← survey home · UK climbs · US climbs · Swiss climbs · Austrian climbs · French climbs

Updates

The survey changes as new data lands and review time allows — and its climb count goes down as well as up, because a measurement that fails scrutiny is retired, not quietly patched. What changed, newest first:

2026-09-26 — Champagne and Lorraine: 109 climbs from the north-east

Nine départements found roads on the IGN LiDAR HD ground model in one evening on eight machines: Ardennes, Aube, Marne, Haute-Marne, Meurthe-et-Moselle, Meuse, Moselle, Pas-de-Calais and Somme, from 8,643 candidate roads. 109 pages clear a ranking door, including two named finishes from the cols wave — Col de la Chapelotte and the D910. Aisne, Nord and Oise wait for IGN to publish their ground model; Pas-de-Calais's coverage is a shade under our floor and is run around rather than parked.

2026-09-26 — The Loire valley and western Paris: 23 climbs from the centre

Eleven départements south and west of Paris found roads on the IGN LiDAR HD ground model in one evening on four machines: Cher, Indre, Indre-et-Loire, Loir-et-Cher, Loiret, Paris, Yvelines, Hauts-de-Seine, Seine-Saint-Denis, Val-de-Marne and Val-d'Oise, from 3,720 candidate roads. The flattest ground surveyed yet: 23 pages clear a ranking door and no named pass exists to give one a finish. Seine-et-Marne, Eure-et-Loir and Essonne wait for IGN to publish their ground model.

2026-09-26 — Normandy and the Pays de la Loire: 94 climbs from the north-west

The first lowland region of France: 10,673 candidate roads found across Calvados, Orne, Eure, Seine-Maritime, Loire-Atlantique, Maine-et-Loire, Mayenne, Sarthe and Vendée, measured on IGN's LiDAR HD ground model on eight machines in one evening, with coverage above 99.5 % in every département. Flat country keeps its climbs short: of more than ten thousand roads measured, 94 clear a ranking door — the Suisse normande, the hills of the Perche and the Alpes mancelles, the Signal de Charlemagne and the cols of the Orne. Brittany and the Manche wait for IGN to publish their ground model. Also on this update: the France page now says which surveys are live rather than which were live a week ago.

2026-09-25 — Provence: 636 climbs from Bouches-du-Rhône, Var and Vaucluse

The rest of France begins with Provence: 8,331 candidate roads found across the three départements, measured on the IGN LiDAR HD ground model on eight machines in one morning, with coverage above 99.8 % in each. 636 pages clear a ranking door across eight blocks and a cols wave — the Dentelles, the Luberon, the Sainte-Baume, the Maures and the Estérel, and the roads up Mont Ventoux. Twenty-six pages read 30 % or steeper over 25 m: five are published as measured after every instrument agreed, four say "pending human review" until a point-cloud cross-section settles a window the profile alone cannot, and seventeen stay unpublished under a rule of their own. Brittany and the Manche wait for IGN to publish their ground model.

2026-09-25 — One road, one page: the better of two measurements now carries the climb

A large survey is measured in blocks, and where two blocks reach across the same road the same climb could be published twice — once with the road's name and once as an unnamed lane, or once in full and once as a shorter piece. The better page now carries the climb: the one that knows the road's name, or, where both do or neither does, the longer one. 1 weaker pages have been withdrawn; each of their addresses now leads to the page that kept the climb, so no link is lost, and every withdrawn measurement stays in our records.

2026-09-25 — Corsica: 471 climbs from the island's LiDAR HD survey

The sixth French massif, and the first Mediterranean island: 4,372 candidate roads found across Corse-du-Sud and Haute-Corse, measured on the IGN 1 m LiDAR on four machines, with coverage complete in both départements (100.00% and 99.96%). 471 pages clear a ranking door across four blocks and a cols wave, including the seam cols and Corsica's named summits. Fourteen pages reading above 30% over 25 m were checked against every instrument and published; three more read that steep but could not be settled from the profile alone — the shape of a driveway apron, a stepped path, or an unstable window under canopy — and now say "pending human review" until a point-cloud cross-section answers them; one page above the Baldwin Street mark, steeper than any road we publish by measurement, waits on a single human eye.

2026-09-25 — Named passes take back the roads named after them

A mountain pass is one climb from the valley floor to the summit, and that is how a col is published here. Where a stretch of the same road had already been published on its own — a piece of the pass road measured before the pass itself was built — the pass now carries the climb and the fragment has been withdrawn. 3 such pages have been folded into the passes that contain them; each of their addresses now leads to the pass, so no link is lost, and every withdrawn measurement stays in our records.

2026-09-24 — One Jura page withdrawn: a lane that turned out to be a grassy bank

Street View showed the mapped lane beside the D436 at Villard-Saint-Sauveur runs on the bank above the road, not the road itself. The survey's road check had matched a gravel track the map carries in the same place. The page is withdrawn and a new rule holds this shape before it is published.

2026-09-24 — The Jura: 140 climbs across Ain, Doubs and Jura, surveyed from IGN LiDAR HD

The fifth French massif: 7,216 candidate roads found from the Jura's three départements, measured on the IGN 1 m LiDAR on four machines. 140 pages clear a ranking door, including the four named cols at the Ain/Jura seam. Two pockets of thin LiDAR coverage — part of the Jura département and the low country of Ain — are shaded as not yet surveyed rather than measured from anything coarser, and are picked up again once IGN flies them. The neighbouring Morvan, in Burgundy, waits on IGN publishing that flight at all; nothing there is measured yet.

2026-09-24 — Named passes take back the roads named after them

A mountain pass is one climb from the valley floor to the summit, and that is how a col is published here. Where a stretch of the same road had already been published on its own — a piece of the pass road measured before the pass itself was built — the pass now carries the climb and the fragment has been withdrawn. 3 such pages have been folded into the passes that contain them; each of their addresses now leads to the pass, so no link is lost, and every withdrawn measurement stays in our records.

2026-09-24 — Five Massif Central pages re-measured from the valley floor to the summit

A valley crossing was found in the published extent of five Cévennes-area pages found by the Massif Central survey: Chemin de Cloux, Chemin de la Ferme, the Col des Mourèzes from Mandagout, and two discovered lanes. Each was re-measured from its low point to the summit above it, the same fix already applied to 183 pages on 2026-09-22. The steepest 25 m of every page stands unmoved; only the length and overall gradient change.

2026-09-22 — The Massif Central: 1,846 climbs across sixteen départements, from the Limousin to the Cévennes

The fourth French massif, measured in a day on eight machines: 65,512 candidate roads found, 44,918 measured on the IGN 1 m LiDAR, 1,797 pages that clear a ranking door — the Sancy, the Cantal, the Cézallier, the Forez, the Pilat, the Aubrac, the Lozère, the Cévennes and the Montagne Noire, with Tarn, Hérault and Aveyron measured this time. The east yields five to six climbs in a hundred roads and the gentle west one or two. IGN has not yet flown the north-west (most of the Creuse, the north of the Haute-Vienne, the Montluçon and Combrailles corners), so the map shades that block as not yet surveyed and no road in it is measured from anything coarser; it is added when IGN flies it. Seven lanes reading above the Baldwin Street mark are held for a person to look at, never ranked by machine. The named cols of the massif followed the same evening: 49 more, the Col de Peyra Taillade and the other Tour de France climbs among them, each built whole from its foot to the col even where a measuring boundary had cut it short, and 17 partial pages absorbed into the col they belonged to (their old addresses redirect).

2026-09-22 — Valley crossings re-measured as the climb out of the valley: 183 pages, steepest section unchanged

A hand check of Wagg Lane in Cornwall found a page that ran from one hilltop to the next across a valley and called itself 826 m at 0.6 %, when the climb is the 400 m out of the stream at 7 %. Two hundred and thirteen live pages had that shape. Each was re-measured from its low point to its summit: on 183 the steepest 25 m came back exactly as published and only the length and overall gradient changed (overall median 2.8 % to 8.0 %); 20 whose steepest figure moved are held as they were, and 2 protected pages wait for a look. New pages with this shape are now refused before publication.

2026-09-21 — A possible bridge or hollow at the steepest section is now said on the page, not corrected

On 235 pages the laser ground model dips and climbs back at or close to the steepest 25 m. That is the shape a bridge leaves when the map does not show one, and it is also the shape of a real hollow that the road goes down into and climbs out of. We re-measured every one of them both ways and the evidence says most are real hollows, so no figure has been changed. Instead each of those pages now says where the dip is and that it is pending human review, the ranking tables carry the same words in their notes, and the page's data file records it. A flag comes off when a person has looked.

2026-09-21 — Every older page re-measured with today's rules: the site reproduces, and 169 figures come down

Overnight we re-measured 17,379 of the site's 17,556 pages on their own published lines with the current code. Ninety-five in a hundred came back within half a point of what they publish, and ninety-seven and a half once a published artefact verdict is allowed to stand where today's check still half-finds it. On 169 pages a mask that exists today reaches a page published before the mask did, and the figure comes down, by about two points at the median; sixty of those no longer clear a ranking door and are listed rather than ranked. Where a re-measure would have RAISED a figure we looked at twelve by eye, rejected eleven, and made that the rule: a rise is never published by machine. One went the other way: Spillman's Pitch in Stroud, where the step the rule had masked is the road itself, now shows its measured 25.7 %. Every re-measured page now records exactly which version of the code produced it.

2026-09-20 — First corrections to the Pyrenees, and three Alpine climbs renamed for the col at their top

A day of eyes on the new French pages. The Route du Bénal in the Ariège, sixth in France this morning, is listed rather than ranked: the mapped line turns straight up a bank where the real road bends gradually, and it waits for a hand-drawn line. A gated field track at Oloron-Sainte-Marie is withdrawn; it is not a road. Three Alpine pages were named after a col their road only passes: they are now the Cime de la Bonette from Jausiers, the Col de Porte from Sarcenas and the Col de la Bonette from Saint-Étienne-de-Tinée, and their old addresses still work. New pages are now held before publication when the steepest stretch sits beside a gap in the official road record, the shape that caught the Bénal out.

2026-09-20 — Some steepest-25 m figures replaced by what our current rules make of the same ground

A check today replayed our current window rules over every published profile and found 106 pages whose headline figure the rules no longer produce — pages published before the September window-texture rule, or measured while that rule could flip on a sub-metre shift of the samples. Where the steepest window turns out to sit on a step, a kerb, a culvert or the edge of a masked artefact rather than on the road, the page now reports the steepest clean 25 m instead, and still shows the raw reading it displaced; 74 of them leave the 20 % rankings. The rule itself now fires only when it fires under three small shifts of the sampling, and a figure it cannot settle keeps its value with a caveat. Four well-known pages, Ffordd Pen Llech among them, keep their published figure until a person has looked. Every deploy now checks that today's code reproduces the figures it publishes.

2026-09-20 — The Pyrenees join France: 954 climbs and 60 cols

France's third massif is measured, from the Atlantic at Hendaye to the Mediterranean at Banyuls. 22,605 roads were measured on IGN's half-metre laser ground model and 954 of them became pages — 554 that rank and 400 that are listed but not ranked. Sixty cols have a page, nearly all of them from both sides and several from three or four: the Tourmalet, the Aubisque, the Aspin, the Peyresourde, the Pailhères, the Portet d'Aspet, the Marie-Blanque, the Soulor, the Balès, the Port de Lers and their neighbours, each measured from the valley floor to the pass rather than in pieces. The Basque cols are here under their Basque names where the road carries one — Col Bagargiak, Col d'Aphanitze, Larraineko mendatea, Col de Burdincurutcheta. The Col du Pourtalet from Laruns, 28 km and 1,487 m of climbing, is the longest climb on the site outside the Alps. Eighty-six roads are missing on purpose: IGN's own model has no data under them, mostly where its survey stops at the Spanish border, and we would rather publish nothing than guess. Five cols on the list have no page — two of them, the Port de Court and the Col du Dragon, because the road carrying the approach's name stops a few metres outside the radius we search at the pass. France now holds 2,954 climbs.

2026-09-20 — The Vosges join France: 160 climbs and twelve cols

France's second massif is measured. 160 climbs from the Vosges — the Ballon d'Alsace, the Grand Ballon, the Schlucht, the Bonhomme, the Calvaire, the Platzerwasel and seven more cols, each measured from the valley floor to the pass rather than in pieces, most of them from both sides. The Grand Ballon from Uffholtz is 21 km and a thousand metres of climbing, the largest French climb outside the Alps. Measured on IGN's half-metre laser ground model against the BD TOPO road record, the same sources as the Alps. Four lanes are missing on purpose: IGN's own model has no data under them, and we would rather publish nothing than guess. The Col du Petit Ballon is not here yet: the roads at the pass carry no name or number in OpenStreetMap, so no route could be built for it. One Vosges page, the Col de Bussang from Bussang, was held back for a day: the laser had read a ravine under an untagged bridge as a 24.6 % ramp on a 4 % road; re-measured with the bridge span masked it reads 5.0 %, and it is now published at that figure. France now holds 2,000 climbs.

2026-09-19 — That letter is called "evidence quality" now

The label we put on the A/B/C/D/U letter this afternoon — "survey grade" — lasted an afternoon. In cycling a bare "grade" means the gradient itself, so a column headed "Grade" next to a gradient column reads as a second steepness number. The letter is now called evidence quality, and the column is headed "Evidence": it says what the figure rests on — how fine the laser data under the road was, and what supporting evidence and checks were available to back the result up. It is still not a measure of accuracy; for that, each page points at how exact a 25 m figure is. Nothing in the data changed.

2026-09-19 — The letter on every page is a survey grade, not a confidence score

Every climb carries a letter — A, B, C, D or U. It used to be labelled "measurement confidence", which read as "this number is accurate". It never meant that. The letter says how the page was processed: whether the ground model was a metre or finer, whether anything had to be masked or moved on the way through, and whether a query is open. An A means nothing needed fixing, which on a page where a check left no record can equally mean nothing was looked at — that is true of well over half the pages carrying an A. So the label is now "survey grade", the column is headed "Grade", and the methodology says plainly what the letter does and does not tell you. For how accurate the numbers are, the page points at the new section on that: a steepest-25 m figure is good to about a percentage point either way. Nothing in the data changed — the downloadable CSV column and the machine-readable documents still use the name "confidence".

2026-09-19 — How exact a 25 m figure is, stated plainly

Every climb page now says that its steepest-25 m figure is good to about one percentage point either way, with a link to a new methodology section that gives the 95 % range from a simulation of the whole measurement across 430 pages in the five countries, and explains why the figure leans slightly high. Two lanes at Pont-ar-gothi confirmed as real, steep roads by eye; one of them returns to the rankings.

2026-09-19 — Faster pages, real region pages, and pictures from a new home

Country pages no longer ship every climb in a hidden table: the Austrian page is now 153 KB instead of 4.7 MB. Every country has real pages for each region and a paged A-to-Z and steepest-first catalogue, so any climb is a link away without a script. Tables scroll on a phone instead of stretching the page. The interactive map loads only when you reach it. Charts and maps are served from assets.howsteep.org, which lets the site grow past the limits of its old home. One Ceredigion lane was withdrawn after a look at the aerial photograph showed a field-edge path, not a road.

2026-09-19 — Site-wide: every page now names all five countries, secure connections, and honest labels

Housekeeping after an outside review of the site. Every place that lists the countries surveyed is now generated from the same list the site is built from, so France appears everywhere the others do. Plain http addresses redirect to https. The machine-readable data on each page states the licences exactly as the licensing page does, describes each source at its true resolution (half a metre in France and Switzerland, a metre elsewhere), and carries a breadcrumb trail. Total-ascent figures are labelled as the rough sums they are. Swiss, Austrian and French charts and data are now cached properly, which makes those pages faster to return to.

2026-09-19 — Wales: 35 held climbs settled with the Welsh Government laser point cloud

Since August, 35 Welsh climbs had been held out of the rankings because a one-metre terrain model cannot tell whether the measured line sits on the road or on the bank beside it. The Welsh Government has now supplied the classified laser point cloud for those road corridors under the Open Government Licence, and a cross-section every 25 metres settled each one. Eighteen climbs keep their measured figure and are released; twelve have a corrected figure after the line was moved onto the road bed the laser found or a driveway was cut off the end; five of those fall below the 20 % ranking door and are listed rather than ranked; two pages come down (a grass track and a lane whose steep part was a driveway); two wait for a line drawn by hand. Every page touched says so and credits the Welsh Government point cloud.

2026-09-18 — A laser answer for 35 Welsh climbs

Thirty-five Welsh pages carried a note saying we could not be sure our measurement line was on the road. The Welsh Government sent us the laser point cloud behind their terrain model, and it can tell a made road bed from the hillside beside it. Eleven pages had a road bed directly under the line and are now back in the rankings at the numbers they already showed. Seven more were passed by eye on the aerial photography. One page is gone: the steep part was a grass track, not a road, so it has been withdrawn. One is now marked as an unpaved track, kept for the record but out of the rankings. Twelve more have been re-lined onto the bed the laser found, or cut where the road turns into a driveway, and re-measured: every one of those twelve figures has changed, most of them downwards, and the reason is on each page. Two of them were steep enough to rank this morning and are not any more. One page measured 29 % on what turned out to be somebody's drive; with the drive taken off it is not a climb at all, and it has gone.

2026-09-18 — Aerial photos on French pages, and a word on what France covers so far

Every French climb page now carries the Aerial button the Swiss, Austrian and US pages have, showing IGN's BD ORTHO national aerial photography behind the measured line. The French section also says plainly what it covers: this first release is the Alps, from Lake Geneva to the Mediterranean; the Vosges, the Pyrenees, the Massif Central and the Jura follow massif by massif, and then the rest of the country as IGN's laser survey completes.

2026-09-18 — France arrives: the Alps, 1,840 climbs

The first French section is live, covering the Alps from the Rhône to the Italian border: 1,840 climbs measured on IGN's LiDAR HD half-metre laser terrain model and checked against the BD TOPO road network, including 127 mountain passes from the valley floor to the col — the Galibier, the Iseran, the Bonette, the Loze, the Cayolle and their neighbours. Every page carries its gradient ladder with the window length of every figure, a surface and access chip, and the same open licences as the rest of the site. The Vosges and the other massifs follow.

2026-09-18 — Named passes take back the roads named after them

A mountain pass is one climb from the valley floor to the summit, and that is how a col is published here. Where a stretch of the same road had already been published on its own — a piece of the pass road measured before the pass itself was built — the pass now carries the climb and the fragment has been withdrawn. 7 such pages have been folded into the passes that contain them; each of their addresses now leads to the pass, so no link is lost, and every withdrawn measurement stays in our records.

2026-09-18 — One road, one page: the better of two measurements now carries the climb

A large survey is measured in blocks, and where two blocks reach across the same road the same climb could be published twice — once with the road's name and once as an unnamed lane, or once in full and once as a shorter piece. The better page now carries the climb: the one that knows the road's name, or, where both do or neither does, the longer one. One weaker page has been withdrawn; its address now leads to the page that kept the climb, so no link is lost, and every withdrawn measurement stays in our records.

2026-09-16 — Where you stand on Ffordd Pen Llech

Every gradient here is measured along the road as the map draws it, and on the Harlech hairpin that line runs on the inside of the bend — the same side the 2019 Guinness survey measured, and the steeper side of a road. Down the middle of the carriageway the same laser model reads about 28 % over 25 metres instead of the 32.6 % the page shows, which is the direction of the correction that moved the record back to Baldwin Street in 2020. The page now says where its line runs and gives both figures, because both surveys were right about the line each of them measured.

2026-09-15 — The rest of the drives at the top

762 more pages ended in a short bend-off: the road turns hard in its last few dozen metres and climbs what is really a private drive. Every one of them has had the drive cut off the end of the route and been re-measured on the road that is left. On most of the pages the steepest 25 metres barely moves, because the drive was not the steepest part of the climb; on seventy of them it falls, and on a few it falls a long way — Grabenkoschakweg at Oberhaag in Styria from 28.6 % to 16.2 %, Chemin de Macheret above Troistorrents in Valais from 35.0 % to 26.2 %, Sonnenbergstrasse at Unteriberg in Schwyz from 24.2 % to 18.1 %. A handful read slightly steeper instead, because re-measuring also redraws the line along the middle of the road. Three pages now meet none of the thresholds for the rankings and are listed rather than ranked. Sixteen more, where the road left after the cut would be under the 200 metres a page needs, have been withdrawn.

2026-09-15 — Roads too short for a page, withdrawn

A road needs 200 metres of climbing to earn a page here — that has been the rule since the survey began, and it is the length below which a gradient says more about a driveway ramp or a kerb than about a hill. 533 pages had been published under it anyway, most of them short unnamed lanes: Switzerland 232, the United States 162, Austria 98 and Britain 41. They have been withdrawn, and their addresses no longer lead anywhere. Three roads under the minimum stay, because a person decided they should: two were looked at and judged by hand, and one — Michaelgate in Lincoln, 185 metres of cobbles — is in a published book of climbs. Every withdrawn page keeps its measurement in our records, and any of them can come back if a book lists it or a person rules it belongs.

2026-09-15 — Lines that cut the bend, redrawn

A second pass over the same problem, closer in. 62 more pages were held because their steepest 25 metres sat three to five metres off the road as the road authority maps it — near enough that the line was plainly on the road, but cutting the inside of a bend just where the hill is steepest, which reads steeper than the road really is. 59 of them could be redrawn along the mapped road — OS Open Roads in Britain, the swissTLM3D network in Switzerland — and were re-measured from scratch on the new line. 41 came back on the road and the same length, and their new numbers are published: 21 clear a ranking door and return to the rankings, and 20 turn out to be ordinary once the line follows the bend and stay listed without a ranking. The rest changed the length of the climb rather than its line, or are still off the road, and are left exactly as they were.

2026-09-15 — Lines that had wandered off the road, redrawn

A page's numbers are only as good as the line they were measured along. 275 pages were held out of the rankings this month because their steepest stretch sat several metres off the road as the road authority maps it — the sign of a measurement line that had slipped onto a bank, a field or a verge, where the ground is steeper than the road. 147 of them could be redrawn along the mapped road — OS Open Roads in Britain, the swissTLM3D network in Switzerland, the state road inventories in the United States — and were re-measured from scratch on the new line. 116 pages now sit within five metres of the mapped road and keep their length, and their new numbers are published: 47 of them clear a ranking door and return to the rankings, while 69 turn out to be ordinary roads once the line is on the road — Elua Street on Oahu falls from 29.2 % to 8.5 % over its steepest 25 m, Wyman Hill Road in Maine from 33.1 % to 20.5 % — and stay listed without a ranking. The remaining 118 pages are a different problem: the route follows the mapped road along almost its whole length and leaves it only where the hill is steepest, so there is nothing to redraw. Those stay listed with their profiles shown as measured while we decide what to do with them.

2026-09-15 — Austria's named passes, measured end to end

Until now the big Tyrolean passes existed here only as fragments — a kilometre or two of road a machine had picked out of a much longer climb. Nine are now measured the way they are ridden, from the foot of the pass to the top: Kühtai from Ötz and from the Sellraintal, the Hahntennjoch from Imst and from the Lechtal, the Piller Höhe from Fließ and from Wenns, the Timmelsjoch Hochalpenstraße above Hochgurgl, and the Silvretta Hochalpenstraße from Galtür and from Partenen. Kühtai from Ötz runs 17.0 km and climbs 1,181 m; the Hahntennjoch from Imst 13.9 km and 1,079 m. Toll roads are ranked here as the public roads they are, with a line on the page saying who charges and whether bicycles pay. Three short fragment pages have been folded into the full climbs they belong to and now redirect there. Eight of the nine rank; the Silvretta's Partenen ramp is published but not ranked, shut by a landslide since 2024. A defect in one block of the national elevation survey has since been fixed, so two more toll glacier roads join them — the Kaunertaler Gletscherstraße, 25.7 km and 1,440 m of climbing from Feichten to the Weißseeferner, and the Ötztaler Gletscherstraße above Sölden, 12.5 km at 11.0 % — and the Timmelsjoch is now measured from Zwieselstein, where the climb really begins, 18.9 km instead of 11.1 km.

2026-09-15 — Eight hairpins, looked at by eye

Eight Austrian pages were held out of the rankings in September because their steepest stretch sat on the inside of a hairpin rather than on the road, and an automatic redraw onto the official road network had not convincingly fixed them. All eight were compared against aerial photography by hand. Five redraws were no better than the line already published and were dropped — those pages keep their published profile and stay listed. One, a lane near Zwillingwand in Salzburg, is much better: its route now follows the mapped road to within half a metre at the steepest point, and the steepest 25 m reads 17.2 % rather than 19.9 %; it returns to the rankings. On one page neither line is on the road where the hill is steepest, so it waits for a route drawn by hand. And one route, in Lower Austria, turned out to follow no visible road at all and has been withdrawn.

2026-09-15 — Measured between house walls

Some village lanes run between house walls a metre or two from the kerb. The laser terrain model behind these pages is bare earth: the buildings are taken out and the ground under them is filled in, and where the gap between two filled-in footprints is narrow that fill can inherit the height of the walls — so the steepest stretch reads the houses, not the road. 3 pages whose steepest 25 m sat in such a gap AND whose profile spiked in a way no road does have been re-measured with the built-up stretch masked. Via Péz Fess in Graubünden falls from 24.1 % to 9.4 % over 25 m and a lane above Orsières from 22.5 % to 10.5 %; both now sit below every ranking door and stay listed. Pearl Street in Tarentum, Pennsylvania keeps its place at 20.3 % (it was published at 23.5 %), its steepest metres now read further up the hill. A further 21 pages run between buildings but climb like roads, and keep their published numbers with a note saying so — among them Chemin du Reposoir at Le Châble, the lane that prompted the check, which reads 20.5 % over 25 m on a 19.1 % hundred metres with house walls 1.9 m apart.

2026-09-15 — The drive at the top is not the climb

Eighteen pages ended in a short bend-off — the road turns sharply in the last few dozen metres and climbs what is in fact a private drive — and on every one of them the steepest 25 m was that drive rather than the road. Seventeen have had the drive cut off the end of the route and been re-measured on what is left; the eighteenth, whose remaining road would be under 200 m, is left alone for now. Every steepest-25 m reading falls: Veia Muntogna da Schons from 23.5 % to 12.1 %, Durchachalmweg from 24.3 % to 16.8 %, Federal Alley in Pennsylvania from 22.6 % to 14.3 %, Ober-Aesch-Weg from 20.2 % to 16.0 %. Four of the seventeen drop below every ranking door and are now listed rather than ranked; the rest were already listed, or still clear a door on the shorter route.

2026-09-15 — Too short to keep a page

24 pages whose climb, once cut back to its crest, was shorter than the 200 m minimum a road needs to have a page here have been withdrawn — 4 British, 13 American, 5 Swiss and 2 Austrian. Each ran a long way past the top of its hill, and the climb itself, from the bottom to the point where the road levels, was under 200 m: a steep pitch rather than a climb. Their addresses no longer resolve.

2026-09-15 — Trimmed to the crest

167 pages whose route ran on along the flat past the top of the climb have been cut back to the crest and re-measured, so the length and the overall gradient describe the climb rather than the climb plus a level run-out. Four pages whose steepest metres were the end of a bridge deck — where the bare-earth laser model reads the ground beneath the abutment, not the road — have been re-measured with the deck masked and now read the road: Accrington Bypass falls from 29.4 % to 4.9 % over 25 m, and the other three likewise drop far below every ranking door and stay listed. Nine pages held since 11 September because a step or bank sat in their steepest window have been re-measured with that texture masked; most barely moved, but a County Down lane falls from 28.9 % to 18.6 % and L Pond Road in Maine from 20.6 % to 14.8 %. Two checks that only some countries had before now cover all four: 101 British and 58 American pages whose steepest 25 m sits well off the mapped road where the coarse terrain model also disagrees with the laser are held for a re-line, and the driveway and power-line checks have run over the whole United States and Britain (10 more US pages held for a driveway at the steepest point, one British page for a power line). Nine trimmed British pages whose re-measured maximum fell below the ranking door are listed rather than ranked.

2026-09-14 — Fifteen hairpins, redrawn along the road

Fifteen pages whose measured line cut across the inside of a hairpin — reading the bank between a switchback's arms instead of the road — have been redrawn along the official road centreline (GIP in Austria, swissTLM3D in Switzerland, OS Open Roads in Wales, the state inventory in Pennsylvania) and re-measured. Most steepest-25 m readings fell: Llanellen to Coed y Prior from 32.25 % to 29.90 %, Route des Larrets from 19.38 % to 16.27 %, Vinaders from 21.30 % to 18.90 %; a few rose slightly, Egg-Gülkestraße from 23.40 % to 24.23 %. Eleven return to the rankings on the new line, four now sit below every ranking door and stay listed. Twelve further hairpin pages could not be re-lined automatically and stay held until a person looks at them.

2026-09-14 — The UK catches up with the instruments

The road-authority check that every US, Swiss and Austrian page has had since it was built — does the steepest 25 m lie on a mapped road? — ran over the UK for the first time, against OS Open Roads. 29 pages whose steepest metres lie off any mapped road and 35 whose steepest window sits at a driveway or track mouth leave the rankings until they are re-lined and re-measured; their profiles stay published as measured. Two more pages (Accrington Bypass, Glenwood Avenue) take their maximum from the end of a bridge deck, where the laser record reads the ground beneath the abutment, and are held with the Pennsylvania road that showed the pattern. The front page's four survey cards now share one layout.

2026-09-13 — Porlock Hill, by name

The A39 from Porlock to Pittcombe Head had been published under an automatic label since August. It is Porlock Hill, the road the sign calls 1 in 4, and now carries that name and its 100 Climbs listing. The village start, where the laser record reads the house walls rather than the road, is still trimmed and waits for a re-measure with a building mask.

2026-09-13 — One front page, four lean country pages

The explanation of the survey lived twice, on the front page and on the UK page, because the UK page was once the whole site. It now lives once, on the front page: how to read a gradient ladder, which hills get a page, how claims are tested. Each country page opens with one sentence saying what it is, carries machine-readable dataset metadata for search engines and language models, and then only what is distinctive to that survey: its elevation model, its road authority, its counts, coverage and sources. Backlinks on every page now list every section.

2026-09-13 — The map line now wears the gradient colours

The route on every climb page map is coloured by the same 25 m gradient bands as the profile and gradient charts, with a key under the map: grey under 10 %, blue to 20 %, orange to 25 %, red above. The colours stay on through the topo, vector and aerial views.

2026-09-13 — Aerial imagery on the map, and why UK pages have none

Swiss, Austrian and US climb pages gain an Aerial button on the map: swisstopo, basemap.at and USGS orthophotos, open national imagery served live with a credit. UK pages say plainly that Britain has no open or government aerial photography at road scale, and the methodology page makes the polite case for changing that, with the LiDAR release of 2015 as the precedent.

2026-09-13 — Austrian restricted roads without a surface record step out of the annex

401 pages on GIP-restricted roads had no surface record under their steepest window. Aerials show most are forestry and alm tracks with a short steep ramp, mountain-bike ground rather than road. They are now listed with their measurements shown, out of the annex and the rankings, until someone confirms a road. Restricted roads with a recorded gravel or paved surface stay in the annex.

2026-09-13 — Eleven record-class Austrian lanes, listed not ranked

Eleven wave-4 routes measure 35 to 44 % over 25 m, at or above the Baldwin Street mark. Results in that class are never ranked by machine: the road authority has no surface record under any of their steepest windows, and the 100 m figures read more like tracks than roads. They are published as listed pages with their measurements shown and a note saying why, until a person confirms the road.

2026-09-13 — Switzerland, one page per road

The same geometry check across the Swiss survey found 704 coincident pairs: the extent pass of 11 September had re-published trimmed routes as new pages beside the originals. 436 duplicates are retired with redirects to the page that stands; where the surviving page had been held and the newer one ranked, the newer measurement now sits under the surviving address. Two new checks ran over every Swiss page: a track or driveway meeting the steepest window (13 pages held) and power lines crossing it (none suspicious).

2026-09-13 — Austria, one page per road

A geometry check across the Austrian survey found 190 pairs of pages whose routes coincide: the same road had been adopted twice, once from the terrain search and once from the road inventory. 125 duplicate pages are retired and redirect to the page that stands (the earlier one, or the longer one where the shorter was a slice with the same steepest 25 m). Six pages the inventory named only by road number carry the Discovered-lane label.

2026-09-13 — Austria's gentler band — 1,775 more climbs

The Austrian passes so far seeded only roads the coarse elevation model read at 15 % or steeper. A fourth wave seeded the 12–15 % band: 9,939 candidates measured on the 1 m BEV ground model, 1,032 pages adopted as measured and a further 743 after their routes were trimmed to the inventoried public run or kept where the road carries on climbing. Austria now lists 4,703 climbs. Two new checks ran over every Austrian page: a driveway or track meeting the steepest window (32 pages held for a look) and overhead power lines crossing it (none suspicious). Six pages the inventory names only by road number wait for a name.

2026-09-12 — Western Tyrol joins the Austrian survey

The Ötztal, Pitztal, Kaunertal, Stubai, Sellrain and Außerfern valleys were missing from the first two Austrian passes — a measuring shard had died quietly and nothing checked the map against the road network. A third wave measured 1,169 routes there and along the Czech border: 544 pages join (Kaunertaler Gletscherstraße, Kühtaistraße, Hahntennjochstraße and Sellrainstraße among them), Austria now has 2,928. Forestry and permit roads whose steepest stretch lies on a restricted segment now rank in the annex with the restriction stated, the same rule Switzerland uses. A coverage check now compares every region cell against its neighbours so a gap like this fails loudly.

2026-09-12 — Humps, steps and culverts masked across every country

The texture rule now also masks a hump: a steepest 25 m that gains more than its surroundings, stops for ten metres and then climbs on — the profile a culvert or a junction apron leaves in a centreline. Every page held yesterday, and every page whose window showed that shape, has been re-measured with the steps masked and re-sited to the sustained gradient: about half return to the rankings at their true value and the rest are listed below the door. Recognised climbs affected read their sustained gradient now (Bwlch Penbarras, Monsal Head, Whinlatter Pass, Newgale Hill among them).

2026-09-11 — One map, and a texture check inside every steepest window

Seven Austrian pages found by eye shared one fault: the steepest 25 m sat on a terrace staircase, a culvert, a field bank or the flank of a span the mask had already caught. The batch now judges the texture inside every reported window and re-sites it like a structure; 655 live pages whose window fails are out of the rankings until re-measured (UK 327, US 125, Switzerland 93, Austria 110) — their profiles stay visible as measured. Austrian surfaces: a page is called paved only when the road register records a surface under at least half the route, so 608 Austrian rows now read "unverified". Austria's extent wave adds its continuations: 2,369 Austrian pages. And one map: every climb of every country on one layer at /map, framed per country by a button.

2026-09-11 — Austria's second wave, and lighter maps

The steep public roads the GIP register knows but the coarse pass missed: 2,404 more Austrian routes surveyed and 589 pages join (437 ranked, 152 listed), plus ten forest, farm and ski-area service tracks at record grades kept as listed pages with their caveat — very steep and unpaved is not a road for ordinary traffic. Austrian pages now number 1,573. Every locator map on the site is now a JPEG instead of a PNG, a quarter of the weight with no visible change.

2026-09-11 — Austria joins the survey

A fourth country, measured on the BEV 1 m terrain model read block by block from the federal server rather than downloaded: the steepest coarse band of the Austrian network, 3,556 routes surveyed and 974 pages published — 534 ranked and 440 listed where the GIP road register or the map left a doubt. Road class, access and, where the register records it, surface come from GIP.at. The famous passes mostly wait for the next band; the Kitzbüheler Horn, the Zillertaler Höhenstraße, the Turracher Straße and the Sölkpass are in this one. Every route considered is in the decisions dataset.

2026-09-11 — Swiss continuations kept, and a spike detector

Where a Swiss climb runs on past the end of its named road onto a lane the road model carries as public, and the continuation still climbs, the measured extent now stands: 451 held pages join as measured, 269 more were re-measured on their named run and 220 of those join too (Swiss pages 4,515). A new check catches a steepest window that is a short spike with a dip just before it and nothing sustained around it — a culvert or bank the line crosses rather than the road: 14 Swiss, 6 UK and 15 US pages are listed until re-sited, and one Emmental lane whose steepest metres sit off the road at a stream crossing.

2026-09-11 — Switzerland, second pass

The gentler coarse bands of the Swiss network (up to 15 % on the coarse surface) measured on swissALTI3D: 19,203 routes surveyed, 1,407 join the Swiss pages — 797 ranked and 610 published but held from the rankings where the road model or the map left a doubt (an unnamed service way, a gate without an access tag). 731 more wait on a re-measure over the run their named road covers, and 233 where the fine and coarse terrain models disagree by 15 points or more stay held. Swiss pages now number 3,844.

2026-09-11 — Waipiʻo Valley Road, listed

The steepest road in Hawaiʻi by reputation joins the US pages on the recognition door: 1,210 m at 21.9 % overall, 28.9 % over 25 m from the 2017 island LiDAR, listed rather than ranked because the county inventory covers only three quarters of the line and the road has been closed to non-resident vehicles since 2022. Published figures for it range from 25 % to 45 %; no 25 m window on the surveyed line reaches 30 %.

2026-09-11 — Switzerland joins the survey

A third country, and the first measured on a 0.5 m national terrain model: 2,437 Swiss climbs from swisstopo's swissALTI3D, with road class, surface and access from the swissTLM3D road model — 1,488 ranked, 949 published but held from the rankings where the map left a doubt (an unnamed farm lane, a gate without an access tag). This first run worked the steepest coarse band, with 345 pages re-measured on the run their named road covers; the long alpine passes mostly wait for the next band. A road under a general traffic restriction (motor traffic limited, cycling normally allowed) ranks with the restriction stated. Every route considered, published or not, is in the decisions dataset.

2026-09-10 — The decisions dataset

Every road the survey has ever considered, and why it stands where it does: 8,996 pages ever published (ranked, held, withdrawn or merged), 139,524 routes measured but never given a page, 16,633 proposed but never measured — each with one reason code from a fixed registry, the rule it applies and the numbers behind it. Tables and per-state geometry at /data/decisions/, the full GeoPackage on the public repository's releases page. ODbL, like the rest of the data.

2026-09-09 — Driveways at the steepest window, and a code-named woods road

A new check asks the map whether a driveway, service way or track meets a route inside its steepest 25 m while that window sits off the state inventory line — the maximum may belong to the driveway. Thirteen US pages are listed for a person's ruling rather than ranked, and AM00 in Maine (a woods road the inventory names only by a code) is listed pending a public-road ruling. Five more unverified pages carry an estimated surface. No page changed its measurements.

2026-09-09 — 46 held candidates join after re-lining to the road inventory

Candidates from the Hawaii, New York and Pennsylvania surveys that had been held because their measured line sat off the mapped road, or the point cloud found no road bed under the maximum, were re-measured on the state inventory's own centreline. 46 clear an adoption door there with the maximum on the line and join the survey (Pennsylvania 26, Hawaii 11, New York 9; 38 ranked, 8 listed for review). The rest stay unpublished: 121 Pennsylvania candidates whose steepest section still sits off the road, 27 with no road bed under the maximum even on the new line.

2026-09-09 — Estimated surfaces where no inventory records one

About 400 US climbs sit on roads whose state inventory carries no surface. For these the page now shows an estimate read from NAIP aerial imagery and the LiDAR return intensity of the road bed, cross-checked against roads the inventories do describe: shown as 'paved (estimated, 86%)' with its confidence, never as verified. Estimates at 80% confidence or above join the paved or gravel views; roughly one in ten at that level will be wrong, and any inventory or reported surface replaces the estimate the moment it arrives.

2026-09-09 — Steepest sections off the mapped road: 198 pages withdrawn, 14 trimmed

A new check reads how far a climb's steepest 25 m sits from the state road inventory's own line. 259 pages were more than 15 m off it — tracks, driveways and spurs beside a real road. Each was re-measured on the run the inventory covers: 14 keep a page at the shorter extent with the maximum on the road (Austin Road, Tiger Valley Road, Fern Street among them), 47 stay listed for review, and 198 clear no door without the tail or still sit off the line and are withdrawn. Shivley Street in Jeannette went with them under a road-review verdict.

2026-09-08 — 170 held candidates join at their inventoried extent

The owed D28-B waves: 1,075 climbs from the New York, New Jersey, Pennsylvania and Hawaii surveys had been held because the state road inventory covered only 50–80% of the measured route. Each was re-measured on the run the inventory does cover. 170 clear an adoption door there and join the survey (Pennsylvania 107, New York 78, Hawaii 8, New Jersey 2; all but two ranked once the inventory confirmed the road as public); 101 stay held because the named road covers under 80% of the trimmed run; 779 clear no door without their uninventoried tail and are not published; 25 sit within 100 m of an existing page and wait on a duplicate check.

2026-09-08 — Private lanes withdrawn in New York and Pennsylvania

256 pages on roads no public authority maintains were withdrawn: 119 New York roads absent from NYSDOT's Local Highway Inventory for their town, and 137 Pennsylvania segments outside the Liquid Fuels programme — the private and homeowner-association lane class the survey does not rank. Each is ledgered with its reason.

2026-09-08 — Tails beyond the listed road trimmed; 148 more pages withdrawn

The second half of the day's inventory rulings. 175 pages whose steepest 25 m sat on a tail beyond the state-inventoried road were re-measured on the inventoried run: 9 keep a page at the shorter extent, 148 clear no adoption door without their tail and are withdrawn (ledgered with reasons), 7 stay listed for review. New York gains 77 verified surfaces from NYSDOT's Local Highway Inventory, the state's listing of town and village roads.

2026-09-08 — Extents trimmed to the listed road; 208 pages withdrawn

Two extent rules applied across the US survey. A climb page that carries a named road must lie on the state inventory's segments of that name: 110 pages were re-measured on that run and keep their place with the shorter extent, 79 whose named run clears no adoption door were withdrawn. And the steepest 25 m window itself must sit on a listed public road: 140 pages whose maximum lay off every inventoried road (private lanes, tracks, driveways) were withdrawn, and 175 whose maximum sits on a tail beyond the road are listed while their extents are trimmed and re-measured. Every withdrawal is recorded with its reason in the retirement ledgers.

2026-09-08 — Hawaii joins the survey

88 Hawaii climbs measured from the state's USGS 1 m LiDAR (three projects flown 2017–2020 over Kauai, Oahu, Molokai, Lanai, Maui and the Big Island): a sweep of the road network plus the steep-segment hunt over the county road centrelines, with the Cycle to the Sun course and the islands' famous access roads listed. The islands publish in two UTM grids (4 and 5) and are measured in each; the US section now shows its regions side by side, Hawaii as its own panel.

2026-09-07 — Pennsylvania joins the survey

2117 Pennsylvania climbs measured from the state's USGS 1 m LiDAR (ten projects flown 2014–2020): a statewide sweep of the road network plus the steep-segment hunt over PennDOT's state-road and local-road inventories, processed in four longitude bands on four machines. The west of the state, published in UTM zone 17, is measured in its own grid. Where a maximum sat off the inventory centreline the line was re-measured on the road before ranking; the map shades the footprint the survey holds.

2026-09-06 — New York joins the survey

865 New York climbs measured from the state's USGS 1 m LiDAR (twenty-four projects flown 2013–2024, covering almost the whole state): a statewide sweep of the road network plus the steep-segment hunt over the NYSDOT Roadway Inventory centrelines, processed in three longitude bands. The west of the state, published only in UTM zone 17, is measured in its own grid. The map shades the footprint the survey holds.

2026-09-06 — Twenty hand-picked climbs return

A tidy-up on 5 September removed measured pages that clear none of the published cutoffs. It also swept up twenty pages that were never there on the cutoffs: the hand-picked founding set (Penglais Hill, Primrose Hill and Padarn Crescent in Aberystwyth, eight signed Welsh roads such as Roman Way and Allt Goch Fawr) and nine climbs adopted by on-sight review in the Welsh trunk sweep. They are back, unchanged, and "a small hand-picked founding set" is now a cutoff in its own right so this cannot recur.

2026-09-06 — New Jersey, wave 2 — the north-east and Delaware Valley counties

194 New Jersey climbs now: the survey's first pass missed two USGS 1 m LiDAR projects (the post-Sandy 2014 flight over Bergen, Hudson, Essex and Union, and the 2015 Delaware Valley flight over Mercer, Burlington, Camden and Gloucester). Their roads are measured in this wave and the coverage map now shades the footprint the survey holds across the state.

2026-09-06 — New Jersey joins the survey — where its 1 m LiDAR reaches

170 New Jersey climbs measured from the state's 2017 and 2018 USGS LiDAR: a statewide sweep of 3,414 candidate roads plus the steep-segment hunt over the NG911 road centrelines. Coverage is partial by the data, not by choice: the north-west six counties and South Jersey are surveyed at 1 m; Bergen, Hudson, Essex, Union, Middlesex, Monmouth and Mercer have no open 1 m survey yet, so no road there is measured. The map shades the footprint the survey holds.

2026-09-06 — Out of beta — a living survey, methodology v1.0

The site-wide "beta" label is retired. Uncertainty is now stated where it lives — provisional names, measurements awaiting confirmation, canopy and access caveats on each page — so a blanket label said less than the pages do. The measurement method is named version 1.0 (September 2026) so it can be cited while the dataset keeps improving; a citation block joins the licensing page.

2026-09-05 — Six pages with an earlier human verdict settled under the adoption bar

Six pages a person had already looked at fell under the adoption doors on their measured numbers. Four stay visible as listed climbs rather than ranked ones, with the reason on the page; two whose spikes were suspected artefacts of a bridge and a retaining wall are withdrawn, since masking those could only lower them further.

2026-09-05 — Three more re-lined climbs restored; three held for a look

The building-footprint sweep prompted by a reader's report found five more routes passing through houses. Three were re-lines onto the laser point cloud that had wandered off the road (Downe Road, Smith Road, Route 39) and are restored to their earlier measurements; three are roads the map itself draws through buildings and are held from the rankings until someone has looked.

2026-09-05 — Two Devon and Somerset lanes restored after a reader's report

A reader reported that Paddons Lane's measured line left the road and ran through houses. It did: a 4 September re-line onto the laser point cloud had followed a garden terrace rather than the lane, and Tadwick Lane had the same fault. Both pages are restored to their earlier measurements (Paddons Lane's steepest 25 m returns to 21.6% from a doubtful 30.2%), every page has been swept for routes passing through building footprints, and the report form now shows a clear acknowledgment when a report is filed.

2026-09-05 — 110 pages withdrawn below the adoption bar; 22 extents corrected

Every page has now been judged against the same adoption doors a new climb must clear (a 20% stretch over 25 m, a 26% one, 1.5 km at 10%, one of a region's thirty biggest rises, or a listing in a recognised guide). 110 pages that clear none of them, most from early machine sweeps, are withdrawn; any can return through a recognition entry or a human verdict. Separately, ten pages whose steepest stretch lay on a second climb beyond the first summit now describe that second climb, and twelve pages whose original traced line no longer survives were re-measured from their published line with the re-measurement disclosed. A route doctor now runs every check that found these classes on every release.

2026-09-05 — 352 climbs now end at their summit

A survey of every page found 379 whose measured route carried on over the top and down the far side, so the page's net gain and overall gradient described a there-and-back rather than the climb (one Cornish lane with a 28% sign read "gain 1.7 m"). Their routes have been re-measured from the original traced lines and cut at the summit: every steepest-section figure is reproduced exactly from the same laser record, only the trailing descent is gone, so net gain, overall gradient and length now describe the ascent. Ten pages whose steepest stretch lay beyond the first summit are held for a human decision on which climb the page should be, and a handful whose original line could not be recovered are left as they were.

2026-09-05 — Report form restored, clean addresses, one reservoir crossing withdrawn

The site moved to a new host on 5 September and the "submit an update" form silently stopped filing reports for about a day; it works again. Every page now lives at a clean address without ".html" (old links still redirect), and the elevation charts say "net gain" and "total ascent" in the same words as the page. A survey of every route against mapped water found one page whose "road" was drawn across Penygarreg Reservoir in the Elan Valley; the laser record there is a water surface, so the page is withdrawn and the road audit now tests for water crossings.

2026-09-04 — 1,135 English climbs from the inventory hunt

England's half of the hunt: every OS Open Roads link read against Terrain 50, the steep ones with no page nearby traced onto OSM roads and measured on the Environment Agency's 1 m LiDAR across two worker boxes overnight. 4,628 candidates measured, 1,220 clear a door. 794 rank outright, 341 are listed with a road-audit note, 27 wait on a duplicate check, and the road inventory's own status lifts most of the listed ones in the same release.

2026-09-04 — 430 New England pages withdrawn as not state-listed roads

A class of early New England pages ran on roads the state road inventories cover by less than half of the route, most of them in Maine, whose inventory lists even private roads. Absence from it means a driveway, a woods road or a camp track that OpenStreetMap drew as a road, so these pages are withdrawn rather than left in limbo. Any of them can return through a recognition entry or a human verdict.

2026-09-04 — 466 held climbs ranked on the word of the road inventories

Most climbs the inventory hunt found were held from the rankings because OpenStreetMap could not vouch for the road: an unnamed service way, an untagged surface, a gate. The road inventories the hunt started from already answer that question, so where OS Open Roads or a state inventory carries the road as public it now ranks, with a note where Open Roads classes it as restricted local access. The remaining holds are pages the network does not cover or that a point-cloud cross-section still has to settle.

2026-09-04 — 283 New England climbs on unpaved roads

A second pass of the inventory hunt admitted OSM "track" ways as seeds, the class most New England gravel roads are mapped as. 597 more roads were measured on the 1 m LiDAR and 290 clear a door; 263 are ranked, 20 are listed pending the road audit's questions, 6 wait on a duplicate check. Half of them read 20% or steeper over 25 m.

2026-09-04 — 636 Welsh and Scottish climbs from the inventory hunt

The hunt that found New England's hidden climbs has run over Britain: every OS Open Roads link was read against Terrain 50, the steep ones with no page nearby were traced onto OSM roads and measured on the 1 m LiDAR. Wales and Scotland land first: 364 are ranked, 272 are listed but held from the rankings until the road audit's questions are answered, 15 wait on a duplicate check. England follows.

2026-09-04 — 284 New England climbs the map never showed

The first climbs from the inventory hunt: every state road-inventory centreline in New England was read against a coarse surface, the steep pieces with no page nearby were traced onto OSM roads and measured on the 1 m LiDAR, and 302 cleared an adoption door. 195 are ranked (most with the canopy caveat), 89 are listed but held from the rankings until the road audit's questions are answered, 15 wait on a duplicate check against a nearby page. Black Mountain Tower Road (Maine) and Gallup Road (Vermont) lead the wave.

2026-09-04 — New England's off-centreline lines re-measured on the point cloud

The road-bed check that settled England's stray lines this morning ran on the USGS point cloud for the 219 New England climbs whose line leaves the state inventory centreline. Seventy keep their page on the corrected line, thirty-four of them back in the rankings after a hold. Eighteen read under 20% on a verified road bed and are withdrawn. Fifty-eight where the cloud cannot vouch for the line keep their page and number but leave the rankings. Seventy-three lie beyond the cloud or needed no shift and stand.

2026-09-04 — England's off-centreline lines re-measured on the point cloud

Seventy-eight English climbs whose measurement line strays from the mapped road were re-lined onto the road bed the Environment Agency point cloud finds and re-measured there. Seventeen keep their page on the corrected line (most ease slightly). Eleven read under 20% on a verified road bed and are withdrawn: the steepness was the offset. Fifty where the cloud cannot vouch for the line under its steepest 25 m keep their page and published number but leave the rankings until a cross-section look.

2026-09-04 — Steep-under-cover pages settled by the point cloud

Twenty-four US climbs whose steepest 5 m reads beyond plausibility under tree cover were checked against the point cloud under their steepest 25 m: seven sit on a plain road bench and rank with the ≈ mark; seventeen leave the rankings until re-lined. A coarse independent surface now cross-checks every ranked US window, and twenty-one climbs it disputes, where the cloud finds no road bench under the maximum, leave the rankings too.

2026-09-04 — Chart cursors realigned; Maybole Bypass withdrawn

The shared cursor on 1,492 climb pages was reading a stale chart calibration, so the dot floated off the curves; the charts are re-rendered with their calibration and the cursor now follows the window centred on it. Maybole Bypass is withdrawn: the A77 bypass opened in 2022 on ground surveyed in 2017–19, so the profile was construction-era earthworks, not the road.

2026-09-04 — England gets a point-cloud arbiter

The road-bed check that settled the New England re-lines now runs on the Environment Agency's national LiDAR point cloud. Three lanes held as "off the mapped network" turn out to sit exactly on a road bed and are back in the rankings; six are re-measured on the bench the cloud finds (Willow Bank 30.6% over 25 m); five with no road bed within 30 m of their line, or under 20% once re-lined, are withdrawn.

2026-09-03 — Four climbs extended to where the road keeps climbing

Fairy Glen (+225 m at the top, 28.4% over 25 m), Hen Ffordd (+426 m), Hawthorn Road (+167 m at the base) and a lane near Orland, Maine (+300 m) are extended along ground the re-walk found still climbing at 10% or more. The published steepest sections are unchanged; only added ground is new.

2026-09-03 — The caveated door, and 85 withdrawals

Climbs whose steepest 25 m is 20% or more but sits under tree cover or beside a bridge mask now stay ranked with the ≈ mark and the caveated tag; the clean-only toggle hides them. 85 US climbs whose best uncovered 25 m falls under 20% once canopy is judged are withdrawn.

2026-09-03 — Canopy judged across the US survey; 205 extents corrected

Every US climb now carries a point-cloud canopy assessment (most had been adopted without one); where the steepest section lies under tree cover the page says so and the ranking marks it. 174 US and 31 UK pages that ran on past the summit or into a flat run-out are cut to the climb; their steepest sections are unchanged, their lengths and overall gradients corrected. Bridge and culvert masks were carried through every re-measurement and verified page by page.

2026-09-03 — Four re-measured pages and one placement fix

Pen Dref and Rosedale Chimney Bank are re-published from re-measurements that reproduce from banked inputs (Pen Dref's published 39% over 5 m was a single artefact; it now reads 23.6%). Two Welsh lanes that clear no adoption door after honest re-measurement are withdrawn. White Downs is re-measured on its recoverable line so the placement mask could apply (32.0% over 5 m is now 26.5%).

2026-09-03 — Canopy caveats restored on 74 corrected climbs

The 74 re-measured New England climbs that stayed after the retirements have been re-run on the same lines with the tree-canopy check the survey applies everywhere else. Forty-eight now carry a canopy assessment again; sixteen headline maxima moved because the steepest window sat under trees or off a road bed and the survey re-sites it to the best measurable stretch. The same check is now running across every US page that never had it.

2026-09-03 — Sixty corrected pages retired — the bar applies to corrections too

Over the last two days 146 climbs were re-measured on lines moved back onto the road. Sixty of them now measure below the survey's adoption bar on every count — typically a 20–22% page reading 15–19% on the corrected line. Their old numbers came from the off-road line, so the pages retire rather than stay at a figure they never earned. The same bar means the same thing in Maine as in Wales only if it is applied to corrections as well.

2026-09-02 — Coverage footprint, clean-only rankings, maps that stand alone

The maps now shade where the survey actually holds open 1 m LiDAR and has screened every road — computed from the tile archive, not asserted — so an empty patch reads as "not yet measurable" rather than "no hills". Every ranking table gains a Clean-only switch that hides caveated maxima. The map library is now served from the site itself, with a static map shown if it or the basemap cannot load.

2026-09-02 — Scope stated exactly, blank States filled, caveats labelled

An outside review caught three things worth fixing at once. The US section now states its scope exactly (the six New England states plus one border road starting in New York) and the site refuses to publish a ranked US climb without its state — 123 recently re-measured pages had shipped with a blank State cell. Caveated maxima carry a visible "caveated" label in every ranking, the cross-survey table says which surveys it covers, and each climb page gains a plain-language measurement card above the raw provenance record.

2026-09-02 — Five pages retired — their steepness was a mapping zig

A new check found route lines that zig sideways where the road does not, so the sampler read a gradient the road never had. Cleaned against OS Open Roads, one climb (Gay's Hill) re-measures and stays; five others fall below the survey's 20% bar and are retired rather than kept at a number they never earned. Seven more stay held for hand geometry.

2026-09-02 — Re-surveyed lines, named twins merged

Twenty-three climbs re-measured on lines traced back onto the road (nine in New England from the LiDAR road bed guided by the state road inventory, fourteen in England and Wales snapped to OS Open Roads); most of the headline gradients came down, a few went up. Nine pages that were the same climb under two names now redirect to the page that keeps it. Twenty-eight lines remain held for re-survey with the reason stated on each page.

2026-09-02 — 55 more climbs re-measured, this time from the LiDAR itself

Where neither the map nor the state road inventory agreed with the ground, the measurement line has now been traced from the LiDAR point cloud's own image of the road bed — and accepted only where two independent checks pass. 55 climbs re-measure on such lines today; corrections ran both directions, including two climbs that turn out steeper than published. The remaining held pages stay held for honest reasons, each stated on its page: the road lies beyond the tracer's reach, the point cloud disputes it, or the survey has no coverage there yet.

2026-09-02 — The same scrutiny for Britain, and 13 duplicate pages merged

The measurement-line check now covers Britain: 34 UK climbs whose steepest published sections sit where the line drifts off the Ordnance Survey road centreline leave the rankings pending re-survey, exactly as their American counterparts did this morning. A corpus-wide geometry sweep also found 13 pages that were duplicates of climbs already published under better names — each now redirects to the page that keeps the climb. And 16 US pages carry a new existence note where the LiDAR point cloud finds no road bed under their steepest sections.

2026-09-02 — 68 US climbs re-measured on corrected lines

The re-survey announced this morning has its first results: 68 US climbs re-measured on lines snapped to their state road inventory's centreline, each snap verified against the LiDAR point cloud's own image of the road bed. Most steepest-section figures came down (the old lines had drifted onto roadside ground); a few went up — the drifted line had been hiding real steepness. Twelve climbs return to the rankings with honest numbers. The remainder of the held set awaits a deeper correction, traced from the LiDAR road bed itself.

2026-09-02 — 64 US climbs held for re-survey

A new cross-check compares every measurement line against the state road inventories' own centrelines. Where a line drifts off the carriageway exactly at a climb's steepest published section — onto a cut bank or roadside ledge the LiDAR faithfully measured — the numbers describe the hillside, not the road. 64 US climbs whose headline gradients sit on such a drift leave the rankings today while their lines are re-surveyed against the inventory centreline; each page stays visible with its hold stated. Honest numbers over big ones.

2026-09-01 — 78 New England climbs recovered the same way

The same access-boundary fix that recovered 175 British climbs today applies across the Atlantic: every New England candidate the whole-route check had rejected was cut at its access boundary and the public remainder measured on its own merits — with one extra safeguard, since American road records vary by state. Each remainder was verified against its state's own road inventory (VTrans, NH GRANIT, Maine E911, MassDOT, CTDOT, RIGIS), calibrated first against roads this survey had already measured. 78 new pages join the US section, each carrying a surface chip from that inventory chain where the records support one — and an honest "unverified" where they don't.

2026-09-01 — 175 climbs the whole-route audit had hidden

When the survey grows a candidate route along the road network, the growth sometimes runs one step too far — onto a private drive, a gated track or a service way — and until now that single appended segment condemned the whole candidate: the access check judges the full route, so the climb was rejected outright and never measured. The survey now cuts such routes at the access boundary and measures the public remainder on its own merits. Re-auditing every England and Wales candidate rejected this way recovered 175 new pages — including recognised climbs like Bamford Clough (the public road clears 20% over 25 m; its famous upper section is private and is not measured here), The Struggle above Ambleside, and a cobbled John Street in Heyrod. Each recovered page records where and how much of its original route was cut. A further set of candidates whose measurements sat suspiciously close to bridges and masked structures is held back for re-measurement rather than published.

2026-08-31 — Lanes the elevation model cannot see

Some lanes run so deeply sunken between banks, walls or hedges that the national elevation model holds no road surface where they go: what it records there is the bank, and a gradient read off it says more about exactly where the route line was drawn than about the hill. The survey now recognises that shape directly — no road bench under the line, and ground falling more than two metres across the metre either side of it — and excludes those stretches, publishing instead the steepest section it can actually resolve. Eighty-eight climbs carry the new exclusion and 42 of them see a published maximum change. The largest is Cotley Lane in Devon, whose steepest 5 m falls from 43.6% to 25.7%: the model never held a carriageway at that spot to measure. As always the displaced reading is kept and shown on the page, and no elevation data was altered.

2026-08-30 — Hillpark Road and The Walk re-measured

Two Scottish climbs are back on native survey lines. The map snapshot they were originally measured against no longer exists anywhere, so they were re-measured against a freshly fetched one — and both returned their published gradients unchanged, which also confirms those roads have not moved in the interval. No published gradient moved.

2026-08-30 — Thirteen climbs re-measured from scratch

Thirteen Welsh climbs whose original measurement runs were lost have been rebuilt from the raw data instead of recovered: the survey re-ran its own discovery and route-growing steps against the same LiDAR tiles and the same dated OpenStreetMap snapshot the pages were first measured from, and regenerated each route from the road up. Every one of the thirteen came back with its published gradients unchanged to two decimal places, which is the strongest evidence yet that the ladders on these pages were right all along. Ayron Street, Brynteg, Heol y Beiliau, Irene Street and Penrhiw Lane are among them.

2026-08-30 — Hardknott Pass and four more return to native lines

Hardknott Pass, Lowfield Lane's neighbours and four other climbs are back on their original survey lines. Their measurement runs were found in the project's cloud archive rather than on any working machine — several eras of output had never been searched. Each reproduces its published ladder exactly, so as with every step of this restoration, no published gradient moved.

2026-08-30 — Nine more restored, Bank Road and Pea Royd Lane among them

Nine further UK climbs are back on their native survey lines, including Bank Road in Matlock, Pea Royd Lane in Stocksbridge and Rosedale Chimney Bank. Their original measurement runs turned out to be sitting in an older set of working outputs that the restoration had not searched, and each one reproduces its published ladder exactly. No published gradient moved.

2026-08-30 — Ten more climbs rejoin the restored set

Ten further UK climbs are back on their native survey lines, including Llandeilo Road, Mountain Road, Rushock, Carr Street and Coronation Street. These were pages whose published gradients came from a masked measurement — a bridge deck, a tunnel mouth or dense canopy, which the survey excludes and measures around — so re-deriving them from the raw terrain could never reproduce the published figure. Re-running the full measurement pipeline, which applies that masking, matched every one of them exactly. As with the rest of the restoration, no published gradient moved.

2026-08-30 — UK pages get the linked charts — and their survey lines back

The linked charts that arrived on the US pages last week are now on the UK ones: hover or touch the map, the elevation profile or the gradient trace and a single cursor follows the same point across all three, reading out distance, elevation and gradient as it goes. Getting there meant a repair first. A recovery in August had rebuilt most UK route lines from the simplified copy used for the "open in geojson.io" links — roughly one point every 40 m — rather than the metre-by-metre line the published gradients were actually measured along. Those original lines have now been restored from the archived measurement runs, and each page's charts rebuilt from the exact samples behind its own published ladder. The published gradients themselves have not moved: where a restored series did not reproduce a page's numbers exactly, that page was left as it was rather than quietly redrawn. 1,606 of 1,661 UK climbs are back on their native survey lines; the rest keep their current charts while their lines are re-measured.

2026-08-30 — The restricted-fraction gate opens — 49 held climbs join

Downe Road's story generalized: an audit found 59 measured English climbs of 20%+ that the adoption pipeline had silently held because part of their route touches restricted-access ways. A window-local check split them: 17 whose steepest section is fully on open road joined with the restriction noted (topped by Mill Lane at 31.2% over 25 m); 25 more return after their private tail ways were trimmed off and the open road re-measured (led by a 38.7% Devon lane and Lindridge Hill at 30.7%); after an imagery review, 7 more join with access notes, 2 are confirmed private and stay out, and 7 — including a 41.8% lane that would top England if it holds up — await on-the-ground verification.

2026-08-30 — London's steepest road joins the survey

A reader-era blog audit (diamond geezer's "London's steepest hills") caught a gap: Downe Road in Cudham was laser-measured months ago — 23.1% over 25 m, the steepest thing we've measured inside Greater London — but never published, having slipped through an adoption pass. It's live now, with the blog credited on the page. The same cross-check confirmed BikeRadar's "7 steepest UK climbs" list against our measurements — including their Afon Ddu in the Conwy Valley, which is the unnamed lane above Dolgarrog this survey found by machine and lists as the UK's steepest at 36.6% over 25 m.

2026-08-29 — Linked charts — see exactly where the steep bit is

Every US climb page now speaks one visual language across its three graphics. The route on the map is painted by gradient — the red stretch on the map is the red spike in the chart — with distance markers every 500 m, and the climb's steepest 25 m is highlighted on the map, the elevation profile and the gradient chart alike. Move your pointer (or finger) over any of the three and a shared marker follows on all of them, reading out distance, elevation and gradient. The home-page ranking's ellipsis rows now say which ranks they span and where the full tables live. UK pages get the same treatment in a following update.

2026-08-29 — The gate-offset sweep — 46 more US lines corrected at their steepest point

A new audit compares every climb's steepest 25 m window against the state's own E911 road centerline: 69 ranked climbs had their headline window measured 8–26 m off the mapped road — a bank or hillside beside it — while the rest of the route tracked the road faithfully. Forty-six re-measured lines validate and go live (Sodom Road falls from 32.9% to 13.9%); most corrected down, but Old Cricket Hill Road's honest line is steeper — 29.5% over 25 m, laser-verified as a built road bed. The rest stay held for deeper review. A same-day hand-check pass with the surveyor's own eyes on Street View settled fifteen more: two routes had begun on the wrong road entirely (Flagg Hill Road's 23.1% ramp belonged to Burnt Hill Road; West Cornwall Road started on a covered-bridge approach), and Taft Road became the first climb corrected by the laser alone — its line rode a stream-gully edge, so we moved it onto the road bed the point cloud itself detects (23.4% becomes 20.5%, and the cross-section now reads clean road the whole way).

2026-08-28 — The US re-line pass — 25 measurement lines moved onto the real road

A LiDAR cross-section audit found routes whose measurement line sat metres off the actual road bed — sampling ditch banks and hillsides instead of tarmac. Twenty-five US climbs are re-measured on lines snapped to their state's own road inventory, each move verified against the laser's detected road bench. Most corrected downward (the steepest window was a ditch), but two corrected up: Worden Road's honest line is 28.7% over 25 m, not 20.5%. Spiegel Road also returns: an OpenStreetMap mapper moved its mapped way onto the real road, and the fresh line measures 22.1% over 25 m. Two same-day Street View catches joined the pass: Warren Hill Road (Jay, ME) was measured up to 20 m off the E911 road at its steepest window (27.0% falls to 15.8%), and Navin Avenue (Lee, MA) began on a bridge deck whose under-span riverbank read as a 34% ramp (23.7% falls to 19.0%), and Richardson Hollow Road (Greenwood, ME) — whose line bulged 15 m onto a roadside bank at its steepest window — drops from 26.8% to 19.7%, now measured on the road's real hill.

2026-08-28 — Community-reviewed pages return, and culvert fixes reach New Hampshire

Two New Hampshire climbs retired by our imagery audit — May Street (Enfield) and Eastman Hill Road (Lebanon) — are back: an OpenStreetMap mapper reviewed our MapRoulette audit tasks and confirmed both roads exist as mapped. Their pages return with fresh measurements and the reviewer credited. The same audit round completed the New England culvert cross-section survey (New Hampshire's tiles were previously unreadable), correcting three live pages whose steepest windows sat on drainage artifacts — Hale Road's headline drops from 20.8% to 12.1% over 25 m.

2026-08-28 — The fame door opens — 47 storied climbs the dials alone had missed

An audit of every National Hill Climb Championship venue and Simon Warren top-100 entry against the full measured corpus found dozens of famous climbs that were surveyed long ago but never paged — the recognition matcher had only ever seen the published site, so a legend that wasn't steep enough for the dials had no door in. It does now. Forty-seven new pages, led by Winnats Pass (venue of eleven national championships), Park Rash's hairpins at 24.8% over 25 m, Mow Cop's Killer Mile, the world's oldest hill-climb course on Yorks Hill, and Vale Street in Bristol at 27.2% — plus Cheddar Gorge, Cat and Fiddle, Holme Moss, Hartside and Michaelgate's cobbles, paged on fame rather than gradient. Two long-published lanes turn out to be Boltby Bank and Langcliffe Scar and now carry their names; recognition chips land on seven more existing pages. Every route ran the full QA battery; three had bend-cut line offsets re-sited to the OS centreline before adoption, which is why Rowsley Bar's ladder says an honest 22.6% rather than the 48% the raw line wanted to claim.

2026-08-27 — England and Scotland join the chain wave — 54 more short, steep streets

The fragment-chain screen swept England and Scotland overnight: 26,979 English fragments chained to 295 candidates and Scotland's to 14, every one through extend, the full QA battery and the line-offset re-site round. 50 England climbs and 5 Scotland climbs (one, on the Wye at Tintern, was the same Welsh road found from both sides of the border — counted once) clear the bar and join the catalogue, led by Eden Mount in Cumbria at 29.8% over 25 m; 245 English and 9 Scottish candidates are held with recorded reasons — off the road network, unresolved line-offset flags, or simply below the page bar. No ranking crowns change.

2026-08-27 — The streets the screen was built for — 22 join Wales, and a new UK crown

A new discovery instrument — the fragment-chain screen — went looking for short steep streets that OpenStreetMap splits into pieces too small for the standard search to score. Its Wales pilot adds 22 climbs, every one through the full extend, QA and measurement-integrity battery (line-offset against the OS centreline, Terrain-50 cross-check, bridge-flank scan, road audit), with flagged candidates re-sited onto the centreline and re-measured or held. Among them: Pen Dref in Harlech (22.1% over 25 m, two streets from Ffordd Pen Llech — reported by a reader of the survey, provisional pending one review) — and a new UK steepest-25 m crown: the hydro service road climbing the Afon Porth-llwyd gorge above Dolgarrog at 36.6%, listed with a plain caveat because its riding rights are unresolved. That is the new policy working as intended: a real road lists, with its restriction stated.

2026-08-27 — Restricted-access climbs join the rankings — with their caveat

A real road now lists with its restriction stated rather than being ranked away. The US toll and state-park roads (Mount Washington Auto Road among them) join the ranking tables carrying a controlled-access note on the row, with details on each page — and the same policy will hold for UK climbs with uncertain rights of way as they arrive. Roads whose physical existence the inventories cannot vouch for remain held until verified: access uncertainty is noted; existence uncertainty is not ranked.

2026-08-27 — US pages stop speaking British — and rankings get a surface toggle

Outside review caught UK boilerplate riding on US pages, and it was right. Every discovered US climb page claimed a cross-check "against the OS road network" — there is no Ordnance Survey in Vermont; those pages now describe the checks the US pipeline actually runs (state road inventory, OpenStreetMap, point-cloud evidence). Footers now credit the sources actually under each survey's numbers — USGS 3DEP and the state road inventories for the US — and the licensing page lists the US sources it owes. The US map now frames all six states (Maine, half the corpus, sat outside the old opening view) and no longer describes overlay layers only the UK map has. The reuse credit becomes "HowSteep — howsteep.org" sitewide, shared pages (format, licensing, about, submit, updates) drop their UK-only titles, the submit form now suggests US climbs too, and the survey cards state both of a country's numbers: pages published and climbs in the public-road rankings. Pipeline working notes no longer ride into published climb documents. And every ranking table gains a surface view: paved roads are the preselected default, with an all-surfaces toggle for gravel and mixed roads (UK climbs are paved by the survey's own scope; US surfaces come from each state's road inventory) — and table notes say plainly that a ranked row may be a brief sharp pitch or a long traditional climb, with the length and overall columns telling you which. The about page now walks through what the Cefn Llan 32% sign would have to mean: the survey finds that slope only at the 2 m scale, exactly where measurement is hardest.

2026-08-26 — US homepage catches up with the UK — and every page gets a date

The US index now mirrors the UK homepage: steepest 10/25/50/100 m tables, largest net rise, and a searchable, sortable catalogue of all published US climbs with state, surface, status and confidence filters. The methodology page gained a "How the US survey differs" section (elevation sources, canopy checks from the point cloud, state road inventories, and the road-existence audit). Every page footer now carries the data-and-reviews-as-of date you are reading here.

2026-08-26 — The road-existence audit — 26 pages retired, the map thanked

A new instrument asks the LiDAR point cloud whether a physically graded road bed exists along each route, with aerial imagery covering the flat ground where geometry alone cannot decide. It audited all 10,672 measured New England routes overnight: roughly 4,600 roads came back physically confirmed, and 26 published pages failed human review of the evidence and were retired — the US count fell from 1,433 to 1,407. Verified map errors were returned to the OpenStreetMap community as a MapRoulette challenge, because the map this survey is built on deserves the findings back.

2026-08-25 — New England at full scale — 1,433 US pages

The discovery waves landed: every published US row now carries a surface chip from its state's own road inventory, rankings hold public roads of any surface with a paved-only filter, roads the inventories cannot vouch for sit in a labelled "road status under review" table rather than the rankings, and each ranked row's steepest window was checked against the inventory individually.

2026-08-23 — HowSteep US launches — New England's founding set

The survey crosses the Atlantic on the same method and the same rules, measured from the public-domain USGS 3DEP 1 m laser surveys: the famous gap roads and hillclimb courses of New England alongside steep lanes only the locals knew.

2026-08-21 — One survey, per-country structure — and settled licensing

The UK survey moved to howsteep.org/uk/ (old links redirect) and the front page now introduces the survey as a whole, with the United States (New England first) in preparation. Licensing settled and published: any individual climb's page and charts are CC BY 4.0, the dataset is ODbL 1.0, and every downloadable climb document now states its licence. The methodology page now documents how the numbers are policed — the masks, the post-batch detector battery, and the human review that adjudicates every flag.

2026-08-20 — Measurement-integrity audit — the climb count went down

A corpus-wide audit retired roughly 95 pages whose headline measurements failed scrutiny (route lines off the carriageway, masked-structure artifacts, roads newer than the elevation survey beneath them) and re-sited 156 routes onto the Ordnance Survey road centreline. 1,500 climbs live. Retirements are public policy: an error we published and caught is removed, not quietly patched.

2026-08-19 — Great-Britain-wide re-pick — 798 pages

The page-selection rules were re-applied across every measured road in Great Britain, growing the site from 523 to 798 pages (England 221→411, Wales 210→272, Scotland 51→59) and rescuing climbs earlier versions had wrongly excluded, Honister Pass among them. Area labels reworked to name where each climb starts.

2026-08-18 — Northern Ireland joins — all four nations measured

Northern Ireland's open river-basin LiDAR reaches roughly a quarter of its roads — enough to measure Belfast's steepest streets, not yet the Antrim coast. Six pages at launch, measured the same way as everywhere else.

2026-08-17 — Scotland launches on a growing record

49 climbs from Scotland's open LiDAR patchwork (the national programme is capturing the rest of the country through 2027). Published exactly what the open record supports — a famous missing hill is almost always waiting for data, not judgement.

2026-08-16 — England launches

212 English climbs measured from Environment Agency LiDAR join the survey after the bulk-tile campaign.

2026-08-13 — Recognition, provisional names and the report form

Climb pages now show which books, championships and road signs match each measurement — recorded as evidence and tested, never copied. Machine-suggested names are marked provisional (°), and a no-account-needed report form collects signs, names, gates and corrections.

2026-08-12 — howsteep.org goes live — the Wales survey

257 Welsh climbs, every gradient stated with its measuring distance, born from one contradictory road sign above Aberystwyth.