I've tried herro on one of our read sets (with R9.4.1 model) and when I compare the reads before and after correction, by aligning them on the reference assembly, the result is very convincing (very few INDEL errors left.) But when I assemble the reads with hifiasm 0.19.8 I lose close to all reads (95%) during the first hifiasm correction step : comparing first and second kmer histograms in hifiasm logs. This is not the case when I assemble HiFi reads.

Which hifiasm options should I change to limit this phenomenon?