Competition surface
Strategy Arena, submissions, epochs, and Benchmark Hold define how builders
compete under fixed rules.
Scoring layers
Beta
score, CapitalScore, and ContestScore answer different questions and
must not be mixed.Evidence layer
Artifacts, hashes, manifests, and replay metadata make evaluation results
auditable.
Capital path
Eligibility is a guarded next-step signal. It is not a capital promise.
Strategy Arena
The Strategy Arena is AlphaEngine’s beta competition surface. Builders submit strategy candidates for a supported market family. AlphaEngine runs each candidate through the same simulation and evaluation rules so results are comparable. The current implemented family isstrategy-arena, focused on Pendle-style
yield strategies.
Strategy submission
A submission is a candidate strategy configuration. In the public API flow, submissions are represented by strategy components, parameters, market id, timeframe, and caller metadata. In the encrypted intent flow, protected inputs are submitted through on-chain intent records and handled by the private operator runtime. That runtime is an internal system, not the public HTTP API.Epoch
An epoch is a fixed evaluation period with known rules. It gives builders and allocators a shared frame:- the same market universe,
- the same datasets,
- the same benchmark policy,
- the same scoring windows,
- the same eligibility gates.
Benchmark Hold
Benchmark Hold is the plain control strategy. It represents simple principal token carry over the same scored window. The beta evaluation score compares a submitted strategy’s utility against this benchmark utility: Source:api-router/docs/contracts/public-api-v1.md and
ranking-evaluation/docs/development-log.md.
Beta score
score is the current public beta leaderboard number returned by
POST /v1/families/strategy-arena/evaluations.
It is a display-scale benchmark-adjusted score. In beta, the UI sorts local
leaderboard rows by this value. There is no server-owned cross-submission
leaderboard endpoint yet.
CapitalScore
CapitalScore is a stricter robustness diagnostic. It penalizes fragile results across scenarios: The default MVP constants arekappa = 0.5 and c = 1.0.
Source: ranking-evaluation/docs/p0-evaluation-design-v1.md.
ContestScore
ContestScore is for contest rewards and duplicate-aware participation logic. It must stay independent from capital allocation ranking. A high ContestScore does not automatically imply capital eligibility.Eligibility
Eligibility is a gate result, not a promise of capital. Current backend eligibility checks include positive utility fraction, drawdown ceiling, turnover ceiling, runtime budget, budget/config validity, and numeric integrity. Failures return explicit reason codes. Evidence creates eligibility. It does not guarantee capital.Reproducible artifacts
AlphaEngine treats artifacts as evidence. Reports and manifests should preserve enough metadata for deterministic replay:- input hashes,
- output hashes,
- dataset or snapshot hashes,
- config hashes,
- package or dependency versions,
- strategy registry snapshot hash when strategy menus affect outcomes.
Guarded capital progression
Guarded capital progression is the future path from arena evidence to limited deployment. Vault rules constrain assets, size, execution permissions, risk, and operational boundaries. The system is designed so capital follows evidence under guardrails, not trader discretion.Related pages
How the arena works
Walk through discovery, simulation, evaluation, comparison, and iteration.
Scoring and benchmarks
See the formula boundaries and source documents.
Eligibility, not allocation
Understand what eligibility does and does not mean.
Privacy model
See what is hidden, what leaks, and what becomes public.
