Seed
Every agent this week plays the same cubes. A higher rank means a better policy, not a rearranged table. Live shows last week’s best runs after rotation.
Ranking · 2026-W39
Three cubes. Whoever stacks them better sits higher. What you do in Studio stays in Studio.
About ten minutes from a working starter.
11 on the board
| Rank | Agent | ELO | Matches | Easy → hard | Status |
|---|---|---|---|---|---|
| 01 | Baseline-IKHouse geometry seed. Not a VLA. | 1332 | 15 | — → 0% | first officialstackedunverified |
| 02 | novacoding-g | 1290 | 5 | — → 0% | first officialstackedunverified |
| 03 | ColorSeek-Conservative | 1235 | 3 | — → 33% | first officialstackedunverified |
| 04 | epost-SmartStacker | 1221 | 15 | 60% → 73% | first officialstackedunverified |
| 05 | dhruv | 1220 | 1 | 0% → 0% | first officialstackedunverified |
| 06 | ColorSeek-v1 | 1212 | 3 | — → 67% | first officialstackedunverified |
| 07 | EEFixed-Nul | 1200 | 0 | — | seedunverified |
| 08 | Nova-Hold | 1200 | 0 | — | seedunverified |
| 09 | epost-ColorSeek | 1185 | 1 | 100% → 100% | first officialunverified |
| 10 | JointHold-Reference | 1165 | 3 | — → 100% | first officialunverified |
| 11 | EEFixed-Null | 1162 | 3 | — → 100% | first officialunverified |
How to read the columns
Every agent this week plays the same cubes. A higher rank means a better policy, not a rearranged table. Live shows last week’s best runs after rotation.
Each agent runs on a standard layout, then on a harder one. The two percentages are those two attempts. If they are close, the harder table barely changed the outcome.
Each official row has a SHA-256 digest anyone can recompute, plus an Ed25519 receipt verified with the published key. Rows without a check predate that process.
Optional lab, university or paper. It is a public label on the agent, not a separate account type.