingen, and for the Frecciarossa from Paris to Milan the planner offers only TGVs. Every list has to run the whole route in order and begin and end where the route does: a list that stops short of an end is left out rather than shown without its terminus. A line’s list from Wikidata is shown only if it holds at least two thirds of the stations OpenStreetMap maps along the same track, and junctions, signal posts, freight yards and depots never appear. Where no source passes, a route keeps no list rather than a wrong one, and each route page says which source its list came from. Changed at the maker’s request.
The map’s panel has a Station names switch. Turned on, it shows the stations of every route the map is showing, from a country’s zoom inwards: a small tick across the line with the station’s name beside it, the busier stations keeping their names where the map is crowded and the others appearing with the next zoom. The map remembers the choice for your next visit, and a link ending in ?stations=1 opens it with the switch on. Until now a route’s stations appeared only while the route was hovered, and the card that comes with the hover could cover some of them. Suggested by a reader.
The Japanese edition reads better on a computer and on a phone. On Windows its Japanese text was drawn in the thin Regular cut of Yu Gothic, grey beside the Latin letters, and is now drawn in the Medium cut Japanese sites use; its smallest labels, 8 to 10 pixels, are a step larger; and labels no longer turn the Latin words inside them into capitals. The map speaks Japanese too: its own place names come from the map tiles’ Japanese names (岐阜市, 静岡市 instead of Gifu, Shizuoka), and the markers at the ends of routes now carry the edition’s names (東京, 大阪) where they carried the English ones. 108 route ends in Japan, China, Korea, Taiwan and Hong Kong that had no Japanese name, and printed in English beside Japanese station names, now have one (長野, 敦賀, 東京（新宿）), each confirmed on Wikidata or read off the terminus station of the route’s own OpenStreetMap relation. On a phone the open panel now stops above the map’s credit line instead of running over it. Changed at the maker’s request.
A route’s photograph now comes with its card instead of after it. While the pointer is on its way to a line, the map starts fetching the photographs of the lines just ahead of it, at most two at a time, and the card asks for its own photograph before anything else the map is loading instead of after it. Measured on thirteen routes, moving the pointer onto each line: the photograph used to arrive a median 64 ms after the card on a fast connection and 202 ms on a slower one like mobile data; it is now there when the card opens on the fast connection, and 53 ms after it on the slower one. Nothing is fetched ahead on a touch screen, which has no pointer to follow, or when the browser asks to save data, and on a computer the photographs themselves are unchanged. Changed at the maker’s request.
On a phone, the photograph on a route’s card is sharper. The card drew it from a copy 500 pixels wide, but on a phone the card spans nearly the whole screen, and a phone’s screen draws two or three of its own pixels for every pixel of a page, so the photograph was stretched to more than twice its width. A touch screen now gets the copy 960 pixels wide, close to all the detail a phone’s screen can show. That copy is about three times the download, a median of 194 KB against 61 KB: measured on eight routes in a browser set up as a phone, the photograph now arrives a median of 28 ms later than before on a fast connection and 145 ms later on a slower one like mobile data. A computer, whose card is narrow enough for the smaller copy to be sharp, keeps that one. Changed at the maker’s request.
The Donate button has moved from the map’s panel to the end of the About page, under a few lines on what a donation pays for. The panel keeps three buttons, Browse routes, Surprise me and About, and is narrower on a computer, 304 pixels instead of 360 (376 in German), so a little more of the map shows beside it; the three sit centred and stretch across the panel’s width. The privacy notice now names the About page as the one place PayPal is reached from. Changed at the maker’s request.
188 photographs are cropped again so that the train stays in the frame on a phone as well as on a desktop. On a phone a route page shows its photograph in a squarer frame, which cuts the sides of a wide picture, and every crop had been aimed only up and down, for the strip on the map’s card, so a train whose nose was near the edge of its picture lost it there. All 1,507 photographs were looked at with every crop they get drawn on the whole picture: 165 now aim sideways as well, and 28 had the card’s strip moved to where the train is. The strip on the card also keeps its shape on the largest phones instead of flattening. Changed at the maker’s request.
The map is lighter to move. The route card, the phone’s bottom sheet and the zoom buttons no longer blur the map behind them, which the browser had to redo on every frame the map moved underneath: measured while dragging the map with a route open, the graphics work fell by about half on a phone’s screen and by a quarter to a third on a laptop. The card is solid now, like the panel. The first route picked in a visit also no longer stalls for about a tenth of a second while the drawing program for its station names is prepared: those names now use the one the map prepares while it loads. Changed at the maker’s request.
Every European route without a timetable check was asked about again, 267 of them. One more, the Harzquerbahn, is now confirmed against a running timetable. The others are funiculars and heritage lines the journey planner does not carry, journeys it offers only with a change of train, or lines where the planner’s train runs along a different track for part of the way; they stay unconfirmed rather than claimed. Four routes added on 23 September are now confirmed against their operators’ own timetables: VIA Rail’s Montréal–Senneterre and Metlink’s Wairarapa Connection along the route shape each operator publishes, and Amtrak’s Mardi Gras Service and Illinois Zephyr by their calling stations. Five more are now confirmed by their operators’ own timetables where the only doubt had come from the journey planner answering about a different train, as its own reply says: Oslo–Stockholm (Entur), Bordeaux–Marseille and Paris–Montpellier (SNCF), and Madrid–Alicante and Madrid–Ourense (Renfe). Found by the ÖBB Scotty re-sweep of every unverified European route.
Six routes in Taiwan are now checked against a running timetable: the Pingxi, Neiwan, Jiji, Shen’ao and South-Link lines and the Tze-Chiang to Taitung. Taiwan Railway publishes its own timetable, one file for each day, and its stations on its open data server under Taiwan’s Open Government Data License, and the check reads it as it reads other railways’ published timetables: for whether a train runs, and whether our line passes the stations it calls at. The Jiji and Shen’ao lines have no train that starts and ends at both of their stations, because the trains run on beyond one end, so the check now judges a line on the part of a longer train between its two ends, and any route on the part of a train that runs on past one of its ends: the Tze-Chiang to Taitung and the South-Link line are confirmed that way. A card for a named train is checked only against its own kind of train, a Tze-Chiang against Tze-Chiang trains and never the local trains on the same rails. The Puyuma and the Taroko stay unconfirmed, because the published timetable does not say which Tze-Chiang trains are Puyuma or Taroko trains. Changed at the maker’s request.
A timetable check that can compare only stations, because the timetable publishes no shape for the route, now needs those stations spread along it: no stretch of more than 250 km without one. Until now the check counted them, and a count says nothing about where they are: the night train from Paris to Nice calls at eight stations, and at none on 875 km of its route, where our line could run along any railway at all. Ten routes lose their timetable claim for this reason, their stations still on our line but too far apart to show which railway it takes between them: the night trains from Paris to Nice and to Briançon, the TGVs from Paris to Nice, Marseille, Montpellier, Barcelona, Quimper and Brest, the TGV Lyria to Zurich, and the Eurostar from London to Brussels, which calls nowhere in the 265 km from London to Lille. Paris–Montpellier, confirmed this morning, is among them. Three routes are newly confirmed because the check now judges a route on the part of a train that runs on past its ends: Brussels–Luxembourg against SNCB’s timetable, and Nahariya–Tel Aviv and the Jezreel Valley line against Israel Railways’. With the six in Taiwan that makes 789 routes confirmed against a running timetable. Found by the atlas’s own audit sweep.
Three lines are drawn again, all three found by reading the stations their Wikipedia articles list. The Hida took the Meitetsu line from Nagoya to Unuma, a way its trains never go, and Gifu lay 13 km from it: it now runs on JR’s Tōkaidō line to Gifu, where the train reverses, and up the Takayama line, 254 km where the card says 256. The Zhengzhou–Xi’an high-speed line followed the older Longhai railway for long stretches, three of its stations between 1.4 and 7.2 km from our line, and now follows the high-speed line itself, 474 km; its card said 523 km and now says 474. The Shinano’s line ran on 9 km past Nagano and doubled back, and now ends at Nagano station. The Hida and the Zhengzhou–Xi’an line are drawn from their OpenStreetMap relations, and all three are measured on OpenStreetMap track. Found by the atlas’s own audit sweep.

26 September 2026

The note in the corner of the page is gone, and so are the Partnership button in the map’s panel and the page both of them opened. The panel keeps its Donate button beside About, and with one button fewer it is narrower on a desktop. The privacy notice no longer lists the time the note kept of when you last closed it. Changed at the maker’s request.

23 September 2026

World Train Map has a Japanese edition, its third after English and German: every page of the atlas under /ja/, with the map, its panel and the route cards in Japanese. Route and place names are taken from Japanese Wikipedia and Wikidata, never made up, and where neither has a Japanese name a route keeps its English one. The descriptions of the routes are machine-translated from the English, as the German ones were. This changelog and the privacy notice stay in English. Changed at the maker’s request.
On an iPhone, Safari now shows the map under its floating toolbar at the foot of the screen and under the clock at the top, where it used to show a strip of plain colour at each end. Nothing moves to make room: the panel, the buttons, the route card and the map’s own controls stand where they stood, and the map is centred on the part of the screen that can be seen, not on the parts under the glass. A tap on the clock, which in Safari scrolls a page to its top, brings the map straight back. Every other browser, and the embeddable map, is unchanged. Changed at the maker’s request.
Also for phones, from the sister atlases: no grey flash when a button is tapped and no rubber band when the map is pulled past its edge, no scroll bar on a route card on a touch screen, no telephone links made out of numbers, and on a phone held on its side the panel, the card and the corner note keep clear of the notch and the home indicator. Changed at the maker’s request.
The map’s panel is arranged the way Movie Scene Map’s is. On a desktop it is one panel: the name with the night and day switch beside it, the language flags, the search box, the buttons on one row, one Filters fold and a line to the other maps at its foot. On a phone it is a bar with the name and the Filters button, the flag and the switch as two squares beside it, and the buttons in two rows under it. On the German map the search box also finds a route by its German name. Changed at the maker’s request.
The note in the corner of the map no longer asks for a donation. It says the atlas is looking for partnerships and sponsorships, and its Learn more button opens a new page that describes what a partnership can look like; the panel has a Partnership button of its own, between About and Donate. The privacy notice says what the note stores, which has not changed. Changed at the maker’s request.
This page is called the changelog now, which is what it had become: it has carried changes to the map beside the fixes to the data since August. Its address and its feed stay where they were. Changed at the maker’s request.
The map now says what it is working towards: every passenger rail service in the world. It used to call itself a curated atlas of notable routes, not every railway on Earth; the line under its name, the About page and the descriptions that search engines read now say what it aims at and how far it has got. The About page was rewritten to describe the site as it is today: how routes are found and added, how each line is checked, and where the names, photographs and station lists come from. Changed at the maker’s request.
Sixteen routes joined the map, 3,258 km of line, about 2,500 km of it never drawn here before: in the United States the Mardi Gras Service along the Gulf Coast and the Illinois Zephyr and Carl Sandburg across the prairie to Quincy; in Canada VIA’s flag-stop train from Montreal to Senneterre; in New Zealand the Wairarapa Connection; in Australia the Bairnsdale line across Gippsland, the Hunter Line to Scone and the Xplorer to Armidale; in Japan the Kounotori to Kinosaki and the Etsumi-Hoku Line to Kuzuryūko; in Korea the AREX express to Incheon Airport; Banja Luka to Doboj in Bosnia and Herzegovina; Macomer to Nuoro in Sardinia; on Java the Banyubiru from Solo to Semarang and the Batara Kresna railbus to Wonogiri; Varna to Dobrich in Bulgaria; and Nador to Taourirt in Morocco. Changed at the maker’s request.
Each of the sixteen is drawn from its own OpenStreetMap relation and passed the same checks as every other line. Fifteen were then measured against the rail graph and sit on real track for their whole length, four metres from it at the median; the sixteenth, Macomer to Nuoro, is narrow gauge, which the rail router does not know, so it rests on its relation alone. Each carries an opening year read from the line’s own article, and a train running on it today, confirmed from the operator or the line’s article rather than assumed from the relation. Found by the atlas’s own audit sweep.
Every photograph on the map was looked at again, 1,491 of them. 273 were replaced by a better picture of the right train on the right line, 58 that had been shown small are now shown full width on their route’s page because Wikimedia Commons holds a large enough copy, and each crop was checked on the card itself, so the card shows the train rather than sky or gravel. Changed at the maker’s request.
The routes that had been checked least were checked again. 48 more are now confirmed against a running timetable, 780 in all, and 40 more lines were measured against the railway track, so every line on the map now carries that measurement except Macomer to Nuoro. Where a journey is made of several trains, the note on how a line was checked now names the main train rather than the first one the journey planner listed. Changed at the maker’s request.
Route pages were redesigned, after the Castle Map’s and Movie Scene Map’s: a wider page with the facts in a column beside the story (above it on a phone), a larger photograph, buttons that open the route on the map, its Wikipedia article and its stations, the questions and answers as folds, and ‘How this line was checked’ beneath ‘Along the way’. Changed at the maker’s request.
Route pages list the stations the train calls at: 9,190 stations on 695 routes, read from each line’s own route relation in OpenStreetMap and drawn as a line with its stops. A route whose relation does not list its stops in order, or whose stops could not all be placed on the line, shows none rather than a wrong list. A station is listed under its own name, not a platform’s: a stop mapped as “Central, Platform 1” reads Central. On the German and Japanese pages a station whose own name is in a script those readers do not use, such as Japanese or Greek on a German page, is given by its English name where OpenStreetMap has one. Changed at the maker’s request.
Route pages show one photograph, the same one the route’s card on the map shows. The gallery of three or four more pictures that followed each page’s story is gone for good. Changed at the maker’s request.
Hover over a route on the map, or tap it on a phone, and its stations appear along it: a small tick across the line at each one, with its name beside it. Where stations crowd at a small zoom, more of them appear as you zoom in. It works on the 695 routes that have a station list; the others show their two ends, as before. Changed at the maker’s request.
The Filters fold has a new section, How often: hourly or more, several a day, once or twice a day, and weekly or seasonal. It is known for 456 of 1,516 routes so far: counted on a weekday in the journey planner, or taken from the pattern the operator publishes, and 172 more routes were counted for it today. While a band is chosen, the routes not counted yet are hidden. The route card shows how often a route runs as a line of its own, and a route page gives the weekday count where no pattern is written. Changed at the maker’s request.

20 September 2026

Eleven railways joined the atlas, 923 km of line of which about 528 km had never been drawn here. In North Zealand the Hornbæk line, the Gribskovbanen’s western arm to Tisvildeleje and the Østbanen across Stevns; in the Netherlands the Utrecht–Rotterdam railway and the West Friesland line north through Purmerend; Santhià to Biella in Piedmont; the Chiltern branch from Hatton into Stratford-upon-Avon; the IC-24 from Charleroi to Leuven; the Lower Silesian sprinter from Forst over the Neisse to Wrocław; the EuroCity Via Regia from Leipzig to Wrocław; and LTG Link from Šilutė along the edge of the Nemunas delta to Kretinga. Found by the atlas’s own audit sweep.
Each of the eleven is drawn from its own OpenStreetMap relation rather than sketched, checked against a scheduled train on the planner, and measured against the rail graph afterwards: all eleven sit on real track for their whole length, three metres from it at the median. Each carries an opening year read from the line’s own article in its own language, and where a route is several railways end to end the year is the one the last of them opened, because that is when the journey first became possible. Found by the atlas’s own audit sweep.
One candidate was dropped on the last check rather than added: a Croatian regional from Virovitica through Daruvar to Banova Jaruga, which the planner answered but which is in fact closed for rebuilding, with buses in place of trains until December. A timetable answer is not the same as a train, and the rule stands that operation is verified before a line goes on the map. Found by the atlas’s own audit sweep.

14 September 2026

A picked route now lands where a phone can see it. The flight that frames a route used to reserve a fixed strip at the top and just under half the screen at the bottom, while the bar and the buttons take more than that at the top and the route card can rise to well over half; the line arrived with its head under the buttons and its foot under the card, and the reader had to drag it out. The page now measures its own bar, buttons and card at the moment of the flight and keeps the route out from under all three. The same measure applies on a desktop, where the panel on the left and the docked card on the right are kept clear too, and to every way of choosing a route: a tap on the line, the search box, Surprise me and a shared link. Changed at the maker’s request.
The chosen route is drawn to be unmistakable: the stroke half as wide again, an outline in the chart’s ink on either side, cream on the night chart and dark on the day chart, and a wider, brighter halo, while every other line holds at half strength so the network around it can still be read. Changed at the maker’s request.
The route tiles are cut at the map engine’s own simplification instead of sixteen times finer. The tiler had been told a tolerance in the wrong unit, so every tile carried points no screen could show; the drawing is the same to the eye, the tiles are smaller, and the graphics chip has far fewer vertices to push through at the world view, which is where every line is on screen at once. Found by the atlas’s own audit sweep.
The map is a flat chart now, the way the Castle Map draws its world, instead of a globe. Opening the atlas shows the whole network from Vancouver to Christchurch across the width of the window, and the chart stays north up: it can no longer be tilted or spun. The globe was also the heaviest thing this map asked of an ordinary laptop’s graphics chip. At the world view every line is on screen at once, and drawing them all onto a sphere, atmosphere included, took most of the time a frame has before the next one is due, which is where the stutter came from while panning and hovering at that view. Changed at the maker’s request.
Hovering or pinning a route no longer hides the rest of the network. The other lines used to fade to five percent while one was lit, a choice made in August so that a selected Margate train would not appear to run on through the Channel Tunnel, and it created the opposite problem: the card could not be read against the lines its route connects to. The others now step back to half their strength and stay readable, and the route in hand is still drawn on top of them at full colour, one and a half times the width, with its own glow. Changed at the maker’s request.

11 September 2026

Nineteen heritage railways join the atlas. In England, the Dartmouth Steam Railway from Paignton to Kingswear, the North Norfolk Railway, the Epping Ongar Railway, the Embsay and Bolton Abbey Steam Railway, the Dean Forest Railway, the Lakeside and Haverthwaite Railway, the Bure Valley Railway, the Chinnor and Princes Risborough Railway, the Cholsey and Wallingford Railway, the Kirklees Light Railway, the Leighton Buzzard Light Railway and Volk’s Electric Railway on the seafront at Brighton, the oldest electric railway still running anywhere. In Wales, the Llangollen Railway, now through to its new station at Corwen, and the Gwili Railway. In Scotland, the Strathspey Railway and the Caledonian Railway at Brechin. On the Isle of Man, the Manx Electric Railway and the Snaefell Mountain Railway, and in New Zealand the Glenbrook Vintage Railway south of Auckland. Found by the atlas’s own audit sweep.
Each of the nineteen is drawn from its own OpenStreetMap relation, and sixteen were redrawn through the atlas’s track checks the same day. Every card carries a short history and the stops and sights along the way, in English and German, with each date and length taken from the line’s own article or its operator. Found by the atlas’s own audit sweep.
Fifty-two route photographs that showed no train were replaced with pictures of a train on the line or at one of its stations. The worst were not railway photographs at all: a map of the line from Zvolen to Vrútky, a map of Romania for Brașov to Zărnești, a ticket stamp for the Tølløse line, a printed timetable for Rome to Ancona, an engraving of the Nordvestbanen’s opening day, a fire engine on the Mühlkreisbahn, an advertising aircraft for the ICE from Frankfurt to Stuttgart and a memorial stone at Kutno. Most of the rest were empty platforms, station buildings and viaducts with nothing on them, among them the Great Western to Bristol, Greater Anglia to Norwich, Eurostar to Amsterdam and the Jungfrau Railway. Found by the atlas’s own audit sweep.
Four routes that had no photograph now have one: the train to Bragado in Argentina, the Luanda railway to Malanje, the Plano Inclinado Gonçalves in Salvador and the Patras suburban line to Kato Achaia. And on the map’s hover card the crop was moved on three photographs that showed a grandstand, a gorge wall or fog instead of the train: the Čierny Hron forest railway, Septemvri to Dobrinishte and Lima to Huancayo. Found by the atlas’s own audit sweep.

7 September 2026

Strakonice to Tábor joins the atlas, and why it was missing is the more useful half. This line and the branch from Karlovy Vary to Loket were both named as absent by josefrichter on Hacker News on 11 August, and agreed to in the same thread, and neither has ever surfaced in the coverage sweep. Today that was measured instead of guessed. The sweep starts from OpenStreetMap route relations, and OSM carries no relation for either line: all 432 route=train relations and all 119 route=railway relations in Czechia were read, and the one match was a quarry siding, which matched only because písek is the Czech word for sand. A railway nobody has grouped into a relation is invisible to a relation-shaped sweep however many trains run on it, and the fix was to ask the timetable rather than the map. České dráhy work Strakonice to Tábor through, no change, an hour and a half, three of them in the morning alone. Reported by josefrichter on Hacker News.
The line itself is drawn from the shape of Sp 2065, the train that runs it: 81 km out of the Otava valley through Písek, over the Vltava at Červená nad Vltavou and past the stone viaducts above Milevsko to Tábor. It sits on real railway for every metre of all nine of its measured legs, a median ten metres from the track a surveyor would find and ninety per cent of it within thirty, and the planner puts our line on its own shape with nothing measurable between the two. The year is the one the line’s own Czech article gives for the whole of it, 1889. The photograph is the railway bridge at Červená, which carries this line over the Orlík reservoir. Found by the atlas’s own audit sweep.
Karlovy Vary to Loket, from the same report, was checked and not added, for a plainer reason: no train runs it through. Every itinerary the planner offers, at every hour it was asked, changes once on the way, and a card here is one journey with two ends. The branch earns a card the day a through train runs it. Reported by josefrichter on Hacker News.
Cork to Tralee was asked for twice and is not being added, which is worth saying out loud rather than leaving unanswered. It is a real service several times a day, and every metre of railway it uses is already on this map: Cork to Mallow belongs to the Dublin to Cork card and Mallow to Tralee to the Dublin to Tralee card. Each railway is drawn once here, which is the same answer the Avant from Granada to Sevilla and the Swanline already got, and the Limerick to Galway card names the train in its own text. Asked for by alan_gr on the OpenStreetMap community forum.
The Torre del Oro, the one train that used to run Cádiz to Barcelona across the whole of Spain, was left waiting in August because its working looked cut back to Córdoba. The planner was asked properly today, at every hour of a weekday, and there is no through train: every itinerary changes at least once. It stays off the map until somebody runs it end to end again, which is what the reporter suspected when he raised it. Raised by alan_gr on the OpenStreetMap community forum.
Two requests from Lemmy in July are answered here rather than quietly not done. One asked for fares on the map. This atlas holds no fares, and no operator publishes them in a shape that could be mapped without inventing the numbers, so there is nothing honest to draw. The other asked for a heat map of how late the trains are, and a third reader pointed at the GTFS realtime feeds that would carry it. Both describe a live map rather than this one: every figure here is a published route fact that will still be true tomorrow, and nothing on this site is fetched while you look at it. For the delays, catenarymaps.org already does much of it. Asked for by corsicanguppy and the16bitgamer on Lemmy.
The contact page now answers the question every channel actually asks. Across Hacker News, Lemmy and the OpenStreetMap forum the most common message this atlas receives is some version of “my line is missing”, and nothing on the site said what makes a suggestion easy to act on, or named the two rules that decide most of them: each railway is drawn once, and every card carries an opening year with a source behind it. Both are on the page now, along with the two things this map cannot show yet (a night train that splits and couples onto another train, and a corridor carrying four operators), so that nobody has to read this log to the end to find them. Found by the atlas’s own audit sweep.

5 September 2026

Thirteen railways that carry a scheduled train and had no line drawn on them joined the map, and every one of them had been waiting on a single missing fact. A sweep measures, country by country, which railways carry a train the planner will confirm and have no atlas line within a kilometre and a half of them; it found 129 such lines across 24 countries, every one of them drawable end to end, and every one of them stopped at the same place, because a card needs the year its line opened and this atlas has never shipped a year it could not source. The years here were read out of each line’s own article in its own language, and where a route is made of sections the year is the one the last section opened, because that is when the journey first became possible. Forty two of the 129 could be dated today, and the thirteen described next came first. All forty two are drawn from their own OpenStreetMap relations, all forty two sit on real railway for every metre, and the furthest any of them strays from the track a surveyor would find is nine metres. Found by the atlas’s own audit sweep.
The longest is the way into Galicia. From Monforte de Lemos a Renfe regional runs 184 km through Sarria and Lugo to A Coruña, on what was for seventy four years the only railway between Galicia and the meseta. It was built from both ends and finished in the middle: Lugo to A Coruña opened on 10 October 1875, the stretch that carries Monforte on 4 September 1883, and Alfonso XII came for the ceremony a few days before that. The second Spanish line is the climb out of the Duero basin into the Sanabria mountains, 238 km from Valladolid through Medina del Campo and Zamora to Puebla de Sanabria, on a railway that took thirty years to build, some of it by Republican prisoners at forced labour, and whose Zamora to Puebla section opened on 23 September 1952. In Portugal the Intercidades from Évora runs 89 km down to Beja over two railways of the 1860s that meet at Casa Branca, the branch up to Évora opened on 14 September 1863 and the main line into Beja on 15 February 1864. Found by the atlas’s own audit sweep.
Five more are Habsburg lines and their successors. Děčín to Liberec, 90 km, was laid in four pieces by four companies between 1869 and 1900, and the last of them climbs to Liberec through five tunnels and over the thirty metre viaduct at Novina, opened on 16 September 1900. Číčenice to Nové Údolí runs 65 km up into the Bohemian Forest: Prachatice was reached in 1893, Volary in 1899 and the Bavarian border in 1910, through trains to Passau stopped in 1951, and the rails beyond Nové Údolí are now a museum line pushed by hand. Hatvan to Somoskőújfalu, 64 km, exists because of the coal at Salgótarján, and opened to the pits on 9 May 1867 and on to the frontier in 1871. Baja to Kiskunhalas, 76 km, is two branch lines that Trianon cut in half and Hungary joined into one, the Baja end from 8 January 1885 and the Kiskunhalas end from 31 October 1903. Békéscsaba to Salonta, 49 km, is the border crossing that the Alföld to Fiume company opened on 14 September 1871 before going broke; Trianon cut it in two as well, but nobody lifted the rails and a railcar still works across. Found by the atlas’s own audit sweep.
The rest are scattered across the continent. The Donauuferbahn, 44 km along the Danube from St. Valentin, is what is left of a line that once ran to Krems: Mauthausen to Grein opened on 4 July 1898, the state line on through the Strudengau in 1909, and passenger trains now turn at St. Nikola-Struden, where the rest was given up in December 2019. The direct railway between Bulgaria’s two Black Sea cities, 212 km from Varna to Burgas, was begun in 1916, abandoned two years later and finished in the war, Karnobat to Komunari in 1942 and the last stretch down the Kamchia valley in 1943. Skövde to Nässjö, 142 km, runs over the Jönköpingsbanan, which was the Southern main line’s own way north until a shorter route through Katrineholm took its traffic in 1874; the whole of it opened on 1 December 1864. Sierpc to Toruń, 80 km, was one of the last railways the Second Republic built, laid between 1935 and 1937, and its bridge over the Drwęca at Lubicz was the first in Poland to carry welded girders. Fossano to San Giuseppe di Cairo, 63 km, is the double track shortcut that replaced the 1874 road to Savona by way of Bra, and it opened in 1933. Found by the atlas’s own audit sweep.
Every one of the nineteen carries a photograph, and each was found by searching for the stations the line actually calls at rather than for the line’s name, then opened and looked at. A Bzmot railcar stands beside an M41 on the Hatvan line itself, in the one photograph on Commons named after that railway. A Talgo comes into Puebla de Sanabria out of the fog, a CP railcar leaves Beja in the evening, a MÁV railcar waits at Salonta on the Romanian side of the border, a wooden shelter stands alone in the pines at Stožec, and a brick viaduct carries the Fossano line over the Stura with the Alps behind it. Every crop was measured against the band the hover card really shows and then rendered as a contact sheet and looked at, which is how one of them was caught: the measurement wanted to move the Sierpc crop down, and moving it cut the arms off the two semaphore signals that are the whole point of the picture. That one is back in the middle of the frame. Found by the atlas’s own audit sweep.
Six more went on the same afternoon, dated the same way. The Almtalbahn climbs 42 km from Wels up the Alm valley to Grünau, built by the city of Wels so that the valley would not be served from Lambach instead, and opened in 1901. The Neusiedler Seebahn crosses the border at Pamhagen on its 49 km from Neusiedl am See to Fertőszentmiklós: it opened on 19 December 1897 with all 109 km of it inside Hungary, and Trianon moved the northern part into Austria in 1921. Biella to Novara, 51 km, was opened by Mussolini on 18 May 1939 and then carried no train for fourteen months because the rolling stock was not ready; it was promised electric traction in 1925 and never got it, and the state took the concession back in 1961. Casale Monferrato to Chivasso, 48 km, is a third class line under the Baccarini law of 1879, and its stretch from Castelrosso to Casale Popolo opened on 30 April 1887. The Markarydsbanan, 37 km, exists because the Southern main line nearly came that way in the 1850s and went east instead; two small companies built it anyway, reaching Vittsjö from Hässleholm in 1890 and Markaryd in 1892. And Balassagyarmat to Vác, 68 km, is two railways that meet at Drégelypalánk, the Ipoly valley line of 15 August 1891 and the line south through the Börzsöny hills that carried its first train on 11 July 1909. Nineteen lines in all, 1,537 km, and the atlas has gone from 1,427 routes to 1,446. Found by the atlas’s own audit sweep.
One year on an existing card was wrong, and dating nineteen new ones is how it was noticed. The Margecany to Telgárt card said its line opened in 1884. That is the year of the first seven kilometres out of Margecany to Gelnica, built as the standard gauge stub of a narrow gauge railway to the Smolník ironworks, and nothing beyond Gelnica existed for another half century: the state line from Červená Skala was begun on 31 May 1931, opened in stages from 1 October 1933, and was put into service end to end on 26 July 1936, which both the Slovak and the Czech articles on the line record. By the rule this atlas dates every card with, the year a journey first became possible, the card now says 1936. The 1884 stub is real, it is simply not the railway a train from Margecany to Telgárt runs on. Found by the atlas’s own audit sweep.
The afternoon went back to the same queue and dated twenty three more, so the day ends at forty two new lines, 3,563 km, and an atlas of 1,469 routes. Six of the twenty three are Austrian, German and Swiss: the Hausruckbahn from Schärding to Attnang, 66 km, the northern half of the Salzkammergutbahn opened on 23 October 1877; the Innkreisbahn from Simbach to Neumarkt-Kallham, 60 km, whose bridge over the Inn opened on 1 June 1871 and which carried the Orient Express from 1883 to 1897; the Linzer Lokalbahn from Peuerbach to Eferding, 25 km, electric from the day it opened and finished on 16 December 1912; the first 22 km of the Franz-Josefs-Bahn out of Vienna, opened in 1870; the AKN line from Ulzburg Süd to Elmshorn, 27 km, completed on 9 June 1907; and Biel to Bern, 34 km, which the Bernese state railway finished over Lyss in 1864. Five are Nordic: Holbæk to Nykøbing on the Odsherredsbanen, 49 km, opened on 18 May 1899; Køge to Faxe Ladeplads on the Østbanen, 30 km, in service since 1 July 1879; Hillerød to Gilleleje on the Gribskovbanen, 26 km, which reached the sea on 14 May 1896; Parikkala to Savonlinna, 32 km, finished on 1 February 1908 from a junction that is now in Russia; and the Tjustbanan, 115 km from Linköping to Västervik, joined up at Bjärka-Säby in 1906 and widened to standard gauge in 1964. Two are Polish: Laskowice to Czersk, 55 km, opened on 2 October 1906, and Czersk to Tuchola, 53 km, a piece of the coal trunk line to Gdynia opened on 9 November 1930. Found by the atlas’s own audit sweep.
The other ten run the length of the continent. Catania to Caltagirone, 88 km, whose whole climb opened on 31 October 1892, drawn only as far as the trains run, because the 1979 extension to Gela has been shut since a viaduct collapsed. The Ponitran from Prievidza to Bratislava, 156 km, made a through journey by the Leopoldov link of 1898. Zaragoza to Barcelona by Caspe, 346 km, the MZA’s second way to the sea, finished in 1894 and opened to traffic in 1895. České Budějovice to Brno, 233 km, on three railways whose last piece opened in 1887, with the planner’s own 4 h 12 for train R 653 as its journey time. Târgu Jiu to Craiova, 106 km, dated 1888 from CFR Călători’s own history of the line, which neither Wikipedia had. Maubeuge to Charleroi, 40 km, where the Nord company’s line met the Belgian one at Erquelinnes on 11 August 1855. Shoeburyness to Fenchurch Street, 63 km, on the Tilbury company’s direct line of 1886. Marylebone to Aylesbury Vale Parkway, 64 km, to a station opened on 14 December 2008 beyond the old end of the line, which is the year that journey became possible even though the rails under it are Victorian. And two the morning had written off, below. Found by the atlas’s own audit sweep.
Two of the three the morning called impossible turned out not to be. Medgidia to Tulcea, 143 km, has no date on any Wikipedia and never had one: it was found in the Romanian press, which marked 85 years since the first train reached Tulcea, and the line’s own timeline runs from a start in 1911 through Babadag on 1 October 1925 to the last section on 25 March 1938, when Tulcea stopped being the only county town in Romania without a railway. Palermo to Trapani, 193 km, was the best row on the list and the one the relation could not carry, because the OpenStreetMap relation for train 388 begins at Piraineto, 21 km short of Palermo Centrale, the suburban stretch belonging to the airport line’s relation instead. The card was drawn from the two relations chained together, and the planner puts the drawing and the train on the same rails to the metre for all 193 km: the long way round the west of Sicily by Castelvetrano, Mazara and Marsala, finished on 5 June 1881 and the only way since the short line by Milo closed in 2013. Eighty of the 129 remain, sixteen of them in Serbia, whose lines have no article that dates them. Found by the atlas’s own audit sweep.
Every one of the twenty three has a photograph, found the same way as this morning’s, by the stations the line calls at and then looked at: an ALn 668 on the beach at Balestrate with the Gulf of Castellammare behind it, a 5047 railcar at Schärding, the Linzer Lokalbahn’s own train with Peuerbach on its blind, a Cityjet at Nußdorf under the vineyards, an SA103 at Rytel with the station name over it, a Class 357 at Shoeburyness, a MUTZ at Bern. Three cards needed a second search with the train’s name instead of the town’s, which is how Vienna, Ulzburg and Bern got a train instead of a bank building, a road and an aerial view. Six crops were moved off centre by the measurement and then looked at on a sheet, and one of them was moved back: the crop the numbers wanted for Tulcea cut the roof off the one station in Romania that is shaped like a boat. The German edition caught up too. Every hook written since 4 September now has its German, 58 strings translated in this session and merged through the same checks the July translation went through, so no route page in the German tree falls back to English any more. Found by the atlas’s own audit sweep.

4 September 2026

Two railways that carry a train and had no line drawn on them joined the map. From Pitești a CFR regional runs 142 km west to Craiova, on rails opened on 9 May 1878 as the Pitești to Craiova to Vârciorova section, six years after the line from Bucharest had reached Pitești and still an unelectrified single track. From Varna a BDŽ regional runs 107 km inland to Shumen through Provadia and Kaspichan: everything east of Kaspichan is the first railway built in what is now Bulgaria, the Ruse to Varna line of 1866, and the last 24 km into Shumen opened on 20 June 1895. Both are drawn from their own OpenStreetMap relations, both are on real railway for every metre of their length, and both sit on the shape of the train the planner runs along them with nothing measurable between the drawing and the service. Neither was found by looking at a map: a sweep measures, country by country, which railways carry a scheduled train and have no atlas line within a kilometre and a half of them, and these two came off the top of that list for Romania and Bulgaria. Found by the atlas’s own audit sweep.
A third line joined them, and it fills the emptiest corner of the map in this part of Europe. Eastern Serbia had one atlas line anywhere near it, the Belgrade to Niš trunk, and nothing at all in the Timok valley. The regional train from Zaječar now runs on the map the way it runs in the timetable: 111 km to Niš in two hours and forty eight minutes, through the Svrljig gorge, where the railway passes through nineteen tunnels. The line was finished in 1922, late for a Serbian main line and for a reason, the work having been interrupted by the First World War and finished afterwards partly by Russian emigre labour. Its own drawing comes from the OpenStreetMap relation for train 2741, it is on real railway for every metre, and the planner puts our line and the train on the same rails with nothing measurable between them. Found by the atlas’s own audit sweep.
And a fourth, which is a border crossing rather than a branch line. The R Zakarpatia leaves Chop in the far west of Ukraine, crosses into Slovakia at Cierna nad Tisou and runs on to Kosice, 104 km in three hours and nineteen minutes. It is the kind of line this atlas exists to show: the two countries use different track gauges, so Chop is where a train changes between them, and a reader looking at the map could not otherwise tell that the crossing carries a scheduled passenger service at all. The Slovak half of it dates itself precisely, because its own line article lists each section as it opened, and the last of them, Kosice to Michalany, opened on 22 October 1873. Found by the atlas’s own audit sweep.
A fifth, and it is the loneliest line added in a long time. The regional train up the Ibar valley from Kraljevo to Rudnica runs 77 km in two hours, past the spa towns of Mataruska Banja and Josanicka Banja, and there was no atlas line within a hundred kilometres of it in any direction. The rails were completed in 1931 to carry ore north out of the Ibar mines, which is the fact that dates the card: Serbian Wikipedia has no article for this railway, so the year comes from the history of the mines it was built to serve, where the line is named in full. Every one of its fifteen stations sits within seven metres of the line as drawn, which is the closest fit of any card added today. Found by the atlas’s own audit sweep.
Two British main lines were missing, and the larger of them is the second railway between London and Birmingham. The atlas drew the West Coast route from Euston and nothing else, so the Chiltern Main Line out of Marylebone, 180 km through High Wycombe, Bicester, Banbury and Leamington Spa to Birmingham Snow Hill, had 164 km of track no line on the map came near. It is a different railway from the one beside it rather than a second drawing of it, and it dates itself: the Great Western built it in three phases and finished it in 1910 with the Bicester cut-off, which shortened the run by thirty kilometres and let it match its rival on two hours. The other is the West Coastway, 73 km along the Sussex and Hampshire shore from Brighton to Portsmouth, built by five companies between 1840 and 1889 and ending on piles out over the water, at a harbour station opened on 2 October 1876. Found by the atlas’s own audit sweep.
Photographs. Eleven routes that had never had one now do, each found by searching for the stations the line actually calls at and then opening the file to look: a railcar at Knin, another at Ormoz, one at Bleiburg, a train at Beli Manastir, Trenord units on the Bergamo lines, a locomotive at Warszawa Zachodnia, a Vouga railcar working to Espinho, a train on the Beira Alta viaduct at Rocamondo, a railcar at Entroncamento and a GySEV Ventus. Four more had a photograph too small to be a hero, under 600 pixels wide, and the worst of them sat on one of the busiest pages here: the Rome to Naples card showed an empty stretch of track with no train on it at all. It now shows a Frecciarossa at Naples. Oslo to Gothenburg has the train that actually runs it, Monte Generoso its rack train on the mountainside, and the Niesenbahn its funicular car on the bridge over the Kander. Every crop was measured against the band the hover card really shows and then looked at, which is how the Chop to Kosice card was caught cropping to a stretch of platform concrete with the train pushed out of frame. Found by the atlas’s own audit sweep.
Then the same question was asked of the whole atlas, and the answer was that the crops are fine and the photographs repeat. Of 1,347 cards whose crop could be compared against a fresh measurement, ten differed enough to look at, and on seven of those the measurement was the thing that was wrong: moving the crop would have pushed the train out of the frame. Those were left alone. What the sweep did find is that 23 photographs were doing duty on 53 different cards, which the atlas has a rule against: one picture of an ICE was the hero of four German routes and one Frecciarossa of four Italian ones. Every group of three and four is broken up, and each card now carries a picture tied to its own line, an ICE on the Hohenzollern bridge at Cologne, a railjet at Budapest Keleti, another at Poertschach above the Woerthersee, a Frecciarossa under the roof at Venezia Santa Lucia. Three more cards were illustrated by photographs with no train in them at all, and two of those are fixed: Sierpc had two semaphore signals and now has the double decker that works the line, and Sarajevo to Banja Luka had an empty building and now has a train crossing the valley. Sixteen cards still share eight pictures between them, because nothing better exists on Commons for the second of each pair, and one of those pairs is deliberate: the Brussels to Luxembourg card was given a photograph of its own, which turned out to show a station entrance and no train, so it went back to sharing a locomotive with its neighbour. A shared photograph that is right beats a private one that is wrong. Found by the atlas’s own audit sweep.

1 September 2026

Seven route pages said their line had been checked against the running timetable and named a bus as the service they were checked against: BusC1880 on the Athens to Kalambaka line, Bus SEV on the Murtalbahn, Bus EV on the Blonay to Chamby and Paar Valley lines, and three more. EV and SEV are rail replacement buses, and a bus follows a road. The checks were made before the verification tool learned to recognise a bus by its name, and nothing had asked them to be earned again. All seven were asked again today. Koblenz to Trier sits on the shape of RB 81 to the metre and keeps its claim with the right train named; the other six get no train from the planner at all, one of them because the Athens to Kalambaka through service does not exist in its timetable, and their pages now say less rather than something wrong. The build refuses to carry a bus forward from here on. Found by the atlas’s own audit sweep.
The sentence that names the checked service printed the planner’s own label as it came: 355 English pages read “R 63 (Zug-Nr. 63)”, eight ended in a trip number after an arrow, two carried a Hebrew label with “<->” inside it, and 31 British lines said their drawing sat “from the shape of ?”, because the planner names no service for them and the record stored the question mark. Each shape now has a sentence: the train number stands alone, a destination reads as a destination, and a line the planner cannot name is “the scheduled train on this line”. Ten pages also listed an OpenStreetMap relation that contributed 0 km to the drawing beside the ones it is made of; a relation that adds nothing is no longer named. Found by the atlas’s own audit sweep.
The official site link on the Nariz del Diablo page led to a poker site. Tren Ecuador’s old domain lapsed and was bought by a gambling operator, so a reader following “the railway’s own site” landed on idnpoker.app; the link is gone until the railway has a site of its own again. The Taieri Gorge page linked a parked domain instead of Dunedin Railways, and the Laos–China Railway page linked a British property company that happens to share the initials LCR. 77 operator links still used plain http; 66 of them now go straight to the https address their operators serve, and the Bluebell Railway, the Museumstoomtram and the Salzburg Festungsbahn links follow the domains those railways moved to. Found by the atlas’s own audit sweep.
Under the photographs, 27 route pages credited people with the furniture of a Commons user page stuck to their names: “TyewongX ( talk · contribs )”, “mattbuck ( category )”, “Rastrojo ( D • ES )”. The credit cleaner now knows that family. On the Sofia to Plovdiv page one gallery photograph had gone blank because Commons renamed the file; it points at the new name. Two cards said a journey took “1 h 0 min”. Found by the atlas’s own audit sweep.
Two readers replying to a map of Europe on X said the same kind of thing about their own corner of the world, and both were right. The first wrote that Algeria's network connects to Tunisia and that the line to Morocco is simply out of service. The Tunisian link was missing here. Trains have run between Tunis and Annaba since 1884, when the branch from Souk Ahras reached the frontier and met the Tunisian network at Ghardimaou, and they stopped in 1995 because of the security situation in Algeria; a revival in 2003 did not last. The first train in almost thirty years left Tunis on 11 August 2024. It is drawn now, 357 km in seven hours, three days a week in each direction, calling at Béja, Jendouba and Ghardimaou on the Tunisian side and Souk Ahras on the Algerian one. The line comes off the worldwide OpenStreetMap rail graph and passes within half a kilometre of all four of those stations, on real railway for every one of its thirty seven measured legs. The Moroccan border is a different matter and the reader is right about that too: the rails are still there and no passenger train has crossed them since the frontier closed in 1994, so there was nothing to add. @yacineen62 on X.
The second reader, writing in Chinese under the same map, said a quick look showed it was missing things like China's direct passenger trains to Vietnam. It was. Through service between the two countries began on 2 August 1955, a year after the rails at Pingxiang reached the border, and it has been broken twice since: once by the collapse in relations, from December 1978 until February 1996, and once by the pandemic, from 6 February 2020 until 25 May 2025. It runs again every night. The train leaves Nanning at 18:05 as T8701, becomes MR2 at the frontier and reaches Gia Lâm in Hanoi at 06:30, which is 396 km in twelve hours and twenty five minutes, an average of 32 km/h, because everybody gets off with their luggage for immigration at Pingxiang and again at Đồng Đăng. Drawn on the OpenStreetMap rail graph through Chongzuo, Pingxiang, Đồng Đăng and Bắc Giang, the four intermediate stops the two operators list, each of them within two hundred metres of the line as it is drawn. @TaiQi23095 on X.
The maker looked at the map and said lines around stations are often drawn in a spiral, and the measurement that followed agreed, 104 times over. At Landivisiau the Paris to Brest line ran west, doubled back east, went west, doubled back again and finally carried on: two tracks about ninety metres apart through the station, and the drawing hopping between them five times, 1.9 kilometres of line across five hundred metres of railway. The cause is the map matcher that snaps relation-less lines onto the railway: it chooses between candidate tracks every few hundred metres, and in a station throat where several run side by side it sometimes picks a different one each time. Its own source code described the effect and only ever guarded against it making a route too long, which a two kilometre weave on a six hundred kilometre route never does; every other check was blind to it, and the one that measures whether a line is on its railway was worse than blind, because every point of a weave IS on real track and it kept answering one hundred per cent. A new instrument now finds any stretch that leaves a point and returns to within sixty metres of it, and removes it only when the stretch never strays more than 150 metres from the line that remains and no station the route calls at depends on it. Forty three routes gave up 168 kilometres of line that went nowhere. Every one was then re-measured against the rail graph, and that re-measurement is what kept this honest: it caught the instrument cutting into the Sargan Eight, the Alishan Forest Railway and the Darjeeling Himalayan, whose loops and reverses hug themselves tighter than any threshold can tell from a weave, and three more routes that measured worse afterwards. All six were put back exactly as they were, and the instrument now refuses them by name. Found by the atlas’s own audit sweep.
The S6 from Lienz to Villach had no card. Its railway did: the original Drautalbahn, opened on 20 November 1871 to join the Südbahn to the Brenner, runs the whole way down the Drau, and the Vienna to Lienz railjet already drew all but about three kilometres of it. What was missing was the service, and it is the one the valley actually uses: twenty two stops, an hour and thirty eight minutes, once an hour, and an S-Bahn number only since the Carinthian timetable change of 14 December 2025, when the old S1 was cut back and this leg was renumbered S6. It is drawn now, 104 km from Lienz past Oberdrauburg, Greifenburg and Spittal to Villach, from OpenStreetMap's own relations and checked against the operator's planner, where it sits on the shape of train 4611 with nothing measurable between the two. @MauMau141674 on X.
Checking that report turned up a railway that really was absent. The Gailtalbahn leaves the Tarvisio line at Arnoldstein and runs west under the Carnic Alps to Hermagor, and 28 km of it were on no card at all: the nearest line the atlas drew passed eleven kilometres from Hermagor station. It opened on 11 August 1894, when 310 people rode the first train over the new bridge across the Gail, and it carries the hourly S4 out of Villach in an hour flat. Drawn from its own OpenStreetMap relation, 48 km, on the shape of train 4205. Neither of these had ever surfaced in the coverage sweep, and the reason is the same for both: the sweep scores a commuter pattern lowest of all the service types and prints only the strongest rows, so Austria's 170 services without a card have never shown an S-Bahn line among the ones it lists. Found by the atlas’s own audit sweep.

31 August 2026

The Málaga to Granada card was filed as a classic railway and dated to 2019, and both were wrong. It is an Avant, and every Avant in this atlas runs on high speed line by definition (the brand exists for medium distance trains on high speed track), so the other three were already filed that way and this one was the odd card out, at 250 km/h like the rest of its fleet. The year is the more interesting mistake. The high speed line from Antequera to Granada did open on 26 June 2019, and the card had taken the LINE’s year for the SERVICE’s: until the chord at Gobantes joined it to the Córdoba to Málaga line, Granada to Málaga meant changing at Antequera, and the direct run only began in April 2022. Dating a service to the year its rails were finished is the same error as taking the opening year off the railway next door, and this atlas has a rule against that one. Reported by alan_gr on the OpenStreetMap community forum.
Two of the three photographs under that same card were of somewhere else entirely: a Renfe unit at Zaragoza Delicias and another at Valladolid Campo Grande, five hundred kilometres and more from any track this line uses. The cause was that the card had no article of its own, so the gallery had been filled from the article about the Avant BRAND, which is a Spain wide subject and illustrated with Spain wide pictures. It has the line’s own article now, the Antequera to Granada high speed line, and its gallery is Andalusian: the Transversal corridor, Antequera Santa Ana, and the line under construction. The deeper reason nobody caught it is fixed too, and it was a hole rather than a judgement. The gate that refuses a photograph taken far from the line can only weigh a photograph that carries coordinates, and five in six do not, so for most of the gallery it had never been able to say anything at all. Commons usually knows anyway, in the categories rather than in the camera: a picture filed under a station’s own category is a picture of that station. The harvester reads those now and judges a photograph by where its subject is, on the same terms as one that carries coordinates, while leaving portraits of trains alone because a photograph of a train claims no place. Fifty three pictures failed that question, and some of them had been sitting there a long time: the Bangkok to Aranyaprathet line was illustrated by Bondi Junction and Sydenham, the Tanzanian Central Line by Loughton and Tottenham Court Road, the Australian Blue Mountains by the Blue Ridge Mountains of Virginia, the road to the Victoria Falls by a Mount Victoria in New South Wales, and Milan to Ventimiglia by four photographs of Great Western coaches at Penzance. Two galleries had nothing honest left in them afterwards and now show nothing, which is the right answer. It is not perfect, and the way it fails is worth knowing: an old photograph is often filed under the collection that holds it rather than the place it shows, so the only spot Commons can offer for a 1900 view of the Brocken railway and an early one of the Semmering is a single point in Zurich, and both were dropped for being Swiss. Reported by alan_gr on the OpenStreetMap community forum.
The same complaint had a second half, and it was the order rather than the contents. A gallery keeps the first four photographs its article offers, in the article’s own order, and the article a card links is very often about a TRAIN rather than a line. A brand article leads with portraits of the train taken wherever the photographer was standing, so article order was handing the slots to pictures of the right train somewhere else, ahead of photographs of this railway. Nothing was being refused; the honest ones were simply queued behind. Now that most photographs can be placed, the ones known to be on the line are taken first, the ones nothing can place come next, and the ones known to be elsewhere come last. No photograph is dropped by this and no gallery lost, but 490 pictures of the line itself moved up into view: the Tashkent to Samarkand card had been illustrated by Bukhara station and now shows Tashkent’s own, Madrid to Salamanca had Zaragoza and Pamplona and now has the Martín Gil viaduct and the tunnel at Colmenar Viejo, and the Moroccan cards stopped sharing one portrait taken at Marrakech. Across the atlas the share of gallery photographs that can be shown to be on the line they illustrate goes from 47 per cent to 57, and the share known to be somewhere else falls by nearly half. Reported by alan_gr on the OpenStreetMap community forum.
Barcelona to Málaga was drawn straight through Madrid Puerta de Atocha while its own description said it skirts Madrid, and the description was the true half. Two services run that city pair and they are not the same railway: one calls at Atocha, and the Renfe AVE this card names does not stop between Zaragoza and Ciudad Real at all. The line had been built by splicing the Madrid to Barcelona relation to the Madrid to Málaga one, and a splice can only join two railways where they meet, which is inside the station the train avoids. It is now drawn from the shape of AVE 3992 itself, and it passes nine kilometres clear of Atocha, which is what skirting Madrid looks like. The card also loses eighteen kilometres of detour it never ran, and its journey time moves from 5 h 52 to 6 h 57, which is what the train takes; the old figure was the one thing here that no reader could have checked against the map, and it was out by an hour. Reported by alan_gr on the OpenStreetMap community forum.

30 August 2026

Two Iberian-gauge lines on the Spanish Mediterranean coast were missing, and only their high-speed neighbours were drawn. Valencia to Alicante is 179 km of 1,668 mm single track through Xàtiva and Villena, opened in 1858 and still worked end to end by the Euromed; Alicante to Murcia is 77 km through the Elche palm groves and the Segura valley, opened in 1884. Both were in OpenStreetMap the whole time, tagged as commuter and regional services, which is exactly the class the atlas’s European coverage sweep hides, so no sweep had ever surfaced them. Both cards are now drawn from the real track and checked against the trains that run them, the Euromed EM 1071 and the Alvia AVL 5668, at nothing worse than a kilometre from either shape. Kaukor (@markoitali47416).
87 route photographs were being cropped through the middle by guesswork. The hover card shows a narrow band cut out of a taller photograph, and where the band lands decides whether you see the train or the sky above it; every photograph added since the last measuring run had been falling back to a blind default. All 87 are measured now, so the band sits on the train. The 59 crops that had been set one by one were left exactly as they were. Found by the atlas’s own audit sweep.

29 August 2026

The Engadine line card said “In service since 1903” directly above its own hook saying the line opened in 1913, and 1913 is the right one. Its article is unambiguous once you read past the first date it mentions: the Rhaetian Railway commissioned the line on 27 July 1903, construction began in the spring of 1909, and “the line was officially opened on 28/29 June 1913”. 1903 is the year the neighbouring Albula line opened, and taking a year off the line next door is the one thing this atlas has a rule against. Wikidata’s own opening-date property says 1903 as well, which is probably how it got in. The Prättigau line moves from 1889 to 1890 for the same reason: its own article says it has connected Landquart with Davos since 1890. Found by the atlas’s own audit sweep.
A card drawing the Emmental line between Bern and Lucerne was labelled as the faster service that takes the other route entirely. The drawing passes Langnau by 390 metres and Wolhusen by 100, and misses Zofingen by 27 kilometres; the card’s own timetable record already named the BLS train, so it had been contradicting its own evidence. Asked again, the planner offers both: IR 2505 runs Bern to Lucerne in 1 h 01 via Zofingen and Sursee, and RE 7 in 1 h 27 via Konolfingen, Langnau, Trubschachen, Escholzmatt, Schüpfheim, Wolhusen and Malters. The card is the second of those and now says so, which drops its average speed from 95 km/h to 66, which is what that railway actually does. Found by the atlas’s own audit sweep.
A Finnish card said it ran from Toijala to Tampere and drew a line from Tampere to Turku: its first point sits 100 metres from Tampere and its last 500 from Turku, while Toijala is 37 kilometres from either end. Turku, half the card’s own name, appeared in neither endpoint. The endpoints now match the line, the distance and the train that was checked against it. Found by the atlas’s own audit sweep.
Twenty Wikipedia links pointed at the wrong article, and one of them had been found by a reader on the Hacker News front page eighteen days earlier and was still live: the Tanzanian Central Line linked the Central line of the London Underground. Ten more named another country’s railway, two named a rail accident rather than a railway, and five landed on a disambiguation page. Every replacement was checked for existence, for redirects and for being the railway rather than a station of the same name. Five cards now carry no link at all, which is the honest answer where no article describes the line: Nigeria’s first standard-gauge railway had been linked to the 2022 attack on it, Egypt’s Talgo service to a Spanish trainset, and a 1966 Iranian service to a high-speed line that has never opened. Found by the atlas’s own audit sweep.
A German route page was headed “Zugangriff zwischen Abuja und Kaduna” (train attack between Abuja and Kaduna). German names are read from the article each card links, so a wrong link becomes a wrong heading, and nobody had looked because the English page was fine. Four more German pages were headed by a station rather than a line. Separately, 32 German pages shared a heading with another page, three of them simply “Railjet”, because the rule that adds the endpoints to an ambiguous name was deciding on the English name and applying the verdict to German. All of those are unique now. Found by the atlas’s own audit sweep.
133 route pages were showing the same photograph twice, once as the headline image and once in the gallery below it. The guard that prevents that compares file names, and one side of the comparison was decoded while the other was not, so it could only ever fail on a file whose name contains an escaped character. Of the 133, every single one had such a character in its name and none did not. 67 photographs were also credited to nobody; their photographers are named now, which most of these licences require. Found by the atlas’s own audit sweep.

28 August 2026

Belgium had nineteen routes touching it and not one ran the line south from Brussels through Nivelles to Charleroi, which is the kind of gap that is obvious to anybody who lives there and invisible to a map that has never been measured against its own country. So every Belgian Intercity relation in OpenStreetMap was laid over the atlas and scored for the kilometres of railway no card drew. Six of them now have cards: Antwerp to Charleroi down the 1843 line past Nivelles, De Panne to Ghent across West Flanders, Antwerp to Hamont over the Campine heath, Antwerp to Ghent through the Waasland, Brussels Airport to Ostend the long way round through Oudenaarde and Roeselare, and Liège to Namur along the Meuse. Each is drawn from its own relation and checked against the shape of the train that runs it: five sit on that shape exactly and the sixth within a hundred metres, and no station on any of them stands more than three hundred metres off the line. Found by the atlas’s own audit sweep.
Three more were refused, and the reason is worth more than the six that shipped. Belgium runs a dozen Intercity series over each other’s rails, so a candidate scored against the atlas alone can look like sixty kilometres of new railway and turn out to be its neighbour’s sixty kilometres. Scored against each other in turn, IC-02 from Antwerp to Ostend is the Antwerp to Ghent and Brussels Airport to Ostend cards laid end to end, the local service from Liège to Namur is the same track as the Meuse card already chosen, and IC-11 from Binche to Turnhout adds fifteen kilometres to lines already drawn. A fourth, Kortrijk to Welkenraedt, has a through train that takes a different railway between the same two ends, and was refused on that. None of the four is a railway this map was missing. Found by the atlas’s own audit sweep.
Two dates were corrected by the checks rather than by reading. Wikidata gives 1844 for line 59, the Antwerp to Ghent railway, and the line’s own article shows that 1844 is when it reached Sint-Niklaas: Ghent came on 9 August 1847, which is the year the card carries. Line 73 to the coast is dated 1855 the same way, and its four sections opened in 1850, 1855, 1858 and finally on 5 February 1870 at De Panne, which is when the journey the card describes first became possible. The lesson is the same in both: an opening date on a staged line is often the date of its first section, not of the through route. Found by the atlas’s own audit sweep.
The service that OpenStreetMap maps from De Panne to Leuven runs the whole way as one train only on Saturdays; on a weekday the planner needs a change at Ghent. A card quotes a journey time, so this one spans De Panne to Ghent instead, which is one train every hour of every day, and which is also where all of its undrawn railway was. Found by the atlas’s own audit sweep.
Poland was the largest trunk gap left in Europe on the atlas’s own measurement, thirty eight cards over a network where more than half the railway had nothing drawn on it. Ten lines now do. Warsaw to Terespol is the spine of them: 212 kilometres east to the Bug on the railway of 1866, the third built in Congress Poland, and one train still runs the whole of it in 2 h 11. With it come Piła to Szczecin across Pomerania, Szczecin to Świnoujście over Wolin island, Olsztyn to Elbląg, Inowrocław to Olsztyn, Sierpc to Nowy Dwór, Lublin to Zamość, Katowice to Zwardoń in the Beskids, Dęblin to Łuków and Kutno to Płock. Nine of the ten sit on their own train’s track at nought or three tenths of a kilometre; the tenth is Warsaw to Terespol, whose train runs 456 kilometres in all and whose reference shape is too coarse to judge a 212 kilometre card, so it carries no timetable claim. It was checked another way instead: the drawn line passes within a kilometre and a half of every station that train calls at, Międzyrzec Podlaski and Biała Podlaska included, which are the towns that belong to this railway and to no other. Found by the atlas’s own audit sweep.
Five candidates were measured and refused, and in Poland the refusals took a tool that did not exist a week ago. A dozen services here run over each other, and each is mapped twice, once per direction, so a candidate scored against the atlas alone can show a hundred kilometres of new railway and be its neighbour’s hundred. Scored against each other in turn, the Nowy Dwór to Sierpc service came out at one new kilometre once its own train’s shorter working was drawn, because they are the same train. Three more were refused for having no through train at all on any date the planner answers: Stalowa Wola to Lublin, Sierpc to Tłuszcz and Tłuszcz to Ostrołęka. The Warsaw to Poznań service by way of Łódź and Kalisz is a real second railway between those two cities and is not drawn, because the planner answers the pair with the main line through Konin and there is no way yet to make it answer the other one. Found by the atlas’s own audit sweep.
Four opening dates were corrected by checking rather than by reading, all of them the same mistake in different clothes: Wikidata gives a railway the year its FIRST section opened. Line 353 is dated 1871, which is when Toruń reached Jabłonowo, while Inowrocław to Toruń and Ostróda to Olsztyn both opened in 1873. Line 69 is dated 1887, which is Bełżec to Hrebenne, while the stretch through Zamość waited until 1916. Line 139 is dated 1852 and was not finished into Katowice until 2 November 1912. Line 26 is dated 1876 and the last piece into Dęblin was laid in 1935. Every one of those years came off the line article’s own list of section dates, which is the source that knows the difference. Found by the atlas’s own audit sweep.
Two of these ten lines publish no opening date on the Polish side at all, and both were dated from the German article about the same railway, which still remembers being built: the line from Piła to Ulikowo went up in three stages and was finished in 1895, and the line to Świnoujście reached the Świna in 1900. The Pomeranian line is the better story of the two. Poland withdrew every passenger train from it on 23 June 2000 and the track between Wałcz and Kalisz Pomorski became unusable; it came back in pieces from 2006, stopped again in 2012, and has run through trains ever since that September. Five pairs a day. Found by the atlas’s own audit sweep.
Saxony was the biggest hole left in Germany and four lines now fill most of it. The Regional-Express from Dresden to Hof is the largest single piece of undrawn railway this atlas has added in one card: 222 kilometres up the Tharandt ramp and along the foot of the Ore Mountains through Freiberg, Chemnitz, Zwickau and the Vogtland, none of which any line here drew. It became one journey in 1869, when the awkward climb from Tharandt to Freiberg was finally carried on to Chemnitz; the pieces either side of it are older, back to the branch of 1845 that first reached Zwickau. With it come the hourly express from Chemnitz to Leipzig on the line of 1872, the run north from Chemnitz down the Zschopau to Riesa and Elsterwerda, which joins a railway of 1852 to one of 1875, and the Vogtlandbahn out of the Zwickau coalfield over the Czech border to Kraslice, opened by a coal railway company in November 1875. Found by the atlas’s own audit sweep.
A fifth was drawn, measured, verified and then held back, which is the more useful half of the day. The RB 80 from Chemnitz to Cranzahl sits on its own track perfectly well and its railway opened in February 1866, but asked for a train next week the planner offers a bus for the whole way and a train only as far as Flöha. A through train does appear seven weeks out, so this is an engineering closure rather than a closed line, and the card is written and waiting rather than shipped. A card quotes a journey time and a bus is not a train. Found by the atlas’s own audit sweep.
Nine more candidates across Europe were refused this week on the same three gates. Zaragoza to Lleida has a train between those cities and it takes the high-speed line, 44 kilometres from the railway the relation describes. Six need a change and so are two railways rather than one card: Madrid to Salamanca, Olomouc to Ostrava, Zwickau to Cheb, Wiener Neustadt to Znojmo, Strasbourg to Sarreguemines, and the TransPennine service from Manchester Airport to Cleethorpes. Riba-roja d’Ebre to Barcelona has both ends in the planner and no itinerary at all between them. And the Dresden to Zittau service turned out to be nought kilometres of new railway once the Liberec train that shares its track was measured first. Found by the atlas’s own audit sweep.
Three more from the same sweep, in three countries. The branch across the Swabian Alb from Gammertingen through Münsingen to Schelklingen and the Danube, opened in 1901 and worked by a railcar in the colours of the Schwäbische Alb-Bahn itself. The regional out of Madrid past El Escorial and over the Guadarrama to Ávila, on the main line towards the French border that was built between 1860 and 1864. And the line out of Oradea over the Hungarian border to Püspökladány, put into service by the Tisza Railway Company on 22 April 1858 and still the way from Transylvania onto the Great Plain. Found by the atlas’s own audit sweep.
Two Romanian candidates were drawn, verified and then set aside for want of a single fact each, which is the gate that stops more cards than any other. Medgidia to Tulcea across the Dobruja is 143 kilometres of railway this map does not draw and a real train runs it in 3 h 14, but its article is a route diagram with no date in it, and the line number article has none either. Pitești to Craiova has no article at all. A card carries the year its railway opened, and an invented one is worse than an absent card. Found by the atlas’s own audit sweep.
The last ten countries were measured and the answer is mostly that there is nothing left. Finland, Denmark, Portugal, Moldova, Lithuania, Latvia, Ireland, Estonia, Norway and Luxembourg between them produce four candidates over forty kilometres. Their tables are topped instead by trains that are barely theirs: the longest row in Lithuania is a Chelyabinsk to Kaliningrad service whose 2,382 undrawn kilometres are almost all in Russia, and the measure credits a relation to the country whose inventory produced it. Moldova’s largest row is the Chișinău to Kyiv train, which this atlas established last week no longer runs. Found by the atlas’s own audit sweep.
One card came out of those ten countries and it is a good one. The Hanko line was Finland’s first private railway, opened in 1873 and paid for by St Petersburg merchants who wanted a harbour that did not freeze; it runs forty minutes down from Karjaa through Tammisaari to the sea. It is drawn as the branch itself rather than as the through service from Helsinki, because the coastal line the Helsinki trains use for their first eighty five kilometres is already on this map and drawing it twice would add nothing. Found by the atlas’s own audit sweep.
Three more were refused in the same pass. The service from Karjaa to Hanko and the one from Helsinki to Hanko are the same railway, and measured against each other the longer one adds nothing at all. Šilutė to Kretinga in Lithuania has a train that runs it in an hour and a half and sits exactly on our drawing, and still has no card, because no article anywhere gives the year that railway opened; it was held for the same reason in August. And the line from Riga towards Minsk resolves at both ends and the planner offers no journey between them at all. Found by the atlas’s own audit sweep.
A hundred and one more route pages can now show that a real timetable was checked against the line they draw, and sixteen have stopped claiming it. The cause of the first number is a bug in this atlas’s own instrument rather than anything about the railways: the planner labels some foreign commuter trains with the same product class it uses for buses, and the checker took that literally, threw the train away as a road vehicle and then reported that no train existed. Finland found it, and once the checker learned to recognise a bus by what the planner CALLS it, a sweep of every European route without a check earned 109 of them one. Found by the atlas’s own audit sweep.
The sixteen that lost the claim lost it honestly and they are worth naming, because a claim withdrawn is more interesting than one added: Madrid to Barcelona, Madrid to Alicante, Madrid to Ourense, Madrid to Santander, Zurich to Munich, Paris to Amsterdam and London to Amsterdam by Eurostar, Paris to Milan, Lisbon to Évora, Brussels to Cologne, Bordeaux to Marseille, Oslo to Stockholm, Paris to Geneva, Paris to Montpellier, the Furka steam railway and the Appenzell line. Each of those lines was redrawn in the middle of August, which makes the check that was made before the redraw a statement about a line that no longer exists. Asked again this week, they do not pass, so the sentence comes off the page until they do. Three others in the same position asked again and passed: Zurich to Milan over the Gotthard, Hamburg to Berlin, and Berlin to Amsterdam. Found by the atlas’s own audit sweep.
Liberec to Dresden, which had been measured and drawn for a week and was waiting on one fact. The Regional-Express crosses from Bohemia into Saxony down the Neisse to Zittau and then works north through Ebersbach and Bischofswerda, and it is not one railway but a chain of Lusatian branches; the last link of that chain, Oberoderwitz to Eibau, opened on 15 October 1879, which is when the journey the card describes first became possible. Every metre of it sits on the train’s own track. Found by the atlas’s own audit sweep.
Sixty six routes had never been measured against real railway, among them everything added in the last fortnight, and they have been now. Sixty five of the sixty six sit on track, at a median of four metres from it. The two that did not are honest about why: the children’s railway in Minsk has no track recorded near it in OpenStreetMap at all, and the classic line from Milan to Bologna agrees with the router for 98 per cent of its length and parts company with it for four and a half kilometres north of the city. Every route in this atlas has now been asked the question. Found by the atlas’s own audit sweep.
Every route in Europe this atlas draws has now been put to a live planner in one pass, all nine hundred odd of them, and the result is reassuring in a way that is worth writing down: of the claims already on these pages, all but one were re-earned against the lines as they stand today. Six hundred and seventy four routes now carry a timetable check, forty seven more than this morning. The one that lost it is Berlin to Basel, which no longer matches the service the planner offers between those cities. Found by the atlas’s own audit sweep.
Twenty two routes gained a photo gallery, which is every route added today that has a Wikipedia article to read them from. Forty six photographs were refused on the way in by a check worth describing: Commons was asked where each picture was taken, and any shot from outside the route’s own countries and more than six hundred kilometres away was dropped. A photograph of a train is not evidence about a railway on the other side of Europe. Found by the atlas’s own audit sweep.
The Mittenwaldbahn was promising a journey an hour faster than it runs. Munich to Innsbruck can be done two ways, over the Brenner side through Kufstein or up through Garmisch and Seefeld, and the planner kept answering with the first one, which is not the railway this card draws. Asked instead about the train that really climbs the Mittenwaldbahn, the drawing turns out to be exact, and the timetable says two hours fifty five rather than the two hours five printed on the card. The card now says what the train does. Found by the atlas’s own audit sweep.
The AVE from Madrid to Zaragoza was drawn out of the wrong Madrid station. Every other card on that railway, Madrid to Barcelona and Madrid to Alicante, starts at Puerta de Atocha, which is where the high speed line begins; this one started at Chamartin, seven kilometres away across the city. The effect was not cosmetic. Asked to check the card, the planner could find no AVE at all leaving from the point the line began, offered six hour regional trains up the old line instead, and the comparison came back two hundred kilometres adrift. The line has been rebuilt from the Madrid to Barcelona trunk it shares, point for point, from Atocha to Zaragoza Delicias. It now measures 307 kilometres against a declared 307 and a planner shape of 307, no station on it stands more than a hundred metres off, and the journey time is corrected from 1 h 15 to 1 h 35, which is what the fastest train of the day takes. Found by the atlas’s own audit sweep.
The RegioJet night train from Prague to Kosice was drawn around the wrong side of Ostrava. Two railways run from there towards Cesky Tesin, one north through Bohumin and one direct through Havirov, and the card took the northern loop while the train takes the direct line. The fault was twenty seven kilometres long and began 353 kilometres into a seven hundred kilometre journey, which is the sort of thing that survives a glance at a map. It was found by measuring the stations the train calls at against the line, Havirov standing ten kilometres away when every other stop was on the rails. Redrawn from the service’s own published shape, the card passes within nine hundred metres of Havirov and no station on it is more than a hundred metres off. Found by the atlas’s own audit sweep.
Ten routes that a live planner disagreed with were worked through one by one, and only the two above turned out to be faults in the atlas. The rest are the planner answering a different question, and saying so is the point of writing them down. Limerick to Ballybrophy is drawn down the Nenagh branch, passing Birdhill, Nenagh, Cloughjordan and Roscrea, which is the railway the card is named for; the planner routes the same two ends the fast way round through Limerick Junction and Thurles, twenty six kilometres away. Tel Aviv to Beersheba is drawn through Lod and Kiryat Gat and the planner answered the coastal service through Ashkelon, and both are real trains. The Caledonian Sleeper is drawn up the Trent Valley while the reference runs it past Birmingham International, which no sleeper does. Madrid to Murcia sits within a kilometre of Cuenca, Albacete and Orihuela, which is the AVE alignment exactly. A card is wrong when the railway is wrong, and a disagreement is not by itself evidence of which side is mistaken. Found by the atlas’s own audit sweep.
Nine cards left Paris down the wrong railway, and they all left it down the same wrong railway. Every TGV out of the Gare de Lyon, to Lyon, Marseille, Nice, Montpellier, Barcelona, Geneva, Zurich and Milan, and the Ouigo with them, was drawn along the LGV Interconnexion Est, which is the line built so that trains from Roissy and Marne la Vallee can join the high speed route without entering Paris. Trains from the Gare de Lyon do not use it. They run out on the classic line through Villeneuve Saint Georges and join the high speed line at Combs la Ville, twenty nine kilometres out. The drawn lines passed within 1.7 km of Brie Comte Robert, which is on the bypass, and four and a half kilometres from Combs la Ville, which is where they should have been. The error was thirty two kilometres long and twelve kilometres wide at its peak, running from twelve kilometres out to forty four. Found by the atlas’s own audit sweep.
The nine were found by grouping the routes a planner disagreed with by the size and the position of the disagreement rather than by name. Five of them reported an identical thirty two kilometres off and thirty kilometres missing, starting at the same place, which is not something five unrelated faults do. Measuring every card that touches the Gare de Lyon against Brie Comte Robert and Combs la Ville then turned up four more that had not been flagged at all. One card was rebuilt from its own train’s published shape and the corrected first fifty five kilometres handed to the other eight, point for point, so the nine now share a single trunk instead of nine slightly different ones. Paris to Lyon and Paris to Geneva sit on the service exactly, Paris to Milan within a hundred metres, and every station on Paris to Marseille is within twenty seven metres of the line. Found by the atlas’s own audit sweep.
The Frecciarossa to Bari and the Frecciargento to Lecce were drawn through the wrong valley between Caserta and Benevento. Two railways cross the foothills there, the main line north through the Volturno valley past Telese and a second route south through the Valle Caudina, and both cards took the southern one. That left them eight kilometres from Caserta, where the trains call, and seventeen kilometres from the track their own train uses, for seventy four kilometres starting a hundred and eighty two kilometres out of Rome. The station list settled it: every stop on the reference lay on the line except Caserta, and a card that misses the one station a train is known to call at is not drawing that train’s railway. Redrawn from the service shape, both now pass Caserta at a hundred metres and Telese at two hundred, while Naples, which neither train goes near, is correctly twenty one kilometres away. Found by the atlas’s own audit sweep.
Three further groups of disagreement were read and deliberately left alone, which is worth writing down because the list they come from is not a defect list. Prague to Cheb and Prague to Zelezna Ruda each stray thirty kilometres on the way out of Prague and both are right: they run down the Berounka valley main line through Radotin, Cernosice and Revnice, ten kilometres from the secondary line through Rudna that the reference takes, and every station the reference calls at is within two hundred metres of the drawn line. A shape that contradicts its own stop list is not evidence. The four short mountain and forest lines in the same list, the Vale of Rheidol, the Cierny Hron forest railway, the Montserrat rack railway and the Train 1900 at Fond de Gras, have no reference at all, because the planner knows no station on any of them and answers with whatever runs along the valley below. Found by the atlas’s own audit sweep.
The AVE from Barcelona to Malaga was very nearly redrawn from the wrong train, and the thing that stopped it is worth more than the change would have been. Asked to check the card, the planner offered an Ouigo, and against the Ouigo the line looked twenty two kilometres wrong on the way out of Barcelona and seventeen kilometres too long overall. Redrawing from that shape produced a line that sat on the Ouigo exactly and measured 1,120 km against a declared 1,121, which looked like a clean repair by every number available. The build refused it. A gate that nobody had needed all day noticed that the new line no longer came within reach of two stations the card is recorded as serving, Lleida Pirineus and Camp de Tarragona, and said so in those words. Asked again with an AVE pinned, which is the train the card is actually named for and the one that calls at those stations, the ORIGINAL line came back sitting on the service with nothing measurable between them. The card was right all along; only the reference was wrong. The line is unchanged, and what remains is a seventeen kilometre stretch six hundred and eighteen kilometres in, around Madrid, where it strays about nine kilometres. That is written down rather than fixed, because the lesson of the afternoon is that a redraw is only as good as the train it is copied from. Found by the atlas’s own audit sweep.
The Frecciarossa from Milan to Bari said it was 786 km long and drew 860, and the train’s own shape also measured 860. Two independent measurements agreeing with each other and disagreeing with the card by seventy four kilometres is not something to leave alone, so the card now says 860. What that one number exposed is that nothing in the atlas had ever compared a card’s declared distance against the line the card actually draws, even though both numbers sit in the same repository and the comparison costs nothing. It does now. Of 1,256 cards carrying both, 979 agree within three per cent and sixty seven are out by a tenth or more, and those sixty seven are now written down and sorted by what the gap probably means: five are excursions quoting a round trip against a one way line, eleven are luxury tours quoting a round advertising figure, and twenty five are ordinary services whose drawn line is longer than the railway is supposed to be. The last group is where the real faults will be, and the four high speed lines in it are the place to start, because the official length of a new railway is published and exact. Found by the atlas’s own audit sweep.
Six more disagreements were read and only two needed anything. The classic line from Milan to Bologna is drawn correctly, through Lodi, Piacenza, Fidenza, Parma, Reggio Emilia and Modena with every one of those stations within about half a kilometre; the planner had answered with a Frecciarossa, which takes the high speed line built beside it, so the reference was the wrong train for a card whose whole point is the old road. Brussels to Cologne, London to Amsterdam and Dublin to Belfast each differ from their reference by four to seven kilometres over a short stretch, which is the width of a pair of railways running side by side, and none of them was changed. Found by the atlas’s own audit sweep.
The Jerusalem card started at Ben Gurion Airport rather than in Tel Aviv. It had been traced from the OpenStreetMap relation for the A1 line, and that railway really does begin at the airport; the train does not. It leaves Tel Aviv Savidor Center and runs down the Ayalon corridor through HaShalom and HaHagana before it reaches the airport, ten kilometres the drawing never had. Those ten kilometres were routed over OpenStreetMap track with the rail router the atlas uses to check every line, and the whole 57 km was then held against Israel Railways’ own published timetable, which puts every calling station of the Jerusalem to Tel Aviv service within 336 metres of it. The card now spans the service it names, and its evidence says the honest thing: 47 of its 57 km follow the relation, not all of them. Found by the atlas’s own audit sweep.
The Alvia from Madrid to San Sebastian was drawn from Atocha, and the Alvia leaves Chamartin, which is this morning’s Madrid to Zaragoza fault the other way round. It also looped west through Medina del Campo before turning back to Olmedo, thirty kilometres of line fifteen kilometres from any track its train uses. Planned from Chamartin with the train itself pinned, the planner’s shape measures 549 km with no change of train, and the card was redrawn from it: every stop the train makes is now on the line to the metre. The declared distance moves from 490 to 549 km and the journey time from four and a half hours to five hours fifty, which is what the timetable’s fastest run says. Two stations came out of the record of calls at the same time, Medina del Campo and Atocha, because the old wrong line had passed both and the train has never called at either; a wrong drawing can admit wrong calls, and the record then refuses the right drawing. Found by the atlas’s own audit sweep.
The Canfranero from Zaragoza had three faults in one line. It started nearly four kilometres from Zaragoza Delicias on the wrong side of the city, it passed four kilometres from Huesca, where the train reverses, and it stopped at Canfranc village, four kilometres short of Canfranc station. Redrawn from its own train’s shape, planned from Delicias, both ends and Huesca are now within two hundred metres, the declared distance is 224 km rather than 180, and the journey time is three hours twenty rather than three hours forty five. Found by the atlas’s own audit sweep.
The Balaton InterCity started two kilometres from any station and took eighteen kilometres to reach Kelenfold, which is seven kilometres of railway from Budapest Deli. The rail router could not draw the missing stretch, because its graph has no direct link between Deli and Kelenfold and answers twenty six kilometres for it, so the stretch was copied point for point from the Tapolca card, which already runs it. The line is 118 km against the published 115, every station is on it, and the journey time is one hour twenty, as the timetable says. Found by the atlas’s own audit sweep.
Four declared distances were wrong where the line was right, and in every case the planner and the drawing agreed against the card. Milan to Reggio Calabria goes from 1,400 to 1,213 km: the only reference is a night train that takes the coastal line past Formia and the loop through Tropea, and without those two detours it measures what the drawing does. Besancon to La Chaux de Fonds goes from 75 to 85, because 75 is the French line to Le Locle and the card runs on to La Chaux de Fonds. Nantes to Quimper goes from 230 to 254, and the Jungfraubahn from 20 to 18, because 20 is the Lauterbrunnen side of the Kleine Scheidegg and the card climbs from Grindelwald. Lisbon to Evora stays as it is: the planner’s shape puts every station the Intercidades calls at on our line, and a shape that contradicts its own stop list is not evidence. Found by the atlas’s own audit sweep.

27 August 2026

Three route pages linked the wrong Wikipedia article, and two of them had built their photo galleries from it. The Byron Bay solar train linked the Boston, Revere Beach and Lynn Railroad, a Massachusetts line closed in 1940; Amtrak’s Crescent linked the New Orleans streetcars, and its gallery was four photographs of streetcars; Amtrak’s Borealis linked the Chicago, St. Paul, Minneapolis and Omaha Railway, a company dissolved in 1957, and its gallery was 1919 steam engines. Each has an article of its own, and the galleries now show the trains the pages are about. The error surfaced while reading every link through its German-language counterpart, which is a better test of what a link is actually about than the title alone. Found by the atlas’s own audit sweep.
The Zalaegerszeg–Boba line is operated by GYSEV, not MÁV-START: the Hungarian article records that GYSEV took over the service on 1 July 2025, and the card now says so. Found by the atlas’s own audit sweep.
Around 770 gallery photographs added this week were crediting “Wikimedia Commons” rather than the photographer, because the credits table had not been refreshed after the photos were. The Creative Commons licences these pictures carry require the author to be named where the author asked for it. All 5,661 photographs the atlas shows now carry the credit Commons publishes for them; 87 name no author because Commons names none. Found by the atlas’s own audit sweep.
Route page titles were saying things twice and, on 67 pages, saying them in two orders: “Avezzano–Roccasecca railway Route Map — Roccasecca to Avezzano by Train”. A page whose name already states both ends now says so once. Eleven pages had shared the bare heading “ICE” and nine the heading “AVE”; each now carries its endpoints, in the shape the atlas already used for “Avant Málaga–Granada”. The “| World Train Map” tail was also dropped from the hub pages, which route pages had never carried, so the words a search result shows are the ones that describe the page. Found by the atlas’s own audit sweep.
The station markers, the white pills with dark rings at every terminus and the small ticks across lone line ends, are now drawn as map symbols sized in screen pixels rather than as tiny line segments sized in degrees. Nothing about them should look different. What changed is underneath: every wheel of the mouse used to rebuild all 1,900 of them and send the lot back through the map engine, and the profiler put that at the top of the list of what made zooming stutter. Now the map engine sizes and rotates them itself, and the set is only rebuilt when the grouping changes, at the two zooms where a city's pill splits into its separate stations. One honest difference: a hub's pill now counts a tiny branch line among its routes even at a zoom where that branch is too short to draw, which changes its size by about the width of a hair. Found by the atlas’s own audit sweep.

26 August 2026

One line tonight rather than four, and the arithmetic behind that is the useful part. The Joban Line runs 364 kilometres up the Pacific coast from Tokyo to Sendai, inland of the Shinkansen and slower than it, opened at Nippori in January 1889 and complete on 1 April 1905, and its whole middle section reopened in March 2020 after nine years shut behind the Fukushima exclusion zone. It is drawn from its own relation and sits on real track for every metre of its length. Japan has seventy nine other routes on this map and this was not one of them. Found by the atlas’s own audit sweep.
Four other candidates were measured, sourced and then set aside, each at a different gate, and listing them is more useful than listing what shipped. The Iida Line has a relation that does not connect end to end, so no path exists inside it and drawing one would mean inventing the joins. The Taiwan Coastal Line opened on 11 October 1922 and runs ninety kilometres from Zhunan to Changhua, and no published timetable for it could be found. Hanoi to Dong Dang has exactly one train a day each way and the only journey time on offer is a traveller's guide saying a little over four hours, which is not a number a card can carry. And Beijing to Kowloon, two thousand three hundred kilometres of undrawn main line, has had no through train since 2020. Found by the atlas’s own audit sweep.
Outside Europe the binding constraint is not the railway and not the geometry. It is that a card quotes a journey time and there is no planner to ask. In Europe the planner answers, names the train and now names the company that runs it, and a candidate goes from a row in a file to a finished card in about ten minutes. Everywhere else each line needs its operator's own timetable found and read, and the honest failure rate of that search is about four in five. That is why tonight is one line, and why the count will keep slowing rather than stopping. Found by the atlas’s own audit sweep.
Three more, and one of them is the first card off a seam that needs a different kind of work. The Binsui line is the eastern arm of the old Chinese Eastern Railway, Harbin out to the Russian border at Suifenhe, running provisionally from November 1901 and opened properly on 14 July 1903; train K7023 takes ten hours and five minutes over it. Beside it two European branches: Golesti to Campulung in Romania, inaugurated on 21 June 1887 and derailed two months later between Mihaesti and Stalpeni, and Hatvan to Ujszasz down the Zagyva through the Jaszsag, opened on 10 March 1873 with four stations on it. The Romanian line published its own journey time as a hundred and fifteen minutes, and the planner independently answered one hour fifty six, which is the sort of agreement that says both are right. Found by the atlas’s own audit sweep.
The European seam is worked out, and it is worth saying exactly what that means rather than leaving it as a feeling. Every railway line in Wikidata that carries an opening date, is fifty kilometres or longer, has an OpenStreetMap relation and is not already on this map has now been put to the planner. What is left in the candidate file is either shorter than twenty five kilometres, has no relation to draw from, or lies outside the region a European journey planner can answer for. Tonight's queue, sorted by how much undrawn track each line carries, ran straight into China, the United States, Brazil and North Korea within a dozen rows. Found by the atlas’s own audit sweep.
Which is why only one Chinese line shipped tonight out of seven measured, and the reason is the journey time rather than the railway. Beijing to Kowloon is two thousand three hundred kilometres of undrawn main line and its through train has been suspended since 2020. Yingtan to Xiamen lost its passenger service between Sanming North and Jiaomei in January 2019, and a map of where trains run should not show it. Shangqiu to Hangzhou and Hengyang to Liuzhou are both open and both busy, and the only journey times findable for them come from trip aggregators rather than from a timetable, which is not the standard the rest of this atlas is held to. They stay out until a published schedule says otherwise. Found by the atlas’s own audit sweep.
Five more, and every one of them drew whole from its relation and sits on real track for the whole of its length. The Frankenwaldbahn climbs out of the Main valley to the Thuringian border at Probstzella by Kronach and Ludwigsstadt, finished in October 1885 and steep enough that heavy trains needed banking engines. The Hohenlohebahn runs up the Kocher from Heilbronn through Schwabisch Hall to Crailsheim, first half open in August 1862 and the rest not until 10 December 1867. The Delmenhorst line crosses the Oldenburg south country by Vechta and Lohne, built in four pieces between 1888 and 1 May 1900. Roccasecca to Avezzano climbs the Liri by Arce, Sora and Balsorano into the Fucino basin, eighteen years from the first section to the last on 20 August 1902. And Mateszalka to Zahony opened on 27 September 1905 as a local line from Nagykaroly to Csap, cut in three by the border of 1920, of which this is the Hungarian piece. Found by the atlas’s own audit sweep.
Of thirty candidates whose opening year has now been checked against the line's own article, eighteen were wrong in Wikidata. Four of tonight's five were: the Frankenwaldbahn is dated 1863 and was finished in 1885, the Hohenlohebahn 1862 and 1867, the Delmenhorst line 1899 and 1900, Roccasecca to Avezzano 1884 and 1902. Only Mateszalka to Zahony was right, and it was right for the reason all the right ones have been right: it opened complete on a single day rather than in stages. The pattern is now firm enough to state as a rule. A precise date in Wikidata is usually the truth. A bare year is usually the first section of something that took another twenty years to finish. Found by the atlas’s own audit sweep.
Three were refused tonight for the mismatch that keeps recurring, and one of them is subtle enough to be worth naming. Agrigento to Palermo has a direct service and it is a bus. Ljubljana to Metlika is a real railway and a real train, but the card would have been a longer version of the Ljubljana to Novo mesto card already here, which the coverage tool said by scoring it thirty five per cent new and naming the card it overlaps. And Olomouc to Sumperk looked clean until the stop list was read: the relation is line 290 through Sternberk and Unicov, and the only direct train goes the other way round through Mohelnice and Zabreh. Same two cities, two railways, and the planner answers with whichever is quicker. Found by the atlas’s own audit sweep.
Four more, and all four drew whole on the first attempt, which has not happened before. The Lahn valley from Wetzlar down to the Rhine through Weilburg, Limburg, Nassau and the spa at Bad Ems, finished on 10 January 1863. The Kinnekullebanan round the flat topped hill by Lidkoping and Mariestad, four small railways stitched together, the last piece laid in 1910 and the whole of it regauged by 1953. Civitanova Marche up the Chienti and Potenza valleys by Macerata and Matelica to Fabriano, begun in 1884 at the far end and joined up in 1888. And the Lemvigbanen, a Danish local line that never belonged to the state, out to the fishing port of Thyboron, finished on 1 November 1899 and privately run to this day. Found by the atlas’s own audit sweep.
The reason all four drew first time is a change in where a card's two ends come from. Until tonight they were station coordinates looked up one at a time, which is slow and also wrong in a specific way: a station is a place people leave from and a relation is a piece of infrastructure, and the two do not always meet, which is why two cards had to be re-spanned earlier today. Now the ends come off the relation itself, as the two points furthest apart in its own geometry, so the drawing starts and finishes exactly where the mapped railway does. `_ask-rels.mts` takes a relation id, finds those ends, asks the planner by coordinate rather than by station name, and prints the fastest direct train with the company that runs it. Found by the atlas’s own audit sweep.
That also fixed the quiet failure that had been losing candidates all evening. Asking a journey planner for Nyiregyhaza, Mateszalka, Knin, Zadar, Turku, Hallnas, Legnica or Macanet-Massanes returns nothing at all, not because no train runs but because the name search does not know them: nine of twenty seven queries in one batch came back as a failure to ask rather than an answer. Asking by coordinate rescued four of them straight away, and it is what found the Kinnekullebanan a second time. That line had been refused this morning because the only train between the two cities it names spends its first eighty kilometres on a main line already drawn here; asked between the railway's OWN two ends instead, there is a train that runs the whole of it in two hours and one minute. Found by the atlas’s own audit sweep.
Three more, and one of them opens a country. Plovdiv to Karlovo is sixty one kilometres north to the Rose Valley under the Balkan range, opened on 28 September 1933, and it is the first Bulgarian branch line on this map rather than another leg of the Sofia main line. Beside it, Oldenburg to Osnabruck, a hundred and eight kilometres across the moors and the Artland by Cloppenburg and Quakenbruck, opened as far as Quakenbruck in October 1875 and through on 15 November 1876; and the Saalbahn from Jena up the Saale gorge by Kahla, Orlamunde and Rudolstadt to Saalfeld, opened 30 April 1874. All three sit on real track for the whole of their length. Found by the atlas’s own audit sweep.
Finding candidates stopped being hand work today. The labels Wikidata gives railway lines are almost always their two ends, in whatever convention the language uses, so `_pairs-from-labels.mts` now parses them and prints a file the planner batch can eat: Ligne de X à Y, Bahnstrecke X–Y, Línia X-Y, X–Y-vasútvonal. Forty candidates a run, spread across countries so one country cannot fill the batch. It guesses, and it says so: a line named after three towns gives the first and the last, which is right for a through route and wrong for one named after a junction in the middle. The planner refuses the rest without being asked. Found by the atlas’s own audit sweep.
Two cards had to move their own ends tonight, and the reason is worth writing down because it will keep happening. A relation is a piece of infrastructure and a station is a place people leave from, and the two do not always meet. The Oldenburg line's relation stops at Osnabruck-Eversburg, four kilometres short of the station the trains run to; the Saalbahn's stops near Grossheringen, ten kilometres short of Naumburg. Rather than stretch a drawing to a station it does not reach, both cards were re-spanned onto ends the relation actually has, which is why one says Jena rather than Naumburg. Drawing to where the trains go and claiming track nobody mapped is the error this avoids. Found by the atlas’s own audit sweep.
Five more, every one of them on its train's own track at p90 zero and on real rails for the whole of its length. The Direttissima under the Apennines is the one worth stopping at: opened complete on 22 April 1934, an eighteen and a half kilometre base tunnel that was then the second longest in the world after the Simplon, and a hundred and eighty kilometres an hour where the old line over the pass managed seventy five. Beside it, Udine to Trieste through Gorizia and out along the Carso, whose last two sections opened three days apart in October 1860; Ljubljana to Jesenice, the Slovenian half of the Rudolf Railway, opened complete on 14 December 1870; Gotha to Leinefelde across the Thuringian Basin, 1870, now rebuilt for a hundred and sixty; and the Danish Northwest line from Roskilde out to the Great Belt at Kalundborg, opened on 30 December 1874. Found by the atlas’s own audit sweep.
One fact has blocked two good lines today and it is now unblocked for good: who runs the train. German regional networks change hands on tender and the web disagrees with itself for years afterwards, so the Saalbahn and the Gotha line were both set aside rather than ship a contested operator. The planner knew all along. It answers from the operators' own data, and its reply carries the company on every leg; `_ask-scotty` now prints it. The Gotha line runs under DB Regio Sudost, the Saalbahn under Abellio Rail Mitteldeutschland, and the Halle to Halberstadt card shipped earlier today under Start Mitteldeutschland is confirmed correct. The operator is no longer a thing that has to be searched for. Found by the atlas’s own audit sweep.
Three candidates were dropped on evidence rather than on doubt. Zvolen to Kosice and Valga to Tartu are both already on this map under other names, which the coverage tool said by naming the existing card rather than by scoring low. The Lille to Fontinettes relation carries five kilometres of track in OpenStreetMap, so there is nothing to draw. And the Lubeck to Puttgarden line's only direct service is a bus, which the Fehmarn works explain and which a map of where trains run should not show as a train. Found by the atlas’s own audit sweep.
Five more. The Verona to Bologna line took thirty seven years to cross the Po plain, begun at Bologna in 1887, over the river at Ostiglia in 1911 and finished into Verona on 1 February 1924, and it is now fifty two minutes end to end. The old Milan to Bologna road through Emilia by Lodi, Piacenza, Parma, Reggio and Modena opened single track in 1861 and was doubled by 1894. Dresden to Gorlitz crosses Saxony through Bautzen and the Lusatian hills, opened in stages from November 1845 and formally complete on 1 September 1847. The Vendsyssel line runs to the ferry quays at Frederikshavn and reached Aalborg itself only in 1879, when the Limfjord bridge opened. And fifty six kilometres up the Zala valley, Boba to Ukk opened in January 1889 and the run on through Zalaegerszeg in October 1890. Found by the atlas’s own audit sweep.
The Milan to Bologna card is the first here that the timetable checker cannot judge, and the reason is the point of the card. It asks whether a drawing lies on the shape of the service that runs it, and it picks the fastest service, which between these two cities is a Frecciarossa on the high speed line built beside the old one. So it reported the card straying five kilometres from a train it was never meant to be on. What settles it instead is the rail audit, which finds the drawing on real track for ninety eight per cent of its length with one disagreement in the throat of Milan Centrale where several parallel tracks are equally plausible, and the Intercity the card names, which takes two hours forty five on the old road while the Frecce take one hour five on the new one. Found by the atlas’s own audit sweep.
Of eighteen candidates whose opening year was checked today, twelve came back wrong. Verona to Bologna is dated 1887 in Wikidata, which is the first fourteen kilometres out of Bologna; Dresden to Gorlitz is dated 1845, which is the first section to Radeberg; the Vendsyssel line is dated 1871, which is the year it reached Norresundby on the far bank of the Limfjord rather than Aalborg. Three were right: Amiens to Rouen, which opened complete on one day, Grossheringen to Saalfeld, which did the same, and Milan to Bologna, whose 1861 really is the year the second half opened and joined the two ends. A date that opens a line is not the date the line opened. Found by the atlas’s own audit sweep.
Four more, and two of them are trunk railways this atlas had somehow never drawn. Turin to Genoa is 165 kilometres over the Apennines through Asti, Alessandria and Novi Ligure, five years of tunnelling that opened in December 1853, and the Giovi tunnel at its summit was the longest in the world when it was bored. Domodossola to Milan is 122 kilometres down the Toce and along the west shore of Lake Maggiore by Stresa and Arona, the Italian approach to the Simplon, reached in 1888. Beside them, two quieter lines: Manresa to Lleida, 117 kilometres over the Segarra plateau by Calaf and Cervera, where the first trains ran in 1860 while the coastal route through Valls took the traffic; and Krefeld to Kleve, 66 kilometres down the Rhine's quiet left bank by Kevelaer and Goch, opened on 1 March 1863. All four are on real track for at least 98.8 per cent of their length, three of them for all of it. Found by the atlas’s own audit sweep.
The checker corrected two of today's four cards before they shipped, and it is the same correction both times: the card had the wrong train. Turin to Genoa was written at one hour fifty five off an Intercity, and the fastest direct run in the timetable is a hundred and one minutes on a Regionale Veloce, so the card says one hour forty one and names the train that actually earns it. Domodossola to Milan came down from one hour forty two to one hour thirty three the same way. A card here quotes the fastest scheduled journey, and the first draft of a card usually quotes whichever train the planner happened to show first. Found by the atlas’s own audit sweep.
The count of wrong opening dates is now ten out of fourteen, and one of today's rejects shows why the check has to be the line's own article rather than the item's label. Wikidata's Ligne de Saint-Germain-des-Fossés à Darsac looked like a clean match for a direct Clermont-Ferrand to Le Puy-en-Velay train, right length, right region, right era. Reading the article, that line runs through Vichy, Thiers, Ambert and La Chaise-Dieu, and the train runs through Issoire, Brioude and Arvant. They are different railways that happen to end near each other. Two more were dropped for the same class of mismatch: the Nahe valley relation turns out to describe a 61 kilometre regional service rather than the line, and the only direct Gothenburg to Skövde train runs the main line this atlas already draws rather than the Västergötland railway the candidate named. Found by the atlas’s own audit sweep.
Four more, and this round the interesting number is how often the free opening date was wrong. Poitiers to La Rochelle, 135 kilometres across the Poitou plain that the Paris TGVs finish on; Bordeaux to Hendaye, 233 kilometres dead straight down the Landes pine forest to the Basque coast; Amiens to Rouen, 117 kilometres of cross country line from Picardy into Normandy; and Halle to Halberstadt, 92 kilometres to the northern edge of the Harz. All four sit on real track for at least 98.3 per cent of their length. Wikidata dated them 1856, 1841, 1867 and 1869. Only Amiens to Rouen was right, and it was right because that line opened complete on a single day, 18 April 1867, instead of in stages. The other three: Poitiers reached Niort in 1856 and La Rochelle the year after, so 1857; Bordeaux got as far as Lamothe in 1841 and to the Spanish border only in 1864; and the 1869 on the Halle line is the day Halberstadt to Vienenburg opened, while the run from Halle itself was finished in 1872. Across eleven candidates checked today the date that came free was wrong seven times. It is a lead worth having, because it names the line and gives you something to check. It is not a fact. Found by the atlas’s own audit sweep.
Three candidates were refused today for the same reason and it is worth naming, because the trap is invisible from either side on its own. A city pair usually has more than one railway between it, and a journey planner answers with the fastest train rather than the line you asked about. Barcelona to Lleida came back with a direct three and a half hour service that runs through Valls, not the Manresa line the card would have drawn. Leipzig to Saalfeld came back direct through Naumburg, not the Gera line. Paris to Tours came back direct through Orleans, not the Vendome line. Each would have put a real train, correctly timed, on the wrong railway. The check that catches it is free: read the stop list the planner returns and see whether the towns on it are the towns the line runs through. Found by the atlas’s own audit sweep.
Two more were dropped before they could become cards. The Fives to Hirson line in northern France has an OpenStreetMap relation carrying a single kilometre of track, so there is nothing to draw from. And the Kinnekullebanan in Sweden is a real gap, but the train that runs it starts in Gothenburg and spends its first eighty kilometres on the main line this atlas already draws, so the card would have been part duplicate. It needs a decision about what it spans rather than more checking. Found by the atlas’s own audit sweep.
Three more from the same list, and the biggest is the whole instep and toe of Italy. The Ionian Railway runs 470 kilometres from Taranto along the arch of the boot by Metaponto, Sibari and Crotone, round the toe past Melito di Porto Salvo and into Reggio Calabria, and one Intercity does it end to end in five hours fifty seven. Nothing on this map went near it: the Frecce to Reggio come down the Tyrrhenian side, and the only other card touching Taranto runs the other way to Bari. Beside it, two much quieter railways. Zafra to Huelva is 184 kilometres laid in 1889 to bring ore out of the Sierra de Aracena, still crossing it past halts named for villages miles from the line, Jabugo, Almonaster, Valdelamusa. And the Stangadalsbanan is 235 kilometres across Smaland from Linkoping to Kalmar, four separate railways stitched together, the last of them opened in 1902 and the through journey possible without changing trains only since 1996. All three sit on their train's own track at p90 zero, and all three are on real rails for at least 98.5 per cent of their length. Found by the atlas’s own audit sweep.
The opening date that comes free with a candidate has now been wrong three times out of seven, so treat it as a lead rather than a fact. Wikidata dates the Ionian Railway to 1866, which is the day Reggio Calabria to Lazzaro opened, a few kilometres of a 470 kilometre railway; the sections were joined at Crotone to Catanzaro Lido on 15 November 1875 and that is when the journey on this card became possible, so 1875 is what it says. The same trap sat under the Warsaw to Katowice classic line, dated 1845, which is the first thirty kilometres to Grodzisk Mazowiecki; the rails reached the Austrian border in 1848. The rule is unchanged and it keeps earning its place: the year a card carries is the year its LAST section opened. Found by the atlas’s own audit sweep.
One candidate was refused for a reason no amount of checking could fix, and it is worth recording because the railway is real. The classic line from Warsaw to Katowice, through Koluszki, Piotrkow Trybunalski and Czestochowa, is genuinely missing from this map: the express card that shares its two cities runs on the Central Main Line through Zawiercie instead, and nothing here passes within eight kilometres of Czestochowa or Piotrkow. But its OpenStreetMap relation is mapped in twenty disconnected pieces with gaps of 83, 34 and 32 kilometres, and no path exists inside it between the two stations. Drawing it would have meant inventing the joins. It stays off until the relation is finished or the line is traced another way. Found by the atlas’s own audit sweep.
Four lines, and they were found by turning the search around. Every hunt for new routes here has started at OpenStreetMap: list the relations mappers have drawn over a country, subtract what this atlas already has, look at what is left. That inventory ran dry, and it can now be said with a number rather than a feeling. This project has cached the full geometry of 5,589 OSM relations over the summer, of which 3,660 are wired into a card and 1,929 are not; walked end to end against every line already published, nine of those 1,929 carry twenty five kilometres of track no card draws, and most of the nine are re routing details inside lines that are already here. So the search moved to Wikidata, where a railway line is an item in its own right rather than a service somebody chose to map. There are 19,436 of them, 7,109 carry a sourced date of official opening, and after subtracting everything this atlas already claims, 1,454 are longer than fifty kilometres. That date is the point. Candidates have been stalling at the last gate all month for want of a published opening year, and starting from Wikidata means the year arrives with the candidate instead of having to be hunted after the work is done. The four that came through every gate today: Szeged to Bekescsaba, ninety six kilometres east across the Great Hungarian Plain, line 135 of 1870, whose rails carry on past Bekescsaba to Salonta over a border drawn in 1920; Poznan to Pila, ninety five kilometres north through Chodziez, built in 1879 to hook Poznan onto the Prussian Eastern Railway; the Main Weser Railway, 199 kilometres from Frankfurt through Giessen and Marburg to Kassel, one of the first trunk railways in Germany; and Rome to Pescara, 237 kilometres over the Apennines by Tivoli, Avezzano and Sulmona, where the Transiberiana d'Italia branches off into the Abruzzo highlands. All four are drawn from their own relation and all four sit on their train's own track at p90 zero, with no station further than a hundred metres from where the timetable says it stops. Found by the atlas’s own audit sweep.
The new method needed a check of its own before any of it could be believed, and the check failed the first time. Deciding whether a candidate is already on this map is now a measurement rather than a judgement: every point of every published line goes into a grid, the candidate is walked at one kilometre steps, and what gets reported is the longest unbroken run that no card comes near. Pointed at seventy six relations this atlas is demonstrably drawn from, which should all read as fully drawn, the median came back at thirteen per cent undrawn and only thirty four of the seventy six sat under ten. Widening the tolerance to three kilometres brought the median to four and fifty three of seventy six under ten, and the rest is structural rather than fixable: a relation is often a network with branches while a card draws one path through it, so branches this atlas rightly does not draw are counted as missing. It shortens fifteen hundred candidates to a few dozen, which is what it is for, and it is not evidence about any single one of them. Found by the atlas’s own audit sweep.
Two numbers off Wikidata turned out to need reading twice, and both would have shipped as facts. Length is a quantity with a unit, and the unit is not always the kilometre: the Bamberg to Rottendorf line publishes 92,153, which is metres, and the first ranking of candidates duly announced eighty four thousand kilometres of undrawn German railway. And an opening date on a line built in stages can be the date of the first section rather than the day the through journey became possible. Wikidata gives the Main Weser Railway as 1848; its own article says the line opened between 1849 and 1852, so the card here says 1852, on the same rule that dated Nancy to Luxembourg to 1859 last week. The lesson is the one this atlas keeps relearning: a fact that arrives already attached to a candidate still has to be read. Found by the atlas’s own audit sweep.
Two candidates were refused rather than drawn, and each was refused for a different reason worth writing down. The 1968 direct railway from Madrid to Burgos passed every geometric test and the planner answered with a train doing the journey in one hour fifty three, and that train runs on the high speed line through Segovia and Valladolid, not on the railway the card would have described. Runnable between two stations is not the same railway. And the Kutno to Pila line, four hundred and ninety five kilometres of it on paper, shares its whole first half with the Warsaw to Bydgoszcz card already on this map, so drawing it would have put a second line on track that is already here. The tool that found the candidates named the offending card itself, which is the part that makes it worth keeping. Found by the atlas’s own audit sweep.

24 August 2026

The Tren Maya was drawn as one line and the railway is a loop, so about two fifths of it was not on this map at all. Palenque up the Gulf side through Campeche and Mérida to Cancún was here; Cancún back down the Caribbean coast through Playa del Carmen, Tulum and Bacalar to Chetumal was not. It is here now, 366 kilometres of it, welded out of the ways Mexican mappers drew and ending ten metres from Chetumal station. The two cards share no track, which is why this is a second card and not a longer first one. Worth saying how it went wrong, because the line that was already drawn was never off the rails: it comes from the relation that maps the Tren Maya's track rather than any train's journey, the relation says so itself in a note, and it branches. Asked for one path between two stations, the tracer found a real one and quietly dropped every branch. The geometry was right and the span was wrong, which is exactly the sort of error a geometry check cannot see. ks2048.
Two Slovenian lines out of Ljubljana, and they were found by asking a question the 22 August sweep had left unasked. That sweep verified the top of its own drawable list and stopped; eleven rows over fifty kilometres were never put to the planner at all. Put to it now, three answered with a real train. Ljubljana to Novo mesto is the Lower Carniola line of 1894, seventy five kilometres over the Grosuplje watershed and down the Krka valley, and Ljubljana to Kocevje is the line of 1893, built for coal and timber, closed to passengers for half a century and carrying them again since 2021. Both drawn from their own relation and both sitting on their train’s own track at p90 zero. The Novo mesto card was written claiming two hours and four minutes and the checker refused it: the fastest direct run in the timetable is ninety nine, and a card here quotes the fastest. It says one hour thirty nine. The third of the three, the Savonlinna branch in Finland, was dropped rather than drawn, because its relation is in nine pieces and paths only forty five kilometres of a railway that takes fifty three minutes to ride. Found by the atlas’s own audit sweep.
Gijon to Puente de los Fierros, 74 kilometres up the Caudal valley to the foot of the Pajares ramp, drawn from Renfe’s own C-1 relation and checked against the planner at p90 0.1 kilometres on train R 12102. It was the last line over fifty kilometres left on the verified European list, held back since 22 August for want of an opening year, because a card here cannot open in a year nobody published and the rule is not to borrow one from the railway next door. Its own sections publish it: Gijon to Pola de Lena on 23 July 1874, Pola de Lena to Puente de los Fierros on 15 May 1881. So the run became possible in 1881, and the 1884 date that keeps turning up in the sources belongs to the Pajares ramp beyond this card, climbing on towards Leon. The two sections sum to 74.975 kilometres against a drawing of 74.0. Found by the atlas’s own audit sweep.
Sicily’s Ionian coast was on this map twice. Two cards ran Messina to Siracusa, both Trenitalia, both scenic, both opened in 1871, both drawn whole from the same OpenStreetMap relation, one calling itself the Ionian Coast Line and the other simply Messina–Siracusa. Neither was wrong about the railway; there was just one railway and two of it. They are one card now, keeping the older page so no link breaks, taking the fuller description off the newer one, and carrying 175 kilometres because that is what the drawing measures rather than the 158 the older card claimed. It was found by asking a question this atlas had never asked: which cards are drawn from the same OSM relation. Seventy four relations are shared, and nearly all of that sharing is honest, because a sleeper and a day train and a luxury tourist train can all run over one line and each deserves its own card. This pair was the one where both cards described the same service. Found by the atlas’s own audit sweep.
A fourth Shosholoza corridor, north to the Limpopo. Johannesburg to Musina, 632 kilometres up through Polokwane and the bushveld, ten hours overnight, on a line that reached the copper town in 1914 and the river crossing into Zimbabwe in 1927. The card stops at Musina and not at Beitbridge, which is what the OpenStreetMap relation is called and twenty kilometres further on, because Musina is where the train stops and a card here describes a service rather than a railway. That cost something: the relation is mapped in four disconnected pieces, and while it can be welded end to end, no path exists inside it to a station in the middle, so the usual tool refused. The line was welded and cut at Musina instead, and then thinned by how far each point sits from the line rather than by dropping every so many, because the first attempt quietly shortened a 633 kilometre railway by fourteen. It sits on real track for 99.9% of its length, four metres off at the median. Found by the atlas’s own audit sweep.
Three more from outside Europe, taking the day's non European additions to five. The Shosholoza Komati runs the old Delagoa Bay line from Johannesburg out through Pretoria and Witbank and down the escarpment to the Mozambique border, 539 kilometres in twelve and a half hours overnight; its first passenger train ran on 18 November 1894 and the line opened formally on New Year's Day 1895, and the photograph on the card comes from the commemorative book printed for that opening. The other two are Buenos Aires radials, each on its own railway out of its own terminus: the San Martin line west to Junin, 254 kilometres, where the Buenos Aires al Pacifico arrived in 1884, and the Sarmiento line to Bragado, 210 kilometres, reached by the Ferrocarril Oeste in 1877 at what its own kilometre posts call km 210. The drawing measuring 210 against a railway that has called it that for a century and a half is the kind of agreement that says the relation is the right one. Bragado ships with no photograph because Commons has none of it. Found by the atlas’s own audit sweep.
Two cards from outside Europe, and they were found a different way. The European sweep leans on a journey planner to prove a train really runs a line, and that planner stops at Europe, so the rest of the world has been searched with one hand tied. The inventory of what OpenStreetMap knows and this atlas does not draw works everywhere, and asked for the notable services it lists with no card, it turned up the Natal main line and the Mitre line north. Johannesburg to Durban is 724 kilometres over the Drakensberg escarpment, leaving at half past eight at night and arriving at seven; the through connection was finished in 1895 and the first train ran into Johannesburg from Durban on 2 January 1896. Buenos Aires to Rosario is 316 kilometres up the Paraná through Campana, Zárate and San Nicolás, and the day it opened, 1 February 1886, was the day the two cities were first joined by rail. Both are drawn from the relation that names the service, at 724 against a card of 725 and 316 against 314. Three other candidates were rejected rather than drawn: Namibia's Windhoek to Keetmanshoop, because the trains have been suspended since 2021 and a map of where trains run should not show one that does not; and two Amtrak routes, the Empire Service and the Palmetto, because they would have put a second line on track the Maple Leaf and the Silver Meteor already draw. Found by the atlas’s own audit sweep.
The last row of the European sweep that was blocked on nothing but a date is on the map: Nancy to Luxembourg, 119 kilometres up the Moselle through Metz and Thionville, drawn from the relation French and Luxembourgish mappers wrote and sitting on its train's own track at p90 zero, with no station further than a metre from where the timetable says it stops. It was held back because a card here cannot open in a year nobody published, and the answer was not to take one from the railway next door: the three sections have their own articles, and they date themselves. Nancy to Metz was inaugurated on 10 July 1850, Metz to Thionville on 16 September 1854, Thionville to Luxembourg on 11 August 1859. The through journey became possible in 1859, so that is the year on the card. Found by the atlas’s own audit sweep.
Three people have now reported a route as missing when it was on the map the whole time, hidden because heritage and luxury start switched off: the Ghan and the Indian Pacific earlier this month, the White Pass & Yukon between Skagway and Carcross this one. All three were right that they could not see the line, which makes it this map's fault and not theirs. The filter panel now says how many of the routes are lit and which shelves are dark, on first load and before anything is opened, with one click to light them all. rwf-rr.

22 August 2026

Two more from the ranked list, both of them scenic lines that had no kilometre of themselves on this map. The Valsugana runs out of Trento past the vineyards, along Lake Caldonazzo and down the gorge of the Brenta to the foot of Monte Grappa, twenty three stops in a little over two hours on a line begun in 1896 and finished in 1910. The Istrian line goes the length of the peninsula at a walking pace, Pula to Buzet in two and a half hours through the red earth and the karst, on the state railway Austria opened to its naval port on 20 September 1876. Ninety five and eighty six kilometres, both drawn from the relation Italian and Croatian mappers wrote, both sitting on their train's own track at p90 zero with the worst calling station a hundred metres away. Found by the atlas’s own audit sweep.
Bosnia was tried and could not be finished tonight. Banja Luka to Doboj is ninety three kilometres of railway with nothing of ours on any of it, which makes it one of the cleanest gaps in Europe, and the planner this atlas is allowed to ask does not answer for either station, so there is no way from here to show that the service runs before drawing it. It stays on the list with the reason attached rather than being drawn on the strength of a relation alone. Found by the atlas’s own audit sweep.
The list this atlas has been working down was ranked by how much railway is missing and never asked whether the railway could be drawn, and the first seven candidates on it turned out to be unusable for the same mechanical reason: the relations OpenStreetMap carries for them are mapped in seventeen to thirty three disconnected pieces, welded across holes as wide as fifty three kilometres. Last night took the clean ones, and a ranking blind to drawability cannot say so. There is now a second question asked before any of the work starts, which is whether a path exists across a relation between its own first and last stop, and the answer is that a relation in thirty pieces can still be perfectly drawable, because a network is not a chain. Of sixty candidates still open, forty five can be drawn, seven need a person to look at them and eight are dead. Found by the atlas’s own audit sweep.
The first one off that list is on the map: the REX 92 from Wiener Neustadt to Fehring, a hundred and thirty two kilometres in two and a half hours over the Wechsel and down through the spa country of eastern Styria. It is the surviving fragment of something much larger. A consortium put a railway from Vienna through Wiener Neustadt, Hartberg and Fürstenfeld to Thessaloniki before the government in 1873, the law passed that May, the Vienna stock market crashed the same spring and the plan collapsed, and the line was built anyway in pieces by different promoters over the next thirty years: Wiener Neustadt to Aspang in 1881, Fehring to Fürstenfeld in 1885, on to Hartberg in 1891, the gap to Friedberg in 1905, and the crossing of the Wechsel itself only in 1910. Boundary stones lettered WSB, for Wien Saloniki Bahn, still stand along it. Drawn from relation 1660006, every metre of it on rail, and on its own train's track at p90 zero with the worst calling station a hundred metres out. Found by the atlas’s own audit sweep.
Catalonia gains the R13 that runs from Barcelona to Lleida the slow way, a hundred and seventy two kilometres in three and three quarter hours: out of the Estació de França and down the coast through Sitges and Vilanova, then inland at Sant Vicenç de Calders and up through Valls and Montblanc to the plain of Lleida. The seaward half opened on 29 December 1881. Drawn from relation 6023370, ninety nine per cent of it on rail, and on the train's own track at p90 zero with no calling station further than a few metres out. The photograph is of Rodalies trains standing in the Estació de França itself, which is where the journey starts. Found by the atlas’s own audit sweep.
Two candidates were authored tonight and thrown away, and both failed on the same question, which is not whether a railway exists but whether a train runs the whole of it. The R14 is the twin of the R13 and goes to Lleida by Tarragona and Reus instead of by Valls; its line is real, its relation draws cleanly at a hundred and ninety nine kilometres, and the planner has no through train on it at all, so the card was measured against the Valls train and found nineteen kilometres adrift before it was deleted. The Austrian R58 from Amstetten to Selzthal was refused here too, and that refusal was WRONG: see the entry below. The R14 stays refused, with the reason written down rather than being drawn on the strength of a relation alone. Found by the atlas’s own audit sweep.
Correcting the entry above, the same evening it was written. The Rudolfsbahn from Amstetten to Selzthal was refused for having no through train, and it has one: two hours and five minutes, one train, every Saturday and twice every Sunday. The mistake was in how it was sampled. Three dates were asked and that felt like enough, but two of them were Tuesdays and the third was a Sunday that fell inside the engineering closure still bustituting the southern half, so a weekend service outside the closure was never once looked at. A line that runs only at weekends is invisible to any number of weekday samples, and the honest description of what happened is not that the planner was unclear but that the question was asked badly. The checker built afterwards now asks a weekday, a Saturday, a Sunday and a date seven weeks out before it will say no. Found by the atlas’s own audit sweep.
And the Rudolfsbahn is on the map, which is the whole point of having found the mistake. A hundred and nineteen kilometres from Amstetten up the Ybbs through Waidhofen, over to Kleinreifling and then along the Enns through the Gesäuse by Hieflau and Admont to Selzthal, in two hours and five minutes. The Kronprinz Rudolf Bahn company built the main line between 1868 and 1873 and its name now survives only for this northern stretch. One train does it on Saturdays and two on Sundays, and on weekdays you still change, which is exactly the pattern that made it look dead. Drawn from relation 1877502, every metre on rail, and on the train's own shape at p90 zero with no calling station further out than the width of a platform. Found by the atlas’s own audit sweep.
Three more off the verified list, all drawn from their relation and all three sitting on their own train at p90 zero with every calling station inside a few metres. The Carinthian line crosses the border at Holmec and runs down the Drava through the old ironworks valleys of Prevalje, Ravne and Dravograd to Maribor, eighty seven kilometres on a single track built between 1857 and 1863 to carry Carinthian steel out to the world. The Mantova to Monselice line goes straight across the Po plain from Lombardy into the Veneto by Nogara, Legnago and Montagnana, eighty four kilometres, its first stretch opened with a special train on 16 May 1885. And the Sibenik line climbs seventy five kilometres from the Adriatic into the Dalmatian hinterland to Knin, on rails opened on 4 October 1877 to bring the coal of the Drnis basin down to the harbour. None of the three has a photograph yet: Commons offers a river at Legnago and a plinthed steam engine at Dravograd, and either would say something about these railways that is not true. Found by the atlas’s own audit sweep.
Six more, all short and all with a sourced date, because the year a line opened is a published number and this atlas does not guess one. The Mühlkreisbahn climbs out of Linz on the north bank of the Danube into the Mühlviertel, opened November 1900. The Triesting valley line runs nineteen kilometres down out of the Vienna Woods to Leobersdorf, the surviving stub of a branch opened as far as Kaumberg in September 1877. The Gleichenberger Bahn is twenty one kilometres of Styrian light railway from the spa down to Feldbach, opened in the presence of the President of Austria in June 1931. The Beira Alta carries the last forty five kilometres of Portugal from Guarda to the Spanish frontier, inaugurated 3 August 1882. Bergamo to Treviglio is the line Bergamo had to fight for, since the Milan to Venice railway was first planned to miss the city, and it opened on 12 October 1857. And the Malpensa Express runs Terminal 2 to Milano Cadorna in three quarters of an hour, opened May 1999. Every one drawn from its relation, every one a hundred per cent on rail, every one on its own train at p90 zero. Found by the atlas’s own audit sweep.
Bergamo to Brescia closes the run at forty nine kilometres. Milan to Venice once went this way, through Bergamo rather than straight across from Treviglio to Rovato, and the whole route opened on 12 October 1857; the direct cut of 1878 took the traffic and left this a local line. On rail from end to end, p90 zero against R 10117. Three candidates were dropped in the same pass on measurement rather than opinion: Ponferrada to Monforte and Pula to Lupoglav are already drawn in their entirety, nought kilometres of new railway between them, and Divaca to Pula is a hundred and four of its hundred and twenty two kilometres already on the map, so what it really wants is for the Istrian card to be re-spanned rather than a second card beside it. Found by the atlas’s own audit sweep.
Three of the twelve stalled cards are unstalled, and the way it was done is worth writing down because the first attempt had been getting wrong answers confidently. Asking a search engine for an article title produced the Vinkovci to Osijek line when the question was Vinkovci to Gunja, and a Lithuanian branch that is not the one the train uses. So the question went to the map instead: walk the service own path, collect every railway line that OpenStreetMap mappers have tagged with a Wikidata identity along it, and read the opening date off Wikidata. That gives an answer about the track the train is standing on rather than about a title that merely sounds right. Out of it: the Linha do Leste from Entroncamento up the Tagus to Abrantes, opened 28 October 1863, twenty nine kilometres in twenty two minutes. Zabok north into the hills of the Hrvatsko Zagorje, twenty four kilometres on a branch of 4 September 1886. And the Linha do Vouga, the last metre gauge railway carrying passengers in Portugal, an hour from the Atlantic at Espinho inland to Oliveira de Azemeis on track opened 21 December 1908. All three on rail from end to end and on their own train at p90 zero. Found by the atlas’s own audit sweep.
Three more, and this time the dates came out of the article each line article actually points at rather than one that merely sounds like it. Vinkovci runs forty nine kilometres south across the Slavonian plain to Gunja on the Bosnian border, a line finished in 1886 whose bridge over the Sava carried it into Bosnia in 1894 and on to Tuzla in 1946. Osijek to Beli Manastir goes half an hour north over the Drava into Baranja, put into service at the end of 1870. And Ormoz crosses the Drava into Croatia to Cakovec in twenty four minutes, on the railway opened on 24 April 1860 to join Budapest to the Vienna to Trieste main line by way of Nagykanizsa and Pragersko. All three a hundred per cent on rail, the Osijek line at a median of one metre from the mapped track, and all three on their own train at p90 zero. Found by the atlas’s own audit sweep.
And the third one tonight is the way out of León to the Bierzo: a hundred and twenty eight kilometres to Ponferrada in under two hours, over the Montes de León by the Rampa de Brañuelas. That climb is the reason the line took so long. The Ferrocarril del Noroeste opened León to Astorga on 16 February 1866 and Astorga to Brañuelas on 17 January 1868, then spent more than a decade failing to get over the mountain, went bankrupt in 1878 and handed the works to the State; the stretch from Brañuelas down to Ponferrada opened only on 4 February 1882, under a company formed to pick up the pieces. Drawn from relation 1878688, every metre on rail, and on the train's own track at p90 zero. Found by the atlas’s own audit sweep.
Zurich to Vienna has been wrong since the beginning of the month and is now right. The railjet calls at St. Gallen and our line ran twenty five kilometres away from it, wandering off the corridor for the first hundred and eighteen kilometres out of Zurich and reaching thirty two kilometres off at worst. The cause was named a week ago and not acted on: none of the thirteen relations attached to that card described the Swiss stretch, so there was nothing correct to redraw it from. Drawing it from the running timetable instead settles it, and the line now sits on the shape of railjet 13477 at p90 zero with every calling station inside a hundred metres. Two facts on the card were wrong as well, and both were found by the redrawing rather than by reading: it is 811 kilometres, not 787, and the journey takes nine hours, not the eight hours ten the card claimed. Reported by alan_gr on the OpenStreetMap community forum.
The Alvia from Vigo to Barcelona joins the map on a reader's suggestion, and at one thousand three hundred and seventy one kilometres in fourteen hours and eight minutes it is the longest scheduled domestic run the atlas draws anywhere. It crosses Spain from the Atlantic to the Mediterranean in a single seat: Santiago de Compostela, Ourense, Monforte de Lemos, the Bierzo at Ponferrada, Astorga, León, Palencia, Burgos, Vitoria, Pamplona, Zaragoza, Lleida and Camp de Tarragona. The spine of it through Galicia and León is the old Palencia to A Coruña railway, finished on 4 September 1883. Suggested by alan_gr on the OpenStreetMap community forum.
That route also caught OpenStreetMap out, which is worth writing down because this atlas usually leans on it. The relation for the Alvia still describes the journey the train used to make, out of the old Vigo-Guixar station and inland by Redondela; the train now leaves from Vigo Urzaiz and runs up the Atlantic coast through Pontevedra and Vilagarcía to Santiago before turning inland. Drawn from the relation, the line missed Santiago de Compostela by sixty five kilometres. It is drawn from the running timetable instead, and the relation is deliberately not registered against the card, so that no later pass can quietly put the line back on a railway the service left behind. Found by the atlas’s own audit sweep.
Málaga to Granada is on the map, and the honest thing to say about it is that most of it was there already. The same reader corrected this atlas two weeks ago for claiming the journey ran entirely on track it draws elsewhere: a stretch really was built to make it possible, and Granada really had no direct train to Málaga before it. Measured rather than argued, five kilometres of the hundred and sixty five are railway the atlas did not draw, the rest being the Córdoba to Málaga and Antequera to Granada high speed lines it already carries. The card is here because the journey is a real one that people make in an hour and twenty three minutes, not because the line is new. The Bobadilla branch of 1874 served Granada for a century and a half; its replacement opened on 26 June 2019. Reported by alan_gr on the OpenStreetMap community forum.

21 August 2026

Every European country the atlas had not yet measured was then measured in one pass: three hundred and eighteen route relations across twenty six countries, each one scored by how many of its kilometres have no atlas line within two hundred and fifty metres. The ranked result is a work list rather than a verdict, and reading it the first thing to say is what it wrongly ranks first. The top three entries are Russian trains to Kaliningrad that happen to pass through Lithuania, and almost all of their two and a half thousand uncovered kilometres are in Russia, not Lithuania: the measure counts a relation against the country whose inventory it came from, which is the right thing to do for a national service and the wrong thing for a transit. Found by the atlas’s own audit sweep.
Two of the genuine leaders were drawn straight away. The Evros line runs two and a half hours north from the port of Alexandroupoli to Ormenio at the top of Greek Thrace, with Turkey out of the right-hand window the whole way, on rails the Oriental Railway opened in 1874 and a border drawn across in 1923; a hundred and seventy four kilometres, and not one of them had a line on this map. The Comboio Raiano is the borderland train across the cork oaks of the Alentejo, Abrantes to Badajoz by way of Ponte de Sor, Crato and Elvas, on the link to Spain that opened in September 1863; a hundred and forty six kilometres, of which a hundred and seventy seven of the relation's hundred and seventy nine were uncovered. Both sit exactly on their train's own track: p90 zero, nothing off the run, nothing missed, worst calling station a hundred metres. Found by the atlas’s own audit sweep.
Two more from the measured lists, both of them the largest single gap left in their country. The Dutch diagonal runs Zwolle to Roosendaal in two and three quarter hours, down the IJssel through Deventer and Zutphen and across Brabant, on the state railway the act of 1860 ordered built: a hundred and sixty eight of its two hundred and fourteen kilometres had no line of ours near them. The Belgian one crosses Flanders corner to corner through the middle of Brussels, Dendermonde to Kortrijk and the French border beyond, over the Halle to Tournai line of 1842: a hundred and fifteen of a hundred and forty eight uncovered. Both come out exactly on their train's own track, p90 zero, nothing off the run and nothing missed. Found by the atlas’s own audit sweep.
Albania was checked and deliberately left empty, which is why it still reads zero per cent. It has three hundred and ninety kilometres of railway and ten route relations in OpenStreetMap, and the operator's own timetable page carries one image where the departure boards should be: a photograph of a train on a viaduct, captioned ON MAINTENANCE, standing in for all three of its services. A country whose railway is not currently running passengers does not get a card, and the honest way to record that is to leave the map blank and say so here. Found by the atlas’s own audit sweep.
Belgium and Germany, measured the same way. Belgium's biggest hole is the Sambre: Intercity 19 from Namur through Charleroi, La Louviere and Mons and over the French border to Lille runs a hundred and forty nine kilometres, and a hundred and thirty nine of them had no line of ours anywhere near them. It is the industrial spine of Wallonia, opened in 1843 to move coal and steel down the valley, and it is now drawn. Germany's is Lusatia: three hours and ten minutes from Leipzig to the Polish border at Frankfurt an der Oder by way of Torgau, Falkenberg, Cottbus and the Neisse at Guben, two hundred and thirty six kilometres of which two hundred and thirty seven of the relation's two hundred and forty one were uncovered. Both verify at p90 zero against their own trains. Found by the atlas’s own audit sweep.
One German railway was measured and held back rather than drawn. The Dresden to Chemnitz to Hof line is the single largest gap the sweep found in Germany, two hundred and twelve unbroken kilometres with nothing on them, and the planner answers Dresden to Hof with a replacement bus as far as Chemnitz, in November as well as today, because that section is closed for electrification. The railway is there and the card is not, because a card here spans a service and the service is currently half a bus. It goes on the list for when the wires are up. Found by the atlas’s own audit sweep.
Then west, where the atlas is fuller and the gaps are subtler, and where a name test is no use at all: a dozen Dutch and Belgian Intercity series run over each other's track, so half of them would be new cards and half the same railway again. The question was made geometric instead. For each candidate relation, how many of its kilometres have no atlas line within two hundred and fifty metres of them. Intercity 2700 from Maastricht to Alkmaar scores zero uncovered kilometres of two hundred and sixty nine, because the atlas already draws that railway; the Maaslijn scores eighty six uncovered of ninety, because it draws none of it. Found by the atlas’s own audit sweep.
Three came out of that, and all three verify as being exactly on the track their train uses. The Maaslijn is the one long Dutch railway that runs north to south rather than in and out of the Randstad, Nijmegen down the east bank of the Meuse to Roermond in an hour and eighteen, single track and still not electrified. The Aberdeen to Inverness line crosses Scotland the other way round, a hundred and seventy four kilometres through Huntly, Keith, Elgin and Nairn on a railway finished in 1858 and mostly single to this day. The Esk Valley line runs an hour and a half from Teesside over the North York Moors and down the river to Whitby, on rails that reached the harbour in 1835 and were worked by horses for their first ten years. Found by the atlas’s own audit sweep.
Romania and Czechia next, and both gaps turned out to be the same shape: a country drawn as spokes from its capital, with the cross country line missing. Romania now has the sixteen hour run from the Banat to Moldavia, Timisoara to Iasi, which never goes near Bucharest: west through Lugoj and Faget, across Transylvania at Deva and Cluj, over the Eastern Carpathians at Vatra Dornei on a crossing finished in 1938, then down through Bukovina to Pascani and Iasi. Its OpenStreetMap relation is in twenty nine disconnected pieces and no path could be found inside it, so this one is drawn from the shape the planner itself returns for the train, and measured afterwards: eight hundred and six kilometres, every calling station within a kilometre and a half of the line, and the whole of it on real track at a median seven metres. Found by the atlas’s own audit sweep.
Czechia had no card at all for the Elbe, which is one of the busiest railways in Europe and the prettiest hour on any train out of Prague. The gap is now filled by the through working from Budapest to Berlin: eleven hours and four countries, up the Danube Bend to Szob, across the border at Sturovo, through Bratislava, Brno and Prague, and then along the sandstone gorge at Bad Schandau on a line the Saxons and the Austrians opened together in 1851. Nine hundred and ninety two kilometres drawn from the relation, and every one of its fifteen calling stations within three and a half kilometres of it. Found by the atlas’s own audit sweep.
Poland came next on the same scorecard, at 29,431 km of railway and thirty nine per cent of it drawn, and it produced one card and one refusal. The Omulew crosses Masuria the short way, Olsztyn to Bialystok in four hours without going near Warsaw, and the middle of it is a single track the Prussian army laid across the Narew in five weeks in the summer of 1915 to move troops to the Eastern Front; the section north of Wielbark had opened as an ordinary branch line in 1900. Every one of its calling stations sits within three and a half kilometres of the line as drawn, and the line came from the relation Polish mappers wrote. Found by the atlas’s own audit sweep.
The refusal is the more useful half. OpenStreetMap has a relation for the Flisak, the eleven hour run from the Silesian coalfield to the sea, and the obvious story to tell about it is the Coal Trunk Line that the Second Polish Republic built between 1926 and 1933 to carry Silesian coal to Gdynia without crossing German territory. The train does not run on it. Asked for the Flisak's own calling list, the planner answers Czestochowa, Radomsko, Piotrkow, Lodz, Kutno, Plock, Sierpc, Torun, Grudziadz, Kwidzyn, Malbork, Tczew, which is the lower Vistula and not the coal main line at all, and the relation itself disagrees with that list twice over about a hundred and thirty kilometres. A card was authored, drawn, measured against the train, and deleted. It is cheaper to check a calling list than to publish a railway that is not there. Found by the atlas’s own audit sweep.
Coverage was being judged from OpenStreetMap, which is an inventory of what somebody has mapped rather than of what a country runs, so the other side of the comparison was fetched: the national operator's own network map for every European country that has a railway. Thirty eight of them, thirty from the railway or its infrastructure manager. ÖBB's Streckennetz dated 14 December 2025, DB's ICE/IC Liniennetz for 2026, NS's Spoorkaart, CFR's Harta generală rețea 2026, PKP PLK's map of every Polish line, SBB's Netzkarte, National Rail's all stations map, Ukrzaliznytsia's own passenger scheme. Then every kilometre of railway OpenStreetMap knows in Europe was measured against every line this atlas draws, at the two hundred and fifty metre threshold the on-rails audit already uses, and the answer is a scorecard rather than an impression. Ukraine is the worst covered real network on the continent: 26,221 km of railway, 5,713 km of it with a line of ours on it, twenty one point eight per cent, from ten cards, nine of which run out of Kyiv. Germany has more uncovered kilometres in absolute terms and a hundred and fifteen cards to Ukraine's ten, which is a different kind of gap and a later problem. Found by the atlas’s own audit sweep.
So Ukraine went first, and three trains that run today were added from relations Ukrainian mappers had already drawn. The Intercity+ from Kyiv to Przemyśl is the first Intercity+ ever run across a border, since 23 December 2016, and since 2022 it is how most of the country reaches the European Union: seven hundred and nineteen kilometres through Vinnytsia, Khmelnytskyi, Ternopil and Lviv in nine hours, and the drawing sits on the train's own track to within a tenth of a kilometre along its whole length. The Svitiaz is the overnight named train to Volhynia, out of Kyiv at 22:50 into Kovel, whose junction the Russian Empire opened in 1873. The Mykolaiv is the named train to the Black Sea shipyards, arriving at a station opened in 1908, and Ukrainian Railways own timetable says it has run daily right through the war. The Svitiaz card is twenty five kilometres shorter than the service, because the relation draws the through line while the train runs into Lutsk and back out again, and that shows in its provenance as a miss run rather than being hidden. Found by the atlas’s own audit sweep.
Two more Ukrainian relations were refused, and the reason is worth writing down because it is the whole point of asking an operator rather than a map. OpenStreetMap carries a relation for train 109, Kherson to Lviv, checked by its mapper in March 2024, and one for train 351, Chișinău to Kyiv, tagged with a nineteen hour fifteen minute journey time. Neither runs. Ukrainian Railways own station board for Kherson lists four trains and all four are the shuttle to Mykolaiv, and the planner offers nothing direct from Chișinău to Kyiv at all. A relation is a drawing of a railway, and a drawing outlives the service on it. Fifteen more Ukrainian relations in the same inventory are Russian trains into occupied Crimea, and those are not services this atlas will draw either. Found by the atlas’s own audit sweep.
The list that decides what gets mapped next could not be read, and that is why some of the largest railways on Earth were not on this map. It is built from OpenStreetMap, which is written by the people who live along the line, so the top of it arrived here as 京哈高速线, 滨洲线, 兰新线, Түрксіб and سكة حديد الحجاز. Every tool in this atlas that asks whether a railway is already drawn asks it by comparing names (the OpenStreetMap relation says it runs from here to there, a card says it runs from here to there, and where the two agree the railway is covered), and that comparison cannot run at all when the two are not written in the same alphabet. The damage was worse than an unsortable list. Taiwan's high speed railway arrived as two hundred and six separate missing services, one per train number, for a corridor this atlas has drawn since it was written; every Korean KTX pattern read as a discovery; and the genuine gaps sat in the same undifferentiated pile, so they could not be found either. The fix is not a romanisation of ours, which would only add a fifth spelling of every station to argue with. OpenStreetMap's own mappers have already written the English, in the name:en tag, and there is a great deal of it: 351,305 distinct names, read from 549,300 tagged stations, towns and villages, downloaded from the OpenStreetMap planet in three queries. Reading every name through that before any test runs took the unreadable pile from five hundred and twenty six services to fourteen, and collapsed the whole list from 1,281 rows to 1,127 as the duplicates finally recognised each other. The native name is still printed beside the English one, because the page you will open next says the native one. Found by the atlas’s own audit sweep.
Three railways came straight off that list, and each of them is drawn from the relation whose name had been hiding it. 宗谷 is the Sōya, the one return working a day from Sapporo to Wakkanai, Japan's northernmost station, five hours and ten minutes up the island in sight of Sakhalin at the end; two hundred and forty two of its four hundred and seven kilometres had no line of ours within two and a half kilometres of them. 滨洲线 is the Binzhou Railway, the west arm of the Chinese Eastern Railway, running since January 1902 from Harbin over the Greater Khingan to the Russian border at Manzhouli, where the Trans-Manchurian crosses to Zabaikalsk: nine hundred and thirty five kilometres, of which one thousand three hundred and five of the one thousand eight hundred and sixty nine mapped kilometres of its double track were uncovered, the biggest single absence the sweep found. 银兰高速线 is the Yinlan high speed line, whose two halves opened three years apart to the day, on 29 December 2019 and 2022, and which brought Yinchuan to Lanzhou down to two hours and fifty six minutes. All three were then measured the way everything here is: 100% of each line on real railway, and a median four, five and seven metres from OpenStreetMap's own track. None of them carries a timetable claim, because the planner this atlas is allowed to ask does not answer for Japan or China, and the journey times on those three cards are the operators' published ones rather than something measured here. Found by the atlas’s own audit sweep.
One number on the Yinlan card needs explaining, because it disagrees with every published figure and it is not a mistake. The line itself is 431 km, Yinchuan to Zhongwei to Lanzhou, and that is what China's railway administration announces. The card says 473, because a card here spans the journey rather than the line, and the last forty six kilometres of the run into Lanzhou West are not on the high speed line at all: they are on the Lanzhou–Zhongchuan Airport intercity railway, which the trains join at Shuping. The drawing says so in its own provenance (427 km from relation 8355929 and 46 km from relation 6143691), and the two add up to the number on the card. Found by the atlas’s own audit sweep.
A rule this atlas has followed since it was written was retired today: draw the classic trunks beside their high-speed twins wherever it is correct to do so. The old rule was that a railway is drawn once: where a high-speed line and the classic railway beside it shared a corridor, only one of them got a line, on the reasoning that two lines through the same geography teach nobody anything. What was wrong with it is that a classic trunk and its high-speed twin are not one railway. The Jingha and the Jingha high-speed line were built ninety years apart, run through different towns and call at different stations, and drawing only the newer one leaves a reader with no idea that the old one is still there carrying the overnight trains. So the unit is the RAILWAY now, not the corridor. What is still refused is the same TRACK drawn twice (two operators over one line, or a card whose whole path is another card's path), because that one is not about tidiness: a second line drawn on top of the first says there are two railways where there is one. The tool that finds what to map next was measuring at two and a half kilometres, which is a question about corridors; it measures at two hundred and fifty metres now, which is a question about railways. Found by the atlas’s own audit sweep.
Three of China's great trunk railways went on the map beside the high-speed lines built to relieve them. The Longhai runs 1,759 km from the Yellow Sea at Lianyungang to Lanzhou; it was begun at Kaifeng in 1905 and took until 1953 to reach its western end, and the Zhengzhou–Xi'an and Baoji–Lanzhou high-speed lines this atlas already drew are the two newest attempts to take the traffic off it. The Jingha runs Beijing to Harbin, and its Tangshan–Tianjin section opened in 1881 as the Kaiping Tramway, which makes it the oldest railway still working in China; the Z15 runs the whole 1,249 km overnight in ten hours and four minutes and the line drawn here comes to 1,246. The Lanxin climbs Wushaoling at 3,000 m, runs the length of the Hexi Corridor past Jiayuguan and reaches Ürümqi through Turpan; it opened in December 1962 and it is what first joined Xinjiang to the Chinese network. Each one now sits beside its high-speed twin on the map, which is the point. Found by the atlas’s own audit sweep.
Drawing them needed a new instrument, and why is worth writing down. Every line here that comes from an OpenStreetMap relation is drawn by welding that relation's ways into a graph and walking the path across it between the card's two ends. That needs the relation to be connected, and China's trunk relations are not: the Longhai is 2,872 ways in 103 disconnected pieces, and the tool answers "no path inside the relation" for Lianyungang to Lanzhou, while answering perfectly for the 505 km from Zhengzhou to Xi'an inside the same relation, 39 m from its nodes. The geometry is all there. Only the connectivity is missing. So the new tool needs none of it: it walks a guide line every 250 m and takes the nearest node of the relation, which makes the guide the load-bearing part. The guide is each line's own station chain, read from Wikidata: 149 stations for the Longhai, one every twelve kilometres, each of them a station OF that railway. That is what keeps it off the high-speed twin, because the twin has its own separate stations: the nearest Longhai node to Luoyang station is on the Longhai, not on the high-speed line twenty kilometres north. Eighty-nine, ninety and seventy-two per cent of those three walks found track. It is honest about what it is: these lines were MATCHED to a relation, not drawn from one, so not one of them claims on its page to have been drawn from an OpenStreetMap relation. Found by the atlas’s own audit sweep.
The same question found an error in a card that has been on this map for months, and when this was written it could not be fixed. The Jingguang High-Speed card, Beijing to Guangzhou, is drawn from OpenStreetMap relation 193064, and 193064 is 京广线, the CLASSIC Beijing–Guangzhou railway, not the high-speed line beside it. 4,417 of that relation's 4,541 mapped kilometres lie within 250 m of the card's own line, which is not proximity, it is the same railway. So a card that says 350 km/h and 2,298 km is drawn along a 160 km/h line built through different towns. The repair needs the high-speed line's own relation, 12265072, and that one is mapped with holes: there is no path through it from Beijing Fengtai to Guangzhou South, and a probe of the track near Chenzhou finds ways that belong to no relation at all. It is also the reason there is no classic Jingguang card in this batch. The classic railway is already drawn on this map, under the wrong name. Found by the atlas’s own audit sweep.
The Jingguang High-Speed card is on the high-speed line now. The item above reported it drawn along 京广线, the classic Beijing–Guangzhou railway beside it, and said the repair was impossible because the high-speed line's own OpenStreetMap relation has holes in it and no path can be walked through it end to end. That was true when it was written and stopped being true a few hours later: the tool written for the Chinese trunks needs no path at all, only a guide to follow, and the guide here is the high-speed line's own 43 stations, which are different places from the classic line's. Snapped to relation 12265072 the card now runs Beijing Fengtai to Guangzhou South on the railway it is named for. The measure of the change is what those two relations now say about us: the high-speed line went from 1,875 of its 4,195 mapped kilometres with no line of ours on them to 398, and the classic line went the other way, from 119 to 2,517. One number on the card moved with it: 2,298 km was the old Beijing West figure and the line is 2,230, which is what Beijing Fengtai to Guangzhou South measures. Found by the atlas’s own audit sweep.
And the Lanxin, added this morning, was drawn with a two hundred kilometre detour in it that nobody would have noticed from the map. Between Lanzhou and Wuwei it measured 485 km where the railway is 277, because it ran west to Xining and came back. The cause was one station: the guide for that line is read from Wikidata, which lists a station on the Lanzhou–Xining railway as being on the Lanxin, and a guide is followed rather than judged. Dropping any station more than 60 km off the line's own axis removes it, and the redraw lands where the mileposts say it should: Zhangye at 494 km against a published 496, Jiayuguan at 717 against 717, Ürümqi at the end. The share of the line that could be matched to real track went from 72% to 93% at the same time, because a guide that follows the railway finds it. Found by the atlas’s own audit sweep.
The same repair was owed on the other pair, and it was owed to this morning's own new card. Beijing to Harbin has a high-speed line and a classic one on this map, and the high-speed card was drawn on neither of them cleanly: 1,366 of the high-speed railway's 2,426 mapped kilometres had no line of ours on them, while 385 km of the new classic Jingha card ran within 250 m of it: two cards drawn on top of each other on a railway that has two. Snapped to relation 3066901 through its own 33 stations, the high-speed card now covers all but 3 km of that relation, and the classic card has the classic railway to itself: 2,149 of the Jingha relation's kilometres are its, and only 93 are the high-speed card's. Neither line moved because a rule changed. They moved because the map was saying something untrue about which railway each train uses. Found by the atlas’s own audit sweep.

20 August 2026

You can now see the railway itself under every line on this map, and judge us with it. The switch is called OpenStreetMap railways and it draws seven hundred and twenty six thousand kilometres of track underneath the routes: the same OpenStreetMap relations these lines were built from, plus five hundred and forty nine railways this atlas has no route for at all. Where one of our lines sits on the grey one it is right, and where it wanders off, it is wrong and you can see exactly where and by how much. Hovering a grey line names the relation it came from and links it, so checking a claim is one click rather than an act of faith. This exists because of two pieces of criticism that could not be answered with a number. jimkats, on the OpenStreetMap community forum, read these routes as having been traced off rendered map tiles. alan_gr, on the same forum, found the Madrid to Murcia line drawn down the conventional railway instead of the high speed one beside it. Both were reasonable readings of a map that asserted its own accuracy and offered no way to test it. jimkats and alan_gr on the OpenStreetMap community forum.
The Hiawatha was on the wrong railway between Chicago and Milwaukee, and the new layer is what made it obvious. The card has always said one hundred and thirty eight kilometres, which is right (it is eighty six miles and Amtrak has run it for decades), but the line drawn on this map was one hundred and sixty one, because for fifty eight of those kilometres it ran down a different railway altogether, a median of two and a half kilometres away from the track the train actually uses and nine kilometres away at the worst point. Amtrak's own service has a relation in OpenStreetMap, listing the ways it runs over, and asking which relations serve the stations at each end of the card found it immediately. Drawn from that relation the line comes to one hundred and thirty eight kilometres, matching the card's own figure to twenty metres, and the measurement afterwards went from 73% of its length on real track to 100%, from two and a half kilometres off the rails at the median to two metres. Found by the atlas’s own audit sweep.
Three more lines were repaired and two repairs were thrown away for measuring worse, which is the same rule as last time pointed at fresh work. The Pearl of the Caucasus cruise train, the worst drawn line on this atlas, had seven stretches replaced with real track and went from 52.8% of its length on a railway to 56.0%, the median distance falling from two hundred and twenty nine metres to one hundred and ninety seven. It is still the worst line here and it says so on its own page. The Vivek Express gained a point and Helsinki to Joensuu a fraction with its median distance from the rails cut from fourteen metres to eleven. Against that, the Deccan Odyssey repair changed not one number in the measurement and Helsinki to Jyväskylä came back further from the railway than it started, 91.1% to 90.1%, so both were put back exactly as they were. Found by the atlas’s own audit sweep.
The comparison layer now draws every railway OpenStreetMap knows, not only the ones this atlas has looked at, and it says which is which. It had been built from the relation cache (the railways the geometry pipeline has fetched), which answers a question about this atlas rather than a question about the world. A reader who switched it on over Kansas or Kazakhstan and saw nothing could not tell whether we had drawn no route there or whether there was no railway there, and those are very different admissions. So the third tier is the planet: one and a half million railway ways, fetched from the OpenStreetMap planet through QLever, welded into a hundred and forty thousand continuous chains, and drawn in the faintest of three greys. The other two are the ones that carry evidence. The brightest is track a route here is drawn from, which is what is left on screen when every route is switched off. The middle one is a railway somebody has described end to end in OpenStreetMap that this atlas has no route for yet, which is where the next routes come from. Both of those name their relation when you hover them; the third cannot, because it is welded from individual ways and the only honest thing to say about a stretch of it is that it is railway. Together they draw 1.73 million kilometres. Found by the atlas’s own audit sweep.
Madrid to Murcia was drawn from the wrong train, and the relation naming the right railway had been in OpenStreetMap the whole time. alan_gr reported the line overshooting the point where the Alicante branch leaves and then taking a sharp turn east onto something that is not a track at all, and he was exactly right about the shape and about the cause. The card was drawn from relation 4777350, which is called AVE Madrid - Alicante: a real service, correctly mapped, going to a different city. So the line followed the Alicante train past the junction at Monforte del Cid and then had to invent a way across to Murcia, which is why it measured five hundred and forty four kilometres against a card that has always said five hundred and twenty nine, and why two of its probe points had no railway within three hundred metres of them. What it should have been drawn from is relation 18342974, Línea de alta velocidad Madrid-Murcia, the line's own record under its ADIF number 046. Spliced from that, the eighty three kilometres past the bifurcation are OpenStreetMap's own nodes to within a metre, the line comes to five hundred and twenty nine kilometres exactly, the probe points with no railway near them are gone, and Renfe's own timetable now puts every station this train calls at within sixty two metres of the drawing rather than a hundred and five. alan_gr on the OpenStreetMap community forum.
Two more routes came off that list, and the way they were chosen matters more than the two of them. Ninety two railways is not a work list, because most of what OpenStreetMap holds cannot be drawn from: a relation is often mapped in disconnected pieces, and welding across a hundred kilometre hole would be a confident fiction. Asking that question of all ninety two leaves twenty seven that describe one continuous railway, and those are the only ones a card can be built from without inventing geometry. Then the service has to actually run, and the atlas has an instrument for that: ÖBB Scotty, the same planner every timetable claim on this site rests on. So: Joensuu to Nurmes, a hundred and sixty kilometres of the Karelian line up the shore of Lake Pielinen, drawn from the relation, lying on real track to three metres, and confirmed by Scotty against train R 761, which also corrected the journey time from the two and a half hours first written down to the two hours and one minute the timetable actually offers. The card stops at Nurmes because the trains do; the rails carry on to Kontiomäki but passenger service north of Nurmes ended in 1993. And Saint Petersburg to Labytnangi, fifty two hours to the last station on the Gulag built Dead Road, the Chum to Labytnangi branch that opened at the end of 1948 and stops on the bank of the Ob across the water from Salekhard. Found by the atlas’s own audit sweep.
Two candidates were built, screened and thrown away, and that is the part worth writing down. Banja Luka to Doboj is ninety three kilometres of electrified railway that Željeznice Republike Srpske list as a daily service in the timetable now in force, and it has a clean relation with no line on it. Chelyabinsk to Kaliningrad is a real seasonal train running this very week. Both were drawn, and both were then asked the only question that decides whether a card can carry a claim, and neither could answer it: Scotty knows no station near either end of either route, so this atlas has no way to check that its line is the train's line. For Bosnia there was no reachable timetable to take a journey time from either, and a card here does not get an invented one. They are candidates still, and the work each needs is now known rather than guessed at. Found by the atlas’s own audit sweep.
And the same comparison, run the other way round, found railways this atlas draws nothing on. Overlaying every relation in the geometry cache on every line the map draws and asking which kilometres have nothing on them turns up ninety two distinct railways with an unbroken stretch of sixty kilometres or more that this atlas has never touched: the Abidjan to Ouagadougou railway at six hundred and fifty five, a Vande Bharat between Bangalore and Hyderabad whose six hundred and twenty five kilometres are uncovered end to end, the Kirandul line through the Araku hills, the long haul from Aktobe to Almaty. The first one added from that list is the southwest arm of the VIA Rail Corridor, Toronto to Windsor: four Corridor cards met at Toronto and every one of them left eastward, so the railway through Brantford, London and Chatham had no line on it. It is drawn from the relation that describes the service and sits three metres from the rails. Found by the atlas’s own audit sweep.

19 August 2026

Two of the worst lines on the atlas were repaired by asking a question that had been written down for weeks and never read. The check that measures how far each line sits from a railway works by routing along real OpenStreetMap track between points taken off our own drawing, and when it cannot reach any track at all it records the exact coordinate where that happened. Four hundred and three of those coordinates were sitting in the cache, describing every stretch of this atlas that runs across open ground, and no tool had ever looked at them. There was already a repair tool that routes a broken stretch along real track and splices the answer back; it was simply pointed at a different defect, the long straight lines left by sparsely placed points, and it had run out of those in July. Pointing it at the off-rail coordinates instead changed nothing about how it draws and everything about what it finds. Japan's Train Suite Shiki-shima went from 28.3% of its length on real track to 91.7%, and the median distance from a railway fell from five hundred and sixty six metres to forty five. The Mumbai to Ahmedabad Vande Bharat went from 33.6% to 84.8%, and from six hundred and seventy five metres to nine. Across thirty routes the average went from 85.3% to 90.0% and the number of lines the atlas admits are off the rails fell from fifty to thirty seven. Found by the atlas’s own audit sweep.
Two repairs were measured, found to be worse, and thrown away. The Stockholm to Åre night train and the Madrid to Murcia AVE both came back very slightly further from the railway than they started (96.0% to 94.7% and 99.0% to 97.5%), and the second is a known and specific failure rather than bad luck: the router prefers the classic line wherever a high-speed railway runs beside one, which is exactly the case in Murcia. Both were restored to what they were. This is written down because it is the whole reason the repair is measured after it is applied rather than trusted because it looked reasonable, and because a tool that improves thirty lines and quietly damages two is not a tool anybody should keep using. Found by the atlas’s own audit sweep.
Sixteen routes lost the claim that they are drawn from a named OpenStreetMap relation, and that is a real cost of the repair rather than a bug. That claim is the strongest provenance on this site (it means somebody who knows the railway listed the exact ways it uses and the atlas took their nodes), and it is attached to a specific drawing. Repairing a stretch of one of those lines makes the old claim a statement about geometry that no longer exists, so it is dropped rather than carried over. The rest of each line is unchanged; the label can be earned back by re-reading the relation against the repaired drawing, which is work for another day and is noted here so the count is not a mystery. Two lines that had never been measurable at all now are: the Lagos to Ibadan standard gauge line reads 100% on track, and Nepal's Janakpur railway 70%. Found by the atlas’s own audit sweep.
Fifteen of those sixteen routes have their relation back, and two of the fifteen turned out not to be bookkeeping. Re-reading a relation against a repaired line is mostly clerical: the stretch it speaks for is drawn again from its nodes, the record is written against the drawing that now exists, and nothing moves by more than a few metres. Thirteen went through in one sweep and two needed the railway named explicitly, because the highest scoring candidate was not the one the card describes. The Marine Liner was different. It came back offering a relation it had never been drawn from, the service's own listing rather than the bridge line it runs over, and drawn from that the Okayama to Takamatsu card went from 89.0% of its length on real track to 100%, six metres from the rails at the median, with the three kilometre stretch that ran clear of any railway gone. Found by the atlas’s own audit sweep.
And the Lagos to Ibadan train had been drawn on the wrong railway. Its last ten kilometres into Ibadan ran along the 1,067 mm colonial line that shares the corridor, which is mapped, real and carries trains, so every measurement this atlas ever took of it said 100% on track and was right to. The card is named for the standard gauge railway that opened in 2021, and that is a different set of rails. What settled it was the gauge in OpenStreetMap's own tags: 1,435 millimetres on the ways the relation lists, 1,067 under our drawing. The line now runs where the card says it does, all 156 kilometres of it from the relation's nodes. Found by the atlas’s own audit sweep.
The sixteenth was put back where it was. The Deccan Odyssey is a cruise train touring a dozen railways rather than a service on one, and the only relation describing any part of it is the Konkan Railway down the coast. Drawing five hundred and thirty seven kilometres of the tour from it read perfectly well and measured worse: 92.3% of the line on real track before, 90.9% after, with seventy eight kilometres of the itinerary swallowed by the coastal alignment. So the drawing was restored, and that card still says nothing about where its line came from, because nothing here has earned the right to say it. Same rule as the two repairs thrown away above, pointed at the opposite kind of work. Found by the atlas’s own audit sweep.
A line the rail router cannot drive was reporting nothing at all, and the reason was a rule that only listened to the second opinion when it disagreed. Where the router has no graph (a railway that opened in 2021, a rack line, a sixty centimetre gauge forest railway), this atlas asks OpenStreetMap directly what track lies under the drawing, and where the two answers differ the direct one wins. The Lagos to Ibadan line had eleven of its sixteen legs refused outright, far too little of it for the router to speak for, and the direct measurement said the whole line lies on mapped track. Because that agreed with the five legs the router had managed, it was thrown away as adding nothing, and the page ended up showing no measurement whatsoever. The direct answer is now taken wherever the router covered less than a third of a route, unless a coordinate on that route has been confirmed to have no railway near it. Two pages gained a measurement they should have had all along: Lagos to Ibadan at 100% and three metres, and the Darjeeling toy train at 100% and eight. Found by the atlas’s own audit sweep.
The Yangon Circular Railway was the worst drawn line on the atlas, and the reason was a question nobody had noticed was unanswerable. Every way this map draws a route from OpenStreetMap asks the same thing: weld the railway's own ways into a network and find the path across it between the card's two ends. That card runs from Yangon back to Yangon. The path between a point and itself is nothing, so the answer came back as zero per cent of the declared distance, and when the tool was told to ignore our endpoints and take the railway from one end of itself to the other, a closed ring has no two ends either; it picked the two furthest apart and returned exactly half the loop. Meanwhile OpenStreetMap's relation for the line had been right the whole time: ninety five member ways, tagged with the railway's own Wikidata identifier, mapped by somebody who knows it. The drawing now walks the ring once, out from the station and back the long way round, and comes to 46.1 kilometres against a card that claims 46. It is fifty one points where it was ten, and it sits five metres from the nodes OpenStreetMap stores. Measured against the track afterwards it went from 8.3% on rail to 100%, the median distance from a railway fell from six hundred and thirteen metres to three, and the longest stretch running clear of any track, seventeen and a half kilometres of it, is gone. Nine other cards in this atlas also start and end in the same city, but every one of them is a cruise train touring a dozen separate railways rather than a circular service on one, so none of them is fixed by this and their pages still say what they are. Found by the atlas’s own audit sweep.
Five more lines are drawn from the OpenStreetMap relations that describe them (the Vivek Express, the Coast Starlight, the Chengdu to Kunming railway, the Seto-Ohashi Marine Liner and Helsinki to Jyväskylä), and it is worth being plain about what that did and did not achieve. Where a relation covers only part of a card the repair is a splice: the stretch it speaks for is replaced and the rest is left alone. The Coast Starlight's relation speaks for two hundred and ninety kilometres of a two thousand two hundred kilometre journey. So these lines are now finer where they were already right, sitting one to twelve metres from OpenStreetMap's own nodes instead of being approximations of them, and four of the five keep the same honest verdict at the bottom of their page, because the stretches that run off the rails are precisely the stretches no relation covers. Helsinki to Jyväskylä improved from 88.3% on rail to 90.2%. The others did not move, and saying they did would be easy and wrong. Found by the atlas’s own audit sweep.
Four proposed repairs were read and thrown away, which is the part of this work that never shows. A tool that scores OpenStreetMap relations against a card will always return its best candidate, and the best candidate is regularly a different railway that happens to share both endpoints. The highest scoring relation for the Hong Kong to Beijing sleeper is the classic Beijing to Guangzhou line rather than the high-speed one the card describes; for Fuzhou to Xiamen it is the conventional railway opened in 2010, not the high-speed line of 2023. The only candidate for Japan's Train Suite Shiki-shima is the Akita Shinkansen's Komachi service, which is a different train doing a different thing. The candidates for the Mumbai to Ahmedabad Vande Bharat are a Rajdhani Express from Delhi, the Ahmedabad metro's Blue Line, and the bullet train corridor that is still being built. All four lines were left as they are. Applying any of them would have moved the drawing onto a real railway, which is exactly what makes the mistake hard to see afterwards. Found by the atlas’s own audit sweep.

18 August 2026

Two Czech expresses join the atlas, and the reason there are only two is worth saying. The list of services OpenStreetMap knows and this map does not draw runs to 1,111 entries, which sounds like years of work and is not: 172 of them are Taiwan's single high-speed line filed once per train number, and most of the rest are domestic line numbers rather than journeys. What survived was small. The R16 climbs from Prague through Plzeň and Klatovy to Železná Ruda-Alžbětín, a station shared with Bavaria across the border, on a line whose last section opened on 20 October 1877 and whose Špičák tunnel stayed the longest in the Czech lands for the next 130 years. The Ex7 runs direct from Prague to Český Krumlov, over rails laid in 1891 to carry Schwarzenberg timber and graphite out of the south, long before anyone went there to look at the town. Both were drawn from their own OpenStreetMap relations, and both journey times come from the ÖBB planner rather than from an estimate: three hours thirty five, and two hours thirty. Neither card claims a top speed, because no figure for either could be verified and inventing one is not allowed here. Found by the atlas’s own audit sweep.
India's longest train had eleven waypoints across Punjab and then a straight line for forty eight kilometres. The Himsagar Express runs 3,785 km from Jammu to Kanyakumari, and most of it was drawn properly from OpenStreetMap, but the stretch coming down from Jammu was a sketch: ten sparsely placed points, one jump of forty eight kilometres, and thirty five kilometres of the drawing running clear of any railway. Twelve of those straight runs have now been routed along the real track and spliced back in, which is a different operation from redrawing the line and a necessary one: asked to redraw the whole route the repair tool refused, correctly, because routing all 3,785 km end to end came back fourteen per cent too long. Only the straights were touched. The line is 1,991 points instead of 1,361, sits nine metres from the rails at the median where it used to sit twenty one, and its longest stretch away from track fell from thirty five kilometres to seventeen. It is still not perfect and the page says so. Found by the atlas’s own audit sweep.
The Trans-Siberian was drawn with one point every ten and a half kilometres, and a straight line between two points ten kilometres apart does not follow a railway that bends. A hundred and six kilometres of it ran off the rails in one continuous stretch, the worst geometry error on the whole atlas, and it was not the map that was wrong: the line came from OpenStreetMap's own relation for the Transsib and was then thinned to fit a budget of nine hundred points per route. That budget was set on the reasoning that a nine thousand kilometre line cannot be drawn finely at any price worth paying, which nobody had ever costed. Costed now: the whole atlas ships as one download, and giving seventeen over-long routes the points their own length asks for adds five kilobytes to it, about seven tenths of one per cent. So they have them. The Trans-Siberian is drawn with 3,228 points instead of 874 and sits a hundred and seventeen metres from the rails where it used to sit six hundred and seventy five; measured against the track afterwards it went from 60% on rail to 99.9%. The Canadian, the California Zephyr, the Indian Pacific, the Empire Builder, the Southwest Chief, the Sunset Limited, the Silver Star, the Silver Meteor, the Texas Eagle, the Coast Starlight, the Ghan's Great Southern, Beijing to Guangzhou, Moscow to Vorkuta, the Vivek Express and the Himsagar Express were all drawn against the same ceiling and are all finer now. Found by the atlas’s own audit sweep.
Almost every route page can now show how far its line actually sits from a railway, and nineteen of them could yesterday. The claim itself is not new (it is measured by routing along OpenStreetMap track between points taken off our own line, so the path that comes back IS the railway and the distance between the two is the answer), but the measurements had been wiped out wholesale when five hundred lines were redrawn from the relations that describe them, and a measurement about a drawing that no longer exists is worth nothing and is correctly thrown away. Re-measuring restores it for 1,284 of 1,288 routes: 1,106 lie on real track from end to end, and where a line passes, the median distance from the rail is four metres. The tail is the interesting part. A rail router has no graph for a five hundred metre cliff railway, so it used to answer for the funiculars, forest railways and sixty centimetre gauge lines by routing along the tram in the street below and scoring them zero; those 162 routes are now measured against the track OpenStreetMap maps under them instead, and their pages say so, because "track a train can travel" and "track somebody has mapped" are not the same sentence. Found by the atlas’s own audit sweep.
Three lines that were sketches are now drawings. The Petit Train de la Rhune aside, which is still wrong and is left saying so, the worst of them were the Mount Hood Railroad, drawn with four points over sixty nine per cent of its own length; the Mary Valley Rattler, six points cutting every corner of the valley; and the Byron Bay solar train, three points covering five kilometres of a three kilometre railway. All three are short, none has a route relation in OpenStreetMap, and none could be traced by anything here: the tools that redraw a line from a relation need a relation, the corridor tracer needs a hand written list of stations, and the rail router has no graph for lines like these. So the missing tool was written. It pulls every railway way around the line, welds them, and walks the shortest path between the card's own two ends, which where one railway runs alone is the railway itself, node for node. It refuses more often than it draws (the ends must land on track, the length must match the card, and a genuine hole in the mapping fails the route rather than being bridged), and two of the five it was pointed at were refused on exactly those grounds. The three it drew went from five, ten and thirty nine per cent on rail to one hundred per cent, a metre or less from the track. Found by the atlas’s own audit sweep.

17 August 2026

The Walderbahnle joins too, and it is here because a bug was found rather than believed. Five kilometres of 760 mm gauge between Bezau and Schwarzenberg is all that survives of the line that ran to Bregenz from 1902, and the first attempt to draw it came back six and a half kilometres long with a one and a half kilometre hole welded across the middle. The hole was ours. A ten metre scrap of station track sitting in the middle of the railway is nearer the growing line's end than the next real piece is, so it got joined first and everything after it welded to the scrap instead of to the line. Fragments under two hundred metres are now ignored when pieces are joined, and the railway measures 4.98 km against the 5.01 its owners publish. Found by the atlas’s own audit sweep.
And four more, three of which are trains you drive onto. The atlas has never had a place for a car shuttle, which is why none of these was here: Le Shuttle under the Channel between Folkestone and Calais, the Vereina under the Silvretta with cars loaded at Klosters for the Engadine, and the Tauernschleuse through the 1909 tunnel from Bockstein to Mallnitz, hourly, the way people have crossed the Alpine main ridge without the pass since 1920. They are filed as classic services, because whatever the product is, each one is a scheduled public railway and that is what the category means here. The fourth is the EuroCity Wawel, Berlin to Przemysl at the Ukrainian border by way of Wroclaw, Opole and Krakow. The Tauern shuttle was the awkward one to draw: its OpenStreetMap relation is in four disconnected pieces with a seven kilometre hole, so no path exists inside it, and the line had to be found on the rail graph instead. The planner then answered for it by name, ASTB9605, which is the service's own number. Found by the atlas’s own audit sweep.
Six more join them, and getting them drawn is the story. Five are German long distance trains reaching towns this atlas had never carried: Dortmund to Oberstdorf, Gera to Dusseldorf, Leipzig out to the ferry pier at Norddeich Mole, Leipzig to the Baltic beach at Warnemunde, and Hamburg over the Rugendamm to Binz. The sixth is the overnight train from Prague to Warsaw. Every one of them was checked against the operator's own planner before it shipped, and the planner named the exact working in four cases, which is why the Gera card carries a photograph of IC 2152 itself. Two things had to be got right rather than assumed. The Oberstdorf train stopped being an InterCity on 12 July 2026, five weeks ago, when an ICE L replaced it on the same timetable, so the card says ICE and shows the new stock rather than the red locomotive that had run it for years. And the Prague to Warsaw sleeper no longer goes the way OpenStreetMap thinks: the map there still routes it through Krakow, while the train runs by Katowice and the fast Central Trunk line, so the drawing follows the stations it calls at today, spliced onto the Katowice to Warsaw line this atlas had already proved, with the stretch between them put on real track by the router. The line the map would have given us was 29 km from the railway for two hundred of those kilometres. Found by the atlas’s own audit sweep.
Five services join the atlas, found by asking what the railways run that this map draws nowhere. The question had only ever been asked one country at a time, which counts a train crossing a border as two gaps and cannot see either clearly; asked of all forty inventoried countries at once it comes back as one list. From it: the Outback Xplorer, a weekly railcar eleven hundred kilometres from Sydney over the Blue Mountains and out past Parkes to the silver town on the South Australian border. Tshiuetin, twice a week up the Labrador iron road to Schefferville with no highway beside it, on the first railway in North America owned and run by First Nations. The Ravenglass and Eskdale, fifteen inch gauge laid for iron ore in 1875 and running up Eskdale from the Cumbrian coast ever since. The Kandertalbahn, Sunday steam from the Rhine plain at Haltingen to Kandern. And the Matravasut, seven hundred and sixty millimetre gauge out of Gyongyos into the Matra hills. Every line is the mappers' own, taken from the route relation that describes the service rather than traced point by point, and every figure on the five cards is the operator's or the article's. Two of the five carry no top speed, because none of the sources gives one and inventing a number to fill a field is how a hover card starts lying. Found by the atlas’s own audit sweep.
The Berlin to Amsterdam line now sits on its train's own track for the whole of its length. It had been drawn as much as three kilometres away from the path ICE 148 takes on the way out of Berlin, seven kilometres of it wrong between four and eleven kilometres in, and the remaining six hundred and thirty were right, which is exactly the kind of error a single average distance cannot see. Drawn again from the six OpenStreetMap relations the service actually runs over, the line and the train's own shape now differ by nothing measurable across six hundred and thirty eight kilometres. That finishes Germany, and what Germany mostly showed is where the work goes next: of the seventeen cards still worth attempting, thirteen had nothing to weld, because welding a line from the railways that describe it needs at least two of them linked to the card and these have one or none. Switzerland said the same thing, and so did Italy and France. The drawing is no longer the weak part. The link between a card and the railways that describe it is. Found by the atlas’s own audit sweep.
Three changes to what a repair has to survive before it is allowed to ship, each written from something that nearly shipped. The station record, which is the only check that knows what a card is supposed to pass through, is now asked before a redraw is applied rather than at the end of the build. It is what caught the Marseille coast line being moved onto the inland railway, but it caught it after the change had been made and recorded, and it now refuses the change instead. It refuses only calls the line currently reaches, so a card already missing a station can still be repaired. Rebuilding that record can also no longer erase the evidence against a bad drawing: the record is built from stations the line passes, so rebuilding it the day after a wrong redraw would quietly drop every call the wrong line had walked away from and leave the build green over the top of it. Those calls are kept now and the build fails on them until somebody decides which is wrong, the line or the record. And a timetable claim now carries the length of the drawing it was measured against, because every other record here already died when the line beneath it changed and this one did not, which is how a claim about the coast line outlived the coast line. Found by the atlas’s own audit sweep.
The Wuppertal Schwebebahn makes no timetable claim, and the reason is worth saying out loud. It is a suspended monorail, a hundred and twenty years old, run by the city's own transport company on a rail hung above the river. Ask any journey planner how to travel between its two ends and it will answer with a Deutsche Bahn regional train, because one runs along the valley floor beneath it, roughly parallel and close enough that measuring one against the other very nearly passes. It is not the same railway and it is not this card's train, so the atlas says nothing about the Schwebebahn's shape rather than something false about it. Two other cards in this run had the same shape of problem, a pair of railways joining the same two cities, and the honest answer in each case was to claim nothing. Found by the atlas’s own audit sweep.
A five kilometre rack railway up a mountain was one command away from being certified against a train that runs along the valley floor. The Montenvers line climbs from Chamonix to the Mer de Glace, no planner we can read knows a station on it, and the nearest answer the timetable could give was the TER along the valley: two kilometres long against the card's five, ending two and a half kilometres away at Les Praz. The check called it "on the right rails". It said so because every tolerance in it was a fixed number tuned on main lines. A line could sit ten kilometres from the service it was being measured against, or draw five kilometres of track that service never touches, and still pass. On a five kilometre card that is the whole railway twice over. The tolerances now scale with the length of the route, with the old fixed numbers kept as a ceiling so that nothing which passed before has been loosened, and a floor of five hundred metres so a funicular is not failed by noise. Montenvers now reads WRONG CORRIDOR and carries no timetable claim at all, which is the honest answer for a railway no timetable we can read describes. Found by the atlas’s own audit sweep.
Three Norwegian services join the atlas and a fourth changes shape. Local line L1 from Spikkestad through Oslo to Lillestrom, the RE10 from Drammen up the east shore of Mjosa to Lillehammer, and Flytoget from Drammen out to the airport. Each is drawn from the OpenStreetMap relation that describes it rather than traced point by point, and each sits on its service's own path with nothing measurable between them. The fourth is the Vestfold Line, which this atlas ran from Oslo to Skien while the train itself runs from Eidsvoll to Skien. Bane NOR titles the line "Skien - Oslo S - Eidsvoll", the planner books it as one train with no changes, and Norway's national timetable names the service "Eidsvoll-Oslo S-Skien". The card now spans what the train spans, two hundred and forty two kilometres rather than a hundred and seventy, and it kept its old address so nothing that links to it breaks. Found by the atlas’s own audit sweep.
The Marseille to Miramas card was rebuilt from the six relations a TER runs over, and the rebuild was wrong. Two railways join those two cities: the Cote Bleue along the shore through Niolon, Carry-le-Rouet and Martigues, which is the line this card is named after and drawn on, and the direct line inland through Vitrolles. The repair matched on the card's two ends, found the inland train, and moved the drawing as much as eighteen kilometres off its own coast, taking a timetable claim with it. What caught it was the station record: sixteen calls the card used to reach, among them every viaduct on the corniche, stood between three and eighteen kilometres from the new line, and the build refuses to ship a drawing that has walked away from its own stops. The coast line is back exactly as it was, and it now makes no timetable claim at all, because the only service the planner offers between those two cities is the one that goes the other way. Found by the atlas’s own audit sweep.
That claim outlived the drawing it was made about, and that was a hole in how this atlas keeps its records. Every other record here is dropped the moment the line it describes is redrawn, and this file says so in as many words. The timetable claim was the one with no such rule: it survived the revert still stating that our line lay a tenth of a kilometre from the shape of TER 87910, while the line actually stored is sixteen kilometres from that train. A carried claim is now dropped when the same service, measured again against the line as it currently stands, disagrees with the number the claim itself states. Not when a planner merely fails to answer, which must never delete evidence, but when the same shape is fetched and measured and says something else. Found by the atlas’s own audit sweep.
One thousand four hundred and seventy four photographs were crediting "Wikimedia Commons" for work done by a person whose name this atlas already held. The gallery ships each picture's original address, which carries a query string on the end, and the routine that reads a file name out of an address ran to the next slash and kept the query string as part of the name. So it asked Commons about a file that cannot exist, wrote the answer down as no author, and then failed to match the same picture again when a page asked who took it. Both ends are mended and every credit has been fetched afresh. Ninety eight per cent of gallery photographs and ninety nine per cent of the large ones now name their photographer, and the remainder are the ones Commons itself publishes no author for. Found by the atlas’s own audit sweep.

16 August 2026

The check that says a line is on the right railway could not see a defect shorter than a tenth of the line, and Barcelona to Málaga is the proof. It leaves Barcelona down the conventional railway and rejoins the high-speed line forty four kilometres later, as much as eighteen kilometres from the train it names, and it PASSED, because the measurement was a ninetieth percentile, twenty four kilometres of an eleven hundred kilometre card is two per cent, and a percentile cannot see two per cent. It had already been reported once and mended once; the mending redrew the line and left the measurement alone, so it came back. The measurement is now a LENGTH rather than a proportion: the longest unbroken stretch where our line is more than two kilometres from the service's own track, and separately the longest stretch of that service's track our line does not draw at all. More than five kilometres of either is a fault, and it reads the same on a ninety kilometre route as on a four thousand kilometre one. Asked again, Barcelona to Málaga fails at fifty two kilometres off the train and seventy kilometres of it missing, and its claim has been withdrawn until the line is redrawn. The card also named "OGV 6420", which is not a train that runs (the direct service is Iryo's IRY 6020), and that is gone too. Two Irish routes checked in the same run stayed at zero on both numbers, which is the point: the instrument is sharper, not louder. Reported by alan_gr on the OpenStreetMap community forum.
The new measurement immediately found the thing it was built to find, on a different route and in the other direction. The Avant from Málaga to Seville reverses at Córdoba, and the atlas did not draw the reversal: eleven kilometres of the train's own path, a hundred and sixty four kilometres in, existed only in the timetable. Every other number on that card looked perfect, which is exactly how a spur goes missing. The same sweep found the AVE from Madrid to Murcia drawn two hundred and forty seven kilometres off the service that runs it, which had been reported as the line sitting south of Cuenca on conventional track and is worse than that; and Zaragoza to Canfranc measuring a hundred and eighty kilometres against a real two hundred and twenty four, with both ends about three kilometres short of the stations they name. Six other Spanish routes were asked the same questions and answered them cleanly, so a card being drawn from only part of a relation does not by itself mean the line is wrong. It means nothing has checked. Reported by alan_gr on the OpenStreetMap community forum.
A hundred and fifty nine routes have lost their timetable check, and the reason is a licence rather than a mistake. Those checks were made against an aggregator: a service that gathers operators' published feeds and republishes them under terms agreed with those operators, which are not terms this atlas can pass on to anybody reading it or downloading its dataset. The rule against using it was made on the eleventh of August and was only half kept: the code stopped producing such checks, and the file the site actually ships still carried a hundred and fifty nine of them, so the rule was true of the next build and not of the live site. They are gone. Every one of those routes keeps everything else it could show: the OpenStreetMap relation its line is drawn from, how far that line was measured from real track, and its Wikipedia article. A hundred and six of them are in countries ÖBB Scotty answers for and can earn the check back properly, one service at a time. The other fifty three are outside any planner this atlas may ask, and their pages now do what the rest of the world's routes already do: link the operator's own timetable and claim nothing about the drawing. What replaces the aggregator everywhere is narrower and defensible: the operators' own public planners, asked about their own trains, used to CHECK a drawing rather than to serve one. Found by the atlas’s own audit sweep.
Eleven railways now check this atlas against their own timetables, and none of them is an aggregator. ÖBB Scotty answers for Europe and answers for nothing else, which left every American, Canadian, Israeli and Antipodean route on the map with no check at all. What those railways do publish, many of them, is their own timetable on their own website: Amtrak, VIA Rail, SNCF, Renfe, Comboios de Portugal, Iarnród Éireann, SNCB, Entur in Norway, the Swiss federal transport office, Israel’s transport ministry, Metlink in Wellington. Each is read to ask two questions and no others (does this train still run, and does our line lie on its path), and nothing from any of them is republished. No departure time, no fare, no stop list and no shape appears on this site or in the open dataset. What appears is a sentence about our own map with the railway’s name beside it, so a reader knows who was asked. That is the whole difference between an operator and an aggregator: the question is not how open the data is, it is whose data it is. Five hundred and four routes now carry a check, from three hundred and ninety eight this morning. Ireland is the one to look at: thirteen of its fifteen routes verified against Iarnród Éireann’s own published geometry, which is the network a reader on the OpenStreetMap forum has spent a fortnight correcting. Found by the atlas’s own audit sweep.
Two of those railways were being marked wrong by arithmetic rather than by evidence. Where an operator publishes no geometry (Renfe and SNCF both publish stations without it), the only reference available is the line joining the stops, and that is a chain of straight cuts across every curve the railway actually takes. It is always shorter than the railway, by five to fifteen per cent on anything with hills in it, and always in the same direction. Measuring that against the distance printed on the card therefore fails a correct line as a matter of subtraction: Renfe scored three of fifty three. Judged instead on what a feed without geometry actually contains (where the stations are, which are exact coordinates that no amount of missing shape can move), it scores twenty, and each of those pages says in as many words that no shape was compared. Two claims of different strength, worded differently, never merged. Found by the atlas’s own audit sweep.
Every timetable check in the atlas was made again, and a quarter of the ones that existed had never been made at all. Two things happened at once. First, the four hundred routes that already carried a check had earned it under the old measurement, the one that could not see a stretch of wrong railway shorter than a tenth of a line, so each was asked again under the new one. Three hundred and sixty four passed and twenty five did not, and those twenty five have lost their claim. Second, and worse, it turned out a refuted claim could not actually be taken away: the builder only knew which routes had PASSED a sweep, so a route proved to diverge simply kept the sentence it already had, because nothing arrived to replace it. All twenty five would have shipped. A fault now removes a claim, while a planner that errors or cannot judge leaves it alone: evidence against is not the same as absence of evidence, and a bad afternoon on somebody else's server must not strip the atlas of its own proof. Found by the atlas’s own audit sweep.
Then the same question was put to every European route that had never been asked, three hundred and seventy nine of them, and a bug in how this atlas talks to ÖBB Scotty came out of it. Forty four routes came back saying the planner knew no station near where the line ended, including Vienna to Budapest, which is not a plausible thing for an Austrian planner to say. It was not saying it. The request asked for the nearest places to a coordinate rather than the nearest stops, so at Wien Hauptbahnhof the eight nearest things Scotty returned were shops: a stationery chain, a health food store, a bakery. There was no station in the list because nothing had asked for one. Asking for stops returns eight stations at the same coordinate, and Vienna to Budapest now checks out on the railjet to within a hundred metres. The atlas now also prefers a stop a TRAIN calls at, because the nearest stop to a main station is routinely the tram stop outside it, and planning a railway journey from a tram stop is how a reference itinerary picks up legs the card never meant. Four hundred and sixty routes now carry a timetable check, every one of them earned under the current measurement rather than inherited from an older one, and sixty more turned out to have real faults that are now on the list. Found by the atlas’s own audit sweep.
Barcelona to Málaga is redrawn, and the reason it took three attempts is worth writing down. Every tool here that adds a relation to a card decides which stretch of the card that relation speaks for by projecting it onto the line already drawn. That works when the line is roughly right and is exactly backwards when it is not: the relation describing the railway the train really uses lay eighteen kilometres from where we had drawn it, so it projected onto nothing, covered nothing, and was dropped for being too small a section. The wrong line was preventing its own correction, and adding the missing relation to the candidates simply produced the same 520 kilometres of the Málaga half, over and over. So the line is no longer corrected; it is rebuilt. The Madrid to Barcelona high-speed line and the Madrid to Málaga one are welded into a single railway network, the card's two ends are found on it, and the drawing is the path between them: 1,138 kilometres, against the 1,138 that Iryo's IRY 6020 publishes for itself, sitting on that train's own shape to within a hundred metres for the whole run with every station it calls at inside a hundred metres of our line. It still has no timetable check, and that is deliberate: six kilometres near Lleida remain where OpenStreetMap and the planner disagree, which is one kilometre over the threshold, so the drawing improves and the claim waits. Reported by alan_gr on the OpenStreetMap community forum.
The AVE from Madrid to Granada had grown a spur near Toledo, and the cause has a name: the card had been linked to OpenStreetMap relation 15616203, which is "Train Avant: Toledo → Madrid", a sixty three kilometre branch service, spliced into a five hundred and sixty eight kilometre card and speaking for eleven per cent of it while the other five hundred and five kilometres kept whatever had been drawn before. The Toledo relation is off the card's candidate list. In its place the line is drawn across the Madrid–Málaga high-speed line as far as Antequera and the Antequera–Granada line beyond it, which is the railway the train actually uses: five hundred and sixty nine kilometres against the card's five hundred and sixty eight. Asked about it afterwards, ÖBB Scotty returns AVE 2076 and our drawing sits on its shape end to end, so unlike Barcelona to Málaga this one has earned its check back. Reported by alan_gr on the OpenStreetMap community forum.
Two more routes were checked because the same fault was suspected of them and it is only true of one. Munich to Verona is clean: four hundred and forty one kilometres against the four hundred and forty two of the EuroCity that runs it, on its shape to the metre, every station exact. Zurich to Vienna is not, and its problem is a missing relation rather than a wrong one: our line runs twenty five kilometres from St. Gallen, which the railjet calls at, and not one of the thirteen relations linked to that card describes the Swiss stretch from Zurich through St. Gallen to St. Margrethen. It cannot be repaired until that relation is found, and saying so is more useful than another redraw of the Austrian half. Found by the atlas’s own audit sweep.
The atlas can now be audited a country at a time, which is how every serious report about it has arrived. Andalucía missing seven services that connect provincial capitals, four branches missing in West Wales, a Greek line drawn from the wrong end, Belgium and Czechia and Poland described as mostly absent: not one of those questions can be asked of a single card, because "what is missing" has no answer inside one. It has an answer inside a country, and OpenStreetMap holds it. For any country the audit now lists what runs there and is not drawn, separately from the corridors that ARE drawn but whose relation was never linked to the card, a distinction that matters, because the second group looks like missing routes and is not. Spain's Málaga to Madrid was reported as having disappeared; it had not, its card had simply lost its link to the railway. The audit also names every card whose relation describes less than nine tenths of it, and every claim resting on such a link. There were two hundred and nine of the first and forty seven of the second across the whole atlas, and they are the work list from here. Found by the atlas’s own audit sweep.
Switzerland could not be checked at all, and the reason was a number too large to write down. A zip file states how big each of its parts is in a four byte field, and above four gigabytes there is no room, so the writer puts a marker in the field and states the real size elsewhere. Switzerland's national timetable does that for every part it contains, including the empty ones. Read without knowing it, all ten claimed to be exactly 4,295 megabytes long (that marker, read as if it were a number), so each was unpacked from its own beginning to the end of the archive. The run burned eleven minutes and several gigabytes and produced nothing at all, twice, and was written off as "probably a memory problem". It was not. Read properly the same file takes ninety seconds, and forty six Swiss routes now carry a check against the Swiss federal transport office's own published timetable: the Gotthard panoramic train, the Seetal line, the Tonkin line, the Furka steam railway, the Rhine Falls line, the sleeper from Zurich to Amsterdam. Norway was a different size problem with the same answer: its route geometry alone is 2.7 gigabytes unpacked, larger than the longest piece of text this machine can hold, so nothing is held any more. Every file in every railway's timetable is now unpacked as a stream and read one line at a time, which costs about a megabyte however large it is. Found by the atlas’s own audit sweep.
SNCF's published timetable stopped moving in January 2025 and this atlas did not notice for eighteen months. The download still answered, still contained trains, and still looked exactly like a timetable. What it no longer contained was a single date in the future. The guard that was supposed to catch that asks whether any service in the file runs in the next thirty days, and when the answer is none it concludes the file carries no calendar and every trip in it must be taken as running, which is true of a feed that ships no calendar at all, and precisely backwards for one whose calendar has expired. Nine routes here were citing a timetable whose newest entry was the twenty second of February 2025. The two cases are told apart now: a feed that has calendar entries and no current ones stops the run and prints the date it stopped at, rather than quietly becoming evidence. SNCF's current download, which carries TGV, Intercités and TER in one file and had been rebuilt the day before, replaces both dead ones, and it answers for fifty four French routes where the pair of expired files managed thirty eight between them. Found by the atlas’s own audit sweep.
The check was asking about the wrong train, and on the Paris to Lyon high speed line it was asking about a train that has never been there. Where several services connect the same two cities something has to decide which of them a card means, and what decided it was whichever service's length came closest to the distance printed on the card. That is a weak test anywhere and a broken one where the railway publishes no route geometry, because then a service's "length" is a chain of straight cuts between its stops, five to fifteen per cent short of the railway, and shorter the fewer stops the train makes. So the Paris to Lyon card was matched to the classic line through Dijon and then failed for lying fifty five kilometres from Dijon, which it does, because the LGV has never gone there. Paris to Lille was condemned the same way by the stopping service through Creil and Longueau. Seven verdicts that were right would have been deleted. Now every service running between a card's two ends is measured and judged, and if any of them fits our line the one calling at the most stations wins, because the hardest one to fit is the one worth citing and an express with two stops proves almost nothing by passing near both of them. If none fits, the closest is reported, so a failure now means that no service running between those two ends follows our line, rather than that one arbitrary service does not. Found by the atlas’s own audit sweep.
A related piece of nonsense went with it. A service counted as a card's own if its termini were within twenty five kilometres of the card's, which is the right allowance for a card ending at a city where the railway's own terminus is a different station in it, and absurd for the Stoosbahn, which is one and seven tenths of a kilometre long: a twenty five kilometre circle round each of its ends swallows half of central Switzerland, and the timetable duly offered a train to Rigi Staffel (a different mountain) as the Stoosbahn. Brienz Rothorn was matched to Giswil and the Rhune rack railway to Bayonne the same way. The allowance now scales with the journey, and eleven mountain railways stopped being told they were somewhere else. Found by the atlas’s own audit sweep.
A reference with no geometry may confirm a line and may no longer condemn one. Confirming is safe whichever service was matched: if every stop of a train that really runs lies within tens of metres of our line, our line is on that railway, and which of the trains sharing it was picked does not matter. Condemning is not, because a failure there has two completely different causes and nothing can tell them apart without a shape to look at. The GoldenPass is one journey on the card and three trains in the timetable, so the only single trip running Montreux to Luzern goes round by Bern. The AVE to Zaragoza does not terminate at Zaragoza (it carries on to Barcelona), so the only services that do end there are the classic ones through Sigüenza. Neither says anything whatever about our drawing, and both would have deleted a claim. Found by the atlas’s own audit sweep.
Where a check rests on stations rather than on a drawn shape, the page now says which of the two reasons applies. A journey planner can hand back a shape too coarse to measure a line against; an operator's own timetable may carry no geometry at all, which is true of Switzerland, Renfe, SNCF and Comboios de Portugal. Telling a reader that the operator's shape was "too coarse" when the operator publishes no shape is inventing one, and forty five pages were about to do it. Separately, the rule deciding which of two verdicts a page should show turned out to do the opposite of what it says: asking the railways on their own, without the planner in the same run, moved a hundred and nineteen routes off ÖBB Scotty's measured shape and onto a weaker stations sentence, purely because the railway with no geometry had been asked last. Five hundred and thirty two routes now carry a timetable check, from five hundred and four, and none of them has traded a stronger claim for a weaker one to get there. Found by the atlas’s own audit sweep.
Six lines are redrawn, and the two worth naming are the Gotthard. Zurich to Milan was thirty eight kilometres away from the EuroCity that runs it; rebuilt across the six relations that make up that railway it now sits on EC 13's own shape end to end, to within a hundred metres, with every station it calls at inside a hundred metres of our line. The railjet from Vienna to Budapest, the Neuchâtel to Pontarlier line and the Schynige Platte rack railway were repaired the same way and each of them earned its check back. The Gotthard Panorama Express went from ninety one kilometres of disagreement to twenty seven and the Vienna to Rome sleeper from three hundred and twenty nine to eighty three, which is a great deal better and still not right, so both are redrawn and neither claims anything. Eleven other Swiss cards could not be attempted at all: the repair welds together every relation somebody has linked to a card, and a card with only one linked relation has nothing to weld. That is the next piece of work, and it is a linking problem rather than a drawing one. Found by the atlas’s own audit sweep.
One redraw was applied and then taken back off, which is worth writing down because the reason is a hole in the test rather than a bad line. A repair only ships if the planner says it is better, and "better" was the total kilometres of railway the two drawings disagree about: how much of our line is off the train's track, plus how much of the train's track we fail to draw. Frecciarossa Rome to Bari came back sixty kilometres better by that measure and thirty nine kilometres longer than the train itself: five hundred and twenty one kilometres drawn against the four hundred and eighty two the service runs, from four hundred and eighty three that had matched it almost exactly. It had cured a corridor by drawing a detour. Both versions are wrong and the new one is wrong in a way that also prints a false distance on the map, so it is reverted and the test now refuses any repair that makes the length disagreement worse. The reference is the same on both sides of that comparison, so it is a like for like measurement rather than a judgement call. It refused two more Italian redraws on its first outing, Rome to Lecce among them. Found by the atlas’s own audit sweep.
A second repair was applied and taken back for a subtler reason: it had earned a check by falling into the weaker of two tests. Frecciarossa Paris to Milan came back holding a timetable claim and twice the disagreement: a ninetieth percentile of one hundred metres before and three point one kilometres after, thirty two kilometres off the service's own track turned into sixty eight. It passed because the shape ÖBB Scotty publishes for that train is too coarse to judge a line against, and when that happens the decision falls to the stations the train calls at, which BOTH drawings sit within two hundred metres of. So the new line won a sentence the old line could also have had, while being measurably further from the railway. Earning a claim is no longer treated as proof that a repair is an improvement: it must not lose ground on the geometry either. Found by the atlas’s own audit sweep.
Five timetable checks were resting on services that call at two stations, and two stations are the card's own ends. That claim reads "our line passes within forty two metres of every station this service calls at", which sounds like something and is nothing: both ends had to be within a few kilometres of ours before the service was considered at all, so four hundred and ninety kilometres of line in between could have gone anywhere and still passed. Eurostar's own feed lists London and Paris and nothing between them. SNCF's direct Paris to Lyon TGV is the same, and the thirteen hundred kilometre Al Andalus manages three calls in total. A check that rests on stations now needs enough of them to constrain the line between: one every two hundred and fifty kilometres, and never fewer than three. Seven claims failed that and are gone. Where a route has a real shape to be measured against, nothing changes; this only governs the weaker sentence. Found by the atlas’s own audit sweep.
Eurostar publishes its own timetable, and it answers a question no national railway can. Four cards on this map cross borders that no single operator covers (Paris to Amsterdam, London to Brussels, London to Amsterdam, Brussels to Cologne), and Eurostar, having absorbed Thalys, runs all of them. Its own file on its own domain is now one of the railways this atlas checks itself against. That came out of a systematic look for others: a catalogue of three and a half thousand public transport feeds worldwide was read against every country where routes here go unchecked, and the honest result is that for the largest gaps there is nothing to find. Japan publishes five open feeds and every one is an urban bus or subway; JR East offers railway data only through a time limited contest and states plainly that it does not include the Shinkansen, and JR West, Central, Kyushu and Hokkaido publish nothing at all, which leaves all seventy eight Japanese routes here, nearly all of them Shinkansen or limited expresses, with no source in existence. China, South Korea, Taiwan, Malaysia and Switzerland have none either. Great Britain has seventeen and all seventeen are buses. Italy's only open railway files are Lombardy's regional operator and Trenitalia's Sardinian services. Those routes will keep saying what they say now: here is the operator's own timetable, and no claim about our drawing. Found by the atlas’s own audit sweep.
Two more lines are redrawn in Germany. The ICE from Hamburg to Berlin was eleven kilometres away from the train that runs it and now sits on ICE 2587's own shape for the whole two hundred and eighty six kilometres, to the metre, with every station exact. The Zugspitze rack railway earned its check against the Bayerische Zugspitzbahn's own service. Germany was two thirds of the way through its list and France had not started when this stopped for the night; both resume from where they are, and the OpenStreetMap answers already fetched are kept, so the next run starts warm rather than from nothing. Found by the atlas’s own audit sweep.

11 August 2026

This site is now World Train Map, at worldtrainmap.com. It was TrainRouter, and that name promised something this has never been. A router plans your journey; it asks where you are, where you are going and when, and hands back a departure. This is an atlas. It draws 1,247 notable passenger routes and tells you what each one is, how far it runs, how long it takes and why it is worth knowing about, and it has never once sold a ticket or read a timetable at you. People arrived expecting a planner and left when they did not find one. The new name is what the thing actually is. Nothing you had is lost: every trainrouter.com address redirects here in a single step and keeps its path, so an old bookmark, an old link in somebody else's article and an old search result all land exactly where they used to, on the same page. That redirect stays up permanently. The map, the data, the route pages and their addresses after the domain are all unchanged, and so is the light or dark setting your browser remembers for this site, which would have been quietly forgotten had the name behind it been tidied up along with everything else. Found by the atlas’s own audit sweep.
The spur up to Kilkenny went missing again, and what is fixed here is the reason it could go missing twice. Dublin to Waterford trains run up to MacDonagh station and reverse back out to the Lavistown junctions; that was drawn on 1 August and gone by the 8th, when 984 lines were redrawn from the OpenStreetMap relation that describes them. A relation describes a RAILWAY. A spur describes what a TRAIN does on it, and redrawing from the relation throws the second fact away. Nothing here noticed, because the four questions this atlas asks of a drawing are its length, its longest straight, the countries it crosses and where its ends sit, and a seven kilometre detour on a hundred and seventy kilometre line answers all four exactly as it does when it is drawn. There is a fifth question now. Every route that links a line article records the stations that article names and this line actually reaches, 6,632 of them across 457 routes, and the build fails if a redraw moves a line off one of them. The spur is back, and the next sweep cannot quietly undo this one. Reported by alan_gr on the OpenStreetMap community forum.
The same question, asked of every line in the atlas, found two more that nobody had reported. Nanning to Guangzhou no longer reached Nanning, which is its own terminus, and Madrid to San Sebastián had stopped calling at Medina del Campo, fifteen kilometres off the line it belongs on. Both are the 8 August redraw again, and both are drawn back the way they were drawn when they worked. Thirty nine detours were dropped that day and only these three were station calls: the rest were lines correctly moved off a railway their trains do not use, the pre sweep Paris to Marseille among them, which ran down the classic line past four RER stops the TGV has never called at. What separates the two is whether the lost piece left the line and came back, or simply went somewhere else. One thing was paid for this: three cards can no longer say they were drawn node for node from one relation, because they are now that relation plus a spur, and a provenance claim that stopped being exactly true is dropped rather than left standing. Found by the atlas’s own audit sweep.
Košice to Prešov joins the atlas, the day it was asked for. Slovakia’s second and third cities are thirty five kilometres apart on a branch the Kaschau–Oderberg company opened on 1 September 1870, leaving the Bohumín main line at Kysak and running north up the Torysa; it was electrified in 1978 and is good for 100 km/h. The line is drawn from the two OpenStreetMap relations that carry it, and checked against the running timetable rather than assumed: the REX does it in thirty seven minutes with no change of train, and our drawing sits 0.2 km from the shape that service publishes. Reported by dvh on Hacker News.
A report that needed no change, written down because the checking is the point. Lisbon to Guarda was said to run through Covilhã, turning east at Entroncamento instead of carrying on to Pampilhosa. That was true while it was true: the Beira Alta was shut for forty two months of works and its trains went round by the Beira Baixa. The line reopened in full on 28 September 2025, the Intercidádes went back to Pampilhosa and Mangualde, and the drawing here sits on the shape CP publishes for that train, within a hundred metres for nine tenths of its length. The four hours eleven minutes on the card is the timetable the line reopened with. Reported by nwales58 on RailUK Forums.
Eighteen routes join in one evening, all between the Baltic and the Adriatic, and every one was proved against a running timetable before it was drawn. A candidate went in only when the public transit feeds showed a through passenger train on it this week; its line comes from the shape that service publishes, and where it rides rails already on the map it holds the existing card’s exact points rather than a second tracing. Poland gains six: the Lublin, Olsztyn and Bydgoszcz InterCity corridors out of Warsaw, Poznań to Gdynia, the Nadodrzanka down the Oder from Wrocław to Szczecin, and the 1867 mountain line to Szklarska Poręba. Czechia gains the second Prague to Brno main over the Vysočina highlands, the rychlík to Trutnov under the Krkonoše, and RegioJet’s R8 to Ostrava. Slovakia gains its southern half: Bratislava to Banská Bystrica, Zvolen to Košice through the karst, and Košice to Humenné. Hungary gains the Kék Hullám along Balaton’s north shore; Germany the seaside Intercity to Norddeich Mole, the Saxony ICE to Dresden and the Remsbahn Intercity to Nuremberg; Austria the promised pair, the railjet to Venice and the EuroCity to Ljubljana. Five more candidates were checked and declined: Warsaw to Łódź is already on the map inside the Wrocław express’s line, and no through train could be proved for the Gdynia to Szczecin coast, Dresden to Wrocław, Olomouc to Ostrava over Krnov, or Budapest to Sátoraljaújhely, so they are not drawn. Vienna to Venice and Vienna to Ljubljana were promised to BafDyce in the dataset repository’s first issue.
A nineteenth route arrived by email the same evening, with its history attached. Braunschweig to Uelzen via Gifhorn was an emigrants’ road before the war, the way from south Germany to the America ships at Bremen and Hamburg; it came out of the war cut to a single unelectrified track, and only new passing loops in recent years brought back an hourly train. The line joins drawn from the shape erixx publishes for the RB47, which the feed shows running the 97 km in about 1 h 46, every hour, with no change of train. The sender called it the RB37; the running timetable says RB47, and the timetable wins, as it does for every card here. The station history in the message, down to the memorial signs on the platform at Gifhorn, is on the route’s own page. Sent in by email; the sender asked not to be named, and is not.

10 August 2026

Eleven hundred lines are now drawn from the relation that describes them rather than from a search for them. Until today most "OSM traced" lines here were found by pulling the track inside a corridor and looking for a path across it, which is inference, and inference picks the wrong railway where two run side by side and rounds what it does find until the line sits a couple of hundred metres off the rail it means. A route relation is the other thing entirely: somebody who knows that railway listed the exact ways it uses, in order. The atlas now asks which relation a train belongs to by asking the stations it calls at, because a relation lists the stops it serves and proximity cannot tell a service from track it happens to share. Of 1,247 routes, 647 are answered by one relation covering the whole card and 176 by a chain of them. 1,093 lines were redrawn from those answers and almost none of it is a change of alignment: 1,098 of the candidates were the same railway drawn coarsely, now a median of four metres from OpenStreetMap instead of an approximation. The atlas draws 1,122 of its lines from a named relation, and measures 1,243 of them against real track at 98 per cent agreement. Found by the atlas’s own audit sweep.
Twenty two routes were deliberately left alone, and it is worth saying why rather than quietly redrawing them. Their relation would move the line more than a kilometre, and that is the signature of a different railway rather than a finer drawing of the same one. Kyiv to Lviv sits 42 km from its candidate and the Roros line 51 km, which is either a badly wrong drawing or a relation for another railway entirely, and nothing here can tell which without somebody reading what the relation is actually called. They keep their current drawing until that is done. A further 424 routes have no relation at all, which is not a failure of the method but a fact about the map: OpenStreetMap has no route relation for much of Africa, Central Asia and South America. Found by the atlas’s own audit sweep.

9 August 2026

The Arlberg railway was linked to the wrong OpenStreetMap relations, and finding out why changed how the atlas looks for them. The card is Zurich to Vienna and it was drawn from three: the Innsbruck to Bludenz line, which is right; the Salzburg to Kufstein to BRENNER line, which this train leaves at Kufstein and which then carries it up a valley it never enters; and a WESTbahn service, a different operator arriving at a different Vienna terminus. All three run near our line, which is exactly how they were chosen, and proximity cannot tell a train from track it happens to share. So the question changed to a better one: look at the departure in OpenStreetMap, see which services call there, and work towards the destination. Stops know which relations contain them. The answer for Zurich to Vienna is a single relation, the railjet’s own, and the card is now drawn from it end to end, 780 km of a declared 787, measuring 99 per cent on real track where the three wrong ones managed 95. One more thing had to give: sixty seven of that relation’s ways are tagged as under construction, a stretch being renewed on a line trains run over hourly, and dropping them cut the relation into three pieces so no path existed between Zurich and Vienna at all. Found by the atlas’s own audit sweep.
The maker photographed four places where a line goes somewhere and comes back: a loop west of Hamburg, another beside Pau, and two forks that turned out to be something else. The atlas could not see any of them. It had a check for a line retracing itself, but it only spoke up at forty kilometres and only on routes already drawn longer than their card, which was built to catch a two hundred kilometre branch in Saudi Arabia and is blind to a five kilometre loop. There is now a finer one: walk every line at hundred metre steps and look for a point it returns to, two samples within a hundred and twenty metres of each other but a kilometre or more apart along the line. It finds a hundred and eighty seven, and it labels them, because the two shapes have different causes: a train that goes out and comes back runs the opposite way at the point it touches itself, while a loop runs parallel. The ICE from Hamburg was carrying eleven kilometres of loop out of the city and back; the night train to Lourdes twenty seven at the far end. Both removed, and both lines now measure on real track for their whole length. The rest are a review list rather than a fault list: a rack railway’s switchbacks are real, and so is a train that reverses at a terminus. Spotted on the map, not by the instrument.
The other two photographs were a different thing, and worth saying so rather than quietly fixing nothing. Neither the line at Innsbruck nor the one at Vienna doubles back at all: at Innsbruck two routes of the same kind meet and are drawn the same colour, so the junction reads as one line forking. At Vienna it is real but harmless, fourteen cards ending at Vienna points up to three and a half kilometres apart, which is Hauptbahnhof and Meidling and Westbahnhof being different stations. Checking every city for that turned up two that are not harmless: a card ending at Tampere a hundred and forty kilometres from where the other Tampere cards end, and one at Ulm a hundred and six kilometres out. Those are on the list to fix. Found by the atlas’s own audit sweep.
The Qinghai–Tibet railway was credited with a top speed of 100 km/h while its own card said 1,956 km in twenty one hours, which is ninety three. No train that stops anywhere averages ninety three per cent of its own maximum for the better part of a day, so one of the two numbers was wrong, and it was the speed: the line runs at 160 between Xining and Golmud and drops to 100 only for the permafrost beyond, and the card had been carrying the slow half’s figure for the whole journey. It now says 160. The same arithmetic caught the Lanzhou to Chongqing railway, where passenger trains are allowed 200 and the card said 160. Thirty six routes average more than three quarters of their claimed maximum and the rest of them check out, including several that look alarming and are not: a funicular really does average its top speed, because it accelerates for a few seconds and then runs at line speed to the summit. Found by the atlas’s own audit sweep.
Two hundred and seventeen more routes have been checked against the running timetable, which is the strongest thing this atlas can say about a line: not where it came from, but that a train really goes that way today and that our drawing sits on the shape the planner returns for it. Every route that had never been asked was asked, first the four hundred and forty three European ones against ÖBB Scotty and then all eight hundred and fifty four against Transitous, which carries the operators’ own published shapes and reaches Japan, North America and Australia where Scotty does not. Five hundred and twenty seven routes are now timetable checked, up from three hundred and ten. Most of the rest were refused rather than failed: a planner often answers with a shape too coarse to judge, under three points per kilometre, and a line cannot be measured against a ruler. None of them turned out to be on the wrong railway. Two of the checks were themselves at fault and are mended: a bus leg now has to be a real share of the journey before it disqualifies an itinerary, which had been refusing the California Zephyr over fourteen kilometres of shuttle in three thousand nine hundred, and one distant station no longer outweighs all the others, which had been refusing the Canadian for a single stop seven kilometres off a line that matched to three hundred metres. And thirteen lines are now settled by their STOPS rather than by a drawn shape, which is a different sentence on those pages and is meant to be. A planner sometimes answers with a ruler between distant stations, 0.08 points per kilometre on the West Coast Main Line, and a real railway cannot be measured against a ruler. The stops are not drawn though, they are coordinates, so where our line passes within a few tens of metres of every station a service calls at and the length agrees, that is said in those words and not dressed up as a measurement of shape. Found by the atlas’s own audit sweep.
The two Bosnian cards named services that do not run, and the operator’s own timetable says what does. Željeznice Federacije BiH publish Sarajevo to Ploče, which is the Talgo down the Neretva to the Adriatic through Mostar and Čapljina, and Sarajevo to Maglaj, which is as far north as their trains now go. There is no Sarajevo to Mostar relation, because Mostar is a stop on the Ploče train, and no Sarajevo to Banja Luka at all. So the first card runs on to Ploče, 194 km in about three and a half hours, drawn from the Sarajevo–Čapljina line and the Croatian M304 beyond the border; and the second becomes Sarajevo to Maglaj, 148 km, on the 1947 youth railway towards the Sava. Both now lie on real track for their whole length, a median seven to nine metres from it. The Maglaj card also loses a photograph of the Neretva bridges and Ploče station, two hundred kilometres from anything it draws. Found by the atlas’s own audit sweep.
Where no planner can be asked, the page now points at the operator’s own timetable instead. ÖBB Scotty stops at the edge of Europe and Transitous carries only the feeds operators chose to publish, so most of the world has nothing to check a line against. Seven hundred and thirty seven routes now carry a link to the railway’s own timetable, from ŽFBiH to Amtrak to the Wellington Cable Car, and each says exactly what it is: a place to look the service up, not a check of our drawing. More than nine in ten routes in the atlas now have either a timetable check or that link. Every URL was fetched before it shipped, and the ones that could not be confirmed were left out rather than guessed at: a missing link costs a reader one search, a wrong one sends them to a stranger. Most came from the operator’s own Wikidata entry, and where the operator could not be found (which is nearly always a preservation society whose website belongs to the railway rather than to the society), the line’s own article answered instead. Found by the atlas’s own audit sweep.
Some of the worst drawn lines in the atlas turned out to be the best, and the instrument was at fault. The measurement asks a rail router to travel between points along each line and compares the two. A router will not refuse a 550 m funicular, it simply goes round: down the hill and along the tram in the street below. So the Dresden funicular was reported as nought per cent on rails and a median 637 m from track, while sitting on the OpenStreetMap relation that describes it to the metre. The same for Fribourg, Tbilisi, Ljubljana castle and about twenty more. A poor score is now a reason to go and ask OpenStreetMap directly what track is under the line, and where the two answers disagree the direct one wins. Every one of those funiculars now reads a hundred per cent on rails, a metre or less from the track, which is what they always were. Found by the atlas’s own audit sweep.
A reader on the Hacker News front page found, in about a minute, that the Tanzanian Central Line linked to the Central line of the London Underground. It is a railway, it does share the word central, and it is four thousand miles away. The first response was a sweep for other links to undergrounds and metros, which found that one and nothing else, and that was the wrong check twice over, because it can only catch wrong articles that happen to be about undergrounds, and it cannot catch the far likelier error of the same railway name in another country. So the question was asked properly instead: where does Wikipedia say each linked article IS, and is that anywhere near the line we draw. Thirteen links were wrong. The Red Arrow, the Moscow night train, pointed at a derailment in Pennsylvania in 1947 that killed twenty four people; the Byron Bay solar train pointed at a railroad in Massachusetts; Bangkok to Aranyaprathet at the Eastern Suburbs line in Sydney; the North Wales coast at the North Coast line in New South Wales; Cape Town at the Southern Highlands line in Australia; two Portuguese Intercidades at a Brazilian one; the Tren de Felipe II at a Mexican train, which now has no link rather than a wrong one. Every one of them was a railway article sharing a name, which is exactly what a keyword guard cannot see. Found by the atlas’s own audit sweep.
The same question, asked of the photographs, was worse. The gallery under each route page is harvested from that route’s own Wikipedia article, which is only ever as good as the article it was pointed at, and Nice to Ventimiglia was illustrated with four pictures of the English sea wall at Dawlish, because its gallery had been taken from "Riviera Line", which is in Devon. The card’s link had been corrected long before; nothing re-checked the pictures. Measured against every route’s own drawn line, a hundred and twenty photographs on eighty four routes sat more than two hundred kilometres away: Guam and Saipan on a New York to Charlotte train, Seattle light rail on a Taiwanese one, the Queen Mary 2 in Boston on the line to Takamatsu, a 1910 photograph of newsboys in St Louis on the Burma railway. The Burma railway’s own link turned out to point at a disambiguation page, which is why its gallery had filled with art about death, Rembrandt’s Anatomy Lesson among it. Three things changed: the gallery now reads the article the card itself links rather than searching for one, it no longer tops that up from a searched article when the card has named one, and a photograph Commons places outside every country the route runs through, hundreds of kilometres from the line, is refused. Fifty seven were refused on the last run. Where that leaves a route with nothing, it now shows nothing, which is the honest answer. Found by the atlas’s own audit sweep.
Sarajevo to Maglaj and Sarajevo to Ploce were showing the same four photographs, Ploce station among them, on a route that ends at Maglaj and never goes there. Reported by alan_gr on the OpenStreetMap forum, who put it charitably as the galleries looking like random photos of trains in and near Bosnia. The cause was that the Maglaj line had no article of its own here, so the harvester searched, found the Ploce line, and gave one route the other’s pictures. It now names the Sarajevo to Samac railway, which is the line it is on. Reported by alan_gr on the OpenStreetMap community forum.
Amsterdam to Berlin said five hours forty two minutes. It is six hours forty four, every day of the five that were checked, and the slowest booked run is seven and a half hours. This was challenged in public by artisinal on Hacker News, who said the high-speed label did not survive the journey time and put the real figure near six and three quarter hours: he was right, and closer than we were. The line is 641 km rather than the 660 the card claimed, so the honest average is 95 km/h. Sweeping the same check across all one hundred and thirty eight high-speed routes found one more card that was wrong in the other direction: Stuttgart to Munich was slower than its own train, and both Transitous and OBB Scotty return one hour fifty eight. Fifteen further flags were investigated and not acted on: in Italy and Turkey the planners answer with regional trains or with a service measuring a hundred per cent more kilometres than the card, which is the wrong-service trap this atlas already knows about, and a flag is not a verdict. Reported by artisinal on Hacker News.
What "high speed" means is now written where the filter is. Thirty one routes filed as classic average faster end to end than most of the high-speed list (London to Newcastle at 165 km/h among them), so the label plainly does not mean the journey is fast. It means the train and its line were built for 200 km/h or more, and every card already shows the speed the journey really averages. Nothing on the map said so until a reader filtered for high speed, got a six-hour train, and said it did not feel like one. Found by the atlas’s own audit sweep.
The conventional Malaga to Sevilla was booked at two hours fifty five, which is not the time of any train that runs that way. The Avant does the city pair in two hours twenty one by going through Cordoba on high-speed track, which is a different railway; the Media Distancia that actually rides Bobadilla, Osuna and Marchena takes three hours forty two, on four checks across the day. Its top speed was blank and is 160, which is what OpenStreetMap tags over half the line. Found by the atlas’s own audit sweep.
Two routes join, both asked for by the people who ride them. Ottawa to Montreal was the busiest gap in the atlas’s Canadian coverage: three Corridor cards met at Montreal and Ottawa and none of them joined the two, which James Brunet wrote in to say, and which kimos had noticed on Hacker News as uneven VIA coverage. It is its own railway through Casselman and Alexandria rather than any part of the Toronto line, five trains a day, the quickest 2 h 5 min. Tartu to Valga was reported missing on GitHub by Carlrs for the right reason: it is how Estonia reaches the rest of Europe by rail, the Baltic Railway’s 1887 branch to meet the Riga line, and the Riga train still uses it. Both are drawn from their own OpenStreetMap route relation, eight and fourteen metres off it, and both were then checked against the operators’ own published timetables, where our line sits on the shape of the train that runs it. Asked for by James Brunet by email and by Carlrs on GitHub.
Barcelona to Malaga left Barcelona down the conventional tracks, taking two improbably sharp turns to get there, and the card claimed it had been checked against the running timetable. Both were true at once, which is the uncomfortable part: the check measured our line against the right service and reported two tenths of a kilometre, because it sampled where the line and the service agreed. Asked properly, OBB Scotty’s own shape for the service put our drawing up to twenty two kilometres out over the first hundred kilometres. It is now redrawn from that shape and sits on it exactly. Reported by alan_gr, who also said Scotty displays the correct route in this area, and was right about that too. Reported by alan_gr on the OpenStreetMap community forum.
On a phone, the only words this map showed on first paint were its name and the line "the world railway map", which is a promise of every railway there is, above a map that carries a tenth of them. It arrived on the Hacker News front page and febusravenga said exactly that: the two lines conflict, and nothing tells you it is curated, "that's why people complain about missing routes". Every other complaint that day was a version of the same thing. The logo lockup is the title now, so there is no second heading to disagree with it, and the line under it says what this is and what it is not, before anything is tapped. Reported by febusravenga on Hacker News.
The rack railway to Kalavryta and the Campos do Jordao line measured two per cent and one per cent on rails, and both are drawn within thirty metres of their own rails for their whole length. OpenStreetMap tags eighty three of the rack railway’s eighty five ways as disused, and every way of the Brazilian line, so the query looking for track simply could not see the track that was there. Disused says nobody runs on the rails, not that the rails are somewhere else, so it now counts. Two things follow. Both lines are drawn from their own relation for the first time. And the tag was right about the service: the Odontotos has been suspended since landslides in March 2026, which the card does not yet say. Found by the atlas’s own audit sweep.
Eighty six routes that no OpenStreetMap relation describes have been snapped onto real track. Until today a line with no relation was drawn by searching the rail network for a path between its two ends, and a search constrained by only two points can pick the wrong railway and cannot tell you it has. These are matched instead: the whole shape of the line we already draw is handed to a matcher that asks which sequence of real track most likely produced it, so every kilometre of the drawing constrains the answer rather than just the ends. It is done in twenty five kilometre legs, and a leg is only taken if it comes back the right length, stays inside the corridor and lands where it started, so a stretch the matcher gets wrong costs that stretch and nothing else. Seventeen hundred and seventy four legs of two thousand six hundred and forty eight were believed, three hundred and one of them moving the line onto track it was not on. Where the matcher was not believed the old drawing stands. Found by the atlas’s own audit sweep.
Twenty four lines that were genuinely in the wrong place have been redrawn from the railway they name, after their endpoints were corrected from the stations’ own coordinates. The Baekdudaegan canyon train was in the wrong valley, sixteen kilometres from Buncheon. The Cierny Hron forest railway was eighteen kilometres out. Kingston Flyer, Nevada Northern, the Skunk Train, Cass, the Oigawa steam train, the Tsugaru stove train, Kurobe gorge, Ruskeala, the Usedom seaside railway, Nariz del Diablo, Trem do Vinho, Maria Fumaca, Montserrat, the Carmelit, Drachenfels, the Trencin electric railway, Bonde de Santa Teresa, Chiemsee, Strasburg, Kampala and the Porrettana all follow their own line now. Each was checked twice before anything moved: once to find the railway and its termini, once by a second reading that tried to disprove the first. Found by the atlas’s own audit sweep.
The night train from Rome to Palermo crossed the Strait of Messina from the wrong shore. It ran past the train ferry at Villa San Giovanni, carried on twelve kilometres down the coast to Reggio Calabria and crossed from there, ten kilometres of open sea instead of the ferry’s seven. It now leaves from the ferry berths themselves and arrives at the ones in Messina, both read from OpenStreetMap. This one fought back twice: the relation that draws the Calabrian line runs to Reggio, so each time the line was corrected the redraw put the spur back, and reported the result as four metres off, because the check only ever measured our line against the relation and never the relation against ours. A line can sit perfectly on a path that goes somewhere it never went. Both directions are measured now. Found by the atlas’s own audit sweep.
Two cards had their distance wrong and it was the distance, not the drawing, that was refusing to let them be redrawn. The Porrettana Express says Pistoia to Porretta Terme, which is forty kilometres of railway, and the card said twenty six. The Cinque Terre Express from Levanto to La Spezia is twenty kilometres, and the card said thirty. Both are corrected, and both lines now come from the relation that describes them. A related nuisance is now understood rather than fixed: several tourist railways declare the distance of the trip you buy, which is there and back, so Cass says thirty five kilometres for a seventeen kilometre climb. Those cards are internally consistent, since their durations are round trips too, so they have been left alone and the drawing is now told the one way length separately. Found by the atlas’s own audit sweep.

8 August 2026

The atlas now goes looking for the railways a route is missing rather than the one it already has. A card is usually several railways end to end, and until today a line that could already name one relation was never asked about again: Munich to Verona was drawn from the Brenner line for a fifth of its length and from our own search for the rest, while the page named that one relation as though it spoke for all of it. Every card is now walked half a kilometre at a time and each point asked how far it sits from the nodes of the relations already linked, and only the stretches nobody speaks for get searched. Munich to Verona answered with the relation for EC 83, which is the train itself, and the card is now drawn from it end to end. The same question turned up the service’s own relation for the Bima, the Howrah Rajdhani, Alaska’s Coastal Classic and the Shiokaze, none of which could be found while the search was pointed only at routes that had nothing. Found by the atlas’s own audit sweep.
Six lines turned out to be drawn along a railway their trains do not use, and in each case the relation is what proved it rather than an opinion about the map. The Paris to Nice TGV ran down the west bank of the Rhône, twenty two kilometres from the line its trains take. The Stockholm night trains to Hamburg and to Berlin left Stockholm on the Nyköping coast line instead of the main line through Katrineholm, thirty five kilometres away at the worst point. The Bima was drawn along Java’s north coast when it runs the southern route through Yogyakarta. The Maple Leaf ran up the wrong bank of the Hudson. The Howrah Rajdhani detoured through Varanasi rather than the Gaya main line the train is named for. And the Harbin to Dalian high speed card was drawn on the conventional Shen Da railway running beside it. Every one of those redraws was then confirmed by the independent measurement: the Paris to Nice line went from 96.2 to 98.9 per cent on real track and its median distance from the rails fell from 52 metres to 14. Found by the atlas’s own audit sweep.
Redraws that a person refused are worth recording too, because a relation running between the same two cities is not the same thing as the railway a card describes. The Gotthard Panorama Express was offered an InterCity that goes through the base tunnel, and the card is the mountain line. The Rokumon, a sightseeing train on the old Shinano line, was offered the Hokuriku Shinkansen beside it. The Hong Kong to Beijing sleeper was offered the conventional Jingguang railway rather than the high speed line it runs on. The Andean Explorer was refused for the opposite reason: it doubles back between Juliaca and Puno on purpose, and the relation would have quietly straightened that out. One redraw was applied and then reverted, because the Málaga to Seville Avant came out thirty kilometres short and eighteen kilometres from Córdoba, and not applying something beats applying something that cannot be defended. Found by the atlas’s own audit sweep.
Three more cards link a Wikipedia article, in Korean and Greek. The two Patras suburban lines link the article that is about exactly those two lines, and the Jungbu Inland card links the line it is named for rather than the train class article an English search kept offering. Found by the atlas’s own audit sweep.
Barcelona to Málaga left Barcelona Sants on a twenty kilometre straight line and then ran down the coastal conventional railway instead of the high speed one, which is what produced the two impossibly sharp turns in the reported screenshot. It is now drawn from the AVE service relation that starts inside Sants rather than the line relation that starts outside it, and from the Madrid to Málaga high speed line for the second half. The uncomfortable part is that the atlas already knew: it measured 96.4 per cent of that line on real track with a fourteen kilometre stretch away from it, while the card said the line had been checked against the running timetable and sat 0.2 km from the shape of IRY 6020. Both were true. The timetable check reports a ninetieth percentile, and a fourteen kilometre error inside eleven hundred kilometres does not move a ninetieth percentile, so a card can be verified and visibly wrong at the same time. The line now measures 99.6 per cent. Reported by alan_gr on the OpenStreetMap community forum.
Madrid to A Coruña began seventeen kilometres east of Madrid, ran further east still, and then jumped twenty six kilometres back to Chamartín before heading northwest. Two waypoints belonging to another corridor had been sitting at the head of the line, and every later tool preserved them: a redraw from the relation keeps the ends it is given, so the error survived being redrawn. They are gone and the route now fits its relation end to end, 98.5 to 99.8 per cent on real track. A scan for the same shape across the whole atlas found eight more lines that begin or end with a jump dwarfing the rest of their geometry. Reported by alan_gr on the OpenStreetMap community forum.
The atlas has stopped claiming to know where a line came from unless it can show you. Until today most route pages said the line was traced onto the OpenStreetMap rail graph, and that sentence was decided by counting how many decimal places the coordinates carried, because the tools that write them round differently. That is a fingerprint of which tool wrote a line last and evidence of nothing about where the line is. It is removed rather than reworded, so 768 pages lost a badge they could not support. What replaced it is better: 1,026 routes now name the OpenStreetMap relations they are drawn from, node for node, with the kilometres each one covers, and the reader can open them. A route is usually several railways end to end, so naming one of them was true of part of the line and read as true of all of it. Found by the atlas’s own audit sweep.
Every line in the atlas has been measured against real railway. Each one was walked in ten kilometre legs on a router that can only travel along OpenStreetMap track, and the drawing compared with what came back, in metres. 947 lie on track from end to end, at a median of four metres from it. 199 diverge somewhere, which is not the same as being wrong, because where a high speed line runs beside its classic one the router can pick the other. 89 have at least one point with no railway within three hundred metres, confirmed against OpenStreetMap rather than taken on the router’s word, and those are a work list rather than a number to hide. The funiculars, forest railways and narrow gauge lines the router has no graph for were measured against the track OpenStreetMap maps around them instead. Two routes could not be measured at all, and say nothing rather than something weaker. Found by the atlas’s own audit sweep.

5 August 2026

Five cards got the right picture, and one of them explains why the others were wrong. The Rauma Line showed a preserved steam engine standing in a yard rather than the BM 93 railcar that runs it through Romsdalen. Sarajevo to Banja Luka showed a black and white Spanish prototype from the 1940s, matched to the route on the word Talgo and nothing else; it now shows a real train on the Doboj to Banja Luka section the service runs over. The Dombóvár to Bátaszék line and the EuroCity to Zagreb both showed station buildings with no train in them, and now show the Bzmot railcar that works the branch and a MÁV InterCity under way on the Gyékényes corridor. Sofia to Varna was wearing the same photograph as Sofia to Plovdiv, which two routes may not do, and now has a class 44 on a passenger train at Varna itself. The common cause: the picture hunt read one Wikipedia article, usually the English one, and the best photograph of a Bosnian or Bulgarian line is almost never there. It now walks the subject to every language edition that has an article on it, and reads the Commons categories too. Found by the atlas’s own photo audit.
The Málaga to Sevilla card was cropped through a cliff. It carries what may be the best photograph on the atlas, a Renfe railcar in the gorge at El Chorro beneath the Caminito del Rey, and the automatic crop centred on the rock face above the train because rock is the busiest thing in the frame. The card now shows the train. Worth recording as a limit of the method rather than a one-off: the crop finder scores each row of pixels for edge energy and colour, which finds a train against sky and loses one against a mountain, and it reported its own uncertainty on this photograph as the lowest of the six measured that day. Where the picture is dramatic and the train is small, a person still has to look. Found by the atlas’s own crop measurement.
Madrid to Murcia was drawn on the wrong railway, and this log said otherwise. On 29 July it was recorded here as redrawn onto the Levante high-speed line through Cuenca. That fix never reached the file: the line shipped all along down the nineteenth century broad gauge route through Alcázar de San Juan, passing 89 km from the Cuenca it claimed to serve. It is now built from the four OSM relations the service actually runs over, the Atocha to Chamartín link, the Torrejón de Velasco trunk, the Levante line and the Monforte del Cid branch, and it passes Cuenca at 1.7 km, Albacete at 0.6 and Villena at 1.6. The card also said 480 km, which was Alicante's number copied across. Murcia's branch leaves the Alicante line 28 km short of Alicante and runs 69 km further south, so it is the longer journey: 529 km, measured off the line itself. A correction that claims more than the map delivers is worse than the error it describes, which is why this entry names what went wrong rather than just what changed. Reported by alan_gr on the OpenStreetMap community forum, twice.
Patras gets its two suburban lines, and the reason they were refused was an answer to a question nobody asked. The refusal said no train runs Rio to Kato Achaia through, which is true and beside the point: the report was for two separate services radiating from Agios Andreas in opposite directions, and OpenStreetMap carries a route relation for each of them. Both are now on the map. Π2 runs eight metre-gauge kilometres north along the gulf to Rio under the Rio-Antirrio bridge, reopened with Agios Andreas on 9 July 2010. Π1 runs twenty west down the old Pyrgos line through the Achaean beach villages to Kato Achaia, which got its trains back on 29 February 2020. Each card is dated by the year its own service reached the end it is named for, which is the reading of start years this forum argued for. One honest note on method: the timetable feed the atlas normally gates new routes on does not cover Greek suburban services at all, so this went in on the strength of the operator's own line brands, the reopening dates and the OSM relations, not a feed check. Neither line has a photograph yet; the only ones on Commons show empty platforms, and an empty platform is not a railway. Reported by jimkats on the OpenStreetMap community forum.
The Sevilla to Huelva card said the neo-Mudéjar terminus at Huelva still stands. It does stand, and that was the most flattering way to put it. The building has been out of use since 2018, was closed for good in March 2019 and lost its roof to a fire in August 2022. The card now says so. Reported by alan_gr on the OpenStreetMap community forum.
Every card photo is now cropped where the picture actually is. The hover card shows a strip a fraction of the height of a 3:2 photograph, so the crop decides whether you see the train or the sky above it. 337 photos had a measured focal point and the other 883 were cut through the middle on a blind guess that they sat slightly low, which is right for a level railway shot and wrong for most others. All 1,220 were measured by scoring each pixel row for edge energy and colour and sliding the visible band to the best position. Measured, 827 of them want their true middle. The 48 focal points set one by one in the July audit were left alone, because a focus chosen for that picture beats a heuristic. Two things also left the card while it was open: the three pixel accent bar across the top and the gradient that faded the photograph into the panel, so the picture now reaches the card's edges and starts it. Found by the atlas’s own audit sweep.
A new page puts the whole atlas in one sortable table. The other reference pages each rank it on a single number and hide the rest, fastest, highest, longest. This one shows distance, scheduled journey time, realized average speed, top speed, opening year and ridership for all 1,231 routes at once, sorts by any of them and filters by route, city or country. A dash means a figure nobody publishes rather than a zero, and it sinks to the bottom whichever way a column is sorted, so an ascending sort can never present "unknown" as the smallest number. It is at trainrouter.com/stats/. Found by the atlas’s own audit sweep.
Every photograph on the atlas now names the person who took it. It used to say "Photo · Wikimedia", which credits a warehouse for work done by a photographer, and the licences these pictures ship under ask for better than that. The author and licence were read from each file's own record on Wikimedia Commons: 4,163 pictures, 4,107 of which name someone. Route pages carry the name and the licence under the hero, every gallery thumbnail now carries its own line rather than one shared note at the bottom of the section, and the map card names the photographer. The prompt for this was finding out that two of the photographs on the atlas, the Málaga to Sevilla line at El Chorro and the Algeciras train in the Serranía de Ronda, were taken by the same reader who had reported both of those cards as wrong. Prompted by alan_gr on the OpenStreetMap community forum.
The map has a search box. Until now the only way to reach a route you already had in mind was to guess where it lived and hover the right line, because the filters answer what kind of journey you want and never which one. Type a route name, a city or a country and the answer is three keystrokes away: name matches first, then endpoints, then countries, famous routes leading each group. Picking one flies the map to it and opens its card, and it searches the whole atlas rather than the part currently lit, so asking for the Orient Express finds it instead of quietly hiding it behind a switched-off filter. Found by the atlas’s own audit sweep.
Five more Welsh lines join the atlas, from the network map its reporter linked. The Rhymney line climbs its own valley from Cardiff through the tunnel under Caerphilly mountain to the head of the valley, on metals the Rhymney Railway opened to passengers in March 1858. The Ebbw Vale line is a coal railway carrying people again, shut in 1962, back to Ebbw Vale Parkway in February 2008 and up to Ebbw Vale Town in May 2015. The Maesteg line turns off the South Wales main line at Bridgend and runs up the Llynfi valley through Tondu, reopened in 1992 after twenty two years without a train. The Vale of Glamorgan line takes the coast to Bridgend through Barry, Rhoose for the airport and Llantwit Major, closed to passengers in June 1964 and reopened in June 2005 with two new stations. The Borderlands line crosses the Dee on the Hawarden swing bridge from Wrexham to Bidston for Liverpool, worked now by the Class 230 units rebuilt from London Underground stock. Each one was drawn by taking the station list of the train that actually runs it from the operator's published timetable, then snapping that chain onto the OpenStreetMap rail graph: the worst of the sixty eight stations on the five lines sits 110 m from the line drawn through it. The Swansea to Cardiff stopping service was looked at again and is still not here, because it runs over main line rails this atlas already draws and the rule is that no railway gets drawn twice. Reported by ceirios on the OpenStreetMap community forum.
The Limerick to Galway card called itself the only intercity route in Ireland that never sees Dublin, which the atlas itself disproves one card away. Limerick to Waterford is on this map and does not go near Dublin, and Cork to Tralee runs direct several times a day without being on the map at all. The card now describes the line instead of ranking it: the Western Rail Corridor, reopened in March 2010 after thirty four years, north from Limerick through Ennis and the edge of the Burren to Galway Bay. Cork to Tralee stays off the atlas for the reason it always was, that both halves of it are already drawn, as Cork to Mallow on the Dublin to Cork line and Mallow to Tralee on the Dublin to Tralee line, and the rule against drawing one railway twice is worth more than one extra card. Reported by alan_gr on the OpenStreetMap community forum.
Six routes were linking to a Wikipedia article about something else. The warning was general, that one identifier can end up standing for two different routes, and it turned out to be true here in the most ordinary way: sixty two articles on this atlas are shared by more than one route, and almost all of that sharing is correct, because one Railjet article really does describe six railjets. The failures are the ones where two things share a word. Al Andalus, a Belle Epoque cruise train, was linked to the high speed line it parks on. The Intercity Riviera from Milan to Ventimiglia was linked to the Night Riviera, which is a sleeper to Cornwall. The Mountaineer in New Hampshire was linked to the Mount Washington Cog Railway, a different railroad on the same mountain. The EuroCity through the Gotthard was linked to the Panorama Express, which is the tourist train. The AVE lines to Alicante and to Zaragoza were both linked to a nineteenth century company whose name happens to contain all three cities. All six now have no link at all, which is this atlas's standing rule: no link beats a wrong link. The six judgements live in the script that resolves these links, so the next run cannot quietly restore them. Reported by NaufalFarras on the OpenStreetMap community forum.
India's longest train was drawn with a 228 km ruler across Andhra Pradesh. The Vivek Express ran down the coast to Vizianagaram, jumped back north east to Brahmapur and returned, because two waypoints placed before the line was traced had been left sitting after the traced run instead of inside it. Both duplicate points the line already passes. Removing them took 400 km of imaginary railway off the map and brought the drawn length from 8 per cent over the published figure to 2 per cent under it. The check that should have caught it was looking at the wrong thing: it flags a straight segment more than twenty times the route's own typical step, which is right for a finely drawn line and blind on a coarse one, because the coarser the drawing the bigger a ruler has to be before it stands out. On a 4,500 km line stepping 13 km at a time, 228 km is only eighteen times the step. The check now also reports any straight segment over 60 km whatever the route's own median says, and that immediately named thirty five more, nearly all of them the long distance lines this atlas has always drawn most coarsely. They are now visible in the audit instead of hidden by a threshold. Found by the atlas’s own audit sweep.
Every traced line was checked against the OpenStreetMap relation that Wikidata names for it, and three narrow gauge lines were redrawn from those relations. Someone on the OpenStreetMap forum guessed that this atlas reads rendered map tiles and copies the shapes it sees. It does not: it asks Overpass for the railway ways themselves and routes along their nodes. But the tracer does find the right relation by NAME and corridor, which is a fair thing to be sceptical of, so this checks it by identifier instead. Each route goes to its Wikipedia article, to the Wikidata item behind it, to the OpenStreetMap relation id recorded there, and the drawn line is measured against that relation every two kilometres. Of 225 routes with usable geometry on both sides, 168 sit within a kilometre of the relation for their whole length. Another 36 follow it and then continue past its end, which is not an error but this atlas's own rule that a card spans the SERVICE while a relation describes the LINE: the Settle to Carlisle line sits 90 metres from its relation at the median and 40 km at the far end, because the train keeps going. That leaves 21 genuinely somewhere else, and they split three ways. Two are Wikidata's error rather than the map's, the Tanzanian Central Line pointing at a relation 7,162 km away. Several more name a neighbouring line rather than the one the service runs on. And seven were short narrow gauge and heritage railways that the corridor tracer has always drawn worst, because the rail router has little or no graph in those places. Three of those seven could be rebuilt straight from their relation and now are: the Lillafüred forest railway, the Šargan Eight and the Tren del Fin del Mundo. The other four were refused by the length and endpoint gates, honestly, rather than forced: the Yangon circular relation is twice the length of the loop the card describes, the Maeklong one is a two node fragment, and Szilvásvárad and Montenvers disagree with their cards by more than the gate allows. Prompted by jimkats and NaufalFarras on the OpenStreetMap community forum.
The number of routes checked against a live timetable went from 264 to 390 in an afternoon, and almost none of that was new checking. It was a test in this atlas's own code applied the wrong way round. A journey planner whose published shape is coarse cannot be used to CONDEMN a line, because a shape drawn as straight hops between distant stops makes every real curve look like a defect, and that guard has been here since the checker was written. It was also refusing to CONFIRM. So 131 European routes sitting a median 150 metres from their reference, with every station on them, were reported as unjudgeable rather than as right. Agreement that close is not something a coarse reference produces by accident: a straight hop between distant stops leaves the real railway by kilometres, so a line matching one to 150 metres is on that railway, and the station test rules out a parallel one. The guard now stands only where the evidence is not already positive on both counts. Retrying the remaining 421 European routes on a second date then confirmed 133 more. Found by the ÖBB Scotty re-sweep of every unverified European route.
The European Sleeper to Prague was drawn on the wrong German railways for 236 km of it. The retry caught it because it finally matched the sleeper itself rather than a daytime alternative: ES 453, no changes, its own length within two per cent of ours, and our line wandering up to 37 km off it east of Berlin and again near the Dutch border. It is redrawn from the train's own shape and now sits within 100 m of it the whole way, with every calling point on the line. Two other flags from the same retry were false alarms of a kind now closed off: the planner had answered a sleeper card with a two change daytime ICE through Cologne and Frankfurt, and an Intercity Direct card with a Eurostar coupled to a EuroCity. A reference that changes trains is a different journey and can no longer report a fault against a card that names one service. Found by the ÖBB Scotty re-sweep of every unverified European route.
Brittany has two arms and the atlas only drew one. Paris to Brest was here; Paris to Quimper was not, although they are the same train for most of the way. The TGV runs out of Montparnasse as one set and splits at Rennes, one half north through Saint-Brieuc and Morlaix to Brest, the other south to the Morbihan coast through Vannes, Auray and Lorient to Quimper in Finistère, 607 km and 3 h 33 at best. The new card shares the Brest card's exact points from Paris as far as Rennes, because as far as Rennes it is the same railway and often the same train, and only becomes its own line where the two halves part. Reported by trial on the OpenStreetMap community forum.
The Milford Haven line was never drawn. Comparing every European route against the running timetable turned up six suspects, and this was the worst of them: the card added on 2 August for Swansea to Milford Haven carried the Pembroke Dock branch instead, all seventy six points of it, identical to the Pembroke card down to the last decimal. The two branches part at Whitland and the copy went the wrong way, so the atlas drew one railway twice and left the other undrawn. It now runs the line it names, through Clunderwen, Clarbeston Road, Haverfordwest and Johnston to Milford Haven, and ends 150 m from that station instead of 7.5 km away. While both were open the shared stretch from Swansea to Whitland was put on one set of points, because two tracings of one railway drift apart at zoom and that is the fault a reader found in Dublin last week. Two more of the six were real and are redrawn from the operators own shapes: the Vouga metre gauge line stopped 9.5 km short of its own terminus at Sernada do Vouga, and the Barcelona to Portbou regional left its corridor after Girona, missing Celra, Bordils, Flassa and Sant Jordi by four to nine kilometres. Found by the ÖBB Scotty sweep of all 730 European routes.
Three of the six suspects were acquitted, and the reason is worth stating because it is the same reason the other three were convicted. Calling points decide it. The Eurostar from Paris to Amsterdam passes all seven of its stations within 200 m, and the Kyiv to Uzhhorod night express passes its stations within a kilometre; where those lines differ from the planner is on stretches with no station on them, which is where two real routings can both be right. Kyiv to Lviv is the same story with one blemish, five of its six calling points exact and a single intermediate stop 12.9 km away. A line that serves every station of the service it names is not on the wrong railway, whatever the average distance between the two drawings says, so none of the three was redrawn on that evidence alone. Found by the atlas’s own audit sweep.
The Avlo from Madrid to Barcelona is gone, because the train is. Renfe stopped running its low cost brand on that route in September 2025 and has not brought it back, so the card described a service that no longer exists. Checked before removing it: on a Tuesday this month the Madrid to Barcelona line carries AVE, Iryo and Ouigo, and no Avlo at all. The reporter also made a second point that is worth more than the first, and the atlas agrees with it: the card should not have existed anyway. Avlo runs on the same rails as the AVE, so the map drew that railway twice, which is exactly what this atlas says it does not do. Following brand variants down to every slightly different service on a line drawn already is a bottomless pit, in his phrase. The old address forwards to the AVE card rather than breaking. Reported by alan_gr on the OpenStreetMap community forum.
Every route page now ends by saying what has actually been checked about it, and refuses to round that up. Three claims are possible and each is a different kind of evidence: the line was traced onto the OpenStreetMap rail graph, or it was drawn from the operator's own published service shape; it was planned on a live journey planner and measured against the shape of the train that came back; and it has a Wikipedia article. A route that cannot show one of these shows nothing rather than a softer word for it. The counts, which are on the about page: 981 of 1,247 lines traced from OpenStreetMap, 170 from an operator shape, 96 whose source this atlas cannot demonstrate, 264 checked against a running timetable, 1,143 with a Wikipedia article. Keeping the two geometry sources apart was the whole difficulty. Nothing in this repository recorded which line came from which tool, and reading it out of the commit history would have been guesswork, because single commits mixed both methods. The answer turned out to be sitting in the coordinates: the three programs that write route geometry round to different precisions, three decimals for the OpenStreetMap tracer, four for the redraw from an operator shape, five for the chord filler, so the rounding is a fingerprint of which one wrote a line last. That inference has one honest weakness, which is that coordinates typed in are coarse too, so a coarse line that is also sparse is reported as unrecorded instead of being credited to a tool it never touched. Found by the atlas’s own audit sweep.
Eleven routes added, and the way they were found is the point. Rather than guess at gaps, every candidate corridor was tested twice: is any line already drawn near both of its ends, and does a direct train actually run it. Most of the first guesses failed the first test, because the atlas already had Warsaw to Gdansk, Krakow to Zakopane, Bratislava to Kosice, Budapest to Pecs, the Santa Claus Express, the Sorlandsbanen, Riga to Daugavpils, the Highland Main Line and Belfast to Derry. Thirty two famous scenic and mountain lines were probed the same way and every one was already here, so the gaps turned out to be ordinary intercity, not spectacle. What survived both tests: the EuroCity from Zurich to Munich, which is 3 h 44 since the line through Lindau was electrified in 2020; the whole east coast of Sicily from Messina under Taormina and Etna to Siracusa; the Intercity from Amsterdam through Rotterdam and over the Hollands Diep to Vlissingen; Bucharest to Sofia over the 1954 Friendship Bridge at Giurgiu; Warsaw to Szczecin; Sofia to Vidin up the Iskar gorge; the sleeper from Kyiv through Lviv to Chernivtsi; Den Helder to Nijmegen coast to border; Zagreb east across Slavonia to Vinkovci; Porto up the Ave valley to Guimaraes; and Antwerp to Hasselt. Each one was drawn from the station list of the train that runs it and then snapped to the OpenStreetMap rail graph, and each was checked back against the operator timetable afterwards. That check earned its keep on the last of them: the Antwerp to Hasselt trace took a 97 km path and missed Diest by 14 km, so it was thrown away and redrawn from the service's own shape at 76 km. Kyiv to Chernivtsi was drawn from the operator shape from the start, because 831 km hung on five station pins is not enough for a tracer to find the right railway. Two departures decided their own cards: the Kyiv train leaves at 20:53 and arrives at 08:00, so it is a night train, and the Bucharest train runs 10:46 to 20:41, so it is not. Found by the atlas’s own audit sweep.
The Canyon Spirit now runs to Salt Lake City, and the atlas stopped it at Moab. The route opened in 2021 as Rocky Mountaineer's first American train, was taken over by the sister brand Canyon Spirit on 21 April 2026 and extended the same season, so it is a three day journey now rather than two, with hotel nights in Glenwood Springs and Moab. The card said 570 km to Moab; it is 911 km to Salt Lake City, climbing Price Canyon under Castle Gate to Soldier Summit and dropping into the Utah Valley at Provo. Two things worth stating plainly. Moab has no station: the railhead is a siding off Highway 313, a motorcoach covers the last miles into town, and on the third day passengers rejoin the train at Thompson Springs, so the line on this map runs 41 km east of Moab rather than into it, and the card says so instead of drawing a railway that is not there. And this train shares its whole route with the California Zephyr, which was drawn separately, up to 1.4 km away from it in the same canyons and not sharing a single point. That is the fault a reader found in Dublin two days ago, where the DART and the Rosslare line scissored apart at Tara Street. Both now use one set of points, and since the new tracing is the better of the two, the Zephyr was moved onto it rather than the other way round: its Denver to Salt Lake City stretch went from 86 points to 243 and every station on it now sits within a kilometre of its own line. Found by the sweep for Wikipedia links pointing at the wrong article.
Checked and left alone: the Wien to Lienz railjet really does go the long way round through Graz. It looks wrong on the map, because the natural way to Carinthia from Vienna was always south west through Bruck an der Mur and Sankt Veit, and this line swings east to Graz first. Since the Koralm tunnel opened in December 2025 that is the route the train takes, and the operator's own timetable spells it out: Wien, Bruck an der Mur, Graz, Weststeiermark, Sankt Paul im Lavanttal, Klagenfurt, Villach, then up the Drau valley. So it passes through Klagenfurt as well as Graz, not instead of it. The drawn line sits a tenth of a kilometre from that train's own published shape for the first 400 km. Two railjets a day run through to Lienz in each direction, which is what the card claims. Found by the atlas’s own audit sweep.

4 August 2026

The funicular shelf became the Mountain shelf, and the rack railways moved onto it. The old category was mechanically correct and practically useless: it held 53 cable-hauled inclines, most of them a few hundred metres of city hillside, while the Jungfraubahn, Pilatus, Snowdon, Pikes Peak, the Gornergrat and the Nilgiri sat filed under Scenic or Heritage as though they were ordinary railways that happened to be pretty. They are not. A rack railway and a funicular are two answers to one question, how do you get a train up a slope too steep for a wheel to grip, and a visitor pointing at Snowdon and at Stoos means the same thing by both. So 31 rack and cog lines joined the 53 cable ones under Mountain. The test is mechanical rather than scenic: rack or cable must be how the line makes its climb. A steep adhesion railway is still an ordinary railway, so the Bernina keeps its 7 per cent and its Scenic label with no rack at all, and a valley transport line with a short rack assist, the Mont-Blanc Express and the Berner Oberland-Bahn, stays where it was. Three lines whose rack was lifted decades ago, the Rübelandbahn, the Erzbergbahn and the Da Lat stub, are judged on what they run on now and stayed put. Aerial cableways remain out entirely; a gondola is not a railway. The category is also lit by default now, which the funiculars never were, because hiding a shelf of city inclines cost nothing and hiding the Jungfraubahn would not. The hub moved from /funicular-railways/ to /mountain-railways/ and the old address forwards. Found by the atlas’s own audit sweep.
Every route now publishes the speed it actually keeps. Distance divided by the fastest scheduled time is the number that decides how long a day takes, and it is typically 60 to 80 per cent of the headline figure a train is certified for: the Tokaido Shinkansen is a 285 km/h train that averages 219, and the Trans-Siberian averages 55 over seven days. 1,205 of the 1,229 routes carry it, on the card and in the facts on every route page. The other 24 refuse it on purpose rather than print a number that lies. A hotel train stables overnight and spends days off the rails, so the Palace on Wheels would read 12 km/h and Rail Cruise Haerang 21, describing an itinerary rather than a journey; a round trip figure covers twice the drawn line. Scheduled trains that simply run for days, the Canadian and Vivek Express among them, are never not under way, so they keep theirs. Checked before shipping: not one published average exceeds its own route’s top speed. Found by the atlas’s own audit sweep.
Photographs are cropped less and aimed better. The card strip and the route page hero were both cut to a fixed shape that suited a desktop window, so on a phone, which is the taller screen, they threw away most of the picture: the hero is now 4:3 on a narrow screen instead of 16:9, the list thumbnails 3:2, and the card strip grows with the card rather than staying 96 pixels tall. Where a photo has no measured focus point the crop now falls slightly low instead of dead centre, because the centre of a level railway photograph is usually sky. And six Indian routes had pictures that showed no train and no railway at all: the Konkan Railway was a patch of dry scrub, the Maharajas’ Express an empty desert, and the Vivek and Grand Trunk expresses were photographs of their own name boards. All six now show the train or the line, among them the toy train in the forest at Matheran and a WAP-7 coming out of the fog with the Grand Trunk behind it. Found by the atlas’s own audit sweep.
Four cards moved to the shelf they belong on. El Expreso de La Robla and the Namibian Desert Express are hotel trains and now sit under Luxury beside the Transcantábrico and Al Ándalus they are siblings of. The Fairy Queen, an 1855 locomotive on a heritage excursion, moved the other way, out of Luxury and into Heritage where the ride being the product is the whole point, and the Schynige Platte railway joined it: 1914 stock climbing to an alpine garden with no village at the top. The funicular category was audited at the same time and nothing needed moving into it: every cable-hauled counterbalanced line in the atlas was already there. What the audit did settle is that the shelf was named too narrowly, which is dealt with in the next entry. Found by the atlas’s own audit sweep.
The complaint that some of the atlas’s writing reads vague and machine-made was taken seriously enough to answer with a full audit: every one of the 1,229 route summaries was read in one sitting. The finding, honestly reported, is that the disease was real but not epidemic. The quoted offender, Dublin to Cork’s premier line since the Famine years, was rewritten this morning, and the read turned up a handful more leaning on the same crutch, which are now fixed: Algeria’s and Bangladesh’s flagship expresses and Belgium’s Viroin valley steam line say what they are instead of claiming premier status, and the same word fell out of two route stories. The rest of the corpus carries dates, engineering and places rather than atmosphere, and was left alone. The atlas also now explains on its about page how routes are chosen, because two readers in two places asked the same fair question: there is no algorithm, the selection is human, and thin coverage of a country is a to-do rather than a verdict. Prompted by alan_gr on the OpenStreetMap community forum, and by jimkats, Rokin and Obi asking how routes get picked.
The Jungang line card caught up with December 2024. It ended at Andong, but since the last stretch of doubled track opened on the twentieth of that month the KTX-Eum has run the whole length of the country, Seoul to Bujeon station in Busan in 3 h 56. The card now spans the full journey: the extension rides the realigned line through Uiseong and Yeongcheon to Gyeongju, checked against every station it calls at, and from Gyeongju to Busan it shares the Donghae line card’s exact drawn points, so the shared railway is one line on the map rather than two. A first attempt traced a plausible-looking relation that turned out to miss Andong by 68 kilometres; the station-by-station check caught it before it shipped, which is what the check is for. Found by the atlas’s own audit sweep.
The Irish line in the water is fixed, and the diagnosis was the reporter’s. The waypoints were right all along, within 100 metres of the operator’s own published shapes. What showed in the sea was simplification: route lines were thinned to a 150 metre tolerance for drawing, so as the basemap gained coastline detail with every zoom step, the line kept its coarse corners and cut across the coves. Drawing tolerance is now 50 metres, and the Dublin coast routes to Rosslare, Greystones and Galway are redrawn from their operators’ shapes at that fidelity. The DART also now reuses the Wexford line’s exact points along the shared coast, so the pair cannot drift apart again. Reported by alan_gr on the OpenStreetMap community forum, pinned down by the maker’s observation that the line stops following the map as you zoom in.
South Korea gains the newest railway the country has opened: the Jungbu Inland line, complete from Chungju through to Mungyeong since 30 November 2024 and worked by KTX-Eum sets running down from Pangyo. 144 km along the rails in about an hour and a half, traced from the line’s own OpenStreetMap relation and drawn through all ten of its stations, past Chungju and the Suanbo hot springs to the old post road town under Mungyeong Saejae. Reading the country properly settled two other things. The KTX to Yeosu on the Jeolla line was already here, so it was never the gap it looked like. And the Jungang line KTX-Eum, whose card still ends at Andong, in truth runs the length of the country to Bujeon in Busan; that re-span is the next one. Raised by humble_boatsman on Lemmy.
Galicia stopped looking cut off. The Madrid AVE card ended at Ourense, which is where the high speed line ends but not where the train does: the S-106 Avril runs on through Santiago de Compostela to the Atlantic at A Coruña. The card now spans the whole journey, 646 km in 3 h 36 min, redrawn point for point from the operator’s own published shape and measuring 644 km against their 646. The rails from Ourense to Santiago and on to A Coruña are drawn for the first time, so the province is joined to the national network instead of floating beside it. Reported by alan_gr on the OpenStreetMap community forum.
Two long metre gauge journeys stopped calling themselves commuter trains. Ferrol to Oviedo, 314 km and 7 h 15 min, and Bilbao to Santander, 119 km over 33 stations, were both labelled Cercanías AM, which in Spain means a short suburban service. Renfe runs them as Media Distancia, lines R-1f and R-3f, with cercanías services overlaid on parts of each rather than the whole. The cards now carry the Media Distancia names. Reported by alan_gr on the OpenStreetMap community forum.
One card stopped writing like a brochure. Dublin to Cork was introduced as Ireland’s premier line since the Famine years, green midlands all the way to the Rebel City, which is atmosphere rather than information. It now says what the journey is: Heuston to Kent in 2 h 15 on Mark 4 push-pull sets, over the 1849 Great Southern and Western main line through Limerick Junction and Mallow. The wider point behind the report, that some of the writing across the atlas reads vague and approximate, is taken and is a larger job than one line. Reported by alan_gr on the OpenStreetMap community forum.
The Western Rail Corridor is redrawn on the real rails into Galway, 180 points where it had 58, and the sea crossing suspected on that approach turned out not to be one. Measured against the world coastline the line reads as wet for six kilometres east of the terminus, but it sits exactly on the railway OpenStreetMap maps there, and the coastline dataset is simply too coarse for the inner bay. Recorded rather than quietly dropped, because the report deserved a real answer either way. Reported by alan_gr on the OpenStreetMap community forum.
The Wikipedia hunt returned for the 111 routes still without an article, this time asking each country in its own script rather than in English. Twenty five more routes found their page and the atlas now links 1,142 of its 1,228 routes across twenty nine language editions: the Tehran to Mashhad line in Persian, the Tunis mainline in Arabic, the Minsk to Gomel corridor under the historic Libava to Romny railway in Belarusian, the Tbilisi Funicular in Georgian, the Katakolo to Olympia line in Greek, the Sofia trunk lines in Bulgarian, the Kyiv night trains to Uzhhorod and Warsaw under their own named articles in Ukrainian, and the Thunderbird, Haruka and Kaiji expresses in English after all, hiding behind disambiguation pages. The eighty six that remain have no article in any language, which is now a verified fact rather than a search failure. Found by the atlas’s own audit sweep.
Thirty more lines carry a published top speed, read off the rails themselves: OpenStreetMap tags the permitted speed on the track, and wherever the tagged stretch covers at least half a route and one figure dominates it, that figure is the card\u2019s. The British values arrive in miles per hour and say so, 90 mph on the Shrewsbury to Chester line, 75 on the West Wales branches. Forty six weaker candidates were rejected for thin coverage rather than shipped. Seven tourist operations also moved to the Heritage shelf where they belong, among them the Porrettana Express with its 1930s coaches, the Sagano torokko on its preserved dead alignment and the Wachaubahn, so the default map shows scheduled railways and the short strokes light up on demand. A fleet wide geometry audit closed the night: every straight line suspected of being a shortcut proved to be straight railway, and the one hand sketch left in the atlas, the Aoniyoshi, now follows its eight real Kintetsu stations. Found by the atlas’s own audit sweep.

3 August 2026

The Channel fix from yesterday did not survive the night, and it was thinner than it read. A shoreline cleanup pass that ran hours later dragged all three Eurostars back through Dover and across the strait at its narrowest point, and under that lay an older fault: the Paris train had been following the classic boat train line through Boulogne and Abbeville inside France all along, because the shape its own data feed publishes is a sketch of the wrong railway. The French leg is now traced from the LGV Nord itself in OpenStreetMap, from the Fréthun portal through the Lille fork to Paris, and the crossing enters the tunnel at the real portals, Folkestone to Coquelles. London to Paris measures 486 km against the published 492, and Brussels and Amsterdam share the drawn line to the fork near Lille, where the two services truly part. Found by the atlas’s own audit sweep.
That same cleanup pass had rewritten 121 route lines it was never asked to touch, thinning curves out of the Brenner, the Bergen line, the fresh Andalucía traces and a hundred more. Its guard only caught lines that got shorter, and the Dover detour had made the Eurostars longer. All 121 lines are restored point for point, and the six shorelines the pass was actually written for keep their fix. Found by the atlas’s own audit sweep.
Around Rosenheim the map showed a braid where the ground holds one railway. The Munich to Rosenheim, Rosenheim to Salzburg and Rosenheim to Kufstein corridors are each drawn once now, and every train that rides them shares the exact points: the Railjet to Vienna and the EC to Innsbruck are redrawn from their operators’ own shapes, and the Brenner railway, the night trains to Rome and Zurich and the Innsbruck to Salzburg Railjet sit on those lines rather than beside them. The Rome night train keeps its real route via Salzburg and Villach, where it joins the Vienna portion. Found by the atlas’s own audit sweep.
Oslo’s western approach is one drawn line for the Bergen, Stavanger and Skien trains as far as Drammen, and the Trondheim and Stockholm trains share their exit to Lillestrøm, so the corridor no longer splits into parallel ghosts at city zoom. The Trondheim line was also redrawn from its operator’s published geometry and got its missing curves back. Found by the atlas’s own audit sweep.
Rome to Bari and Rome to Lecce ride the same rails as far as Bari, and the map now draws them that way. The Lecce route is retraced from its Frecciarossa’s own shape, Bari shares those exact points to the junction, and the knot the two independent sketches used to tie east of Benevento is gone. Found by the atlas’s own audit sweep.
The Hefei to Fuzhou high speed line carried sixty phantom kilometres past its own terminus, out to Fuzhou South and back, ending in the middle of a city block. It is retraced from OpenStreetMap end to end, stops at Fuzhou station as the real trains do, and measures 830 km. Found by the atlas’s own audit sweep.
The Prague to Pardubice corridor is drawn once and shared by the Ostrava Pendolino, the Košice night train and the Vienna and Brno expresses, which used to cross each other in a braid east of Prague. Found by the atlas’s own audit sweep.
The Coastal Pacific is traced onto New Zealand’s Main North Line, metre for metre along the Kaikōura shore it used to wade beside. The Posázavský Pacifik was checked against OpenStreetMap and was already right. The Vienna to Bratislava REX’s dead straight run across the Marchfeld was suspected as a drawing error and verified as real track, 43 metres off the drawn line over 33 kilometres. And the Iyonada Monogatari’s card mixed up its two courses, quoting the Yawatahama distance for the Matsuyama to Iyo-Ozu run; it now says 55 km, the length of the coastal line it rides. Found by the atlas’s own audit sweep.
Checked, with thanks, and no change: Brightline West is under construction and joins the atlas when there is a train to ride. The Las Vegas Monorail and the casino trams are urban people movers, the Tucson streetcar is city transit, and Union Pacific runs freight; this atlas maps scheduled passenger train journeys. The suggestions are logged so nothing is forgotten when Las Vegas gets its train. Reported by Flap_Slimy_Outward on the OpenStreetMap community forum.
Three Norwegian lines join the atlas, each drawn from its operator’s own published geometry and proved against the live timetable: the original 1879 Østfold line from Oslo to Ski, hugging the fjord through Nordstrand, now the all stations companion to the new tunnel; the 1882 Østre linje from Oslo through Askim and Mysen to Rakkestad; and the 1917 Bratsbergbanen from Porsgrunn through Skien and Nordagutu to Notodden, which shares the drawn Sørland main line for the stretch it truly rides. Found by the atlas’s own audit sweep.
With the local line drawn, the Halden and Gothenburg trains moved onto the railway they have actually used since December 2022: the Follo tunnel, eleven minutes under the ridge from Oslo to Ski. The old shore line through Nordstrand now belongs to the local trains that serve it, and the two railways show on the map as the two distinct lines they are. Found by the atlas’s own audit sweep.
Route pages link their Wikipedia article wherever a confident match exists: 850 of the 1,232 routes now carry the link, up from 693. Every one of the new matches was vetted before shipping, and the rejects show why that pass exists; among the candidates the automatic search had proposed were an interstellar comet for the Marrakech night train, a Pantera album for the Great Southern and a 1970 crime film for the Train Rouge. Routes without a confident article keep no link rather than a wrong one. Three lines also gained a published top speed from the same sweep. Suggested by alan_gr on the OpenStreetMap community forum.
The duplicate hunt then swept every remaining route by endpoint, by name and by drawn geometry, and found the other two thirds of the same July mistake: the batch that had added Korea’s three missing KTX arteries had in fact added all three twice, because none of them were missing. KTX Seoul to Busan and KTX Honam already lived on the map under their older cards. The older cards keep their pages and inherit the best of both twins: the flagship high speed alignment through Gyeongju and Ulsan that the newer tracing had drawn correctly, the 89 million passengers a year figure, the on route photographs, and Honam’s true 370 km. The retired addresses forward to the survivors. The same sweep exposed a wrong railway nobody had reported: the ITX Saemaeul and Mugunghwa cards, classic line trains both, had been drawn on the high speed line via Gyeongju ever since a July tracing pass picked the wrong parallel track. Both are redrawn onto the classic Gyeongbu main line through Suwon, Daejeon and Miryang, 436 km against the timetable’s 442, checked station by station against where the real trains actually call. The map also draws sooner now: the route network is fetched and decoded on a second thread while the map engine does its own setup, and appears about a second earlier on a mid range phone. Found by the atlas’s own audit sweep.
That worldwide pass earned its keep twice over, because reading all 1,232 routes in one sitting exposed two cards that were the same journey twice. The Seoul to Gangneung KTX had been added in July 2026 as a missing artery when it had been on the map since the ninth of that month, and the Budapest to Bucharest Ister night train arrived a second time inside a batch of reader suggestions. Each pair shared the same termini and the same train, against the atlas rule that every railway is drawn once. The older card keeps the page in both cases, adopting the better sleeper photograph the newer Ister card had brought with it, and the two retired addresses now redirect to the surviving ones rather than break. Found by the atlas’s own audit sweep.
The Wikipedia pass then went around the world in twenty three languages. The morning had left 382 routes unlinked because their lines do not answer to an English search, so each was searched again in its own languages and every candidate judged one by one before shipping. 271 more routes now carry their article, 1,121 of the 1,232: the Ister night train in Hungarian, the Whoosh in Indonesian, the Gangneung KTX in Korean, Japan’s sightseeing fleet from the Tohoku Emotion to the steam Paleo Express in Japanese, and the narrow gauge lines of Romania and the Carpathian Tram of Ukraine in their own. The same jury then re examined the older links and twenty three did not survive it: Amsterdam to Frankfurt pointed at a London station, the Kyiv to Dnipro intercity at a Kyiv metro stop, both LNER Azuma cards at a 1935 steam locomotive class, the Mont Blanc Express at the wrong mountain railway, and the German Heath Railway at Haywards Heath in Sussex. Eighteen were moved onto the article they should always have carried and five now go without rather than wrong. Routes whose line has no article in any language keep no link at all, which remains true of most night trains across Central Asia. Found by the atlas’s own audit sweep.

2 August 2026

Every route in the atlas was verified in one overnight sweep, each one measured against the shape its operator publishes and the timetable its train keeps. 382 routes were confirmed sitting on their exact railway. Ten were proven wrong and redrawn from the operator shape the same night: the Saphir Odoriko had cut the corner at Ofuna and the Spacia X missed Kasukabe; Kyiv to Lviv and Kyiv to Uzhhorod each rode the wrong one of two real Ukrainian corridors; the Danish Lyn missed the new line over Koge Nord; the Swedish X2000 south strayed mid country; Vienna to Bratislava was drawn via Kittsee when the REX runs via Marchegg; Munich to Lindau rode the Memmingen line when this card is the Allgau route through Kempten; Oslo to Stockholm had swollen to 805 km; and Paris to Milan carried a splice overreach at Lyon. Three cards then took their distance from the shape itself. A dozen more flags were investigated and proved to be the checker fooled, not the map: both French night trains are correctly on their classic lines while the reference chained daytime services, the Arlberg reference rode a documented diversion, and the classic line references answering for high speed trains were recognised for what they are. All of it is recorded here rather than assumed. Found by the atlas’s own overnight verification sweep.
The Channel had grown a knot at Dover. All three Eurostars were drawn through Dover town and into the sea from there. The real trains enter the tunnel at Folkestone, twelve kilometres west, and the Paris line also strayed up to 96 km from its railway inside France. All three are redrawn. Paris follows its operator’s own shape, and Brussels and Amsterdam share its exact points as far as Lille. The sketched Dover Javelin was retraced too. The North Kent Javelin returns as its own card, because moving Margate to the Canterbury line had silently undrawn the Faversham corridor. Both Javelin paths to Margate now exist, as they do in the timetable. Found by the atlas’s own audit sweep.
The Brenner railway had been traced at three kilometre strides, which reads as a polygon in the Adige valley. It now follows the Railjet’s own published shape at eighteen points per kilometre, and the Venice Simplon Orient Express beside it shares those exact points for 252 km over the pass. The same sweep retraced the Dover and North Kent Javelins, the Nanyuki Safari Train, the Bangkok to Aranyaprathet line, the Shatabdi to Amritsar and the Kuranda Scenic Railway onto real track. The Nanyuki card had also said 177 km for a 237 km line. Two routes were checked and kept as they were: the Aoniyoshi, because the tracer found the wrong branch, and Cass, because its mountain spur is not yet mapped. Under it all sat one more file fault: 210 waypoint rows separated by a bare carriage return with no newline, invisible to every line based tool. All mended. Found by the atlas’s own audit sweep.
On the newest Spanish reports: the doubled Granada to Antequera lines were already gone in the build that went live minutes after the report, because the old conventional leg left with the Mr Henderson change. And Madrid to Málaga never disappeared. The card and its 511 km line are intact and live. What looked like an absence was a deploy caught mid flight. Reported by alan_gr on the OpenStreetMap community forum.
alan_gr’s fourth report in two days covered more than the DART. Limerick to Waterford described a reversal at Limerick Junction. That was true when through trains ran, but today’s journey is a change of trains there, and the card now says so. The Enterprise no longer claims to be only De Dietrich sets, since railcars work half the hourly service. The DART card names the fleet rather than one EMU class. His question about start years, service year or line year, is a fair one. Where a named service is the card, the atlas now leans service year. And the photo that replaced the wrong Mr Henderson picture turned out, delightfully, to have been taken by alan_gr himself. Reported by alan_gr on the OpenStreetMap community forum.
Mr Henderson gets his railway back, at its right length. The card had named the whole Algeciras to Granada journey after a nickname that belongs only to the British built Algeciras to Bobadilla line, and its 3 h 30 was the time to Antequera, not Granada. The card now spans Algeciras to Antequera Santa Ana, the modern junction beside old Bobadilla. One train, about three hours. The Antequera to Granada leg, which no conventional train has run for years, is gone from the card. Reported by alan_gr on the OpenStreetMap community forum.
Andalucía fills in, from a reader’s systematic list of what was missing. Five routes join. The conventional Málaga to Sevilla through El Chorro’s gorge beneath the Caminito del Rey. Almería to Granada across the Hoya de Guadix, drawn from its train’s own shape. Jaén to Cádiz, one train across all of Andalucía. Jaén to Madrid over the Despeñaperros pass. And Sevilla to Zafra into the Sierra Morena. The Avant Málaga to Sevilla he also asked for was already on the map. Three were checked and not added. Málaga to Granada and Granada to Sevilla by Avant run entirely over rails this atlas already draws, and the atlas draws each railway once. The weekly Huelva to Zafra and the currently shortened Torre de Oro wait until their timetables settle. The Linares to Almería line also stops claiming to be the only rail route into Almería, which the Granada trains disprove daily. Reported by alan_gr on the OpenStreetMap community forum.
West Wales joins the atlas: the 1856 line to Milford Haven, the Pembroke Dock branch past Tenby’s walls and beaches, and the 1906 boat train line to Fishguard Harbour, where the ferry sails for the Rosslare this atlas already draws from the Irish side. The Pembroke branch shares the Milford line’s exact points as far as Whitland, so the shared trunk is drawn once. The Swanline was checked and not added, because every metre of the Swansea to Cardiff main line is already on the map. Suggested by ceirios on the OpenStreetMap community forum.
The whole atlas went through a water audit and came out drier. Some 650 routes had points or chords sitting in sea or lake, nearly all of it too small to see, and each was snapped to land and rounded along its shore. Genuine crossings such as the Severn Tunnel, the Mälaren bridge, the Ivankovo causeway and Thane creek were left exactly where they belong. The Shiki shima cruise train, whose loop crossed the Tsugaru strait as a straight line through open sea, now runs the Seikan tunnel path both ways. The audit also earned its keep a second time: the fixing pass first flattened 410 mountain and narrow gauge lines by re rounding their curves, and Darjeeling lost a third of its length. The atlas’s own length audit caught it before anything shipped, and every flattened line was restored point for point. Found by the atlas’s water audit.
Eleven route pairs that each drew the same railway twice now share one line. The whole TGV Sud Est family, Marseille, Lyon, OUIGO, Montpellier, Barcelona and the Frecciarossa to Milan, rides a single tracing of the high speed line, redrawn first from the operator’s own shape. That redraw also restored the missing approach at Aix. Paris to Strasbourg lends its line to Luxembourg and Frankfurt. Madrid to Barcelona lends its line to the Avlo, Zaragoza and Málaga cards. Brest lends its line to Rennes. The remaining same track pairs are archived with the method that fixes them. Found by the atlas’s parallel-pair survey.
Two catches from the maker’s own screenshots. Opening a route card dimmed every other line to 16 percent opacity. That silences a teal regional line but not gold: with the Margate Javelin selected, the Eurostars beside it still read as lit, and the route appeared to continue through the Channel Tunnel into France. Dimmed lines now drop to 5 percent. The Margate card had also been quietly redrawn onto the Faversham pattern by the day’s audit while its own text still told the Ashford story. It is back on the route its description narrates: HS1 to Ashford, then the 1846 Canterbury line to the Thanet coast, 147 km, passing 10 km clear of Faversham. Found by the atlas’s own audit sweep.
The DART and the Rosslare line drifted apart a second time, and the second report exposed the real disease. Each line was correct against its own operator shape. The fault was that two dense shapes of the same coastal track were simplified independently, and two simplifications can sit up to 700 m apart between their points, worst right at Tara Street. The cure is structural, not another retrace. The DART now reuses the Wexford line’s exact points from the city centre to Greystones, so the two can never diverge again. It is the same shared geometry rule the atlas already applies to branches off a trunk. Reported by alan_gr on the OpenStreetMap community forum, whose second look caught what the first fix had papered over.
The TGV to Brest was wrong from the first metre. It left Paris from Gare du Nord, the wrong station, made two U turns inside the city and then rode the old line all the way west. The train really leaves Montparnasse and takes the LGV Atlantique and the 2017 Bretagne Pays de la Loire line to Rennes, then the north coast railway through Saint Brieuc and Morlaix. Redrawn onto that path from the OSM route relation. The distance came down from 620 to 612 km. Reported by brunoct on the OpenStreetMap community forum.
Nightjet Vienna to Rome claimed to be in service since 2021. The through night train is far older. It ran jointly under ÖBB and Trenitalia as the EuroNight Allegro Tosca from 1993, with predecessors before that, so the card now says 1993. Both Rome night train cards also say what the timetable knows: the Vienna and Munich portions run as one train south of Villach, where they are joined and split. Reported by dieterdreist on the OpenStreetMap community forum.
The Linha do Algarve card stopped at Faro, drawing only the western half of the line. The eastern half through the Ria Formosa lagoons to Tavira and Vila Real de Santo António is now drawn, and the whole line is one card. It has been one journey since through trains started running past Faro in July. The fastest through train in the current timetable does the 140 km in 2 h 20, and the drawn line sits on the operator’s own published shape to within 100 m. Suggested by alan_gr on the OpenStreetMap community forum.
Three more Greek routes, again the day a local asked. The Proastiakos card now runs past Kiato along the Gulf of Corinth to Aigio, where its trains have gone since 2020. That is 197 km measured along the rails. Athens to Chalkida joins, drawn from its own train’s published shape down to the Euripus strait. So does the Piraeus to Athens Airport line, the Olympic era link east along the motorway median, redrawn from the operator’s shape after a first sketched attempt cut corners. Suggested by jimkats on the OpenStreetMap community forum.
The Konkan Railway card stopped at Goa, though the line it is named for runs on down the coast. The card now spans Mumbai to Mangalore, about 914 km and 16 hours on the through trains. The retrace that extended it also repaired what it found: the old drawing measured 741 km to Goa where the journey is 581, wandering off the alignment. The new line sits on the OSM rail graph for 90 percent of its length. Reported by StillAlive on Lemmy.
Leeds to Morecambe joins the atlas: the 1850 Little North Western through Bentham country. Traced from the Bentham Line’s own OSM relation, 113 km exactly against the published distance, with the timetable’s fastest through train at 2 h 05. Suggested by FluidBeef on Lemmy.
Four more reports were checked and needed no change, which is worth recording. The Keighley and Worth Valley Railway is already on the map, under Heritage, a category that is off by default. The Ghan and the Indian Pacific were already filed under Luxury, exactly where a reader said they belonged. Cork to Tralee stays absent as a card of its own, because every metre of it is already drawn and the atlas draws each railway once. And the Patras suburban railcars were checked but not added. No train runs Rio to Kato Achaia through. The two overlapping legs meet at Agios Andreas, and this atlas does not draw a through card where no through train exists. Raised by FluidBeef and mcheva on Lemmy, and by alan_gr and jimkats on the OpenStreetMap community forum.
The relation search now reads the link OSM itself keeps. Mappers tag route relations with the line’s own Wikidata item far more widely than Wikidata points back at OSM: 314 of this atlas’s 690 line items sit tagged on a rail relation, against 220 the other way, and about 130 lines can only be found from the OSM side. So the tracer asks for that tag before guessing by name, and it now accepts funicular, narrow gauge, monorail and light rail relations too. The retrace sweep this powered redrew nine short drawn routes onto their real track, among them the Darjeeling Himalayan, whose loops had been flattened by rounding, Tren a las Nubes, the Hiram Bingham, KTX Eum Seoul to Andong, the Cox’s Bazar Express and the Hue to Da Nang crossing of the Hai Van pass. Suggested by Woazboat on the OpenStreetMap community forum.
Shrewsbury to Chester joins the atlas: the 1848 border line over the Ceiriog at Chirk and the great Cefn viaduct, through Wrexham and down the Dee to Chester. Traced from its OSM relation and finished into Chester station past Saltney Junction, where the relation itself stops. Asked for by trigpoint on the OpenStreetMap community forum.
Three Greek cards say honest years, and the Meteora line stops pretending. Piraeus to Kiato reads 2005, when its suburban service began, not 2010. Athens to Chalkida reads 2009, the year the suburban service started, though the branch’s tracks date to 1904. And the Kalambaka card no longer implies a through train from Athens that does not exist. It is now the Palaiofarsalos to Kalambaka branch it really is: 80 km, about two hours in the current timetable, redrawn from the railcar’s own published shape. That redraw also put the line through Palaiofarsalos station properly for the first time. The old drawing had missed the junction by seven kilometres, and the 300 km it used to share with the Athens to Thessaloniki card are no longer drawn twice. Reported by jimkats on the OpenStreetMap community forum, his second round of corrections in two days.
A water audit now sweeps every drawn line against the coastline, and its first run caught a fault the same afternoon it was introduced. The Konkan Railway’s new Mumbai end crossed the harbour in a straight line, left behind by a router failure earlier that day. It now runs round the water through Thane and Panvel, as the trains do. The other full atlas checks came back clean. Every straight looking segment is verified real track, and the one line still cutting corners offshore, the Shiki shima cruise loop, was recorded as a known limit rather than left silently wrong. Found by the atlas’s water audit.
The same audit caught six cards whose distance the service itself disproves. Measured against the operators’ own published shapes: IC61 Basel to Interlaken is 160 km, not 200. ICE Frankfurt to Stuttgart is 184, not 210. The Salzkammergutbahn is 107, not 130. The Côte Bleue line is 53, not 61. The Nightjet to Amsterdam is 840 measured along its drawn rails, not a round 1,000. And KTX Eum’s 268 km predated the Jungang line’s straightening; the relation traced line measures 221. The Glacier Express was also missing its run into Chur and back out, and Łódź to Kutno cut corners. Both are redrawn from their operators’ own shapes. Found by the atlas’s own audit sweep.

1 August 2026

Dublin to Waterford trains call at Kilkenny by running up a spur to MacDonagh station and reversing back out to the Lavistown junctions. The map drew the junction triangle without the side trip. The spur is now drawn both ways, as the train runs it. The same pass added the short reversals the map had skipped at Killarney, where trains back out through the headshunt, and at Limerick Junction. The Waterford card also carried 158 km and 1 h 55 min. Measured along the rails as the train actually runs them, the route is 179 km, and the fastest train in the current timetable takes 2 h 10 min. Reported by alan_gr on the OpenStreetMap community forum.
The Limerick to Ballybrophy line joins the atlas: the Nenagh branch, built in stages from 1857 and through when the Birdhill to Nenagh gap closed in June 1864. Two trains a day each way, about 1 h 45 min over 92 km, drawn from the service’s own published shape and verified against the live timetable. Suggested by alan_gr on the OpenStreetMap community forum.
Four routes joined the atlas the day locals asked for them on the OpenStreetMap forum. Thessaloniki to Florina, on the 1894 Salonika to Monastir line, verified against the timetable as a direct three hour run. Sevilla to Huelva, the 1880 ore line to the Atlantic, with its direct Media Distancia and Intercity trains. The Ballina branch, half an hour down the Moy valley to Manulla Junction. And Limerick to Waterford, three trains a day each way across the Suir valley. Each was traced from OSM rail geometry through its real stations. Suggested by jimkats (Florina) and alan_gr (the other three) on the OpenStreetMap community forum.
Dublin to Cork left from the wrong station. The line was drawn out of Connolly and around the north of the city, a loop only a few commuter trains make. Cork trains leave Heuston, like the Tralee service the map already had right. Redrawn from Heuston down the shared mainline. Reported by alan_gr on the OpenStreetMap community forum.
The DART and the Dublin to Rosslare line are one railway down the coast, but the map drew them as two, drifting up to 1.8 km apart and clipping across the sea at the headlands. Both are now traced from their own OSM route relations, so they sit on the same track and the coast keeps its shape. Reported by alan_gr on the OpenStreetMap community forum.
The three high speed routes into Málaga each stopped a different distance short of the terminus. All three now end at María Zambrano, as every train does. Reported by alan_gr on the OpenStreetMap community forum.
Mr Henderson’s Railway wore a photo of the wrong railway: a train at Benalúa de Guadix, on the Granada to Almería line east of Granada. The photo is now the daily Algeciras to Madrid train in the Serranía de Ronda, taken after Benaoján Montejaque, which is the line the page is about. The card also pretended the 1892 through journey still exists. It takes two trains today, changing at Antequera, and the card now says so. Reported by alan_gr on the OpenStreetMap community forum.
Athens to Kalambaka was drawn over the old mountain line through Bralos and shown as one through train. Neither is true any more. The remnant service on the old line below Lamia is replaced by buses, and the journey today runs the new Tithorea to Domokos line through the Kallidromo and Othrys tunnels to Palaiofarsalos, where a railcar for Kalambaka takes over. The line is retraced onto the new alignment from OSM rail geometry. The card now says two trains with a change at Palaiofarsalos, and the distance came down from 375 km to 353. Reported by jimkats on the OpenStreetMap community forum, within hours of the atlas being posted there.
The Athens to Thessaloniki mainline had the same flaw in milder form. It was mostly on the new alignment, but wandered off it around Domokos and missed Palaiofarsalos, a junction station every InterCity passes, by seven kilometres. Retraced. 502 km became 481. Found by the atlas’s own audit sweep.
The GoldenPass Line was credited to two companies when it takes three. The card named the MOB and the Zentralbahn, leaving out the BLS, which owns the standard gauge middle from Zweisimmen to Interlaken Ost and has jointly run the GoldenPass Express with the MOB since December 2022. The point behind the correction was larger. This is one brand over three railways and two gauges, not one train. That is why the card says two trains and roughly five and a quarter hours rather than quoting a through service. Reported by in_luzern on RailUK Forums.
The GoldenPass Line had no story behind it, only a one line summary of which trains run where. It now carries the history it was named for: the MOB reaching Zweisimmen in 1905 over the 1,279 m Saanenmöser, and the Golden Mountain Pullman Express of June 1931, run with the Wagons Lits company. A genuine Pullman train on metre gauge track, with four salon cars built specially for it. It lasted one summer. The Depression was at its peak, the British and American visitors it was built for never came, and the carriages ended up on the Rhaetian Railway in 1939. The gauge break at Zweisimmen that helped kill it outlived the train by ninety years. Told to me by Gordon on RailUK Forums, who also sent the 1931 poster.

30 July 2026

Three fixes that day came from the maker’s own use of the site rather than a reader. The endpoint markers that littered empty terrain as orphaned grey ticks at low zoom are now tied to route legibility. 79 of them were culled in the reported viewport, and all are back by street zoom. The EC Transalpin was re spanned Zurich to Graz, the example that set the standing rule that a card spans the service, not the track segment. And two regionals named that morning, Bolzano to Merano and Plzeň to České Budějovice, joined the atlas drawn from their operators’ published shapes. Found by the atlas’s own audit sweep.
The night train revival page named Paris to Vienna as a service back after a 14 year gap. ÖBB withdrew that Nightjet in December 2025, and this atlas has never held it, so the page was crediting the revival with a train nobody can catch. It also described Paris to Berlin as a relit Nightjet, when the Paris to Berlin sleeper in this atlas is European Sleeper’s. Both examples are now taken from the revival list itself rather than written into the prose, so the prose cannot outlive the data again. Found by the atlas’s own audit sweep.

29 July 2026

Paris to Nice was drawn through the Morvan. The line wandered off the high speed line after Melun and spent 240 km crossing Auxerre, Clamecy and the hills north of Nevers before rejoining at Le Creusot, country the TGV has never touched. It now follows the service the card names, which leaves the high speed line near Avignon and runs the classic Mediterranean coast railway through Toulon, Saint Raphaël, Cannes and Antibes. The distance came down with it. 1,088 km was the old route, and the train measures 971. Reported by nwales58 on RailUK Forums.
The Pyrenees line out of Toulouse started in the wrong place and stopped short at the far end. It began at Place François Verdier, followed the tramway across the city centre and picked up the alignment of a metro line that is still being built to reach Matabiau. At the southern end it gave up 4 km short of Latour de Carol, in the mountains above Porta. It now starts at Matabiau and ends at Latour de Carol Enveitg. Reported by nwales58 on RailUK Forums.
The AVEs to Valencia and Alicante were drawn on the old broad gauge railways they replaced. Both now follow the Levante high speed line through Cuenca. Alicante was missing it by 90 km at Cuenca Fernando Zóbel, and its distance was the old route’s 455 km rather than 480. Reported by nwales58 on RailUK Forums.
Alfa Pendular Lisbon to Porto was drawn up a line of its own, missing Entroncamento by 39 km. It now follows the Linha do Norte the train actually runs on, through Entroncamento, Coimbra and Aveiro. Reported by nwales58 on RailUK Forums.
Frecciarossa Rome to Bari left the real railway around Caserta and cut 19 km wide of it before rejoining near Benevento. It no longer shared the alignment with the Frecciargento to Lecce as it does on the ground. Both now run over the same metals between Caserta and Benevento. Distance corrected 450 to 480 km. Reported by nwales58 on RailUK Forums.
The Hokuriku Shinkansen stopped at Kanazawa. It has run to Tsuruga since March 2024, so the whole Fukui prefecture leg was missing: 125 km through Komatsu, Kaga Onsen, Awara Onsen, Fukui and Echizen Takefu, traced off the line itself. The card now spans Tokyo to Tsuruga, 576 km in 3 h 8 min. The two other faults reported on this line were checked and are not there. The line does reach Jōetsumyōkō before turning west, and it does follow the Chikuma valley through Iiyama and Nagano. Every station on the route sits within 1.8 km of the drawn line. Reported by sbracegirdle on GitHub.
ICE International Amsterdam to Frankfurt no longer says Class 406. The ICE 3neo, Class 408, took the route over in June 2024. Reported by DjDaan111 on GitHub.
Bulgaria’s fastest train was nearly twice as fast on this site as in the timetable. Sofia to Plovdiv was recorded at 1 h 30 min, which made it a 104 km/h average and put Bulgaria 39th in the world. No such train runs. The fastest through service in the published timetable takes 2 h 45 min, and the rebuilding works between Elin Pelin and Kostenets are why. Bulgaria’s fastest is now the Burgas train at 76 km/h, and the country ranks 59th. Reported by bozhodimitrov on Reddit.
The fastest trains page now carries a second ranking of its own: the fastest train in each country that never crosses a border. The first table credits an international service to every country it serves. That is the right answer to the question of what is the fastest train you can board there, but it produces a United Kingdom represented by a Franco British train earning most of its average on French track. Rather than argue for one reading, the page publishes both. Read domestically, the UK sits 13th on LNER’s London to Newcastle at 165 km/h, next to Germany at 163, which is exactly where the readers who raised it said it would land. Raised by BeatTheMeatles420, Matangitrainhater, funky_galileo, lexonid and Every-Progress-1117 on Reddit.
Four reports were checked and no change was made, which is worth recording too. Lisbon to Guarda already turns at Entroncamento and matches its train to within 100 m. Rome to Lecce was already on the right railway. It was Rome to Bari beside it that was wrong. Romania’s Bucharest to Constanța really does average 106 km/h. The timetable says 2 h 09 against the card’s 2 h 07. And Sweden’s X2000 is right at 3 h 05. The planner answers that query with a slower train via Borås, 33 km longer, which is a different railway rather than a correction. Found by the atlas’s own audit sweep.
Madrid to Murcia was the last of the Levante AVEs still drawn on the old broad gauge railway. It now follows the high speed line through Cuenca, Albacete and Villena and down the new Monforte del Cid branch. It is worth saying why this one took a second attempt. The timetable offers two real answers to Madrid to Murcia, a direct AVE from Chamartín and a longer one from Atocha that reverses at Alicante, and the first attempt redrew it as the wrong one of the two. Found by the atlas’s own audit sweep.
Clicking through to the map from a route page now lights that route up and opens its card. It used to fly you to the right part of the world and leave you there with every line drawn the same, which is no help at all if you arrived wanting to see one of them. Found by the atlas’s own audit sweep.
Warsaw to Wrocław was redrawn from its train’s published shape. Worth recording because of what it settled. The line still contains a 54 km dead straight section, and it is now provably not our error, because the operator’s own shape has it. A straight segment on this map can mean a railway nobody has traced, or a source that draws that stretch coarsely, and until now there was no way to tell those apart. Found by the atlas’s own audit sweep.
A new check does this work from now on. Every route inside timetable coverage can be measured against the published shape of the train that runs it, station by station, which is what caught all of the above. The earlier checks could only ask whether a line was plausible, not whether it was the right railway. It also reads the journey time back from the timetable, which is what exposed Bulgaria. Found by the atlas’s own audit sweep.

28 July 2026

Sixty routes that showed no top speed now show one. Regional lines across France, Finland, Hungary, Germany, Sweden, Denmark, Italy, Latvia, Estonia and Switzerland were carrying a blank where every other route states its line speed. Each figure is the published speed limit for that line, taken from Wikidata or the line’s own national Wikipedia article, and checked against the timetable so no line claims a speed the schedule already beats. Found by the atlas’s own audit sweep.
Ten top speeds corrected where the card claimed a speed the line does not have. Among them the Tauern railway, 160 down to 130. The Pustertalbahn, 100 down to 80. The Circumetnea, 70 down to 60. The Chur to Arosa line, 55 down to 35, which is what its street running through Chur allows. And the Bānītis narrow gauge, 40 down to 25. Each was changed only where Wikidata and the line’s own national Wikipedia agreed. Found by the atlas’s own audit sweep.
The Konkan Railway was drawn a quarter short: 557 km of line for a 741 km journey, cutting inland where the real railway threads the coast between the Western Ghats and the sea. It is now traced along its actual alignment through Ratnagiri, Kankavli and Madgaon. Found by the atlas’s own audit sweep.
The photo on nearly 300 route cards is now cropped on the train rather than through the middle of the frame. The card shows a wide strip of a tall photo, so a blind centre crop often framed the sky above the train or the ballast below it. Every photo was measured and the crop aimed at the subject. Found by the atlas’s own audit sweep.
The site’s own title and description had been advertising 700 plus train routes across 118 countries since the atlas passed a thousand. They now read from the atlas itself, 1,166 routes across 119 countries at the time, and are rebuilt on every deploy so they cannot go stale again. Found by the atlas’s own audit sweep.
The Cockle Train’s photo had been showing nothing at all. Wikimedia renamed the file and the link went dead. Replaced with the same locomotive at the same place by the same photographer. Found by the atlas’s own audit sweep.
The map now opens filling the screen. The opening zoom had been calculated for a flat map while the atlas actually draws a globe, which the map engine renders about three times smaller. The world arrived as a small marble with half the window empty around it, most noticeably on phones. Found by the atlas’s own audit sweep.
Twelve more regional lines now state a top speed, each one a published figure that the speed limit tagged on that line’s own track agrees with: the Rennes and Périgueux lines, Reims to Laon, Clisson to Cholet, Olten to Lucerne, Amiens to Laon, the Kystbanen, Saintes to Royan, Pieksämäki to Joensuu, Lollandsbanen, Haapamäki to Seinäjoki and Linares Baeza to Almería. Two candidates were dropped rather than published. The Seetalbahn’s 40 km/h describes the narrow alignment it ran on before 2010, and Burgdorf to Thun’s 50 comes from a sentence about the line opening in 1899. Found by the atlas’s own audit sweep.

24 July 2026

Added the Southeastern Highspeed Javelin services, Britain’s only domestic high speed trains and the other users of HS1 alongside the Eurostar that was already mapped: London St Pancras to Dover via Ashford and Folkestone, and to Margate via Canterbury West. Reported by westcoaster on RailUK Forums.
HS1 redrawn on its real alignment while adding those services: the Stratford International curve, the Thames tunnel at Swanscombe, the Medway viaduct and the North Downs tunnel. All five trains that use the line, three Eurostars and both Southeastern branches, now share it point for point. The Eurostars to Brussels and Amsterdam used to cut a straight line from London to the coast, missing Ashford by 8 km. Found by the atlas’s own audit sweep.
The about page now states plainly what this map is: a selection of the world’s notable railway routes and train services, not a map of every railway. There was no such explanation anywhere on the site before. Suggested by Gordon on RailUK Forums.
Added the point that a named train rarely has a railway to itself. The Glacier Express runs over the same metals as the local services working the Rhône and Rhine valleys all day. Mapping only the famous train leaves readers thinking it runs on a special railway unconnected to the ordinary one. Suggested by Gordon on RailUK Forums.
The about page now sends readers to a journey planner for real times and fares. It points to bahn.de, which covers most of Europe rather than only Germany, instead of merely warning that the figures here are not live. Suggested by Gordon on RailUK Forums.
The whole about page rewritten, from a line by line draft Gordon wrote out and sent unasked. He is a linguist, writer and editor by trade, and specialises in turning translated English into correct English. The opening line, the AI versus human paragraph, the sources paragraph and the errors get fixed section, which now says that any error or omission can be reported and that every major modification is dated and credited. What was a list of disclaimers reads as an invitation. The wording on that page is substantially his, not mine. Written by Gordon on RailUK Forums, one of the largest single contributions anyone has made to this site.

23 July 2026

Brenner Railway redrawn onto its real alignment: the Wipptal climb past St. Jodok, the Eisack valley through Franzensfeste, Brixen and Bozen, and the Adige gorge at Peri. 140 traced points where eight straight chords used to cut across the valleys. Found by the atlas’s own audit sweep.
Pustertalbahn traced through the Rienz and Drau valleys. Its western end now forks off the Brenner line exactly at Franzensfeste, as it does in reality. Found by the atlas’s own audit sweep.
Oslo to Göteborg no longer runs down the Bohusbanan coast, partly through open water, which Oslo trains never touch. It now follows the real route: the Østfold line via Moss, Fredrikstad, Sarpsborg and Halden, then Dalsland, Öxnered and Trollhättan. Found by the atlas’s own audit sweep.
Oslo to Stockholm had jumped from Oslo to Karlstad in one 150 km straight line. It now follows the Kongsvinger line up the Glomma valley, then the current SJ routing via Karlstad, Örebro and Västerås. Found by the atlas’s own audit sweep.
Venice Simplon Orient Express, Vy Oslo to Halden and X2000 Stockholm to Göteborg aligned onto the same corrected Brenner, Østfold and Laxå to Hallsberg geometry, so shared corridors draw as one line instead of drifting apart. Found by the atlas’s own audit sweep.
TrainRouter now opens in light mode. The original dark look is one click away on the sun and moon toggle, and your choice is remembered. Found by the atlas’s own audit sweep.

22 July 2026

Removed a duplicate Saphir Odoriko entry. The Izu luxury train had been listed twice, once as Tokyo to Izukyu Shimoda and once as Tokyo to Shimoda. Found during a multilingual Wikipedia audit of all 852 routes. Found by the atlas’s own audit sweep.

20 July 2026

GoldenPass reframed honestly as the GoldenPass Line: a two train journey, the gauge changing GPX Montreux to Interlaken Ost and then Zentralbahn’s Luzern to Interlaken Express, not one continuous service. Reported by in_luzern on RailUK Forums.
Gotthard Panorama Express now notes it runs summer only, once a day, with a seat surcharge. Reported by in_luzern on RailUK Forums.
Added the Treno Gottardo: SOB’s everyday hourly panorama train over the 1882 Gotthard mountain line, Basel and Zürich to Locarno, no surcharge. Suggested by in_luzern on RailUK Forums.
Seventy three routes added in a systematic mesh audit, the connecting intercity corridors between the famous lines: Wien to Lienz via the new Koralm tunnel, Prague to České Budějovice, München to Innsbruck, the Deutsches Eck, the EC Transalpin via Kitzbühel, Hamburg to Köln, Zürich to Genève, the Normandy and Nantes diagonals, the Basque and Cantabrian Alvias, Rome to Genoa, Bari to Matera, the Highland and Brighton main lines, Dublin to Tralee, five Japanese limited expresses, Korea’s new Donghae line and more. Suggested corridors that failed verification were left out: the Paris to Wien Nightjet was withdrawn in December 2025, the Gäubahn faces the Stuttgart 21 cut, and Turin to Aosta is closed for electrification. Found by the atlas’s own audit sweep.
Eight European regional lines added after live verification: the Aubrac from Clermont to Béziers over Garabit, the Ligne des Horlogers, the Fichtelbergbahn, the Usedomer Bäderbahn, Devon’s Tarka Line, the Hel Peninsula line, Corsica’s Bastia to Calvi and Dublin to Westport. Cercedilla to Cotos was skipped. It has been closed for works since May 2024 with no reopening date. Found by the atlas’s own audit sweep.
Highest point figures added for 19 mountain lines, from the Jungfraubahn at 3,454 m to Lima to Huancayo at 4,781 m and Pikes Peak at 4,302 m, for the new highest railways ranking. Found by the atlas’s own audit sweep.

19 July 2026

Amsterdam to Groningen now notes that the Almere to Lelystad stretch past the Oostvaardersplassen was voted the Netherlands’ most beautiful train journey by NS passengers in 2016, with 37 percent of 4,394 votes. Suggested by mat64Hidde on Somda.

18 July 2026

Helsinki to Oulu updated. The double deck IC25 at 5 h 24 is now the fastest, and the former Russia Allegro sets run alongside as Pendolino Plus. The old Sm3 Pendolino entry was stale. Reported by rf_ioliver on RailUK Forums.
Helsinki to Turku no longer claims shoreline views. The Rantarata runs mostly inland despite its name. Reported by rf_ioliver on RailUK Forums.
Helsinki to Kuopio rephrased to the Savo line through the lake district, noting the continuation to Iisalmi and Kajaani. Reported by rf_ioliver on RailUK Forums.
Jokioinen Museum Railway now credits Finnish, German, Soviet and Belgian builders, not Belgium alone. Reported by rf_ioliver on RailUK Forums.
Added the Porvoo Museum Railway: summer Dm7 trains from Helsinki to Porvoo. Suggested by rf_ioliver on RailUK Forums.
New Luxury category: 17 all inclusive hotel and cruise trains, among them the VSOE, Rovos, the Ghan and Seven Stars, moved out of Night and Scenic, so those categories now count scheduled services only. The night train figures were recomputed accordingly. Suggested by Ant1966 on RailUK Forums.
Seven routes added on reader suggestions: Geneva to Brig, TransPennine Liverpool to Newcastle, Intercités Bordeaux to Toulouse to Marseille, Intercity Milan to Ventimiglia, Frecciarossa Adriatica Milan to Bari, and the Ister and Chopin night trains. Suggested by Ant1966 on RailUK Forums.
CrossCountry Aberdeen to Plymouth redrawn on the coastal main line via Berwick between Edinburgh and Newcastle, and over the Dawlish sea wall between Exeter and Plymouth, instead of inland chords. Reported by Ant1966 on RailUK Forums.
Night Riviera and Cornish Riviera Express geometry aligned onto the one real line west of Exeter, through Dawlish, Newton Abbot, Totnes, Plymouth and Bodmin, removing phantom parallel routings. Reported by Ant1966 on RailUK Forums.
Paris to Lille TGV no longer loops via Charles de Gaulle airport. It follows the LGV Nord straight, like the Eurostars beside it. Reported by Ant1966 on RailUK Forums.
Frecciarossa Paris to Milan and TGV Paris to Montpellier now follow the LGV Sud Est alignment to Lyon instead of a straight chord. Montpellier also departs Gare de Lyon, not Gare du Nord. Reported by Ant1966 on RailUK Forums.
Italo Milan to Rome now follows the Direttissima south of Florence, matching the Frecciarossa alignment. Reported by Ant1966 on RailUK Forums.
Amsterdam to Groningen redrawn via the Hanzelijn. The map still showed the pre 2013 routing via Amersfoort. Reported by Bemined on RailUK Forums.
IC Amsterdam to Maastricht service top speed corrected from 160 to 140 km/h. Reported by Oldercouple on RailUK Forums.
Amsterdam to Groningen rolling stock updated to the mixed ICM and ICNG fleet running since June 2026. Found by the atlas’s own audit sweep.
Voralpen Express description no longer calls St. Gallen a lake city. The Romanshorn leg ended in 2013. Reported by in_luzern on RailUK Forums.

16 July 2026

UK routes now name the home nations, England, Scotland and Wales, instead of United Kingdom. Prompted by RailUK Forums feedback.

15 July 2026

Caledonian Sleeper rerouted via Edinburgh Waverley with its reversal. Distance corrected from 966 to 911 km, duration to the booked 12 h 42 min. Reported by pokemonsuper9 on RailUK Forums.
Eurostar marketing durations replaced with real scheduled times. Found by the atlas’s own audit sweep.
17 straight line distances replaced with actual route lengths. Found by the atlas’s own audit sweep.
Four withdrawn services removed. The Trans Mongolian was re scoped to the K23 and K24 trains that actually run. Found by the atlas’s own audit sweep.
US Acela average corrected to 110 km/h, behind 32 countries, not 33, on the fastest trains page. Found by the atlas’s own audit sweep.