Weighted rank
How high does it usually finish? Each game pays by final place, and first and last count most. This is the default ranking.
King+1
Sub-King+0.2
Sub-Kooz−0.2
Kooz−1
Evenly matched bots average 0.
17 bots ranked three ways from the same games, including 6 fixed benchmarks. Updated 2026-11-05 17:19 UTC.
King +1Sub-King +0.2Sub-Kooz −0.2Kooz −1
What the numbers mean · How it works · Code on GitHub
Open “how it plays” under any bot to see how it decides.
| # | Bot | Weighted rank | King only | Score delta | Finishing places | Against each benchmark |
|---|---|---|---|---|---|---|
| 1 | Self-taught + look-ahead The Self-taught model bids alone and picks its best cards, then tries each one on imagined hands. By Essam. Code name: bitter-search. | 1,866 | 1,695 | 4,640 | King 40% | Yosif's trained bot + look-ahead +0.0426%-5 · Random +0.6050%+2463 · Basic rules +0.5750%+178 · Look-ahead +0.2538%+65 · Look-ahead, tuned rules +0.3642%+85 · Yosif's trained bot +0.1632%+31 2,000 games, last played 2026-11-05 |
how it plays | ||||||
| 2 | Yosif's trained bot + look-ahead benchmark Yosif's trained models choose which cards are worth considering. It then tries each one on imagined hands. By Yosif Elessawi. Code name: Hybrid (B-Tig). | 1,804 | 1,685 | 4,623 | King 43% | |
how it plays | ||||||
| 3 | Self-taught One trained model makes every decision. It began by playing randomly and learned only by playing itself, with no strategy written in and nothing copied. By Essam. Code name: bitter-net. | 1,804 | 1,629 | 4,566 | King 36% | Yosif's trained bot + look-ahead +0.0024%-15 · Random +0.6050%+2643 · Basic rules +0.4544%+142 · Look-ahead +0.2837%+66 · Look-ahead, tuned rules +0.2536%+52 · Yosif's trained bot +0.1224%-1 2,000 games, last played 2026-11-05 |
how it plays | ||||||
| 4 | Yosif's bidding, our cards + look-ahead Yosif's look-ahead bot with its card model swapped for Copycat with practice. It keeps Yosif's bidding. By Essam. Code name: rl-hybrid. | 1,787 | 1,680 | 4,602 | King 40% | Yosif's trained bot + look-ahead -0.0124%-2 · Random +0.6050%+2593 · Basic rules +0.5549%+206 · Look-ahead +0.3241%+97 · Look-ahead, tuned rules +0.3341%+103 · Yosif's trained bot +0.1635%+43 2,000 games, last played 2026-11-04 |
how it plays | ||||||
| 5 | Copycat with practice + look-ahead The practised model picks its 2 best cards, then Look-ahead tries them on imagined hands. By Essam. Code name: rl-pimc. | 1,760 | 1,611 | 4,564 | King 38% | Yosif's trained bot + look-ahead -0.0321%-11 · Random +0.6050%+2487 · Basic rules +0.5849%+191 · Look-ahead +0.3441%+84 · Look-ahead, tuned rules +0.3742%+96 · Yosif's trained bot +0.1126%+38 2,000 games, last played 2026-11-04 |
how it plays | ||||||
| 6 | Copycat with practice Copycat's model, improved further by playing itself. It bids with Basic rules. By Essam. Code name: rl-cards. | 1,691 | 1,531 | 4,497 | King 37% | Random +0.6050%+2677 · Basic rules +0.5549%+159 · Look-ahead +0.2437%+60 · Look-ahead, tuned rules +0.3041%+89 · Yosif's trained bot +0.0628%+20 · Yosif's trained bot + look-ahead -0.1016%-25 2,000 games, last played 2026-11-04 |
how it plays | ||||||
| 7 | Yosif's trained bot benchmark Two trained models, one for bidding and one for cards. It started by copying Look-ahead, then improved by playing itself. By Yosif Elessawi. Code name: RL v2. | 1,621 | 1,510 | 4,363 | King 36% | |
how it plays | ||||||
| 8 | Look-ahead benchmark Imagines what the others might hold, plays each card to the end of the round using Basic rules, and picks what scored best. By Yosif Elessawi. Code name: PIMC. | 1,455 | 1,375 | 4,148 | King 31% | |
how it plays | ||||||
| 9 | Copycat + look-ahead Copycat's model picks the most promising cards, so Look-ahead can spend its effort trying only those. By Essam. Code name: net-pimc. | 1,434 | 1,349 | 4,147 | King 29% | Random +0.6050%+2562 · Basic rules +0.5348%+175 · Look-ahead -0.0321%-12 · Look-ahead, tuned rules +0.1228%+27 · Yosif's trained bot -0.1318%-30 · Yosif's trained bot + look-ahead -0.309%-88 960 games, last played 2026-11-04 |
how it plays | ||||||
| 10 | Look-ahead, tuned rules benchmark Like Look-ahead, but it plays each card out with rules tuned for that job. By Yosif Elessawi. Code name: Search bot. | 1,381 | 1,319 | 4,112 | King 29% | |
how it plays | ||||||
| 11 | Copycat A trained model that learned its card play by copying Look-ahead. It bids with Basic rules and does no look-ahead itself. By Essam. Code name: pimc-clone. | 1,310 | 1,270 | 3,965 | King 26% | Random +0.6050%+2915 · Basic rules +0.5149%+142 · Look-ahead -0.1419%-29 · Look-ahead, tuned rules -0.0519%-8 · Yosif's trained bot -0.2810%-78 · Yosif's trained bot + look-ahead -0.347%-117 960 games, last played 2026-11-04 |
how it plays | ||||||
| 12 | Rules + trump count Basic rules, except it stops chasing tricks once its unbeatable trumps cover what it still needs. By Essam. Code name: table-pressure. | 1,024 | 975 | 3,354 | King 16% | Random +0.6050%+2370 · Basic rules +0.0929%+23 · Look-ahead -0.467%-126 · Look-ahead, tuned rules -0.454%-127 · Yosif's trained bot -0.511%-173 · Yosif's trained bot + look-ahead -0.493%-158 960 games, last played 2026-11-04 |
how it plays | ||||||
| 13 | Basic rules benchmark Follows a short list of rules a person wrote. By Yosif Elessawi. Code name: Heuristic. | 985 | 975 | 3,131 | King 11% | |
how it plays | ||||||
| 14 | Cautious rules Basic rules, set to take fewer risks. By Essam. Code name: cautious-heuristic. | 962 | 943 | 3,207 | King 14% | Random +0.5950%+2248 · Basic rules +0.0026%-1 · Look-ahead -0.551%-169 · Look-ahead, tuned rules -0.531%-163 · Yosif's trained bot -0.542%-197 · Yosif's trained bot + look-ahead -0.541%-191 960 games, last played 2026-11-04 |
how it plays | ||||||
| 15 | Always highest card A simple example bot. It never bids and always plays its highest card. By Essam. Code name: highcard-wire. | 726 | 708 | 2,738 | King 8% | Random +0.4844%+3323956 · Basic rules -0.580%-810 · Look-ahead -0.600%-1104 · Look-ahead, tuned rules -0.600%-1112 · Yosif's trained bot -0.581%-1184 · Yosif's trained bot + look-ahead -0.600%-1321 960 games, last played 2026-11-04 |
how it plays | ||||||
| 16 | Always lowest card A simple example bot. It never bids and always plays its lowest card. By Essam. Code name: lowcard-wire. | 726 | 708 | 2,738 | King 7% | Random +0.4843%+1729261 · Basic rules -0.600%-934 · Look-ahead -0.600%-1087 · Look-ahead, tuned rules -0.600%-1051 · Yosif's trained bot -0.581%-1212 · Yosif's trained bot + look-ahead -0.590%-1232 960 games, last played 2026-11-04 |
how it plays | ||||||
| 17 | Random benchmark Plays any legal move. It's the bottom of the scale. By Yosif Elessawi. Code name: Random. | 400 | 400 | 400 | King 0% | |
how it plays | ||||||
What the numbers mean
Bots are ranked in three ways, using the same games. Each way asks a different question.
King only
How often does it come first? Nothing else counts.
King1
Sub-King0
Sub-Kooz0
Kooz0
Evenly matched bots come first 25% of the time.
Score delta
How many points does it win by? This is its average game score minus its opponent's. One huge game can move it.
its score−opponent's score
Evenly matched bots average 0.
Reading the standings
- The standings show each of the three as a rating, like chess Elo. A bot rated 200 points higher wins about 76% of sets. Random is fixed at 400, and ratings compare only within a column.
- A set is one deal played four times with the seats rotated, so lucky cards cancel out.
- Benchmark bots are fixed, and every other bot is measured against them.
- An early result comes from only a few test matches, so treat a small lead as a tie.