Uses a multi-layer strategy that combines trend, momentum, volume, sentiment, and on-chain data to decide whether to go long, short, or stay out. Every trade must pass a confidence threshold and strict risk controls.
Built from quantitative trading principles and continuously refined through backtesting and market learning
Running a test executes your pipeline against real data and calls the AI — this costs
20 gold.
Edit Cut
The LLM receives this text as a system instruction at this stage of the pipeline.
Players are sorted best → worst. The filter trims from the bottom before injecting into the LLM.
At 0% strength, all player rows are passed to the LLM.
The bottom 0% are removed,
leaving the top 100%.
—
The indicator data is passed as context to the AI — the AI decides how to interpret and use it.
At 0% picks are sorted by signal with no eliminations. Increase to also cut the bottom N%.
Indicators are fetched for tokens surviving at this pipeline stage. Unsupported pairs are skipped silently. Candle interval is selected automatically from the tournament window.
—
At 50% strength, the bottom
50% of tokens are removed.
The top 50% pass to the next stage.
—
Sort cuts reorder tokens by score — highest-scoring tokens move to the top of the list.
No tokens are removed; all pass through to the next stage.
The model used to make the final pick selection and generate the in-character justification.
This node is fixed and cannot be configured. It is always present in the pipeline.
