About ForgeCertify
ForgeCertify was built by SKYFIRE Forge and is operated as an independent, vendor-neutral certification body.
We publish everything: the tests, the thresholds, the methodology, and the results — including our own.
Origin
ForgeCAD needed a rigorous way to validate its geometry kernel — one that went beyond "does it crash"
to "is the math actually correct." We searched for an existing standard and found none that covered
the full range of operations real engineering CAD work demands: sub-micron boolean accuracy, physical
property fidelity against closed-form solutions, round-trip I/O without silent geometry degradation.
So we built one. And then we made it public, because a test suite that only one vendor can run
is not a standard — it's marketing.
ForgeCertify is operated with a clear principle: the entity that created the test suite is not exempt
from it. ForgeCAD's kernel results are published to the same public ledger as everyone else's. The same
thresholds apply. The same signing chain is required. There are no carve-outs.
Independence principles
Open specification
Every test in the suite is public. The pass/fail thresholds are documented. Anyone can read the source, reproduce any result, or challenge a threshold.
No cherry-picking
Certification requires running the full suite. There is no mechanism to exclude categories or individual tests. Skip counts are reported and explained — skips for missing capability are different from failures, and both are disclosed.
Results are immutable
Once published, a result is never edited. If a new test is added and a previously-certified kernel fails it, that is disclosed publicly — the original certificate remains valid for the scope it covered, and re-certification is required for the new suite version.
Creator is not exempt
ForgeCAD's ForgeKernel is held to exactly the same thresholds as every other kernel. We do not pre-screen results before publishing. If ForgeKernel fails a test, that failure appears on the public ledger.
Community contributions
The test suite grows through external contributions. Accepted submissions are credited. We prioritize tests that expose real engineering failure modes over tests that favor any particular kernel architecture.
Scoring methodology
Each test has one of three outcomes:
PASSED
The kernel produced output within the specified tolerance. Tolerances are defined per-test and documented in the source — e.g. volume accuracy within 0.001%, boolean result within 1 µm of reference geometry.
SKIPPED
The test requires a capability the kernel does not declare (e.g. analytic_primitives for exact volume tests against mesh-only kernels). Skips are not failures — they reflect the kernel's declared capability tier. Skip reasons are always reported.
FAILED
The kernel produced output outside tolerance, raised an unexpected exception, or produced geometry that violated a hard invariant (e.g. negative volume, non-watertight mesh). Failures block certification for the affected capability tier.
Capability tiers
Kernels declare a capability tier that gates which tests they are expected to pass:
| Tier | Typical kernels | Capabilities |
|---|
| COMMUNITY | manifold3d, CGAL, Blender | Boolean, mesh healing, STL export, basic inertia. No analytic primitives, no NURBS. |
| STUDIO | Mid-range commercial kernels | COMMUNITY + NURBS curves/surfaces, parametric features, 2D constraint solving. |
| PRO | OpenCASCADE, CadQuery, FreeCAD, Parasolid | STUDIO + exact analytic primitives, loft/sweep/blend, full round-trip I/O, GD&T. |
| FORGE | ForgeKernel | PRO + IGA, convergent modeling, geodesic, medial axis, G2/G3 continuity, CAM, FEA/CFD mesh export. |
Test suite versioning
The suite is versioned with a MANIFEST.json signed with Ed25519. Every test file is hashed into the manifest. If a test is added, changed, or removed, the manifest version increments and the signature changes. A certificate always states which suite version was used — older certificates remain valid for that scope.
Kernels are encouraged to re-certify against each new suite version. The public ledger shows the suite version alongside every result.
Questions about methodology? Want to challenge a threshold?
Email us or
submit a test. We welcome scrutiny.