The commentator kept the readings the winners burned

Claim (verbatim)

Zheng Xuan (鄭玄, 127-200) annotated the classics at the moment the jinwen (今文, New Text) and guwen (古文, Old Text) recensions were fusing, and his commentaries are a deliberate reservoir of the losing readings. Glossing the Three Rites and the Odes, he routinely records the variant he is not adopting, with fixed formulae - gushu (故書 'the old text has...'), jinwen...wei... (今文...為... 'the New Text writes X as Y'), guwen...wei... (古文...為...) - preserving, inside a surviving commentary, the exact graph-level readings of recensions that as independent texts did not survive at all. The mechanism is that a commentary's job is adjudication, and adjudication must quote the rejected party; so the losing recensions live on as citations inside the very work that superseded them. That makes Zheng Xuan's apparatus countable: the density of these variant-citation formulae measures how much lost-recension text a surviving commentary is carrying. Prediction restated: in the Zhouli commentary the gushu formula alone recurs at least 150 times, and the pooled variant-citation rate outruns a single-recension control commentary tenfold. The recensions that lost the philological war did not vanish; they were interred, legibly, in the footnotes of the recension that won.

Prediction clause (verbatim)

Prediction: counting the recension-variant formulae (故書; 今文...為; 古文...為, and their fixed variants) in Zheng Xuan's commentary to the Zhouli (周禮注), the gushu (故書) formula alone will occur at least 150 times, and the pooled variant-citation rate per 1,000 graphs of commentary will exceed that of a single-recension control commentary - He Yan's Lunyu jijie (何晏 論語集解) - by at least a factor of ten (primary clause: the >=150 gushu count together with the >=10x rate ratio against the single-recension control; the verdict follows it).

Kill-dataset (verbatim)

Kill: the Chinese Text Project (ctext.org) or Kanseki Repository full text of the Zhouli with Zheng Xuan's commentary (周禮鄭注), string-counting the pinned variant formulae and normalizing by commentary length, with He Yan's Lunyu jijie (論語集解) as the single-recension control.

Novelty / leakage triage

anticipated in the literature — this exact test has never been run

Zheng Xuan's practice of recording rejected recension readings with fixed formulae (故書; 今文...為; 古文...為) is a classic topic of Chinese philology, and Duan Yucai's 周禮漢讀考 is a whole monograph on exactly these 故書/今古文 readings in the Zhouli commentary. But the primary clause is a formula count and rate ratio — 故書 occurring >=150 times, and the pooled variant-citation rate >=10x that of He Yan's Lunyu jijie control — and the discipline flags formula counts over in-print materials as adjacent. No published source states the >=150 count together with the >=10x normalized ratio against that named control. The apparatus is described exhaustively; the specific counts and the control-normalized rate are un-run.

• 段玉裁 (Duan Yucai), 周禮漢讀考 (on the 故書/今古文 readings in Zheng Xuan's Zhouli commentary)
• 皮錫瑞 (Pi Xirui), 經學歷史 (on the New Text / Old Text recensions)

Predictions

No prediction registered yet.