1 · Source versions and input identity一 · 来源版本与输入同一性
A reported date or version label does not pin exact bytes. Unknown input fixity stays unknown. Reuse permission is scoped to the identified source or release; public availability is not itself a reuse grant.已记录的日期或版本名不能确定具体字节。未经验证的输入指纹仍标为未知。再利用许可仅适用于所指明的来源或版本；公开可访问并不等于获得再利用许可。
2 · Regeneration chains (raw → dataset → finding)二 · 再生成链（原始 → 数据集 → 结论）
Each listed dataset has an output checksum. It verifies that output's bytes, independently of whether the original input can be acquired and identified. The chain names the intended regeneration script; it is not proof of a successful external replay.每个列出的数据集都有输出校验和，用于核对该输出的字节，与原始输入能否获取和确认是两回事。构建链列出预定的再生成脚本，并不证明外部复现已成功。
3 · Honest limits三 · 如实声明的边界
Exact replay requires a source-specific input identity, an available input, the appropriate reuse terms and the documented build environment. Current unknowns are explicit above. In particular, the conflicting earlier EDCS rights summaries have not established a source-specific grant; EDCS reuse remains unknown, requiring review. LIST v0.3's identified Zenodo release declares CC BY 4.0 with attribution; this does not establish the identity of the input used here or clear every derivative. The ML benchmark has separate harness documentation at
eval/REPRODUCIBILITY.md.
精确复现需要特定来源的输入同一性证据、可获取的输入、适用的再利用条款及有记录的构建环境。当前未知项已在上表明确标示。尤其是，先前相互冲突的 EDCS 权利摘要不能证明特定来源的许可；EDCS 再利用仍为未知，待审阅。所指明的 LIST v0.3 Zenodo 版本声明采用 CC BY 4.0，须署名；这不能证明本站所用输入的同一性，也不能为全部衍生成果作出许可认定。ML 基准另有评测装置文档 eval/REPRODUCIBILITY.md。