TIB v1.0: Timeline Intervention Benchmark: propensity-logged benchmarks for e-commerce intervention bandits

TIB (Timeline Intervention Benchmark) v1.0: the first public, propensity-logged benchmark family for session-level e-commerce intervention bandits (six actions including a do-nothing control). Contains: sample CSVs and SHA-256 checksums for ten canonical seeds of two byte-reproducible generated tiers (tib-synth: sequence-pattern regime with per-action counterfactual rewards; tib-twin: 30-field production-style schema with realized and counterfactual outcomes and exact logged propensities); a deterministic build recipe for a semi-real tier over public RetailRocket timelines; a normative evaluation protocol (multi-estimator reporting with match rates, ten-seed confidence intervals, full-row-count rule, full-pass sampling for arm-sorted files); a datasheet following Gebru et al.; Croissant machine-readable metadata; and a 13-policy baseline leaderboard. All data regenerates offline from committed seeds and verifies byte-for-byte against the published checksums. Companion preprint: "Timeline-Weighted Contextual Bandits for Real-Time E-Commerce Intervention: Representation, Benchmarks, and an Honest Off-Policy Study" (DOI linked in Related works).