The classics kept a classroom; the diviners did not

Claim (verbatim)

Within the Hanshu Yiwenzhi (漢書藝文志) the six divisions were not equally survivable, and the reason is institutional custody rather than the merit of the books. The Liuyi lue (六藝略, Six Arts / classics) fed directly into the state examination and teaching system, which recopied and commented its texts without interruption; the technical divisions - bingshu (兵書, military), shushu (術數, numbers-and-divination), fangji (方技, recipes and medicine) - and the minor schools of the Zhuzi lue (諸子略) had no such permanent classroom, only intermittent private demand. A book survives where an institution is obligated to reproduce it. So within one catalogue the classics should survive whole at a high rate while the divinatory and military divisions approach total loss, and among the ten schools the same split should recur: Confucian (儒) and Daoist (道), canonized and taught, keep multiple whole works; Agriculturalists (農家), Logicians (名家), and Minor Talk (小說家) collapse to bare titles and quotation. Prediction restated: the classics survive whole at least five times as often as the pooled technical divisions, and three named minor schools survive at essentially zero. The asymmetry is a fingerprint of who was obliged to hold the brush, not of what the books were worth.

Prediction clause (verbatim)

Prediction: computing whole-text survival division by division, the Liuyi lue (六藝略) classics survival rate will exceed the pooled survival rate of the bingshu (兵書), shushu (術數) and fangji (方技) divisions by at least a factor of five, and among the Zhuzi lue schools the Confucian (儒家) and Daoist (道家) will each retain multiple whole transmitted works while the Agriculturalists (農家), Logicians (名家) and Minor Talk (小說家) retain essentially none (primary clause: the >=5x classics-over-technical whole-survival ratio; the verdict follows it); survival is scored the same way for every division against a fixed full-text corpus, and the test voids for coverage if the division and school of fewer than 500 entries can be read from the treatise's headers.

Kill-dataset (verbatim)

Kill: the Chinese Text Project (ctext.org) Hanshu Yiwenzhi (漢書藝文志), reading each entry's division and school from the treatise's own headers, with whole-survival scored identically for every division against ctext's pre-Tang full-text library and the Siku quanshu holdings.

Novelty / leakage triage

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

The institutional-custody pattern — classics recopied by the teaching apparatus while the 兵書/術數/方技 divisions and the minor schools (農家/名家/小說家) collapse to titles and quotation — is standard Yiwenzhi scholarship and is qualitatively established. The primary clause, however, pins a >=5x classics-over-pooled-technical whole-survival ratio scored identically across divisions, exactly a per-division ratio of the kind the discipline flags as adjacent when only the materials are in print. No published study reports this specific division-by-division survival ratio against a fixed corpus. Nearest prior art (the annotated-catalogue tradition and Yu Jiaxi on how ancient books were transmitted and lost) works the neighborhood without running the test.

• 顧實 (Gu Shi), 漢書藝文志講疏 (Shanghai: Commercial Press, 1924)
• 余嘉錫 (Yu Jiaxi), 古書通例
• 姚振宗 (Yao Zhenzong), 漢書藝文志條理 (in 二十五史補編)

Predictions

No prediction registered yet.