On main at 3f23e36, Engine.decide compares total in-flight plus waiting decisions with LOCALJEV_MAX_QUEUE, while README documents that setting as waiting decisions before HTTP 529. With maxInflight=1 and maxQueue=1, a deterministic mock holds the first inference request and the second call is immediately rejected as overloaded, leaving the configured queue slot unused. Expected: allow up to maxQueue waiting decisions in addition to up to maxInflight active decisions. Actual: maxQueue limits total admitted decisions, shrinking usable queue space by the active count. This causes earlier overload responses under load; no interface or default needs to change.
On main at 3f23e36, Engine.decide compares total in-flight plus waiting decisions with LOCALJEV_MAX_QUEUE, while README documents that setting as waiting decisions before HTTP 529. With maxInflight=1 and maxQueue=1, a deterministic mock holds the first inference request and the second call is immediately rejected as overloaded, leaving the configured queue slot unused. Expected: allow up to maxQueue waiting decisions in addition to up to maxInflight active decisions. Actual: maxQueue limits total admitted decisions, shrinking usable queue space by the active count. This causes earlier overload responses under load; no interface or default needs to change.