Guides

Writing Good Prompts

How to write clear, unambiguous prompts that maximize your market rating.

What Makes a Good Prompt?

A good prompt clearly defines what information should be extracted from the sources. The resolution agent (an LLM) should be able to read the prompt and know exactly what to look for without any ambiguity.

Key Principles

Be Specific

Use precise language that leaves no room for interpretation.

Bad:"Did the stock go up?"

Good:"What was the closing price of AAPL on March 15, 2026?"

Include Time References

Always specify exact dates, times, and time zones when relevant.

Bad:"Who won the election?"

Good:"Who won the 2024 US Presidential Election according to the official results?"

Define the Output Format

Tell the agent exactly what format the answer should be in.

Bad:"How did the team perform?"

Good:"What was the final score of the Super Bowl? Return as 'TeamA X - Y TeamB'."

Avoid Subjective Terms

Words like "significant", "major", "many", or "best" are ambiguous.

Bad:"Was there significant rainfall?"

Good:"What was the total rainfall in mm for London on March 20, 2026?"

Prompt Patterns

Use these patterns for common market types:

Sports Results

"Who won the [Event Name] on [Date]? Return the winner's full name."

Price Data

"What was the [opening/closing] price of [Asset] on [Date] in [Currency]?"

Election Results

"Who won the [Election Name] according to the official certified results?"

Numerical Data

"What was the [metric] for [entity] in [time period]? Return as a number."

Common Mistakes

>Using relative time references ("tomorrow", "next week")
>Asking multiple questions in one prompt
>Assuming context the agent won't have
>Using abbreviations without definition
>Not specifying units for numerical answers