DEV Community

Sam Ellis
Sam Ellis

Posted on

What makes a photo scavenger hunt feel fair rather than random

Every photo scavenger hunt has a moment where the player looks up from the screen, turns around slowly, and says "that's not right". The app is working. The GPS is right. The photo is on their screen. And yet the moment feels arbitrary, and when it finally clicks they are not sure whether they earned it or got lucky.

We have built a lot of hunts now, and we have watched this moment happen more times than we can count. It is the single thing that separates a hunt that feels generous from one that feels like a guessing game. And almost none of it is about GPS accuracy or image matching. It is about how the clue is written.

The rule we keep coming back to

A clue should describe something a person could notice on purpose, not something a camera would only catch by accident.

That sounds obvious. It is not. Most of the early clues we wrote were descriptions of the photo: "a red door", "a blue bicycle against a brick wall", "a cat sitting on a bench". Those are descriptions of pixels. They are only findable if you happen to point the camera in the right direction and zoom in on the right corner of the frame.

The clues that work describe the place: "the red door is next to the shop that sells umbrellas", "look for the bicycle with the basket full of flowers". Now you are not hunting for an object, you are hunting for a relationship between objects. Your eyes have to sweep the scene and reason about it.

The difference is that the second version can be solved by someone who is slightly to the left of the right answer. The first version cannot.

Clue writing is not copywriting

We used to think of this as a copy problem, so we tried to be clever. Clever clues are a trap. They are fun to write and miserable to play.

"What does a door with no building have?" is a clever clue. It is also a clue that requires the player to think about ontology before they can look at anything. On a good day it produces a satisfying aha. On a bad day it produces a player staring at a phone with the app closed, wondering whether it is worth the battery.

Plain, concrete, slightly boring clues work better:

  • "the green bench with a missing slat, on the corner with the bakery"
  • "the yellow door two numbers down from the one with the bicycle sign"
  • "the fountain with the statue of a man holding a fish"

Every word there does work. There is no riddle to solve. There is only a place to find.

Length is not the enemy, ambiguity is

There is a persistent belief that clues should be short. Short clues are good when the place is distinctive and the match is unambiguous. They are bad when the location has several similar features and only one of them is the right one.

The rule we use instead: make it long enough to be unambiguous, and no longer. Every extra sentence you write is a sentence you have to keep true. Every feature you list is another feature that the real world has to actually have. If you write "the red door next to the bakery on a street with a church tower in the background", you have now committed to a church tower. If the tower is on the other side of the square, you have made the hunt worse, not better, because the player will look for the tower, fail to find it, and start doubting everything else.

If two clues feel equally good, take the shorter one. If one clue feels ambiguous, add detail about the surroundings, never about the object itself. More object detail narrows the pixel match. More surroundings detail narrows the search.

Write the clue from the photograph, not the map

Another thing we got wrong early: writing clues from the map, from the list of coordinates, from the idea of the place. A map has no smell and no sound.

Now we write the clue by looking at the photograph that the player will be matching. We open the image on the actual phone-sized crop, we look at the edges, we ask what a person standing there would notice that is outside the frame. The bakery sign is usually just outside. The wet pavement reflecting the window is usually in the corner. The person with the dog usually passes through.

Those edge details are the best clue material there is, because they are exactly the things the photo matching will not reward you for. Matching rewards you for finding the red door. Noticing the bakery next to it is the actual game.

The fairness test

Before we ship a hunt, we run the clue past three questions:

  1. Could someone who has never seen the photo describe this location from the clue alone? If not, the clue is describing a picture, not a place.
  2. Is every feature in the clue true of the real location? We check each one on the ground, in the conditions the player will be in.
  3. If the player is standing ten metres away, does the clue tell them which way to go? If they would have to walk around the block to work it out, that is exploration, not hunting, and it is not what people signed up for.

If a clue fails any of the three, we rewrite it. Usually the fix is to swap an object detail for a surroundings detail.

What this is for

None of this is theoretical. It is the difference between a hunt where people finish and immediately want another one, and a hunt where people finish and shrug.

We have written up how the rest of it works - the GPS accuracy, the camera matching tolerance, the offline behaviour, all of it - in the Ping Hunt docs. If you want to try one before you build one, there are hunts already running in Copenhagen and Berlin.

But if you take one thing from this: describe the place, not the photograph. Everything else follows from that.

Top comments (0)