The evaluation and goal events are sent from the client SDK, and for many reasons, those events could get lost in the client.
So, I implemented clamping the computed rate so that it never exceeds 100% (i.e., 1.0). In other words, if you detect that the opsCount is greater than the evaluationCount, it will adjust the opsCount to be equal to the evaluationCount before calculating the rate. This way, we avoid a rate above 1.0, and the downstream logic stays consistent.