Buffer

A buffer is a stock of spare resources placed where variance threatens output — in a factory, extra parts kept in front of the bottleneck workstation so it never starves when an upstream station has a slow hour or breaks down. Because statistical fluctuations combine with dependent events, a balanced plant cannot absorb variance; the bottleneck idles and that lost time is lost throughput forever. So Goldratt’s rule: buffers protect the constraint, while non-constraints get excess capacity. The two are interchangeable to a degree — more upstream excess capacity lets you run a smaller buffer, and a bigger buffer demands less excess capacity — so you choose a mix by comparing their costs against the value of protected output. Notably, only excess capacity can replenish a buffer after a bad streak; a balanced plant can only wait on lucky positive variance, which bad luck keeps cancelling.

CF adds an epistemological point: a buffer presupposes that error correction is happening at a lower level. When a machine breaks (error) it gets repaired (correction), and that correction itself consumes time, parts, and effort. Without spare resources, any effort spent correcting one problem would starve something else — so slack is what lets error correction run without cascading disruption. This is why CF treats a buffer as kin to a margin of error: deliberate slack that keeps a good-enough system stable under variance, rather than waste to be trimmed. In project management (drum-buffer-rope, critical chain), CF endorses aggregating buffers globally — one buffer for the whole project beats padding every task — because aggregated fluctuations even out.