Hey, the calculation behind “AI Share of Committed Code” has been refined over time to make it more accurate, so values for the same period may look different now than what you saw before.
A few things can also affect the numbers:
Whether you’re using the same filters (for example, production branch only vs all branches)
Automated code formatting can break some diff signatures, which affects tracking
The metric is scored on-device, so it depends on commits being analyzed on the same machine where the AI code was written
I have some other questions or requirement, could cursor share the details how to calculate the value?
For example, how to calculate the code merge/rebase,cherry-pick, whether calcluate the mardkown, package-lock.json such kind of “document“ or “configuration“ files.
If a repo has such commit history: A(in cursor)→B (not in Cursor) → C(merge from other branch), how do cursor calculate them?
I noticed cursor agent-trace.dev specification, but have no any further update on it.
Our team AI coverage is about 80%, it is a very high figure in the industry (google 50%,Microsoft 46%), it’s almost unbelievable.I need more data to double check whether it calculates as what I expected