Some pipetting problems look like accuracy problems and are really physics problems wearing an accuracy costume. The correction curve is such a good hammer that everything starts to look like a nail it can drive, and nowhere is that more misleading than at the bottom of the volume range. A recurring forum story makes the point better than any spec sheet: someone trying to place a few microliters into empty wells, watching most of the dispenses simply fail, and reaching for calibration when the answer was never in the curve at all.
The story worth learning from
The specific case, from a thread on one of the automation forums, was 3 microliters of reagent into empty wells of a PCR plate using 10 microliter tips. The success rate was around one in ten. The person had been optimizing, adjusting, and by any reading treating it as something a better class would fix. It was not. The volume was arriving unreliably because of what happens physically when a tiny droplet meets a dry surface it does not want to leave the tip for, and no correction point changes that. You cannot correct a transfer that does not consistently happen in the first place.
Why the curve is powerless here
A correction curve adjusts how much the instrument aims for, assuming the transfer occurs and the error is a steady bias. At capillary scale into an empty well, neither assumption holds. The droplet is small enough that surface tension and the tip geometry dominate, so the failure is not that you delivered the wrong volume, it is that the liquid stayed in the tip, or broke up, or beaded on the wall. That is variability and outright miss, not bias, and the curve has no grip on it. Pointing calibration at it is the classic wrong tool: the number you are adjusting is not the number that is failing.
The fixes that actually worked
What turned that thread around was physical, not numerical, and the moves generalize to almost any low-volume struggle.
- Tip size: the single biggest change was moving from 10 microliter tips to 50 microliter tips. The larger tip gave more consistent behavior at the same small volume, which sounds backward until you remember the geometry near the tip opening matters more than nominal capacity.
- Taught geometry: teach the instrument the true bottom of the well using the actual head and the actual tips, and start the dispense close to it, around a millimeter, so the droplet has a surface to touch rather than a fall to survive.
- Blowout: enlarge the blowout so the tip is actively cleared at the end, which let the same setup reach reliable single-microliter dispenses that were impossible before.
- Pre-fill where you can: dispensing into liquid already in the well, rather than a dry surface, gives the droplet something to merge with, so pre-adding buffer is a legitimate fix when the protocol allows it.
None of these is a correction point. All of them change whether and how the liquid leaves the tip, which is the thing that was actually broken.
A rule for reaching for the curve
The general lesson outlives the specific numbers. Before you touch a correction curve, ask whether your problem is a wrong amount or an unreliable event. If the same request sometimes works and sometimes does not, you have an event problem, and events are fixed with tips, geometry, heights, and blowout. Only once the transfer happens the same way every time, so the error settles into a steady bias, does the curve become the right tool. Correct a reliable transfer, never an unreliable one.
No correction curve can fix a transfer that does not reliably happen. At low volumes into empty wells, change the tip, the taught height, and the blowout first, and correct only what is already repeatable.