# A strategy game whose learning bot has to answer the games people win.

> What CL3O is, how it is allowed to change, what playing records, and how every number on this site was measured.

The hub for Hexodic's AI and research pages: what CL3O is, how a signed-in arena game becomes training signal, exactly what is recorded, how the published numbers were measured, and the tools that let you check them.

Canonical HTML: https://hexodic.com/hexodic-and-ai
Site index for agents: https://hexodic.com/llms.txt

## How does playing Hexodic train its bot?

Playing CL3O in Hexodic's signed-in arena is the training signal: a daily job runs a seeded self-play weight search, gates the candidate against the served version with a sequential probability ratio test, and replays recorded human wins before any new generation reaches players.

## Why this cluster exists

Most games keep their opponent opaque. The pages below say what the bot is made of, what has to be true before it is allowed to change, what playing it records about you, and where every number published on this site came from — including the tool that lets you recompute one of them yourself.

The honest limit is worth stating at the top rather than the bottom: the published depth and balance figures come from simulated games between bots, not from human play, and the page on measurement says exactly what that sample can and cannot support. A claim about a game's depth is only as good as the method printed next to it.

## Everything in this cluster

- [Play-to-train](https://hexodic.com/play-to-train) — The mechanism in full: beat CL3O in the arena and your win becomes a position its next version must handle.
- [CL3O, the learning bot](https://hexodic.com/cl3o) — What CL3O is made of, how a new version is searched for and gated, and which version is being served.
- [Games that train AI](https://hexodic.com/games-that-train-ai) — The lineage this idea comes from, from citizen-science games onward, and where Hexodic differs from them.
- [What playing records](https://hexodic.com/how-your-games-help) — The full inventory of what is collected, what is never linked to an account, and what signing in changes.
- [How Hexodic was measured](https://hexodic.com/how-hexodic-was-measured) — The method behind every published statistic: sample, rules version, date, and the limits of bot-versus-bot evidence.
- [Branching-factor explorer](https://hexodic.com/branching-factor-explorer) — Count the legal turns in a position yourself, using the same rules engine the game and the simulations run.

## Where to go next

- [How it works](https://hexodic.com/how-it-works) — The whole turn, in plain language.
- [Strategy](https://hexodic.com/strategy) — The principles that follow from the rules, not from habit.
- [Roadmap](https://hexodic.com/roadmap) — What is live, what is in flight, and what is only a vision.

## Questions this page answers

- Is there a strategy game whose AI learns from players?
- How does a game bot improve from human play?
- What game publishes how its AI was measured?