When a liquid class underperforms, the useful first question is not how bad but in what way. Accuracy and precision are different failures. A class can hit the wrong volume consistently, or scatter around the right one, and the fix for each is different. Confusing the two is how people spend an afternoon adjusting a flow rate that was never the problem.
Two different questions
Accuracy asks whether the average delivered volume matches the target. It is about bias, a systematic offset that pushes every dispense the same way. Precision asks how tightly the individual dispenses cluster together, regardless of whether that cluster sits on the target. It is about scatter, the random variation between replicates. A class can be precise and inaccurate, tightly grouped around the wrong volume, or accurate and imprecise, correct on average but all over the place. You want both, and you diagnose them separately.
How to see each in the data
Weigh a set of replicate dispenses and convert to volume. The mean tells you accuracy: how far it sits from target is the bias. The spread tells you precision, usually reported as a coefficient of variation, the standard deviation divided by the mean. Report both, because a single average hides the scatter and a single scatter number hides the bias.
Which knobs move which
Once you know which failure you have, the parameter to reach for is fairly predictable.
- To fix accuracy, correct the systematic offset: the calibration or correction curve, and over-aspiration to compensate for liquid retained in the tip. These shift the mean without necessarily changing the scatter.
- To fix precision, stabilize the transfer: slower flow rates, longer settling and post-aspiration delays, well-sized air gaps, level tracking, and pre-wetting the tip so the first dispense is not an outlier. These tighten the cluster.
- Watch for parameters that help one and hurt the other, and change one thing at a time so you can see which number moved.
Why it matters downstream
Bias and scatter propagate differently into your results. A systematic volume error skews concentrations and shifts a standard curve, so it biases the answer. Poor precision inflates variance, widens error bars, and can bury a real effect in noise. An assay needs both under control, and knowing which one a class lacks tells you what your data can and cannot yet support.
Chase the right failure. Adjust the calibration for a biased class and the transfer stability for a noisy one, and never assume a single average has told you which you have.