We argue that the ratio $S_3 =mathrm{^3_Lambda H} / (mathrm{^3He} times frac{Lambda}{p})$ is a good representation of the local correlation between baryon number and strangeness, and therefore is a valuable tool to probe the nature of the dense matter created in high energy heavy-ion collision: quark gluon plasma or hadron gas. A multiphase transport model (AMPT) plus a dynamical coalescence model is used to elucidate our arguments. We find that AMPT with string melting predicts an increase of $S_3$ with increasing beam energy, and is consistent with experimental data, while AMPT with only hadronic scattering results in a low $S_3$ throughout the energy range from AGS to RHIC, and fails to describe the experimental data.