Data Vintages and Evidence Granularity in Kalshi Event Contracts

An observed prediction-market result does not identify when publicly available evidence became sufficient under the contract. We study this distinction using a 30-day captured Kalshi corpus, a metadata projection of 1,029,600 distinct market identifiers, and a field-specific census of 6,842,078 lifecycle receipt-message associations. Primary rules are available for all projected markets and secondary rules for 998,771. A document-based analysis distinguishes the selection of a first publication from a finalized data vintage: among 931 captured contracts naming Carbon Arc, 248 select a first reported value and 683 select a finalized value at an explicit calendar cutoff. These are contract counts, not independent releases. An eight-case sports diagnostic and direct decoding of four preserved web responses separate extraction failure, missing values, and insufficient predicate granularity. In one case, a preserved homepage contains a final score of Syracuse 13-Pittsburgh 27. A fixed, source-linked family query yields 264 contract records; the score supports conditional evaluation of the normal-branch numerical predicates of 78 full-game contracts, but does not establish governing-source authority, first publication, or the previously selected third-quarter predicate. Finally, nested determination and settlement timestamps are present in all corresponding extracted message associations, despite much lower availability of the top-level sending timestamp. The results show why rule availability, predicate sufficiency, source admissibility, and temporal identification must be evaluated separately. We report reproducible documentary and computational findings, not an independently adjudicated clock distribution or an estimate of venue delay.