> For the complete documentation index, see [llms.txt](https://docs.delphi.fyi/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.delphi.fyi/how-ai-settlement-works.md). # How Automated Settlement Works How markets resolve automatically after close, including model review, verification, reports, and no-outcome payouts. ## Automated Settlement Automated Settlement resolves every new Delphi market after trading closes, including markets opened through Predict Anything. Each market's question, outcomes, data sources and settlement criteria are set before trading starts. Delphi then reviews the available evidence against them and submits the final result on-chain. Creators do not run a model or select a winner after close. Traders can inspect the configuration and resulting settlement report. {% hint style="info" %} Automated Settlement applies only to markets marked **Automated Settlement** in the market view. Existing **Legacy** markets use the previous manual settlement flow. {% endhint %} ### Before You Trade Every automated market shows the inputs that guide its settlement: * **Question:** The real-world question being answered. * **Outcomes:** The results the workflow can select. * **Data Sources:** The sources used to review evidence. * **Settlement Criteria:** The rules that map evidence to an outcome. * **Analyst Models:** The models that review available evidence. * **Judge Model:** The model that selects the final result. * **Trading Close:** When new trades stop. Review these details before trading. Clear questions, complete outcomes, reliable sources, and precise criteria make markets easier to resolve. On markets opened through [Predict Anything](/predict-anything.md), these terms are generated before the market exists. Delphi drafts the settlement criteria and data sources while it prices your prediction, and shows them on the forecast card. If it can't produce terms that public sources could settle, the prediction is marked *Forecast Only* and no market is opened. {% hint style="warning" %} Automated Settlement cannot fix an ambiguous market. Incomplete questions, outcomes, or criteria may prevent a winning outcome. {% endhint %} ### The Automated Settlement Process After trading closes, the market moves through six stages: 1. **Trading closes.** No more positions can be bought or sold. 2. **Delphi waits for evidence.** Settlement starts at the offset chosen when the market was created, 1 to 12 hours after close, so official results and reliable reporting have time to appear. 3. **Analyst models review the market.** They assess the rules, sources, and available evidence. 4. **The judge evaluates the reports.** It selects the outcome that best matches the declared rules. 5. **The result is submitted on-chain.** The workflow sends the selected outcome to Delphi's contracts. 6. **The market page updates.** It shows the outcome, payouts, settlement details, and available reasoning. Creators do not need to take settlement action after an automated market closes.
### Settlement Timing Automated Settlement starts after trading closes. It is not instant. The workflow waits for evidence, reviews model outputs, and submits the result on-chain. Settlement starts 1 to 12 hours after close (8 hours by default), and every market settles within 48 hours of trading close. The market contract enforces this window. Markets are set to close before the outcome is knowable. If trading stayed open until the answer was public, prices would drift to near-certain and trading would stop meaning anything. {% hint style="info" %} Trading close ends trading. It does not always end the evidence window. Evidence published later can still matter when it describes an event within the market's rules. {% endhint %} ### How the Models Work Together Automated Settlement separates evidence review from the final decision. #### Analyst Models Analyst models independently review the market and its available evidence. They: * Find evidence from the declared sources. * Apply the settlement criteria. * Explain which outcome the evidence supports. * Identify incomplete or conflicting evidence. #### Judge Model The judge receives the analyst reports and makes the final structured decision. It evaluates the reports against the market's question, outcomes, and settlement criteria. Separating these roles reduces reliance on one model response. It also creates a clearer decision record. ### What Is Verifiable? The final judge runs through [Gensyn's REE (Reproducible Execution Environment)](https://docs.gensyn.ai/tech). REE produces a [receipt](https://docs.gensyn.ai/tech/ree/receipts) that records the model, inputs, output, and computation hash. You can use the receipt to reproduce the judge's computation. It confirms that the recorded model produced the recorded output from the recorded inputs. Verification proves computational integrity. It does not guarantee that: * Every source was factually correct. * The question and outcomes were well designed. * The analyst reports included every relevant piece of evidence. * The selected outcome represents an objective or universal truth. The receipt verifies the computation. The market's rules, sources, and settlement report let you evaluate the decision. {% hint style="warning" %} Verifiable does not guarantee a correct settlement. It means the recorded computation can be independently checked. {% endhint %} ### Understanding the Settlement Report After settlement, select *View Settlement Report* in the AI Settlement section. Depending on the available artifacts, the report may include: * The final outcome. * Analyst and judge model roles. * The market's data sources and settlement criteria. * Settlement reasoning and the on-chain transaction. Settlement reasoning explains how the evidence was assessed against the market's rules. Delphi displays it only when a settlement artifact is available.
### What Winners Receive When a market settles to a winning outcome, winning shares become claimable from the market page or *Claimable* in *Activity:* * On LMSR markets, each winning share pays $1. * On DPM markets, winning shares split the pool. ### What Does "No Outcome" Mean? Sometimes a market closes without a winning outcome. Delphi displays this state as *No Outcome.* When this happens: * No listed outcome wins. * Positions pay at the final market prices when trading closed. On LMSR markets, prices freeze at close, so a share priced at 12% pays $0.12. * Payouts may be higher or lower than the original purchase amount. * Proceeds appear in the **Payouts** section of [Activity](/activity.md). This is not a refund of the original purchase amount. For example, imagine **Yes** costs `$0.70` and **No** costs `$0.30` when trading closes. If the market has no outcome, Yes positions pay at `$0.70` per unit and No positions pay at `$0.30` per unit. No position receives a winner-takes-all payout. {% hint style="info" %} **No outcome** differs from a market removed for a policy violation. Removed markets display **Removed** in the app. {% endhint %} ### Frequently Asked Questions These answers cover automated settlement. For other common questions, see [Help](/help.md).
Does the creator choose the winner? No. The creator defines the market and reviews its suggested sources and criteria. The Automated Settlement workflow selects the final outcome after close.
Can the creator change the rules after the market goes live? No. The settlement inputs lock when the market is created. Traders rely on the published configuration.
Does Automated Settlement guarantee the correct answer? No. It evaluates available evidence against declared rules. Models, sources, and market definitions can still be incomplete or wrong.
Why does settlement take time? Official evidence may not be available at trading close. Delphi waits for evidence, runs the analyst and judge workflow, and submits the result on-chain, within 48 hours of close
Where can I see why a market settled? Open the market and select *View Settlement Report.* Settlement reasoning appears when a supporting artifact is available.
What happens if Delphi cannot declare a winning outcome? The market displays *No Outcome.* Positions pay at their final market prices, and proceeds appear under *Payouts.*
Can an automated market still be removed? Yes. Policy enforcement is separate from settlement. Policy-flagged markets display *Removed.*
Who settles markets opened through Predict Anything? The same automated workflow. Delphi created the market, but it doesn't pick the winner, and its opening position settles alongside everyone else's.
### Next Steps * Learn how to [create a clear market](/create-a-market.md). * Learn how to [explore markets and trade](/explore-and-trade.md). * Review positions and payouts in [Activity](/activity.md). * Read more about [Gensyn REE](https://docs.gensyn.ai/tech). --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter: ``` GET https://docs.delphi.fyi/how-ai-settlement-works.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.