The moment a correction curve makes the average land on target feels like the finish line. It is not. A curve that fits the run you built it from can still be wrong on the next run, because fitting always flatters the data it was fitted to. The step that separates a class you can trust from one you merely hope about is the confirmation run, and the published calibration work treats it as non-negotiable: screen, adjust, and then confirm, with the confirmation carrying more weight than the fit.
Why fitting is not confirming
When you measure a few volumes and set a curve to match, you have proven the curve describes those measurements. That is circular. Some of what the curve absorbed was the real relationship and some was the noise of that particular run, and only a fresh measurement can tell you how much was which. This is why the calibration study confirmed its adjustments with a separate run and, tellingly, with more replicates than the fit used. The fit can be quick and rough. The confirmation has to be clean enough to believe, because it is the number you will actually rely on.
What a good confirmation looks like
A confirmation is not just re-running the same dispense. It is deliberately set up to catch the ways a curve fools you.
- Fresh setup: re-teach or re-mount rather than continuing the exact session you tuned in, so you are not just re-measuring one lucky configuration.
- More replicates: take more shots per volume than you did while fitting, so the confirmation averages down the noise the fit may have chased.
- Across the range: verify at several volumes, not only your most common one, because a curve can be true at the point you fitted hardest and drift between points.
- A fit-quality bar: hold the calibration line to a real standard, a coefficient of determination above about 0.99, so you are confirming a clean relationship and not blessing a scatter of points that happen to average out.
If the confirmation sits inside your method's tolerance for both trueness and spread across the volumes you run, the curve has earned trust. If it does not, the fit found noise, and you go back a step rather than shipping it.
The technique that cancels your curve
There is a subtle trap that a confirmation run will expose if you let it, and miss if you are careless. Pre-wetting the tip changes how much liquid the first transfer holds back, and Hamilton's guidance notes that pre-wetting can effectively nullify the correction curve, making transfers consistent in a way that overlaps with what the curve was compensating for. The danger is a mismatch between how you tuned and how you run. Fit a curve with no pre-wet and then enable pre-wetting in production, and you have changed the very shortfall the curve was built to cancel, so a curve that verified perfectly now over-delivers.
The rule is simple: verify under the exact conditions you will run. Same pre-wet setting, same tips, same labware, same temperature. A confirmation run that differs from production in any of these is confirming a class you are not going to use.
Record the confirmation, not just the curve
A confirmation is evidence, and evidence only counts if it survives. Write down the measured trueness and spread, the volumes tested, the tip, the labware, the temperature, and the conditions including whether pre-wetting was on. Months later, when a result looks off, that record is what lets you ask whether the class drifted or the run changed, instead of re-tuning from a blank slate on a hunch.
Fitting a curve proves it matches the data you fitted. A confirmation run on a fresh setup, under the conditions you will actually run, proves it will hold. Only the second one is worth trusting.