One history wears another's skin, up to a point

Claim (verbatim)

The Hanshu (漢書, Ban Gu, finished c. 111) and the Shiji (史記, Sima Qian, c. 91 BCE) run over the same early-Han ground twice, and the second historian did not re-report it - he re-copied it. For the period they share, from the founding of Han to roughly the reign of Emperor Wu, Ban Gu transcribed long stretches of Sima Qian nearly verbatim, editing archaisms, inserting documents, and re-sorting the material into his annals-and-tables frame, then continued alone past the Shiji's horizon. This makes the pair a controlled transmission experiment already sitting in the corpus: verbatim overlap should be dense in the shared span and fall off a cliff exactly where the Shiji's coverage ends, because after that point Ban Gu had no Sima Qian in front of him to copy. The overlap is thus not a fact about either author's prose but about where one text physically had the other on the desk. Prediction restated: the parallel annals and biographies of pre-Emperor-Wu figures share verbatim n-grams at least ten times as densely as any Hanshu chapter covering the decades after the Shiji stops - a step at the Shiji horizon, not a gradient. Read as citation geometry, the Hanshu's debt to the Shiji is measurable to the chapter, and its edge is the edge of the older book.

Prediction clause (verbatim)

Prediction: computing shared long character-n-grams (shingles of length >=15 graphs) between the two texts chapter by chapter, the parallel biographies and annals of pre-104-BCE figures (for example Xiao He 蕭何, Zhang Liang 張良, Han Xin 韓信, and the Gaozu/Gaodi annals 高祖本紀 / 高帝紀) will show verbatim-overlap density at least ten times that of any Hanshu chapter covering events after the Shiji's coverage ends (primary clause: the >=10x overlap-density ratio across the Shiji horizon; the verdict follows it); n-gram length is fixed in advance to guard against tokenization variance.

Kill-dataset (verbatim)

Kill: the Chinese Text Project (ctext.org) full texts of the Shiji (史記) and Hanshu (漢書), computing chapter-level shared-shingle density at a fixed n-gram length and comparing chapters inside versus outside the Shiji's chronological coverage.

Novelty / leakage triage

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

The Hanshu's heavy verbatim borrowing from the Shiji across the shared early-Han span is textbook historiography, and the parallel passages have a long comparative tradition — Ni Si's 班馬異同 systematically set the two texts side by side. But the primary clause pins n-gram shingle geometry: shared >=15-graph shingles at >=10x density inside versus outside the Shiji's coverage horizon, a step not a gradient. That is the discipline's canonical 'n-gram geometry = adjacent' case — the phenomenon is in print, the shingle-density computation across the horizon is un-run. Adjacent; the neighborhood is worked qualitatively and in traditional parallel-collation, not as citation geometry.

• 倪思 (Ni Si, trad. attrib.), 班馬異同 (a systematic collation of the Shiji–Hanshu parallel passages)

Predictions

No prediction registered yet.