Building a daily game that lasts: 393 objects, 855 rounds
A daily puzzle has to be different every day and the same difficulty every day. Those two requirements pull against each other, and most of the design work behind this game is the compromise between them.
The numbers so far
171 sets have been published: 120 for the daily game, 33 for Pop Culture and 18 for Geography. Five rounds each, which comes to 855 rounds. Behind them sit 393 measured objects, of which 196 belong to the daily pool, 152 to Pop Culture and 118 to Geography, with some appearing in more than one.
The arithmetic that has to work is simple to state and awkward to satisfy: enough variety that no two days feel alike, enough consistency that a score on one day means the same as a score on another.
Not every object can do every job
A round has two objects with different jobs. One is the reference, held at true scale, and the other is the target, which you resize. Those jobs have different requirements.
A reference has to be something the player can size from memory. It needs a familiar, unambiguous silhouette and a size most people have an opinion about. A sheet of A4, a person, a bus, a wine bottle. Of the 393 objects, 235 are marked as usable in that role.
A target can be anything, including things nobody has any intuition for, because the whole point is that you are guessing. 276 objects are marked usable as targets. The overlap is large, and the objects that can only be one or the other are the interesting cases: a garden ant makes a terrible reference and a fine target.
The constraint that does the most work: the two objects in a round have to be within a workable ratio of each other. Too close and the round is a coin flip. Too far and the target shrinks to nothing on screen and the guess becomes a shrug.
Where the ratio band sits
Rounds cluster between about one and five. That is not an arbitrary choice: it is where human estimation is still doing something. A whale shark at 12.1 m against a T. rex at 12 m is 1.01, which is about as close as the set gets and makes for a round that almost everyone gets right for the wrong reason. The Eiffel Tower at 330 m against the Millennium Falcon at 34.75 m is 9.50, near the top of what stays playable.
Beyond about twenty the round stops testing judgement. The player knows the answer is very small, drags the handle until the shape is very small, and scores reasonably without having estimated anything. A game full of those is easy and boring in the same move.
Why pairings are checked rather than random
With 393 objects there are tens of thousands of possible pairs, and most of them are nonsense. Random pairing would produce rounds that are unplayable, rounds that are trivial, and rounds where the two objects have nothing to do with each other and the comparison teaches nothing.
So the pairings are drawn from combinations that have been checked to work: the ratio is sane, both objects belong to the same game, and the two are things somebody might plausibly want to compare. 525 distinct pairings have actually been served across the 855 rounds, which means the good ones recur.
Recurrence is deliberate. A wine bottle against a sheet of A4 has come up eleven times, and A4 against an apple twenty-seven. Those are the calibration rounds: familiar objects that let a player check their eye against something they genuinely know, which makes the unfamiliar rounds in the same set fairer.
Scoring on one axis
A round is scored on the relative error between the size you committed to and the true one. That is a linear measure, which matters: scoring on area would square the error and make a near miss feel like a catastrophe.
It also means only one dimension is ever judged. The object carries one figure and one axis, and your drag is compared against that. The alternative, scoring on both dimensions, would require every object to have two authoritative figures, and for most of them the second figure does not exist in any published source.
Everyone gets the same set
A set belongs to its date, not to the player. That is what makes a shared score comparable, and it is also why an archived date replayed today is identical to the day it went out.
The cost is that a bad set stays bad forever. A round where both objects are obscure, or where the ratio turns out to sit in a dead zone, cannot be quietly swapped out afterwards without breaking the promise that the archive is what was published. That is a reasonable price and it puts the pressure where it belongs, on the checking that happens before a set goes out.
Three games rather than one
Splitting the pool into three was the decision that made the whole thing sustainable. The daily game holds real objects at human and architectural scale. Pop Culture holds characters and craft, where the figures are stated rather than measured. Geography holds the 118 outlines built from mapping data.
Keeping them separate does two things. It stops a round from putting a measured animal against an invented character, which is a comparison with one firm number and one soft one. And it lets each game have its own difficulty: Geography is harder because almost nobody has a calibrated sense of a country's span, and it would drag the daily game's scores down if mixed in.
What running out looks like
At five rounds a day the daily pool of 196 objects and its checked pairings do not last forever, and pretending otherwise would be silly. The honest answer is that the catalogue grows: 393 objects now, 585 outlines drawn, and 275 of the objects carry a stated fact about their size beyond the raw figure.
The other answer is that repetition is not the enemy people assume. A pairing that returns after four months is a genuinely different round, because you have forgotten your previous answer and your eye has changed. What kills a daily game is not repetition. It is rounds that were never worth playing the first time.