Raphael Liu’s Jev skill

The claim

A community skill and Python client for one bounded Jev decision, with offline validation and a separation between model output and authority to act.

Our read

The MIT-licensed standard-library client uses the TypeSafe endpoint, an environment key, a bounded timeout, and no automatic retries. Live calls transmit the supplied evidence. Author benchmarks were not reproduced and do not establish that a cascade preserves host-agent accuracy.

Source reviewed: Primary sources support the stated scope. Not installed or inference-tested by Typed Recipes.

Test status

We haven't reproduced this yet. When we do, measured accuracy, latency, and cost will appear here next to the claimed figures.

Related entries