Locus Core

A high-performance mathematical modeling kernel combining exact analytical B-Rep topology, bounded multiprocessing compute workers, AST-sandboxed math parsing, and ISO-10303 STEP solid synthesis.

Request early access Technology & Architecture

✓ Multi-Process Compute Pool ✓ AST-Sandboxed Math Evaluation ✓ Sub-millisecond LRU Cache

Interactive 3D • Drag to Orbit Model: NACA 0015 Airfoil

Chebyshev-spaced NACA 4-Digit Equations

Internal Engineering

The Four Pillars of Locus Core

Bounded computation, validated inputs, and representation-aware exports support geometry workflows. Resource limits and geometric accuracy require validation against the intended workload.

1. Bounded Worker Multiprocessing Pool

Configured geometry operations use isolated worker processes managed by ComputePool. Admission, deadlines, and resource limits bound their execution; isolation coverage depends on the request path.

pool = ComputePool(max_workers=4, per_job_timeout=15.0) result = pool.submit('create', payload).result(timeout=15.0)

2. AST-Sandboxed Math Parsing

Parametric dimensions, NACA thickness equations, and mathematical profile curves are evaluated using an Abstract Syntax Tree (AST) validator. Disallows arbitrary execution, imports, or variable access while supporting high-speed mathematical functions.

def safe_eval(expr, scope): tree = ast.parse(expr, mode='eval') # Validates safe operators: Add, Sub, Mult, Div, Pow, sin, cos return Evaluator(scope).visit(tree)

3. SHA-256 LRU Model Cache

Every geometric operation is uniquely fingerprinted using a canonical SHA-256 digest of its normalized parameters. Identical requests bypass the worker pool completely, reusing retained geometry when the source parameters and evaluation settings match.

digest = sha256(canonical_json(payload)).hexdigest() if cache.has(digest): return cache.get(digest) # reuse a retained result

4. Topological Manifold Verification

Geometry checks cover watertightness, winding, signed volume, and degeneracy on supported paths. Topology varies with cavities and genus; an Euler characteristic of two is not a universal requirement. Supported parameter domains and downstream imports still require validation.

assert mesh.is_watertight assert signed_volume(mesh) > 1e-6 assert len(mesh.non_manifold_edges) == 0

Performance scope

Measure the workflow that matters

Model generation, surface evaluation, exports, network transfer, and browser rendering contribute different costs. We are not publishing latency or throughput guarantees without a reproducible workload, hardware configuration, and measurement method.