Two almost corner free routes in central Seoul

Two routes and what each is for

Both Seoul routes are paved, and neither is a loop.

The Cheonggyecheon Stream Path is 4.11 km and does not finish where it starts. The Gyeongui Line Forest Park route is 1.96 km, and at 1,956 m it is the shortest curated route in the pack, at the other end of which sits London's Richmond Park Tamsin Trail at 11,569 m. The two do different jobs.

Distance is not the difference that matters when you train alone. It is the number of decisions. Across 4.11 km the stream path carries one turn cue. The forest park route carries none. Once you are on either path, the route data leaves you one decision: when to stop.

The Cheonggyecheon stream path

Four kilometres and change, paved throughout.

One turn cue, and it ends somewhere other than the start. Decide your return before you set off, because the route does not bring you back.

The stream is longer than the route, and the city's own page carries two different figures for it. The Seoul stream page describes the stream as 10.84 kilometres long in its overview, while its restoration project table gives a total length of 8.12 km with a restored length of 5.84 km, fetched 2 August 2026. Either figure leaves the curated 4.11 km covering part of the restored stretch, so there is water beyond both endpoints when you want more.

From Cheonggye Plaza the stream passes under a total of 22 bridges before it flows into the Hangang River, per Visit Seoul, fetched 2 August 2026. Count them out and match the count coming back, and you have a split structure that needs no screen.

Rain is the hazard on this path. The city page warns that the water level rises very quickly on rainy days and the trails become submerged without warning, tells visitors to evacuate through the nearest exits when it rains, and states that areas below bridges are especially hazardous because floodgates can open, in the same stream page, fetched 2 August 2026. On a wet day the bridges are the thing not to shelter under, and the exits up to street level are the thing to know in advance.

The Gyeongui Line Forest Park

Under two kilometres, paved, no turn cues.

It also ends away from its start. A warm up or a shakeout. One leg of something longer, and not a session on its own unless you repeat it.

Again the park outruns the route. Seoul's guide describes a 6.3 km linear park from Gajwa Station in Yeonnam-dong, Mapo, to the Wonhyo 1-dong Community Service Center, passing Hyochang Park Station in Yongsan, in its park entry, fetched 2 August 2026. The curated route covers under a third of that, so the simplest way to extend it is to keep going.

The park exists because a railway left. The city describes it as a defunct section of the Gyeongui Line turned into a greenway after the line was put underground, on the Seoul Metropolitan Government page, fetched 2 August 2026. A line built for trains bends gently, and the stored route reflects it: zero turn cues across the whole 1.96 km.

Go early if you want the path to yourself. Visit Seoul's park entry calls the Yeonnam-dong section a popular gathering spot where young people lay out mats and enjoy beer. Good evening, poor tempo run.

A route with no corners is a fact, not a defect

Zero turns is stored as zero turns.

Runflake says so on the course screen instead of going quiet: This course has no turn guidance. Follow the route on the map. The line shows once your position has been matched to the route. The silence is labelled, not left for you to interpret.

That is deliberately a different sentence from the failure case: Turn guidance could not be loaded for this course, so this run has no turn cues. Both leave you with no cues for the whole run, and they mean different things. One says the path has no corners. The other says the turn data did not arrive, so the map is the only guide you have.

Zero here comes out of a rule you can check. A bend only counts as a turn if the direction of travel changes by at least 55 degrees, measured 35 m before and after, and the same 55 degrees still holds over a wider 56 m window. This stretch of converted railway never changes direction that hard, so the honest output is nothing. How the cues behave where there are corners is covered in city routes and turn cues.

Signal along a sunken stream path

Start with what nobody here knows. There is no measurement of how a phone behaves along either of these two paths, not in the route data and not in this article. Anyone who tells you how many satellites you hold beside a particular bridge is guessing.

What is documented is the general mechanism. The United States government's GPS accuracy page lists satellite signal blockage due to buildings, bridges and trees among the local factors affecting accuracy, and states that GPS enabled smartphones are typically accurate to within a 4.9 m radius under open sky, fetched 2 August 2026. Read it in reverse and the useful part appears: the accuracy your phone reports is a moving number, and it moves for reasons unrelated to how you are running. Two of the factors it names, buildings and bridges, describe this ground, since the stream runs beneath the bridges counted earlier and between the blocks of central Seoul. How much either one moves the number on this path is unmeasured, and stays unmeasured until someone runs it with a logger.

So a fixed off route corridor is the wrong shape for the problem. The corridor is 2.5 times the phone's own reported accuracy, with a floor of 45 m and a ceiling of 80 m, so it tightens when the phone claims confidence and widens when it does not.

A moment outside it is not enough either. You are declared off route only after 12 continuous seconds outside, or 40 m of route progress while outside. The run keeps recording and the app says so: You are {N} {unit} off the route. Keep running. Guidance resumes when you are back on it. That is a report and not navigation. No new path is calculated and nothing points back towards the route, so returning to it is your job and a paper map or the stream itself is the better guide. If signal goes altogether, course judging stops instead of blaming you: No usable signal, so the course is not being checked. Your run keeps recording.

Those thresholds were derived in a GPS simulator run over the route pack, and the source states plainly that there are no real device logs behind them. Take them as a design position, not a field result. Receiver behaviour in built up ground is covered in GPS in tunnels and urban canyons.

Where the route data comes from

The shapes are not drawn in house. They come from OpenStreetMap and are credited to OpenStreetMap contributors under the ODbL licence. The OpenStreetMap copyright page states that the project is open data, licensed under the Open Data Commons Open Database License by the OpenStreetMap Foundation, fetched 2 August 2026.

Everything else on a route record is derived from that shape rather than typed beside it. Turn instructions are computed from each route's own shape by the turn detector before publishing, not written by hand and not taken from a navigation provider.

Elevation is simply absent.

It was never measured, and an unmeasured value is stored as unknown instead of zero. That is why this article says nothing about how hard either route climbs.

Adding distance without leaving the path

Neither route finishes where it starts.

So the first lever is the obvious one. Out and back turns 4.11 km into 8.22 km and 1.96 km into 3.92 km, on ground you have already inspected.

Running a stored course backwards switches turn cues off completely instead of mirroring them, and the app tells you: You are running this course in reverse, so turn cues are off. Bear left and U turn are not mirror images, and a wrong instruction at a junction is worse than none. On the forest park route you lose nothing. On the stream path you lose one cue on the way home.

1. Double back at the far endRun to the stored finish, turn, and retrace. The surface does not change and the length is known before you leave, so the only open question is where in the two legs you put the effort.
2. Run past the end of the stored routeThe forest park route covers under a third of the 6.3 km park, and the stream route covers part of a longer restored stretch. The extra distance is already underfoot.
3. Put the work inside the distanceInterval phases end on a time or on a distance, and one of the four fixed presets is eight 400 m repeats with 200 m of recovery, so track repeats are a phase the app counts and not a stopwatch guess. There is no custom builder, so the four presets are the whole choice. Distance goals also carry one failure mode: if signal stops, the distance stops advancing and the phase never ends by itself, which is why the run screen keeps a control that ends the intervals while the run carries on recording.

For a path with one corner in four kilometres, that is enough to build a training block from. What the route gives you is a surface and a length measured along its own shape, and the rest is yours.