> I don't think it has the ability to select arbitrary edges and apply such features (I could be wrong).
That's actually how it works. Chamfers and fillets are handled by OpenCascade, so they can be applied to arbitrary edges. They only exist on the solid model though, so they are not integrated with the solver.
I'd be curious to see how native chamfers/fillets in solvespace would work, I imagine some amazing things could be done if they'd be part of the solver.
I've seen commercial products that use socketed devboards inside. If anything, it's just an indication that the hobbyist and professional spaces are slowly converging.
Please be aware that this is not a first-party open source release of a previously secret internal spec/implementation, but rather the result of an impressive reverse-engineering effort by Christian Nöding, whose videos about this project have been posted on here as well. Still, minor kudos to Behringer for giving the official permission and sharing some internals to make this possible.
For the METR rating (first half of the article), it is indeed 50% success rate at completing the task. The win rate only applies to the GDPval rating (second half of the article).
In this specific case, they started digging through a HDD image that someone else pulled. They're still trying to get hold of an actual Redbox machine to investigate the hardware as well.
That's actually how it works. Chamfers and fillets are handled by OpenCascade, so they can be applied to arbitrary edges. They only exist on the solid model though, so they are not integrated with the solver.
I'd be curious to see how native chamfers/fillets in solvespace would work, I imagine some amazing things could be done if they'd be part of the solver.