01
The digital model and the physical object are often miles apart
A CAD assembly may exist on an engineering workstation while the physical system exists on a factory floor, job site, customer location, or mobile device. The camera that observes the object usually does not have the native CAD application, full project vault, or engineering workstation beside it.
That separation creates a context problem. Computer vision can detect shapes and objects, but it may not know which engineered entities are expected, how those entities relate, which variations are valid, or what source geometry supports a match.
The research question is how to carry enough engineering context to the field without carrying the entire engineering environment.
02
STX as an engineering-context package
STX is a Stratic Systems research format under development, not an industry standard. Its proposed role is to package a controlled representation of engineering context for use outside the originating CAD system.
The package can contain stable entity identity, normalized geometry or geometric descriptors, assembly relationships, coordinate frames, metadata, provenance, and optional visualization assets. The exact schema should remain versioned and extensible as real use cases expose what information is necessary.
The design principle is separation: the field system receives the context required for its task without automatically receiving the complete source project or proprietary CAD history.
03
Vision should observe; geometry should verify
Field vision and deterministic geometry solve different parts of the problem.
A vision model can propose that a camera observation resembles a known component. Geometry-derived descriptors can then help verify whether dimensions, proportions, feature placement, orientation, or relationships are consistent with the expected entity.
The architecture should preserve both results: what the vision system observed and how strongly the engineering representation supports or contradicts the match.
04
A possible field loop
The first useful loop does not require a robot. It can be a controlled camera experiment with a small physical assembly and known engineering ground truth.
05
This extends the digital thread into observation
NIST's digital-thread research focuses on linking authoritative product information across design, manufacturing, and inspection. Related NIST test-bed work has examined cyber-physical infrastructure connecting CAx environments with manufacturing systems through product models.
STX explores a narrower question inside that larger direction: what portable representation is useful when an AI or vision system needs engineering context at the point of observation?
The goal is interoperability of understanding rather than remote control of the original CAD application.
06
Security and provenance are part of the format
A portable engineering package should make its boundaries explicit. The receiver should be able to determine who or what created the package, which source version it represents, which entities are included, and whether the package has been modified.
Future work may include signing, integrity verification, access policy, selective disclosure, and stronger provenance mechanics. Those are research requirements rather than finished claims.
This is especially important if STX eventually moves between organizations, devices, autonomous systems, or inspection workflows.
07
The larger objective
The long-term value of STX is not the file extension itself. It is the ability to separate engineering knowledge from one application session and make a controlled portion of that knowledge available to other intelligent systems.
If a camera, mobile device, local AI model, inspection station, or future robot can understand the same entities and relationships established by the engineering system, CAD stops being only a design artifact. It becomes a source of physical context.
That bridge—from engineering definition to physical observation—is one of the key requirements for practical physical intelligence.