Every solver result here traces to a hashed artifact.
Four case studies. One is a failure, one is an unfinished trend. Both stay on this page because an engine you can check is only credible if you can also check it when it's wrong, or still in progress.
The bracket that survives where the conventional part fails
Torsion SF 2.055 vs 0.427 — same solver, same material. Heavier, and we say so. CalculiX-certified
A combustion-chamber sector coupon, solver-verified and print-ready
Inconel 625, curved-sector FEA, pore-flow LBM, print-ready 3MF/STL. Report claims machine-compiled: 10/10 ALL_TRACED.
The run that failed honestly (0 for 12)
The proxy promised twelve winners. The real solver failed all twelve. Nothing shipped — and the system quantified its own blind spot.
Nine campaigns. The search got better each time.
5.49 kg → 4.54 kg → 2.40 kg, every point CalculiX-certified. Not a baseline beat — the declared baseline fails outright. validated_sidecar