Abstract
This disclosure describes a software development kit that makes AI applications, agents and robot software conformant to a safety standard as they are written, rather than tested for conformance afterwards. The kit offers no direct path to actuators, data, networks, money or other agents: every such action is made through a governed call that carries a declared purpose and is submitted to the system's runtime governance point. At build time, the kit analyses the code and derives the application's declarations, such as the actions it can take, the data it can read, the sensors it uses and the limits it needs; a build that uses a capability not declared, or declares one it does not use, fails. The kit pins each build to a version of the standard's clauses and runs the executable tests bound to those clauses against the application, in a local emulator of the governor, reset and safety plane, on every build. A passing build is packaged as a governed software bundle that binds the code and model to their policy, declarations and test evidence, with signed build provenance linking source, dependencies, bundle and model lineage. Hooks for incident reporting, conformance evidence and privacy defaults are built in and cannot be switched off in certified builds. When the standard issues an urgent clause, the kit flags affected applications and re-runs their tests. Developers thus receive, at each build, a statement of which conformance level the application meets and why.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Wang, Edward D. H., "SI SDK™: A Development Kit that Builds Conformance into AI Applications by Construction, Routing Every Action through Governed Calls, Deriving Declarations from the Code, Running the Standard's Executable Tests in Every Build, and Emitting a Signed, Certifiable Software Bundle", Technical Disclosure Commons, (October 20, 2026)
https://www.tdcommons.org/dpubs_series/12323