Each trial shuffles a full deck, deals 8 cards to each player, then deals alternately until the deck is empty — so you finish with 30 of the 60 cards. Each trial then measures, per colour, how many cards you got and what they are worth. Both are reported by rank: the five values are sorted, so group 1 is whichever colour came out worst that trial and group 5 whichever came out best. The groups are ranks, not colours — group 1 is red in one trial, blue in the next.
Each game is solitaire — no opponent yet. Take 8 cards, then every turn play one card to a venture and draw one, until the draw pile empties. A computer discards only when it cannot legally play anything, and never draws from the discards.
The same computers with an opponent across the table, and perfect information on both sides. Every deal is played twice with the seats swapped, so nothing here can be first-player advantage. Solitaire cannot express half of Venture — with nobody to receive them, discards are free — so this is where a computer that prices both sides means anything.
Every computer against every other, all on the same footing. A duel settles one pairing; this settles the whole field at once — who is actually strongest, who only looks strong because of who they were measured against, and which pairings are still too close to call. Each pairing plays the same deals with the seats swapped, exactly as a duel does.
Write a computer in a small language shaped like Python. A for
loop holds a set of cards and every line inside narrows it — so
if change in potential min: leaves the one card that costs the least, and
play plays it. Saved computers join the list in the Computers section and are
measured by the same runner as the built-in ones.