A Provenance-Preserving Scientific Claim Ledger: Versioned Evidence, Supersession, and Auditable Correction Without Historical Rewriting

Scientific knowledge systems routinely revise claims as new evidence appears. Yet many research workflows represent claims as mutable text fields or database rows whose latest value silently replaces earlier states. This creates an audit problem: correction can erase provenance, supersession can obscure which evidence supported which formulation, and later interpretations can inherit support that was collected for materially different claims. This paper presents a domain-specific architecture for scientific claim records that integrates bitemporal non-erasure semantics (Snodgrass 1999; SQL:2011), PROV-O-aligned provenance tracking (Lebo et al. 2013), nanopublication-inspired evidence typing (Groth et al. 2010), and an event-ID replay rule layered on event sourcing (Fowler 2005). The integration includes design elements not identified together in the cited systems: a typed evidence-inheritance function I() for supersession events, a five-state administrative status vocabulary independent of evidentiary state, machine-generated claim lifecycle management via PROPOSED_CLAIM, and typed SPLIT/MERGE operations with provenance threading.