Corpus Analysis

The eleven corpora

Topic-family records

Two crowns per topic family: the record (most dicta in the family outright) and the signature (largest share of the holder's own classified corpus, floor of 5 items). Every cell links straight to the holder's teachings, pre-filtered to that family.

Raw record counts skew toward the biggest corpora — a sage with thousands of dicta can out-count everyone in most families on volume alone. The signature column is the honest counterweight: it normalizes by each sage's own corpus size, which is why the two columns often disagree.

Era drift: what the eras talk about

Each topic's share of its era's corpus (tannaitic era = zugot + tannaim, 5,831 dicta; amoraim, 16,405 dicta), for the corpus-wide leading topics. The clearest signal: ritual purity dominates tannaitic speech and nearly vanishes from amoraic speech, while meta-legal topics (legal precedent) appear only among the amoraim.

Dispute pairs

Sage pairs most often named together in dispute-classified segments (counted once per segment). Co-occurrence in a dispute, not outcomes.

Abaye and Rava, head to head

Counted honestly, this is co-occurrence, not psak: 258 segments carry both Abaye and Rava as attributed speakers. Beyond those, Abaye is the attributed speaker in 19 more segments that merely name Rava, and Rava in 17 that merely name Abaye — nearly even. The asymmetry is in total output: 3392 dicta for Rava against 2101 for Abaye. Which position "won" a given dispute is a question these counts do not answer.

Methods — what these numbers are, and are not

The miner counts dicta expressly attributed by formula (אמר X, X אומר, תני X, משום X, …) in the merged printed text of the Mishnah and Talmud Bavli only — no Tosefta, Yerushalmi, or midrashim, and no stories about a sage. Attribution follows the letter of the printed text (רבא and רבה are never merged). Classification (kind/type/topic/summary) is machine-generated and topic tags are free text, normalized here only for case and whitespace; the classifier's "unclear" tag marks passages too fragmentary to classify and is excluded from every topic ranking (those items are surfaced as fragmentary counts on the sage pages instead). Transmission-chain counts exclude every chain whose attribution carries ואיתימא. Dispute partners and dispute pairs count co-occurrence of recognized names in the same segment — sages missing from the alias table (Shammai, Beit Hillel/Beit Shammai as houses, most tannaim) are invisible to those counts, so all pair numbers are lower bounds. None of these numbers state halachic outcomes.