mtcodec: Reproduction package for Compression and Encoding Cost in Context Clustered Huffman and rANS Coding

Abstract

This software and evidence package supports the manuscript “Compression and Encoding Cost in Context Clustered Huffman and rANS Coding” by Richard Leinecker, University of Central Florida, prepared for submission to the Data Compression Conference. The study compares deterministic context reassignment with native-policy and matched-cost Brotli-derived merge/remap controls under common static Huffman and range asymmetric numeral systems (rANS) serializers. It evaluates complete codec-body size, encoding and decoding throughput, and requested heap allocation using 41 corpus entries and 720 sampled blocks of nominal sizes 16, 64, and 256 KiB. Under the tested implementations and workloads, reassignment with estimated, header-inclusive rANS selection encodes approximately 8–18 times faster than native-policy merging, with mean-rate penalties of approximately 0.6–2.1%. These measurements characterize the compression–encoding-cost trade-off within the study’s controlled framework. The package contains frozen source code, tests, experimental protocols, input manifests, raw measurements, allocation records, expected outputs, and reproduction scripts. The public evidence archive preserves measurement CSVs and scientific files unchanged, with documented redactions of operational paths and machine identifiers. An analysis script regenerates all four manuscript tables and the 36 points in its figure from saved measurements. Completed validation includes publisher reacquisition and SHA-256 verification of all 41 corpus entries, verification of 771 input members across 87 workloads, saved-results regeneration, and a Windows build/smoke test passing 52 unit checks, 144 native-oracle comparisons, and five expected-failure controls. The original DCC005 experiment completed successfully; the new publication wrapper’s separate audit and full timing phases have not been executed. Version: v1.0.0-dcc-submission.Experimental session: DCC005-20260926-112604-bdea7596. Corpus data are acquired separately. Applicable CC0 and MIT notices are retained and documented in the package.