Global competition · 2026-08-v1
Human + AI leaderboard
Every entry is scored on gauntlet-01. Human scores come from server-replayed move sequences; AI scores come from the same scripted-round harness. This is a reproducible playground, not research-tier evidence.
| # | Player | Type | Verified score | Moves |
|---|---|---|---|---|
| 1 | Gray throughput extended-state policy | ai | 245,697 | 75 |
| 2 | Open-loop beam public policy | ai | 212,213 | 65 |
| 3 | Greedy 1-ply public policy | ai | 207,956 | 65 |
| 4 | Expectimax D2 public policy | ai | 191,989 | 60 |
| 5 | Risk-sensitive D2 public policy | ai | 160,237 | 50 |
| 6 | MCTS public policy | ai | 158,187 | 50 |
| 7 | Rollout H8 public policy | ai | 122,527 | 40 |
| 8 | Sparse expectimax D2 public policy | ai | 120,968 | 40 |
Scripted-round leaderboard
Every policy plays the exact same predetermined rounds: the visible disc sequence is fixed by move number, and every gray disc hides a fixed value that takes its place when revealed. Two policies therefore face identical randomness, move for move. Generated 8/21/2026, 11:18:54 PM. How the benchmark works →
Playground evidence only — scripted rounds are not a research tier and support no qualification claim (see docs/benchmarks.md).
| # | Policy | Mean | Median | Min | Max | Moves | G01 |
|---|---|---|---|---|---|---|---|
| 1 | Gray throughputextended state heuristic-search | 245,697 | 0 | 245,697 | 245,697 | 75.0 | 246k |
| 2 | Open-loop beam heuristic-search | 212,213 | 0 | 212,213 | 212,213 | 65.0 | 212k |
| 3 | Greedy 1-ply heuristic-search | 207,956 | 0 | 207,956 | 207,956 | 65.0 | 208k |
| 4 | Expectimax D2 fair-expectimax | 191,989 | 0 | 191,989 | 191,989 | 60.0 | 192k |
| 5 | Risk-sensitive D2 heuristic-search | 160,237 | 0 | 160,237 | 160,237 | 50.0 | 160k |
| 6 | MCTS tree-search | 158,187 | 0 | 158,187 | 158,187 | 50.0 | 158k |
| 7 | Rollout H8 heuristic-search | 122,527 | 0 | 122,527 | 122,527 | 40.0 | 123k |
| 8 | Sparse expectimax D2 heuristic-search | 120,968 | 0 | 120,968 | 120,968 | 40.0 | 121k |
Flow diagnostics
| Policy | Clears / move | Reveals / move | Max chain | Censored | Illegal | Compute |
|---|---|---|---|---|---|---|
| Gray throughput | 1.95/ 2.40 target | 1.03/ 1.40 target | 7 | 0 | 0 | 0.3s |
| Open-loop beam | 1.71/ 2.40 target | 0.83/ 1.40 target | 6 | 0 | 0 | 0.9s |
| Greedy 1-ply | 1.72/ 2.40 target | 0.85/ 1.40 target | 4 | 0 | 0 | 1.2s |
| Expectimax D2 | 1.68/ 2.40 target | 0.88/ 1.40 target | 6 | 0 | 0 | 5.7s |
| Risk-sensitive D2 | 1.74/ 2.40 target | 0.96/ 1.40 target | 6 | 0 | 0 | 5.7s |
| MCTS | 1.70/ 2.40 target | 0.86/ 1.40 target | 5 | 0 | 0 | 3.3s |
| Rollout H8 | 1.32/ 2.40 target | 0.53/ 1.40 target | 5 | 0 | 0 | 1.0s |
| Sparse expectimax D2 | 1.55/ 2.40 target | 0.82/ 1.40 target | 5 | 0 | 0 | 0.4s |
The 2.4 clears / 1.4 reveals per move figures are diagnostic targets from limited task-record runs, not proven thresholds. Click any score above to replay that game move by move.