← Back to Archive

Part count is a reliability number

Why every additional component multiplies the ways a thing can fail.

Every additional component in an assembly is a fresh, independent chance for something to go wrong, and because an assembly's overall reliability is the combined product of every individual part's own reliability rather than a simple average of them, adding parts drags the whole system's reliability down even when every single part being added is, on its own, entirely dependable.

What is really going on

An assembly works only if every one of its components works, which means overall reliability is not decided by the average part's quality, it is decided by multiplying every part's individual chance of working correctly together, since the whole system fails the moment any single one of them does not. That distinction between averaging and multiplying is the entire subject of this article, and it is easy to miss because everyday language about reliability leans so heavily on the former. A component that is individually reliable the overwhelming majority of the time still contributes a small, nonzero chance of being the one that fails on any given unit, and that small chance compounds directly with every other component's own small chance the moment it is added to the same assembly, an effect invisible when looking at any one part alone and unavoidable once the full parts count is actually multiplied out.

The string-of-fairy-lights comparison

An old-style string of fairy lights wired so that every bulb depends on every other bulb in the same circuit will go dark entirely the moment even one single bulb fails, regardless of how many dozens of other bulbs on the same string are still working perfectly well. Doubling the length of that string by adding twice as many bulbs roughly doubles the number of individual chances for a single failed bulb to plunge the whole string dark, rather than doubling the light's usefulness for free, since each additional bulb is one more independent opportunity for the same kind of failure to strike. An assembly built from many components is wired together in exactly this same all-or-nothing sense functionally, even though its parts are mechanical rather than electrical, since the assembly as a whole works only for as long as every one of its individual pieces keeps working, and every part added lengthens that same vulnerable string by one more bulb. Nobody stringing lights thinks of an extra bulb as a free improvement, and nobody adding a component to an assembly should think of it that way either. Modern LED sets are sometimes wired in small parallel groups specifically to avoid this, a handful of bulbs bundled together so that one failure only darkens its own short segment rather than the entire string, a real design choice that trades a little extra wiring complexity for exactly the kind of protection the old series-wired string never had.

Why even a very reliable part still drags the total down

A component that works correctly ninety-nine times out of a hundred sounds, described in isolation, like an entirely trustworthy part, and it is, taken alone. Combine a hundred such components into a single assembly, each contributing that same small, independent chance of being the failure, and the overall assembly's reliability drops to a meaningfully lower figure than any individual part's own ninety-nine percent would suggest, since the chance that all hundred happen to work correctly together on the same unit is considerably smaller than the chance that any one of them does. This is not a flaw in any individual component, it is simple multiplication applied honestly across every part actually present, and it is the exact reason a genuinely complex assembly with many components can struggle to reach the same overall reliability a much simpler assembly achieves almost by default. That same multiplication does not stop at the level of a single assembly either, since a product built from several such assemblies wired together in the identical all-or-nothing way multiplies their already-reduced reliabilities against each other in turn, which is why a genuinely complex product, several subassemblies each already carrying its own internal part count, can end up considerably less reliable overall than any single subassembly's own figure would suggest.

The number that matters here

An assembly built from a hundred components each individually reliable ninety-nine times out of a hundred works correctly, overall, well under nine times out of ten once every component's own small failure chance is multiplied together honestly, a sobering gap between how trustworthy each individual part looks in isolation and how trustworthy the finished assembly built from a hundred of them actually turns out to be.

Where redundancy changes the arithmetic

Not every additional component drags reliability down, since a part added deliberately as a genuine backup, one of two pumps either of which alone can keep a system running, changes the arithmetic in the opposite direction, because now the system only fails if both fail together, a considerably rarer event than either failing alone. This is a completely different situation from the ordinary case this article has been describing, where every added part sits in series with the rest of the assembly and the whole depends on every single one, and confusing the two is a real and costly mistake, adding a component in the hope of improving reliability when its actual wiring into the system, functionally in series rather than genuinely in parallel as a backup, means it can only ever make the total worse. Genuine redundancy is not a free upgrade either, since a second pump still has to be bought, installed, plumbed and maintained, the same accumulated cost the previous article in this set already described, so the decision to add a deliberate backup is a real trade, buying a specific, quantifiable improvement in reliability at a specific, real cost rather than simply adding a part and hoping it helps.

Why this matters in practice

Improving reliability by making individual components more dependable is a real, valid strategy, but it fights against diminishing returns the moment part count itself is left unaddressed, since a hundred components each improved to a slightly higher individual reliability still has to survive the same multiplication across all hundred of them. Reducing the number of components in the first place attacks the problem from the other, often more effective direction, shrinking the total number of independent chances for failure rather than only trying to make each individual chance smaller, which is exactly why part-count reduction shows up again and again as a genuine reliability strategy in its own right, not merely as a cost-saving measure that happens to have a pleasant side effect. The two strategies are not in competition either, a design that both removes unnecessary components and improves the reliability of the ones that genuinely remain is simply attacking the same multiplication from both directions at once.

More on Deleting parts