Rounding errors, repeated enough times, can bring down real systems

Parker explains how seemingly negligible rounding errors, repeated across millions of calculations or compounded over time, can produce consequences wildly disproportionate to the original inaccuracy. He details the Vancouver Stock Exchange index of the early 1980s, recalculated after every trade using a value truncated rather than properly rounded; over thousands of trades a day, this tiny systematic truncation compounded until the index had drifted enormously below where it should have been within months, requiring an embarrassing public correction.

Parker uses this and similar cases to illustrate a broader principle: an error trivially small in isolation can become significant once applied repeatedly and consistently in the same direction, rather than randomly canceling out. This is distinct from random measurement noise, which tends to average out over many trials.

The lesson extends to modern computing and financial systems, where floating-point rounding decisions, often invisible to end users, can silently accumulate into meaningful discrepancies if not carefully managed.

Takeaway: an error too small to notice once can become impossible to ignore after a million repetitions in the same direction.