How does deterrence operate when security commitments are deliberately ambiguous? This project argues that deterrence outcomes depend not primarily on formal assurances or signalling credibility, but on how challengers infer the location of the patron's intervention threshold under uncertainty. When commitment structures are imprecise, challengers cannot read the intervention boundary from institutional form — they must infer it from structural signals and observed patron behaviour. A formal model shows that this inference problem makes incremental probing rational, generates deterrent effects through the belief channel independently of formal commitments, and renders deterrence failure partly endogenous to patron behaviour. The argument is examined through structured plausibility probes of Russia and China as challenger cases, combining formal modelling with qualitative process tracing.

How do client states navigate the alignment-hedging continuum when patron tolerance is never fully specified? This project argues that variation in client alignment behaviour within hierarchical security relationships is shaped by beliefs about patron tolerance thresholds rather than material constraints alone. When patrons preserve deliberate ambiguity about the consequences of client autonomy, clients cannot determine their optimal position through deduction from structural conditions. They must instead infer it through sequential policy probing and Bayesian belief updating. A formal model identifies two analytically distinct thresholds — an abandonment threshold and an entrapment threshold — that together define the client's feasible zone of strategic manoeuvre. The argument is examined through structured process tracing of the 2019 GSOMIA episode in South Korea and the S-400 procurement controversy in Türkiye, two cases that provide direct empirical traction on the inference mechanism under conditions of genuine threshold uncertainty.

Do nuclear weapons prevent interstate conflict or merely constrain how far it escalates? This project argues that nuclear capability operates as an escalation ceiling rather than a deterrence shield — dampening the relationship between perceived threat and violence intensity rather than reducing baseline conflict probability. The ceiling effect is not a general property of nuclear weapons but a specific property of mutually nuclear relationships: it requires both sides of a conflict to face nuclear risk simultaneously, and is most binding precisely when perceived threats are most severe. Using dyadic and actor-level data from the International Crisis Behavior Project and an ordered probit specification that separates four nuclear dyadic configurations, the analysis finds that the interaction between threat gravity and nuclear configuration is negative and significant only in mutually nuclear dyads, where the predicted probability of full-scale war falls from 18 percent at low threat to 6 percent at existential threat levels — compared to 45 percent in non-nuclear dyads at the same threat level. Neither asymmetric configuration produces a significant ceiling effect, suggesting that the escalation-constraining benefits of nuclear weapons are absent in precisely the dyadic relationships that non-proliferation regimes are designed to preserve.