Evidence-Preserving Recovery for Regional Identity Synchronization

Abstract

Regional identity synchronization may propagate selected changes while keeping existing regional systems authoritative. We study an external, change-only coordinator connected through thin regional machine-to-machine API adapters. With monotonic source revisions and conditional updates, an interrupted request can remain unattributable: two executions have equal declared, normalized restart-API observations although different requests caused the observed change. Our evidence-preserving checkpoint recovery contract separates past-request attribution from the future synchronization frontier. Request-bound receipts can confirm past success; otherwise, a validated adapter-signed checkpoint and current scoped permission may enable a new conditional request while preserving unresolved requests and unknown history. A locally fixed loopback HTTP study compared permanent hold, a separately implemented API event consumer, and a custom Git-like evidence-graph consumer whose runtime does not use Git: three arms, six scenarios and three repeats, totaling 54 synthetic cases on one Mac mini. Permanent hold resumed 0/12 recoverable cases; event and Git-like consumers each resumed 12/12 and used 213 HTTP requests over 18 cases. A separate eight-case process-stop rerun confirmed all future updates using signed adapter receipts and source revision/value readback; six old requests were not marked successful and two remained indeterminate. These bounded results assume trusted regional adapters and honest coordinator execution and audit retention. Source mutation and adapter receipt persistence remain non-atomic; effect-time revocation requires a source authorization fence. The contribution is an integration contract and capability boundary, not a general safety or legal-compliance proof. In the main comparison, both recovery consumers had equal recovery results and API counts. This record contains the manuscript PDF and its LaTeX source. The 15-page technical report expands a study described in an abstract accepted for the IWSEC 2026 Poster Session. Experimental code and result artifacts are not included in this record. This technical report has not undergone journal or full-paper conference peer review.