There is an emerging interest in generating robust algorithmic recourse that would remain valid if the model is updated or changed even slightly. Towards finding robust algorithmic recourse (or counterfactual explanations), existing literature often assumes that the original model m and the new model M are bounded in the parameter space, i.e., ||Params(M)-Params(m)||<. however="" models="" can="" often="" change="" significantly="" in="" the="" parameter="" space="" with="" little="" to="" no="" their="" predictions="" or="" accuracy="" on="" given="" dataset.="" this="" work="" we="" introduce="" a="" mathematical="" abstraction="" termed="" naturally-occurring="" model="" which="" allows="" for="" arbitrary="" changes="" such="" that="" points="" lie="" data="" manifold="" is="" limited.="" next="" propose="" measure="" call="" stability="" quantify="" robustness="" of="" counterfactuals="" potential="" differentiable="" e.g.="" neural="" networks.="" our="" main="" contribution="" show="" sufficiently="" high="" value="" as="" defined="" by="" will="" remain="" valid="" after="" probability="" concentration="" bounds="" lipschitz="" function="" independent="" gaussians="" since="" quantification="" depends="" local="" constant="" around="" point="" not="" always="" available="" also="" examine="" estimators="" proposed="" and="" derive="" fundamental="" lower="" bound="" sample="" size="" required="" have="" precise="" estimate.="" explore="" methods="" using="" measures="" generate="" robust="" are="" close="" realistic="" changes.="" has="" interesting="" connections="" multiplicity="" known="" rashomon="" effect.="">