Setting aside the idea of purposely loaded dice....

Statistically, most of us just don't need to worry about the minor bias that arises from the usual manufacturing process for dice. This imbalance is notable when you roll the die a statistically relevant number of times. You can do a Chi-squared analysis on a hundred rolls, for example. For lesser numbers of rolls, you can't tell the bias from just randomness.

Chi² is valid from about 5 rolls per side onward; the more rolls the more accurate, 10 per side is more so... Keeping in mind that confidence increases with the square root of number of trials, and 30 trials is the minimum

It's easy to do... Pick your multiplier. 5 is the minimum for validity, but more is better.

Roll the die a number of times equal to multiplier times sides, recording the number of times it falls on each face.

now, you have a list. For each item, find the difference from the multiplier, then square that.

sum the squares, divide by the number of times rolled.

If the value is under 1, it's fair. If it's over the multiplier, it's most likely notably unfair.

Note that properly, you divide the squares by the expected (which is also our multiplier) then sum, then divide by number of sides, but its easier just to handle that all at the end.