# Create a Market ## The Market Creation Process Anyone can create a prediction market on Delphi. The creation flow has six steps: **\[1]** Details, **\[2]** Outcomes, **\[3]** Economics, **\[4]** Settlement, **\[5]** Timing, and **\[6]** Review. Click *Create Market* in the top-right corner to get started. ### Step 1: Details Give your market a clear, specific question that can be definitively resolved by a certain date. For example: *"Will the Temperature in St. Augustine, FL Reach 102 Degrees Fahrenheit by May 18th, 2026?"* The more specific and unambiguous the question, the better, whereas vague questions lead to messy settlements. Then, select one of the available categories based on the market's question. ### Step 2: Outcomes This is where you define the possible outcomes for your market: you need at least two, with a maximum of 20 outcomes permitted. For a straightforward yes/no question, "Yes" and "No" will be prefilled. If your market has more nuance, you can add custom outcomes (up to 90 characters each). ### Step 3: Economics This is where you set up the financial parameters of your market. #### Initial Liquidity This is the liquidity pool that gets trading started. The more liquidity you provide, the more attractive the market is to traders. {% hint style="info" %} A portion of your initial liquidity is reserved as a refundable deposit. You get it back when you settle the market, but forfeit it if you [fail to settle within the deadline.](https://docs.gensyn.ai/testnet/delphi/settling-a-market) {% endhint %} #### Trading Fee Set the percentage fee charged on every trade in your market. The default is 2%, but you can adjust it. As the creator, you'll receive 1.5% of the fees on total traded volume within your market. ### Step 4: Settlement Settlement configuration is the most important step as it determines how your market will ultimately be resolved. There are three things to set here: the **\[1]** AI model, your **\[2]** data sources, and the **\[3]** settlement prompt. #### Judge Model Choose the AI model that will evaluate the outcome. You have two paths: **\[1]** non-verifiable models or **\[2]** verifiable models. *Non-verifiable models* (e.g., GPT-5.3, Gemini 3, Claude 4.6) are cloud-hosted LLMs with tool access. They can look up data sources directly, which makes settlement convenient. However, because they run on third-party infrastructure, results aren't bitwise reproducible and someone else can't re-run the exact same computation and guarantee an identical result. *Verifiable models* (e.g., Qwen-2.5-14B-Instruct) run inside [Gensyn's REE (Reproducible Execution Environment).](https://app.gitbook.com/s/jHECdpSAZDuPfU2oZmM2/) When you settle with a verifiable model, you produce a receipt containing the model, prompt, data, and a hash. Anyone can take that receipt and independently verify the result on their own machine. This is what makes a market "Verifiable." {% hint style="info" %} For a deeper explanation of how this works, see [How AI Settlement Works.](https://docs.gensyn.ai/testnet/delphi/how-ai-settlement-works) {% endhint %} #### Data Sources Select or input the data sources the judge model should reference when determining the outcome (e.g., CoinMarketCap, CNN, CNBC, ESPN, Bloomberg, Reuters, Associated Press, X/Twitter). You can also type in custom data sources. These will be displayed on your market's page, so it's a good way to let potential participants know what sources you'll be using for your data. {% hint style="warning" %} For verifiable models, keep in mind that you'll need to manually retrieve data from these sources at settlement time and include it in your prompt, since local models can't access the internet. {% endhint %} #### Settlement Prompt Write the prompt the judge model will use to determine the outcome. Be explicit: tell the model the question, what the possible outcomes are, and exactly how to decide. How you write the prompt depends on the type of model you selected. #### For Non-Verifiable (cloud-hosted) Models These models have tool access, so your prompt should instruct the model to fetch the data it needs from your specified data sources. For example: > "You are an AI judge settling a prediction market. The question is: Did it rain in Los Angeles on April 8, 2026? Review the web page provided from the Weather Underground and determine whether there was precipitation. Precipitation greater than 0 means the market should resolve to Yes. Precipitation of 0 means the market should resolve to No. Give your final answer as a single Yes or No." The model will use its tools to access the data source and determine the outcome directly. #### For Verifiable (local) Models: These models run locally and do not have internet access or tool access. Your prompt should be written so that at settlement time, you can append the raw data you manually retrieved from your specified sources. The model will then evaluate that data to produce its result. Using the same weather market example (but with a local model) the prompt would be: > "You are an AI judge settling a prediction market. The question is: Did it rain in Los Angeles on April 8, 2026? You will be provided with the raw content of a weather data source below. Review the data and determine whether there was precipitation. Precipitation greater than 0 means the market should resolve to Yes. Precipitation of 0 means the market should resolve to No. Give your final answer as a single Yes or No. Web page data below:" This is the prompt as you'd write it during market creation. Notice it ends with "Web page data below:"... That's the placeholder for where the data will go. When it's time to settle the market, you go to your specified data source (in this case, Weather Underground), retrieve the relevant data, and append it to the end of your prompt. The final prompt you run through REE would look something like this: > "You are an AI judge settling a prediction market. The question is: Did it rain in Los Angeles on April 8, 2026? ... Web page data below: \ > **Source: Weather Underground, Los Angeles, CA, April 8, 2026 Precipitation: 0.02 in."**\ > \[rest of settlement prompt, as above] The verifiable model reads the appended data, evaluates it against the criteria in your prompt, and produces the settlement result. See [Settling a Market](https://docs.gensyn.ai/testnet/delphi/settling-a-market) for the full settlement flow. ### Step 5: Timing Set when your market closes to new trades. There are some quick-set times available, or you can choose a custom date and time. The settlement deadline is automatically set to 24 hours after the market closes. This is the window you have to settle the market. If you miss this deadline, the market will auto-settle and you'll end up forfeiting your entire liquidity contribution, and all participants are refunded. ### Step 6: Review This is the final step in the market creation flow, and is a page containing everything you've configured so far: **\[1]** category, **\[2]** question, **\[3]** outcomes, **\[4]** trading fee, **\[5]** close time, **\[6]** settlement deadline, **\[7]** settlement type (reproducible or non-reproducible), **\[8]** initial liquidity, and the **\[9]** market creation fee. Review everything, click confirm, and your market goes live! ### After Your Market Goes Live Once your market is active, traders can start buying and selling positions on the outcomes. You can monitor your market's activity from the [Activity](https://docs.gensyn.ai/testnet/delphi/activity) page under the *My Markets* tab. When the market closes, you'll need to settle it. See [Settling a Market](https://docs.gensyn.ai/testnet/delphi/settling-a-market) for the full walkthrough of the post-close settlement flow.