Score states

A weak link should change the state

Hover a state card, or focus it with the keyboard, to reveal what the state means.

Limitations

What the product should not overclaim

Scores are point-in-time

Quotes, detector signals, and freshness checks can change after a response is generated. A score is evidence for a decision record, not a standing guarantee.

Unsupported coverage stays unsupported

Unsupported tokens, chains, bridges, routes, or wallet assets are not inferred. Routescore should return unsupported, partial, degraded, stale, or unavailable states instead of hiding gaps.

Risk categories stay separate

Execution quality, bridge risk, oracle risk, contract risk, yield risk, and alert risk are different problems. A single score must not hide a separate critical risk.

Calibration needs live evidence

Internal Brier, ECE, and model-quality reports are not public performance claims until enough reviewed labels, live quote samples, and methodology checks exist.

APIs and MCP do not execute

The REST API and MCP server return decision-support evidence. They do not custody assets, route funds, broadcast transactions, or recommend financial actions.

Commercial influence is disclosed

Paid placement, protocol influence, partner-provided data, or score overrides must be disclosed and kept separate from scoring methodology.

Audit trail

Where to verify the contract

Read the methodology — inspect the scoring inputs, weights, signs, caveats, and version history.
Read the risk disclosure — understand where modeled decision support ends and user judgment begins.
Inspect the API and MCP envelope — verify the read-only schemas, coverage states, and machine-readable caveats.
Check source status — confirm which dependencies are current, degraded, stale, or unavailable.