Firts of all: Great project!

For the calibration of the ELO Levels I see two main problems. First, you need a lot of games to estimate the ELO rating. I'd say 20 games to estimate an ELO with deviance of +-100 which is still a lot. Because you want to test not only one rating, one opponent and one setup the best way would be to do the testing automated, either by implementing a UCI or xboard interface to GUI or engine.

The second problem is that the ELO values on different websites and of different Engines with adjustable strength don't match each other at all. If you want to calibrate to human or FIDE ELO, IMHO the ELO of the dedicated units mentioned on this page match best. You can use the engines of CB-Emu for calibration. If you implement uci or xboard interface to the engine you can do this automated within Arena.

I did a lot of testing in the past. The ELO levels of Shredder and HIARCS are well calibrated. Both only reduce the calculated nodes and do not blunder on purpose. Here are the tests: https://open-chess.org/viewtopic.php?p=17735#p17735