SMILE classifiers/regressors in Earth Engine

I am working on a project intended for publication that leans heavily on Earth Engine for much of the workflow: assembling, compositing, and mosaicking lots of images from multiple data sources, stacking these together into a multiband raster that can then be fed to a ML algorithm. I have been frustrated by the performance and limited documentation of EE’s in-house classifiers and regressors in the past (especially their old version of Random Forest), and so my workflow up to now has consisted of exporting data prepared in EE for final analysis in Python using sklearn. It works, but it is cumbersome and it adds a lengthy extra step of getting the data out of Earth Engine before the already lengthy step of running a classifier can even get started.

To my delight, Earth Engine recently replaced their old implementations of common machine learning classifiers and regressors with new versions based on SMILE (website, github). Having looked through the SMILE documentation I am impressed, and I’ve tinkered with the tools in EE and they give the results I expect, and fast. However, there is obviously (and unavoidably) some code sitting between the native SMILE algorithms and their implementation in EE’s code editor–arguments are slightly different, the classifiers are called using ee.Classifier, etc.

My question is this: Does anybody know if these algorithms are truly just SMILE wearing Earth Engine clothing to fit in, or has EE altered them in some more fundamental way? In other words, are these "SMILE classifiers" or are they "classifiers based on/adapted from SMILE"?

I want to believe that the replacement of the old classifiers with SMILE will allow me to just do the whole analysis in EE because it would make my life much easier. But I want even more to be confident that the tools I’m using work exactly how I think they work.