That is a good question! Unlike mProphet for DIA, there is not really a peak-level q value that can be applied to limit error at the signal extraction level. I have found that curious, but I think it is reasonably "state-of-the-art".
I believe that usually in label-free MS1 extraction MS/MS spectrum match statistics are used to decide on a set of MS/MS spectra believed to have an acceptable error rate. Then these are used with retention time alignment to extract signal from the MS1, and that signal is assumed to also have an acceptable (though unquantified) error rate. Which is a somewhat problematic assumption, I think.
Often, second-level filters, such as minimum signal or 2 peptides per protein, are applied to the signal extraction data to reduce error rates in the protein fold-change results which are often the end goal of the MS1 extraction data. But there is not a simple statistical cut-off value to control error rate in this transition between MS/MS to peptide matching and signal extraction.
In practice, however, good error control in peptide-spectrum matching, signal filtering, and criteria like minimum 2 peptides per protein help produce protein fold-change results with reasonable error rates, since you also get some control of error rate again in the fold-change calculation with adjusted p values for that test. And if you use a minimum 2 peptides per protein then interference and noise are much less likely to result in a statistically significant fold-change at the protein level.
We explored this a little in a paper on a recent ABRF iPRG study here:
http://pubs.acs.org/doi/abs/10.1021/acs.jproteome.6b00881
Hope this helps. Sorry I don't have a better story for you.
--Brendan