@mykola_kondratiuk This matches all differences in the conversation generated by the agent against the committed code, so it will be 90% counted by the agent. Presumably, a modern workflow where the developer writes an entire description and the agent writes 90% of the code would still be more efficient and faster than traditional coding.
This is not meant to judge good or bad, but rather a metric to track and compare across all developers in the team, so developers can learn from each other.
Along with ratings, you can also track which models work best in each project/language/genre. I think this would be more valuable with a future team server that aggregates data for the entire team of developers.
<a href