Research papers on autonomous systems, memory, governance, and infrastructure.https://serban.ai/enhttps://serban.ai/papers/thermodynamic-memory-model/https://serban.ai/papers/thermodynamic-memory-model/We propose a thermodynamic framework for machine memory where information undergoes phase transitions analogous to physical matter. Ephemeral observations enter as gas-phase particles with high entropy. Through Boltzmann-weighted consolidation, related memories cluster into liquid-phase structures. Sufficiently reinforced patterns crystallize into solid-phase permanent knowledge. The model provides formal temperature and pressure parameters that control consolidation rate, forgetting curves, and retrieval energy. We demonstrate that this framework produces memory systems with emergent properties absent from static vector stores: graceful degradation under load, natural deduplication through crystallization, and energy-proportional retrieval where frequently accessed memories become cheaper to recall.Sun, 01 Mar 2026 00:00:00 GMTthermodynamic-infopersistent-cognitionhttps://serban.ai/papers/organism-state-machines/https://serban.ai/papers/organism-state-machines/Long-lived autonomous software systems require lifecycle management beyond binary alive/dead states and hard wall-clock timeouts. We introduce the Organism State Machine (OSM), a formal model with six canonical states: running, sleeping, degraded, autohealing, quarantined, and stopped. Transitions between states are governed by compulsory heartbeat watchdogs and typed fault/outcome events rather than arbitrary timers. We show that OSM-governed systems exhibit superior fault tolerance compared to timeout-based approaches: failed cycles become learning events rather than termination triggers, stale work degrades gracefully before being failed and redispatched, and the system maintains liveness guarantees without human intervention. The model is implemented in the OSMOS open specification and validated against fleet workloads of 50-500 concurrent autonomous agents.Sun, 15 Feb 2026 00:00:00 GMTorganism-computingerror-intelligence