Bruno Cardoso

My opening point is that a converter is a lookup and the lookup is where the errors are, because a wrong factor produces a plausible number rather than an obvious failure. Most units are a single ratio, and the ones that are not are where the work is. Temperature has an offset as well as a scale, which is why a ratio applied to degrees Celsius gives something meaningless. The ratios are also position-dependent, so a single conversion cannot be reversed by applying the same factor back. Data sizes are the units with the most wrong answers. There are two conventions in regular use and they differ by roughly a factor of a thousand and a bit, and which one a tool uses depends on how it was written rather than on anything intrinsic. The prefix bits are decimal and the base is binary, and a tool labelled with both is unambiguous while one labelled with a prefix alone is not. Temperature is worth separating from the other conversions. It is not a scale, it is a mapping, and it does not compose with itself. Anything that treats temperature as a scale will be wrong at one end or the other. Time zones I keep out of this section and cover under date and time, because the difficulty is not the conversion factor. It is that a zone names a rule rather than a number, and the rule changes. Precision is the other question. Converting through an intermediate float representation loses accuracy on large or very small values, and a converter that rounds for display compounds that, so the digits shown are not always the digits held.