Ask anyone what a kilogram is in pounds and you will hear 2.2. It is a fine answer for a bag of flour and a wrong one for anything that gets multiplied. The real figure is 2.20462262, because a pound is defined as exactly 0.45359237 kilograms. A definition fixed by international agreement in 1959 and unchanged since.
The error in 2.2 is 0.2097%, which sounds like nothing and behaves like a percentage. It scales with whatever you multiply it by and it always points the same way. Convert 50 kg and you are 0.23 lb light. Convert a tonne and you are 4.6 lb light. Convert a 24-tonne container load and you book freight for 52,800 lb instead of 52,911 lb. An error of 111 pounds, an entire passenger worth of weight that exists on the scale and not on the paperwork.
The direction matters as much as the size. 2.2 is always low, so the error never cancels itself out; across a manifest of forty pallets it points the same way forty times.
More decimal places is not the fix
The instinctive repair is a longer short factor, and it can make things worse. 2.205 looks more careful than 2.2 and it overshoots: the same 24 tonnes comes out at 52,920 lb, nine pounds high where 2.2 was 111 pounds low. 2.2046 undershoots by half a pound. Each truncation is its own distinct wrong answer, and which one you get depends on where somebody happened to stop typing.
The reciprocal is worse and far more common. Going the other way, people reach for 0.45 to turn pounds into kilograms. That is 0.79% off: nearly four times the error of 2.2, because 0.45359237 cut to two decimal places loses proportionally more than 2.20462262 cut to one. The short factor that feels safest is the one carrying the largest error.
There is a defensible answer to how many digits are enough, and it depends on the largest input a tool will ever meet rather than on taste. Roughly speaking each decimal place in the factor buys one more digit of accuracy in the result: 2.2 is good to three significant figures, 2.2046 to five, 2.20462262 to nine. For a customs declaration rounded to whole pounds, five is more than plenty. The reason the full figure gets carried anyway is that the tool has no idea which of those two jobs it is doing, and an exact definition costs nothing to hold on to.
Round last, always
The rule every converter here follows: carry the full-precision factor through the arithmetic and round once, at display time, to the precision the reader asked for. Rounding the factor first bakes the error into every result that will ever come out of it.
This is a rule rather than a preference because the two operations do not commute. Rounding and then multiplying is not the same as multiplying and then rounding, and the gap between them grows with the multiplier, which is precisely the situation a converter is in, because the multiplier is whatever the visitor typed and it could be anything.
It compounds when conversions chain. Kilograms to pounds to stone is two multiplications; round after each and the second one scales the first one’s error along with the value. Any route through an intermediate unit does the same: square metres to square feet by way of metres to feet, or litres per 100 km to miles per gallon, which passes through three separate definitions and is where most fuel-economy converters quietly stop agreeing with each other. The rounding note covers what happens when the same value is rounded at two stages of a calculation.
Why the definitions are exact in the first place
None of this would be worth arguing about if a pound were an approximation. It is not, and neither are its relatives: an inch is exactly 2.54 cm, a mile is exactly 1,609.344 m, a US gallon is exactly 3.785411784 litres. Every one of those is a definition rather than a measurement, agreed once so that the two systems hold a fixed relationship instead of a drifting one.
The 1959 agreement is where that came from. Before it the American and the imperial inch were defined by slightly different routes and disagreed in the sixth decimal place. Invisible for carpentry and awkward for precision engineering shipped across the Atlantic. The agreement picked one value and both sides adopted it.
What it left behind is the instructive part. The US survey foot, used for land surveying, stayed defined as 1200/3937 of a metre rather than 0.3048 — a difference of two parts per million. That is 20 centimetres over 100 kilometres, which is nothing on a garden fence and a boundary dispute across a county. It was finally retired at the end of 2022, and the reason it outlived by six decades the agreement that replaced it is that redrawing every land record is harder than living with two slightly different feet.
Five decimal places on the reference table
This is why the conversion pages print five. Not because anyone weighs flour that precisely, but so that the number you copy is the number the definition gives, and any rounding after that point is yours, made once, deliberately, at a precision you chose for a reason.
It is also why a converter should show the factor and not only the answer. An answer you cannot reconstruct is one you have to trust. A factor you can check against NIST is one you can verify in the time it takes to open a second tab, which is the difference between a tool and a black box.