Automate the edges, not just the happy path

Most genomics labs pour effort into automating the happy path. In practice, the bottlenecks are the 5% exceptions: misreads, partial aspirates, bent foils, odd sample volumes. I’ve seen robots idle while humans chase down a bad barcode or a plate that doesn’t quite meet spec.

At Helix, our biggest throughput gains came from engineering the edges. We added simple checks and containment: inline imaging to catch seal issues, redundant ID verification before pooling, a quarantine lane for out-of-range volumes, and clear human-in-the-loop prompts. Fewer re-runs, steadier takt time.

The metric that moved behavior: time-to-containment for exceptions. We track first-pass yield by lot and instrument, and alert on variance rather than absolute thresholds. Stitching LIMS, robot logs, and QC into one timeline made root cause blindingly obvious.

What exception signals or controls paid off most in your lab? If you had to pick one add-on to tame off-nominal events, what would it be?