What they probably do is to run ArchR for quality assessment and cell filtering, and then they load the remaining cells into signac. \ Signac (like seurat) does not include any doublet inference tool, while ArchR does: addDoubletScores + filterDoublets

This is how the doublet info is stored inside and ArchR project:

>exp_Integrated@cellColData[,c("DoubletScore","DoubletEnrichment")]
DataFrame with 11477 rows and 2 columns
                               DoubletScore DoubletEnrichment
                                    <array>           <array>
WT#TAGGAGGCAAGGAACC-1       23.423640545554               2.9
WT#GTCGTAAAGATTAGAC-1      11.2245245084216  2.33333333333333
WT#ACGTGGCAGTGGTGTG-1      7.36953484802665               2.1
WT#AGCTGTAGTTTGTCTT-1      16.7685480948753               2.6
...

You can extract the list of UMIs that pass the filtering and use it to load the data in signac.