Contradiction-Conditioned Replanning

A hierarchical LLM planner pairs with a Bayesian module that tracks a teammate's evolving strategy; instead of replanning on every new observation or ignoring a strategy shift entirely, the agent interrupts its current skill only when the partner's actions directly contradict the currently inferred skill.

Properties

family: Coordination & Scaling
maturity: frontier
diagram: loop
whyMultiAgent: The belief being tracked is a model of a distinct partner's strategy, not the agent's own state; contradiction can only arise between one agent's inferred model and another agent's independently generated actions, so the interrupt condition is structurally undefined for a single agent acting alone. Distinct from the already-tracked progressive-autonomy pattern, which tracks an executor's own trustworthiness over time to gate a supervision mode: here the Bayesian model tracks a teammate's task strategy, and the trigger for action is contradiction between inferred partner intent and observed partner behaviour, not accumulated trust.
mast: [Cat 2]
useCases: [Ad hoc teaming with new teammates whose strategy shifts mid-task][Human-agent or agent-agent collaboration in cooperative games (e.g. Overcooked)][Reducing wasted replanning cost in long-horizon coordination]
source: Bayesian Partner Modelling enables Adaptive Replanning for LLM Coordination · arXiv:2608.18490

Provenance

Source: https://arxiv.org/abs/2608.18490

Contradiction-Conditioned Replanning

Properties

family: Coordination & Scaling
maturity: frontier
diagram: loop
whyMultiAgent: The belief being tracked is a model of a distinct partner's strategy, not the agent's own state; contradiction can only arise between one agent's inferred model and another agent's independently generated actions, so the interrupt condition is structurally undefined for a single agent acting alone. Distinct from the already-tracked progressive-autonomy pattern, which tracks an executor's own trustworthiness over time to gate a supervision mode: here the Bayesian model tracks a teammate's task strategy, and the trigger for action is contradiction between inferred partner intent and observed partner behaviour, not accumulated trust.
mast: [Cat 2]
useCases: [Ad hoc teaming with new teammates whose strategy shifts mid-task][Human-agent or agent-agent collaboration in cooperative games (e.g. Overcooked)][Reducing wasted replanning cost in long-horizon coordination]
source: Bayesian Partner Modelling enables Adaptive Replanning for LLM Coordination · arXiv:2608.18490

Provenance

Source: https://arxiv.org/abs/2608.18490