Terminal-state theorems for a two-stage Maki--Thompson rumor model: final size, stifler composition, and non-identifiability from pooled statistics

Abstract

Rumor models of the Daley--Kendall and Maki--Thompson family anchor the statistical mechanics of social contagion, yet the stifler composition a rumor leaves behind has resisted closed-form analysis. Inspired by the asymmetric two-step pairing (A2SP) process of step-growth polymerization, we import its common-factor cancellation into a minimal two-stage MT variant in which re-contacted spreaders are re-activated into a senior class at rate $αk_1$ rather than stifled, and obtain exact terminal-state theorems. (i) Saturation: the rumor exhausts the ignorant pool for every stifling-to-recruitment ratio $μ>0$, with a three-way tail classification and a logarithmically critical line at $μ=1$. (ii) Final size: the terminal stifler pool splits as $(x_1,x_2)=(μ/(α+μ),α/(α+μ))$, independent of seed size; the fingerprint inverts linearly ($α=μx_2/x_1$) and matches Gillespie simulations within $0.95\%$. (iii) Pooled versus history-resolved: pooled statistics are exactly $α$-free on any static graph; only the history-resolved composition carries $α$. Restoring the classic teller-retires channel breaks the exact closure beyond quantified gates (contact rates of $1$--$3\%$ of $k_1$ for spreader channels at $μ\leq1.5$, up to $32\%$ at $μ=2$), and on configuration-model networks the composition bias decays as $\langle k\rangle^{-0.71}$. An exploratory inversion of 354,930 Higgs--Twitter retweet episodes, screened by an identity check and four negative controls, yields $\hatα=2.21\%$ (joint cluster-bootstrap CI $[1.6\%,3.0\%]$; $\approx3.7\%$ after calibration at the matched threshold). Terminal compositions identify this effective re-activation rate, an upper bound on contact-coupled re-activation; the mechanism itself requires timing information beyond the terminal state.