The new games have been added without being consistent with what is already there:

1. a version of stripped down poker is created in a new function in tests/games.py but that new code still refers to Myerson not Reiley; moreover it ignores the existing create_stripped_down_poker_efg. For reconciling create_one_card_poker_lacking_outcome_efg() with create_stripped_down_poker_efg it seems that you might want to add a third variant. In that case you might want to go from a boolean flag in to some kind of "type" argument, which if None would give the standard one.

2. the new games added with non-terminal or missing outcomes etc. do not have variants with all terminal outcomes; note that the existing functions from Master for the relevant tests with and without non-terminal outcomes (create_stripped_down_poker_efg, create_matching_pennies_efg, and create_kuhn_poker_efg) all have an argument that allows to create a standard version with only (and all) terminal outcomes and a non-standard one. The tests then include a case for the standard version, and a xfail variant created via the same function in game with an appropriate argument.

Here's an example:

        (
                games.create_2x2_zero_sum_efg(),
                [[["1/2", "1/2"]], [["1/2", "1/2"]]]
        ),
        pytest.param(
            games.create_2x2_zero_sum_efg(missing_term_outcome=True),
            [[["1/2", "1/2"]], [["1/2", "1/2"]]],
            marks=pytest.mark.xfail(reason="Problem with missing terminal outcomes")
        ),

Here's another example:

        (games.create_matching_pennies_efg(),
         [[["1/2", "1/2"]], [["1/2", "1/2"]]]),
        pytest.param(
            games.create_matching_pennies_efg(with_neutral_outcome=True),
            [[["1/2", "1/2"]], [["1/2", "1/2"]]],
            marks=pytest.mark.xfail(reason="Problem with missing outcomes")
        ),

There's also examples for LP and LCP solve in mixed behaviors for stipped-down poker and Kuhn poker.

In these examples the two versions, with and without non-terminal outcomes, may or may not give the same answer (note e.g. that currently kunh_poker_lcp_first_mixed_strategy_prof() is used in the xfailing and standard tests. I believe you have an example (maybe one of these games even) where making such a variation changed the output of the LP or LCP solver, which is not impossible (namely that the solver finds a different solution even though the mathematical program getting solved is equivelent in terms of the set of solutions).

I presume you put the "expected outputs" for the xfailing variants based on a standard variant anyhow? Or did you already have working code for these xfailing variants? Either way, please add tests for the standard variants as well.

Note that we also include for stripped-down poker and Kuhn the terminal-only and non-terminal examples for solving in mixed strategies, here from test_lcp_strategy_rational:

        (games.create_stripped_down_poker_efg(), [[["1/3", "2/3", 0, 0], ["2/3", "1/3"]]], None),
        (
                games.create_stripped_down_poker_efg(nonterm_outcomes=True),
                [[["1/3", "2/3", 0, 0], ["2/3", "1/3"]]],
                None
        ),
        (games.create_kuhn_poker_efg(), [games.kuhn_poker_lcp_first_mixed_strategy_prof()], 1),
        (
                games.create_kuhn_poker_efg(nonterm_outcomes=True),
                [games.kuhn_poker_lcp_first_mixed_strategy_prof()],
                1
        ),

Better, but before I carefully check all the cases can you please use everywhere in the expected profiles:

• 1 and 0 instead of 1.0 and 0.0
• fractions as strings instead of floating point long fractions, e.g. "1/3" instead of 0.333333 and "2/3" instead of 0.66666667.

The tests already do a floating point comparison, and we should encode the exact equilibria as the target of what we would like to find, even for enumpoly which only works in float. Note all the other tests are done using integers without quotes or fractions in quotes. Please stay consistent with that.