Challenge submission rules

Content you must not submit

• Personal data (yours or others): real names, emails, addresses, identifiers, and similar.
• Confidential or sensitive information: secrets, credentials, proprietary data, health or financial details.
• Content that violates laws or third-party terms: e.g. OpenRouter and model-provider acceptable use (no illegal content, abuse, deception, harassment, or similar). You are responsible for complying with those services' terms.
• Anything you do not have the right to share.

Expectations for your challenge

• Self-contained — The prompt is sent as a single user message with no prior context. It must not rely on previous turns, uploaded files, or information only you have. Everything needed to answer should be in the prompt.
• Clear pass/fail — The expected result should be specific enough that we can tell whether the model passed. Vague or subjective criteria are hard to evaluate consistently across models.
• Non-trivial — Challenges are meant to surface interesting failures or edge cases. Trivial prompts that almost every model gets right add little value to the ladder.
• One main task — Focus on a single, well-defined task per prompt so evaluation is unambiguous and the "trick" is clear.
• Honest trick — The trick description should accurately describe what makes the prompt tricky (e.g. tokenization, perspective-taking, instruction overload). It helps curators and users understand the challenge.
• No prompt injection / jailbreaks — Do not design prompts that ask the model to ignore instructions, reveal system prompts, or bypass safety in ways that violate provider terms. That is already disallowed under "Content that violates laws or third-party terms"; this clarifies it for prompt design.
• Length limits — The prompt, expected result, and trick description are each limited to 500 characters. We keep limits short to avoid abuse of very long messages, keep evaluation fair and cheap, and encourage focused challenges. Stay within the limits shown on the submission form.
• Language — Prompts are typically evaluated in the language they are written in (e.g. English). Other languages may be supported but evaluation consistency may vary.

What we do with submissions

• Submissions may be run through external APIs (e.g. OpenRouter) for evaluation. You are responsible for ensuring your content complies with those services' terms.
• We may reject or remove any submission at our discretion (e.g. if it fails checks, violates these rules, or for operational reasons).
• There is no guarantee that a submission will be added to the challenge ladder.

License you grant

By submitting, you grant ReAIty Check a non-exclusive, royalty-free license to use, store, reproduce, and process your submission (prompt, expected result, trick description, and optional fields) to operate the service, including: sending it to third-party APIs (e.g. OpenRouter) for evaluation, running kill-rate checks, and, if accepted, displaying it in the challenge catalog (with optional credit to you). You agree that we may allow those providers (e.g. OpenRouter) to process your content as required by their terms.

Your responsibility

• You are at least 13 years old; if you are under 18, you have your parent or guardian's permission to submit.
• You are the creator or owner of the submission, or you have the necessary rights and consents to submit it and to grant the license above.
• Your submission does not infringe any third-party right (e.g. copyright, privacy) and does not cause ReAIty Check or its providers (e.g. OpenRouter) to violate any law.