Biological cognition as viability-constrained expansion of controllable futures

Biological cognition is often described through homeostasis, prediction, exploration, or control. However, none of these principles alone specifies how a living system should act while preserving viability and enlarging its future possibilities. We tested whether living-cognition-like policy selection can be operationalized as viability-constrained expansion of controllable futures. A minimal grid-world model was implemented in which agents evaluated finite-horizon policies under internal viability dynamics, hazards, resources, movement costs, and reachable-endpoint controllability. Seven reduced policy-selection models were compared across 1000 critical environments containing viability risk, expansion opportunity, and viability-control conflict. The viability-constrained endpoint and joint viability-expansion models produced the highest joint viable-expansion scores. At the family level, viability-constrained expansion substantially outperformed models based on homeostatic preservation, avoidance of previously unvisited states, novelty seeking, random admissible action, or unrestricted controllability. Unrestricted controllability frequently violated viability, whereas homeostasis-only and uncertainty-minimizing selection preserved viability but produced limited or zero expansion. The family-level advantage remained positive across the default configuration and 47 prespecified parameter variations. These findings support viability-constrained future expansion as a candidate minimal operational principle for living-cognition-like policy selection in state-constrained agents. Under the tested environmental and computational conditions, the framework is a computational proof-of-principle and does not establish the biological or neural mechanisms of cognition.