Gensyn Launches Evidence Mining Competition for AI Agents

Semi-realistic illustration of an AI agent mining evidence on a live Delphi market interface with USDC rewards.

Gensyn will open its Agent Competition: Evidence Mining on November 23, testing whether autonomous agents can find information that improves forecasts on live Delphi prediction markets. According to the official Gensyn announcement, agents can submit sourced evidence and receive USDC when that information proves useful to Delphi’s forecasting system. The competition rewards information quality rather than compute expenditure or trading performance.

The mining window runs through December 11, with $7,250 in additional competition prizes. Gensyn says the top three participants will divide $6,000, while five $250 bounties are reserved for useful feedback. Evidence rewards are separate from the leaderboard prize pool, creating a direct payment mechanism for useful agent-produced research during the competition itself.

Agents Compete on Evidence Quality

The workflow gives agents a narrower objective than the earlier Delphi Agent Arena trading competition. They search for information relevant to active prediction markets, submit evidence through Delphi and receive a decision, rationale and potential reward within minutes. Gensyn says novelty and speed matter, with duplicate information receiving no reward. An agent therefore has to identify information that the forecasting system does not already know rather than simply retrieve large amounts of public data.

Technical infrastructure for that process has already appeared in the official Delphi SDK. Version 2.2.0 introduced an EvidenceClient capable of submitting sourced evidence, checking submission status and monitoring results, alongside a starter agent that selects markets, finds dated information and waits for a judgment. The competition turns those API capabilities into an explicit economic loop: research, submit evidence, receive evaluation and potentially earn USDC.

That represents a different use of the infrastructure behind Gensyn’s previous Agent Arena experiments, where autonomous systems interacted directly with prediction-market trading and settlement. Gensyn has also created grants for developers building around Delphi and agent infrastructure. Evidence Mining shifts the agent’s economic role from primarily trading information to producing information that other forecasting systems can consume.

Competition Tests an Agent Information Market

The model gives AI agents a measurable task without implying that their research is automatically reliable. Submissions are evaluated before rewards are assigned, and Gensyn says agents compete on successful, accurate and new information. Payment is conditional on the usefulness of evidence rather than simply proving that an agent completed a search or consumed computational resources.

That distinction fits a broader push toward economically active agents, including architectures where software can trade, purchase services or operate within bounded financial permissions. The Chain Observer has tracked similar experimentation through policy-limited smart accounts for AI agents, but Evidence Mining focuses on information production rather than transaction authority. Its economic output is a piece of sourced evidence whose value depends on whether it improves an existing forecast.

The November 23 launch remains a time-limited competition rather than proof of a sustainable market for autonomous research. What it provides is a concrete experiment in pricing machine-generated information: agents search public sources, Delphi evaluates the contribution, and useful evidence can generate an immediate USDC reward. The results after December 11 should provide a clearer measure of whether that incentive structure produces consistently novel and useful information rather than simply higher submission volume.

Find Us on Socials

Join Our
Newsletter

Subscribe to get latest crypto news!

Latest News

You may also like

Motes Adds Aave Lending to AI Agent Stack

The Chain Observer
Privacy Overview

This website uses cookies so that we can provide you with the best user experience possible. Cookie information is stored in your browser and performs functions such as recognising you when you return to our website and helping our team to understand which sections of the website you find most interesting and useful.