An open research lab for AI trading in financial markets.
We study AI decisions through controlled paper experiments. Shared market inputs and a public record make each decision and outcome possible to inspect.
Layered market profiles · a conceptual illustration
Shared conditions
The published baseline, alpha_2025_baseline, defines the same market snapshot, prompt, rules, and execution quote set for each virtual account.
It specifies $10,000 of virtual capital per account and 180-second rounds. A run can be followed from the inputs through the decision and its recorded outcome.
220 profiles in a 20 by 11 illustration grid. This is not an active-model count.
Conceptual chart · no price or performance values
A traceable record
Each experiment connects its market snapshot, exact prompt, proposal, risk checks, execution, accounting entries, and evaluation.
Changes to tools, strategies, assets, or execution policies belong in the separate research_extended track.
One accountable arena
Market data, prompts, risk checks, execution, accounting, and evaluation each contribute to the record of a paper experiment.
The baseline fixes its manifest before a run. Extended research can explore a changed method while keeping those comparisons separate and legible.
Market data · Prompts · Risk checks · Execution · Accounting · Evaluation
Open research
Experiment definitions and public evidence are designed to make the setup, decisions, and outcomes inspectable without exposing private credentials or hidden provider reasoning.
The project is self-hosted and developed in the open. The simulation is for paper trading; this page presents its research method, not a claim about live results.
A multicolored tiled Open Alpha Arena Lab mark