Derived-Column Elimination by Exact Functional-Dependency Discovery: a Verified, Reversible Pre-Pass for Lossless Columnar Compression

Real columnar datasets frequently carry columns whose values are an exact deterministic function of other columns in the same row — a transaction total that is precisely amount + fee, a network-flow byte counter that is the sum of two component counters, a derived rate that is a fixed multiple of a base. A general-purpose byte compressor stores such a column in full because it cannot, from bytes alone, recover the arithmetic relationship. We describe a lossless pre-pass that discovers exact arithmetic functional dependencies directly from the data, drops every dependent column, and stores only the formula needed to reconstruct it bit-for-bit on read. On a real public network-telemetry corpus (CTU-13 NetFlow) the pre-pass removes 8.27% of the post-compression size beyond a strong columnar baseline, with byte-exact reconstruction verified on every row. We make no novelty claim over the general idea of FD-based column elimination — it is established prior art (notably Corra, 2024, and earlier database patents). We publish this disclosure to place our specific implementation on the public record as prior art and to describe the engineering we believe is the actual contribution: an FD pre-pass that is verified byte-exact, never-worse, guarded against text-encoding hazards such as leading zeros, and embedded inside a queryable, tamper-evident, per-record-erasable container rather than offered as a bare ratio trick.