Dead volume, and the sample that contaminates the next one
Where liquid sits when it should have left, and what it costs.
Dead volume is liquid sitting inside a manifold in a cavity, corner or side branch that the main flow does not actually pass through on its way from inlet to outlet, and because that liquid is never swept by the bulk of the flow it lingers there long after a fresh sample has supposedly replaced it, bleeding a trace of the old sample into every one that follows.
The puddle below the teapot spout
Pouring from a teapot, tipped as far as anyone dares, always leaves a shallow puddle below the level of the spout, since pouring cannot lift liquid that sits lower than the opening it must leave through. That puddle looks negligible until a different tea is brewed in the same pot, when a trace of the previous brew turns up in the first cup. A quick rinse poured out the same way leaves the same puddle behind, which is why anyone who cares about the next brew rinses with more water, or tips the pot from a different angle. A manifold's dead volume behaves the same way: too small to see, sitting below the reach of whatever flush runs between samples, and reappearing as a faint trace of the previous sample once a sensitive enough measurement looks for it.
In a manifold, the flow takes the path of least resistance, usually the shortest connection between inlet and outlet. A feature off to the side of that path (a small chamber housing a valve seat, a shallow recess left by how a mould had to be built, a branch that only occasionally carries flow) receives little of the bulk current. Liquid in those spaces is replaced slowly, by diffusion and whatever weak secondary current the geometry allows, so a real volume of the previous sample can still be there after the main flow has moved on.
A trace that halves and halves again
Old liquid trapped in a side cavity does not empty in one clean exchange. It bleeds out across several cycles, each carrying away a share of what remains, so the contamination fades with each sample instead of ending sharply. Suppose a cavity gives up half its old contents on every cycle. After one cycle half the original trace remains, after five cycles about 3 percent, and after ten cycles about 0.1 percent, still present and still measurable by a sensitive enough method. A rinse procedure that assumes carryover is gone after one or two flushes is betting against that arithmetic. The trail is also awkward to catch: a single obviously contaminated result invites investigation, while a thin, diminishing trace spread across the next several results leaves each one looking close enough to correct.
Working with the decay, then designing it out
Where a cavity cannot be redesigned away, the practical response is to work with the slow decay. That means running a deliberately oversized flush of several channel volumes between samples, and ordering a run so that a sample expected to carry only a trace of something runs before a sample known to carry a great deal of it, so that any carryover moves in the direction that does least damage. Both are cheaper than a full manifold redesign once a design is built and in service.
Dead volume stops mattering only where every feature sits directly in the swept path of the main flow, with nothing branching off it that does not also carry a comparable share of that flow. Achieving that usually means moving a feature, for instance putting a valve seat directly in the main channel, since extra flush volume through a channel that never reached a given pocket will never finish the job a change of geometry can. A manifold with no dead volume left in it is the finished version of the simple routed path that started this set, with every side feature either brought into the swept flow or removed.