Kyrgyz Automatic Punctuation Restoration Benchmark: reproduction archive

Frozen text pools, canonical split manifest, archived word-level predictions, training configurations, reports and code snapshot for "Automatic Punctuation Restoration for Kyrgyz: A Two-Source Benchmark and Evaluation Study" (ACM TALLIP, under review). Two written sources (Kyrgyz Wikipedia and a Kyrgyz news corpus), S/C/P word labels with Q/E kept in an extended audit, document-level splits after near-duplicate grouping. See README.md for file hashes, use and per-component licenses.