What was done
In Epoch AI's MirrorCode results, Claude Opus 4.6 reimplemented gotree, a bioinformatics toolkit of about 16,900 lines of Go, from its executable and tests alone, passing 2,000 of 2,001 tests. Earlier Opus models scored 15 to 63%.
From the source
Where the human stepped in
Not reported yet. If this is your project, tell us what the AI got wrong and how you fixed it.
