Jian Lin—Non-Interference Preservation Theorem (NIPT)

We formalize a non-interference guarantee for a layered audit stack: a frozen mathematical CORE, auxiliary PERIMETER modules, and META-theory guards. We prove that if (i) CORE parameters are read-only, (ii) evidence is stored in isolated namespaces with committed hashes, and (iii) terminal adjudication is a monotone function of declared leaves that never writes into CORE, then no PERIMETER/META outcome (PASS/FAIL/ALERT/INAD) can modify CORE constants nor invalidate CORE-level conclusions. Conversely, CORE-level acceptance depends only on declared, hash-committed summaries of auxiliary leaves, not on auxiliary implementation details.