RAMLab: Bounded Exact Memory Compression on a Narrow Benchmark Surface

This archive contains the RAMLab code, frozen workloads, benchmark scripts, reports, and manuscript for a bounded exact memory-compression study. RAMLab is evaluated on a fixed archived comparison surface consisting of three controller-stress workloads, one runnable adapter, and seven bounded proxy adapters. On that surface, the reported Exp70 configuration reaches 5.352x–6.005x effective capacity and exceeds the bounded Baryon proxy on effective capacity, while Baryon retains a lower bytes-touched proxy. The archive also includes a separate mini hotset service slice that serves as a narrow live proof-of-mechanism: under fixed dataset, API, traces, and budgets, RAMLab serves more exact requests before budget failure than raw while all served responses hash-match truth. The claims in this release are intentionally narrow and are limited to the reported archived surface; this archive does not claim operating-system RAM replacement, universal workload wins, or a production-ready memory layer.