Five nets, one shared core

Claim (verbatim)

The great medieval subhasita anthologies were assembled in different centuries and different courts: Vidyakara's Subhasitaratnakosa in Pala Bengal (c. 1100), Sridharadasa's Saduktikarnamrta at the Sena court (1205), Jalhana's Suktimuktavali in the Deccan (1257-58), Vallabhadeva's Subhasitavali in Kashmir (fifteenth century), and Sarngadhara's Sarngadharapaddhati (1363). Had they all been dipping into one stable pan-Indian canon their poet-rosters would largely coincide; they do not. Each anthologist sampled mainly the poetry circulating around him, so the sets of named poets should overlap only in a thin canonical core (Kalidasa, Amaru, Bhartrhari, Rajasekhara, Murari and a few dozen more) and diverge nearly everywhere else. That overlap geometry then carries a survival payload, because pan-Indian fame and textual survival draw on the same circulation: the poets common to all the anthologies are precisely the ones whose works survive, while the poets unique to a single anthology are the drowned regional tail. Intersection is the canon and the symmetric difference is the graveyard, and the two can be told apart mechanically once the rosters are laid side by side.

Prediction clause (verbatim)

Prediction: taking the named-poet sets of at least four major anthologies (Subhasitaratnakosa, Saduktikarnamrta, Suktimuktavali and Sarngadharapaddhati), the pairwise Jaccard overlap of poet-rosters will be below 0.25 for every pair, while the poets in the intersection of all four will survive as authors of an independently transmitted work at a rate at least three times that of anthology-unique poets; the sub-0.25 pairwise Jaccard is the deciding clause and the survival ratio follows it. Poet identities are normalized across editions before matching, with honorific and spelling variants merged by hand, and the test voids for coverage if fewer than three anthologies yield a usable poet-index.

Kill-dataset (verbatim)

Kill (not yet built): the kavi-indexes of the printed editions (Kosambi and Gokhale, Harvard Oriental Series 42; Sures Chandra Banerji's Saduktikarnamrta; Embar Krishnamacharya's Suktimuktavali, Gaekwad's Oriental Series 82; Peterson's Sarngadharapaddhati) reconciled poet by poet, with survival scored against the New Catalogus Catalogorum, the poet identities controlled by Sternbach's Descriptive Catalogue of Poets Quoted in Sanskrit Anthologies and Inscriptions, and the SARIT/GRETIL text layer confirming attributions.

Novelty / leakage triage

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

Sternbach's Descriptive Catalogue reconciles the anthology poet-rosters name by name and his History-of-Indian-Literature fascicle describes each collection's date and regional sampling, so the materials and even the identity-normalization exist in print. The editors' introductions (Banerji for the Saduktikarnamrta, Krishnamacharya for the Suktimuktavali) note the regional bias of each roster. But the pairwise Jaccard geometry has never been computed, and the every-pair-below-0.25 clause is genuinely uncertain: the Subhasitaratnakosa and the Saduktikarnamrta are two eastern nets barely a century apart and could overlap far more heavily than the pan-Indian pairs. The threefold survival ratio for the intersection poets is likewise un-run against the New Catalogus Catalogorum.

• L. Sternbach, A Descriptive Catalogue of Poets Quoted in Sanskrit Anthologies and Inscriptions, 2 vols. (Wiesbaden, 1978-1980)
• L. Sternbach, Subhasita, Gnomic and Didactic Literature, A History of Indian Literature IV.1 (Wiesbaden, 1974)
• S. C. Banerji (ed.), Sadukti-karnamrta of Sridharadasa (Calcutta, 1965), introduction
• E. Krishnamacharya (ed.), Suktimuktavali of Bhagadatta Jalhana, Gaekwad's Oriental Series 82 (Baroda, 1938), introduction and kavi-index

Predictions

No prediction registered yet.