Three duplicate-filter rows separate raw reads from the unique count used downstream. Exact arithmetic here means exact results for the stated model inputs; measured inputs still carry uncertainty and significant-figure limits.

highlighted = computed this step

Duplicate filtering is subtraction, not a label

The low row keeps 20 raw reads minus 4 duplicates, so unique reads are 16.

20    4=1620\;-\;4=16
Duplicate filter ledgerRaw read count minus duplicate read count pins the unique-read count.raw=20duplicates=4unique=16accepted=1

Different raw rows can land on the same unique count

Raw 28 with 4 duplicates and raw 30 with 6 duplicates both give 24 unique reads.

28    4=2430    6=2428\;-\;4=24\quad30\;-\;6=24
Duplicate filter ledgerRaw read count minus duplicate read count pins the unique-read count.raw=28duplicates=4unique=24accepted=1

Coverage must use unique reads

The accepted coverage row later uses 24 unique reads, not the raw count 30.

coverage source=2430\text{coverage source}=24\ne30
Duplicate filter ledgerRaw read count minus duplicate read count pins the unique-read count.raw=30duplicates=6unique=24accepted=1