A Snap Distance Is Two Numbers
Joining star to star failed in four pairs of ten: the line grabbed a star it was only passing. The fix was two snap radii — wide at intent, narrow in transit.
HappyBirthTown is a pixel town that friends fill with letters, photos, and decorations and hand to the birthday person as a gift. One of the things a friend can leave is a constellation. There is a telescope in the town; step up to it and you get a round patch of night sky with stars already in it, you draw something, and what you drew hangs in a strip beside everyone else's drawings.
That device has now had four designs. It was a text box. Then it was joining stars with lines. Then it was a free curve drawn anywhere on the sky. On 21 September it went back to joining stars — and going back is the part worth writing down, because the argument for the free curve had not been wrong.
The free curve sold the wrong thing
The case for free drawing was expressive range. If you can only connect the stars that are there, the set of possible pictures is whatever the star layout permits. A free curve can be anything. That is a true observation, and it is the kind of reasoning that wins a design discussion.
What it missed was what the device was for. The sentence this feature sells is everyone looked at the same sky and drew it differently. That sentence needs the sky. It needs the stars to be fixed, shared, and the thing being interpreted. With a free curve the stars turn into wallpaper, the sky becomes a background, and what is left is a drawing app with a dark theme. Nobody is interpreting a shared sky; they are decorating a canvas.
Constellations are, historically, the act of joining given points. People who did it thousands of years ago had the same stars and produced different animals. The constraint is not a limitation of the tool. It is the tool's name. That was the reason to go back: not that the free curve was less expressive — it was more — but that it stopped being a constellation.
There is a quieter benefit too. A blank canvas is intimidating on a phone; most people draw nothing. Points already placed invite a first stroke. The constraint lowers the cost of starting.
Going back, we got it wrong once more
The reimplementation kept the earlier rule: a stroke that starts on a star and passes near other stars picks them up in order, so one gesture becomes one bent line. Stars it joins light up. The path of the finger is not kept; only which stars were joined.
Testing this, straight lines kept bending. You would aim from one star to another across a gap, and somewhere in between the line would hook a star you were only passing and turn. In four pairs out of ten the drawn line was not the one intended. The grab distance was a single number, and it was tuned generously so that a thumb could actually land on a small star.
That generosity was correct at the ends and wrong in the middle. When you press down or lift off, you mean the star you are near — a wide radius helps you, because aiming a fingertip at a few pixels is hard. When you are dragging through the middle of a stroke, a star near the path is incidental — a wide radius hurts you, because it attaches things you did not choose.
So the radius became two. Press and release are intent, and get a generous distance. Dragging through is transit, and gets a narrow one. Tap-tap-tap to join stars one at a time, or sweep once through several; both still work, and a stroke left open waits until you tap empty sky to close it.
Where a magnet should be strong
The general rule is small and I have not seen it written down. Every snapping tool — alignment guides, grid snap, node connection, magnetic selection — has a snap distance, and it is usually one value. It should be two, because the same gesture contains two different kinds of moment. Where the hand is committing, be generous: a tight radius there makes people miss what they aimed at. Where the hand is only passing, be strict: a loose radius there makes the tool grab things they never wanted. One number cannot serve both; it will be too strict at the ends or too loose in the middle.
And the larger rule, the one that sent us back the third time: expressive range is not always a good. There are objects whose constraint is their identity, and widening what they can do makes them into something else with a smaller name. A tool that lets you draw anything on the sky is not a better constellation tool. It is not a constellation tool at all.