SETL v2.0: Definition, Reference Architecture and Canonical Governed Verification Record

This release package publishes SETL v2.0, the Standardized Evaluation & Trust Layer. SETL is an open architectural verification layer for producing governed, evidentially grounded and inspectable records of AI-assisted cognitive work. Its governing principle is: “The output is not the record.” The package contains: (1) a concise SETL v2.0 definition; (2) the SETL v2.0 Reference Architecture Definition, documenting the evolution from the v1.1 verification primitive to a running reference implementation in Nivel; and (3) the canonical governed verification record wb-20260901T215552Z-12e76987. The canonical specimen preserves the actual evidential, verification and disposition state of a completed Nivel execution. Its final governed disposition is BLOCK, demonstrating that SETL records the basis for a result and its governed outcome rather than treating a model output as sufficient evidence.