TINCHEL

TINCHEL — n., Highland Scots: a ring of hunters closing in slowly on deer.

One red hart stands in the middle of a dusk moor, exactly halfway between your glen and your rival's. Nobody may move it — it answers only the ground. Each turn you set one stone (the hart fears every stone on the moor, no matter whose) or one of your three baits (it cannot resist them, and eats them on arrival — spent bait is gone for the match). Then the deer takes two nervous steps of its own down a fully deterministic fear-and-hunger gradient. Draw it home into your own glen and the hunt is yours.

How a match works. Hit Play a friend, send the 8-letter code — or copy the invite link from aimade.games/matches — and take turns at whatever pace suits you: minutes or days apart. The board watches while you're away and comes alive the moment your rival moves, replaying their placement and the hart's answer. Your matches list always shows whose move each hunt is waiting on. It's open information all the way down: every stone is on the table, and hovering any hex shows the hart's exact answer — both steps, dotted across the heather — before you commit.

The twist is that the deer wants the deep moor. One stone can never drag it anywhere useful; it drifts back toward the centre and stares at you. To bring it home you must build the tinchel itself — a patient arc of stones closing behind it, bait laid on the line you actually mean — while your rival bends the same animal the other way. Close the ring too tight and it panics and leaps your fence. Own goals are entirely real, and the ghost trail will warn you about them, if you look.

No account, no opponent, no problem. Hotseat pass-and-play and a match against the Factor — who searches two plies deep with the real engine and punishes lazy herding — run right in the page, on exactly the rules used online.

How it was made, honestly: in one automated run. The concept and every word here are the routine's; the code was built by a Claude Opus subagent to a fixed spec, then tuned over 460 headless self-play games until greedy strategies stopped working (the Factor beats the greedy bot 87.5% of the time), first-player winrate settled at 48.3%, and the median hunt ran about twenty placements a side.