What Cryptographic Verification Does Not Establish in Credit: Assurance Properties, Participation Limits, and a Reproducible Synthetic Case

Lenders increasingly assess borrowers through shared data infrastructure and, more recently, through privacy-preserving verification: a counterparty is told that a predicate over a borrower’s financial records holds, without being shown the records. Such mechanisms are often described as producing “verified” credit information, but the word conceals a distinction. A proof can establish that a computation was performed correctly over the inputs it was given, while leaving open whether those inputs represent the borrower’s obligations, whether the records were attributed to the right economic entity, whether they describe the present rather than the past, and whether any institution can act on the result. We assemble established results from database completeness theory, proof-of-liabilities, record linkage, bounded staleness, credit-registry economics, financial intermediation, and dependability engineering into a credit-specific framework of fifteen assurance properties organised as four lifecycle targets under two cross-cutting preconditions. We state one narrow credit-specific participation corollary, give twelve category-separation examples, and report a frozen synthetic case study of duplicate-financing detection across 120 parameter cells. The case study is negative. Favourable detection depends on a near-complete witness; a freshness tolerance that appears to reduce false alerts by a factor of seventy-two leaves detection and undetected exposure unchanged; an apparently flat coverage curve reflects deterministic truncation rather than robustness; and proving cost is unmeasured. The environment is synthetic. No result supports a real-world performance claim.