Why two identical moulds make different parts

Cavity-to-cavity variation, and why it matters for tolerance.

Two identical moulds, or two identical cavities within the same larger tool, make slightly different parts because no two cavities are ever machined to genuinely identical dimensions, and no two positions within the same tool receive molten material at exactly the same temperature, pressure and timing, so small differences accumulate between them even when every input to the process was, as far as anyone could measure, held the same.

Cupcakes from the corner of the tray

Baking a tray of cupcakes from a single bowl of batter, in a single oven, at a single set temperature, still produces cupcakes that are not quite identical, and the differences are rarely random. The ones in the corners of the tray often come out a little more browned or a touch smaller than the ones in the middle, because their position exposes them to slightly different heat, the corners running hotter or catching more moving air than the sheltered middle. Nobody treated them differently. Every cupcake came from the same batter and the same bake, yet position alone produced a small, consistent difference between one part of the tray and another, and the same thing happens between the cavities of a production tool.

Where the differences between cavities come from

A production mould with several cavities is machined so that each one is meant to produce an identical part, and modern machining gets them extremely close, though never perfectly so. The cutter that machined one cavity was in a very slightly different state of wear by the time it machined the next, and the equipment used to measure each cavity carries its own small margin of uncertainty.

Material does not arrive at every position identically either. The runner system that carries molten plastic or metal to each cavity has its own geometry, and a cavity sitting further from the point where material enters the tool tends to fill slightly later, slightly cooler, or under slightly lower pressure than one sitting closer, even though both were fed from the same shot at the same moment. Each difference is small, and they do not cancel out as random noise would. A cavity that sits further from the feed point runs slightly cooler on every single shot, producing a small, repeatable bias between its parts and its neighbours' parts, like the corner cupcakes that come out browner every time the tray goes in.

Sampling that never reaches the bad cavity

Because this variation is systematic, a quality check that measures a few parts pulled from wherever is convenient can miss it entirely. On an eight-cavity tool, five parts picked at random have roughly an even chance (seven eighths multiplied by itself five times, a little over one half) of including nothing at all from one particular cavity, so a single cavity drifting outside tolerance on every shot can slip through the sampling as often as it is caught. A proper check records which cavity each measured part came from, so that a bias specific to one position shows up as a pattern instead of being averaged invisibly into a total that looks perfectly healthy.

This is also why a production tolerance has to allow for more than the process's average behaviour. A tolerance that only reflects the best-performing cavity will be violated repeatedly by a worse one elsewhere in the same tool, and a tool that looks well within specification on average can still be steadily producing out-of-tolerance parts from its weakest cavity, one part in every eight if that cavity is the only offender.

Two sources of error, kept apart

A properly conducted tolerance analysis treats cavity-to-cavity variation as its own separate source of error, stacked on top of the variation any single cavity shows from shot to shot, instead of folding both into one combined number that hides which is responsible when a part turns up out of specification. A tolerance set by measuring a handful of parts from one favourably placed cavity during early trials can look comfortably generous right up until full production starts feeding parts from every cavity into the same inspection stream. At that point the tolerance that felt safe starts being broken by parts from whichever cavity was never sampled during those trials.

Single cavities and forgiving features

Cavity-to-cavity bias matters far less on a single-cavity tool, where every part comes from the same physical cavity and the same feed geometry every time, leaving shot-to-shot process variation as the only real source of scatter. It also matters less for features that are insensitive to small dimensional shifts, a cosmetic surface with no mating requirement, for instance, where the variation cavity position introduces sits well within what the part can tolerate and nobody ever needs to trace it back to its source.