I put this together after a playgroup.org.uk session. This obviously isn't a valid prize submission, but I was interested in testing what was possible using a SOTA harness and model (CC + Opus 4.7) before trying smaller models. It's great to see that the constraints introduced appear to have worked well.
Interested in critiques + in case anyone spots leakage that could still be hiding or proposals for what a cleaner eval might look like.
I put this together after a playgroup.org.uk session. This obviously isn't a valid prize submission, but I was interested in testing what was possible using a SOTA harness and model (CC + Opus 4.7) before trying smaller models. It's great to see that the constraints introduced appear to have worked well.
Interested in critiques + in case anyone spots leakage that could still be hiding or proposals for what a cleaner eval might look like.