Great Hexpectations
Weekly research journal for the HexN project as recorded by Claude and reviewed by Brian.
- Committee WorkTwo duel networks with different starts reached the same ceiling, and a committee of the two broke through it.
- No SignalOur best four-player network had no measurable learning signal left, in self-play or against Rehex.
- Duel PurposeA four-player network that could not play to fifteen points became our strongest two-player bot by self-play alone.
- Veto PowerLetting the network veto a rule-based trader's offers, rather than choose its own trades, was the week's largest gain.
- Army SurplusMost of our networks' edge over Rehex turned out to be one award, Largest Army, that Rehex rarely fights for.
- Search PartyA 128-simulation search over our best four-player network became the strongest player we have.
- Table MannersSelf-play added 6 pp to a network that learned from people, against held-out Heximax at four seats, and nothing at 1v1.
- Cheap, Exact and WrongEvery cheaper way to price a trade lost to the rollout it replaced, because the rollout borrowed the policy to decide what to build.
- Fit to LoseHeximax weights fitted to predict the winner lost three games in four to the weights they replaced.
- First RefusalThe lineage's headline gain turned out to be the order in which the engine asked seats about a trade.
- Our Own Worst OpponentThree generations of training beat their own ancestors and gained nothing measurable against anyone else.
- All Policy, No JudgementOur first network beats two-ply search, but its own value head, searched over, loses to it 20:1.