CodeMarkBench: Canonical Raw Results and Release Bundle for Source-Code Watermarking Reliability Evaluation

This Zenodo record archives the rerun-backed raw results and sanitized release bundle for CodeMarkBench, a benchmark for evaluating source-code watermarking reliability for LLM-based code generation. The archived canonical matrix is the completed single-host 8-GPU run of suite_all_models_methods: run_count=140, success_count=140, failed_count=0, and execution_mode=single_host_canonical. The canonical matrix index SHA-256 is 7bf4cc0a125b0a9bf9e1cb8efb8cdfd089a3521c4f58b7daf399edc862d35c0b. The GitHub companion repository contains the source code, documentation, canonical inputs, environment capture, and materialized summary tables/figures. This Zenodo record carries the large raw matrix artifact plus a sanitized release bundle so reviewers can regenerate the tracked summary exports or restore the project on a fresh cloud host. GitHub companion commit: 8270ef18c393d9c9b9d05d9c1b7bedac83977a48. Raw archive SHA-256: 29d0c20a5f5e99cc24d61e7479e4d788565161c78c3660e560412eb502d38a2d. Sanitized bundle SHA-256: 44d4f05511d82e69545b7f2403f67fbdd45f6585a5c337f301c063614b8f8ed1.