Getting people tocome back to alearning app.

The first Everywhen, shipped in mid-July, was a place-anchored learning app. You opened it, a place lit up on the map, and you got a two-minute story about it, ideally listened to on a walk. One place, one takeaway, one question. The premise still drives everything: memory is exceptionally good at hanging facts on locations (the method of loci), so every fact in Everywhen is pinned to the spot where it happened.

The loop was tidy and the content was good. What it lacked was pull. Explore was built on “pull, don’t push”: nothing appeared unless you asked. A daily notification named a random place. Tomorrow looked exactly like today, so there was no particular reason to come back to it, and in my own use I mostly didn’t.

Explore, the original front door. It is still in the app, but it is no longer the reason to open it.

The day after 1.0 shipped, I wrote a new first line for the spec: “The retention engine: one shared daily round of four place-guessing puzzles that ends in learning.” It was built and merged that same day, and Today became the first tab.

Guessing before reading also does real work. The guess is a retrieval attempt, a moment of “I think I know this”, and the story then lands on a question you already care about. The game feeds the Passport, the atlas of every place you have collected. It is not a parallel universe.

A day runs easy to hard, and each place is worth more than the last: 100, 200, 300 and 400 points. Knowing the obscure place pays the most, which is the right incentive for a learning game, and the maximum comes out a clean 1,000. A fifth bonus place is optional and never counts toward the streak or the shared score, so the core day stays comparable for everyone.

The first when-input was one slider from 4.5 billion years ago to today, log-scaled so the sixties and the age of the dinosaurs felt equally reachable. It looked elegant and was physically impossible to use: the full-points window was about half a pixel of track at every era. It became two steps, a range and then a year inside it, shipped the same day the daily did. Try both.

When was Pompeii buried?

Pompeii, Italy · full points within ±97 years · try both inputs

First design · one slider

67,000 years ago

4.5 billion years agotoday

Off by 65,135 years.

1.4 px wide: the whole full-points window on this track.

Shipped · range, then year

Before historyAntiquityMedievalEarly ModernIndustrialModern

Pick a range

 

Six ranges, oldest first. The slider wakes up in the one you pick.

Same answer, same tolerance, same rules as the app. Only the input changed.

The original where-score had, in my own note, “a cheap top and a cruel middle”. Anything within about 64 km showed a perfect score, while a guess 2,000 km along the Amazon earned a quarter and 5,000 km earned almost nothing. In late October I reshaped the curve so it falls fast near the answer, keeping a top score meaningful, and flattens across the middle, so the right continent keeps real credit.

The reveal says one line, right away, in the voice of a delighted friend across the table. It is never scoreboard-speak. When you nail it, it says so. When you don’t, the puzzle takes the blame: the place was hiding, and being wrong is how these stick. The hardest place of the day stays forgiving-funny, because nobody is meant to get that one cold.

“You felt the era. The map played tricks.”

Every reveal ends in a real place and what happened there.

A list of 930 places is a lot of names to tell apart. In late October every place got its own small line drawing: a bridge, a crater, a stepwell, a chimp. They sit beside each place in the Library and its collections, drawn in the colour of the place’s realm, so a shelf of places has a shape before you read a word. Find one, or filter by realm.

A share is the one part of Everywhen that leaves the app and lands in someone else’s messages. So each card is designed for whoever receives it, which means two people at once: a friend who plays, and a stranger who doesn’t. Every card is the same 4:5 object on iPhone, Android and the web. Tap through them, then flip the thread to dark.

“See your guesses” lets you step back through the day’s four places: your pin and the answer on the map, your scores, then the place’s story. The whole design is one sheet over a map, and almost all of its quality is in how it moves. Every rule below is in the shipped app, with the same constants on iOS, Android and the web.

You

‹Place 4 of 4×

The temple you enter by going down

Rani ki Vav, Patan, Gujarat, India

You felt the era. The map played tricks.

Where146/200522 km away

When175/200around AD 1050

Read story

At Rani ki Vav, the grand architectural approach leads down into the ground. Built in the eleventh century as a memorial, the stepwell made access to water into a carefully designed descent, with stairs, pavilions, and carved panels along the way.

Its seven levels were conceived as an inverted temple. Water was both a practical resource and something sacred, so the structure joined engineering to religious imagery. More than five hundred major sculptures and many smaller ones filled the walls around the descending visitor.

Flooding and changes to the nearby Saraswati River eventually buried the well in silt. That ended its original function but helped protect the carvings for centuries.

The recovered stepwell makes a daily necessity feel monumental. Instead of raising a temple above the plain, its builders created a richly worked space beneath it.

Photo: Bernard Gagnon, CC BY-SA 3.0

Try it

Drag it, flick it, push it past the top.

Release–px/s

Lands atHalfwhere it opens

SpringSettlesk 200 · c 30

Every release is projected to where it would coast to a stop, then settles on the nearest of three stops: Open, Half and Peek.

• Show the physicsThe three stops, live speed, and where a flick lands
• Break itIgnore the flickSnap to wherever you let go, the way most sheets do
• Break itA bouncy springWhat a wobble feels like on a sheet holding a story

Reset the sheet

1. Project the flick. A release is projected to where it would coast to a stop, using the same deceleration as scrolling (0.998 per millisecond). A firm flick from the bottom reaches the top; a lazy one doesn’t.
2. Three stops. Half shows the whole reveal panel and is where every place opens. Peek gives the map nearly the whole screen. Open is the story.
3. Settle, don’t bounce. A critically damped spring that starts at your finger’s speed. A wobble on a sheet holding a story reads as a toy. Reduce Motion gets a stiffer one.
4. Bend, don’t wall. Past either end the sheet follows at a shrinking fraction of your finger, so an edge feels soft instead of broken.
5. One surface. Pushing up moves the sheet to the top, then the same gesture scrolls the story. Pulling down scrolls the story back, then brings the sheet down.
6. The map holds still. It is framed once, for the half stop, and never re-centres as the sheet moves.
7. One haptic, at commitment. A light tick when the sheet commits to a new stop, never just for passing one.

One day of getting it right

1. Nov 1

   “See your guesses” arrives: a way back into the day’s four places from the scorecard.

2. Nov 5, 15:58

   First design: pull the panel up and the story rises over it as a separate page. It read as a second page stacked on the first, and I reverted it 26 minutes later.

3. 16:33

   A half sheet over the map instead. Push it up and the photo and story are right there below the scores.

4. 16:59

   The motion rules arrive: project the flick, settle on a critically damped spring, bend past the ends instead of hitting a wall.

5. 18:23

   A third, lower stop, so the map can have nearly the whole screen when you want to look at it.

6. 18:52

   The map stops moving when the sheet does. Every camera fix re-centred it a beat later, so the review map became a native map view with its margins split from its camera.

7. 21:07

   The rule I’d written down (“stop at the top, pause, then scroll”) was wrong in the hand. Reversed: the sheet and its story are one surface.

8. Nov 6

   Driving it by touch in the simulator found it didn’t move at all on the newest iOS, and that Read story was a dead tap. Both fixed before 1.9.2.

Everywhen is three separate clients: SwiftUI on iPhone and iPad, Kotlin and Compose on Android, and React on the web at everywhen.cc. The game rules are shared, not the interface: scoring and streak fixtures are generated from the Swift code and checked by the Android and web tests, so a day scores identically everywhere.

Everywhen has no analytics SDK, by choice, so I can’t show you a retention curve. What I can show is the reasoning, the craft, and a habit of playing my own game every day and fixing what it teaches me. If I added consented measurement, these are the questions I’d want answered first:

• How many people play on day 2 and day 7, and which round they stop at.
• Whether a healed streak keeps people longer than a broken one.
• How often a share turns into a new player’s first round.
• Whether people go back to stories in the Passport, or mostly return for the next game.