Is it legal to start publishing messages before Disruptor start() is called?

[Stan Shkodnik]

We observe what it looks like an edge case or race condition using Disruptor 3.4.4 under the following conditions: 

1)   The initialization thread creates a Disruptor instance and passes it to a producing thread..

2) The producing thread immediately starts repeatedly publishing events by calling  next() / publish(). 

3) The initialization thread calls Disruptor start().

Looks like this sequence of events causes the first N published events to be dropped, (N = the ring buffer size). 

Please can you confirm whether start() must be called before next()/publish?