It hides, and comes back. The question it is asking is whether you know when — a sound before the return means something a sound after it does not.
— what the microphone hears · ▎your sounds · ▎its return · ▎seeking · ▎hello
Coalitions: —
Each row is a player; each column is somebody they can hear. The number is how often that player's sounds arrived before the row player's return, with the count it is out of — because a rate of 1 from one round is not a finding. A row that tracks nobody is being left out; two players tracking each other are a coalition.
Peekaboo with no words in it. Something hides, and comes back, and the whole question in the game is whether you know when — because a sound made before the return means something a sound made after it cannot. Reacting proves you were listening; arriving early can only come from a model of when the return was due. That is the rung where responsive starts to look like intelligent, and it is the smallest game that reaches it.
It will not start a game just because you are there. It says hello once and waits. If you do not answer, that is a no — it stops asking and leaves you alone for half a minute. Being detected is not consenting.
You can take the other role. Chatter at it for a while and then go quiet: it will notice you have hidden and come looking. Coming back when it calls is the reciprocal of anticipating its return, and the stronger of the two, because hiding has to be chosen.
Everyone in the room has a voice that means something. Each player keeps its own register and timbre — who it is, fixed. The third word beside it is its stance: what its sound is doing right now, which changes as it goes. Tired means its director keeps backing off; insistent means it is calling after somebody who hid; leave-me-alone means it was declined and is staying out of the way. The sound is a readout of a state, not decoration on top of one. The registers are kept distinct so you can hear who is who — and so the measurements can tell them apart, because an anticipation rate averaged over several partners is an average of unrelated things.
What to distrust. A partner vocalising at random still scores about 0.22 on the anticipation measure, so treat any rate near that as noise rather than prediction. And the rate shown per player is still pooled over everyone it can hear, so it inflates as the room fills — more sounds means more of them landing inside an anticipation window by luck. The per-partner records are being kept and are the honest version; this list does not show them yet. With three others in the room, treat anything under about 0.5 as noise.