Version 0.3.0 came out on 21 October 2026. Since 0.2.0 we kept checking the solver in several independent ways and found sixteen more defects that could give a wrong result. They are fixed, and each one has a published safety advisory saying what was wrong, who is affected and what to re-check.
Re-run your projects
These are the cases that can move a design result the most. Each advisory has the full worked case and what to do with results you already delivered.
| Advisory | Case | Correct | Before |
|---|---|---|---|
| SA-2026-005 | Gable roof rafter: axial force at the base (kN) | -56.67 | -41.67 |
| SA-2026-007 | 1.5 m cruciform column: critical load (kN) | 2993 | 5860 |
| SA-2026-014 | Renumbered space frame: plastic collapse factor | 176.47 | 424.3 |
| SA-2026-015 | Guyed mast: tension in the guy (kN) | 5.18 | 4.34 |
| SA-2026-013 | Cantilever of curved shells: moment at the root (kN·m/m) | 35.0 | 0 |
The other advisories cover the response spectrum with CQC, plane models solved in 3D with nodal moments, supports changed from spring to rigid, arches outside the X–Z plane, the p-y, t-z and q-z soil curves, the solid shell in thin plates and, for those who use the engine or its API directly, material-nonlinear analysis, curved MITC9 elements and temperature loads on members.
Warned instead of silently wrong
- A result that a still-open defect could affect carries a warning naming it, in your language. One remains: material-nonlinear analysis is a whole-member secant model.
- When a temperature load does not state the expansion coefficient or the section depth, the result says which value was assumed.
- What has no solution is refused: a moment on a joint nothing can turn, a shell load on an element of another type, a soil curve without the inputs its reference needs.
Checked more deeply
- End-to-end certificates for frames and trusses: an independent checker rebuilds the stiffness and the loads from the model and certifies the displacements against the model, not only against the matrices the engine assembled. It already found three cases where engine and model disagreed.
- The same model always gives the same numbers, bit for bit, and every report, workbook and CSV records which version and which data produced it.
- CIRSOC 201 concrete design comes from one implementation in the engine.
- Mesh error estimates for plates and shells, shown as estimates, never as certified.