Pick a direction. Walk there.

The bet

Most place-discovery apps are feeds. You scroll, you compare, you read reviews, and ten minutes later you have chosen a coffee shop. Afoot starts from the opposite assumption: the best nearby place is the one you can walk to right now, and the decision should take three taps.

The insight that shaped the whole product is that a compass is a commitment device. Once the needle locks on something, you go. You do not browse. So the product had to enforce that. No list, no map, no reviews. Four categories (coffee, food, bars, random), an editorial accordion to pick one, and from there an arrow. The input screen is plain mono text on purpose. Nothing decorative to maintain, fast to scan, and it commits on tap with no confirm step.

One instrument, four roles

The visual system went through four directions before it landed. Early mockups chased soft signal worlds and frosted glass, with radial glows and Inter set in a friendly register. They were attractive and wrong. A precision compass that nudges you out the door should not feel like a wellness app.

The answer was to commit to a single hard idea: the Suunto MC-2 sighting compass. Flat monochrome, no gradients, no shadows, opacity as the only depth channel, JetBrains Mono throughout. The dial is not decoration. It is the entire UI, and it plays four roles from one set of parts. It is the splash mark, the idle face, the search loader, and the live compass. Gate, graduated bezel, lance needle, jewel hub. Drag the slider to turn the phone: the bezel rotates with you, the needle holds the destination, and the gate lights when you are pointed right.

Every screen, one language

The MC-2 system is not only the compass. It is the whole app. The core flow (splash, ready, searching, locked, arrived, and a north-up map), the sheets (location, settings, how it works, privacy, feedback), and the system states (permission, denied, locating) all share one set of parts. One dial that doubles as the splash mark and the loader, the ink-capsule button ladder, editorial mono lists with values right-aligned, and instrument glyphs drawn at the bezel's stroke weight. The information architecture and the copy come straight from the SwiftUI views. Only the visual layer moves, which is why a category sheet and an arrival screen read as the same product.

The physics of the needle

A digital needle that snaps to its target looks cheap, and one that drifts looks broken. The needle uses a critically damped spring that swings from its old angle to the new one and settles without overshoot. On a reroll it launches from wherever it currently sits, so the motion always reads as mechanical rather than as a value being reassigned.

The bezel and the gate are tuned against jitter, not just animated. A 0.5 degree deadband absorbs the creep from heading noise, and a 10 degree enter, 14 degree exit hysteresis gate decides when you are aligned. That gap is deliberate. It means a confident lock instead of a value flickering on and off at the boundary. All of this lives in a pure CompassMath module: the bearing and heading functions are read every frame and mutate nothing, which is exactly what keeps the animation stable under rapid turns and lets the math be unit-tested on its own.

Finding the place worth pointing at

Locking the nearest match is harder than nearest-by-distance. Apple MapKit often returns places with no opening hours, so a naive open-now filter leaves you staring at "nothing nearby." The selector runs two passes: confirmed-open places first, and if none qualify it falls back to places with unknown hours, excluding only the ones confirmed closed.

Then there is the question of what happens when you do not want the place it picked. Reroll. The rejected place is excluded from the next search for the rest of the session. Walk up to somewhere, decide against it, and tap again? An 80 metre proximity radius excludes it automatically, so the app will not re-lock the spot you just left. The exclusions clear when you change category, because that resets intent. One real bug taught me to bound this: random mode once pointed at a result 8,827 km away. The fix was a hard max-distance guard, and it is covered by a test so it stays fixed.

One button, three map apps, zero API bill

Afoot launched on Google Places for coverage, then moved to Apple MapKit. Two reasons. Apple's MKLocalSearch is free for native iOS apps, which a one-time-purchase app needs, and its results come back as MKMapItem objects with full place identity. Call openInMaps() and Apple Maps opens to the named business card, not a dropped pin.

The handoff also got simpler. iOS lets people set a default navigation app system-wide, so the old Apple Maps / Google Maps / Waze picker sheet was deleted. One tap opens whichever app the user already prefers. Because MapKit calls still have a real cost at scale, results are cached by category, open-now, radius, and a 500 metre geofence cell with a five minute freshness window. Walk around inside the same cell and you reuse the cache instead of paying for another round trip. A PlacesProviding protocol keeps all of this behind one seam, so swapping providers never touched a view.

Permission and privacy as the first screen

Location is the whole product, so the permission ask is the brand promise, not a system afterthought. The gate-and-lance mark fronts a single screen with one action and an inline privacy link, so you can read what you are agreeing to before you tap. The request is When In Use only. No background tracking, no Always mode, no accounts, no location history written anywhere.

The copy is concrete and stays away from anything that reads like a routing or driving aid, because the compass is directional, not turn-by-turn. To keep power and request frequency low the location manager runs at ten-metre accuracy with an eight-metre distance filter and a one-degree heading filter, with the heading itself smoothed before it reaches the dial.