AI & Computingpreprint2026-08-09

Parallel-Chain Provenance with Structural Merge-Absence

Open access0 citations

Abstract

Multi-authority systems often need to retain several legitimate histories about the same external object without silently collapsing those histories into one lineage. This paper reports a scoped prototype built around an ordered history carrier whose public state-producing construction and rewrite operations take at most one parent history; multi-input methods are comparison queries, and no reviewed method constructs a descendant of two parent histories. The prototype is exercised through eight application wrappers: patent prosecution, version control, software bills of materials, AI model lineage, genomic variant interpretation, psychiatric assessment, credit-rating actions, and bank credit intelligence. A fresh run on 8 August 2026 passed 1,329 scripted regression checks across the eight wrappers. These are custom, shared-scaffold checks, not 1,329 independent experiments or formal proofs. Seven wrapper classes declare a record-and-classify posture; the version-control wrapper declares operations-with-merge and records three-way merge membership at the application layer because the carrier does not represent two-parent ancestry. Separately, an abstract snoc-list lemma was checked with Lean and Z3: two incomparable histories cannot both be prefixes of a single unary history. No refinement proof connecting that lemma to the complete Python implementation is present in the audited eight-file formal inventory, and the audited TLA+ files report no model-checking or theorem-proving result. The bank wrapper exposes content-excluding count and regulator-projection interfaces, but it does not implement authentication, private set intersection, or cryptographic confidentiality. The contribution is therefore a documented prototype with an author-operated, hash-scoped evidence record and a precisely bounded non-fusion interface pattern—not a general impossibility theorem, a production security guarantee, or an empirical validation of eight external domains.

// Source

View paper (DOI)Open access versionOpenAlexZenodo (CERN European Organization for Nuclear Research)Published 2026-08-09

Authors: Berkan Karakus