When applied to statistical systems showing an arctic curve phenomenon, the tangent method assumes that a modification of the most external path does not affect the arctic curve. We strengthen this statement and also make it more concrete by observing a factorization property: if Z
n+k
denotes a refined partition function of a system of n + k non-crossing paths, with the endpoints of the k most external paths possibly displaced, then at dominant order in n, it factorizes as where is the contribution of the k most external paths. Moreover if the shape of the arctic curve is known, we find that the asymptotic value of is fully computable in terms of the large deviation function L introduced in Debin et al (2019 J. Stat. Mech. 113107) (also called Lagrangean function). We present detailed verifications of the factorization in the Aztec diamond and for alternating sign matrices by using exact lattice results. Reversing the argument, we reformulate the tangent method in a way that no longer requires the extension of the domain, and which reveals the hidden role of the L function. As a by-product, the factorization property provides an efficient way to conjecture the asymptotics of multirefined partition functions.