The author is building a high-performance operating system called V.E.L.O.C.I.T.Y.-OS, aiming to run entirely within the CPU's L3 cache. To achieve maximum speed for scalar operations, they are now compiling abstract syntax tree (AST) blocks directly into raw x86-64 machine instructions at runtime. This approach bypasses the overhead of closure chains, which were previously a bottleneck for non-massive computations. They developed a scalar detector to identify blocks containing only basic scalar operations to trigger this direct compilation.The JIT compiler then emits machine code bytes into an executable memory page, including a carefully crafted function prologue for register allocation and stack framing. Local variables are mapped directly to CPU registers R12D through R15D to minimize memory access. A critical bug involving REX prefixes during register loading and storing was identified and fixed.Furthermore, a component called SCEV-Lite was implemented to perform algebraic loop solving during JIT compilation. If a loop follows predictable arithmetic patterns, SCEV-Lite algebraically solves for the final values, transforming potentially millions of loop iterations into a constant-time, five-instruction calculation. This significantly accelerates loop execution.A pre-pass called pre_register_variables was introduced to correctly map dynamic loop variables to the stack before assembly generation, resolving an issue where such variables were being written back as zero. Benchmarks showed an incredible speedup of 198,937x for an induction loop compiled with this native JIT compared to the interpreter. This optimization collapses execution layers, mirroring the two-tier architecture of modern CPU microcode.The next steps involve implementing classic compiler optimization passes like constant folding, propagation, loop unrolling, and dead code elimination to further refine code generation.