How we design a fair spot the difference puzzle

The definition we work to

A difference is fair when a player who looks directly at it will see it. That sounds obvious, and it rules out most of what makes puzzles frustrating elsewhere: single-pixel edits, changes hidden under a watermark, a shade of blue two steps along from another shade of blue, or a change so small that a thumb covers it.

It also means fairness is measurable. If the requirement is that a change is visible when looked at, we can express it as minimum size, minimum contrast and minimum distance from other changes, then check every scene against those numbers before it ever reaches a player.

Zone size: the thumb rule

Every changed area has to be comfortably tappable on a phone held in one hand. In practice that means a change occupies a meaningful share of the frame, not a stray detail, and the tap target around it is larger than the visible change so a near-miss still counts as a find.

This constraint shapes what kinds of change are even possible. It is why our scenes change whole objects rather than object details: a whole mushroom can be removed at thumb scale, a mushroom's spot cannot.

Spacing: differences must not cluster

Two changes sitting on top of each other read as one change, which means a player can legitimately find both and only be credited for one. So scenes enforce a minimum separation between changed areas, expressed as a percentage of the frame. In our photo studio that separation is an adjustable setting, because the right value differs between a wide landscape and a close-up interior.

Spacing has a pleasant side effect: it spreads differences across the whole scene, which rewards a systematic scan and punishes camping in one corner.

Contrast: no invisible colour swaps

A recoloured object is only a fair difference if the new colour is clearly a different colour. Our drawn scenes pick replacement colours from separated points in the palette rather than neighbouring shades, and our photo puzzles measure the actual pixel difference between the two images and discard anything below a contrast floor.

This is the check we tightened most recently. Faint tint shifts technically registered as differences and consistently produced the "that was it?" reaction, so the threshold went up and the edits themselves got bolder: objects added and removed instead of gently tinted.

Difficulty comes from structure, not from cruelty

Once fairness is fixed, difficulty has to come from somewhere else. We use four levers: how many differences there are, how densely the scene is populated, how much objects repeat, and which change types are used. A sparse desert scene with two moved objects is harder than a crowded market with five removed ones, and neither one breaks a fairness rule.

That is why our expert pack is the emptiest one on the site. Restraint makes changes quiet without making them invisible.

The gates a scene has to pass

Before a puzzle is published it runs through automated checks. Any failure blocks publishing rather than producing a warning.

• The scene contains the expected number of measurable differences.
• Every changed zone is at least thumb-sized.
• Zones are far enough apart to be tapped independently.
• Every zone sits fully inside the visible frame.
• Edits stayed local, so the whole image did not shift or re-render.
• Contrast at each zone is above the visibility floor.

Determinism, and why it matters to you

Our drawn puzzles are generated from a seed, which means the same puzzle URL always rebuilds the exact same scene, on any device, forever. That is what makes leaderboard times comparable and shared links trustworthy: when someone tells you they solved a puzzle in nineteen seconds, they solved the puzzle you are looking at.

It also makes review practical. A failing scene can be reproduced exactly, inspected, repaired and re-checked rather than vaguely regenerated and hoped over.

What we reject

Scenes get thrown away constantly, and the most common reasons are dull rather than dramatic: the differences ended up bunched in one corner, the edit shifted the whole image, or the change count came out too low after the contrast floor removed the weak ones. Rejecting those is cheap. Publishing them is not, because an unfair puzzle costs a player's trust and trust is slow to rebuild.