Editing Template:Convert/doc (section)

===Rounding temperatures: ''°C'', ''°F'' and ''K''<span class="anchor" id="Rounding temperatures"></span>===
In temperatures, the conversion will be rounded either to the precision comparable to that of the input value or to that which would give three significant figures when expressed in kelvins, whichever is more precise.

#{{#invoke:DemoTemplate|convert|10,000|C|F K}}
#{{#invoke:DemoTemplate|convert|10,000.1|C|F K}}
#{{#invoke:DemoTemplate|convert|-272|C|F K}}
#{{#invoke:DemoTemplate|convert|-272|C}}
#{{#invoke:DemoTemplate|convert|100|C|F K}}
#{{#invoke:DemoTemplate|convert|0|C|F K}}

The precision of the input number in example (1) is one digit, but the precision of its kelvins expression is three, so the precision of the Fahrenheit conversion is ''made'' three (made 180...) . (1) and (2) seem to belie the fact that a {{convert|0.1|C-change|F-change|disp=x| is a |abbr=off}}, and make the 32 degrees difference shown in (1) begin to seem off somehow.
Result (1) seems off until you set the [[significant figures]] yourself with {{para|sigfig}}:
:{{#invoke:DemoTemplate|convert|10000|C|sigfig=5}}
or you set the precision ''positionally'', relative to the decimal point (zero being ''at'' the decimal point):
:{{#invoke:DemoTemplate|convert|10000|C|0}}

The precision of the input number in example (2) is six, so the precision of the Fahrenheit output is six now, whereas before kelvins had determined it to be three. Examples (3) and (4) show how this can be hidden and generate questions, but it occurs there because the kelvins conversion generated the fractional parts. (Before it was the input number that generated the fractional part.)  In example (3) the three input digits converted into five significant output digits because of the two digits after the decimal point, generated by the kelvins conversion. This happened again in (5), but in (6) decimal fractions were neither given as input nor induced by the kelvins conversion.