We Simulated 1,800 Games of Spades: Win Rates, Bidding Accuracy and Nil Success
Real numbers from 1,800 games played by the actual Spedes engine across all six modes: which side wins, how often bids set, and when a Nil is worth trying.
Most Spades strategy articles online repeat the same received wisdom. This one does not — every figure below comes from 1,800 complete games played by the actual Spedes engine, the same code that runs when you press Play on this site. We let the production AI personas bid and play against each other across all six modes, 300 matches per mode, and recorded what actually happened.
Why publish the raw numbers? Because "bid aggressively" and "avoid bags" are opinions until you can measure them. Some of what we found confirmed the standard advice. Some of it — the Joker set rate in particular — surprised us, and changed how we tune the mode. If you play Spedes, these are the numbers your games actually produce.
Want to reproduce this yourself? The script is in the repository at scripts/collect-stats.mjs. It runs the real engine headlessly and prints the same JSON table below. Nothing here is hand-entered.
300 finished matches per mode, all seats played by AI. "Bid make rate" is the share of non-Nil bids that were made exactly or exceeded; "set rate" is the share that failed. Nil figures are per seat per match.
Partnership: 150: 5.6: 58.8%: 41.2%: 30.6%: 61.0%: 3.70. Solo 3: 200: 4.3: 90.8%: 9.2%: 2.9%: 19.2%: 1.23. Solo 4: 150: 5.9: 65.6%: 34.4%: 33.9%: 45.0%: 5.40. Mirror: 200: 5.4: 80.2%: 19.8%: 32.7%: 58.9%: 2.60. Joker: 200: 8.8: 42.7%: 57.3%: 21.5%: 77.9%: 7.21. Whiz: 150: 4.1: 68.8%: 31.2%: 16.1%: 70.5%: 2.42
The single most striking number in the whole set: in Joker mode, 57.3% of bids are set , and only 42.7% are made. No other mode comes close — the next-worst is Partnership at 41.2% set. Joker games also run the longest by a wide margin (8.8 hands on average, versus 4.1 for Whiz) and trip the bag penalty most often (7.21 per match).
This is a direct consequence of the two jokers being super-trumps. They absorb tricks that would otherwise go to the highest spade, which means the relationship between "cards I think I can win" and "tricks I actually win" is far less predictable than in standard Spades. A hand that looks like a comfortable 4-bid in Partnership frequently comes up short in Joker, because a joker snatches a trick you had counted on.
Practical takeaway: if you bid the same way in Joker that you do in Partnership, you will set far more often than you are used to. The data says bid one lower than your Partnership instinct in Joker, and treat a Nil as a genuinely premium play — it succeeds 77.9% of the time, the highest of any mode.
Solo 3 is the outlier at the other end: 90.8% of bids are made , and only 9.2% set. Nil is almost never attempted (2.9%) and almost never works when it is (19.2%). With only 17 cards each and 3 players, every hand is short and the outcome is unusually calculable — there is simply less room for a surprise to swing a trick.
The win distribution is also the most even of any mode (34.3% / 28.0% / 37.7%), which is what you want from a three-player free-for-all: no seat is structurally advantaged. If you find Solo 4 or Partnership feels luck-driven, Solo 3 is the mode where skill at exact hand evaluation pays off most, because the field is so much more predictable.
Nil success varies by a factor of four across modes, which is the kind of spread that should change how you play each one:
Joker: 21.5%: 77.9%: Premium play — try it. Whiz: 16.1%: 70.5%: Strong when the spade count allows. Partnership: 30.6%: 61.0%: Good odds, well worth attempting. Mirror: 32.7%: 58.9%: Solid, and the overtrick penalty makes it attractive. Solo 4: 33.9%: 45.0%: Coin flip — bid it only with a genuinely void hand. Solo 3: 2.9%: 19.2%: Do not. The field is too short to hide 13 cards
Solo 4 is the interesting failure case. Players attempt Nil more often there than in any other mode (33.9%) yet succeed least often of the four-player modes (45.0%). With no partner to cover a broken Nil and four sets of hands competing for the same 13 tricks, a Nil that would be safe in Partnership gets broken almost as often as it survives. The lesson: Solo 4 rewards Nil discipline, not Nil enthusiasm.
Partnership trips the 10-bag penalty 3.7 times per match on average — more than any mode except Joker and Solo 4. Combined with the 41.2% set rate, this is the clearest evidence in the whole dataset for the standard advice that bag management decides casual Spades: the penalty is not a rare disaster, it is a recurring cost that quietly drains points from the side that keeps taking "one more trick because it was there."
Mirror is the deliberate counter-example. Because Mirror charges −5 for every overtrick immediately rather than accumulating bags, the bag penalty never fires — and the set rate drops to 19.8%, the second-lowest in the set. The mode does exactly what it was designed to do: convert a slow, hidden cost into an immediate, visible one, and in doing so make precision bidding pay.
Across every mode, no seat or side is structurally favoured. The two-side modes land within a few points of 50/50 (Partnership 53/47, Mirror exactly 50/50, Whiz 52/48), and the multi-side modes spread their wins across all seats with no seat dominating. This is not an accident — it is the result of the dealer rotating every hand, which is why the last-to-bid informational advantage does not compound into a permanent edge.
The one caveat is Joker, where the first side wins 57.7% to the second side's 42.3%. That is a wider gap than any other two-side mode, and it is almost certainly the jokers again: whoever leads into a trick contest that the super-trumps can decide has a measurable first-mover edge. It is not large enough to make the mode unfair, but it is large enough that we are watching it as we tune.
Transparency matters more than a confident-sounding number, so here is exactly what produced the table:
- Engine: the production src/game/engine.js — the identical code that runs in your browser. Not a simplified model.
- Players: the six production AI personas (Hana, Dex, Ingrid and the rest), each with its own bidding tendencies and playing habits. Seat 0 plays with no persona, matching how a local practice table seats the human.
- Volume: 300 finished matches per mode, 1,800 total. Each match runs until a side reaches the target score, capped at 80 hands.
- Tier: the Standard preset for every mode (Partnership 150, Solo 3 and Mirror 200, and so on), so the numbers reflect the default table you are dealt.
- What counts as a set: a non-Nil bid where tricks won fell short of the bid. Overtricks count as made, not as a separate category.
One honest limitation: these are all-AI games. A human table bids and plays differently — people bluff, tilt, and miscount in ways the personas do not. The relative picture (Joker is hard, Solo 3 is calculable, Nil is mode-dependent) will hold, but treat the exact percentages as a baseline, not a prophecy.
- Playing Joker? Bid a trick lower than your Partnership instinct, and take the Nil when you have a void — it lands 78% of the time.
- Playing Solo 3? Trust your count. The field is predictable enough that precise bidding wins, and almost nobody should be bidding Nil.
- Playing Solo 4? Be stingy with Nil. A 45% success rate means you are flipping a coin with your match on the line.
- Playing Partnership? Watch the bag count from hand one. At 3.7 penalties per match, the slow bleed is real, not theoretical.
- Playing Mirror? Bid exactly. The −5 overtrick charge makes one extra trick cost more than it would anywhere else.
Every one of these is a claim you can test at a real table — open a free Partnership game and watch whether the pattern holds for you. That, more than any single statistic, is the point of measuring it in the first place. If you want the rules behind any of these modes before you play, start with the complete rules guide , or see how each mode differs .
Frequently asked questions
Are these statistics from real games or a model?
Real games. Every match was played by the actual Spedes engine — the same src/game/engine.js that runs in your browser — with the production AI personas bidding and playing. Nothing is estimated or modelled; the script simply records what the engine does over 1,800 games.
Can I reproduce these numbers myself?
Yes. The script is in the repository at scripts/collect-stats.mjs. Run it with the number of rounds you want (node scripts/collect-stats.mjs 300) and it prints the same table. Because the engine uses a random deal each time, your exact percentages will differ slightly, but the pattern across modes will match.
Which Spades mode is the hardest to bid in?
Joker, by a wide margin. 57.3% of bids are set in Joker, versus 41.2% in the next-hardest mode (Partnership). The two jokers act as super-trumps and unpredictably steal tricks you had counted on, which makes hand evaluation far less reliable than in standard Spades.
When should I bid Nil?
It depends heavily on the mode. Nil succeeds 77.9% of the time in Joker and 70.5% in Whiz, making it a strong play there. In Partnership it is a solid 61%. But in Solo 4 it only works 45% of the time, and in Solo 3 just 19.2% — so in those two modes a Nil is closer to a gamble than a strategy.
Is any seat or side unfairly advantaged?
No. Across all six modes the win rates are balanced — the two-side modes land within a few points of 50/50 and the multi-side modes spread wins evenly across seats. The dealer rotates every hand, which prevents the last-to-bid informational advantage from becoming a permanent edge. The only mild exception is Joker, where the leading side wins 57.7%, a gap we are monitoring.
Continue reading
About · Contact · Privacy · Terms