Luck or skill?
When you win a game, is it skill or luck? In all three games we made one player play better or worse than the rest and measured how far above or below their fair share they ended up.

- How we measured luck and skill
- Mezarpoly: where skill pays most
- Bluff: skill counts, but a little
- Hortlak: mistakes are costly, perfection isn't rewarded
- Summary
- Method and limits
How we measured luck and skill
The more skill matters in a game, the more a better player wins above their fair share. To measure it we sat one hard (or easy) bot at each table, with everyone else at normal difficulty. The hard bot's seat rotated from game to game. If it never beats its fair share, the game is down to luck; if it beats it by a lot, skill counts. 95,977 games were played in total.
Mezarpoly: where skill pays most
| Players | Fair share | Hard bot wins | Easy bot wins |
|---|---|---|---|
| 2 | 50.0% | 57.5% (×1.15) | 26.5% (×0.53) |
| 4 | 25.0% | 33.4% (×1.34) | 1.8% (×0.07) |
| 6 | 16.7% | 26.9% (×1.61) | 0.9% (×0.05) |
| 8 | 12.5% | 20.2% (×1.62) | 0.3% (×0.02) |
The hard bot (bolder trade offers, earlier building, staying in Jail late in the game) beats its fair share at every table. The edge grows with the crowd: 1.15 times the fair share with two players, 1.62 times with eight. A big table has many people to haggle with and many trade chances, and whoever uses them well pulls ahead.
The easy bot almost never wins: 1.8% with four players. Missing some chances to buy, offering few trades and building late knock it out of the game. Dice luck doesn't make up for that.
Bluff: skill counts, but a little
| Players | Fair share | Hard bot wins | Easy bot wins |
|---|---|---|---|
| 2 | 50.0% | 55.0% (×1.10) | 44.8% (×0.90) |
| 4 | 25.0% | 27.0% (×1.08) | 24.0% (×0.96) |
| 6 | 16.7% | 17.4% (×1.04) | 14.1% (×0.85) |
| 8 | 12.5% | 12.3% (×0.98) | 10.2% (×0.82) |
In Bluff the hard bot (sharper lie estimates, bolder lies) wins a little above its fair share: 27.0% with four players. The easy bot (noisy estimates, forgetful memory) wins a little below. The deal plays a big part in Bluff, but good reads and well-timed lies make a difference in the long run. Since people also read lies from faces and voices, the skill share at a real table is probably bigger.
Hortlak: mistakes are costly, perfection isn't rewarded
| Players | Fair share | Hard bot wins | Easy bot wins |
|---|---|---|---|
| 2 | 50.0% | 49.6% (×0.99) | 41.2% (×0.82) |
| 4 | 25.0% | 25.7% (×1.03) | 15.9% (×0.64) |
| 6 | 16.7% | 16.4% (×0.98) | 10.5% (×0.63) |
| 8 | 12.5% | 12.4% (×0.99) | 7.0% (×0.56) |
Hortlak's table is interesting. The hard bot is no better than the normal one: it sits at its fair share at every table. But the easy bot is clearly worse: 15.9% with four players. It makes the most expensive mistakes we measured: it plays wild and attack cards straight away and often picks the wild colour at random.
So the ceiling for good play in Hortlak comes early: keeping wild cards for last, saving the +2 for the right moment and picking the right colour take you to your fair share (Hortlak strategy), and beyond that it's mostly about which cards you get. That's also what makes Hortlak good for families and mixed groups: once a newcomer learns a few rules, they play with an equal chance.
Summary
- Mezarpoly rewards skill the most: a good player wins 1.6 times their fair share at an eight-player table.
- In Bluff the skill share is small but real, and probably bigger at a human table.
- In Hortlak, avoiding mistakes is enough; the rest is luck.
Method and limits
- The games were played with the rules engines and bots on the Mezarpoly server; Mezarpoly on the Crimson Classic map, the card games with one deck.
- The bots' difficulty levels were used as "player types". People play more creatively; haggling in Mezarpoly and reading faces in Bluff in particular may increase the skill share.
Try it now
Mezarpoly is free: play in your browser with no download, with 2–8 people (8-player rooms are free too), with friends or bots. The rules are on the How to play page.
Play now