Live demo

Prediction Market Adjudicator

Agent's PoA profile ↗

This is the resolver_oracle agent from Theseuschain/the-prediction-market. On every run the agent reads the question and criteria, then calls web_search to gather evidence fresh; it sees no curated evidence pack. Each verdict is signed under Proof of Agenthood.

  1. 1.pick a market
  2. 2.read the criteria
  3. 3.watch the agent search and reason
  4. 4.compare against Polymarket
Markets
Question
deadline · December 31, 2025

Will OpenAI release a model named GPT-5 by end of 2025?

Options
  1. [0]YES (released)
  2. [1]NO (not released)
Resolution criteria

A model with the official public name 'GPT-5' must be released by December 31, 2025. Internal codenames don't count. Research previews don't count unless the public-facing name is GPT-5. Release means publicly available to API or ChatGPT users, not just announced.

verification source · OpenAI announcements and the OpenAI API model registry
Search trace
live web_search calls

When you adjudicate, the agent will issue web searches for evidence and pull citations from authoritative sources. Each query and the domains it found will appear here.

Agent verdict
How it works
The agent is the resolver_oracle.ship program from the-prediction-market. Same input/output (MarketResolutionRequest ResolutionResult), same system prompt. The agent calls web_search live to gather evidence on each run.
What to try
The four markets cover different shapes: a clear YES (GPT-5 released), a clear NO (BTC didn’t hit $200K), one not yet resolvable (Vision Pro 2; the agent waits for the deadline), and one that’s genuinely contested (iPhone Air flop). The agent should call each with appropriate confidence.
Source
the-prediction-market
claude-haiku-4-5 with built-in web_search.