Press once and get somewhere to walk to, picked out of a scatter of random points around where you are standing. Built for the Mudita Kompakt and its E Ink screen.
Yorimichi is 寄り道 — the stop you make on the way, the turning you take that was not on the route. Which is the whole of it: a place nothing in your day would otherwise have taken you to.
Not a fork. Written from scratch in Kotlin and Jetpack Compose, using Mudita's own MMD design system so it looks like the apps the phone already ships with.
 |
 |
 |
 |
Version 0.1.0, and honest about it: the maths is unit tested, every screen has been driven on an emulator, and the signed release has been installed and exercised — but nobody has actually walked it yet. Not once. It has never been outdoors.
So the things still unknown are the ones only a walk can answer. How long the GPS takes to fix under trees. Whether an arrow that updates in five-degree steps is steady enough to follow on an E Ink panel, or whether it judders. Whether the line about how thick the scatter was reads as useful or as noise when you are standing in a field wondering why you came.
It is public at this stage precisely because of that. If you take it out, what happened is the most useful thing you could send back.
- Scatters a thousand points evenly over the disc around you, estimates how thickly they fell across it, and sends you to the thickest patch — or the thinnest, if you would rather have the empty part.
- Draws an arrow, a distance and the coordinate, and nothing else. There is no map in this
app and there will not be one. A greyscale basemap that pans and zooms is the worst thing
this panel can be asked to draw, and the phone already has an offline map that does it
properly. The coordinate is handed to that one with a
geo:intent. - Says how thick the patch actually was, against what an even scatter would have given. Most runs come back somewhere ordinary, and it says so rather than dressing the answer up.
- Says how good the fix under it is, and how old, because a walk scattered around where the phone was twenty minutes ago is a different walk.
- Walks of ¼ mile to 5 miles, or 500 m to 10 km. Two words of settings otherwise.
The idea is not new and it did not come from an app. On 14 April 1921 the Paris Dadaists held the first of what they called excursions et visites — a guided tour of the church of Saint-Julien-le-Pauvre, chosen for being nothing in particular. Their flyer says the point of it plainly, in Andrew Green's translation:
the Dadaists, wishing to correct the ineptitude of unreliable guides and cicerones, have decided to undertake a series of visits to selected places, especially those that don't really have a reason to exist.
Their tour guide read entries chosen at random from the Larousse. Three years later Breton, Aragon, Morise and Vitrac took a train to Blois — a town picked at random off a map — and walked out into the countryside for several days, which is this app with a pin instead of a generator. The Situationists turned the same impulse into a procedure and called it the dérive; La Monte Young wrote the whole of Composition 1960 #10 as "Draw a straight line and follow it", and Stanley Brouwn's entire 1962 piece reads "a walk from a to b".
So this is a score generator, not an instrument. It writes you a one-line instruction of that same shape — go to 35.88498, −82.81982 — and the walk is the work.
Which is also why it does not dress up the answer. The 1921 excursion was a washout: rain, fifty-odd people, the promised band never arrived, the auction was cancelled and the onlookers drifted off. Everyone involved thought it a failure and it is now considered the first work of walking art. A hundred years of this practice already know the outing is usually a let-down, and say so. An app that promised you otherwise would be the only one in the line that did.
The numbers here come from the phone's own generator. They are not quantum, nothing is fetched from a server, and no claim is made that an intention held while pressing the button reaches the arithmetic — that last part being the one thing this app declines to inherit from Randonautica, which is otherwise its direct ancestor.
Declining it costs nothing, because the tradition already has a better account of the same experience. Breton called it hasard objectif: the coincidence that arrives looking preordained, "the true precipitate of desire". That is a claim about the person who finds the meaning, not about the machine that picked the number — and it survives knowing exactly how the number was picked, which the other claim does not.
One: ACCESS_FINE_LOCATION, to know where to scatter the points around. There is no
INTERNET permission, so nothing can leave the phone even by accident, and nothing about
where you are or where you were sent is written down. The details, and how to check them
rather than take them on trust, are in PRIVACY.md.
Three earlier projects implement the same idea, and this one owes them the method:
- OpenRando (MIT) — the clearest short statement of the algorithm.
- pyrandonaut (GPL-3.0-or-later) — the same in Python, against scipy's own kernel density estimate.
- LibreRandonaut (GPL-3.0-only) — the Android one, and where the idea of testing an attractor rather than just reporting it comes from.
None of their code is in here. Three things are done differently: the scatter and the density estimate happen in metres rather than degrees, because a kernel that is round in degrees is an ellipse on the ground; the Gaussian kernel is separated per axis, which is one exponential per cell per axis instead of one per pair; and the answer carries a measure of how concentrated it really was.
Issues and pull requests are welcome. The things that would help most, roughly in order:
- Reports from actual walks — where, how far, whether the arrow was followable, whether the fix was any good under cover.
- Other devices. This is built for one phone's screen, 4.3" and monochrome. It should run on any Android 12 or later, but it has never been seen on anything else.
- The honest-number problem.
concentrationcompares the local density to the disc's average using the tenth nearest neighbour. That is a defensible choice rather than the only one, and if you know a better statistic for "is this cluster actually anything", say so.
Two things this app will not take, so nobody wastes an afternoon on them: a map drawn inside the app, and any claim that intention influences the numbers. The README says why for both.
GNU General Public License v3.0 only. Copyright wander wildwood.
Icons are from Material Symbols, Apache 2.0.