Mycelial Cache: Topology-Aware Eviction Through Use-Dependent Structural Reinforcement

The Mycelial Cache is a topology-aware cache primitive that maintains a selforganizing weighted mesh over entries, where access patterns create weighted edges throughHebbian strengthening and the mesh topology informs eviction. The primitive targets a recurring constraint class in caching: workloads where co-access structure exists (entries accessedtogether form clusters, with certain entries serving as bridges between clusters) and where classical policies that score entries independently by recency or frequency are structurally blind tothat topology. Inspired by six convergent biological analogs across four phyla (fungal mycelialnetworks, neural synaptic plasticity, bone remodeling, vascular adaptation, slime mold transportnetworks, and ant pheromone trails), the cache uses composite eviction scoring that protectsstructurally important entries. Deterministic replay tests show 83.7% hit rate versus 82.6%(LRU) and 80.1% (moka) on 1K co-access cluster workloads, with 73.0% versus 66.2% (LRU)and 71.5% (moka) under a 1K workload shift. Criterion throughput at 100K shows a mixed costprofile: get() is 544.12 ns for Mycelial versus 12.13 ns for lru and 495.96 ns for moka; set() is1,174.25 ns versus 109.29 ns and 1,773.93 ns respectively. The current implementation exposesbridge-aware eviction, mesh introspection, and fever-based adaptation; predictive prefetchingand topology-following invalidation remain future work.