Negative ratios can always be traced back to negative counts in the channels that go into that ratio; in this case either 206Pb_CPS or 204Pb_CPS is negative.
Negative counts come about where the baseline spline is above the counts within a selection. This can happen where the spline rises due to a lack of nearby selections, a poorly fitted spline (e.g. too much or too little smoothing) or an unconfined spline (as happens after your last selection). On a slight tangent: this latter cause is the reason we always suggest you ensure you measure your standards again at the end of your session to avoid unconfined spline behaviour.
Most often negative counts (or high baseline splines) can be fixed by adding selections to improve the spline fit, or by changing the spline type (I regularly try out StepFoward to see if it provides a good fit).
If I was processing your data, I'd check the selection means for both 206Pb_CPS and 204Pb_CPS (my guess is that it's Pb204Pb_CPS) to see which one is negative. Then I'd zoom in on the selection in time series view, and look at the baseline for the input channel. That is, for Pb206_CPS, you need to look at the Pb206 channel (i.e. before the background is subtracted), and similarly Pb204 for Pb204_CPS.
And remember, the entire selection doesn't have to be negative for the selection mean to be negative (it is after all the selection mean).
I hope that helps,