So here How to achieve dimensional accuracy of printed parts the Cartesian printer approach to accurate sizing is covered. And deltas get a footnote of being a whole other ball of wax - so let's look at that wax ball.
I'm currently resurrecting, or erecting, depending how one looks at it, a poorly documented Chinese printer (purchased in China by a Chinese student and abandoned in the US after graduation) that appears to be based off a Rostock mini. Z 204mm, X Y 100mm radius or 200 mm diameter, Bowden tube.
It appears to be equipped with an absurdly small nozzle (lacking a good way to check that precisely yet, feeding suggests possibly 0.1mm, and no, the vertical resolution is not nearly enough to make that in any way reasonable), which is obviously fixable if the rest of it can be made to work.
Yesterday we got it to the point of (very tediously with tiny nozzle) spitting out a test cube, which was 17 by 17 by 20 mm - undersized in XY, accurate in Z. The last bit makes me pretty sure the steps/mm are right on the steppers, but obviously something is off in the geometry (measured, no documentation for this exact printer can be found, at least by non-Chinese-speaking/reading me - it being utterly un-branded does not help.)
A second test cube was produced with scaling set to 20/17 (1.176) in the XY directions, and that seems to be accurate at least to non-precision measurements. I'm now contemplating "what likely needs tweaked, and in what direction" for the delta geometry - I'd say the rod eye-to-eye measurement (85mm, IIRC) is fairly decent, the offset from rod mount at carriage to rod center is not too terrible, and I have low confidence in the rod mount at printhead to nozzle - so that's the one I suspect most.
tl, dr:But I lack an intuitive understanding of how each of those parameters (offset at carriage, offset at printhead, and arm length) would affect the printed size. I'll come back and edit in what we currently think each is. Rather than depending on scaling in the slicer I'd like to seek the "right numbers" but when at the limits of available measurement precision, having an idea how and in what direction incorrect offsets or arm length affect the print size would be good, rather than just blindly changing numbers and hoping.
I also have some other "print quality" issues and a bed leveling issue which will be other questions when I can sort that out.