Small research teams may need to review structured image records without transferring operational files to a third-party service. This study redesigns and reproducibly evaluates a lightweight Flask platform backed by local JSON files. An isolated copy of a legacy prototype was hardened with hashed passwords, controlled registration, CSRF and role checks, structured validation, atomic replacement, validated restore, safer export and paths, login throttling, security headers, and failure-closed parsing. Evaluation used generated metadata and no operational images or accounts. All 12 functional tests passed, including an injected atomic-replacement failure that preserved the previous repository. The hardened copy satisfied all 15 audit-derived controls, versus none in the baseline; this trace is not a security certification. Across three sequential sessions, all 360 clean-start, in-process concurrency scenarios passed, allocating 32,400 synthetic records with zero duplicate identifiers. At 20 clients, median session-level P95 was 760.2 ms for batch allocation and 157.1 ms for saving one record. A separate 120-scenario loopback-HTTP run produced request-level P95 values of 913.8 and 202.7 ms. Three scale sessions yielded 900 observations from 100 to 10,000 records. A negative boundary experiment then started two or four independent processes against the same generated repository. None of 20 clean starts met the integrity criterion; 18 produced duplicate identifiers and assignment-to-master ownership mismatches, and nine also contained a write error. The evidence supports only the tested single-host, single-process configuration and empirically rejects treating its process-local lock as a multi-process transaction. Usability, review quality, production-network performance, certification, and legal compliance remain untested.