Bayesian probabilistic modeling for four-top production at the LHC

Monte Carlo (MC) generators are crucial for analyzing data in particle collider experiments. However, often even a small mismatch between the MC simulations and measurements can undermine interpretation of results. This is particularly important context LHC searches rare physics processes within beyond standard model (SM). One ultimate SM currently being explored at LHC, $pp\to t\bar tt \bar t$ with its large multi-dimensional phase-space an ideal testing ground to explore new ways reduce impact potential mismodelling on experimental We propose novel statistical method capable disentangling 4-top signal from dominant backgrounds same-sign dilepton channel, while simultaneously correcting possible imperfections modelling most relevant discriminating observables -- jet multiplicity distributions. A Bayesian mixture multinomials used light-jet $b$-jet multiplicities under assumption their conditional independence. The background distributions generated deliberately mistuned simulator as priors. posterior distributions, well fractions, then learned using inference. demonstrate that our mitigate effects mismodellings realistic $t\bar tt\bar search, leading corrected better approximate underlying truth-level spectra.