verified sessions next to your aether benchmark (free)
hey Siddhartha, saw aether-coder and that you built your own benchmark for it. people who build benchmarks actually measure things, so I'll pitch data with the check included.
66 recorded sessions of an AI fixing real bugs in real repos. every command ran for real, errors and retries kept, tests passing at the end, and each session is scored against the actual human fix. verify.py re-checks hashes and counts offline, so you can audit before you train.
honest note: my sessions are Python library work, not Aether. but the process layer is the same everywhere: explore, plan, edit, run, fail, fix. built to improve coding, tool calling, agentic execution, and recovery from failure.
it's for continued pretraining, not SFT. the first 100 cost nothing, so all 66 are free right now:
https://github.com/Sumama-Jameel/megabit-trajectories
if it ever runs next to your benchmark I'd like the result, either direction.
Sumama