Contract Retrieval and Payoff Differences across Kalshi and Polymarket

Prediction-market contracts can share a topic without sharing a payoff predicate. We examine the separation between candidate retrieval and contractual comparison in a fixed 30-day captured corpus comprising 1,029,600 Kalshi market representations and 1,226 Polymarket market identifiers under 100 observed parent events. The Polymarket frame contains 2,082 market and 198 event projection versions; it is the observed first-page collection, not a venue-wide census. An initial lexical search reports 3,117,473 candidate pairs. A subsequent field-aware search produces 1,486,341 delivered relation rows across 75 nonempty parent anchors. Its size is exactly the sum of parent-block Cartesian products; one Bitcoin block accounts for 78.77 percent. We then inspect a fixed hash-ranked sample of 294 pairs with full contractual texts and version histories. Eighty-eight pair records contain an unresolved literal option/person/party placeholder in the selected Polymarket question and are retained separately. The remaining 206 receive a proposition-first documentary screen identifying explicit differences in the recorded predicates, rather than an automated acceptance label. Examples distinguish entity homonyms, an event from its mention, a party winner from a vote-share threshold, a single race from a conjunction, and a path barrier from a fixed-time benchmark value. We provide exact candidate denominators, a reproducible reading ledger and source-bound counterexamples. The results evaluate the executed retrieval relation and its information requirements; they do not estimate platform-wide payoff equivalence, human-review reliability, or cross-venue finality speed.