Game selection

How many

Four games: enough for diverse games, but not too expensive.

Popularity

A game should be somewhat established and be known to some extent, but should be too small for labs to have properly trained on its strategy.

Difficulty

The game has to be really hard.

Luck

Games with luck must use seeds with 50%–50% odds.

A cheap oracle

There has to be an existing engine that is near-perfect and computationally reasonable to run.

Chosen

The following games meet all criteria:

Dropped Xiangqi: measured ideal-move rates (~80–90%) were too easy.

Look