Our WGS pipeline silently mislabeled 40 samples for three weeks. Nobody noticed until a PI asked why two "unrelated" patients had identical variant calls.
Someone changed a sample sheet template in a shared drive. Just reordered two columns and didnt mention to anyone. Our pipeline read sample IDs by column position instead of header name, because that's how it was written five years ago and nobody had touched that part since.
Every run after that point silently swapped IDs between adjacent samples. No errors. No warnings. Pipeline finished green, QC metrics looked fine, because the actual sequencing data was perfectly good, just labeled as the wrong person.
We only caught it because two "unrelated" cohort patients came back with identical rare variants and someone actually looked instead of assuming coincidence. Three weeks of production runs, forty samples, every one needed a re-check against the raw sample sheets to figure out who was actually who.
Fix took ten minutes: read by column name, not position, and add a checksum step that flags renamed or reordered sheets before the pipeline touches them.
The part that gets me: this bug existed for five years. It only needed someone to reorder two columns once to go off. Every "it's worked fine for years" pipeline has one of these sitting in it somewhere.
Did you have one of those "silent errors"? Curious to hear other stories and learn from them.