Known issues
Where the API's answers are known to fall short today, how much, and what to do until it is fixed. Checked 12 October 2026. When one is fixed it says so here and on the changelog.
Race pace pages and the pace plan can differ over gorges Fixed on 7 October 2026
Since 7 October 2026 the pace plan removes terrain outliers in gorges, as route analysis does. The race page's pace calculator reads the course file without that step, so over gorges its climb, and so its splits, can differ from the pace plan's for the same course.
- Of 23 courses with race pages, 22 agree within 2 % of the finish time at every 5 km mark; UTA22 reads 1,409 m on the page against 1,258 m in the pace plan, its marks 2.2 % apart; measured 7 October 2026.
Meanwhile: Fixed on 7 October 2026: for a course file TrailSplits serves, the race page now reads the pace plan's climb and splits, so both agree on all 23 courses checked (UTA22: 1,258 m on both).
No terrain heights in the southern Caucasus: climbs read 0 m Fixed on 12 October 2026
Over Armenia, Azerbaijan, southern Georgia (Tbilisi included) and the neighbouring areas around Kars and Tabriz, the terrain model answers 0 m. Heights, climbs and hike times along lines there rest on that hole.
- 0 m at every point of a 0.5° grid over about 38.5–42.3° N and 42.7–49.7° E; every Armenian sample line (36) and 4 of 22 Georgian ones read 0 m at start, middle and end; checked 6 October 2026.
Meanwhile: Fixed on 12 October 2026: the terrain model fills the area from SRTM (credited), and the planet layer is float-first. Yerevan reads 1,001 m, Tbilisi 607 m and Ganja 425 m; no summit in the 278-peak heights check reads 0 m any more.
Search sometimes misses names in another language, exonyms and typos Open: being worked on
Place search finds a name as it is written locally well, but often misses the same place by its name in another language ("Genfersee" for the Léman), by an exonym ("Lake Lucerne"), or with a small typo.
- Right place in the first three: 98 % for names as written locally (57 queries), 88 % for names in another language (26), 77 % for exonyms (13), 77 % with a typo (13); measured 8 October 2026, after the search release with other-language names and ranking v2 (3 October: 90 %, 38 %, 23 %, 54 %).
Meanwhile: Send the name as it is written locally where you can, and pass bias_bbox for the area your user is looking at.
Hike times run short outside Switzerland Open: correction waits for more data
Hiking times come from a model calibrated on Swiss marked paths. Elsewhere they are often shorter than the times route operators publish.
- Against 221 routes with operator times in eight countries: median difference 15.4 %, and our times 12.9 % shorter on average; measured 12 October 2026.
- Switzerland 5.4 % (40 routes); Greece about 28 % shorter (40); Britain about 25 % shorter (9 routes, a small sample).
- Norway: no verified walking-time truth. Of 10 routes checked against the operators' own pages, none matched: most pages give the return trip, or differ from the times mapped on the routes, so we cannot yet say how far our times are off there (checked 7 October 2026).
Meanwhile: Each time now says which countries the line crosses and how many published legs from each are in the calibration (time.region), with notes for Britain, Greece and thinly covered countries. In Trentino, route analysis also gives the times the local alpine club publishes for its own trails (time.models, since 6 October 2026); elsewhere there is no published convention yet. Outside Switzerland, treat the times as optimistic. Where a call takes a pace (route weather, go windows, route cards), a pace_factor above 1 or your own leg times fit your users better.
Measured stage times run short of the operators' times Open: correction waits for more data
Where an operator, hut or official trail association publishes a time for a stage, our measured stage time for the same line is usually shorter. Our times are moving time without breaks, while published times usually include short stops. stages-huts/1 also still carries each guide's own stage time, which the guide pages no longer quote; it runs shorter still.
- Against operators' stated times on 43 stages in 3 long-distance guides, most of them on the Kungsleden (12 October 2026): our measured stage times typically 15 % shorter; the guides' own stage times, kept in stages-huts/1 for comparison, typically 27 % shorter (44 stages).
- The lengths agree closely: the typical stage is 2 % shorter than the official length (61 stages, 12 October 2026), so the gap is time, not distance.
Meanwhile: Where the operator states a time, the guide page shows that time (39 of the 44 stages that state one), and stages-huts/1 carries it as the operator's time. Plan a day from the operator's time where there is one, and read ours as moving time.
Exits: a road had one map point, not one per kilometre Fixed on 5 October 2026
Route exits gives, every kilometre, the walking minutes and metres to the nearest road. Each road appears once in exits[], at the point where its best way out meets it. Other kilometres that lead to the same road may meet it somewhere else, so a line drawn from them to that point is not the way a walker would go. Fixed on 5 October 2026: each kilometre’s road entry in along[] now carries meet {lon, lat}, where that kilometre’s own way out meets the road.
- Against official Swiss roads and stops on 32 stretches: a road was found wherever one lies within 500 m (124 points); 10 of 171 road walks reach the road elsewhere than its map point; measured 3 October 2026.
Meanwhile: Draw each kilometre’s way out to along[].road.meet; exits[] still gives the point of the line’s best way out to that road. Stops, villages and huts are single points and were not affected. Remember a mapped way out is never a safe one: lifts and pass buses can be seasonal, and a hut is shelter.
Near sharp summits, the evening shade came too early Fixed on 3 October 2026
With sun: true, route weather says where the line is in sun or shade. On sharp summits it read the setting sun as hidden by terrain up to an hour before sunset, where the horizon is open. Fixed on 3 October 2026: a summit no longer shades itself.
- Against 2 m lidar, every judged check at both summits now agrees: the Säntis in 27 of 27 checks (27 of 37 after the first fix), the Niesen in 45 of 45 checks; measured 3 October 2026.
- Across the headline set, the sun-or-shade state agrees for 96.6 % (1,384 of 1,433 points), up from 95.9 %; measured 3 October 2026.
Meanwhile: Still true: at a sharp summit the stated horizon angle can be too high (about 14° on the Säntis, where lidar gives about 0°), because the terrain cell averages in the cliff below. There the state reads borderline rather than shade.
Flat routes read a little climb that is not there Fixed on 3 October 2026
On flat city and coastal courses, route analysis read some climb from bridges and waterside edges in the terrain model. Since 3 October 2026, rises under 4 m on a line that stays flat are ignored (method route-smoothed-100m+flat4).
- Extra climb on flat courses: 0.21 m per km (median of 5 courses), down from 2.77; 4 of 5 flat courses within the rule, up from 1; measured 3 October 2026.
- Of the 5 courses, the Gold Coast marathon now reads 71 m against an official 62 m, where it read 179 m.
Meanwhile: Still open: a bridge more than 4 m high can add climb (Kaunas reads 214 m against 95 m). On flat lines outside Switzerland, where the terrain is the Copernicus surface model, answers now carry climb_low_confidence: treat their climb as approximate.
Something wrong that is not here? Tell support@trailsplits.com. The live answer times are on the status page.