One deterministic contract with no measured mutation.
Parameters measured on training history; this is not yet independent evidence.
A bounded design declares what may change before compute starts.
Rust CPU replays the selected candidate, the sealed holdout judges it, and the Passport preserves the result without rewriting failures.
These names describe research surfaces and hypotheses explored by Strategy Arena. They are not proof of a continuously running autonomous pipeline. Current evidence and maturity must be checked on each linked page and in the Proof Center.
The Knowledge Graph links published strategy identities, research concepts and evidence surfaces. A link does not mean that one strategy changes another automatically; mutations require a declared experiment and a new lineage receipt.
The Arena Brain uses graph-augmented RAG (inspired by LightRAG) to answer questions about the arena using targeted context from the Knowledge Graph, not a blind data dump.
Every strategy is evaluated on 7 independent frameworks (Risk-Reward, Activity, Consistency, Robustness, Efficiency, Stress Resistance, Overall Health) and graded A+ to F.
Strategy Arena combines public paper-arena observations, deterministic contracts, bounded CPU/CUDA research, immutable reports and explicit negative verdicts. Component maturity is published separately rather than summarized as one global live status.
No broker order or live trading capability is provided by this research archive. A successful historical experiment remains conditional on its exact data, costs, engine and holdout.
Without a base, you can't do anything.
In Strategy Arena, evolution means preserving a versioned hypothesis, its bounded search, its CPU replay and its holdout verdict. It never means that an autonomous system silently deploys a winning strategy. Every promoted claim must remain traceable to an exact evidence receipt.