Error rate 100x higher than it should be with 76-bp error model?

I'm getting that Flux Simulator's substitution rate across all read bases is about 2.5%, far higher than the Phred score plot at http://sammeth.net/confluence/display/SIM/4.5.4+-+Error+Models suggests. My strong suspicion is that somewhere critical in the code, a factor of 100 is missing because of some confusion between working with percentages and not. I already see one place where this confusion occurs: http://sammeth.net/fisheye/browse/barna/barna.simulator/src/main/java/barna/flux/simulator/Sequencer.java, line 301: the variable avg is a proportion of bases that are mutated, not a percentage as claimed by the log.

I have confirmed for a couple of simulations that the mean error rate computed from quality scores agrees with the substitution rate.