sswj_zh_ma_aligned supplement: language-model boundary review audit trail and quality checks

Supplement to the dataset sswj_zh_ma_aligned (doi:10.5281/zenodo.23292069), a sentence-aligned Manchu–Chinese parallel corpus of the 御製翻譯四書五經. It holds the exact instructions given to the language-model agents that reviewed sentence boundaries, the full audit trail of their 8,305 changes and 4,564 notes in 1,284 chunks, the 3,479 replayable manual corrections, and the results of the quality checks reported in the data paper: 3,865 agent-graded pairs, the human inspection of all 61 pairs graded partial or wrong, a human check of 100 agent-passed pairs, and the ten-page word-accuracy check (398 words). All pair-level files are keyed by the dataset's sentence_id. See README.md for the file layout.