# What is Delphi?

## Delphi, an AI-settled Information Marketplace

[Delphi](https://delphi.gensyn.ai/) is an information market. It allows users to create their own markets in any topic, and others can buy and sell positions on the possible outcomes. Think of it like a marketplace for predictions, where participants commit capital to the outcomes they believe are most likely, and the market's odds shift in real time based on collective sentiment.

When a market creator sets up their market, they configure an AI model and a settlement prompt that will be used to determine the final outcome once the market closes. This means resolution isn't dependent on a centralized team or a single trusted oracle, but rather it's driven by an AI judge.

### Verifiable Settlement

Delphi takes this a step further with **verifiable settlement**, powered by [Gensyn's REE (Reproducible Execution Environment).](https://app.gitbook.com/s/jHECdpSAZDuPfU2oZmM2/)

When a market creator chooses a verifiable model, the settlement process produces a [receipt](https://app.gitbook.com/s/jHECdpSAZDuPfU2oZmM2/ree/receipts), a cryptographic record containing the model, the prompt, the data provided, and a hash of the computation. Anyone can take that receipt and independently re-run the exact same computation on their own machine to confirm the result by using REE in [reproducible](https://app.gitbook.com/s/jHECdpSAZDuPfU2oZmM2/ree/examples#production-reproducible-inference-with-a-real-model) mode.

{% hint style="warning" %}
*Verifiable settlement is optional.*&#x20;

Market creators can also choose to settle markets using cloud-hosted LLMs from providers like OpenAI, Gemini, and more.&#x20;
{% endhint %}

The key point of verifiable settlement is its assurance of accurate computation based on the provided input data and prompt. It guarantees that the AI model has performed the expected calculations reliably to maintain computational integrity. If the data is credible, the entire process is verified, but the main focus is on producing correct and reproducible computations, regardless of the data.

*Not all markets use verifiable models.* Creators can also choose non-verifiable cloud-hosted models (like GPT-4o, Gemini, or Claude) that have tool access and can look up data sources directly. These are convenient but aren't bitwise reproducible, so independent verification isn't possible in the same way.

{% hint style="success" %}
Markets that use verifiable settlement are tagged with a **Verifiable** label so you can identify them at a glance.
{% endhint %}

### How It Works

Delphi has a straightforward lifecycle:

1. A creator makes a market by posing a clear question, defining possible outcomes, seeding initial liquidity, and configuring the AI settlement model and prompt.
2. Participants trade by buying positions on the outcomes they believe are most likely. Odds shift dynamically as capital flows in.
3. The market closes at a predetermined time, and no more trades can be placed.
4. The creator settles the market by running the settlement prompt through the configured AI model, pasting the result, and selecting the final outcome.
5. Winners claim their payouts based on their share of the winning outcome pool.

If a creator fails to settle within the deadline (24 hours after close), the market auto-settles. In this scenario, the creator forfeits their liquidity contribution, and all participants are refunded.

{% hint style="info" %}
Read more about [How AI Settlement Works.](https://docs.gensyn.ai/testnet/delphi/how-ai-settlement-works)
{% endhint %}

### What's Next

Ready to get started? [Connect your wallet and pick up some $TEST tokens](https://docs.gensyn.ai/testnet/get-started#connect-your-wallet) and make your first information market on Delphi today.
