Question: stable cache format for reading per-test-case metric scores

Hi,

I have written a DeepEval adapter for an external tool that binds results to a pre-registration published before the run.

Before depending on it: is the cache file shape stable across minor versions, and is there a better documented route to per-test-case metric scores than reading it directly?

The adapter extracts values and computes nothing; aggregates are computed downstream from per-row values using fixed-precision decimals.

https://gist.github.com/GautamTalksDev/89898dc6c9670beba627efc33faac163