The scoring has a cliff on both sides
In The Fox in the Forest you score by how many of the 13 tricks you take. Taking three or fewer is "humble" and worth 6 points. Seven to nine is "victorious", also 6. Everything in between scores 1 to 3, and taking ten or more is "greedy": zero.
Because there are exactly 13 tricks, the two of you are tied together. If you stay humble, your opponent took at least ten and scores nothing. If you are victorious, they are stuck in the middle. A round is really a fight over who ends up in the bad band.
Points for the round, by tricks taken
Treasure (the 7s) adds a point per 7 on top of this.
Show the numbers
| Tricks | Points |
|---|---|
| 0 | 6 |
| 1 | 6 |
| 2 | 6 |
| 3 | 6 |
| 4 | 1 |
| 5 | 2 |
| 6 | 3 |
| 7 | 6 |
| 8 | 6 |
| 9 | 6 |
| 10 | 0 |
| 11 | 0 |
| 12 | 0 |
| 13 | 0 |
Trying to lose works, until your opponent notices
We started with blunt strategies. One bot always tries to win the trick in front of it. Another always tries to lose. A third follows a simple plan: estimate how strong the hand is, then aim for humble or victorious. The strongest bots search: before each card they imagine 24 ways the hidden cards could be dealt, play every option to the end of the round, and pick the card that scores best on average.
The first result looks like a strategy: always trying to lose beats always trying to win 88% of the time. But that is a bit like beating someone at chess who only moves forward. The trick-grabbing bot takes every trick you hand it, so losing is easy. The searching bots, which actually think, win around eight in ten games against everything else.
Win rate against the other six bots
Games to 21 points, each pairing played hundreds of times with alternating dealer.
Show the numbers
| Bot | Average win rate | Games played |
|---|---|---|
| Searcher, starts humble | 81% | 3,600 |
| Searcher, decides per hand | 81% | 3,600 |
| Searcher, starts victorious | 79% | 3,600 |
| Rule-following planner | 37% | 1,800 |
| Always tries to lose | 34% | 1,800 |
| Always tries to win | 21% | 1,800 |
| Random legal cards | 19% | 1,800 |
So the real question is: can you still stay humble when the other player fights back? We sent a stubborn loser (a bot that dumps every trick it can) against three opponents and counted where its rounds ended. Against the trick-grabber it stays humble in 79% of rounds. Against the simple planner, 27%. Against the searching bot, which happily hands tricks back, only 14%: the stubborn loser gets stuck in the 4-6 middle in 69% of rounds and loses 4.0 points per round.
A player who starts humble but adapts does much better against the same opponent. The searcher with a humble opening plan stays humble only 17% of the time too, but when it notices that it is not going to work, it turns around and lands victorious in 34% of rounds. That is how you break even against a good player.
Where the rounds end for a player who wants to lose
Share of rounds by tricks taken. 3,200 rounds per matchup.
- Humble (0-3)
- Middle (4-6)
- Victorious (7-9)
- Greedy (10+)
Show the numbers
| Matchup | Humble | Middle | Victorious | Greedy | Points per round vs opponent |
|---|---|---|---|---|---|
| Stubborn loser vs trick-grabber | 79% | 20% | 1% | 0% | +2.02 |
| Stubborn loser vs planner | 27% | 36% | 21% | 17% | -0.35 |
| Stubborn loser vs searcher | 14% | 69% | 7% | 10% | -4.04 |
| Adaptive humble vs searcher | 17% | 33% | 34% | 16% | +0.08 |
Humble or victorious? The same hand, played twice
Then there is the question of what to aim for when you pick up your 13 cards. So we dealt 9,000 hands and played every one of them twice against the same searching opponent: once with a bot that starts out aiming for humble, once with a bot that starts out aiming for victorious. Same cards, same decree, same opponent. Both bots switch plans when the round forces them to.
The surprise: over all hands it barely matters (+0.02 points per round for starting humble, -0.04 for starting victorious). Up to five trumps the two plans are within noise of each other. With six or more trumps, the plan that sounds obvious is the wrong one. Going for victory with a fistful of trumps wins too many tricks too early and slides into greedy. Starting low lets the strong hand land in seven to nine on its own.
Average point difference per round, by trumps in your starting hand
Your round score minus your opponent's. Same deals, played with each opening plan.
- Start humble
- Start victorious
Show the numbers
| Trumps | Hands | Start humble | Start victorious |
|---|---|---|---|
| 1 | 180 | -0.02 | +0.12 |
| 2 | 850 | +0.23 | +0.31 |
| 3 | 2,029 | -0.09 | +0.06 |
| 4 | 2,705 | -0.06 | -0.14 |
| 5 | 2,091 | +0.03 | +0.08 |
| 6 | 879 | +0.25 | -0.54 |
| 7 | 222 | +0.27 | -0.26 |
One more thing the twin experiment shows: if you could see the future and always pick the right plan, you would gain 2.5 points per round. None of the simple rules we tried (counting trumps, counting high trumps, combinations) captures more than a sliver of that. The plan you start with matters much less than how you adjust trick by trick.
Which cards help most with all this is its own question, and a fun one: the best card in the deck is a 3.
How to play it at the table
- Open low and aim humble by default. It scores as well as going for victory on average, and it is the safer plan.
- Don't chase victory because you have lots of trumps. Let them win the tricks they are forced to win.
- Watch your opponent. If they start dumping tricks on you, humble is not going to happen: switch before you are stuck at four to six.
- Once you have four tricks, humble is gone: now push hard for seven, because the middle band is where points go to die.
An independent check: an earlier analysis of 903 games between Monte Carlo AIs found that 30% of the six-point rounds were humble and 70% victorious. Our searching bots land at 33% and 67%, with different code and a different search method.
Leading the first trick is a small disadvantage (-0.18 points per round), so being the dealer is slightly better. Over a game to 21 that is close to nothing.
How we tested
The rules engine covers the full base game: the decree card, following suit, all six card abilities (including the Witch as the only 9 in a trick and the Monarch forcing the 1 or the highest card) and the scoring bands. Edge cases are checked by automated tests. The searching bot uses Perfect Information Monte Carlo: for each decision it samples 24 hands the opponent could hold, consistent with what has been played and which suits they have shown to be out of, and plays every option to the end of the round.
Limits worth knowing: the bots are strong but not perfect. The searcher decides Fox swaps and Woodcutter discards with simple rules instead of searching them, which probably undersells those two cards. Results for hands with seven or more trumps rest on a few hundred deals. The plan experiment used 9,000 hands. In full games to 21 the two plans are even: the humble-plan searcher won 49% of 1,200 games against the victory-plan searcher, which is why we only call humble the better plan for strong hands. Every run uses a fixed random seed, so the numbers reproduce exactly. More on how we test.
