About the challenge
Case Closed is an asynchronous agentic hackathon.
Your mission: build an autonomous detective agent capable of solving procedurally generated murder mysteries.
Your agent will interact with a custom investigation environment through a simple HTTP API. It must gather evidence, interview witnesses, reason about timelines, and ultimately identify:
- The culprit
- The means
- The evidence-backed timeline
Actions costs budget. Seeds presents new mysteries. The best agents will balance reasoning, exploration, and efficiency.
We strongly encourage participants to build using Railtracks, though teams are free to incorporate additional tools, models, and frameworks as desired.
Unlike traditional LLM benchmarks, scoring is fully deterministic. Submissions are evaluated against hidden ground-truth solutions using evidence attribution and mechanical scoring.
Get Started
- Register your team
- Read the API documentation
- Open a mystery seed
- Build your detective agent
- Submit your repository
- Compete on the live leaderboard
The environment can be accessed entirely through HTTP requests—no SDK required.
Requirements
What to Build
Build an autonomous detective agent capable of solving Case Closed mysteries through the official API.
Your agent should:
- Navigate the investigation world
- Collect and manage evidence
- Interview witnesses
- Reason about timelines
- Operate within a limited action budget
- Submit a final verdict identifying:
- The culprit
- The means
- The supporting evidence chain
Participants are strongly encouraged to use Railtracks, though additional tools, models, and frameworks may be incorporated.
What to Submit
Coming soon...
Prizes
Top Score
Second Score
Third Score
Devpost Achievements
Submitting to this hackathon could earn you:
Judges
Mechanical Scoring
Judging Criteria
-
Correct Culprit
-
Credited Timeline Steps
-
Correct Means
-
Efficiency Bonus
Questions? Email the hackathon manager
Invite others to compete
This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.
