Buffers are boring in the best possible way. You dial in a class once, it behaves the same on Monday as it did in validation, and the only thing that changes between runs is the label on the tube. Biological fluids are the opposite. Whole blood, plasma, and serum are alive with the history of the patient they came from, and that history shows up as viscosity you did not choose, protein that foams and films, and solids that block a tip mid-aspiration. A class written for water will not just deliver the wrong volume from these fluids. It will fail in ways that look like a working run until someone checks the result.
The angle worth taking here is that a good blood or serum class is less about a clever set of speeds and more about respecting how little you control. The sample varies. The surface lies to your sensor. The tube is layered. Your job is to build a class that tolerates that variation where it can and detects it where it cannot, so that a bad sample becomes a flagged sample rather than a silently wrong number.
Every tube is a different liquid
The first thing to internalize is that there is no single viscosity for whole blood. Hematocrit, the fraction of the sample that is packed red cells, swings widely from patient to patient, and it drags the viscosity along with it. A polycythemic sample can be visibly thicker than a normal one, and an anemic sample thinner. Plasma and serum are gentler but still protein-rich, so they sit above water on the scale and vary with the donor. You are not tuning for a liquid. You are tuning for a distribution of liquids, and the class has to hold across the whole spread rather than nailing the mean.
That reframing has a practical consequence. You slow everything down and you leave margin. Aspiration that is gentle enough for the thickest sample you expect will over-perform on the thinnest, and that asymmetry is fine. The reverse is not: a speed tuned for average blood will shear and short-deliver on a thick one, and you will never see it in the numbers because the volume looks plausible. When the sample can be anything within a range, tune for the hard end of the range and accept the mild cost everywhere else.
Protein foams, and the film it leaves behind
Protein is a surfactant, and surfactants foam. Aspirate serum too fast, or blow out hard into a well, and you whip air into the liquid and raise a head of foam that no amount of settling time fully clears on the timescale of a run. Foam is not cosmetic. It sits on top of the liquid as a low-density fake surface, and it wrecks anything that tries to find the real one.
The subtler problem is the film. Protein-laden liquid wets the inside of a tip and clings there as a thin coating that drains slowly, so the tip carries more liquid than the piston displacement says it should and delivers more than you intended. Run after run, that film builds a small positive bias into your volumes, a drift that is easy to blame on calibration when the real cause is the sample coating the walls. Gentle aspiration, an honest settling delay, and a blowout that pushes liquid rather than atomizing it all reduce both the foam and the film. You are trying to move the fluid without beating air into it.
Clots, fibrin, and the tip that blocks
Whole blood clots, and serum tubes throw fibrin strands even when they are not supposed to. A clot does not politely announce itself. It gets drawn toward the tip orifice, lodges there, and the aspiration that was pulling liquid is suddenly pulling almost nothing. If the system has no way to notice, it will proceed to a dispense that delivers a fraction of the target volume into a well that looks, to a camera or a human, exactly like every other well.
This is the strongest argument for pressure-based monitoring during aspiration. A tip drawing liquid at a known speed produces a characteristic pressure signature, and a clot changes that signature the moment it seats. A system watching the pressure trace can abort, flag the sample, and move on rather than delivering short. For clinical and diagnostic work, that detection is not a nicety. A short-delivered patient sample is a wrong result with a name attached to it, and catching it at the tip is far cheaper than catching it three steps downstream, if you catch it at all.
Layered tubes and the immersion problem
Centrifuged samples are stratified on purpose. Plasma or serum sits on top, a gel separator or a buffy coat sits in the middle, and packed cells sit at the bottom. Your target is usually the top layer, and getting it cleanly means putting the tip in the right place and keeping it there as the level drops. Go too deep and you touch the gel or draw up cells, contaminating the aspirate with exactly the material the centrifuge was meant to remove. Stay too shallow and you break the surface, suck air, and foam the sample on the way in.
The technique that makes this work is level following: the tip tracks the falling surface as it aspirates, staying a small, controlled distance below it rather than plunging to a fixed depth. Do that well and you stay in the intended layer through the whole draw. Do it badly, or not at all, and a class that worked on a full tube will dip into the wrong layer as the volume drops on a nearly empty one.
Sensing an unreliable surface
Finding the liquid level is where blood and serum punish the obvious approach. Capacitive level detection works by sensing the change as a conductive liquid touches the tip, and blood is conductive, so in principle it is a natural fit. In practice, foam and film sabotage it. A protein film bridging the tip reads as liquid before the tip reaches the real surface, and a foam head triggers the sensor on air dressed up as fluid. You get a false trigger, the tip stops high, and the aspiration starts in foam or in nothing.
Pressure-based detection is the alternative, sensing the physical resistance of the surface rather than its electrical properties, and it is often less fooled by the film that defeats a capacitive sensor. Neither is a cure on its own. The reliable setups pair careful, foam-avoiding technique with a detection method chosen for the sample, and they treat a level-detection error as a reason to stop rather than a reason to guess.
Carryover is a wrong-result risk, not just contamination
Between patient samples the stakes on carryover change character. A trace of one buffer in another buffer is a purity problem. A trace of one patient in another patient is a clinical error, and no wash protocol is trusted enough to make sample reuse acceptable in that setting. Fresh tips per sample, no reuse across patients, and disciplined containment are the norm precisely because the cost of getting it wrong is a mislabeled result rather than a noisy one. Add to that the plain fact that these are biohazardous fluids, and the containment discipline is doing double duty for the operator and for the data.
With a buffer, a bad transfer gives you a bad number. With a patient sample, a bad transfer gives you a confident number attached to the wrong person.