Call variants across a whole genome on a laptop, in RAM you declare up front, and get results that reproduce byte-for-byte — with a receipt to prove it. Most variant callers spend memory that grows with your data, so "will this finish on my machine?" is something you find out the hard way. Rosalind inverts that: you state a budget, and it tells you *before committing a byte* whether the job fits, then honors that ceiling while it runs. It streams a coordinate-sorted BAM one read at a time, reads the reference from a compact memory-mapped index (no second copy of the genome in RAM), and keeps its working set proportional to *local read depth* rather than file size. Every run prints — and records — the memory it actually used. Where the evidence is too thin to be sure, it **abstains** instead of guessing.
0 commit comments