Five years of Kalshi temperature markets across seven U.S. cities, read hour by hour as a probability distribution over tomorrow's daily high. At the end of the market's first hour of trading its root-mean-square error sits about 10 percent below the National Blend of Models, the most accurate single public product, and it still leads by 11 percent at the NBM's final bulletin on the target morning. The advantage is largest where forecasting is hardest, in Philadelphia, Chicago, and New York in winter when cold-front timing matters, and nearly vanishes in Miami where tropical air keeps highs predictable. An event study settles the direction of information flow: public forecast revisions travel toward the market roughly four times further than the market travels toward them, so prices do not jump when the National Weather Service publishes.

Distinguishes payoff-space no-arbitrage, which follows from terminal payoffs, from protocol-executable no-arbitrage, which depends on what position transformations traders can actually perform. Using Polymarket's negative-risk markets and the NegRisk Adapter, the authors reconstruct depth-aware portfolio values and transaction histories to measure payoff-bound violations: an estimated $1.12 million in arbitrage profit across two realization channels. Violations concentrate on the unsupported YES side, while adapter-supported NO-side violations are rarer and shorter-lived. The takeaway: market efficiency depends not only on payoff structure, but on whether protocols expose payoff equivalences as executable primitives.

Analyzes 23 million moneyline trades on Kalshi across major sport leagues and finds two systematic ways market prices fail as probabilities. Calibration is not static: parameters sit near perfect-calibration values mid-contract but depart sharply as expiry approaches, with the final ten minutes producing a step-like Prelec curve consistent with insurance-demand behavior by traders holding losing positions. Cross-game parlays are systematically overpriced relative to the product of their leg prices, with overpricing growing in leg count and a separate market-level markup at the parlay-pricing stage. The practical takeaway: treating prediction-market prices as probabilities requires conditioning on time to expiry and product type, not just price.

Reichenbach and Walther ask when a prediction market price should be trusted as a probability. Working through Polymarket's contract history, they separate price informativeness from calibration and bias, testing whether quoted odds track realized outcomes. The paper treats trust in prices as an empirical property rather than a design assumption — and shows where calibration and systematic bias distort the read.