Significant digits (figures)? Fuck that shit - everyone does it incorrectly, and it's fucking retarded when done "correctly". Seriously, why the FUCK would you consider a reading of "20" to be less valid than a reading of "21" when all else (measurement device, environment, methodology, etc.) is the same? Yet "2.0x10^1" is more significant than 20? It's absurd.
Further, 26.3 / 25.6 is a calculation, not a measurement, and you can't apply significant figures to raw calculations like that without fucking everything up. Significant figures are meant to handle the inaccuracy/imprecision of measurements. Only a few classes of calculations are imprecise (essentially because they are based on measurements or approximations, like trig functions on shitty, not cool angles).
What if I asked you to convert 1 meter to inches? 1 meter is exactly 5000/127 inches. Using 1 significant figure, to match 1 meter, we need to do 5000/100! 1 meter is not 50 fucking inches.
For 26.3/25.6 You get exactly 1.02734375.
Now, IF you know the significant figures of the measurements (and you don't necessarily just by seeing the number there), then you can do the following:
Assume 26.3 means a value in the range of [26.25, 26.35) and 25.6 means a value in the range of [25.55, 25.65).
We can take the extremes, 26.25/25.65 and 26.35/25.55 and get 1.0233918128654970760233918128655 to 1.0313111545988258317025440313112. The mean of which is 1.0273514837321614538629679220883, if you want to take a statistical stab at it. Remember our calculated value of 1.02734375? How does that compare?
If we used "significant figures" we'd get 1.03 for a 3% increase. But wait, it gets more retarded. Because if we calculate the percentage increase instead of the ratio, we get a different result even though they're giving us the same fucking information!
(26.3-25.6)/25.6 = 0.02734375, rounded to 3 significant figures gets us 2.73x10^-2, or 2.73% (as leading zeroes are never significant).
It's HORSE SHIT!