There is a thread on one of the lab automation forums that is worth reading not for the answer but for the confession at the top of it. An experienced engineer, someone who had been running decks for years, admits that he had misunderstood how correction curves work the entire time. He is not alone, and the misunderstanding is specific enough to be worth naming, because almost everyone who builds a class arrives at it and almost nobody is told they are wrong until a dispense comes out where they did not expect it.
The intuition that feels right and is not
The mistake is to treat a correction curve as a lookup table of fixed outputs. You measure that asking for 100 microliters delivers 99, so you add a point that says 100 maps to 101, and you assume that from now on every time the method asks for 100 the instrument will simply dispense 101. One volume in, one corrected volume out. It reads like a dictionary, and a dictionary is exactly the wrong model.
The engineer in that thread laid out the intuition cleanly with a small table: 0 maps to 0, 100 maps to 101, 500 maps to 600, 1000 maps to 1200. His question was the honest one. If I dispense 100 over and over, do I always get 101? Or does the number shift depending on what the plunger is doing? The person who answered confirmed the doubt directly: adjusting the point at 100 is not how you control the amount delivered at 100 in the way you think.
What the curve actually does
A correction curve is a continuous relationship, not a set of pinned values. The points you enter define a line, or a series of connected segments, and the instrument reads a correction off that line for whatever volume is in play at the moment. Between two points it interpolates, so a request that lands between your measured volumes gets a correction blended from its neighbors rather than a value you typed in directly.
The consequence that trips people up is that the correction applied to a given transfer depends on where that transfer sits on the curve, which is governed by plunger travel, not by the label on the step. The point you set at 100 shapes the segment around 100. It does not stamp a guaranteed 101 onto every step that happens to be nominally 100, because the plunger may be moving through a different part of its range when that step executes.
Multi-dispense makes the error visible
Nowhere does the fixed-output model fall apart faster than in multi-dispensing. Picture aspirating 1000 microliters once and dispensing it as ten steps of 100. Under the dictionary model each of those ten steps is a 100 dispense and should get the 100 correction. Under the real model the plunger starts full and empties progressively, so the first step happens near the top of the curve and the last step happens near the bottom, and each one draws its correction from the part of the curve its plunger position actually touches.
This is why the seasoned advice is to build the curve against the real volumes the tip moves through, not against the aliquot size. If you correct only at 100 because that is your aliquot, you have tuned a single point on a line that the multi-dispense sequence rides all the way from full to empty. The right first, wrong last is the classic symptom, and it comes straight from correcting the label instead of the travel.
How to actually control a dispense
If the curve is not a dictionary, how do you make a specific transfer land on target? You measure across the range and let the curve learn the true relationship, then you let interpolation do its job.
- Measure at several volumes: dispense replicates across your working range and weigh them, so the curve is anchored by real data at more than one place.
- Bracket the volumes you care about: put points on both sides of a critical volume so the segment through it is defined by measurement rather than extrapolation.
- Match the curve to the mode: for single transfers, correct at the volumes you run; for multi-dispense, correct against the descending volumes in the tip.
- Re-measure after you edit: confirm the segment moved the way you intended, because a point you nudge changes the slope of its neighbors too.
A correction curve is a line the plunger rides, not a table it looks up. Correct the travel, not the label, and multi-dispense stops surprising you.